CNN-LSTM short-term load forecasting based on the K-Medoids clustering and grid method to extract load curve features

Efficient and accurate short-term load forecasting is an important guarantee of safe, stable and economic operation of a power system. Given the large prediction errors of peak and valley loads, this paper proposes a convolutional neural network and long short term memory network (CNN-LSTM) hybrid prediction model based on a grid method to extract load curve features. First, the K-Medoids algorithm is used to cluster the daily load curves, and each cluster center is taken as the typical daily load curve. The typical daily load curve is divided into several sections by the grid method and numbered successively to extract the features of the load curve. Then, the characteristics of each typical daily load curve and the historical data of corresponding load types are reconstructed into a new feature set and input into the CNN-LSTM hybrid neural network. The features among the data are mined by the CNN to form a new feature vector, which is then input into LSTM for prediction. Finally, the 2012—2013 power load data set in New England is taken as an example for simulation verification. The results show that the load prediction accuracy of the proposed method is improved for different dates, and the prediction accuracy of peak and valley loads is effectively improved while the average forecast accuracy of the daily load is also improved. © 2023 Power System Protection and Control Press. All rights reserved.