Research On Short-term Power Load Forecasting Method Based On Neural Network Combination Model

With the rapid development of smart grids,the production and life of human society have higher and higher requirements for the stable operation of the system,so short-term load forecasting has gradually become an research topic.Short-term power load forecasting results help to formulate reasonable grid dispatching plans,arrange power equipment maintenance,guide response to emergencies,and ensure stable operation of the power system.However,due to the load forecasting characteristics and the existence of load influencing factors,the short-term power load forecasting results will be biased.Therefore,in order to obtain short-term power load forecasting results,it is necessary to propose new forecasting methods.This thesis proposes two short-term power load forecasting models,CEEMDAN-NARX and Fisher-LSTM,respectively,studies and analyzes the load forecasting characteristics and the role of influencing factors before short-term power load forecasting,and uses actual sample data to make predictions.The validity of this verification model.The main content of this thesis includes:(1)Research and analyze the effect of power load forecasting characteristics on short-term load forecasting and the influence of factors on load.Due to problems such as power equipment failures,the collected sample data will have some problems.Therefore,several solutions are introduced for abnormal and missing data.In order to obtain accurate short-term power load forecasting results,the data normalization method is introduced in detail.Finally,several evaluation indicators of short-term load forecasting are introduced and the basic steps of short-term load forecasting are briefly described.(2)Considering the uncertainty,non-linearity,and various external influencing factors of power load forecasting,the CEEMDAN-NARX short-term load forecasting model is constructed.The model first uses the CEEMDAN algorithm to decompose historical power load data into several IMF components and residual components;then NARX neural network uses its dynamic feedback function and memory to correlate historical loads with future loads,and predicts IMF components and residual components separately;at the same time,experimental analysis shows that the short-term load forecasting model can solve the electrical load Uncertainty and other issues,thereby further improving the prediction accuracy.(3)Constructed a Fisher-LSTM short-term power load forecasting model.The model uses Fisher information to process meteorological data.The Fisher information method can observe changes in the system state.When the system state changes,if the accumulated temperature effect occurs,huge changes in temperature need to be processed in time,so that the temperature data can be reasonably input to In the prediction model;then use the load data and weather data to train the LSTM short-term load forecasting model,and at the same time adjust the model parameters through the verification set;finally use the trained short-term load forecasting model to predict the test set,and the Fisher-LSTM short-term load forecasting model can be verified according to the experimental results Can effectively improve the prediction accuracy.(4)Analyze the forecasting situation of different short-term power load forecasting models through comparative experiments.First,compare and analyze the CEEMDAN-NARX short-term load forecasting model with other types of prediction models.The experimental results verify that the CEEMDAN-NARX short-term load forecasting model has higher prediction accuracy than other models;then compare and analyze the Fisher-LSTM short-term load forecasting model with SVM,RNN,etc.The prediction results of the model show that the Fisher-LSTM short-term load forecasting model has the best prediction effect.Finally,the CEEMDAN-NARX short-term load forecasting model and the Fisher-LSTM short-term load forecasting model are compared and analyzed.Among them,the Fisher-LSTM short-term load forecasting model is used in the sample data with obvious temperature product effect.Concentration is more advantageous,and the prediction effect of the CEEMDAN-NARX short-term load forecasting model and the Fisher-LSTM short-term load forecasting model is at the same level when the system is stable.