Outage-Aware Predictive Energy Management for Seamless Industrial Operation on Weak Grids: Learned Forecasting and Risk-Scaled Reserves for Grid, Diesel, Solar and Battery Dispatch

Industrial sites on weak grids face frequent, random outages, yet most energy management systems either assume the grid is always available or react to outages only after they occur. This study aims to co-optimize grid exchange, diesel generation, photovoltaic (PV) output and battery storage across grid-connected and islanded operation. An outage-aware model predictive controller (OA-MPC) combines gradient-boosted PV and load forecasters, a logistic outage-onset model with a fitted two-state Markov repair process, and a soft battery reserve that always holds the diesel cold-start ride-through energy and adds sustained-deficit energy in proportion to the predicted onset probability. A one-year, hourly, 20-scenario Monte Carlo study of a 1.2 MWp PV, 2 MWh battery and 800 kW diesel plant serving an 800–1200 kW industrial load compares OA-MPC with rule-based dispatch, an outage-blind MPC and a perfect-forecast MPC. Relative to the outage-blind MPC, OA-MPC reduces total unserved load by 66.8 ± 4.7% and critical unserved load by 96.9 ± 7.1%, cuts total cost including interruption by 12.1 ± 0.9% and diesel starts by 18.8 ± 2.0%, while raising operating cost by only 0.4 ± 0.1%; annual diesel fuel is statistically unchanged. Relative to rule-based dispatch, OA-MPC lowers operating cost by 5.3 ± 0.2% and diesel fuel by 8.0 ± 2.4%. OA-MPC closes 70.3% of the unserved-load gap between the outage-blind and perfect-forecast controllers, and its advantage persists across outage frequencies and forecast-error levels. The remaining gap to the perfect-forecast bound lies mainly in diesel use and reflects the value of knowing outage timing and duration in advance. Future work should test the controller with measured outage, load and solar data from operating industrial sites.