Physics-Informed Predictive Dispatch and Energy Orchestration for Hybrid Microgrids

Abstract

Smart energy management in hybrid micro grid is a difficult task due to the variability of renewable generation, load demand and electricity prices. A Physics-Informed Predictive Dispatch and Energy Orchestration (PIPEO) framework is developed in this paper that combines a Physics-Guided Long Short-Term Memory (PG-LSTM) forecasting model with Model Predictive Control (MPC) for coordinated scheduling of PV, wind power, battery storage, diesel generation, and utility-grid exchange systems. The PG-LSTM fuses historical operation variables with physical constraints to produce forecasts, whereas the MPC uses the PG-LSTM estimates to generate power-balance, state-of-charge, generator and grid-exchange limit-constrained dispatch commands. Normal operation, reduction of renewable generation, load increase, grid outage, and forecast uncertainty has been simulated in a 30-day MATLAB R2024b scenario with a time stamp of 15 minutes. Compared to conventional MPC, the daily operation cost is decreased from 398.6 USD to 342.4 USD, renewable utilisation increases from 88.6% to 94.2%, renewable curtailment is reduced from 9.8% to 4.1% and carbon emissions are reduced from 149.3 kg/day to 116.7 kg/day. The proposed framework additionally connects an obtainment-reliability worth of 94.6%. The proposed PIPEO framework has been thoroughly tested using MATLAB simulation with a variety of typical hybrid microgrid operating scenarios. The presented study is limited to simulation level validation, but a structured practical validation pathway that includes the processor in the loop (PIL), real-time digital simulation (RTDS), controller hardware-in-the-loop (CHIL), and operation hybrid microgrid evaluation is proposed for future research.