Enhanced model predictive control using artificial neural networks for grid-connected two-level inverters

Integrating artificial intelligence (AI) to improve the performance of photovoltaic systems is compelling for two main reasons. First, AI technologies are making significant progress and are widely applied across various fields, such as medicine and agriculture. Second, the world is experiencing an increasing demand for energy while simultaneously facing the challenge of reducing reliance on traditional energy sources that exacerbate carbon emissions and climate change. This emphasizes the critical importance of transitioning to sustainable and clean energy solutions. In this context, photovoltaic panels have emerged as one of the most promising and widely adopted renewable energy sources. As the global reliance on solar energy grows, enhancing the efficiency and reliability of solar energy systems has become imperative. However, this transition is accompanied by significant challenges, particularly in integrating these systems with existing power grids. The complexities involved in achieving seamless integration, maintaining consistent performance, and preventing system failures necessitate the adoption of advanced technologies. These challenges are further exacerbated by the need to manage fluctuations in energy production, optimize energy storage, and ensure the long-term sustainability of solar power. This paper presents a novel approach for controlling ANN-MPC solar inverters, leveraging artificial neural networks (ANN) to address the challenges related to the efficiency of photovoltaic systems and their integration with the grid, ultimately reducing dependence on conventional energy sources. Initially, the ANN-based model is trained using large datasets derived from the system inputs and outputs, taking into account all critical operating conditions, such as sudden changes in current. Following the training process, the model acquires extensive data, enabling it to adapt to all transient conditions the system may encounter. The model continuously evolves to improve the efficiency of grid-connected photovoltaic systems. Additionally, experimental validation and performance testing of the proposed model are conducted to verify the effectiveness and flexibility of the ANN-based approach. The experimental results confirm the proposed approach's ability to enhance the performance of grid-connected electrical systems, reduce computational costs, and achieve model-free control.