From: Anomaly detection in quasi-periodic energy consumption data series: a comparison of algorithms
Algorithm | Configuration parameters |
---|---|
Basic Statistics | Pearson product-momentum correlation coefficient minimum value (0.2) |
AR | p order in [25, 305] |
ARIMA | p order in [25, 305], d = 1, q = 0 |
LOF | number of neighbours in [1, 300] |
OC SVM | gamma in [0.001, 0.9726], tol in [\(10^{-10}\), 0.1], nu in [0.001, 0.5] |
ISOF | number of trees in [20, 200], max samples in [150, 400] |
GRU | 2 GRU layers both with 32 units. Training: 500 epochs with patience = 30 and batch size = 64. |
LSTM | 2 LSTM layers both with 32 units. Training: 500 epochs with patience = 30 and batch size = 64. |
GRU-MS | 2 GRU layers with 64 and 32 units, and 10 units for the output layer. Training: 500 epochs with patience = 30 a batch size = 64. |
LSTM-MS | 2 LSTM layers with 64 and 32 units, and 10 units for the output layer. Training: 500 epochs with patience = 30 and batch size = 64. |
GRU-AE | 2 GRU layers with 128 and 64 units. Training: 500 epochs with patience = 30 and batch size = 64. |
LSTM-AE | 2 LSTM layers with 128 and 64 units. Training: 500 epochs with patience = 30 and batch size = 64. |