Solar irradiance forecasting models using machine learning techniques and digital twin: A case study with comparison

Abstract

The ever-increasing demand for energy and power consumption due to population growth, economic expansion, and evolving consumer choices has led to the need for renewable energy sources. Traditional energy sources such as coal, oil, and gas have contributed to global pollution and have adverse effects on human health. As a result, the use of renewable energy for power generation has increased tremendously. One such area of research is solar irradiation prediction, which utilizes Artificial Intelligence and Machine Learning techniques. With the use of real-time predicted data, the digital twins are intended to add value to the organization by identifying and preventing problems, predicting performance, and improving operations. This paper provides an overview of various learning methods used for predicting irradiance and presents a new ensemble solar irradiance forecasting model that combines eight machine learning models to ensure model diversity. The model's most critical factors for predicting irradiance include temperature, cloudiness index, relative humidity, and day of the week. To conduct a comprehensive analysis, the proposed 8-Stacking Regression Cross Validation (8 STR-CV) model was tested using data from three different climatic zones in India. The model's high accuracy scores of 98.8% for Visakhapatnam, 98% for Nagpur, and 97.8% for the mountainous region make it a valuable tool for future prediction in various sectors, including power generation and utilization planning.