基于二分类Logistic和多元回归的新疆电线积冰及厚度预报研究

Research on the Prediction of Wire Icing and Ice Thickness in Xinjiang Based on Binary Logistic Regression and Multivariate Regression

摘要:

利用新疆1985-2015年电线积冰及相关气象数据，基于多元回归方法，建立新疆电线积冰预报模型。通过分析，得出以下结论：（1）新疆电线积冰日数整体上北疆多于南疆，山区多于盆地，其中沿天山一带（包括东、西、中沿天山一带）最多。最大标准积冰厚度北疆北部最大；其次是南疆西部。塔里木盆地南缘积冰日数和最大标准冰厚都是全疆最小区域。（2）利用二分类Logistics回归模型对电线积冰是否发生进行预测，准确率达89.49%，模型效果较好；再利用多元回归模型对积冰厚度进行预测，典型积冰天气过程个例积冰厚度预测平均绝对误差为0.98mm，模型效果较好。（3）重现期推算结果显示新疆冰区呈现北疆北部地区，塔城地区，沿天山中、西部区域，南疆偏西地区积冰厚度较大；塔里木盆地及周边、东疆大部积冰厚度较小的分布特征。

Abstract:

Based on the wire icing data and related meteorological data in Xinjiang from 1985 to 2015, a wire ice accumulation forecasting model for Xinjiang was established based on the multiple regression method. Through analysis, the following conclusions were drawn: (1) Xinjiang’s wire icing days are generally more in northern Xinjiang than in southern Xinjiang, and more mountainous areas than basins, with the most along the Tianshan Mountains (including the east, west, and central Tianshan Mountains). The maximum standard ice accretion thickness is the largest in the northern part of northern Xinjiang; followed by the western part of southern Xinjiang. The number of days of ice accumulation and the maximum standard ice thickness on the southern margin of the Tarim Basin are the smallest areas in Xinjiang. (2) The two-class Logistics regression model was used to predict the occurrence of ice accretion on electric wires, the accuracy rate is 89.49%, and the model effect was good; then multiple regression model was used to predict the thickness of ice accretion. The average absolute error of thickness prediction of the typical ice accretion weather process was 0.98mm, and the model effect was good. (3) The calculation results of the return period show that the ice thickness was big in the northern part of northern Xinjiang, the Tacheng area, the central and western regions along the Tianshan Mountains, the western part of southern Xinjiang; the ice thickness was small in the Tarim Basin and surrounding areas, and most of the eastern Xinjiang.