Forecasting Kp Index Using a Hybrid Machine Learning Model Based on Random Forest and Sequence-to-sequence

In this work, we develop a Kp forecasting model based on the random forest and the sequence-to-sequence. Our model forecasts the Kp index every 3 hours in advance through two phases. In the first phase, the model predicts that the geomagnetic storm will occur (storm: Kp≧5) or not (non-storm: Kp <5) using the random forest. In the second phase, the model forecasts the Kp index using the random forest for storm cases or the sequence-to-sequence for non-storm cases. For input data we use solar wind parameters (proton density, proton velocity, Bz and total B) from ACE, preliminary Kp from NOAA, the Boyle index, and the dynamic pressure. We divide chronologically these data into the training data (2000~2012) and the test data (2013~2014). Major results of this study are as follows. First, our model successfully forecasts the Kp index. Second, our model shows better results than other machine learning algorithms such as CNN, simple LSTM etc. Also, our results are better than those of Tan et al.(2018) who used the LSTM and a simple regression method. Third, statistical scores for test datasets are as follows: 0.57 for root-mean-square-error, 0.43 for mean-absolute-error, and 0.87 for correlation-coefficient.

This work was supported by Institute for Information & communications Technology Promotion(IITP) grant funded by the Korea government(MSIP) (2018-0-01422, Study on analysis and prediction technique of solar flares).