利用BP神经网络预测AE指数

采用的预报模式是一种全连接的BP网络模型,利用太阳风及行星际磁场的观测数据预报AE指数.神经网络输入选用ACE卫星数据,取5 min平均值,通过比较,选用4个预报参量.构造了预报参量时续为20 min,40 min和60 min依次递增的三个网络,分别进行训练和预测,并对行星际参量对AE指数影响的时续性进行了探讨.预报结果表明,全连接BP神经网络在AE指数的短期预报中是比较有效的,同时还提出了需要进一步改进的环节.

Forecasting Auroral Electrojet Activity With BP Neural Networks

Magnetosphere substorm is the result of couplings between solar wind and magneto- sphere.In general,as AE,AL etc indices are used to inspect the turbulence in polar region when sub storms occur,they are the target indices in space environment prediction.The AE index is predicted by data of solar wind and interplanetary magnetic field(IMF),with a back-propagation neural network,which is all-joint one.The data of input comes from ACE satellite,combined into 5 min resolution.There are 4 input variables:The B_y,B_z components of interplanetary magnetic field,the velocity of solar wind and the density of solar wind proton.The 3 networks with the length of input time series of 20,40 and 60 minutes are constructed,trained and train them separately. Then the time series of variables on influencing the AE index is discussed.The predictions show that our network model can forecast the trend of fluctuation of AE index,having wonderful veracity in quantitatively computing index value.The correlation between four input variables(B_y,B_z,v,n) and AE index is very good.As the time extends,the forecasting precision needs some improvements.