Day-ahead Demand Response Potential Forecasting Model Based on BILSTM for Load Aggregators

With the continuous development of social economy, the exhibition of stable, continuous and efficient power supply has become the focus of the global energy field. However, the traditional load aggregator model can not deal with irregular time series data and variable-length series flexibly, which makes the traditional model face certain difficulties in adapting to practical application scenarios. Therefore, considering the real-time and efficiency of the model, a demand response (DR) potential prediction model based on bidirectional long-term and short-term memory network (BiLSTM) is proposed. Firstly, customers are divided into different types according to their willingness to reduce economic costs and respond, and then a Home Energy Management System (HEMS) model considering electricity price incentives is constructed. Secondly, the Light gradient boosting machine (LightGBM) model is used to rank several influencing features according to their importance, which reduces the redundancy of input features. Finally, on the basis of selecting features, BILSTM model is introduced to predict the DR potential of load aggregators (Las). The results of the simulation-based experimentation unequivocally demonstrate the efficacy of the proposed approach.