Short Term Load Forecasting Based On Combined Model

As the energy crisis is becoming more and more serious,the power system needs to make effective improvement in dispatching,and the premise of optimal dispatching is to accurately predict the short-term power load.With the development of society,the influence factors of modern power load are gradually increasing.People need to analyze the power load more fully and clarify the relationship between its internal and external influence factors in order to build a better prediction model.The in-depth development of smart grid provides a guarantee for solving the above problems.Because the traditional single statistical forecasting model is relatively simple and the forecasting accuracy is low,the main hot topics in the field of power load forecasting are deep neural network and combined forecasting model.In order to improve the accuracy of short-term power load forecasting,based on the research theories and achievements at home and abroad,this paper studies the existing problems of short-term power load forecasting.In order to ensure the prediction accuracy of the prediction model,the original power data is preprocessed,and the time,weather conditions and historical load data highly related to the prediction data are selected as the input of the prediction model.The VMD-LSTM-MLR model based on set variational mode decomposition(VMD),long-term and short-term memory(LSTM)neural network and multiple linear regression(MLR)is used to predict power load.VMD-LSTM-MLR model uses VMD to decompose power data into eigenmode functions with different frequency characteristics.According to the zero crossing rate,the decomposed eigenmode functions are divided into high frequency components and low frequency components.Then the LSTM neural network which can maximize the estimation of high frequency components is used to predict the high frequency part.The low frequency part is predicted by multiple linear regression which can easily obtain the accurate load forecasting value of low frequency component.Finally,the predicted high-frequency data and low-frequency data are superimposed to obtain a complete prediction result.Through the experiment,it is found that the combination of LSTM neural network and Light GBM can obtain the advantages of the two models at the same time,improve the mining effect of long time series information by Light GBM,improve the prediction ability of LSTM neural network for discontinuous feature data.Therefore,the combination model of LSTM and Light GBM is used to replace the LSTM neural network in VMD-LSTM-MLR model to get VMD-LSTM-Light GBM-MLR model.VMD-LSTM-MLR model and VMD-LSTM-Light GBM-MLR model are used to predict the electric power data of a city in Jiangsu Province,and good results are obtained.In order to further verify the practicability of these two models in big data samples,VMD-LSTM-MLR model and VMD-LSTM-Light GBM-MLR model are used to predict the electric power data of the 9th electrical mathematical modeling competition,good prediction results are also obtained.