Machine learning-based economic model predictive control for energy hubs with variable energy efficiencies

The effective management of multi-energy systems (MES) is crucial for optimizing resource allocation and reducing operational costs, given the nonlinear interdependencies among multiple energy carriers. This paper proposes a data-driven economic model predictive control (MPC) framework for the optimal operation of an energy hub (EH) integrating electricity, heat, and gas networks. Unlike conventional model-based approaches that rely on fixed or simplified energy efficiency assumptions, the proposed method employs an artificial neural network (ANN) to learn and approximate the nonlinear efficiency characteristics of energy conversion and storage. By incorporating ANN predictions into the economic MPC framework, the proposed approach enables multi-step dynamic forecasting, ensuring constraint satisfaction, real-time state updates, and online optimization. The optimization and machine learning toolkit (OMLT) is adopted to efficiently solve the resulting optimization problem. Comparative studies demonstrate that ANN-based economic MPC achieves a near-optimal control performance, with an economic cost gap of only 1.2% compared to the nonlinear EH model. Additionally, it reduces energy costs by 15.5% compared to traditional MPC with fixed efficiency assumptions. These results highlight that ANN-EMPC effectively captures the nonlinear efficiency variations of EH components, leading to more adaptive, data-driven decision-making. The proposed framework offers a scalable and computationally efficient solution for real-time multi-energy system management, bridging the gap between physics-based modeling and data-driven control strategies. • A data-driven economic MPC framework to facilitate the energy management of EHs. • Fast decision-making to ANN-based optimization problems with long time predictions. • Balanced distribution of energy resources among EHs with low operational costs.