Hybrid Electricity Consumption Prediction Based on Spatiotemporal Correlation

Abstract

Background: Electricity consumption forecast is an important basis for the power system to achieve regional electricity balance and electricity spot market transactions. Objective: In view of the fact that many electricity consumption prediction models do not make good use of the correlation of data in time dimension and space dimension, this paper proposes a day-ahead forecasting model based on spatiotemporal correction, which further improves the forecasting accuracy of electricity demand. Methods: Firstly, the long short-term memory (LSTM) model is used to construct the forecasting model. Secondly, from the perspectives of time correlation and space correlation, meanwhile considering calendar factors and meteorological factors, the K-Nearest Neighbors (KNN) model is taken to construct correction models, which can correct the forecasting results of LSTM. Results: According to the analysis of power consumption data of 9 areas in New England, the mean absolute percentage error (MAPE), mean absolute error (MAE), and root mean square error (RMSE) of time dimension correction model are reduced by 0.35%, 5.87% and 5.06%, and the 3 evaluation metrics in space dimension are decreased by 0.52%, 6.82% and 7.06% on average. Conclusion: The results prove that the models proposed in this paper are effective.