Optimization strategy of property energy management based on artificial intelligence

Abstract

This study focuses on the design and optimization of property energy management systems, aiming to improve energy efficiency, reduce waste, and enhance user comfort and satisfaction through intelligent means. The research background is based on the urgency of energy conservation and emission reduction, and the rise of smart property management models on a global scale, especially the increasing demand for energy efficiency monitoring, predictive analysis, automated control, and user engagement. To address the urgent need for energy conservation and emission reduction, particularly in the realm of property management, this study designed and optimized a property energy management system. The core of the research is a systematic energy management framework that encompasses efficient monitoring, intelligent predictive analytics using techniques such as Long Short-Term Memory (LSTM) networks for energy consumption forecasting, automated control, user-friendly interfaces, and system safety. An empirical case study was conducted at a large-scale commercial complex, confirming the effectiveness of the system. Through intelligent transformation, specifically the optimization of air conditioning and lighting systems using advanced technologies like frequency modulation and LED lighting, a total energy saving rate of 25% was achieved. The annual economic savings exceeded 1.25 million yuan, and user satisfaction was significantly improved. During the research process, several limitations and challenges were encountered, including data quality issues and scalability concerns. These limitations were addressed through rigorous data preprocessing and validation, ensuring the robustness of the findings and their applicability to similar environments. The results demonstrate the potential of integrating artificial intelligence and machine learning techniques into property energy management systems, paving the way for more sustainable and efficient buildings. This revised abstract includes more specific details about the technologies used, such as LSTM networks, and mentions the limitations and challenges faced during the research. It also emphasizes the practical application and scalability of the system.

Introduction

In today’s society, with the acceleration of urbanization and the rapid development of information technology, energy demand increases rapidly, energy supply tension and environmental pollution problems become increasingly prominent. Especially in the real estate and property management sectors, energy consumption accounts for a significant proportion, becoming an important factor affecting environmental sustainability and operating costs of enterprises. The traditional property management mode relies on manual operation and experience judgment, which is inefficient in energy allocation, use efficiency monitoring and fault prevention, and is difficult to meet the needs of efficient and green development of modern cities (Si et al. 2022).

In recent years, with the rise of the concept of “smart city construction”, the use of high-tech means to optimize energy management and achieve efficient allocation and utilization of resources has become a global consensus. Therefore, this study focuses on the application of artificial intelligence in property management to optimize energy management, aiming to explore a new way to improve energy efficiency and promote environmental sustainability. This research not only has important theoretical value, providing a new perspective and methodology for interdisciplinary research in the field of property management and energy, but also has far-reaching significance in promoting the intelligent transformation of the property management industry and achieving win-win economic and environmental benefits at the practical level (Imran, Iqbal and Kim 2022). The framework of the integrated energy intelligent management platform is shown in Fig. 1.

Fig. 1

Integrated energy smart management platform

Figure 1 shows a flow chart depicting the modules and their interrelationships related to energy data collection, regional energy consumption performance analysis, household-by-house detection and labeling, and property early warning. Together, these modules form an energy management system that collects and analyzes energy data to optimize energy use and improve efficiency.

At present, the application of artificial intelligence in the field of property energy management is in a rapid development stage. Some advanced property management systems have begun to integrate AI technology to achieve refined management of building energy consumption. For example, using machine learning algorithms to analyze historical energy consumption data can accurately predict future energy demand, thereby optimizing energy procurement plans and avoiding waste of resources. The popularity of Internet of Things technology makes it possible to interconnect devices, and the real-time collection of sensor data provides rich decision-making basis for the system, making intelligent regulation of air conditioning, lighting and other systems a reality, effectively reducing unnecessary energy consumption. In addition, the fault prediction and diagnosis function based on artificial intelligence is gradually improving, which can issue early warning at the early stage of equipment abnormality, greatly reducing the energy waste and maintenance cost caused by sudden failure (Zhang et al. 2023). However, despite the initial success, the application of artificial intelligence in property energy management still faces many challenges. These include, but are not limited to, uneven data quality and lack of uniform standards. The algorithm model is not universal and adaptable enough to be directly replicated in different types of property projects. as well as user privacy protection and technical security issues. The existence of these problems limits the full potential of artificial intelligence technology, and there is an urgent need to find solutions through scientific research (Pant et al. 2024).

Existing property management systems rely heavily on manual operations, which are inefficient and error-prone. Manual document processing, tracking transactions and maintaining updated asset inventories require significant time and human resources. As the business grew, the limitations of these manual processes became apparent, such as frequent data entry errors, delayed information updates, and difficulty responding quickly to market changes. By introducing artificial intelligence technology, system performance can be significantly improved. AI automates repetitive tasks, reduces human error, and ensures high accuracy and timeliness of data. In addition, machine learning algorithms can analyze historical data to predict future trends in property values, providing valuable insights for decision makers. Therefore, in order to overcome the limitations of existing systems and achieve more efficient asset management, adopting AI solutions is imperative.

The core of this study aims to address existing challenges and deepen the potential of AI technologies in the field of property energy management. Specifically, the research strives to achieve three major goals: First, through continuous optimization of algorithm models, significantly improve the accuracy of energy consumption prediction, provide more scientific and reliable data support for energy planning, and ensure the accuracy of energy efficiency analysis. Second, explore the adaptive enhancement strategy of AI system, so that it can better adapt to the unique needs of different property environments, and promote personalized and dynamically adjusted energy management solutions. Moreover, data security and privacy protection are emphasized, and effective protection mechanisms are established within the framework of big data analysis applications (Kim et al. 2018).

In order to achieve these goals, the research content has been carefully designed to cover four key aspects: First, in-depth study of data integration and preprocessing methods, effective integration of IoT device data, meteorological information, historical energy consumption records and other sources of information, and high-standard preprocessing, laying a solid foundation for subsequent in-depth analysis. The second is to devote itself to the innovative construction of intelligent prediction models, using cutting-edge technologies of machine learning and deep learning to develop models to accurately capture energy demand trends, while comprehensively considering external variables such as seasonal changes and special events to enhance the practical value and flexibility of the models.

Literature review

Property energy management

Property energy management is an important means of improving building energy efficiency, reducing operating costs and promoting environmental sustainability. In recent years, with the advancement of technology and the increasing emphasis on energy conservation and emission reduction, the field of property energy management has experienced significant changes and progress, but it is also facing a series of challenges. According to the research of Di Matteo et al. (Matteo et al. 2019), modern property energy management systems generally adopt automatic control technology to realize the basic adjustment of building internal environmental parameters (such as temperature, humidity, illumination, etc.). However, most of these systems rely on preset programs and lack the ability to dynamically adjust to real-time environmental changes and user behavior, resulting in inefficient energy use. In buildings equipped with advanced building automation systems, energy waste is still prevalent due to lack of effective data integration and analysis (Li et al. 2019).

With the development of Internet of Things (IoT) and big data technology, property energy management is gradually shifting to data driven (Peña-Torres et al. 2022). It is mentioned that by deploying a large number of sensors, property managers can obtain real-time energy consumption data within the building, which makes fine management possible (Zhang et al. 2023). However, the quality and integrity of data remains a challenge, and as noted in Qi et al. (Qi et al. 2018), data cleansing, integration, and accurate energy modeling require more attention.

At present, property energy management strategies tend to diversify, not only focusing on hardware and system optimization, but also focusing on user behavior guidance. The research of Gbadega and Sun (Gbadega and Sun 2022) shows that energy-saving incentive mechanism designed by combining behavioral science principles can effectively motivate tenants to change their energy use habits and achieve energy conservation objectives. In addition, energy management systems based on cloud platforms have also begun to gain favor, such as the cloud platform proposed by Cao (Cao 2022), which can realize remote monitoring, data analysis and policy deployment, enhancing the flexibility and efficiency of management (Grzywinski 2022).

In recent years, with the development of Internet of Things (IoT) and big data technology, energy management in property management is gradually shifting to a data-driven approach (Zhan et al. 2022; Oliveira et al. 2019). According to recent research of Siau K and Wang (Siau and Wang 2020), by deploying a large number of sensors, property managers can obtain real-time energy consumption data inside buildings, thus achieving refined management. However, data quality and integrity remain a challenge, and as noted in Zhao et al. (Zhao et al. 2019), data cleansing, integration, and accurate energy modeling require more attention. In addition, although advanced building automation systems can provide a large amount of data, the lack of effective data analysis tools still leads to widespread energy waste (Li et al. 2022). At present, property management strategies tend to diversify, not only focusing on hardware and system optimization, but also focusing on user behavior guidance. The research of Xiang et al. (Xiang et al. 2022) shows that the energy-saving incentive mechanism designed by combining behavioral science principles can effectively motivate tenants to change their energy use habits and achieve the goal of energy conservation and emission reduction. In addition, energy management systems based on cloud platforms are gradually gaining favor, such as the cloud platform proposed by Naik and Satapathy (Naik and Satapathy 2021), which can realize remote monitoring, data analysis and policy deployment, enhancing the flexibility and efficiency of management. Despite the opportunities presented by technological advances, energy management in property management continues to face issues such as low technology integration, user privacy and data security (An et al. 2022).

Despite the opportunities presented by technological advances, the property energy management sector still faces several challenges. First of all, the technology integration is not high, and the information island phenomenon between systems is common, which limits the full use of data (Zhan et al. 2022). The importance of cross-system integration was highlighted as key to achieving efficient energy management. Secondly, user privacy and data security issues are becoming increasingly prominent. How to collect and analyze a large amount of data while ensuring user privacy is an urgent problem to be solved in current research and practice. Finally, the lack of standardization and standardization leads to poor compatibility between different systems, increasing administrative difficulty and cost (Oliveira et al. 2019).

In the field of building energy management, multiple strategies are applied to improve energy efficiency and reduce energy waste. These strategies typically include model predictive control (MPC), demand-side management (DSM), optimization algorithms, and fault detection and diagnosis (FDD). Each strategy has its own unique advantages and can complement each other to achieve more efficient energy management. Model Predictive Control (MPC) is an advanced control strategy that predicts the behavior of a system over time based on mathematical models. In this way, MPC can optimize energy use and ensure that conditions such as temperature and humidity in buildings remain within set target ranges. MPC is particularly useful in response to changing external conditions because it can adjust control actions in real time to meet current optimal conditions Demand-side management (DSM) refers to the use of incentives to guide users to change their electricity usage habits to reduce peak power demand. This can be achieved by intelligently scheduling the run-time of equipment, such as delaying the start-up time of non-critical equipment or adjusting the set-point of an air conditioning system. DSM not only helps to reduce electricity costs, but also reduces stress on the grid and improves overall energy efficiency. Optimization algorithms are used to find optimal energy use scenarios. These algorithms may include genetic algorithms, particle swarm optimization (PSO), or other heuristic methods. Through optimization algorithms, the most economical energy consumption pattern can be automatically calculated, while ensuring that the comfort level inside the building is not affected. Fault detection and diagnosis (FDD) technology is used to monitor system performance and identify potential problems in time. FDD can help locate the source of a fault quickly and fix it before it becomes a major problem. This not only avoids expensive maintenance costs, but also ensures that the system always operates efficiently.

Application of artificial intelligence in energy management

In recent years, artificial intelligence (AI) technologies, especially machine learning (ML) and deep learning (DL), have shown great potential and influence in the field of energy management, opening up new paths to improve energy efficiency, reduce costs and promote sustainable development.

AI technology plays a central role in energy demand forecasting, learning from historical data to accurately predict future energy consumption trends. For example, (Siau and Wang 2020) deep learning models were applied to predict the electricity consumption of commercial buildings, and the results showed that deep neural network (DNN) models significantly improved the prediction accuracy compared to traditional statistical methods, helping managers to make more reasonable energy procurement and scheduling decisions (Zhao et al. 2019). AI has also demonstrated a strong ability to optimize energy allocation and efficiency. An intelligent air conditioning control system based on reinforcement learning (RL) was developed, which can dynamically adjust the air conditioning operation strategy according to indoor and outdoor environmental conditions, effectively reducing energy consumption by more than 20% while ensuring indoor comfort. This shows that AI can achieve refined management of energy use and improve overall energy efficiency. AI also plays a key role in real-time monitoring and fault prevention (Li et al. 2022). Machine learning algorithms are used to detect abnormal data in smart grids, which can identify equipment failures or energy theft behaviors in time and improve system security and stability (Xiang et al. 2022). These applications reduce energy waste and unplanned downtime due to equipment failures. AI technology is also being used to automate control and energy dispatch systems for faster, more accurate responses (Zhang et al. 2023). A microgrid energy management system integrated with deep reinforcement learning is designed, which can automatically adjust energy allocation strategies according to load changes and renewable energy generation, maximize the use of renewable resources and reduce fossil energy dependence [19–20]. AI also plays a role in understanding user behavior patterns, making it possible to develop personalized energy-saving strategies. Using machine learning to analyze residents ‘electricity consumption behavior, a set of intelligent energy-saving recommendation system based on user preferences is designed, which effectively promotes users to participate in energy-saving activities by providing customized energy-saving suggestions (Wang et al. 2022).

With the rapid development of information technology, more and more intelligent technologies are introduced into the building energy management system. The PLUG-N-HARVEST ICT platform is a prime example of how intelligent technologies can be combined with existing building management systems to achieve greater energy efficiency. The platform collects real-time data on building energy use by connecting sensors, controllers and other smart devices. These data are then transmitted to a central server for processing. Through the software interface, the data can be analyzed and the corresponding control measures can be taken. For example, a smart lighting system can automatically turn off lights when sensors detect that an area is empty. The ICT platform enables the various systems within the building to work together. For example, HVAC systems can adjust their operating modes based on real-time weather forecasts, while energy management systems can intelligently adjust their electricity schedules based on price fluctuations in the electricity market. In this way, ICT platforms can minimize energy consumption while maintaining user comfort. In order to demonstrate the effectiveness of the PLUG-N-HARVEST platform, the researchers conducted several field tests. In these tests, the platform was successful in reducing energy consumption while improving the indoor environmental quality of the building. We can see how this integrated solution works in the real world through the demonstration of practical application cases.

While AI has a promising future in energy management, there are challenges, including data quality issues, inadequate model interpretability, and privacy and data security issues. For example, Parvin et al. (Parvin et al. 2021) emphasizes that when using big data for energy management, how to ensure the accuracy and integrity of the data, and how to effectively use the data while protecting user privacy, is the focus of future research.

The design of smart home energy management systems (HEMS) requires consideration of multiple requirements. To ensure the usability and user experience of the system, designers must identify which features are most important and give them due attention during system development. The user interface of HEMS should be intuitive and easy to use, even for users without technical background. The interface should provide a clear display of information that allows users to easily view current energy consumption and adjust settings in simple steps. A high level of automation is an important feature of modern HEMS. For example, the system can automatically adjust the indoor temperature to adapt to different seasonal changes; or automatically turn on or off appliances according to user habits. These automated features not only save energy, but also improve the convenience of life. As one of the core objectives of HEMS, energy-saving effect is an important index to evaluate system performance. The system should have the ability to monitor energy consumption and provide energy-saving recommendations or take energy-saving measures automatically. For example, when a room is detected to be empty for a long time, the system can automatically turn off air conditioning and lighting. When designing HEMS, you also need to consider priorities among different requirements. For example, in some cases, users may be more concerned about the energy efficiency of the system than the degree of automation. By assigning weights to different requirements, we can better balance various functions in the design process and ensure that the system can meet the main user needs. By taking these requirements into account, designers can create practical and efficient HEMS that not only improve occupant comfort, but also effectively reduce energy consumption and contribute to environmental protection.

In recent years, artificial intelligence (AI) technologies, especially machine learning (ML) and deep learning (DL), have shown great potential and influence in energy management, opening up new ways to improve energy efficiency, reduce costs and promote sustainable development. According to the latest research (Wang et al. 2022), AI technology plays a central role in energy demand forecasting, accurately predicting future energy consumption trends by learning from historical data. For example, using a deep learning model to predict the power consumption of commercial buildings, the results show that compared with traditional statistical methods, the deep neural network (DNN) model significantly improves the prediction accuracy and helps managers to make more reasonable energy procurement and scheduling decisions (Parvin et al. 2021). AI has also played a significant role in optimizing energy distribution and efficiency. An intelligent air conditioning control system based on reinforcement learning (RL) has been developed, which can dynamically adjust the air conditioning operation strategy according to indoor and outdoor environmental conditions, effectively reduce energy consumption by more than 20% and ensure indoor comfort (Liu et al. 2023). In addition, AI can also play an important role in real-time monitoring and failure prevention. Machine learning algorithm is used to detect abnormal data in smart grid, which can timely identify equipment failure or energy theft behavior, and improve the security and stability of the system (Khayyam et al. 2018). These applications reduce wasted energy and unplanned downtime due to equipment failures.

Property energy management system architecture design

System requirements analysis

Fig. 2

Requirements hierarchy framework

The design of the property energy management system should be based on comprehensive demand analysis to ensure that the system can effectively solve practical problems and improve energy management efficiency. The specific requirements framework is shown in Fig. 2. Demand analysis should cover the following key dimensions: (1) Efficient energy monitoring: The system needs to have the ability to monitor all kinds of energy (such as electricity, water and gas) consumption in real time, including but not limited to equipment energy consumption, regional energy consumption and time consumption, so as to ensure the accuracy and timeliness of data. (2) Intelligent prediction and analysis: Using historical data and environmental parameters, the system needs to be able to predict energy demand trends, conduct energy efficiency analysis, identify energy consumption anomalies, and provide data support for decision-making. (3) Automatic control and optimization: automatic scheduling and control based on prediction results, optimizing energy allocation and reducing waste, while considering the balance between comfort and economy. (4) User-friendly interface: It provides intuitive and easy-to-use interface for property management personnel and end users to view energy consumption reports, management strategies, early warning information, etc.

System architecture design

Based on the above requirements, the property energy management system architecture is usually divided into four levels: perception layer, network layer, platform layer and service layer, and its specific framework is shown in Fig. 3.

Fig. 3

Model architecture

Figure 3 depicts the modules and their interrelationships related to energy data collection, regional energy consumption performance analysis, household-to-house detection and tagging, and property early warning. Together, these modules form an energy management system that collects and analyzes energy data to optimize energy use and improve efficiency.

Perception layer is the front end of system data acquisition, mainly including various sensors and intelligent devices. These devices collect data on energy consumption, environmental parameters (e.g. temperature, humidity, light), and equipment status. For example, smart meters are used to measure energy consumption, temperature and humidity sensors monitor environmental conditions, and Internet of Things (IoT) sensors can be installed on critical devices to monitor their operating status in real time. The design of this layer needs to consider the rationality of sensor layout, data accuracy and maintenance convenience.

In the building energy management system, the service layer is located between the network layer and the platform layer, which plays a vital role as a bridge. It not only interacts with the underlying data communications infrastructure, but also interfaces with the higher-level data processing and decision-making components. The functions of the service layer include, but are not limited to, data aggregation, cleansing, format conversion, and providing standardized service interfaces to higher-level components. This hierarchical structure helps to improve the expansibility and flexibility of the system.

The network layer is responsible for data transmission and communication, ensuring that information collected by the perception layer can be safely and efficiently transmitted to the platform layer. This layer usually uses Internet of Things technologies, such as Wi-Fi, Bluetooth, LoRa, Zigbee and other wireless communication protocols, as well as wired networks to build stable data transmission channels. Communication stability, bandwidth requirements, network security, and interoperability with other network systems were considered (Liu et al. 2023).

The platform layer is the “brain” of the system, responsible for receiving, processing, storing, and analyzing data. It usually consists of a data warehouse, a big data processing platform, and machine learning and artificial intelligence algorithm engines. Data warehouse is used to store mass historical data. Big data processing platforms leverage stream and batch processing technologies to process real-time and historical data. AI algorithms are responsible for advanced functions such as data mining, pattern recognition, and predictive analysis. The platform layer should be designed for high availability, high performance computing and flexible algorithm integration.

The service layer is a user-oriented interface that provides visual energy consumption reports, early warning notifications, policy recommendations, and system management. The service layer design needs to focus on user experience, provide an interface that is easy to understand and operate, and support mobile device access, so that property managers can easily understand and respond to energy usage anytime and anywhere. In addition, the service layer should also support API interfaces to facilitate integration with other management systems (e.g. property management systems, financial systems) (Khayyam et al. 2018).

Key technology selection and integration scheme

In terms of technology selection, the system adopts a series of advanced and practical technical components to ensure that every link from data acquisition to processing analysis to service deployment has high efficiency and reliability. For data collection, we chose high-precision, low-power smart devices such as smart meters and environmental sensors that accurately capture energy usage and provide a high-quality data base for subsequent analysis. Communication technologies were chosen with the actual needs of the project in mind, utilizing long-range, low-power communication protocols such as LoRa or NB-IoT to serve the broad coverage needs of large properties, while Wi-Fi or Zigbee technologies are suitable for high-speed, flexible connectivity in small or indoor environments. In big data processing, Hadoop and Spark frameworks are used to process large-scale data sets, and Kafka and Flume message queue technologies are used to realize real-time and efficient data transmission. In the part of artificial intelligence algorithm, we use TensorFlow, PyTorch and other deep learning frameworks to build prediction models, and integrate classical machine learning algorithms such as random forest and support vector machine to optimize the models in order to achieve the best prediction effect. In terms of cloud platform services, according to cost-benefit analysis, consider using AWS, Azure, Alibaba Cloud and other public cloud services, or customize private cloud solutions according to specific needs, and make full use of elastic computing, big data processing and DevOps tools of cloud platforms to improve system operation efficiency and agility of development and operation (Liu et al. 2018).

In terms of integration scheme design, we adopt a highly modular system architecture, dividing the whole system into multiple independent modules such as data acquisition, data analysis, user interface, etc. This design is not only conducive to system maintenance and flexible upgrade, but also convenient for module addition, subtraction and adjustment according to different business requirements. To facilitate seamless integration between systems, we have developed clear API interface standards to ensure efficient communication between modules and easy integration with third-party systems. In terms of security, the system deploys strict data encryption transmission mechanism, configures firewall, and integrates intrusion detection system to ensure data security and user privacy in an all-round way. In terms of technical architecture, Docker containerization technology and Kubernetes orchestration tools are introduced. Through microservice architecture design, the scalability and flexibility of the system are greatly enhanced to facilitate rapid response to business changes. In addition, continuous integration/continuous deployment (CI/CD) processes have been established to automate code testing, building, and deployment, accelerating software iteration and ensuring continuous and stable system operation. The comprehensive application of this series of technologies and integrated solutions has laid a solid foundation for building an efficient, safe and easy to maintain property energy management system (Padmanabhan and Radhika 2021; Köbis and Tammer 2023).

Integrating AI technology into a broader system architecture is critical for property energy management. For example, by combining machine learning algorithms with existing building automation systems, real-time monitoring and forecasting of energy demand can be achieved, resulting in more efficient allocation of resources. Specifically, AI can help predict peak energy demand and adjust HVAC system operating patterns accordingly to reduce unnecessary energy consumption. In addition, by using reinforcement learning algorithms, the system can automatically adjust settings based on user preferences and behavior patterns, further improving energy efficiency. The application of these technologies not only helps save costs, but also improves the user experience and reduces carbon emissions. For example, by dynamically adjusting the settings of lighting and HVAC systems, energy waste can be reduced without sacrificing comfort, resulting in cost savings and efficiency gains.

To demonstrate how AI technology can be applied to property energy management, consider a specific case study. Consider a large commercial office building equipped with an advanced energy management system that integrates machine learning algorithms, specifically Long Short Term Memory Networks (LSTM) with attention mechanisms. The system first collects data from various sensors, including temperature, humidity, light intensity and occupancy. The data preprocessing phase includes data cleansing, normalization, and feature engineering to ensure the quality of the data input into the model. Next, the LSTM model with attention mechanism is trained on historical energy consumption data to predict future energy demand. The attention mechanism helps the model focus on the points in time that are most relevant to the prediction, thereby improving prediction accuracy. After training, the model can dynamically adjust energy use plans based on factors such as weather forecasts, weekday/weekend patterns, and specific events. The implementation of such systems has been shown to reduce energy consumption by up to 15 per cent, not only reducing operating costs, but also improving the efficiency and sustainability of energy use.

Application strategy of artificial intelligence in property energy management

Energy consumption prediction model

In this study, we employed a variety of advanced artificial intelligence (AI) models and algorithms to improve the performance of energy management systems. These models and techniques are mainly used to predict future energy consumption, optimize energy allocation and dynamically adjust energy use strategies. The following are detailed descriptions of the main models used in this study and their application scenarios. We used a Long Short Term Memory Network (LSTM), a special type of recurrent neural network (RNN) that specializes in processing time series data. In this study, the LSTM model was used to predict future energy consumption. By training on historical energy consumption data, models can capture trends and periodic patterns in the data to generate accurate load curve predictions, which are critical for planning energy demand ahead of time. As a model-based control strategy, model predictive control (MPC) is used to optimize the control strategy in this study. MPC uses predictive models to determine optimal control sequences that minimize energy consumption while maintaining stable system operation based on predicted energy demand and supply. Reinforcement learning (RL) plays an important role in energy management systems as a machine learning method that enables agents to learn optimal strategies through interaction in unknown environments. RL is able to dynamically adjust the operating state of the equipment to adapt to changing energy prices and user needs, thereby maximizing energy savings. Particle swarm optimization (PSO), as a heuristic search algorithm, is applied to find the optimal energy allocation scheme. PSO ensures the rational allocation of energy resources among subsystems to achieve the goal of energy saving and emission reduction. To ensure the validity of these models and algorithms, we perform a series of data preprocessing steps to improve the quality of the raw data. These steps include data cleaning, missing value handling, outlier detection, data transformation, and feature selection. Through these pre-treatments, we ensure that the data sets used for analysis are of high quality, thus improving the reliability and validity of the results. This high-quality data provides a solid foundation for AI models to more accurately predict and optimize the behavior of energy management systems.

In the field of property energy management, accurate prediction of energy consumption is an important prerequisite for optimizing energy distribution, reducing waste and improving energy efficiency. Traditional time series prediction models such as ARIMA (Autoregressive Integral Moving Average) may have limitations when dealing with series data with long-term dependence. In contrast, long short-term memory network (LSTM), as a special recurrent neural network (RNN), solves the long-term dependence problem effectively through its unique memory unit mechanism and becomes a popular choice in energy consumption prediction. However, the native LSTM model still has room for optimization in the face of high-dimensional data, noise disturbances, and complex nonlinear relationships. Therefore, this section introduces an improved LSTM-based energy consumption prediction model, which further improves prediction accuracy and model robustness by integrating attention mechanism, multi-step prediction strategy and feature selection method (Wang et al. 2023).

Assuming thatis the hidden state of LSTM at time step t and${x_t}$is the corresponding input, the attention weight calculation formula is shown in Eq. (1).

${e_t}$ is obtained by applying the hyperbolic tangent function tanh after linearly combining the hidden state h_t and the current input ${x_t}$ with the weight matrices ${W_h}$ and ${W_x}$ and the offset term ${b_a}$.${\alpha _{\text{t}}}$ is obtained by inputting ${e_t}$ into the softmax function, which converts the output of the tanh function into a probability distribution of time steps, i.e. attention weights.

$$\begin{gathered} {e_t}={\text{tanh}}({W_h}{h_t}+{W_x}{x_t}+{b_a}) \hfill \\ {\alpha _t}=\frac{{\exp ({e_t})}}{{\sum\limits_{{k=1}}^{T} {\exp } ({e_k})}} \hfill \\ {{\tilde {h}}_t}=\sum\limits_{{k=1}}^{T} {{\alpha _k}} {h_k} \hfill \\ \end{gathered} $$

(1)

where, and are weight matrices, bias terms, attention-weighted hidden states for subsequent prediction outputs (Chen et al. 2019).

Traditional single-step forecasting methods only predict energy consumption at the next time point, while multi-step forecasting directly outputs predicted values at multiple time points in the future, which can provide more comprehensive forward-looking information for energy management. The improved LSTM model improves the consistency and practicality of prediction by introducing a multi-step prediction strategy that uses current and historical information to predict future sequences (Khalaf 2020; Kumari and Singh 2021).

Before model input, reasonable feature selection and preprocessing are key to improve prediction performance. Firstly, principal component analysis (PCA) is used to screen out the characteristics closely related to energy consumption, such as historical energy consumption, weather conditions, holiday information, etc. Secondly, normalization or standardization of data is carried out to eliminate dimensional effects and ensure the stability of model training. The principal component analysis formula used is shown in Eq. (2).

$${X^\prime }=U\Sigma {V^T}$$

(2)

where, is the dimensionality reduced data, U is the left singular matrix, is the diagonal matrix, contains eigenvalues, and is the right singular matrix.

The model training adopts mean square error (MSE) or root mean square error (RMSE) as loss function, and updates model parameters through optimization algorithms such as back propagation and gradient descent. To avoid overfitting, regularization terms can be introduced or techniques such as early stop strategy and Dropout can be used. Loss function definition Formula (3).

$${\text{Loss}}=\frac{1}{N}\sum\limits_{{i=1}}^{N} {{{({y_i} - {{\hat {y}}_i})}^2}} $$

(3)

Through the above improvements, the LSTM model integrated with attention mechanism not only enhances the ability of the model to capture key information, but also improves the practicability of the prediction. The feature selection and preprocessing ensure the quality of the model input, so that the whole prediction system shows better performance in complex and dynamically changing property energy consumption scenarios. In the future, continuous optimization of model parameters and structure in combination with feedback from actual application scenarios will further improve prediction accuracy and generalization ability of models, providing more accurate and efficient decision support for property energy management.

In this study, we constructed an LSTM model consisting of two layers of LSTM units, each containing 128 neurons, to capture long-term dependencies in time series data. The attention mechanism is embedded on the output of the LSTM layer with a weight matrix of dimension 32 to enhance the model’s focus on critical time points in the sequence. For model training, we set the initial learning rate to 0.001 and adopt the learning rate decay strategy, which reduces the learning rate by 0.1 times every 10 epochs. During the training process, the data size of each batch is 64 samples, and a total of 50 epochs are trained. To prevent overfitting, we added dropout layers between the LSTM layers with dropout rate set to 0.5. In addition, to optimize model performance, we used Adam Optimizer, where β1 was set to 0.9, β2 was set to 0.999, and epsilon parameter was 1e-08. All of these parameters were adjusted through multiple experiments to achieve optimal model performance. The detailed settings and adjustments of the relevant parameters are documented in the public code.

Equipment energy efficiency optimization control strategy

Equipment efficiency optimization control strategies are designed to automatically adjust equipment operating parameters through intelligent algorithms to minimize energy consumption while maintaining or improving equipment efficiency. In property energy management, it is particularly important to adopt advanced control strategies for high energy consumption equipment such as air conditioners, lighting, elevators, etc.

MPC is a closed-loop control strategy that predicts future states based on a dynamic model of the plant and adjusts current control inputs based on this prediction to optimize performance indicators for a future period. The core idea of MPC is to continuously optimize the control strategy on a rolling basis, and its control law can be expressed as Eq. (4).

Here, u(k) represents the control actions at time step k, and N is the prediction horizon. J(i | k) is the stage cost function, which evaluates the performance of the system at time step i given the initial conditions at time step k. λ is a scalar weighting factor, and ${J_{{\text{terminal}}}}(k+N|k)$ is the terminal cost function, which penalizes deviations from the desired state at the end of the prediction horizon.

$${u^*}(k)=\arg {\hbox{min} _{u(k),...,u(k+N - 1)}}\sum\limits_{{i=k}}^{{k+N - 1}} J (i|k)+\lambda {J_{{\text{terminal}}}}(k+N|k)$$

(4)

where,${u^{(k)}}$is the optimal control input at time k, N is the predicted horizon, J(i| k) is the instantaneous cost function based on the current state versus future time i,$\lambda {J_{{\text{terminal}}}}(k+N|k)$is the endpoint cost, is used to account for the system state at the end of the prediction,$\lambda$and weights.

In equipment efficiency optimization, RL learns optimal strategies by constantly interacting with the environment, without explicit models, and is suitable for complex nonlinear systems. Take Q-learning as an example, and its update rule is Formula (5).

$$Q({s_t},{a_t}) \leftarrow Q({s_t},{a_t})+\alpha [{r_{t+1}}+\gamma {\hbox{max} _{a^{\prime}}}Q({s_{t+1}},a^{\prime}) - Q({s_t},{a_t})$$

(5)

where, represents the expected payoff of taking action in a state, is the learning rate, is the immediate reward, is the discount factor, is the maximum expected payoff of the next state, and reflects the direction of policy optimization.

PSO is a swarm intelligence-based optimization algorithm, which is suitable for finding global optimal solutions to nonlinear and multimodal problems. PSO can be used for parameter tuning in equipment energy efficiency optimization, such as finding the most energy-efficient air conditioning temperature setting. The basic update formula is Formula (6).

$$\begin{gathered} {v_{i,d}}(t+1)={v_{i,d}}(t)+{c_1}{r_1}(pbes{t_i} - {x_{i,d}}(t))+{c_2}{r_2}(gbest - {x_{i,d}}(t)) \hfill \\ {x_{i,d}}(t+1)={x_{i,d}}(t)+{v_{i,d}}(t+1) \hfill \\ \end{gathered} $$

(6)

where, and are the velocity and position ofthe ithparticle in the dth dimension, and are learning factors, and are random numbers, and represent the optimal solution and global optimal solution of the particle itself, respectively.

We construct a fusion model, whichcan integrate the above threemethods incombination with the specific scene of the property, and its framework is shown in Fig. 4. MPC is used for short-term precision control, RL is used for dynamic strategy learning to adapt to environmental changes, and PSO can be used as a parameter optimization tool to help determine key parameters in control strategies. In addition, an online learning mechanism is introduced to enable the model to continuously adjust and optimize itself according to newly collected data, improving the flexibility and adaptability of the control strategy.

Figure 4 shows three different control strategies: MPC, RL, and PSO. These strategies depend on online learning mechanisms for adaptive adjustment. MPC focuses on short-term precision control, RL focuses on dynamic strategy learning, and PSO focuses on parameter optimization. In addition, there is a layer of “self-tuning and optimization” that may integrate all of the above strategies to achieve a more comprehensive automated control system.

Fig. 4

Fusion model framework

Load dispatching and energy allocation algorithm

The core of load dispatching and energy distribution lies in how to meet energy demand and optimize energy use. This process is often translated into mathematical optimization problems, where the objective function can be to minimize total energy consumption, minimize costs, or combine energy efficiency and user satisfaction. Consider a simplified model, assuming that there is a group of devices that consume energy over time, then the total energy consumption objective function can be expressed as Eq. (7).

$$\hbox{min} \sum\limits_{{i \in I}} {\sum\limits_{{t \in T}} {{P_{i,t}}} } $$

(7)

The purpose of Eq. 7 is to minimize the total power consumption of all devices i in set I over all time periods t.

Under a series of constraints, such as equipment operation constraints, energy supply constraints, user demand, etc., optimization algorithm is used to solve the problem.

Genetic algorithm is a heuristic search method that optimizes by simulating natural selection and genetic mechanisms. In the load scheduling problem, each load allocation policy can be regarded as a chromosome, represented by coding (e.g. binary coding), and a fitness function can be designed to evaluate the quality of the policy, such as the reduction of total energy consumption. The basic operations of GA include selection, crossover and mutation.

Firstly, individuals are selected probabilistically according to fitness values to enter the next generation, and individuals with high fitness have a higher probability of being selected, as shown in Eq. (8). where P(i) is the probability that the ith individual is selected, F(i) is its fitness value, and N is the population size. Two individuals are then selected for chromosome exchange to produce new offspring, which can be expressed in Eq. (10). $cpoint$For crossover points, decide where to start swapping genes. Then random changes are made to certain genes of individuals to increase diversity. where is the probability of variation, generating random numbers between [0, 1).

$$P(i)=\frac{{F(i)}}{{\sum\limits_{{j=1}}^{N} F (j)}}$$

(8)

$$Child=Paren{t_1}[:cpoint]+Paren{t_2}[cpoint:]$$

(9)

$$Chil{d_j}=\left\{ {\begin{array}{*{20}{l}} {1,}&{{\text{if }}random()<{p_m}{\text{ and }}Chil{d_j}=0} \\ {0,}&{{\text{if }}random()<{p_m}{\text{ and }}Chil{d_j}=1} \\ {Chil{d_j},}&{{\text{otherwise}}} \end{array}} \right.$$

(10)

In Eq. 8, P (i) is the fitness value of device i, which is calculated as the ratio of its energy consumption F (i) to the total energy consumption F (j) of all devices in the population. The fitness value is used to evaluate the performance of each individual in genetic algorithm.

Equation 9 combines two parents (Parent_1 and Parent_2) to create a Child. Progenies inherit traits from both parents according to the cpoint. Crossover is a key operation in genetic algorithms, which allows information exchange between parents.

As shown in Eq. 10, mutation is performed on the offspring (Child) with a certain probability (p_m). If the random number generated is less than p_m and the bit in the offspring is 0, it is flipped to 1. Similarly, if the bit is 1 and the random number is less than p_m, it is flipped to 0. Otherwise, the bit remains unchanged. Mutation introduces diversity into the population and helps prevent premature convergence.

Anomaly detection and fault warning

GBM is an iteration-based decision tree algorithm that builds multiple weak learners (usually decision trees) and gradually reduces prediction errors in a gradient descent manner. In anomaly detection and fault warning systems, GBM can be used to learn energy consumption patterns under normal conditions, thereby identifying behavior deviating from this pattern as an anomaly signal or fault warning.

The model learning process for GBM can be formalized as a process that minimizes a loss function L, which measures the deviation of the predicted value from the true value y, as shown in Eq. (12). ${F_{m - 1}}$denotes the cumulative prediction of the first m − 1 weak learners, is the prediction of the currently added weak learner on the sample, and L can be mean squared error (MSE), absolute error (LAD), or other loss function.

$$L(y,\hat {y})=\sum\limits_{{i=1}}^{n} L ({y_i},{F_{m - 1}}({x_i})+h({x_i}))$$

(11)

Once the model training is complete, for new observed data, an anomaly score can be calculated by comparing the difference between the model predicted values and the actual values. The higher the anomaly score, the more the data point deviates from the normal pattern, which may be an anomaly or fault signal, as shown in Eq. (12).

$${\text{Anomaly Score}}=|y - \hat {y}|/{\sigma _{\hat {y}}}$$

(12)

where is the standard deviation of the predicted values, used to normalize the scores to fit the volatility of the data.

Equation 11 relate to gradient boosting machines and anomaly detection. Equation (12) defines the loss function for the GBM, where y is the true label, \hat{y} is the predicted label, n is the number of samples, F_{m-1} is the prediction from the previous iteration, and h(x_i) is the contribution of the new weak learner. The GBM iteratively adds weak learners to minimize the loss function until convergence.

Equation (12) computes the anomaly score, which measures the deviation of the actual value (y) from the predicted value (\hat{y}). A high anomaly score indicates that the sample deviates significantly from the expected behavior and may be considered anomalous. The standard deviation (\sigma_y) is used to normalize the difference between y and \hat{y}.

Empirical research and case analysis

Subject selection and data Collection

This study selects a large-scale commercial complex located in a city in eastern China as the research object. The complex includes office area, shopping center, hotel and entertainment facilities, with a total area of about 300,000 square meters and a huge annual power consumption. In order to ensure the comprehensiveness and accuracy of the data, we adopted a diversified data collection strategy, covering the following aspects: (1) Energy consumption records: 24 h of electricity, water and natural gas consumption data for a consecutive year, totaling 365 days. (2) Environmental parameters: temperature, humidity, light intensity and other environmental monitoring data, used to analyze the impact of environmental factors on energy consumption. (3) Equipment operation log: operation time and status record of air conditioning system, lighting system, elevator and escalator. (4) Personnel flow data: Use the personnel access records obtained by the monitoring system and access control system to analyze the relationship between human flow and energy consumption.

Table 1 provides examples of energy consumption records for electricity, water, and natural gas usage on five randomly selected dates and time intervals. The table includes the date, time interval, and corresponding energy consumption values in kWh, m³, and m³, respectively. This data can be used to analyze energy consumption patterns throughout the day, identify peaks and valleys, and inform the development of energy-saving strategies.

Table 1 Examples of energy consumption records

Table 1 provides records of energy consumption for five randomly selected dates and time intervals, relating to electricity, water and natural gas consumption. Each data point provides basic information for energy management optimization and helps to understand the characteristics and changes of energy use over time. For example, during the period from 14:00 to 15:00 on March 15, 2023, the records show that the electricity consumption is 320kWh, the water consumption is 12 m³, and the natural gas consumption is 22 m³. These data can be used to analyze the energy consumption pattern of a specific period, whether it is related to working hours, temperatures or special events, and then guide the formulation of energy saving measures.

Method implementation steps

In implementing energy management optimization, we first perform data preprocessing, which is the cornerstone of analytical accuracy. This phase involves cleaning outliers, such as eliminating recorded errors or extreme values, filling in missing values to maintain data continuity, and unifying time series to ensure that all data are aligned on the same time scale. Then, a baseline model of energy consumption is constructed based on historical data, and trends and seasonal changes of energy consumption are identified through time series analysis, which provides reliable reference for subsequent prediction and strategy formulation. Subsequently, we adopted advanced load dispatching and energy allocation algorithms, as well as the strategies described in Section Load dispatching and energy allocation algorithm, to fine-tune energy use management. In conjunction with the gradient hoist (GBM) technology in Section Anomaly detection and fault warning, anomaly detection is performed to monitor energy consumption patterns in real time, identify and respond to potential equipment failures or abnormal consumption in a timely manner, and ensure stable operation of the energy system.

Case analysis

This project focuses on the intelligent transformation of the air conditioning system and lighting system, introducing frequency conversion technology and LED lighting, and optimizing the control strategy by using algorithms, such as dynamically adjusting the set temperature of the air conditioner and adjusting the lighting brightness according to the flow of people.

Table 2 Monthly total energy consumption data in 2023 (unit: 10,000 kWh)

Table 2 summarizes the monthly energy consumption for the entire year, broken down into electricity, water, natural gas, and total energy consumption. The data is presented in units of 10,000 kWh, m³, and 10,000 kWh, respectively. By comparing the data across different months, researchers can observe seasonal trends and develop energy conservation strategies tailored to specific periods, such as increased energy consumption during the summer months due to higher cooling demands.

Table 2 summarizes the monthly energy consumption for the whole year, broken down into electricity consumption, water consumption, natural gas consumption and total energy consumption. By comparing the data of different months, we can clearly see the trend of energy consumption over time, such as the summer (July and August) total energy consumption reaches the highest, reflecting the summer cooling demand caused by the increase in energy consumption peak. This data is critical for building baseline models of energy consumption, identifying seasonal variations and developing energy conservation strategies.

Table 3 Comparison of energy consumption in five random zones (kWh)

Table 3 compares electricity consumption for five different areas (Office A, Central Business Area, Dining Area, Parking Lot, Outdoor Lighting Area) for specific months from January to December. The data show the trend and difference of energy consumption in each region over time, providing a basis for the implementation of refined energy management. For example, the commercial center area and office area A have high annual power consumption, suggesting that these two areas may be the focus of energy-saving transformation. Although the overall energy consumption of restaurant area, parking lot and outdoor lighting area is low, there are obvious seasonal changes, and there is also room for optimization. Table 3 compares electricity consumption for five distinct areas (Office A, Central Business Area, Dining Area, Parking Lot, and Outdoor Lighting Area) for specific months from January to December. The data shows the trend and differences in energy consumption across regions over time, providing a basis for implementing refined energy management. For instance, the Commercial Center and Office A exhibit high annual power consumption, indicating they could be prioritized for energy-saving transformations. Despite lower overall energy consumption in the Restaurant Area, Parking Lot, and Outdoor Lighting Area, there are noticeable seasonal changes offering opportunities for optimization.

Result analysis and energy saving effect evaluation

Quantitative analysis of energy saving effect

Table 4 presents the energy-saving effects of retrofits on air conditioning, lighting, and other systems, specifying the proportion of energy consumption reduction achieved through intelligent retrofits and strategic adjustments. The energy saving rate is calculated annually, representing the reduction in energy consumption compared to the estimated annual energy consumption after implementing energy-saving measures. This table provides a comprehensive view of the effectiveness of the implemented energy-saving strategies. Table 4 summarizes the energy-saving effects of the retrofits on air conditioning, lighting and other systems, and specifies the proportion of energy consumption reduction achieved through intelligent retrofits and strategic adjustments. This not only verifies the effectiveness of optimization measures, but also provides basic data for subsequent cost savings calculations.

Table 4 Summary of overall energy saving effect

The energy saving rate in Table 4 is calculated yearly. This means that the percentages represent the reduction in energy consumption relative to the estimated annual energy consumption, after implementing the energy-saving measures. The energy saving rate for each system (Air Conditioning, Lighting, and Other) reflects the percentage decrease in energy consumption achieved over the course of the year. The total energy savings and energy saving rate provide an overview of the effectiveness of the implemented energy-saving strategies.

Economic benefit and environmental impact assessment

Table 5 Overview of economic benefits

Table 5 highlights the positive impact of energy conservation and emission reduction measures on the financial health of enterprises by quantifying the direct economic benefits of energy conservation and maintenance cost savings due to improved equipment efficiency. The amounts are presented in thousands of Chinese Yuan (¥). Table 5 highlights the positive impact of energy conservation and emission reduction measures on the financial health of enterprises by quantifying the direct economic benefits of energy conservation and the maintenance cost savings due to improved equipment efficiency.

User satisfaction and acceptance survey

Through questionnaire survey (1000 valid questionnaires were recovered) and on-site interview, the user’s feeling and acceptance of the reconstructed environment were evaluated. The results show that 92% of respondents consider the indoor environment more comfortable, as shown in Table 6. 88% of shoppers said the lighting was appropriate and the visual experience was better. 76% of respondents expressed appreciation and support for the energy saving and emission reduction measures adopted by shopping malls.

Table 6 Results of user satisfaction survey

Table 6 evaluates the effect of the transformation from the perspective of end-users. The high percentage of satisfaction reflects that the transformation measures are not only technically successful, but also recognized by users, which is of great significance for the continuous promotion and implementation of energy conservation and emission reduction measures. User support for comfort, lighting, and energy-saving and emission-reduction initiatives are important soft metrics for measuring project success.

Table 6 presents the results of a user satisfaction survey, showing the percentage of respondents who rated their satisfaction with the reconstructed environment, lighting effect, and awareness of energy conservation and emission reduction measures. The survey results demonstrate a high level of satisfaction among users, with 92% considering the indoor environment more comfortable and 88% finding the lighting appropriate and visually pleasing. Additionally, 76% of respondents expressed appreciation and support for the energy-saving and emission reduction measures adopted by the shopping mall.

After a series of intelligent transformation and strategy adjustment, the project achieved remarkable energy saving effect in air conditioning, lighting and other systems, with a total energy saving rate of 25% and energy consumption saving of 75,000 kWh. The economic benefit analysis shows that the direct electricity cost savings of 1.2 million yuan, plus the maintenance cost savings of 50,000 yuan due to the improvement of equipment efficiency, bring a total economic benefit of 1.25 million yuan, demonstrating the positive contribution of energy conservation and emission reduction measures to the economic benefits of enterprises. The user satisfaction survey shows that 92% of users believe that environmental comfort has improved, 88% of customers are satisfied with lighting improvements, and 76% of respondents support energy conservation and emission reduction initiatives, indicating that technological transformation has not only improved energy efficiency, but also greatly enhanced user experience, laying a solid foundation for social acceptance and sustainable development of the project. To sum up, the project has achieved remarkable results in technical practice, economic benefits and social benefits.

Table 7 Summary of overall energy saving effect

Table 7 summarizes the estimated energy consumption, actual energy consumption, total energy savings, and energy savings rates for the different systems over a year. By comparing the estimated energy consumption with the actual energy consumption, the effect of energy saving measures implemented in each system can be seen. Energy savings were 20% for air conditioning systems, 40% for lighting systems, and 10% for other systems, for an overall energy savings of 25%.

Table 8 Overview of economic benefits

Table 8 provides an overview of the economic benefits of energy conservation projects. The total economic benefit can be calculated from the savings in electricity and maintenance costs. As shown in the table, through energy saving measures, electricity saved 1.2 million yuan, maintenance cost saved 50,000 yuan, so the total economic benefit of 1.25 million yuan.

Table 9 Results of user satisfaction survey

Table 9 shows the results of the user satisfaction survey. The survey items include environmental comfort, lighting effects and perception of energy conservation and emission reduction measures. By collecting user satisfaction scores for each project, it can be seen that most users are satisfied or very satisfied with the environmental comfort and lighting effects, while the perception of energy saving and emission reduction measures is relatively scattered.

Table 10 Statistical Analysis of Energy Consumption Variability

Table 10 provides a statistical analysis of the variability of monthly energy consumption. The monthly average energy consumption, standard deviation, and 95% confidence interval are provided to help understand the trend and range of energy consumption over time. For example, the average energy consumption in July and August is high, and the standard deviation is also large, indicating that the energy consumption fluctuates greatly in these two months.

Table 11 Regression Analysis of Energy Consumption

Table 11 shows the results of the regression analysis of the energy consumption. By analyzing the impact of different predictor variables (e.g., temperature, humidity, occupancy level, time period) on energy consumption, coefficients, p-values, and R-squared values can be derived for each variable. Negative coefficients for temperature indicate a decrease in energy consumption as the temperature increases, while positive coefficients for humidity, occupancy level, and time period indicate an increase in energy consumption as these variables increase. The R-squared value indicates how well the model explains the energy consumption change, for example, occupancy level explains the energy consumption change the most, reaching 60%.

Conclusion

With the global climate change and resource shortage, energy conservation and emission reduction has become one of the key strategies to promote sustainable development. Energy consumption accounts for a significant proportion, particularly in the real estate and property management sectors, especially large commercial complexes, where energy management efficiency directly affects operating costs and environmental footprint. In view of this, this study focuses on the design and optimization of property energy management systems, aiming to improve energy efficiency and reduce waste through intelligent means, while enhancing user comfort and satisfaction. The research background is based on the urgency of energy conservation and emission reduction, and the rise of smart property management models on a global scale, especially the increasing demand for energy efficiency monitoring, predictive analysis, automated control and user engagement. With the global climate change and resource shortage, energy conservation and emission reduction has become one of the key strategies to promote sustainable development. Energy consumption accounts for a significant proportion, particularly in the real estate and property management sectors, especially large commercial complexes, where energy management efficiency directly affects operating costs and environmental footprint. In view of this, this study focuses on the design and optimization of property energy management systems, aiming to improve energy efficiency and reduce waste through intelligent means, while enhancing user comfort and satisfaction. The research background is based on the urgency of energy conservation and emission reduction, and the rise of smart property management models on a global scale, especially the increasing demand for energy efficiency monitoring, predictive analysis, automated control and user engagement. Through empirical research and case analysis, this study successfully implemented the intelligent transformation project, especially in the optimization of air conditioning and lighting system, significantly reduced energy consumption, the total energy saving rate reached 25%. The economic benefit analysis shows that the annual electricity cost is saved by 1.2 million yuan, the maintenance cost is saved by 50,000 yuan, and the total economic benefit reaches 1.25 million yuan, which strongly proves the economic feasibility of energy saving and emission reduction measures. User satisfaction surveys show that environmental comfort, lighting effects and recognition of energy conservation and emission reduction initiatives have increased significantly after the renovation, indicating that technical measures have won wide praise from end users while improving energy efficiency.

This study successfully achieves the expected energy-saving goal, and verifies the effectiveness of the system through user satisfaction surveys. Future research will focus on exploring advanced artificial intelligence technologies to further improve the prediction accuracy and adaptability of the system. In addition, we plan to extend the system to more types of building attributes, such as different climatic zones and building uses, to enhance its universality and scope of application. These explorations will bring innovative solutions to the field of building energy efficiency, ensuring the continued relevance of research and its contribution to sustainable development.