Short-term Load Forecasting Based On Multi-factor And EEMD-LSTM-MLR Method

The safety and economic operation of the power grid have become crucial with the development of the power market and the gradual improvement of user requirements.The accurate short-term electric load forecasting can effectively guarantee the safe operation of the power grid,reduce the cost of power generation,meet the needs of users and improve the social and economic benefits.At the same time,with the increasingly severe energy crisis and environmental pollution problems,the power system urgently needs to make effective innovations in operation and dispatch.Accurate short-term load forecasting is a necessary prerequisite for optimal dispatch of power system.Corresponding to the traditional way of dispatching by regulating the output of generating units,the traditional method of load forecasting mainly aims at regional loads.The development of smart grid provides technical and theoretical support for users to participate in power system operation and dispatch,but the traditional load forecasting method is less practical for user load with strong randomness.Electric load has obvious periodic characteristics,meanwhile,it is affected by many complex external factors such as weather,economy,holidays,observation errors,as well as uncertainties in user behavior.Power system loads show strong stochastic characteristics locally in the basic overall periodicity law.This kind of random feature includes the part caused by external factors and the part caused by user's behavior.Combining these uncertainties,the difficulty of load forecasting is increased,which makes it difficult to guarantee the validity and accuracy of load forecasting only from a single point of view.Therefore,from two perspectives,this paper fully excavates the law of load self-change,while fully considering the impact of external factors on load change,and combines the two aspects to optimize the final results.On the one hand,through the calculation of Pearson correlation coefficient,the related characteristics of load forecasting are screened,and a multi-input LSTM neural network deep learning model is constructed,which takes into account the external factors such as temperature,weather,holidays and weeks,to improve the accuracy of load forecasting.On the other hand,in order to solve the problem that the load is highly random and the prediction accuracy is not high,this paper improves the accuracy of short-term electric load forecast by constructing ELM(EEMD-LSTM-MLR)prediction method.Ensemble Empirical Mode Decomposition(EEMD)is used to decompose the electric load data into different intrinsic mode functions(IMF)from frequency high to low.Then,the multiple linear regression(MLR)method and the LSTM neural network method are used to predict the low frequency part and the high frequency part respectively.Finally,the predicted results are combined to obtain the complete prediction result.ELM prediction method not only can accurately predict the trend of electric load,but also can effectively predict the local features of randomness.The reciprocal variance method and Shapley value method are used to optimize the combination of the two prediction results,which makes the overall model more comprehensive and more stable,and further improves the accuracy of the prediction.