基于机器学习算法的西部方向气候模式预测订正研究

Prediction correction of western climate model based on machine learning algorithm

基于气候预测对西部方向环境保障的重要性,针对高原地区气候模式准确度不高的现实困境,采用大数据挖掘技术,充分处理气候系统非线性统计特征。首先利用随机森林,对气候模式融合网格数据进行订正;而后将订正网格进行EOF分解,采用信息流算法挖掘环流因子与时间序列因果关系,构建不同模态下高影响因子集;采用随机森林进行建模,预报不同模态的时间序列;最后还原预报的格点场,完成模式格点数据修订。结果表明,经机器学习算法修订后的气候模式预报准确度、预报技巧显著提高,同时,模型预报的稳定度也有较大提升。本研究基于机器学习算法进行气象大数据挖掘,提升气候模式预测效能,旨在为提升西部方向气候预测水平提供方法思路。

Based on the importance of climate prediction to support the battlefield environment in the western,and aimed at the realistic dilemma of low accuracy in plateau-climate model,this paper adopts big data mining technology to fully deal with the nonlinear statistical characteristics of the climate system.Firstly,the random forest is used to correct the data of climate model fusion grid.Then,EOF is used to analyze the corrected grid,and the information flow algorithm is also used to mine the causal relationship between circulation factors and time series,in order to construct the high-impact factor subsets in different modes.Finally,it models with random forest predicts time series of different modes,then restores the predicted grid field and completes the revision of model grid data.The results suggest that the forecasting accuracy and skills of modified climate model by machine learning algorithm have been significantly improved,as well as the stability of model prediction.This research based on machine learning algorithm for big data mining improves the efficiency of prediction model.It Aims at providing methods and ideas for improving the level of climate prediction in the western.