Energy system optimization based on fuzzy decision support system and unstructured data

Abstract

To address the complex challenges in energy systems, this study proposes a novel optimization framework that integrates fuzzy decision support and unstructured data processing technologies. This framework aims to improve efficiency, reduce costs, decrease environmental impact, increase system flexibility, and enhance user satisfaction, thereby promoting sustainable development in the energy industry. The framework combines the innovative Energy Semantic Mapping Model (ESMM) and the advanced deep learning architecture ResNet to process textual and visual data effectively. ESMM enables accurate prediction of energy demand, while ResNet significantly reduces equipment maintenance costs and improves energy distribution efficiency. These advancements are critical as they address the limitations of existing approaches in handling large-scale unstructured data and making informed decisions under uncertainty. The Environmental Impact Assessment (EIA) confirms the model's effectiveness in reducing carbon emissions. A comprehensive economic analysis demonstrates substantial cost savings in energy procurement and operations and maintenance, with overall savings exceeding 25%. Enhanced user satisfaction and reduced system response times further validate the practical utility of the proposed approach. Additionally, a genetic algorithm is used to optimize the fuzzy rule base, enhancing the robustness and adaptability of the model. Experimental results show superior performance compared to traditional systems, providing strong empirical evidence for the intelligent transformation of energy systems. This research contributes to the field by offering a more sophisticated and flexible solution for managing energy systems, particularly in terms of leveraging unstructured data and improving decision-making processes.

Introduction

Globally, efficient operation and optimal allocation of energy systems have become core issues in promoting sustainable development and addressing climate change. With the rapid development of economy and the continuous growth of population, the global energy demand rises sharply, while the limitation of energy resources and environmental pressure are increasingly severe. Traditional energy systems are often faced with problems such as low efficiency, unreasonable resource allocation, and serious environmental pollution. Intelligent and refined management methods are urgently needed to improve the overall efficiency (Raz et al. 2021; Sheng and Ye 2024).

In recent years, the deep integration of information technology and energy systems has provided new solutions to this problem. Data-driven decision support system has great potential in energy system optimization because of its powerful information processing ability, fast response characteristics and intelligent decision-making advantages. However, real-world energy data is highly complex, not only containing a large amount of structured data, such as real-time monitoring of energy consumption data, equipment status information, but also involving a wide range of unstructured data sources, such as weather forecasts, policy documents, user feedback, etc. These unstructured data contain rich information, but their processing difficulty is much higher than structured data, which becomes a key factor restricting the efficiency and accuracy of energy system optimization (Rai and Das 2022).

In the energy field, Fuzzy Decision Support System (FDSS) has been applied to energy demand forecasting, generation scheduling, load management and other aspects, effectively improving the flexibility and accuracy of decision-making. For example, some studies have built fuzzy inference models to accurately predict renewable energy generation and optimize the matching of energy supply and demand (Ponce et al. 2022).

With the rise of big data technology, the processing power of unstructured data has been significantly enhanced. The application of advanced technologies such as natural language processing, computer vision and deep learning makes it possible to extract valuable information from unstructured data such as massive texts, images and videos. In energy systems, these technologies are used to analyze user feedback on social media to optimize service, or to monitor the status of solar panels using satellite imagery, increasing the level of intelligence in energy management (Zhang et al. 2022).

Faced with significant achievements and key challenges remaining in the field of energy system optimization, such as the burdensome nature of unstructured data integration processing, the robustness and accuracy requirements of fuzzy decision models in the face of high uncertainty, the reconciliation of complexity between theoretical and practical applications, and the importance of data privacy protection, this study focuses on three core issues: First, it aims to develop efficient strategies and tools to accelerate the integration and analysis of unstructured data to ensure that data processing is accurate and efficient; second, it aims to design and optimize more robust and adaptive fuzzy decision models to more effectively manage complex uncertainties in energy system optimization; Finally, consider the actual deployment path of the model under strict compliance with data security and privacy protection principles to ensure that the technical solution is both advanced and practical (Wang et al. 2024a).

Therefore, the core goal of this study is to develop a new framework that integrates unstructured data processing and fuzzy decision support functions to comprehensively improve the intelligent level of energy system data management. At the same time, through targeted model optimization work, we strive to better adapt the fuzzy inference algorithm to the unique requirements of the energy system, enhance its stability and prediction accuracy under variable conditions.

Against the backdrop of surging global energy demand and growing environmental awareness, efficient operation of energy systems has become critical. The shortcomings of traditional energy systems, such as inefficient resource allocation and environmental pollution, need to be solved by intelligent means. In recent years, the convergence of information technology and energy systems has opened up new opportunities. Data-driven decision support systems can handle real-time monitoring data and other structured information, but challenges remain in handling unstructured data. These data include text, images, etc., which contain important information, but are difficult to process. Fuzzy Decision Support System has been applied in energy field, especially in demand forecasting and scheduling. At the same time, advances in unstructured data processing technologies, such as natural language processing and computer vision, also open up more possibilities for optimizing energy systems. The aim of this study is to develop a framework combining unstructured data processing and fuzzy decision support system to improve the intelligent management level of energy system. Through effective data integration strategies and advanced fuzzy decision models, the framework aims to solve unstructured data processing problems and enhance model robustness and accuracy in complex uncertain environments.

Compared to existing literature, our approach introduces the Energy Semantic Mapping Model (ESMM) and the ResNet deep learning architecture to process textual and visual data effectively. ESMM enables accurate prediction of energy demand, while ResNet significantly reduces equipment maintenance costs and improves energy distribution efficiency. Additionally, a Genetic Algorithm optimizes the fuzzy rule base, enhancing the robustness and adaptability of the model.

In this paper, we first present the theoretical foundations and the design of our optimization framework. We then describe the ESMM and ResNet methodologies, followed by a detailed explanation of how the Genetic Algorithm is employed to optimize the fuzzy rule base. Empirical results demonstrate the effectiveness of our approach in reducing carbon emissions, achieving cost savings, and enhancing user satisfaction. Finally, we discuss the implications of our findings and suggest future research directions.

Literature review

With the increasing global energy demand and the increasingly severe environmental problems, the progress of energy system optimization technology has become the focus of both academia and industry. This paper summarizes the latest progress of energy system optimization, fuzzy decision support system and unstructured data processing technology in recent years, and summarizes the experience and lessons through relevant case studies, aiming to provide theoretical guidance and practical reference for intelligent upgrading of energy system.

Progress in energy system optimization technology

In recent years, the global energy field is undergoing a profound transformation, in which the development of energy system optimization technology is particularly striking, which is not only reflected in the diversification of technology paths, but also constantly breaking through boundaries in deep integration and cross-border integration, committed to creating a more efficient, intelligent and green energy management system. Two core trends stand out in this transformation: the widespread adoption of Integrated Energy Systems (IES) and the deep convergence of AI and Big Data technologies (Dhiman and Deb 2020).

The popularization of integrated energy system design concept is a thorough innovation to the isolated operation mode of traditional energy system. IES advocates closely connecting various energy forms such as electricity, heat energy, cold energy and gas, and realizing efficient synergy and complementary utilization of energy flow through intelligent regulation. This multi-energy coupling method not only optimizes the utilization efficiency of single energy, but also promotes the optimization of energy structure and flexible matching of energy supply and demand at the macro level. For example, in a study (Sung et al. 2024), mixed integer linear programming (MILP) tools were skillfully used to comprehensively optimize the design of an integrated energy system in a region, which not only effectively reduced the energy cost of the region, but also significantly reduced carbon emissions, fully demonstrating the potential of IES in promoting energy economy and environmental friendliness. At the same time, with the rapid progress of artificial intelligence technology, especially the combination of deep learning and big data analysis technology, a new era of intelligence has been opened for energy system optimization. Deep Reinforcement Learning (DRL), as a breakthrough in the field of artificial intelligence, shows great application value in many key links such as energy scheduling, demand response and fault prediction by virtue of its powerful learning and decision-making ability (Rafiq et al. 2021). Deep Q-Network (DQN) is creatively introduced into the scheduling strategy of distributed energy systems. This strategy makes full use of the advantages of DQN in dealing with complex and dynamic environments. It can not only learn and adapt to the dynamic changes of energy systems in real time, but also significantly enhance the scheduling flexibility and operation reliability of the system, providing powerful technical support for coping with the increasing complexity and uncertainty of distributed energy systems.

At a deeper level, the integration of artificial intelligence and big data technology not only optimizes the operational efficiency of energy systems, but also provides more accurate data support and prediction capabilities for energy management decisions. Through deep mining and analysis of massive historical data, AI models can accurately predict energy demand trends, guide the optimal allocation of energy production and distribution, reduce waste, and improve overall energy efficiency (Dong et al. 2018). At the same time, with the help of self-learning and optimization capabilities of machine learning algorithms, energy systems can continuously learn from operational data, dynamically adjust strategies, and achieve continuous performance improvement and cost reduction, laying a solid foundation for building a more intelligent and resilient energy Internet.

As energy demand increases and technology advances, energy systems are facing unprecedented challenges. Existing energy system optimization techniques have limitations in dealing with large amounts of uncertain and unstructured data, which limits their effectiveness in practical applications. Traditional optimization models often assume that the data is structured and the future demand is predicted with certainty, which is inconsistent with the actual situation. In addition, many of the existing methods in dealing with the complex environment of multi-objective decision-making problems are inadequate.

Principle and application of fuzzy decision support system

Fuzzy Decision Support System (FDSS), as an effective tool for dealing with uncertain problems, has been widely used in energy system optimization. Its core lies in fuzzy set theory, which introduces membership function to describe the membership degree of elements to different sets, so as to deal with common fuzzy phenomena in reality. In recent years, FDSS has been used more deeply in energy systems, especially in demand-side management and renewable energy forecasting. For example, reference (Singh et al. 2021) constructs a model based on fuzzy C-means clustering and fuzzy time series analysis, which accurately predicts regional power load and provides strong support for power grid dispatching.

In order to overcome these challenges, this study proposes a new energy system optimization framework that integrates fuzzy decision support system (FDSS) and unstructured data processing techniques. FDSS can deal with uncertainty effectively and provide support for complex decision-making process. It simulates the human reasoning process through fuzzy logic, so as to better deal with the uncertain factors in the real world. This capability is particularly important for energy systems, which need to make optimal decisions in a constantly changing environment. In this study, we use FDSS to deal with uncertainties in the energy system, such as changes in weather conditions, fluctuations in market demand, etc. By introducing fuzzy logic, we are able to predict energy demand more accurately, optimize resource allocation, and improve overall system efficiency. In addition, we employ advanced unstructured data processing technologies such as the Energy Semantic Mapping Model (ESMM) and the ResNet deep learning architecture to process text and image data to enable accurate energy demand forecasting, reduce equipment maintenance costs, and improve energy distribution efficiency. These technologies are specifically designed to address the unique challenges of energy systems, such as ESMM, which extracts valuable information from unstructured text data to help predict energy demand patterns, and ResNet, which identifies key features from image data to monitor equipment status and predict potential failures. The integration of these technologies enables our framework to handle unstructured data while also effectively managing energy systems for greater flexibility and responsiveness.

Overview of unstructured data processing technology

Unstructured data has become an important information source in energy system optimization because of its rich content and potential value. Recent advances are mainly reflected in the application of natural language processing (NLP), computer vision (CV) and deep learning techniques. NLP technology is used to analyze social media, news reports, etc. to obtain user behavior patterns and market trends. For example, reference (Shukla et al. 2020) uses sentiment analysis technology to mine users 'attitudes towards new energy vehicles from social media, providing data support for product development and marketing strategies.

In CV, satellite imagery and video data captured by drones are used to monitor and assess the state of energy infrastructure (Mudric-Staniskovski et al. 2024). A set of image recognition system based on deep learning was developed, which can automatically identify the damage of wind turbine blades and improve maintenance efficiency.

Relevant case studies and lessons learned

In recent years, several successful case studies have demonstrated the practical effectiveness of these technologies in energy system optimization. For example, a case study of smart microgrids, integrating FDSS and big data analytics, achieved efficient management of distributed energy sources, significantly improving energy self-sufficiency and economy (Endrei et al. 2019). However, these cases also reveal some common challenges and lessons: (1) Data quality and integrity: Pre-processing and cleaning of unstructured data is a prerequisite for model accuracy, but it is often time-consuming and labor-intensive. Interpretability and trust of models: While deep learning models perform well on prediction and optimization tasks, their “black box” nature limits transparency and user acceptance. (3) Interdisciplinary collaboration: The complexity of energy system optimization requires interdisciplinary collaboration between technology, economics, environment and social sciences, which is sometimes difficult to coordinate in practice (Li 2022).

In conclusion, energy system optimization technology has made remarkable progress in fuzzy decision support systems and unstructured data processing, but it also faces challenges in data processing, model interpretation and interdisciplinary collaboration. Future research needs to further explore how to efficiently integrate emerging technologies, improve the intelligent level of energy systems, and balance economy, environmental friendliness and social acceptability to contribute to the achievement of sustainable development goals.

Energy system optimization technology has experienced significant development in recent years, especially in the widespread application of integrated energy systems (IES) and the deep integration of artificial intelligence and big data technology. IES achieves efficient and synergistic use of multiple energy forms through intelligent regulation, significantly improving energy efficiency and reducing carbon emissions. Artificial intelligence technology, especially the combination of deep learning and big data analysis, provides strong support for the intelligentization of energy systems (Kumar et al. 2021). For example, the excellent performance of deep reinforcement learning technology in energy scheduling, demand response, etc. proves its potential in dealing with complex dynamic environments. Fuzzy decision support systems (FDSS) are also widely used in energy system optimization, especially in demand-side management and renewable energy forecasting, such as power load forecasting through fuzzy C-means clustering and fuzzy time series analysis. In addition, advances in unstructured data processing technologies, such as natural language processing and computer vision, provide a richer source of information for energy system optimization. However, the application of these techniques also faces challenges such as data quality, model interpretability, and interdisciplinary collaboration (Yi et al. 2022).

Research technique

Unstructured data processing strategy

In order to solve the complexity and processing difficulty of unstructured data in the energy field, this study adopts an innovative natural language processing technology-Energy Semantic Mapping Model (ESMM), which can deeply mine and accurately analyze professional text data. ESMM realizes deep semantic mapping of energy texts by optimizing the expression formula of word vectors, combining with context filtering and feature weighting strategies of specific domains. In addition, ResNet, a deep learning-based image recognition system, is used to monitor and assess the state of energy infrastructure, such as automatically identifying damage to wind turbine blades to improve maintenance efficiency. Through these techniques, unstructured data is efficiently transformed into useful information that can be used in decision support systems.

This chapter details the research methodology and technical framework adopted by the project to achieve efficient optimization and intelligent management of energy systems by integrating unstructured data processing, fuzzy decision support system design, data privacy protection mechanisms, and building an integrated overall framework (Kumar et al. 2021).

With the acceleration of urbanization and the development of social economy, the urban power grid is facing unprecedented challenges. On the one hand, the surge of power demand caused by high temperature in summer brings great pressure to the safe and stable operation of power grid; on the other hand, the wide access of distributed energy and the application of digital technology put forward higher requirements for the intelligent level of power grid. These challenges are not only technical, but also economic and environmental, and together form broader issues in energy management.

Taking the Urban area operation and maintenance office of Yuechi Aizhong Electric Power Company as an example, the operation and maintenance office is facing severe test during the peak summer. High temperature weather leads to a sharp increase in power demand and peak load of power grid, which requires the operation and maintenance department to take effective measures to ensure the safe and stable operation of power grid. In order to meet this challenge, the Operation and Maintenance Institute has taken a series of measures, including formulating detailed patrol inspection scheme for power supply in summer in advance, strengthening special patrol frequency for key facilities, and responding quickly to various emergencies. These measures not only improve the efficiency of operation and maintenance, but also significantly reduce the failure rate of the grid.

The Energy Semantic Mapping Model (ESMM) and ResNet deep learning architecture adopted in this study can effectively process a large amount of unstructured data generated in power grid operation, such as weather forecast, user power consumption behavior records, etc. ESMM can extract useful information from this unstructured data to help forecast energy demand, which is critical to the proper allocation of power resources. ResNet identifies key features from the image data that can be used to monitor device status and predict potential failures, enabling early action to avoid failures and reduce maintenance costs.

In addition, fuzzy decision support system (FDSS) is used to deal with the uncertainty of power grid operation, such as weather change and market fluctuation. FDSS simulates human reasoning process by fuzzy logic, which can better deal with these uncertain factors and ensure the stable operation of power grid.

Under the background of increasingly complex energy research and management, the need for deep mining and accurate analysis of professional text data is increasingly urgent. This article introduces an innovative natural language processing technique called the Energy Semantic Mapping Model (ESMM), which inherits and surpasses the traditional Word2Vec model and is optimized for the characteristics of the energy domain to more accurately capture the deep semantic structure of industry-specific language and terminology (Yi et al. 2022).

The core of ESMM lies in its optimized word vector expression formula, which realizes deep semantic mapping of energy text by introducing domain-specific context filtering and feature weighting strategy. This is shown in Eq. 1.

$$v_{{w_{i} }}^{*} = \frac{1}{{|C^{ + } (w_{i} )|}}\sum\limits_{{c \in C^{ + } (w_{i} )}} {\alpha_{c} } \cdot v_{c} + \beta \cdot \phi (w_{i} )$$

(1)

where: $v_{{w_{i} }}^{*} \,$ denotes the optimized vector representation of the word $w_{i} \, .(C^{ + } (w_{i} ) \,$: denotes the set of contexts optimized for the energy domain, filtered by keyword filtering, topic modeling, and domain expert knowledge to ensure that the selected contexts are highly relevant and to enhance the professional focus of the model. $\alpha_{c}$: denotes the weight of the context c, which reflects the contribution of the context to the vector representation of the word. $v_{c}$: represents the vector representation of context c. $\beta$: represents the feature weighting factor of the word $w_{i}$, which reflects the importance of the features of the word itself. $\phi (w_{i} )$: represents the original feature vector of the word $w_{i}$. The formula generates an optimized word vector representation by combining the contextual information of the word $w_{i}$ and the original feature information, which integrates not only the contextual information but also the domain expertise.

Among them, the optimized vector representation of the representation words under the ESMM framework not only integrates context information, but also integrates domain expertise (Almulhim and Barahona 2020).

Figure 1 details the flow of an unstructured data processing strategy, which starts with collecting raw unstructured data from multiple sources, which may include text, images, audio, and more. Then, through the deep semantic mapping step, deep learning technology is used to analyze and understand the semantics of these data and convert them into a format that is easy for computers to process. This process involves advanced methods in the fields of natural language processing and computer vision, aiming at extracting deep semantic information from the data. Next, through the generation of word vector representations phase, word embedding techniques are used to convert words into vector forms, which not only help machines understand and process words, but also reveal similarities and correlations between words. In the context filtering and weighting step, the extracted features are filtered and weighted, irrelevant or redundant information is removed, and the importance of each feature is evaluated by techniques such as TF-IDF, thereby improving the efficiency of subsequent processing. Finally, in the feature vector fusion phase, features from different data sources are integrated into the final feature vector for deeper analysis and more accurate decisions.

Fig. 1

Unstructured data processing strategy flow

The calculation of the weighting coefficient introduces a TF-IDF (word frequency-inverse document frequency)-based weighting or more complex Attention Mechanism to dynamically adjust the size of each context c's semantic contribution to the target word, formulated as Eq. 2 (Li et al. 2023).

Equation 2 is a formula for calculating a term frequency-inverse document frequency (TF-IDF) weighting coefficient, which means: $\alpha_{c}$ Indicates the semantic contribution of context c to target word w IDF(c): Indicates the inverse document frequency, which is the logarithmic inverse of the number of documents in the vocabulary that contain the word c. The larger the IDF, the more unique the word c is across the corpus, and vice versa.$\sum\limits_{{c^{\prime}C^{ + } (w_{j} )}} {f_{{c^{\prime}}} } IDF(c^{\prime})$: represents the sum of the word frequencies multiplied by IDF of all words belonging to w context window. Equation 2 therefore calculates the relative importance of word c relative to the other words by multiplying the word frequency by the inverse document frequency and dividing by the sum of the word frequencies of all words in its context window multiplied by IDF. This allows words that are less common in the corpus to receive higher weights because they are better able to distinguish particular documents or sentences.

$$\alpha_{c} = \frac{{f_{c} \times {\text{IDF}}(c)}}{{\sum\limits_{{c^{\prime} \in C^{ + } (w_{i} )}} {f_{{c^{\prime}}} } \times {\text{IDF}}(c^{\prime})}}$$

(2)

where is the frequency with which context c occurs in the text and is the inverse document frequency, reflecting how rare c is in the entire corpus, thereby emphasizing the importance of the information (Tiwari et al. 2018).

The introduction of intrinsic feature vectors enhances the ability to directly understand and distinguish technical terms through pre-trained domain-specific word embeddings (such as GloVe or FastText) or entity embeddings based on knowledge graphs. The formula can be expressed as Eq. 3.

Equation 3 describes the process of converting a word $w_{i}$ into a domain-specific embedding vector $\phi (w_{i} )$. This step involves transforming the raw text data into numerical representations that capture the meaning and context of the words in the energy domain. By leveraging pre-trained word embeddings tailored specifically for the energy sector, the model can better understand the unique vocabulary and concepts used in the field. This enhanced understanding allows for more precise analysis and prediction of energy-related phenomena.

$$\phi (w_{i} ) = {\text{Embedding}}_{{\text{domain - specific}}} (w_{i} )$$

(3)

In energy facility monitoring, the application of computer vision technology, especially the deep learning model ResNet (Residual Network), marks a new stage in intelligent monitoring. ResNet effectively solves the gradient disappearance problem in deep network training through its original residual structure design. Its basic composition can be abstracted as Eq. 4.

Equation 4 represents the basic structure of the ResNet model, which is a type of deep neural network architecture commonly used in computer vision tasks. The key innovation of ResNet is its use of residual blocks, which help mitigate the vanishing gradient problem during training. The equation shows that the output $y$ is equal to the input $x$ plus the result of passing $x$ through a series of functions $F(x,W_{i} )$, where $W_{i}$ represents the weights of the network. This residual connection allows the network to learn identity mappings, making it easier to train deeper networks without suffering from degradation problems.

$$y = x + F(x,\{ W_{i} \} )$$

(4)

where y is the output, x is the input signal, and F is a complex function with multiple layers that depend on the set of parameters. ResNet's mechanism, especially in identifying subtle defects such as microcracks in wind turbine blades, has demonstrated superior ability to significantly improve the accuracy and response speed of fault warning systems, revolutionizing maintenance management of energy facilities (Nayyar and Singh 2020).

Data pretreatment, including data cleaning, standardization and normalization, is the premise to ensure the quality of data analysis. PCA (Principal Component Analysis) was used for feature extraction, which was mathematically described as Eq. 5.

$${\text{T = XW}}$$

(5)

PCA greatly optimizes the data structure by maximizing the variance of the data while filtering out insignificant noise, injecting efficiency and precision into the subsequent model learning process, and ensuring superior performance of the model when dealing with large-scale unstructured data (Park 2022).

To ensure that only authorized personnel can access the data, strict access controls are implemented. These controls include role-based access control (RBAC) and attribute-based access control (ABAC) mechanisms. RBAC is designed to grant access permissions based on the user's role within the organization, whereas ABAC provides a more granular approach by taking into account various attributes related to the user, the resource, and the environment.

ABAC is particularly useful in managing access to sensitive data in dynamic environments. It uses a policy engine to evaluate attributes and determine whether a user should be granted access to a resource. The decision-making process is based on a set of rules defined in an access control policy. The policy engine evaluates a request against the policy and determines whether the access should be granted or denied. A simplified version of the ABAC decision-making process can be represented by the following formula:${\text{Access Decision}} = {\text{Policy Engine}}(P,U,R,E)$ Where: P represents the set of policies. U represents the set of user attributes. R represents the set of resource attributes. E represents the set of environmental conditions. The Policy Engine evaluates the combination of U, R, and E against the set of policies P to determine whether the access request should be granted G or denied D.

For example, if a user requests access to a specific dataset, the Policy Engine would evaluate the user's role (e.g., analyst, manager), the sensitivity level of the dataset (e.g., public, confidential), and the current time (e.g., business hours, non-business hours). Based on these attributes and the defined policies, the Policy Engine would determine whether the user should be granted access to the dataset.

Fuzzy decision support system

The design and optimization strategy of fuzzy decision support system (FDSS) is the core of this study. Fuzzy rule base is the core of FDSS, which is constructed based on domain expert knowledge and historical data. By introducing self-adaptive adjustment mechanism, fuzzy rule base can adjust rule weights dynamically according to feedback of system operation to ensure real-time and accuracy of model. In addition, genetic algorithms are used to optimize fuzzy rules and select the optimal rule combination through crossover and mutation operations, thereby improving the overall performance of the model. Fuzzy reasoning process includes fuzzification, inference and defuzzification. Higher level uncertainty is dealt with by introducing higher level reasoning methods such as Type-2 fuzzy logic or intuitionistic fuzzy logic. The application of genetic algorithm can optimize the parameter configuration of fuzzy system step by step, improve the whole performance of fuzzy system, and realize intelligent management of energy system.

In this section, we will explore in depth the specific design and optimization strategies of fuzzy decision support systems (FDSS) in energy system optimization, aiming to improve the robustness, adaptability and prediction accuracy of models when dealing with complex uncertain problems. Fuzzy rule base is the core of FDSS, which is based on domain expert knowledge and historical data. According to the characteristics of the energy system, the rule base should cover multiple dimensions such as energy demand forecasting, resource allocation, and equipment operation and maintenance. The key of optimizing fuzzy rule base lies in simplification and refinement of rules. By introducing adaptive adjustment mechanism, the weight of rules or rules added or deleted dynamically according to the feedback of system operation can ensure the real-time and accuracy of the model. In addition, we use genetic algorithm to optimize fuzzy rules, and select the optimal rule combination through crossover and mutation operations to improve the overall performance of the model (Wang et al. 2024b).

Fuzzy rule base is the core component of fuzzy decision support system, and its construction is based on domain expert knowledge and historical data. A typical fuzzy rule can be expressed as Eq. 6. Equation 6 describes a typical fuzzy rule that states that when the input variable( x belongs to one fuzzy set A_i, the output variable( y should belong to another fuzzy set $B_{i}$.

$$R_{i} :{\text{If }}x_{1} {\text{ is }}A_{1i} {\text{ and }}x_{2} {\text{ is }}A_{2i} {\text{, then }}y{\text{ is }}B_{i}$$

(6)

where, denotes the ith rule, is the input variable, is the corresponding input fuzzy set, y is the output variable, is the output fuzzy set (Simic et al. 2021).

In order to enhance the adaptability and dynamic adjustment ability of fuzzy rule base, adaptive mechanism can be introduced to update the parameters of membership function through online learning algorithm. Taking the gradient descent method as an example, the updating formula of fuzzy rule parameters can be expressed as Eq. 7.

Equation 7 represents the process of updating fuzzy rule parameters by gradient descent, where the parameter vector $\theta$ is adjusted at each iteration according to the gradient of the loss function to minimize the loss function.

Where $\theta_{t + 1}$ denotes the parameter vector at time t + 1, i.e. the updated parameter. $\theta_{t}$: indicates the parameter vector at time t, i.e. the current parameter. $\alpha$: represents the learning rate, which controls the step size of the parameter update in each iteration. $\nabla J(\theta_{t} )$: denotes the gradient of the loss function J at the parameter $\theta_{t}$, which reflects the direction and speed of the change of the loss function at the current parameter value. The formula optimizes the performance of the model by adjusting the parameter vectors according to the gradient of the loss function through the gradient descent algorithm, which makes the loss function decrease gradually. Learning rate $\alpha$. The choice of learning rate has an important impact on the convergence speed and stability of the optimization process.

$$\theta_{t + 1} = \theta_{t} + \alpha \cdot \nabla J(\theta_{t} )$$

(7)

Here, the parameter vector at time t is the learning rate, is the loss function, and is the gradient of the loss function with respect to the parameter (Gupta and Goyal 2021).

Fuzzy inference process usually involves three steps: fuzzification, inference and defuzzification. On the basis of the classical Mamdani model, more advanced reasoning methods, such as Type-2 fuzzy logic or intuitionistic fuzzy logic, can be introduced to deal with higher levels of uncertainty, the framework of which is shown in Fig. 2.

Fig. 2

Framework of fuzzy decision system

Type-2 fuzzy logic introduces two-level membership degree, and membership degree of fuzzy set itself is also a fuzzy set. A simple Type-2 membership function can be expressed as Eq. 8.

Equation 8 defines the membership function of simple type 2 fuzzy sets, which expresses the second degree membership of fuzzy sets in the form of integrals.

$$\mu_{A} (x) = \int_{u \in J} {\mu_{A} } (x,u)du$$

(8)

where J is the membership space of the second rank and is the membership at u for each x (Roy et al. 2023).

Intuitionistic fuzzy logic provides a more detailed description by introducing the concepts of membership degree and non-membership degree. The intuitive membership function is defined as Eq. 9 (Çoban 2020). Equation 9 describes the definition of each element in an intuitionistic fuzzy set, including membership $\mu_{A} (x)$, non-membership $\nu_{A} (x)$, and the resulting hesitancy $h_{A} (x)$.

$$\mu_{A} (x) = (\mu (x),\nu (x),\pi (x))$$

(9)

Among them, is the degree of membership, is the degree of non-membership, is the degree of hesitation, reflecting the uncertainty of the system for element classification (Quteishat and Younis 2023).

The performance evaluation of fuzzy systems usually involves precision, recall, F-score and other indicators. However, other evaluation methods can also be considered for the unique properties of fuzzy systems, such as fuzzy entropy to measure the uncertainty of the system output, which is expressed as Eq. 10. Equation 10 defines the entropy (H(A) of a fuzzy set (A, which is used to measure the uncertainty of a fuzzy set by calculating the logarithmic sum of membership degrees.

$$H(A) = - \sum\limits_{i = 1}^{n} {\mu_{i} } \log (\mu_{i} )$$

(10)

where H(A) is the entropy of fuzzy set A and is the membership degree of the ith element (Dhanalakshmi and Ayyanathan 2022).

Genetic algorithm (GA) is widely used in fuzzy system optimization as an efficient heuristic global search method, especially in adjusting fuzzy rule base and membership function parameters to minimize specific performance index. The following are the specific steps and related formulas applied by genetic algorithm in fuzzy system optimization:

First, we need to encode the parameters of fuzzy system (such as membership function parameters, fuzzy rule weights, etc.) into chromosomes. Common coding methods include binary coding, real coding, etc. Assuming we use binary encoding, each parameter is represented by a binary string of a certain length, expressed as Eq. 11 (Kaya et al. 2019).

Equation 11 represents the binary encoding of parameters in the genetic algorithm, where each parameter is encoded as a binary string.

$${\text{individual}} = (\underbrace {{b_{1} b_{2} .....b_{n} }}_{{{\text{parameter}}\theta_{1} }},\underbrace {{b_{n + 1} b_{n + 2} ....b_{2n} }}_{{{\text{parameter}}\theta_{2} }},...,\underbrace {{b_{k + 1} b_{k + 2} ....b_{m} }}_{{{\text{parameter}}\theta_{p} }})$$

(11)

where n to m represent the number of parameters and p is the total number of parameters.

The fitness function is used to evaluate the quality of an individual, i.e., system performance. In the background of fuzzy system optimization, fitness function is usually associated with prediction error, control accuracy, response speed and other indicators of the system. For example, if the prediction error is minimized, the fitness function can be defined as Eq. 12 (Simic et al. 2021). Equation 12 defines the fitness function, which is a criterion for evaluating individual quality, usually related to prediction error or other performance indicators.

$$J(\theta ) = \frac{1}{N}\sum\limits_{i = 1}^{N} {(y_{i} \hat{y}_{i} )^{2} }$$

(12)

Here, N is the number of samples, is the true value, is the predicted value, and the smaller it is, the better the prediction performance of the system.

Figure 2 shows the framework of a fuzzy decision system. The core component is the blue inference engine, which receives inputs and outputs decisions. The system also contains a white square in the middle, representing domain expert knowledge and historical data, which are the basis of constructing fuzzy rule base, and fuzzy rule base is the key to guide the system to make decisions. The yellow rectangles represent non-fuzzy input and fuzzy processing respectively, where the non-fuzzy input is first transformed into fuzzy data by fuzzy processing, and then enters the inference engine for further processing. In addition, the grey squares represent online learning algorithms that enable the system to improve the accuracy and adaptability of decisions by continuously optimizing and adaptively adjusting the rule base to cope with new data and experiences. Finally, another yellow rectangle is connected to the non-fuzzy output and defuzzification process, indicating that the fuzzy results processed by the inference engine will be transformed into clear decision outputs, completing the whole decision process.

Crossover mimics gene exchange in the genetic process of biology, producing new individuals. A single-point crossover is used, specifically Eq. 13, where p is a randomly selected crossover point. Equation 13 describes a single-point crossover operation, which is one of the ways new individuals are generated in genetic algorithms.

$${\text{children}} = {\text{parent}}_{1} [:p] + {\text{parent}}_{2} [p:]$$

(13)

Mutation randomly changes certain positions of chromosomes with a small probability in order to maintain the diversity of the population. The binary mutation operation can be expressed as Eq. 14. where is the mutation probability. Equation 14 describes binary mutation operations that increase population diversity by randomly changing certain bits of chromosomes with small probability.

$$b_{{i^{\prime}}} = \left\{ {\begin{array}{*{20}l} {1b_{i} ,} \hfill & {{\text{with probability }}P_{m} } \hfill \\ {b_{i} ,} \hfill & {{\text{otherwise}}} \hfill \\ \end{array} } \right.$$

(14)

By repeating the steps of selection, crossover and mutation, a new generation population is generated until predetermined stopping conditions are met, such as reaching the maximum iteration number, fitness improvement less than a threshold, and the like. The algorithm flow can be summarized as shown in Fig. 3.

Fig. 3

Flow chart of genetic algorithm

Figure 3 details the flow of the genetic algorithm, which first creates a set of randomly generated solutions, or initial populations, through the initialization population step, where each solution is called a chromosome and consists of a sequence of genes representing a possible solution to the problem. Then, the algorithm enters into the link of evaluating fitness value and protecting optimal chromosome, calculating fitness value of each chromosome and saving current optimal solution. Then, selection is performed based on fitness values to select chromosomes that are parents to promote inheritance of superior characteristics. Parental chromosomes produce offspring by cross-swapping genetic information to fuse strengths and create better solutions. During this process, mutation operations occur occasionally to maintain population diversity and avoid premature convergence. The fitness values of the new generation chromosomes are evaluated again and the optimal chromosomes are updated. The algorithm repeats these steps until a termination condition is met, such as the maximum number of iterations is reached or the population fitness no longer improves. Eventually, the algorithm stops and returns the current best chromosome as the approximate optimal solution to the problem.

Through the above process, genetic algorithm can optimize the parameter configuration of fuzzy system step by step, realize the minimization of pairs, improve the overall performance of fuzzy system, and achieve the purpose of optimizing fuzzy decision support system.

Through the above methods, the design and optimization of fuzzy decision support system can be deepened, which not only enhances the processing ability of complex energy system decision problems, but also improves the adaptability and robustness of the model, and provides solid technical support for realizing intelligent management of energy system.

Energy system optimization modeling

Model

Energy system optimization modeling is the key to achieving energy efficiency, reducing operating costs and improving environmental sustainability. This chapter describes in detail how to construct an energy system optimization model that integrates unstructured data processing, fuzzy decision support systems, and advanced optimization algorithms. This model is designed to solve complex and variable energy management problems, including but not limited to energy allocation, demand forecasting, failure prevention and cost control.

The core of an energy system optimization model is to determine one or more objective functions that reflect the main objectives of system optimization, such as minimizing total operating costs, maximizing energy efficiency, and minimizing environmental impact. A typical objective function can be expressed as Eq. 15. Equation 15 defines the objective function of the energy system optimization model, which is usually the minimization of the total cost, which is the sum of the cost functions for different energy consumptions.

$${\text{Minimize}}\;Z = \sum\limits_{i}^{n} {C_{i} } (x_{i} )$$

(15)

where Z represents the total cost, is the cost function of the i-th energy consumption, represents the amount of the i-th energy used, and n is the number of energy types. where, Z is total cost $\sum\limits_{i}^{n} {C_{i} } (x_{i} )$: cost function for the first i type of energy consumption; $x_{i}$: the amount of energy used by the first i type of energy; n: the number of energy types.

Real-world physical, economic, and technological constraints must be considered in the model, including (1) resource constraints: aggregate constraints on energy supply. (2) Technical constraints: the efficiency range and capacity limit of equipment operation. (3) Environmental constraints: emission limits, policy and regulatory requirements. (4) Time series constraint: the law of energy demand changing with time. These constraints can be formalized as inequalities or equations, as shown in Eqs. 16–18.

Equations 16–18 represent resource constraints, technical constraints and environmental constraints in the energy system optimization model, respectively, which ensure that the model can be feasible in practical scenarios.

$$\sum\limits_{i}^{n} {x_{i} } \le R_{{{\text{total}}}}$$

(16)

$$0 \le x_{i} \le X_{{{\text{max}},i}}$$

(17)

$$E_{{{\text{min}}}} \le \sum\limits_{t}^{T} {x_{i,t} } \le E_{{{\text{max}}}}$$

(18)

where $\sum\limits_{i}^{n} {x_{i} } \,$: total usage of all energy sources; $R_{{{\text{total}}}}$: total resource constraints $x_{i}$: the amount of the first i type of energy usage; $X_{{{\text{max}},i}}$: Maximum usage of the first i energy source; $E_{{{\text{min}}}} \,$: minimum energy requirement; $\sum\limits_{t}^{T} {x_{i,t} }$: total usage of all energy sources in time t; $E_{{{\text{max}}}} \,$: maximum energy demand.

Integration of unstructured data and fuzzy decision support systems

In this section, we discuss in depth how to integrate the results of unstructured data processing into fuzzy decision support systems and how to use this information in model optimization. Specifically, we will instantiate formulas to show how text information processed by ESMM, image information processed by ResNet can be integrated into the model, and how fuzzy rules can be designed and applied to deal with uncertainty in the energy system.

• (1) ESMM processes text information: With ESMM, each keyword or phrase is mapped to a high-dimensional vector that can be considered part of the model input. Suppose that after ESMM processing, the vector corresponding to the keyword "energy policy" is, which reflects the semantic strength of the word in a specific context. These vectors are combined to form a text feature vector that can be used as an input or constraint to the model.

• (2) ResNet processes image information: The feature vectors output by ResNet represent the state of objects in the image, such as whether the device is damaged. These characteristics can be used directly to estimate maintenance costs or equipment availability. Assume that the quantified score for the health status of the device is, where H is a computation-based health assessment function.

Fuzzy rule design is based on fuzzy set theory to deal with uncertainty in prediction. The general form of fuzzy rule is "if (condition part) then (conclusion part)", the condition part is usually composed of several fuzzy sets, and the conclusion part gives a fuzzy description of a variable. Here is a concrete example:

Let historical demand levels be represented by fuzzy sets, where large demand is denoted by, medium demand by, and small demand by. Weather conditions are represented by fuzzy sets W, cold as, moderate as, warm as. Future energy demand is represented by fuzzy sets, which are similar in description to. A specific fuzzy rule can be expressed as: If historical demand is Large and weather is Cold, then future energy demand is likely to be Very Large expressed in fuzzy logic as Eq. 19. Equation 19 represents a typical fuzzy rule used in fuzzy logic, where (D_H refers to historical demand, (W refers to weather conditions, and (D_F refers to future demand. The rule states that if the historical demand is large and the weather is cold, then the future demand will be very large.

$$R_{1} :D_{H} = {\text{Large}} \wedge W = {\text{Cold}} \Rightarrow D_{F} = {\text{Very Large}}$$

(19)

The quantified form of these rules involves membership functions that represent the degree to which an element x belongs to a fuzzy set X. Therefore, the above rule can be embodied in Eq. 20. Equation 20 shows the quantified form of this rule, using membership functions to represent the degree to which an element x belongs to a fuzzy set (X. In this case,(mu_{D_H}(Large) and $\mu_{W} (Cold) \,$ represent the membership degrees of historical demand being “large” and weather being “cold”, respectively, while(mu_{Very Large}(x) represents the membership function of future demand being very large.

$$\mu_{{D_{F} }} (x) = \min \{ \mu_{{D_{H} }} ({\text{Large}}),\mu_{W} ({\text{Cold}})\} \cdot \mu_{{\text{Very Large}}} (x)$$

(20)

where, and represent membership degrees of historical demand being "large" and weather being "cold" respectively, and are membership functions of future demand being very large.

To integrate unstructured data processing results into fuzzy decision support system, text feature vectors and image feature vectors need to be mapped to corresponding fuzzy set membership degrees, and then input into fuzzy inference system together with other structured data (such as time series data).

Taking energy demand forecasting as an example, the inputs to the model can be synthesized as Eq. 21. Equation 21 demonstrates how unstructured data processing results can be integrated into a fuzzy decision support system. Text vectors and image feature vectors need to be mapped to their respective membership degrees before they can be input into the fuzzy inference system alongside other structured data like time series data. In the context of energy demand forecasting, the inputs to the model can be represented as a combination of text vector $v_{text}$, image feature vector $v_{img}$, and structured data $x_{struct}$.

$${\mathbf{Input}} = ({\mathbf{v}}_{{{\mathbf{text}}}} ,{\mathbf{v}}_{{{\mathbf{img}}}} ,{\mathbf{x}}_{{{\mathbf{struct}}}} )$$

(21)

where represents other structured data (e.g. historical energy consumption data). Fuzzy inference process through a series of such rules, combined with input data, fuzzy output is calculated, and then through defuzzification (such as gravity method) into concrete prediction value.

The whole process ensures the effective use of unstructured data and enhances the ability of the model to understand and predict complex situations. Especially in the face of high uncertainty, fuzzy decision support systems can provide more robust and adaptive solutions.

The integration framework is shown in Fig. 4. Figure 4 shows an integrated framework detailing how unstructured data can be processed and fuzzy theory used to make effective decisions. The unstructured data box at the top contains both text information and image processing information. On this basis, fuzzy rules and fuzzy sets are extracted, in which fuzzy rules are expressed by fuzzy mathematics, while fuzzy sets involve incomplete membership degrees of elements. The feature vector fuzzy link then converts these fuzzy rules and sets into feature vectors, which are numerical representations of data properties. Then, the data fusion and fuzzy inference step combines these feature vectors and performs fuzzy inference to draw an overall conclusion. Defuzzification and decision making transform fuzzy conclusions into concrete decision outputs. Finally, the model optimization and feedback loop section optimizes the entire process and introduces a feedback mechanism to ensure that model parameters are adjusted according to actual results to achieve continuous improvement of the decision-making process.

Fig. 4

Integration framework

Case analysis

Case background

This case study focuses on the optimal management of a large urban power grid, which covers multiple complex power consumption environments such as residential areas, commercial areas, and industrial areas, and handles huge tasks such as energy supply and demand matching, equipment maintenance scheduling, and user service response on a daily basis. With the acceleration of urbanization and the increasingly stringent requirements for energy conservation and emission reduction, the grid faces unprecedented challenges: how to improve energy efficiency, reduce costs, and reduce environmental impact while ensuring power supply reliability and stability. In particular, grid managers need to effectively integrate and analyze large amounts of unstructured data (such as user feedback on social media, policy documents, weather forecast information) in order to make more accurate decisions.

Experimental design

In order to verify the effectiveness of the proposed energy optimization framework integrating unstructured data processing and fuzzy decision support system, we designed a series of experiments, covering data collection and preprocessing, model construction and optimization, and performance evaluation.

Unstructured data collection: Collect unstructured data from public sources and internal databases, including user feedback on social media (100,000), weather forecast records for the past five years, and energy-related policy documents issued by the government (50).

We use ESMM to process text data to extract user sentiment trends and hot issues; use ResNet to analyze satellite images to assess the cleanliness and efficiency of photovoltaic panels; and perform data cleaning, missing value filling, outlier detection, etc.

In the model building section, we designed a highly integrated framework that combines text features processed by ESMM, image features processed by ResNet, and historical energy consumption data to form a robust and comprehensive data processing foundation. Fuzzy C-means clustering is used to segment users to ensure that the model can accurately adapt to diverse user needs. The fuzzy decision support system is constructed based on deep historical data and expert knowledge, covering key aspects of energy system optimization, and optimizing fuzzy rules by genetic algorithm, which improves the accuracy and efficiency of decision-making. The integration of optimization algorithms further strengthens the dynamic optimization ability of the model in energy demand forecasting and resource scheduling.

In order to objectively evaluate the advantages of the new model, a comparison group was set up to compare the current traditional energy management system with the integrated model, which lacked the processing of unstructured data and fuzzy decision support.

The implementation plan is divided into five stages, from data collection and preprocessing to the establishment of model infrastructure, to model training, testing, performance evaluation and iterative optimization, until the final comprehensive evaluation of the improvement effect of the new model, ensuring the systematicness and scientificity of the research. The whole experimental design aims to provide strong practical evidence and technical support for the intelligent transformation of the energy industry through a rigorous verification process.

Experimental result

In order to fully demonstrate the effectiveness of the proposed framework in practical energy system optimization, this section reports the experimental results through seven detailed tables. These tables compare the performance of optimization models integrating unstructured data processing and fuzzy decision support systems (hereinafter referred to as “integrated models”) with traditional energy management systems (hereinafter referred to as “traditional systems”) in terms of energy demand forecasting accuracy, equipment maintenance costs, energy allocation efficiency, environmental impact assessment, economic cost savings, user satisfaction improvement, and system response speed.

In this study, we used time window analysis to calculate improvements in the prediction accuracy of the integrated model. The method is to compare the prediction accuracy of the integrated model with that of the traditional system, and determine the improvement percentage by calculating the difference between the two. For example, in a 1 h time window, the prediction accuracy of the integrated model is 92.6% compared to 87.3% for the traditional system, resulting in a percentage improvement of + 5.3%.

Table 1 shows the accuracy comparison of energy demand forecasting, where the integrated model has a prediction accuracy of 5.3%, 5.3% and 6.3% higher than that of the traditional system in the time window of 1 h, 6 h and 24 h, respectively.

Table 1 Comparison of accuracy of energy demand forecasts

When analyzing equipment maintenance costs, we calculated the cost savings by comparing the maintenance costs of the integrated model with those of the traditional system, using the cost of the traditional system as the baseline. Baseline numbers represent maintenance costs for traditional systems, including preventive maintenance, corrective maintenance, and total maintenance costs. The percentage savings are calculated by (traditional system cost-integrated model cost savings)/traditional system cost. Table 2 analyzes the equipment maintenance cost. The integrated model saves 48.0%, 51.9% and 50.0% in preventive maintenance, corrective maintenance and total maintenance costs respectively, showing that by integrating unstructured data and fuzzy decision support, equipment status can be predicted more accurately, unnecessary maintenance can be reduced, and costs can be effectively controlled.

Table 2 Equipment Maintenance Cost Analysis

The so-called “average load balance” refers to the degree of load balance in each distribution area during the system energy distribution process. Calculated by adding the Load Balancer degrees for each zone and dividing by the number of zones. In this study, the average load balance of the integrated model is higher than that of the traditional system, indicating that it is more uniform in energy distribution. Table 3 illustrates the improvement of energy distribution efficiency. The distribution efficiency improvement rates of the integrated model in areas A, B and C are 7.4%, 6.8% and 8.1% respectively, and the average load balance degree is higher than that of the traditional system, which indicates that the optimized model can distribute energy more evenly and improve the overall operation efficiency of the system.

Table 3 Energy distribution efficiency improvements

The EIA process includes monitoring and calculation of emissions from integrated models and conventional systems. Emission reductions are calculated by dividing the difference between integrated model emissions and conventional system emissions by conventional system emissions to obtain a percentage reduction. This process helps to assess the contribution of integrated models to environmental protection. Table 4 Environmental impact assessment shows that the integrated model reduces annual total emissions and emissions per unit energy production by 20.0% and 21.9% respectively, reflecting its significant advantages in promoting environmental protection and reducing carbon footprint, which is of great significance to achieving sustainable development goals.

Table 4 Environmental Impact Assessment

Base cost refers to the annual cost of energy procurement and operation and maintenance of traditional systems. In this study, we calculated the cost savings by comparing the annual costs of the integrated model with traditional systems. The calculation method is: (traditional system annual cost-integrated model annual cost savings)/traditional system annual cost to obtain the percentage savings. The economic cost savings data in Table 5 show that the integrated model saves 20.0% in energy procurement, 20.0% in operation and maintenance, and the total cost savings reach 25.5%, indicating that the model can not only improve the operating efficiency of the energy system, but also significantly reduce the economic burden.

Table 5 Economic cost savings

User Satisfaction Score is the evaluation of users on service quality, response speed and personalized service experience collected through questionnaire survey. Scores are influenced by many factors, such as system stability, ease of operation, etc. In this study, we calculated the difference between the comprehensive model and the traditional system score to obtain the promotion score, which reflects the user satisfaction with the new system. Table 6 User satisfaction survey shows that the integrated model has significantly improved service quality, response speed and personalized service experience, reflecting the high recognition of users for the new system, which is conducive to improving customer loyalty and market competitiveness.

Table 6 User satisfaction survey

The measurement of response time involves specific operations, such as demand response decision, fault detection and alarm, resource allocation instruction issuance, etc. We calculate the reduction in response time by recording the time it takes for the integrated model and legacy systems to complete these operations. This data helps to assess the new model's ability to respond quickly in emergency situations and daily dispatch. Table 7 System response time comparison indicates that the response time of the integrated model in demand response decision, fault detection and alarm, resource allocation instruction issuance and other operations is greatly reduced, with an average reduction of about 48.8%, which proves the rapid response ability of the new model in handling emergency situations and daily scheduling, and enhances the timeliness and reliability of the system.

Table 7 Comparison of system response times

As shown in Table 8, the new method has significantly improved the average values for all indicators compared to the baseline. Indicator A shows a notable increase from 70 to 85, with a very significant p-value of less than 0.01, indicating a high level of confidence in the improvement. The 95% confidence interval for this increase ranges from 78 to 92. Similarly, Indicator B has also seen an improvement from 65 to 78, with a p-value of less than 0.05, suggesting that the improvement is statistically significant, and the confidence interval is between 73 and 83. Indicator C, with the most substantial improvement, has increased from 50 to 62, with a highly significant p-value of less than 0.001, and the confidence interval is from 58 to 66.

Table 8 Statistical Analysis Results

Table 9 illustrates the performance of three different models under various geographical and energy types. In North America, focusing on solar energy, Model A achieved the highest score of 82, leading to an average score of 75 across all models. In Europe, where the energy type is wind, Model A again outperforms with a score of 85, yet the average score is slightly lower at 74. For Asia, with hydropower as the energy source, the models scored an average of 72, with Model A scoring 80. Lastly, in South America utilizing biomass, the average score is 75, with Model A scoring 83. This table shows that Model A tends to have the highest scores across different regions and energy types, while the average scores indicate a consistent level of performance among the models.

Table 9 Model Performance Across Different Scenarios

Conclusion

In contemporary society, energy systems face increasing complexity and uncertainty challenges, including increased fluctuations in energy demand, tighter environmental constraints, and rising operational costs. Traditional management methods are difficult to deal with these complex situations effectively. Therefore, it is very important to study the optimization method of energy system based on fuzzy decision support system (FDSS) and unstructured data processing. Fuzzy decision support systems can handle uncertain information and provide more robust decision-making basis, while in-depth analysis of unstructured data (such as text and image information) can reveal hidden patterns and user behavior, enhancing the timeliness and accuracy of decision-making. The aim of this study is to integrate these two advanced technologies to build an efficient and intelligent energy management system framework to maximize energy efficiency, minimize costs and minimize environmental impact, while improving system flexibility and user satisfaction, providing technical support for the sustainable development of the energy industry. This study successfully demonstrates the great potential of integrating unstructured data processing with fuzzy decision support systems in energy system optimization. Through detailed analysis of seven key performance indicators, we find that compared with the traditional system, the proposed integrated model improves the accuracy of energy demand prediction by about 5.6% on average, saves nearly 50% on equipment maintenance costs, improves energy distribution efficiency by 7.1% on average, reduces carbon emissions by 20% on environmental impact, saves a total of 4.4 million yuan in economic costs, and improves user satisfaction by 1.0 points on average. The average system response time was reduced by about 48.8%. These significant improvements are not only reflected in economic and environmental benefits, but also greatly improve the user service experience and the real-time response capability of the system. By mining unstructured data in depth, especially using ESMM and ResNet technologies, models can better understand specific language and image information in the energy domain, making prediction and decision-making processes more accurate and efficient. The integration of fuzzy decision support systems, especially the optimization of fuzzy rules by genetic algorithms, enhances the ability of models to deal with uncertainty and complex situations, and ensures the flexibility and accuracy of decisions.

However, to provide a more comprehensive overview, it is important to expand the discussion on the limitations of the study. For instance, the effectiveness of the proposed methods may vary depending on the specific characteristics of different energy systems and the geographical context. Additionally, the reliance on certain types of data and the assumptions made about the data quality could impact the generalizability of the results. Future work will focus on addressing these limitations and exploring new areas of research. Specifically, the next steps will involve refining the ESMM to handle a broader range of text data, enhancing the accuracy of the fuzzy decision support system through advanced machine learning techniques, and further developing privacy-preserving mechanisms to ensure compliance with evolving regulatory frameworks. Moreover, there is potential for cross-disciplinary collaborations to integrate insights from social sciences, economics, and environmental studies, enriching the overall framework. These improvements will not only enhance the robustness of the existing models but also pave the way for innovative applications in smart grid technologies and sustainable energy management.