DeepEMS: Multimodal optimal energy management of microgrid systems based on a hybrid multi‐stage machine learning model

Abstract The effective management of microgrids is important towards transition to sustainable energy paradigm. By optimizing the utilization of different energy sources, such as solar photovoltaic panels and energy storages, it improves the reliability of the grid and develops resiliency in dealing with of challenges of unexpected variations in demand. To this end, the proposed paper presents DeepEMS, a system developed to manage the energy of microgrids through the incorporation of diverse intelligent algorithms. DeepEMS provides dynamic microgrid management through the utilization of Bidirectional Long Short‐Term Memory (BiLSTM) networks, Sliding Linear Programming (SLP), and Random Forest (RF). By implementing these methodologies, DeepEMS can optimize energy consumption throughout the microgrid by dynamically identifying and coordinating the needs of various energy sources. DeepEMS achieves precise multimodal optimization and facilitates integration of storage systems, grid interactions, and renewable energy sources (RES), as demonstrated by simulations and data analytics. DeepEMS presented performance in control, resource allocation, management, and grid utilization. Furthermore, in a comparative analysis with alternative intelligent models including XGBoost, Light GBM, RF, and Decision Trees, DeepEMS consistently demonstrated higher performance as measured by several key performance indicators (KPIs).