Research Article
Optimization of a Low-Power Solar Inverter using AI-Based Reinforcement Learning
@INPROCEEDINGS{10.4108/eai.16-9-2025.2361034, author={Sukarman Purba and Bakti Dwi Waluyo and Wanapri Pangaribuan and Selly Annisa Binti Zulkarnain}, title={Optimization of a Low-Power Solar Inverter using AI-Based Reinforcement Learning}, proceedings={Proceedings of the 7th International Conference on Innovation in Education, Science, and Culture, ICIESC 2025, 16 September 2025, Medan, Indonesia}, publisher={EAI}, proceedings_a={ICIESC}, year={2026}, month={3}, keywords={low-power inverter reinforcement learning solar photovoltaic total harmonic distortion energy efficiency deep q-network}, doi={10.4108/eai.16-9-2025.2361034} }- Sukarman Purba
Bakti Dwi Waluyo
Wanapri Pangaribuan
Selly Annisa Binti Zulkarnain
Year: 2026
Optimization of a Low-Power Solar Inverter using AI-Based Reinforcement Learning
ICIESC
EAI
DOI: 10.4108/eai.16-9-2025.2361034
Abstract
The rapid growth of rooftop photovoltaic (PV) installations demands compact, high-efficiency inverters that maintain power quality under varying solar conditions. This study proposes a reinforcement learning control strategy for a low-power single-phase inverter. A Deep Q-Network (DQN) agent is trained to minimize total harmonic distortion (THD) and switching losses while tracking a sinusoidal reference across a wide range of solar irradiances. Following a two-stage methodology, a detailed MATLAB/Simulink model was used for offline agent training. The learned policy was then deployed on a 32-bit microcontroller and validated on a 500 W hardware prototype. Experimental results show the proposed controller achieves a 1.8% THD, outperforming conventional PI (3.9%) and model-predictive (2.4%) controllers. Efficiency improved by 2.7% due to optimized switching, and dynamic response to irradiance changes was 35% faster.
