Research On Short-term Load Forecasting Method Based On Deep Learning Neural Networks

Short-term power load forecasting is not only a key part of power system dispatch,but also one of the important work of electricity marketing,power grid planning and other management departments;deep learning is an artificial intelligence method that has been widely valued in recent years;this paper selects deep learning recurrent neural networks The three most typical models are designed to study the performance of their models for short-term power load forecasting,including power load data preprocessing,prediction model feature selection,model parameter determination,and comparison of prediction characteristics from shallow network to deep network,etc.To explore and prove the applicability of deep learning neural networks in short-term power load forecasting,and ultimately improve the accuracy of short-term load forecasting.The work done in this paper and the innovations in the research results obtained are as follows:First,it analyzes the research status of power system short-term load forecasting at home and abroad,analyzes the current common methods,characteristics,advantages and existing problems of power system load forecasting,and introduces the research value of power big data.The classification and characteristics of power load are summarized,including the main factors affecting it and the performance evaluation indicators of the prediction model,data preprocessing methods,and basic process steps of prediction.Secondly,the basic principles of deep learning recurrent neural networks are introduced;through the comparison of deep learning networks and shallow networks,the advantages of deep learning are demonstrated;the special status of activation functions in neural networks is explained,and the perceptron model and Multi-layer perceptron model;summarize and analyze the structure and model characteristics of three typical models,namely RNN(recursive or recurrent neural network)model,LSTM(long short-term memory network)model,LSTM(long short-term memory network based on Attention mechanism))Neural network model;on this basis,clarified the adaptability of deep learning recurrent neural network to power system load forecasting problem.Then,based on the measured power load data,the three typical models of power load prediction method model construction and comparative testing and analysis are completed.By comparing the prediction results,the LSTM neural network short-term load forecasting model experiment based on the Attention mechanism has an ideal optimization effect.Finally,the full text is summarized,and further follow-up research work is prospected.The research results of this paper have certain scientific significance and practical value for improving the accuracy and reliability of short-term power load forecasting.