A Load Classification Method Based on SA-LSTM Network Considering Category Imbalance Processing

In order to master the regularity of user power consumption and enhance the service quality for electricity users, the refined classification of user load is of great significance for power company. Therefore, in load classification research, a load classification method based on Self Attention Long Short Term Memory (SA-LSTM) neural network considering category imbalance processing is proposed to address the imbalance of load categories and the performance defects of the classification model. First, for addressing the issue of load category imbalance, the generation adversarial network (GAN) based on Wasserstein distance is constructed to oversample the minority category samples; second, for enhancing the classification accuracy, the LSTM load classification model improved by Self Attention mechanism is built to identify the load type of user; third, the typical power consumption pattern is obtained by calculating the centroid of the classification results. The effectiveness and accuracy of the proposed method are evaluated in an experiment by the classification of actual load data measured in a certain place.