Power load prediction of smart grid based on deep learning

作     者：Siyu Xiang Cao Zhen Jian Peng Linghao Zhang Zhengguo Pu 

作者机构：State Grid Sichuan Electric Power Research Institute Chengdu Sichuan 610095 China School of Computer Science and Engineering University of Electronic Science and Technology of China Chengdu 611700 China College of Computer Science Sichuan University Chengdu 610065 China Anhui Jiyuan Software Co. Ltd. Hefei 230088 China 

出 版 物：《Procedia Computer Science》 (计算机科学会议集)

年 卷 期：2023年第228卷

页      面：762-773页

学科分类：08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

主　　题：Real-time electricity price CNN Input features Power load forecasting 

摘      要：Real electricity costs, weather, and historical load data are added as new reference data when analyzing the characteristics of the load change policy to improve the forecasting model. A load forecasting model for smart grid based on convolutional neural network(CNN) is proposed. The model first analyzes the characteristics of the variable model and determines the forecast model inputs using two methods of correlation analysis to understand the impact of actual energy costs, weather and other factors on the changes. Second, since Feedforward neural network is deficient in processing the correlation information between loads, the prediction model developed by the author was studied as CNN. After completing the training, input two sets of real datasets for final data analysis and comparison, and the results show that, LSTM, BP, SVR[UNK]MAPE[UNK]RMSE respectively 0.49%, 0.57%, 0.83%, 1.44% and 106.08 MW, 113.82MW, 169.77 MW, 285.61 MW. It can be seen that the prediction model proposed by the author has the highest accuracy and the best fit with the trend of actual load changes. It is proved that the CNN model has certain advantages when dealing with load forecasting problems related to time series.