AI-Driven Renewable Energy Forecasting and Smart Grid Optimization for Energy Transition in Bangladesh

This study is among the first to build and evaluate an end-to-end AI-driven renewable energy forecasting + smart grid optimization framework specifically for Bangladesh’s energy context — a low-resource setting largely absent from the existing ML forecasting literature (which is dominated by developed-country studies).

Key contributions:

Comparative model evaluation — benchmarks five forecasting approaches (Random Forest, ANN, SVR, Linear Regression, ARIMA) on a 5-year, 43,800-instance dataset combining NASA POWER, BMD, and BPDB data; Random Forest wins with 92.4% accuracy, a 57% MAE improvement over the ARIMA baseline.
Localized feature engineering pipeline — tailored to Bangladesh’s meteorological and grid conditions (cyclical time encodings, rolling lag features).
Quantified smart grid gains — demonstrates concrete operational impact when AI forecasting drives dispatch: +31% grid stability, −68% renewable curtailment, −36% CO₂ intensity, forecast horizon extended from 1 to 24 hours.
Practical deployment roadmap — translates the technical results into a policy-relevant argument (e.g., ~4.2M tonnes CO₂ avoided/year at scale), directly tied to Bangladesh’s 2041 Mujib Climate Prosperity Plan target of 40% renewables.

The core novelty is less “a new algorithm” and more rigorous, quantitative evidence — in an underexplored geographic context — that pairing AI forecasting with grid optimization unlocks large, measurable gains, giving policymakers and grid operators a concrete case for investing in digital infrastructure alongside physical renewable capacity.