Comparison of ARIMA, LSTM, BiLSTM and GRU for Prediction Model of S4 over Kototabang-Indonesia Faruk Afero (a*), Falin Wu (b*), Varuliantor Dear (*c)
a) SNARS Laboratory, School of Instrumentation and Optoelectronic Engineering
Beihang University
Beijing, P.R.China
*farukafero[at]buaa.edu.cn
b)SNARS Laboratory, School of Instrumentation and Optoelectronic Engineering
Beihang University
Beijing, P.R.China
*falin.wu[at]buaa.edu.cn
c) Research Center for Space Science
National Research and Innovation Agency
Bandung, Indonesia
*varuliantor.dear[at]brin.go.id
Abstract
An accurate prediction of time series data is important for space weather information services including for forecasting ionospheric scintillation. Scintillation activity usually increases with increasing solar activity. During periods of maximum solar activity, the ionosphere structure usually changes dramatically which will significantly affect radio wave propagation. It is challenging to build a prediction model of ionospheric scintillation under extreme space weather conditions. In this paper, we conduct a comparison study of scintillation prediction using Autoregressive Integrated Moving Average (ARIMA), Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU). In order to take a challenge, a month data of scintillation over Kototabang-Indonesia (0.3S -100.35E) on January 2014 which the strong solar activity was happening is used as a training dataset. A qualitative comparison between the predicted output and the actual future was conducted to evaluate the model performance. The prediction results using these four methods show a reasonable scintillation model and BiLSTM has the best performance among the others methods. These results indicate that BiLSTM has the potential to be applied to predicting the scintillation by considering the lowest RMSE and MAE are 0,0900 and 0,1011 respectively.
Keywords: ARIMA, BiLSTM, GRU, Ionospheric Scintillation, LSTM, Machine Learning, Space Weather, Time Series Prediction.
