Optimal planning for integrated electricity and heat systems using CNN-BiLSTM-Attention network forecasts

Abstract

This paper investigates optimal planning of integrated electricity and heat systems (IEHS) based on convolutional neural network-bidirectional long short term memory with attention mechanism (CNN-BiLSTM-Attention) network forecasts. Firstly, CNN and BiLSTM are employed to extract the spatio-temporal features of IEHS data, and the attention mechanism automatically assigns corresponding weights to BiLSTM to distinguish the importance of different time load sequences, then a combined CNN-BiLSTM-Attention network is constructed, which allows for a more accurate load forecasting. Furthermore, we establish optimal planning strategy to improve efficient operation and energy efficiency of the IEHS, in which the upper-level planning model with the optimization objective of minimizing comprehensive operation costs is formulated, the lower-level efficiency model aiming to balance revenue and cost is considered, then genetic algorithm and iterative solution strategy of Cplex solver are applied to optimize the bi-level objective functions. Simulation results show that the proposed CNN-BiLSTM-Attention model obtained smaller RMSE, MAE and MAPE compared to several other forecasting models. In addition, the optimization method could obtain global optimal solution of the bi-level objective functions, which improves the energy utilization efficiency of the IEHS.