Machine learning and physics-based model hybridization to assess the impact of climate change on single- and dual-axis tracking solar-concentrated photovoltaic systems

This study explores the hybridization of machine learning models with analytical techniques to evaluate the impact of climate change on active power predictions for single-axis (PAS) and dual-axis (PAD) tracking solar-concentrated photovoltaic (CPV) systems in Alice Springs, Australia. It integrates meteorological factors, including global solar irradiation, extraterrestrial solar radiation, clearness index, ambient temperature, and relative humidity, to assess their influence on system performance. Descriptive statistics for PAS and PAD reveal slight variations in these factors, with PAD experiencing higher values for solar irradiation and ambient temperature. To enhance forecasting accuracy, machine learning models, including SARIMA, CARIMA, MLP, RBF, boosting (BOT), and bagging (BAG), are utilized to predict future performance of CPV systems in Alice Springs. These models consider historical weather data and system performance metrics to make accurate predictions for optimal system operation and maintenance. A novel hybrid model, CARIMA-SARIMA-LG, integrates SARIMA, CARIMA, and the logistic distribution (LG) model, showing exceptional performance. The CARIMA-SARIMA-LG model achieves R² values of 0.9833% for PAS and 0.9777% for PAD during training, surpassing other models in error metrics such as MAPE, RMSE, and nRMSE. In contrast, traditional machine learning models like MLP and RBF exhibit diminished predictive capabilities. Ensemble methods, while useful, do not achieve the accuracy of statistical models. Furthermore, the study evaluates the efficacy of the CARIMA-SARIMA-LG hybrid model across five African locations (Kano, Accra, Johannesburg, Nairobi, and Dar es Salaam), demonstrating PAD's adaptability to diverse climates. The findings illustrate PAD's outstanding accuracy and adaptability across a variety of climatic situations, providing breakthrough insights for renewable energy generation and grid integration. The influence of changing climate conditions on solar photovoltaic systems in Alice Springs, Australia, across SSP126, SSP245, and SSP585 scenarios demonstrates notable differences in active power generation for both single-axis tracking (PAS) and dual-axis tracking (PAD) systems. Under SSP126, the projections indicate slight variations in power, peaking in DJF with an increase of +3.145% from 2015 to 2030, followed by a decrease of −3.775% in the late century from 2071 to 2099. PAD systems show comparable seasonal patterns, experiencing increases in the initial phases (+5.463%, DJF, 2015–2030) followed by declines as the century progresses (−0.578%, ANN, 2071–2099). Under the SSP245 scenario, moderate losses are projected, with PAS facing a decrease of −2.138% (annual average, 2031–2050) and PAD encountering significant declines by the period of 2071–2099, amounting to −5.001% (annual average). SSP585 indicates the most drastic declines, with PAS facing a reduction of −3.294% (ANN, 2015–2030) and PAD seeing a decrease of −5.731% (ANN, 2015–2030), intensifying energy security challenges amid persistently elevated emissions. The findings highlight the critical importance of implementing robust mitigation measures, fostering technological innovations, and developing adaptive energy strategies to secure reliable solar energy production in the face of changing climate conditions. Advancements in tracking systems and energy storage technologies may mitigate certain adverse effects associated with climate change, especially in more severe scenarios such as SSP585.