Short-term power load combination prediction by using KNN and BiLSTM

Abstract

The traditional short-term load forecasting is not accurate in the extraction of meteorological factors, which leads to make prediction accuracy low. To fully explore the influence of meteorological factors on electrical load and effectively utilize the advantages of deep learning technology in nonlinear fitting, a short-term electrical load forecasting method based on two-way long short-term memory network taking meteorological factors into account is proposed in this paper. After the outliers of the original data were removed and standardized, the key factors affecting the power load were fully excavated by using the K-Neighbor-Nearest (KNN) algorithm, and the data sequence was reconstructed. After setting the hyperparameter of the neural network, the Bidirectional long short-term memory (BiLSTM ) network model is built to realize the short-term high-precision prediction of power load. The simulation results show that, compared with BiLSTM and LSTM, the combined method of KNN and BiLSTM mentioned in this paper has higher prediction accuracy.