融合时空特征的PCA-PSO-SVM臭氧(O3)预测方法研究

针对目前臭氧预测方法未能考虑臭氧污染的区域性和在时间周期内的强自相关性的问题，提出一种融合时空特征的PCA-PSO-SVM臭氧组合预测模型.利用小波分析和系统聚类提取臭氧时间序列波动特征和站点空间分布相似性特征，并通过主成分分析和粒子群算法优化的支持向量机组合模型（PCA-PSO-SVM）对臭氧日最大8h平均浓度进行预测，以2016~2018年杭州市大气污染物观测数据和气象数据进行实验验证.结果表明：融合时空特征的PCA-PSO-SVM模型预测精度有较大提升，与未融合时空特征的PCA-PSO-SVM模型相比，精度提升19%.气象因素中温度对臭氧预测效果影响最大，在气象预报数据存在一定误差的情况下，提出的模型仍得到较高精度的预测效果，具备较好的鲁棒性.

Research on PCA-PSO-SVM ozone prediction considering spatial-temporal features

The current ozone predicting method usually lacks the effect of the spatial covering of ozone pollution coupled with its strong self-correlation within a certain period. To compensate such a deficiency, a PCA-PSO-SVM based model of ozone combining prediction considering spatial-temporal features was proposed. Using wavelet analysis and system clustering, the fluctuation characteristics in time series and spatial distributing similarity of the ozone was extracted, and the maximum daily 8-hour average concentration of the ozone was predicted resorting to PCA-PSO-SVM model which composes of the principal component analysis (PCA) and particle swarm optimization based support vector machine (PSO-SVM). The model was verified by the experiment with the historical data of atmospheric pollutants and meteorological situation from 2016 to 2018 in urban Hangzhou. The results showed that the ozone predicting accuracy by the PCA-PSO-SVM model considering spatial-temporal features was significantly improved. Compared with the PCA-PSO-SVM model without spatial-temporal features, the predicting accuracy was raised 19%. The experiment also proved that the temperature exerts the largest influence among all the meteorological factors on the ozone prediction. The proposed model was showed its robustness to obtain high predicting accuracy even in case of the weak weather forecast.