TPE-XGBOOST与LassoLars组合下PM2.5浓度分解集成预测模型研究

在我国当前大气重污染的环境下,PM2.5浓度的预警预报工作显得尤为重要.由于PM2.5浓度时间序列具有高度复杂性与随机性等特点,且传统的PM2.5浓度分解集成预测方法没有考虑空气质量因素与气象因素的信息,仅靠PM2.5浓度的历史值难以准确对其精准预测.本文在对历史数据的分解下,对高频数据引入TPE-XGBOOST模型,对低频数据引入LassoLars模型,结合空气质量因素与气象因素反映分解特征的变化趋势,对PM2.5浓度时间序列展开预测研究.通过实验,该组合模型显示出了良好的预测效果,且相对于单一分解集成预测模型有较大的预测精度提升.

An integrated prediction model of PM2.5 concentration based on TPE-XGBOOST and LassoLars

Because of serious atmospheric pollution, the early warning and forecasting of PM_2.5 concentration is particularly important. Due to the high complexity and randomness of the time series of PM_2.5 concentration, the traditional integrated PM_2.5 concentration decomposition prediction method does not take the air quality and meteorological factors into account. Thus, it is difficult to predict the PM_2.5 concentration accurately only by the historical value. By decomposing the historical data, this paper introduced the TPE-XGBOOST model for high-frequency data and LassoLars model for low-frequency data, combined air quality and meteorological factors to reflect the variation trend of decomposition characteristics, and made prediction for the time series of PM_2.5 concentration. Through the experiment, the model shows good prediction effect, and has higher prediction accuracy compared with the single decomposition integrated prediction model.