Hybrid ANN WRF solar radiation forecasting in data limited tropical region

Abstract

Understanding solar radiation is vital for optimizing the integration of solar energy systems, particularly in regions with diverse topographical features. West Java Province in Indonesia, characterized by its varied topography and substantial solar potential, serves as an ideal case study for advanced predictive modelling. This study investigates the potential for solar energy development in West Java Province by estimating solar radiation using Artificial Neural Network (ANN) and hybrid methods to identify the optimal configuration model and analyze its spatial distribution. Solar radiation measurements were collected from five locations, with the two best locations selected for data processing using data from January to December 2022. The dataset was divided into 70 % training data and 30 % testing data. The optimal ANN configuration for the Lowland location was 6-30-1, yielding an RMSE of 135.8 W/m², rRMSE of 54.8 %, MBE of 15.9 W/m², and rMBE of 0.064 %. For the Highland location, the optimal configuration was 5-40-1, with an RMSE of 156.7 W/m², rRMSE of 49.29 %, MBE of 7.75 W/m², and rMBE of 0.024 %. The model's overall estimation error ranged from 48–50 %. Integrating the ANN model with WRF improved accuracy in the Highland area by 2 %. Spatial distribution analysis indicated that lower-altitude areas experience higher solar radiation intensity, while higher-altitude areas receive lower radiation due to specific atmospheric conditions influenced by the province's varying altitudes.