Optimization of abnormal sequence labeling for electricity consumption data based on mask-enhanced self training algorithm

The power consumption data on the load side is multi windowed and contains abnormal sequences, resulting in mask operation time delay and low hit rate. A power consumption data anomaly sequence annotation optimization method based on mask enhanced self training algorithm is proposed for this purpose. Based on the Kalman filter algorithm, pulse filtering is applied to power data, and a mask autoencoder model based on power data vector projection is constructed to perform a thermal mask autoencoder on the filtered data sequence; Enhance the mask autoencoder model using attenuation mask algorithm, and optimize the labeling of abnormal sequences in electricity consumption data by combining window segmentation and SVM algorithm. The experimental results show that the masked entity mapping effect of this method is good, and it can capture the structural information between sequences. The masking operation has a small time delay and can accurately extract the fluctuation mutation values of power data. The hit rate of abnormal markers is very high.