Techno-economic optimization and machine learning forecasting benchmark of a hybrid renewable microgrid for sustainable humanitarian energy access in a refugee settlement
Year: 2026
Type: article
Abstract: More than 600,000 Rohingya refugees live in the flood-prone hillsides of the Kutupalong-Balukhali camp complex in Cox's Bazar, Bangladesh, where electricity access is limited, intermittent and partly dependent on diesel generators. This study designs a grid-connected hybrid renewable energy system (HRES) for a representative electrified cluster of about 10,000 shelters and their community and commercial facilities in Balukhali (7,536 kWh/day) and, in a second and separate step, benchmarks machine learning (ML) and deep learning (DL) forecasting models on the simulated operation of the optimized system. HOMER Pro was first used to compare PV, wind turbine (WT), battery (BESS) and grid combinations. The optimal PV-WT-BESS-grid system has a net present cost of $1.51 million, a cost of energy of $0.0263/kWh that includes net-metering export revenue, a 73.5% renewable fraction, an annual operating cost of $18,508 and an unmet load of 0.03%. Its operational emissions (706 t CO₂/yr) are 59.5% lower than those of a grid-supplied baseline and, in an indicative screening, about 70% to 76% lower than those of an equivalent diesel supply. The 8,760-h dataset of the optimal system was then used to benchmark five models (MLR, SVR, RF, GBM and LSTM) for 24-h-ahead load forecasting and five models (MLR, SVM, RF, GBM and ANN) for PV and WT power estimation. GBM gave the most accurate load forecasts (R² = 0.932), while ANN and GBM best reproduced the simulated PV (R² = 0.995) and WT (R² = 0.992) output. Because the models were trained on HOMER-generated data with realized rather than forecast weather inputs, these scores are upper bounds and a methodological benchmark, not field-validated accuracy, and the forecasts do not alter the optimal sizing. Sensitivity analysis shows that the real discount rate and solar irradiation dominate the cost of energy. The framework offers a transparent starting point for humanitarian microgrid planning that should be validated with metered demand and measured weather and generation data before deployment.
Source: Energy Nexus
Language: English
Subfield: Energy Engineering and Power Technology
Field: Energy
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,200
OpenAlex ID: W7219675952