Machine-Learning-Augmented Sizing and Dispatch of PV-Grid-BESS Hybrid Microgrids to Achieve 100% Reliability at Lower LCOE: A Replicable Framework for Critical Facilities

• Author(s): Sean Tapiwa Kabera; Munashe Naphtali Mupa; Patronella Siphatisiwe Mtemeli; Peter Mangoro; Tazvitya Aubrey Chihota
• Paper ID: 1714253
• Page: 1085-1094
• Published Date: 19-02-2026
• Published In: Iconic Research And Engineering Journals
• Publisher: IRE Journals
• e-ISSN: 2456-8880
• Volume/Issue: Volume 9 Issue 8 February-2026

Abstract

The present paper presents a scalable machine-learning-enhanced model of sizing and dispatching photovoltaic (PV)-grid-battery energy storage system (BESS) hybrid microgrids with the objective of attaining 100 percent reliability and minimum levelized cost of energy (LCOE) of critical or essential facilities, including hospitals and data centers, in the United States. In an era when grid vulnerability has increased due to extreme weather, cyberattacks, and increasing demand, renewable integration is a sustainable solution, but intermittency remains a challenge. Conventional deterministic and stochastic approaches have tended to produce oversized systems, expensive systems, and inefficient reliability because of poor management of uncertainties, such as weather variability, tariffs, and outages. The given framework fills these gaps by expanding the previous reliability-cost optimization into a U.S.-oriented blueprint in line with the Department of Energy (DOE) resilience programs, such as Grace Resilience and Innovation Partnerships (GRIP). Single-diode PV equations, dynamics of kinetic BESS, net metering, and the stochastic Monte Carlo simulations of uncertainties are included in the system modeling. ML augmentation uses XGBoost surrogates to optimize multi-objective (sizing at LPSP=0) and real-time dispatch (PPO/TD3) using reward functions based on cost and reliability. The replicability will be guaranteed by step-by-step procedures that are adjusted to the regional policies, incentives (e.g., IRA tax credits), and software such as HOMER Pro, DER-CAM, and the information of NSRDB, OpenEI, and EIA. Case studies confirm the strategy: a Californian hospital is able to reduce its LCOE (15%, from 0.085 to 0.072/kWh) with 100 percent reliability, using a high solar-to-net metering; a data center in Texas does the same, cutting LCOE by 15 percent (0.092 to 0.078/kWh) during a hurricane risk. Sensitivity analysis verifies that it is strong against degradation and outages. There are also data dependencies and computation requirements, and the consequences are the implications of scalable deployments of DOE-aligned outage reduction in national costs. Real-world pilots and digital twins will be used in the future.

Keywords

Hybrid Microgrids, PV-Grid-BESS, Machine Learning, Reinforcement Learning, LCOE Optimization, Reliability Metrics, Grid Resilience, DOE GRIP, Net Metering, Critical Facilities

Citations

IRE Journals:
Sean Tapiwa Kabera, Munashe Naphtali Mupa, Patronella Siphatisiwe Mtemeli, Peter Mangoro, Tazvitya Aubrey Chihota "Machine-Learning-Augmented Sizing and Dispatch of PV-Grid-BESS Hybrid Microgrids to Achieve 100% Reliability at Lower LCOE: A Replicable Framework for Critical Facilities" Iconic Research And Engineering Journals Volume 9 Issue 8 2026 Page 1085-1094

IEEE:
Sean Tapiwa Kabera, Munashe Naphtali Mupa, Patronella Siphatisiwe Mtemeli, Peter Mangoro, Tazvitya Aubrey Chihota "Machine-Learning-Augmented Sizing and Dispatch of PV-Grid-BESS Hybrid Microgrids to Achieve 100% Reliability at Lower LCOE: A Replicable Framework for Critical Facilities" Iconic Research And Engineering Journals, 9(8)