Saturated Load Forecasting Based on Attention-LSTM Correction Model

Accurately predicting saturated load enables utilities to estimate the peak load demand after urban construction development, thereby optimizing the construction sequence and resource allocation during the transformation of distribution network infrastructure and it plays a crucial role in the effective planning and operation of the distribution network. In this paper, the traditional saturated load prediction method based on load density is improved and a saturated load prediction method based on attention-long short-term memory (LSTM) correction is proposed. First, the feature vectors describing the factors affecting saturated load are established, and an LSTM model is constructed to mine the nonlinear mapping relationship between saturated load and feature vectors. Then, an attention mechanism is embedded into the LSTM model, which enables the model to extract the main influencing factors of saturated load, thus improving the accuracy of predicting saturated load and the generalization ability of the model. Finally, the effectiveness of the proposed method is verified by using load data from four actual substations in the cities of China to simulate load forecasting. The simulation results show that compared with the traditional method based on logistic curve fitting, the proposed method has higher accuracy in predicting the saturation load value and saturation year of high-load development level in the forecasted area.