Treffer: Multi-scale reinforcement learning framework for development policy optimization: Evidence from energy poverty alleviation.

Development policy implementation faces fundamental challenges of resource allocation under uncertainty, heterogeneous population needs, and dynamic intervention effectiveness. Traditional static optimization approaches fail to address these complexities, resulting in suboptimal outcomes across health, education, energy, and social protection sectors. This study develops a multi-scale reinforcement learning framework for adaptive development policy optimization, integrating multi-armed bandits for regional allocation, contextual bandits for household targeting, and deep reinforcement learning for sequential optimization. The framework addresses hierarchical decision-making structures common across development programs while incorporating uncertainty quantification and adaptive learning mechanisms. Empirical validation employs energy poverty alleviation in Ghana as proof-of-concept, utilizing comprehensive data from 4800 households across 16 administrative regions. Thompson Sampling multi-armed bandits achieve 94.8 % optimal efficiency in regional targeting, representing 15–25 % improvement over conventional allocation approaches. The integrated framework demonstrates 23 % enhancement in poverty reduction compared to demographic targeting methods, with substantial cost-effectiveness gains for typical program budgets. Robustness analysis confirms enhanced stability across data degradation scenarios and external shocks. The framework establishes transferable methodological foundations for adaptive development policy optimization, extending beyond energy poverty to health program targeting, education resource allocation, and social protection system design. Implementation guidelines address varying institutional capacities and data availability scenarios across development contexts. The modular design enables progressive advancement as organizational capabilities develop while providing immediate benefits through robust algorithmic implementations. • First reinforcement learning framework for energy poverty policy optimization. • Thompson Sampling achieves 94.8 % efficiency in regional resource allocation. • 15–25 % improvement over conventional allocation methods demonstrated. • Comprehensive validation using 4800 households across Ghana's 16 regions. [ABSTRACT FROM AUTHOR]

Copyright of Renewable & Sustainable Energy Reviews is the property of Pergamon Press - An Imprint of Elsevier Science and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)