Electricity Consumption Prediction Based On Autoregressive Kalman Filtering

Abstract

Electricity consumption prediction is crucial for energy suppliers and industrial companies as it aids in optimizing energy planning and reducing energy consumption losses. Existing methods primarily focus on the time series relationships of individual nodes or components, overlooking the spatial structure of node groups, which leads to insufficient prediction accuracy. To overcome this limitation, we propose an autoregressive Kalman filtering (AKF) method for electricity consumption prediction. Our primary contribution lies in the innovative design of the Kalman filter observation equation in AKF, which finely adjusts the initial predictions of the autoregressive (AR) model based on the hierarchical structure of equipment. This approach comprehensively considers the interrelationships among equipment levels, significantly enhancing prediction accuracy. Specifically, we first utilize the autoregressive model to capture the autocorrelation of the sequence, forming the basis for constructing the state equation in the Kalman filter. In designing the observation equation, we simplify the model and reduce the complexity of parameter estimation by setting the sum of predicted electricity consumption values of sub-node components as the observed value for the total node components. To validate the effectiveness of our proposed method, experiments were conducted using real electricity consumption data from Foshan Ceramic Factory. The results demonstrate significant improvements in prediction accuracy compared to baseline methods such as BP, LSTM, GA-BP, PSO-SVM, and AR.