Design of Short-Term Power Load Forecasting Model Based on Deep Neural Network

Abstract

In power system operation and planning, the accuracy of short-term power load forecasting is very important. Because of its powerful data processing and modeling ability, deep neural network has become an effective tool to accurately predict short-term power load. In this study, a short-term power load prediction model based on deep neural network is designed, which adopts deep long short-term memory and threshold period unit model, and combines Boosting algorithm for model fusion. The results show that the average absolute percentage error of the model fused by Boosting algorithm is 0.07%, which is 1.02% lower than the average weight method and 0.59% lower than the reciprocal error method. Boosting fusion model can effectively reduce the overall prediction error and maintain high stability of prediction error at peak, plateau and time sampling points, so as to achieve good prediction effect. Specifically, the MAPE of the model fused using Boosting algorithm is 0.07% (95% confidence), which is 1.14% higher than the average weight method and 0.79% higher than the reciprocal error method. The design of short-term power load forecasting model based on deep neural network can provide more accurate prediction for power system operation and planning, and help to improve the operation efficiency and reliability of power system. At the same time, the design and application of this model also provide a new idea and method for the application of deep learning in power system. The introduction of Boosting algorithm further improves the prediction accuracy and stability of the model, which is a major innovation in model design.