异常值对钟差预报模型的影响分析

Influence of Outliers on Clock Difference Prediction Model

为了分析异常值对钟差预报模型的影响程度,首先,通过对钟差数据进行绘图分析来识别钟差数据中的异常值;其次,利用中位数法(MAD)和一种基于中位数的小波阈值法钟差数据预处理策略(WMAD)分别对钟差数据中的异常值进行处理;最后,利用处理前、后的钟差数据建模预报钟差,并分析各模型预报的效果。结果表明:较小的异常值对二次多项式和灰色模型2种模型预报的效果影响不大,但会影响时间序列、卡尔曼滤波及小波神经网络3种模型的预报精度。

In order to analyze the influence of abnormal value on clock bias prediction model,first of all,it identifies the outliers in the data through the study of the drawing of clock bias data,then uses the method of Median Absolute Deviation(MAD)and a kind of wavelet threshold method based on Median Absolute Deviation data preprocessing strategies(WMAD)to deal with the outliers in the clock bias data.Finally,the clock bias data before and after processing are used to model and predict the clock bias,and the prediction effect of each model is analyzed.The results show that small outliers quadratic polynomial(QP)and gray model(GM(1,1))have little influence on the prediction effect,but will affect the accuracy of time series(ARIMA),kalman filter(KF)and wavelet neural network(WNN)models.