A Study of Microgrid Scheduling Methods for Bidirectional LSTM Networks Under Constraints

Microgrids, as a crucial component of distributed energy systems, encounter intricate optimization and scheduling issues. Aiming to improve the effectiveness of deep learning methods in micro grids when constrained, this paper proposes a solution to enhance the Bidirectional LSTM model. It involves generating new data and training a second Bidirectional LSTM model with added power constraints for devices like energy storage and power generation. These added constraints are incorporated to refine the initial prediction results of the Bidirectional LSTM model. Compared to the model trained only once, the proposed method significantly decreases both the mean square deviation and the average threshold exceeding value of the prediction data. It also reduces the potential risk of scheduling scenarios that exceed the affordability of the microgrid under constrained conditions.