Coronal And Heliospheric Solar Wind Modeling Using Multiple Observations And Artificial Neural Network

In recent years,human society is increasingly affected by space weather.With the advance of space exploration technology and the continuous understanding on the physical process of space weather,an important subject of space weather research arises as integrating more observational data and physical mechanism into numerical model and developing efficient prediction method.The main work of this paper is to study the large-scale structure of solar wind by means of statistical analysis and numerical simulation,and to develop numerical model of coronal and heliospheric solar wind based on multiple observations,so as to improve the level of space weather forecasting.The global distributions of magnetic field and plasma parameters on the source surface are important for coronal and heliospheric modeling.In this paper,a method for determining the global distribution of solar wind density and velocity on the source surface by analyzing in-situ observations is introduced.Then,we mainly describe the establishment of a new method to obtain the self-consistent global distribution of solar wind parameters on the source surface based on multiple observations and neural network tactic.The magnetogram and polarized brightness(pB)observations are used to derive the magnetic field and electron density on the source surface respectively.Then,an artificial neural network(ANN)machine learning technique is applied to establish an empirical relation between the solar wind velocity with both the magnetic field properties and the electron density.The ANN is trained with interplanetary scintillation(IPS)velocity data,and is validated to be more reliable than the WSA model for reconstructing the global distribution of solar wind velocity,especially at high latitudes.The plasma temperature distribution is derived by solving a simplified one-dimensional MHD system on the source surface.Therefore,using this method can obtain the global distribution for all the parameters self-consistently based on magnetogram and polarized brightness observations.Next,a hybrid intelligent source surface model applying the ANN tactic for predicting solar wind speed near the Earth is presented in this paper.The model is a hybrid system merging various observational and theoretical information as input.Different inputs are tested including individual parameters and their combinations in order to select an optimum.Then,the optimal model is implemented for prediction.The prediction is validated by both error analysis and event-based analysis from 2007 to 2016.The overall correlation coefficient is 0.74,the root-mean-square error is 68 km/s,and the probability for detecting a high-speed-event is 68%.Then,a method for predicting the Ap index of geomagnetic disturbance using the nonlinear auto-regression with external input(NARX)neural network is proposed.The external inputs adopt the source surface characteristics derived from solar observations.The prediction result of Ap index from CR2181 to CR2190 shows that the prediction accuracy of the NARX model is higher than the 27-day persistent method.For the magnetic disturbance days with Ap index greater than 10 and 15,both the possibility of detection and false alarm rate of the NARX model are better than the 27-day persistent method.These results further demonstrate that the source surface characteristics based on multiple observations can predict the large scale solar wind condition near the Earth properly.Three-dimensional magnetohydrodynamics(MHD)modeling is a key method for studying the interplanetary solar wind.Finally,this paper develops a new solar wind MHD model driven by multiple observations.The computation region of this model is from 0.1 astronomical unit(AU)to 1 AU.The model solves the ideal MHD equations in a six-component grid system by using the TVD Lax-Friedrich scheme.The boundary conditions of the model are given according to the self-consistent source surface structure based on multiple observations.The model is used to simulate the threedimensional(3D)interplanetary solar wind during CR2062.The simulation results well reproduce the large-scale structure of solar wind and show rich observational characteristics,and is in good agreement with the actual observations from OMNI and Ulysses.Therefore,the model is helpful to provide more realistic 3D interplanetary solar wind.