Random Forest Algorithm for Solar Forecasting in Jamshedpur – India

Machine learning has created a wide collection of solar forecasting works because of its recent advancement. For machine learning models, choosing the appropriate hyper-parameter configuration directly affects the model's performance. It frequently causes a thorough understanding of machine learning methods as well as proper hyper-parameter optimization strategies. Hence, this paper provides a Random Forest model and two other optimization techniques for solar radiation forecasting: Grid Search and Random Search, and the performance of these two are compared. Experiments are conducted on datasets of Diffuse Horizontal Irradiance (DHI), and Direct Normal Irradiance (DNI) including 9 other parameters, for Jamshedpur (Jharkhand). Widespread data analysis is done using a correlation plot, box plot, and histogram. These models predict using the similar day approach, which assumes that the sun and earth are in the same location on a comparable day in prior years. The results suggest that Random Search outperforms (R ² = 0.93063 for DHI and 0.85956 for DNI) the Grid Search (R ² = 0.93056 for DHI and 0.85727 for DNI) in terms of accuracy.