A data-driven load forecasting method for incentive demand response

Abstract

The participation of incentive demand response (IDR) can improve power grid flexibility, and reduce peak shaving pressure. However, the further development of incentive demand response services will be limited by the uncertainty of user response behavior. To tackle this problem, this paper proposes a data-driven load forecasting method for IDR, which considers consumer behavior. Firstly, we describe the power market auxiliary service operation mechanism for load aggregators, through which the ability of demand-side resources to respond to auxiliary services is improved. Then, add an attention layer to improve the traditional long and short memory (LSTM) model, and propose an IDR load forecasting method based on this model, which considers the model’s learning of historical data of user behavior. Finally, establish a simulation model to verify the effectiveness of the improved LSTM model forecasting. The proposed method is simulated and compared with the traditional LSTM method and k-nearest neighbor prediction method. The results show that compared with the other two methods, the mean error, root means square error, and mean absolute percentage error in the improved LSTM method are reduced by 7.31 MW, 8.32 MW, 2.06% and 10.62 MW, 13.46 MW and 3.09%, respectively, which effectively improves the accuracy of forecasting and increase the participation of demand-side resources in the power auxiliary service market.

Keywords

• User behavior forecasting
• Power big data
• Data-driven
• Incentive demand response
• Power market