基于聚类分型的随机森林O3浓度预测研究

Research on prediction of O3 concentration in random forest based on clustering classification

摘要:

近地面O3是光化学污染的主要成分之一，对于人体健康以及生态系统具有较大危害。为了准确预测上海市O3浓度的变化情况，本研究基于上海市2014—2020年6种空气污染物浓度的监测数据以及同期的天气预报数据，提出了一种经模糊C均值聚类算法优化的随机森林O3浓度预测模型。首先利用互相关分析的方法筛选出两个聚类因子，随后利用模糊C均值聚类算法将O3浓度分为三种类型，最后利用随机森林建立O3浓度预测模型，对比聚类前后的预测效果。研究结果表明，前1日的O3浓度和PM10浓度对预测日的O3浓度影响最大，且O3浓度变化受月份的影响明显。经模糊C均值聚类之后，O3_8h浓度预测结果的平均绝对误差和均方根误差分别减小了10.5％和8.8％。随机森林提升了O3浓度的预测效果，且聚类后模型的决定系数R2增加，说明该模型对上海市O3污染预测具有较高的实际应用价值。

Abstract:

Ground-level O3 is a major component of photochemical pollution and poses significant risks to human health and ecosystems. To accurately predict O3 concentration variations in Shanghai, this study proposed a random forest prediction model for O3 concentrations optimized through the fuzzy C-means clustering algorithm, based on monitoring data of six air pollutant and weather forecast data from 2014 to 2020 in Shanghai. Firstly, two clustering factors were selected using cross-correlation analysis. Then, O3 concentration was categorized into three types using the fuzzy C-means clustering algorithm. Finally, a random forest model was established to predict O3 concentration, and the predictive performance before and after clustering was compared. The results show that the O3 concentration and PM10 concentration of the previous one day have the greatest influence on the O3 concentration of the prediction day, and O3 concentration variation is notably affected by the month. After fuzzy C-means clustering, the mean absolute error and root mean square error of the predicted O3_8h concentration decreased by 10.5% and 8.8%, respectively. The random forest model improves the accuracy of O3 concentration prediction, and the coefficient of determination R2 increases after clustering, demonstrating that this model has high practical value for predicting O3 pollution in Shanghai.