Ultra-short-term Power Anomaly Detection of Wind, Solar, and Energy Storage Power Stations Based on Multi-dimensional AI Data Fusion

The traditional ultra-short-term electricity statistical model relies only on limited statistical features for anomaly detection, which limits the accuracy of anomaly detection. To this end, a new approach for ultra-short-term power anomaly detection in wind, solar, and energy storage power plants based on multidimensional artificial intelligence data fusion is proposed, which extensively collects relevant multidimensional data using multiple types of sensors. After standardized processing to eliminate dimensional differences, the deep learning network is used to assign weights according to the importance of each dimensional data to complete the fusion. Based on the fused data, a multi-level neural network is constructed to extract time dimension features at multiple scales. Combined feature construction is introduced to mine the nonlinear relationship of different power dimensions, and the two types of features are normalized and spliced to extract ultra-short-term power features. Using a hybrid model of LSTM and autoencoder to pre-process and fuse features, partition the dataset, and achieve ultra-short-term power anomaly detection in wind power plants, experiments have shown that it can improve detection accuracy and reduce errors.