A Method of Anomaly Data Identification and Correction for Industrial and Commercial Consumers’ Electricity Loads

Industrial and commercial users are numerous and widely distributed, with electricity consumption behaviors exhibiting time series characteristics. Complete and accurate electricity load data contribute to optimizing adjustable load resource scheduling, as well as providing data support for load forecasting and demand response efforts. To enhance data restoration precision, a novel anomaly data identification and correction model is proposed, combining the K-means++ algorithm, bidirectional Long-Short-Term-Memory (Bi-LSTM), and Dropout technique. Initially, load data from industrial and commercial users are grouped into similar clusters using the K-means++ algorithm and the elbow rule, followed by preliminary screening based on clustered typical curves and predefined feasible domain filtering rules. Subsequently, utilizing the Bi-LSTM method and Dropout technique, multiple sets of prediction samples are generated. Finally, by employing Bayesian inference to generate posterior distributions, typical user load feasibility domains are updated, completing the further identification and correction of abnormal values, closer to the true value. In the case analysis section, the proposed model is compared with other methods to validate its suitability for identifying and rectifying abnormal load data among industrial and commercial users.