Predicting Photovoltaic Generation Using a Hybrid Model Combining Autoregressive Integrated Moving Average and Support Vector Regression

The increasing adoption of solar energy as a renewable resource necessitates the development of precise models for anticipating solar photovoltaic (PV) power generation. Numerous variables, including local climate, time of day, season, and geography, influence the production of solar PV power. This research is centered on the creation of a prediction model, utilizing 2019 solar radiation data from the Himawari-8 satellite, with General Santos City as the area of interest. ERA5 Reanalysis data were employed to consider additional factors like ambient temperature, cloud cover, precipitation, and relative humidity, impacting power production. Deep Learning, employing a Convolutional Neural Network (CNN) architecture, was utilized to construct the Autoregressive Integrated Moving Average (ARIMA) model. Further optimization was achieved through Support Vector Regression, enhancing the accuracy of predicting irradiance values. Evaluation metrics, including Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), demonstrated the model's efficacy with a low MAE of 2.19 and a low RMSE of 3.04%. Future enhancements may involve expanding environmental parameters and augmenting datasets to further refine the predictive capabilities of the model.