Application of an LSTM-Based Forecaster in a Model Predictive Controller of a Microgrid

Following the establishment of renewable energy communities (REC), it has become essential to develop control systems for energy management. In fact, efficiently managing generators, loads, and storage systems is important to promote self-consumption. In this context, accurate prediction of both generation and consumption profiles, as well as market prices, becomes crucial. Indeed, accurate forecasts allow for defining optimal schedules for energy trading, thus minimizing costs. In this perspective, the current work presents a model predictive control (MPC)-based controller for managing two microgrids. Additionally, to exploit the ability of MPC to execute control actions based on future forecasts, a market energy price prediction model is implemented. This predictor leverages LSTM artificial neural networks (NNs) capable of capturing complex temporal dependency patterns in time series to make predictions. The study analyzes two scenarios. The first considers only generation and load profiles while the second incorporates market prices. In particular, the latter scenario compares the performance of the LSTM-based model with a traditional autoregressive integrated moving average (ARIMA) model. The results obtained are very promising and pave the way for new analysis methodologies.