Data-driven demand prediction based on integrated LSTM model

Abstract

The flexibility of the power grid can be significantly promoted by the participation of power customers in dispatch. However, as the uncertainty of customer behavior,the development of demand response services is limited. To solve this problem,the framework of incentive-based demand response is constructed in this paper. The way that load aggregators integrate demand-side resources to participate in the power market is elaborated. And the behavior of power customers responding to incentive policies is transformed into demand elasticity. Then,a data-driven demand elasticity prediction method based on the integrated long short-term memory (LSTM) is proposed. Meanwhile,to improve the performance of the prediction model,the original data is smoothed and scaled,and the weight coefficients of the loss function are added. The simulation results show that,compared with the traditional LSTM algorithm and the k-proximity prediction method,the average forecasting error with the proposed model for the demand elasticity is reduced by 5.33% and 28.8%,and mean absolute percentage error (MAPE) for the total load prediction is reduced by 2.06% and 3.09%. Additionally,based on integrated LSTM,the influence of smoothing and scaling data preprocessing on prediction accuracy is analyzed. The results show that the prediction accuracy can be significantly promoted by data preprocessing.

Keywords

• integrated long short-term memory (lstm) algorithm
• demand elasticity
• data preprocessing
• power market
• incentive-based demand response
• data-driven