摘要
采用最大重叠小波分解与重构方法,将影响O3小时浓度的不同时间尺度的物化过程分离出来,以提高序列的光滑性.同时,选择合适的传统时间序列模型(如ARIMA模型等)来描述不同过程的序列特征,并分别拟合预报.最后,在建模中引入24h季节项,以实现提前24h一次性预测未来1d的O3逐时浓度.结果表明,预报的平均相对误差为12.92%,平均绝对误差和均方根误差分别为10.04μg·m-3和13.98μg·m-3,预报值与实测值的相关系数和匹配指数分别为0.96和0.98.随着预测期的延长,预报误差仍处于可接受范围内.该方法同样适用于每日最大O3小时浓度预报,研究结果为发布天气预报式的空气质量预报提供了新思路,便于公众规划出行并减少大气污染暴露.
It is well known that different dynamic processes in different time scales governs O3 concentration. Time-series of hourly concentration of O3 was decomposed into different time-scale components using maximum overlap wavelet transform. These components could be fitted by proper time series models （ARIMA, etc）. In this study, 24-hour season module was introduced to obtain 24-step-ahead prediction of O3 hourly concentration in a day. The mean absolute error, mean absolute percentage error and root mean squared error of the predictions are calculated as 12.92%, 10.04 μg·m^-3 and 13.98 μg·m^-3, respectively. The correlation coefficient and index of agreement could reach 0.96 and 0.98. Additionally, the result of model performance is still reasonable as the prediction time extends. This model is also able to predict the daily maximum concentration of O3. In this way, this study provides a new method in air quality forecast to help people plan their outdoor activities and avoid air pollution exposure.
出处
《环境科学学报》
CAS
CSCD
北大核心
2013年第2期339-345,共7页
Acta Scientiae Circumstantiae
基金
山东省东营市生态文明战略规划研究项目~~
