Demand Side Load Forecasting Method Based on STL-RFE and DeepAR Model

In order to ensure the stable operation of the power grid, accurate power load forecasting must be carried out. The power load data has obvious time series characteristics and is affected by a variety of nonlinear factors, resulting in a high degree of randomness. In order to improve the prediction accuracy of power load, a demand side load forecasting method based on STL (Seasonal and Trend decomposition using Loess) - RFE (Recursive Feature Elimination) and DeepAR (Deep Autoregressive) model is proposed in this paper. This paper first normalizes the historical data, then processes the abnormal data through STL decomposition to expand the data set, and uses RFE to screen the feature data set to form the best feature combination. Finally, the DeepAR model is used to predict the load power and evaluate the results. Based on the typical demand side load sample data of a park, this paper predicts the adjustable load in this area. Compared with the single DeepAR model, the STL-RFE and DeepAR models can effectively improve the accuracy of load forecasting. Compared with other models, the RMSE of the prediction results is reduced by at least 0.0158, the MAE is reduced by at least 0.0147, and the MAPE is reduced by at least 1.4473 %.