Research On Electricity Price Forecast Based On LSTM And XGBoost Combined Model

From the analysis of the current development of the electricity market,the electricity price is closely related to the interests of market participants,so the electricity price forecasting has become an urgent problem to be solved.Under the change of electricity price,the supply side power generation faces new development problems,and affects the use of the user side.As the basis of market leverage adjustment,electricity price management can play key functions and effects through accurate and reasonable electricity price forecast.At present,the utilization rate of electricity price forecasting method is not high in the periodic change law,and the step length forecasting is relatively short,which leads to the deviation of electricity price forecasting itself.In order to accurately predict the short-term electricity price in the electricity market,this paper has done the following work:First,in order to verify the effectiveness of the LSTM-XGBoost model,this article first uses French data to conduct short-term electricity price forecasting research.Using data from the French electricity market from January 1,2019 to December 31,2020,collect and integrate single points every hour,that is,24 points can be integrated every day,and the corresponding dimension is recorded as €,create an independent training set training model,and carry out January2021.The prediction model of BP,LSTM,XGBoost and LSTM-XGBoost was compared with the actual value on the 1st,and the prediction model LSTM-XGBoost,which was combined by the reciprocal error method,had the highest prediction accuracy.The error rate of LSTM-XGBoost with an RMSE of0.74 is much lower than that of BP,LSTM’s 3.80,1.25,and XGBoost’s 0.88,so that the LSTM-XGBoost combined model has the highest prediction accuracy.Secondly,the electricity price data of the Austrian power market from November 2020 to January 2021 is selected for the LSTM and XGBoost combined forecasting model to complete the test.The results show that: based on LSTM and XGBoost for model combined testing,the MAPE standard is1.61%,which is obviously better than the single The prediction model LSTM is4% lower than the XGBoost model 1.83%.Finally,by collecting and integrating data changes in the USA PJM power market,based on the principle of deep learning,also select electricity price data from November 2020 to January 2021,a targeted combination model based on GRU and XGBoost is proposed to analyze short-term power price changes and predict the corresponding results.Through the application of the GRU-XGBoost model,the experimental quality of a single model is compared and analyzed.The results show that not all single model combinations can promote the improvement of prediction accuracy.Therefore,a reasonable choice to participate in the combination of a single model can effectively control research errors.The average error MAE value corresponding to the GRU-XGBoost model is 1.25 higher than that of GRU and XGBoost 1.07 and 0.56,while the average MAE value of the LSTM and XGBoost combined prediction model is0.41,which is significantly lower than the 0.89 and 0.56 of the single prediction model LSTM and XGBoost.It is proved that the combined model is significantly better than the GRU-XGBoost combined model.The combination model proposed in this paper effectively improves the accuracy of short-term electricity price forecasting,and has a strong universality.It can be applied to the short-term electricity price forecasting of the electricity market and provides a strong decision-making basis for market participants and regulators.