Long Short-Term Memory Model for Time Series Prediction and Forecast of Solar Radiation and other Weather Parameters

Abstract

Interest in the solar radiation and associated meteorological variables have been growing over the years due to their effect on energy generation, agriculture and food security, Ozone layer and other industrial applications. Hence, forecasting these variables in long-term and short-terms using various fusions of measured weather parameters has become a major research focus in many regions. However, developing and selecting an accurate model for the prediction of solar radiation based on several weather parameters is still a challenging task. In this study, a time series model for predicting and forecasting Solar Radiation and other weather parameters was developed using Long Short-Term Memory (LSTM). A publicly available dataset of Mowo, Osun State, Nigeria containing Longitude and Latitude, Elevation, Maximum Temperature, Minimum Temperature, Precipitation, Wind speed, Relative Humidity, Solar Radiation and Tropospheric Ozone was collected. The LSTM model was trained with diverse compositions like the number of layers, the number of neurons in each layer, training epochs, and optimization algorithms. Results showed the model had 97%-99% correlation coefficient between actual and predicted values and 99.3%-99.9% prediction accuracy on the test datasets. The LSTM model also forecasted each of the trained variables for a ten-year period between 2015 to 2025 accurately.