基于集成学习的O3的质量浓度预测模型

为准确预测O₃的质量浓度及其发展趋势，分析其诱发因素，提出一种基于集成学习的O₃的质量浓度预测模型。以北京市2015—2016年O₃污染物的质量浓度及气象因素数据为基础，提出并建立面向O₃污染物的质量浓度预测的特征选择-集成学习多层预测模型，在对数据进行缺失值填补及异常值分析的基础上，利用Pearson相关分析和Lasso回归分析同时对清理后的气象资料数据进行特征选择，以消除数据冗余，提高预测精度;提出基于自组织映射神经网络self-organizing featuremap, SOFM和Elman神经网络Elman neural network, ENN的集成学习算法，利用SOFM对样本数据进行聚类以实现样本的合理分布后，使用ENN进行仿真训练来预测O₃的质量浓度。试验结果表明：采用Pearson-Lasso特征选择和SOFM样本聚类对数据做前期处理后，ENN的预测精度由74.6%提高到82.1%，能够改善基于ENN的O₃污染物的质量浓度的预测准确率。

An integrated learning approach for O3 mass concentration prediction model

In order to accurately predict O₃ mass concentration and development trend and to analyze inducing factors, an O₃ mass concentration prediction model based on integrated learning was proposed. A multilayer FS-IL model for the O₃ pollutant mass concentration was established in accordance with the data of O₃ pollutant mass concentration and meteorological factors from 2015 to 2016 in Beijing, on the basis of missing value filling and outlier analysis, Pearson correlation analysis and Lasso regression analysis were used to select features of the cleaned meteorological data to eliminate data redundancy and improve prediction accuracy; an integrated learning algorithm based on self-organizing featuremap (SOFM)-Elman neural network (ENN) was proposed. After clustering sample data with SOFM to realize reasonable distribution of samples, ENN was used for simulation training to predict O₃ mass concentration. The experimental results showed that the accuracy of ENN-based O₃ pollutant mass concentration prediction was improved from 74.6% to 82.1% after the preliminary processing of data with Pearson-Lasso feature selection and SOFM sample clustering.