基于多元分解的大气污染深度学习预测方法

Deep Learning Prediction Method of Air Pollution Based on Multivariate Decomposition

为了有效地控制和治理大气污染,合理预测污染物在大气中浓度,对于提前采取预防措施、有效管理污染活动发挥着重大作用。针对多变量非线性、复杂的时间序列,以及多因素影响预测浓度的问题,提出一种基于多变量分解的非平稳时间序列深度预测方法。首先,确定主要预测变量,并对主变量进行STL(Seasonal and Trend decomposition using Loess, STL)分解得到3个分变量;其次,通过皮尔逊相关系数得出与主变量较为相关的几个相关变量;最后,对3个分变量和其余相关变量进行CNN+BiLSTM混合网络预测并融合得到主变量预测值。通过对太原环境监测站提出的工业园区的大气数据进行实验,预测结果表明,上述预测模型对工业园区大气污染预测模型的优化,并为工业园区大气环境污染防治对策的制定提供准确、及时的数据依据。

In order to effectively control and control air pollution, predicting reasonably the concentration of pollutants in the atmosphere plays an important role in taking preventive measures in advance and effectively managing pollution activities. Aiming at the problem of multivariable nonlinear and complex time series and the influence of multiple factors on the prediction concentration, this paper proposes a non-stationary time series depth prediction method based on multivariate decomposition. Firstly, the main predictive variables were determined, and Seasonal and Trend decomposition using Loess(STL) was applied to decompose the main variables into three variables;secondly, the Pearson correlation coefficient was used to obtain several related variables which were more related to the main variable;finally, the three variables and other related variables were predicted by CNN+bilstm hybrid network, and the predicted values of the main variables were obtained by fusion. Through the experiment of the atmospheric data of Industrial Park proposed by Taiyuan Environmental Monitoring Station, the prediction results show that the prediction model optimizes the prediction model of air pollution in Industrial Park, and provides accurate and timely data basis for the formulation of prevention and Control Countermeasures of air pollution in industrial park.