基于多源数据的榆林市PM_(2.5)预测模型研究

Study on Yulin PM_(2.5) prediction model based on multi-source data

在PM_(2.5)监测过程中,常出现同一地点、同一排放源浓度监测结果差异较大的情况,受地面环保监测站点数量、分布的限制,很难反映变化情况。本文利用榆林市2018至2020年气象站逐小时数据和环保监测站的PM_(2.5)数据,先对站点进行时空匹配,并与PM_(2.5)进行相关性分析,再通过建立循环神经网络对PM_(2.5)浓度进行预测。结果表明:PM_(2.5)与气象要素线性相关性从强到弱,分别为温度>湿度>气压>风速,相关系数R分别为0.187、0.164、0.151、0.018;基于多源数据的循环神经网络可对PM_(2.5)预测,建立的实验组中,实验组1应用最好,平均决定系数r^(2)为0.836、平均相对误差为0.16,实验组4应用最差,平均决定系数r^(2)为0.537、平均相对误差为0.2。

In the process of PM_(2.5)monitoring, it is often the case that the monitoring results of the same location and the same emission source differ greatly. Due to the limitations of the number and distribution of ground environmental monitoring stations, it is difficult to reflect the change. In this paper, hourly data of Yulin meteorological stations and PM_(2.5)data of environmental monitoring stations from 2018 to 2020 were used to match the stations in time and space, analyze the correlation between the stations and PM_(2.5), and predict the PM_(2.5)concentration by establishing a cyclic neural network. The results show that the linear correlation between PM_(2.5)and meteorological elements is from strong to weak, namely, temperature>humidity>air pressure>wind speed, and the correlation coefficients R are 0.187, 0.164, 0.151, 0.018 respectively;The cyclic neural network based on multi-source data can predict PM_(2.5). In the established experimental groups, experimental group 1 has the best application, with an average determination coefficient r^(2) of 0.836, an average relative error of 0.16, and experimental group 4 has the worst application, with an average determination coefficient r^(2) of 0.537, and an average relative error of 0.2.