融合多元影响因素的配电台区 BiLSTM 负荷预测方法

Load forecasting for transformer districts is the key to meeting the power supply-demand balance and hence plays a significant role in guiding the early warning，emergency response，and economic operation of power systems. However，satisfactory short- and medium-term forecasted results for transformer districts are unavailable using conventional methods since the daily load forecasting is affected by various coupling factors. To improve the generalization capability of load forecasting for transformer districts，a bidirectional long short-term memory (BiLSTM)load forecasting model is proposed，which introduces principal component analysis(PCA)and electricity consumption behavior analysis. First，to achieve the dimensionality reduction and information correction of input variables containing redundant，missing，and error components，the PCA method is used to analyze the principal components of preselected external factors that affect the power load. Second，to eliminate the effects of behavior differences，the electricity consumption behavior of different categories of consumers is extracted and grouped using the fuzzy C-means(FCM) method based on genetic algorithm(GA)，according to historical data. Third，to improve the generalization capability，BiLSTM forecasting models，which are optimized by the stochastic weight averaging (SWA)algorithm，are established for each category of consumers to forecast the daily load of the entire year. Finally，the load forecasted results for transformer districts are obtained using a linear superposition of load forecasted data from all categories of consumers. Based on the historical data of a low-voltage transformer district in China in the last four years，the forecasted annual electricity consumption of the PCA-BiLSTM model is closer to the real data and better reflects the electricity demand of consumers，compared with those obtained from other conventional LSTM models. Moreover，the classified forecasting method based on PCA and electricity consumption behavior analysis can effectively improve the accuracy of the forecasted results.