A Neural Network-Based Conversion Loss Model with Hard Constraints for Energy Management

The evolving microgrid technology integrates various converters for varieties of energy sources and applications. In modern energy management systems (EMS), the increasing number of power conversion processes between energy sources introduces additional decision variables, which subsequently increase the complexity of the resulting optimization problems. Most existing conversion loss models are too complex to fit in optimization problems. This paper presents a neural network-based linear surrogate model for the accurate and efficient approximation of power conversion losses. In energy management problems, a primary concern of the neural network-based surrogate models is that the neural networks may violate the optimization constraints due to their black-box nature. In this study, the proposed neural network model is trained with the augmented Lagrangian method to enforce additional hard constraints on the network input/output variables. Moreover, the trained neural network is reformulated as a mixed-integer linear programming (MILP) model, allowing the model to be used in energy management problems that can be efficiently solved using MILP solvers. The case study results demonstrate that the proposed model is capable of approximating the conversion loss with small absolute errors while satisfying the additional hard constraints. In addition, the resulting MILP model can be solved efficiently using state-of-the-art MILP solvers.