Artificial neural network-based flexible power point tracking for enhanced dynamic response of grid-connected photovoltaic systems

The widespread adoption of photovoltaic (PV) systems has accelerated in recent years due to technological maturation, reduced manufacturing costs, and enhanced power electronics enabling reliable grid integration. Additionally, the global transition toward low-carbon energy sources and the implementation of favorable energy policies have reinforced this trend. However, the intrinsic variability of PV generation driven by climatic and load fluctuations remains a challenge for maintaining grid stability. To address this, a Flexible Power Point Tracking (FPPT) algorithm based on Artificial Neural Networks (ANN) aimed at improving the dynamic response of grid-connected PV systems is proposed in this paper. A PV system, modeled in MATLAB/SIMULINK R2022b, simulates a two-stage, grid-connected PV generator with a rated capacity of 3.68 kWp and programmable local loads. The ANN was trained using real-world data collected in Santo Domingo, Dominican Republic, via MATLAB’s NNSTART toolbox. Validation was conducted through Controller Hardware-in-the-Loop (CHIL) simulations. The results reveal that the proposed ANN-FPPT algorithm accurately and consistently adjusts to the changes in irradiance received by the PV array with 99.67 % accuracy, while the traditional P&O-CPO featured 92.34 %, based on the Mean Absolute Percentage Error (MAPE). The proposed approach demonstrated superior responsiveness and stability, especially when operating to the left of the maximum power point, where it maintained higher accuracy across a wider load range.