Research On Load Prediction Methods In Energy Internet New Operation Scenarios

With the development of the Energy Internet,a high proportion of renewable energy,distributed power,flexible loads,etc.are gradually connected to the power grid,which has a certain degree of impact on the operation of the power grid in the new operating scenario.In the new operation scenario,the user’s electricity consumption has changed from a single passive electricity consumption to various forms of active participation.The form of electricity consumption has become more diversified,and the uncertainty of the user’s electricity consumption has gradually increased.A certain degree of impact has occurred,and accurate load forecasting is conducive to the rational arrangement of the power generation plan at the source end and the real-time adjustment of the transaction price of the power grid.Therefore,in order to make the power grid more rationally arrange the power sale plan and further improve the user-side response management,it is of great significance to carry out the load forecast of electric vehicle charging stations and residential users in the new operation scenario.First of all,with the development of the Energy Internet,more and more users are participating in the operation of the power grid.The short-term load forecasting of users can provide theoretical basis for grid decision-making,such as the allocation of power generation plans and the establishment of electricity prices.Load forecasting is one of the important tasks of the power grid.Secondly,in view of the load forecasting problem in the new operation scenario,this paper proposes the SAE-ELM algorithm based on the stack autoencoder and extreme learning machine algorithm in deep learning to realize the load forecasting of electric vehicle charging stations.The electric load data of the electric vehicle charging station is complex,and the average missing method is used to pre-process the abnormal missing data;the factors that affect the load prediction of the charging station are analyzed,such as historical load data,temperature and humidity,and day type,and construct simulation experiment data;the key parameters in SAE-ELM are optimized,such as the learning rate,the number of model layers,and the number of nodes,to improve the model prediction effect.By using MSE,RMSE,MAE,and MAPE measurement error methods,the SAE-ELM algorithm proposed in this paper is compared with SAE-BP and ELM.The results show that the SAE-ELM-based charging station load prediction algorithm proposed in this paper has higher prediction accuracy.Finally,when the SAE-ELM algorithm is used to forecast the load of ordinary residents,the prediction effect is not good.In view of this problem,this article continues the idea of codecs and combines the advantages of LSTM in processing time series,and proposes a Seq2seq codec algorithm based on LSTM to realize the load prediction for residential users.The autocorrelation coefficient method is used to determine the order of the model that affects the residential electricity load.For the key parameters of the model,the appropriate Attention mechanism and residual mechanism are selected,and the number of layers,nodes,and learning rate of the model are determined to improve the accuracy of model prediction.By using MSE,RMSE,MAE,and MAPE measurement error methods,the proposed method is compared with RNN,LSTM,and GRU.The results show that the LSTM-based Seq2seq codec model is used to minimize the load prediction error for residential users.At the same time,the model has been used for small-scale residential users and the load forecasting of electric vehicle charging stations has achieved high accuracy.