A Neural Networks-based Maximum Power Point Tracker with Improved Dynamics for Variable DC-Link Grid-Connected Photovoltaic Power Plants

This paper proposes a maximum power point (MPP) tracking algorithm based on neural networks to correctly track the MPP even under abrupt changes in solar irradiance and to improve the dynamic performance across the dc capacitor utilized in the power converter that serves as an interphase to connect photovoltaic power plants into the ac grid. Traditional maximum power point tracking algorithms such as "perturb and observe" (P&O) and "incremental conductance" (IC) are able to track the point of maximum power in most cases. However, they can fail under rapid changing atmospheric conditions. Furthermore, in architectures with a power converter operated at variable dc-link, P&O and IC-based algorithms provide a step-like voltage reference which translates into a repetitive overshoot across the dc capacitor. This may negatively affect the lifespan of the capacitor and affect the overall dynamic of the system. This paper develops a neural network (NN) algorithm that overcome the aforementioned issues, including thorough modelling and control of the various components involved in the realization of a grid-connected photovoltaic power plant. The approach is validated via detailed computer simulations on experimental data.