Kp Index Prediction Based On Similarity Algorithm Of Machine Learning

Geomagnetic storm is an important disastrous space environmental disturbance event,which has an important impact on satellite systems,communication systems,navigation systems,and power systems.The solar wind is the direct cause of geomagnetic storms.The timely and accurate analysis and judgment of the solar wind and forecasting of possible geomagnetic storms can enable us to reduce the damage caused to humans by the severe disturbance of the space environment to a greater extent.This paper constructs two Kp index short-term forecast models.The first model uses XGBoost algorithm,single-layer perceptron and similarity algorithm which is widely used in machine learning,to find similarities from the time series changes of characteristic parameters by solar wind data and interplanetary parameter data accumulated by the Ace satellite.And then,guide the geomagnetic Kp index forecast in the next 1-3 hours,through historical similar cases’ follow-up impact.So that it can carry out the geomagnetic storm warning.With the aid of the Akasufo solar windmagnetospheric energy coupling function,Model 2 uses the solar wind and interplanetary data and the Akasofu coupling function to calculate the energy input from the solar wind to the Earth’s magnetosphere,and calculates the similarity with the input energy to find historical solar winds with similar effects to the earth..Furthermore,the Kp index forecast is guided by the follow-up influence of similar historical cases.According to actual business needs,the two models can be combined for mixed recommendation.Two models were used to test 40 solar wind cases that caused large magnetic storms,40 solar wind cases that caused small magnetic storms,and 40 solar wind cases that were not affected by geomagnetic storms randomly selected from 1998 to 2019.The recommended results show that both two models can successfully recommend historical solar wind cases that have similar geomagnetic effects to the input solar wind.The correlation coefficients between the predicted values of the two models and the measured values are 0.93 and 0.84.The root mean square errors of the predicted results are 0.79 and 1.38,,and the average absolute errors of the predictions are 0.65 and 1.14.Different from the traditional forecast model,the recommended model in this paper can not only give the geomagnetic Kp index after the recommended solar wind event as a forecast output,but also has better correlation and fewer input parameters,and it is no longer limited to artificial neural networks’ black-box process.It can give a clearer and more intuitive comparison of historical solar wind and input solar wind characteristic parameters in time series,so that forecasters can better integrate their own forecasting experience into the forecast.