Research on Data and Error Driven Method in Forecasting of Industrial Power Consumption

An incremental approach to identify and forecast the main influencing factors of industry electricity consumption based on data-driven and error correction is proposed. Firstly, at the data-driven level, the maximum information coefficient and maximum correlation-minimum redundancy are applied to initially identify the main influencing factors from the perspective of correlation and redundancy. Second, at the error-driven level, a random forest approach is used for error fine screening. Once again, the structure of the industry electricity consumption forecasting model considering the progressive main influencing factors identification and classification modeling is constructed. The self-organizing mapping method is applied to cluster the multivariate data sample set, so as to realize the classification of training samples, and the Back Propagation (BP) model is constructed to classify and forecast the corresponding screened main influencing factors and historical electricity consumption data, so as to improve the computational performance of the industry's electricity forecasting algorithm and the prediction accuracy. Finally, the typical industrial power consumption in a region in the north of China is taken as the forecasting object to illustrate the advantages of the proposed method.