DOI: 10.11648/j.ajece.20261002.12 ISSN: 2640-0502

Adaptive Hybrid MPPT with Tabular Reinforcement Learning for Adaptive PI Gain Tuning in a PV-Fed Buck–Boost Converter

Olusesan Orebanjo, Stephen Ukagu, Dennis Okonkwo, Osamudiamen Amasowomwan, Ikenna Onyegbadue, Dedacus Ohaegbuchi, Lucky Odejimi
Show PDF Cite
Maximum power point tracking (MPPT) is essential for efficient photovoltaic (PV) operation under varying irradiance and temperature. Conventional methods such as Incremental Conductance (INC) remain attractive for their simplicity, yet fixed proportional–integral (PI) gains limit dynamic performance when environmental conditions change rapidly. Under partial shading the power–voltage characteristic can also contain multiple local maxima, making reliable global tracking more difficult. This paper proposes a hybrid MPPT architecture in which a tabular Q-learning agent adaptively tunes the PI gains of a PV-fed buck–boost converter while INC retains responsibility for generating the operating-point reference. The learning state is defined from PV voltage, power and their increments; actions are bounded gain steps constrained by small-signal stability analysis; and the reward balances power improvement against excessive gain movement. Because the learning agent only supervises the inner-loop gains rather than replacing the entire control law, the design preserves the interpretability of classical INC while introducing operating-condition-dependent adaptation. The controller is evaluated in simulation against fixed-gain INC and direct-duty-cycle Q-learning under six irradiance profiles, with additional temperature, partial-shading, ablation and hyperparameter-sensitivity tests. Results demonstrate that the hybrid architecture achieves competitive tracking performance while maintaining adaptive PI gains within stability bounds. Ablation confirms that joint adaptation of both gains outperforms single-gain cases, and the sensitivity study identifies a robust hyper-parameter region. The study confirms the feasibility of lightweight, hierarchical reinforcement learning for adaptive PI regulation in PV converters and provides a practical framework for online simulation-based gain tuning under dynamic environmental conditions.

More from our Archive