Autonomous Microgrid Optimization: Design, Simulation, and Field Evaluation

Thesis Abstract

This study addresses the operational and economic challenges of rural electrification by advancing autonomous microgrid systems that can self-optimize generation, storage, and demand management under uncertain solar and load conditions. The core problem is to design a robust optimization framework that delivers reliable power, minimizes total lifecycle cost, and ensures resilience to component failures and weather variability. The aim is to develop and evaluate an integrated design, simulation, and field-validated workflow that achieves real-time energy management, scalable architecture, and transparent performance reporting. Specific objectives are (1) to formulate a multi-objective optimization model for autonomous microgrids that jointly minimizes operating cost, carbon footprint, and voltage/thermal losses while satisfying reliability constraints; (2) to implement a decision-support architecture combining model predictive control, reinforcement learning-based dispatch, and distributed energy resources (DER) scheduling; (3) to validate the framework in a high-fidelity simulation environment with stochastic solar irradiance and load profiles; (4) to pilot the approach on a 1.2 MW community microgrid with 350 kWh storage and 600 kW PV capacity, deploying fault-tolerant communication and cyber-physical security layers; (5) to evaluate performance across scenarios including grid-t connected, islanded, and mixed operation modes; and (6) to derive governance, maintenance, and policy implications for scalable deployment in similar settings. The methodology adopts a design–implementation–evaluation sequence built on three strands. First, a quantitative optimization design is developed using a mixed-integer linear programming (MILP) formulation for day-ahead and real-time dispatch, augmented by a non-linear program (NLP) for power-flow fidelity and voltage regulation; a model predictive control (MPC) layer provides receding-horizon decisions, while a reinforcement learning (RL) agent fine-tunes dispatch under unmodeled dynamics. Second, the population comprises operational microgrids in rural or peri-urban contexts with at least two DER types (PV and battery storage) and controllable loads; the primary field pilot includes a community microgrid in Southeast Asia consisting of 1.2 MW solar capacity, 600 kWh storage, and critical load segments for healthcare and water supply. A purposive sampling approach identifies participating households and commercial loads, with 60 demand profiles collected via smart meters over a 12-month period to capture seasonality. Third, data collection employs instrumented sensors for voltage, frequency, state-of-charge, solar irradiance, ambient temperature, and operational logs; governing documents and component specifications are archived for reliability assessment. Validity and reliability are established through instrument calibration, triplicate measurements, and cross-validation with utility-scale data. Data analysis uses regression analysis to model relationship between storage dispatch and losses, ANOVA to compare performance across operation modes, time-series analysis for demand-supply balance, and scenario analysis to stress-test the optimization framework. The conceptual framework integrates IEEE 2030.5 standards for interoperability, the Theory of Planned Behavior to examine user acceptance of automated control, and the Resilience Theory to assess fault tolerance, culminating in a conceptual model that links optimization decisions to reliability and sustainability outcomes. Expected findings indicate that the integrated MPC–RL–MILP framework achieves a reduction in total operating cost by 18–25% relative to baseline heuristic dispatch, while reducing energy losses by 6–12% and maintaining voltage within ±5% of nominal limits under islanded and grid-connected scenarios. Field results are anticipated to show that autonomous decision-making maintains at least 99.9% system availability during peak load events, with storage cycling aligned to solar variability to improve self-consumption by 22–30%. Sensitivity analyses are expected to reveal critical thresholds for storage sizing, PV oversizing, and communication latency beyond which reliability benefits decline. The study contributes to knowledge by delivering a rigorously tested, scalable design–simulation–field evaluation methodology for autonomous microgrids, integrating control theory with practical field constraints and policy considerations. It advances theoretical understanding of multi-objective energy management under uncertainty, demonstrates empirically how RL can complement model-based optimization in microgrid contexts, and provides transferable guidelines for replication in similar settings globally. The main conclusion is that a hybrid control architecture, when implemented with robust cyber-physical security and stakeholder engagement, can deliver reliable, cost-effective, and sustainable autonomous microgrids. Recommendations include the adoption of standardized interoperability protocols, investment in latency-tolerant communication architectures, iterative hardware-in-the-loop testing prior to deployment, and policy frameworks that incentivize reliability and resilience in remote electrification programs.