Microgrid energy management strategy using deep learning neural network

Abstract

Abstract Under the background of a large number of uncertain new energy connected to microgrids, traditional model driven methods could be hard to address the speed problem of microgrid optimization. Therefore, we come up with a data-model hybrid driven microgrid scheduling strategy. This strategy includes three stages: day-ahead dispatching, preparing for day-in dispatching and day-in dispatching. In the day-ahead dispatching stage, basing on the day-ahead forecast data of renewable energy, the mixed integer linear programming (MILP) is used to solve the day-ahead dispatching plan. In the preparing for intraday scheduling stage, a neural network is introduced as the day scheduling model, new energy power and load power regard as inputs to the neural network, controllable units in the microgrid as outputs to train the neural network. During the intraday scheduling phase, neural networks are used to optimize the scheduling of controllable units in microgrids based on ultra short term predictions and pre day predictions. Finally, the effectiveness of the raised method was verified through numerical analysis.