Machine Learning Approach for Solar Forecasting

Abstract

The Republic of India, which has the world’s second-largest population, faces difficulties in fulfilling expanding energy demands by lowering the use of fossil fuels. In addition, the energy sector’s use of fossil fuels adds to global warming. expressly responsible for the manufacturing of hazardous substances, distribution, and consumption of energy. The sun, an unlimited energy source, can be exploited as an alternative to meet this growing demand. Furthermore, technological advancements in chemistry, material science, and solid-state Physics have improved the efficiency of photovoltaic (PV) modules, resulting in a variety of products. topologies with varying performance characteristics, as well as the addition of sun-fueled plants to a portfolio of the power market. Despite its numerous benefits and popularity as a renew- able energy source, even the best solar panels eventu- ally lose their effectiveness. Because solar cells are prone to damage from weather-related accidents, temperature variations, and UV exposure over time, inspections are required to maintain cell performance levels and minimize financial losses. This Thesis uses solar power generation and weather data from a solar plant to gain insights, address issues, and predict/forecast plant output for improved grid management/stability. In this the- sis, we propose the development and evaluation of machine learning regression models for predicting the rate of solar output, measured as a percentage of baseline capacity. Three distinct models, including Linear Regression, Stacked Ensemble Model, and Light Gradient Boosting Machine (LGBM), were constructed and rigorously assessed. These advanced models demonstrate superior predictive capabilities, offering promising avenues for enhancing solar energy forecasting accuracy and facilitating informed decision-making in renewable energy integration and management systems. By using these techniques, the research improves the accuracy of output projections, optimizes grid management, and contributes to the overall stability of solar power generation systems. The research objective is to pioneer the advancement of solar energy forecasting through the development and meticulous evaluation of machine learning methodologies. Focusing on the utilization of weather data as the primary input, the study seeks to harness the predictive potential of stacked ensemble models and LightGBM algorithms. Through this endeavor, the thesis aims to contribute significantly to the enhancement of renewable energy management practices, offering actionable insights for stakeholders in energy planning, grid optimization, and sustainability initiatives. This study takes advantage of the possibility of real-time estimation of solar power in PV systems. The estimation results demonstrate a significant improvement in terms of the productivity of solar panels.