Comparison of Echo State Network with ANN-Based Forecasting Model for Solar Power Generation Forecasting

Utilization of electricity from solar and wind has increased lately, due to which its integration with grid support and reliability is our priority. This can only be possible when we have accurate forecasting models, which have the potential to handle nonlinear data. Forecasting helps us to make decisions for coordination among devices and scheduling which ensures reliability. Solar power exhibits nonlinear characteristics because its output power is dependent on nature. Machine learning models can handle nonlinear data very efficiently. Hence, an echo state network-based (ESN) forecasting model is developed for forecasting solar power and made a comparison with the conventional ML model, i.e., the ANN model. To evaluate the model performance, a statistical method is used such as mean absolute error (MAE), mean square error (MSE), mean absolute percentage error (MAPE), and correlation of determination (R2). The model performance is good if it shows the minimum value for MAE, MSE, and MAPE and the maximum value for R2, which is obtained for ESN-based forecasting model. Hence, we can conclude that the ESN performance is better than ANN-based prediction model. Forecasting is done on solar power generation to predict a day ahead, a week ahead, two weeks ahead, and a month ahead.