Machine Learning Algorithm and Modeling in Solar Irradiance Forecasting

Forecasting solar irradiance is a very important requirement for determining the potential of the energy produced, especially to maintain the continuity of the availability of electrical energy. Creating a solar irradiance forecasting modeling algorithm is important to obtain the degree of closeness between the predicted and actual results. This study proposes a novel method to solve solar irradiance prediction by using a machine learning model based on the Bidirectional Long Short-Term Memory (BiLSTM) with one hidden layer and three input data variables and involves setting hyperparameters. The processed data is primary data obtained from the measurement results of three sensors over a certain period. The training technique and model validation use the backpropagation algorithm by adjusting the learning rate value. The proposed BiLSTM model has achieved convergence at epoch 183
^rd
and no overfitting occurs. Furthermore, the comparison between the BiLSTM model and the Baseline model shows that the BiLSTM model has better performance in predicting solar irradiance one minute ahead with mean absolute error (MAE) and root mean squared error (RMSE) values of 0.004326917696744204 W/m
²
and 0.00917678491314492 W/m
²
respectively. At the same time, the Baseline model shows MAE of 0.01438286076453574 W/m
²
and RMSE of 0.033326815578768605 W/m
²
.