A WOA-ATTENTION-LSTM Based MPC in LVDN Energy Consumption Control Under User Comfort Consideration

This paper has proposed an energy-saving control scheme for operating a user energy system in a low voltage distributed network. On the basis of improved long short-term memory (LSTM) neural network, a model predictive control optimization is developed targeting towards the duel objectives of both user comfort and energy saving. The innovation of the study include: 1. by using whale algorithm to optimize super parameters of an LSTM artificial neural network, and introducing attention mechanism to an LSTM full connection layer, the network has ensured the model training speed while keeping prediction accuracy and avoiding over-fitting issue; 2. The MPC controller with AI-based prediction model is used to operate the complex user energy consumption system. It is evidenced that the proposed scheme can ensure the control accuracy due to precise prediction as well as feedback mechanism; 3. The test results show that the proposed MPC achieves 7% energy saving while following the user comfort index PMV accurately.