Hybrid RBF Neural Network and Multi-Level Genetic Algorithm for MPPT Optimization in Photovoltaic Systems with IHGM-SEPIC Converter: Experimental Validation

Resumen

Abstract– This paper presents a hybrid Maximum Power Point Tracking (MPPT) system that combines Radial Basis Function (RBF) neural networks with a multi-level Genetic Algorithm (GA) for photovoltaic (PV) system optimization. The proposed approach is implemented on embedded hardware (ESP32-S3 and Raspberry Pi 5) coupled with an Interleaved High-Gain Modified SEPIC (IHGM-SEPIC) converter, achieving a measured efficiency of 98.87%. A complete mathematical model of the IHGM-SEPIC topology is derived, comprising 16 state equations. The GA simultaneously optimizes converter parameters, MPPT control parameters, and RBF network weights using BLX-α crossover, adaptive mutation, and tournament selection with a multi-objective fitness function. Experimental validation was conducted using a 100W monocrystalline panel under real Andean highland conditions (Pamplona, Colombia, 2340 (m.a.s.l.) over a dataset of 5,629 records. Results demonstrate 4.67% improvement over conventional P&O and 7.37% over commercial controllers in tracking efficiency, with a response time of 0.8 seconds versus 3.5 seconds for P&O. Statistical validation from N=30 GA executions confirms convergence reliability with σ = 0.35%.