Enhancing photovoltaic system performance through NARX-LSTM forecasting and neuro-controller based MPPT techniques

Abstract

The increasing worldwide energy demand in transportation, residential, and industrial sectors, along with the adverse environmental impacts of continued fossil fuel use, has made renewable energy sources critically important. Renewable energy, which includes wind, geothermal, biomass, solar, and hydropower, is crucial for enhancing energy security, decreasing reliance on non-renewable resources, and addressing global environmental deterioration. Between 2022 and 2023, the total global capacity for renewable energy achieved an important achievement of 3, 381 gigawatts (GW). Within this range, solar energy emerged as a prominent source, with an installed capacity of 1, 053 GW, ranking second only to hydropower, which leads with a capacity of 1, 392 GW. Over the past decade, the installed solar capacity has experienced substantial growth, escalating from 140, 514 megawatts (MW) in 2013 to 1, 053 GW by 2022. This dramatic expansion highlights the decreasing costs and enhancing efficiency of solar technologies, reflecting their growing feasibility to meet surging energy demands sustainably. Despite the rapid growth and adoption of solar power, the overall performance of photovoltaic (PV) systems is significantly influenced by various environmental factors. These include solar irradiance, temperature, and atmospheric conditions, which can vary widely depending on geographic and climatic contexts. Addressing this through continuous technological innovation and system design improvements is essential for maximizing the potential of solar energy to meet the world's growing energy needs effectively and sustainably. In this context, this thesis presents advanced methodologies to enhance the performance of PV systems through two main approaches. Firstly, it develops a hybrid Nonlinear Autoregressive with Exogenous Inputs (NARX) and Long Short-Term Memory (LSTM) model for accurate daily solar irradiance forecasting. This model effectively handles the complex, non-linear relationships inherent in weather-related data, significantly improving prediction accuracy and system responsiveness to environmental changes. Secondly, the thesis implements neuro-controller-based Maximum Power Point Tracking (MPPT) techniques, which dynamically optimize energy capture and conversion efficiency, adapting in real-time to varying environmental conditions. Additionally, the thesis incorporates a theoretical application study of a 30 MWp grid-connected solar PV power plant in Djibouti.