Indirect Forecasting of Hourly PV Power Generation Based on a Hybrid Model Combining Data Analysis and Machine Learning Technique

Abstract

This work presents an indirect way to predict hourly PV power generation. Changes in solar irradiance significantly affect PV power output, although temperature changes have a relatively less impact. This study develops a hybrid model for estimating solar irradiance values using data analysis and machine learning techniques. On the other hand, the hourly temperature is predicted using a basic persistence model. Ensemble empirical mode decomposition (EEMD) breaks the original GHI series into several orthogonal subseries termed intrinsic mode functions (IMFs). A forecasting model based on an ML technique is developed to predict all the IMFs. This study compares two distinct learning-based ML models for solar irradiance and power forecasting, viz. artificial neural network (ANN): a neural network-based ML technique, and extreme gradient boosting (XGBoost): an ensemble learning-based ML technique. Finally, the PV power generation is computed based on a mathematical model by utilizing forecasted solar irradiance and temperature values in Delhi, India. EEMD–ANN reported an improved forecast precision by reducing the RMSE and MAE by 15.86% and 17.81%, respectively, compared to the EEMD–XGBoost. The corresponding RMSE, MAE, and R² score of EEMD–ANN in predicting hourly solar irradiance values are 38.93 W/m², 26.47 W/m², and 0.977, respectively.