Artificial Neural Network-Based Solar Irradiance Forecasting

The solar PV industry has grown at a very fast rate in the renewable energy sector over the last one to two decades. With the current pace of growth, it is expected to replace the use of fossil fuels worldwide to a major extent, by 2050. However, to deliver proper PV power output to the national grid structure, fluctuations in solar energy production can disrupt grid stability. In this research work, the authors have performed a predictive model for solar irradiance based on Artificial Neural Networks (ANNs), based on the various climatic parameters, including temperature, humidity, and cloud cover, as input features. An ANN model was developed from scratch using both Gradient Descent (GD) and Stochastic Gradient Descent (SGD), the solar PV industry has experienced rapid growth in the renewable energy sector over the past one to two decades. With the current rate of development, it is projected to largely replace fossil fuels worldwide by 2050, if not entirely. However, fluctuations in solar energy production can disrupt grid stability when delivering power to the national grid. In this research, the authors developed a predictive model for solar irradiance using Artificial Neural Networks (ANNs), based on various climatic parameters, including temperature, humidity, and cloud cover as input features. An ANN model was built from scratch using both Gradient Descent and Stochastic Gradient Descent optimization techniques to enhance performance. Experimental results show that both methods achieved effective convergence; however, the SGD-based approach resulted in lower mean squared error (MSE) values, indicating greater accuracy in predicting solar irradiance. To evaluate the model’s effectiveness, its performance was compared with a baseline linear regression model. Evaluation metrics such as MSE, MAE, and RMSE were employed. The results reveal that the ANN model outperforms the linear regression model, exhibiting lower error rates across all metrics.