Red neuronal imitando un EMS basado en optimización: prueba de concepto en un PLC

Source abstract

This paper presents a proof-of-concept of an artificial-intelligence-based energy management system (EMS) for smart grids with battery storage. A multilayer perceptron (MLP) is trained to approximate the optimal control decisions of a model predictive control (MPC) EMS that solves a mixed-integer linear programming (MILP) problem at each time step. The AI model, trained on 267,894 input-output pairs generated by the reference optimiser, replicates the optimal control policy with an RMSE of 0.073 and an inference time approximately 21 times lower than the optimiser. The network is fully implemented in CODESYS on a simulated PLC, demonstrating its operation on a standard IEC 61131-3 industrial controller. Results validate the feasibility of replacing a MILP optimiser with a neural network for real-time energy management under limited computational resources.