SOLAR IRRADIANCE FORECASTING USING GLOBAL POSITIONING SYSTEM (GPS) DERIVED TOTAL ELECTRON CONTENT (TEC)

Abstract

Solar energy is one of the most significant energy sources and the only potential energy source capable of providing the world's required extra energy over the next few decades. Due of its intermittency because of weather variations, the integration of renewable energy (RE), such as solar energy, into the electrical network is a challenge for grid operators. Conversely, the installed capacity of solar photovoltaic (PV) globally continues to rise. In Malaysia, the average monthly daily solar radiation is 4,000-5,000 W/m², with the average monthly sunshine duration ranging from 4 to 8
h. Thus, forecasting is becoming an effective resource for network grid operators to control the output of solar photovoltaic (PV) energy. Solar radiation measurement will decrease when ionosphere total electron content (TEC) decreases. This is because free electrons forming in the ionosphere are strongly dependent on the solar radiation. This study aims to investigate the interaction between TEC and solar irradiance for further use in solar irradiance forecasting. In order to obtain the TEC, GPS data was extracted in order to substitute into the calculations. The interaction between TEC and solar irradiance was done using neural net fitting. The overall correlation coefficient, R obtained is 0.91, which indicates a good fit. In addition, future TEC values were predicted using the auto regressive integrated moving average (ARIMA) and long short-term memory (LSTM) models. The predicted TEC values were further fed into the trained neural network to forecast future solar irradiance. In order to compare the accuracy of model, root mean square error (RMSE) was used to evaluate the results. It was found that the ARIMA model is better for solar irradiance forecasting as it has lower RMSE as compared to the LSTM model.
Keywords: Global Positioning System (GPS); ionosphere; total electron content (TEC); solar irradiance; time series prediction.

Description