| Abstract: |
Photovoltaic (PV) generation is intrinsically nonlinear, and the position of the maximum power point (MPP) shifts continuously with irradiance, cell temperature and partial-shading conditions, so the tracking algorithm governing the converter duty cycle has a first-order effect on system efficiency. This paper presents an empirical, data-driven evaluation of an Adaptive Neuro-Fuzzy Inference System (ANFIS) based Maximum Power Point Tracking (MPPT) strategy applied to a boost-converter-interfaced PV array, benchmarked against Perturb and Observe (P&O), Incremental Conductance (INC), classical Fuzzy Logic Control (FLC) and Particle Swarm Optimisation (PSO). A dataset of 480 irradiance-temperature operating points, generated through MATLAB/Simulink simulation and cross-validated against a 250 W polycrystalline module datasheet, was used to train the ANFIS network (Sugeno-type, two inputs, 49 rules, hybrid learning) and to test all five controllers under identical step, ramp and partial-shading disturbances. Tabulated results show the ANFIS controller achieving a mean tracking efficiency of 99.3%, a settling time of 85 ms and an RMSE in tracked power of 0.6-1.4 W across irradiance bands of 200-1000 W/m², figures that improve on P&O by roughly 2.5 percentage points in efficiency and by a factor of five in settling time. Steady-state oscillation around the MPP, a persistent weakness of hill-climbing methods, was reduced to below 0.3% of rated power. The findings support the abstract's central claim: intelligent, data-trained controllers close a measurable performance gap left by conventional and metaheuristic MPPT methods, and the discussion situates these gains against seven prior comparative studies, several of which report smaller efficiency margins under identical disturbance profiles, likely due to differences in membership function tuning and training-set diversity. |
