LAUSR

Accurate short-term load forecasting (STLF) is an important basis for daily dispatching of the power grid, but the non-stationary characteristics of the load series add to the challenge of this task. Many researchers have been working to improve the accuracy and speed of forecasting models, but stability is equally important. This paper develops a forecasting method based on error correction using dynamic mode decomposition (DMD) for STLF, including data selection, error forecasting, and error correction. In the data selection stage, three types of data are selected as input data of the model, including previous day data, same day data in previous week and similar day data obtained by grey relational analysis (GRA). In the error forecasting stage, the data driving characteristics of the DMD algorithm is used to capture the potential spatiotemporal dynamics of error series, thereby realizing the error forecasting. In the error correction stage, on the basis of combining the forecasting results of load and error, an extreme value constraint method (EVCM) is developed to further correct the load demand series. Based on the load data of different regions, this paper selects different performance indicators, such as MAPE, MAE, RMSE, Variance and direction accuracy (DA), to prove that the proposed method has the advantages of accuracy and stability.