基于DWT-GRU模型的天津市NO2浓度预测研究

Study on NO Concentration Prediction in Tianjin Based on DWT-GRU Model

为达到准确预测城市空气污染浓度的目的,文章通过结合离散小波分解和门控循环单元构建集成神经网络模型(DWT-GRU),利用2014年1月1日-2019年6月30日的6种主要大气污染物浓度数据及同期的气象信息进行训练,获得最优模型结构,从而实现对未来1 d天津市NO2日均浓度的预测。首先将输入数据利用离散型小波变换分解信号,提高输入变量的数据维度,然后利用GRU进行特征学习,通过与其他模型进行对比,表明组合模型能够提供更高的预测精准度和更好的泛化能力。最后分析了政策引导、产业发展、居民生活对天津市NO2浓度的年度变化及季节性变化的影响。

This paper elaborates on the building of a high precision and stability concentration prediction model to study the daily average concentration change of NO2 in Tianjin City. For this purpose, concentration data of six major atmospheric pollutants were collected from January 1 of 2014 to June 30 of 2019, as well as the meteorological information during the same period. An integrated neural network model of the optimal model structure based on the discrete wavelet decomposition and the door control cycle units(DWT-GRU) was constructed, which was used for predicting the next day’s daily average concentration of NO2 in Tianjin. The procedure went as follows: firstly, the input data was decomposed into signals by using discrete wavelet transform to improve the data dimension of the input variables;then GRU was used for feature learning, and compared with other models,it proved that this model was capable to provide higher prediction accuracy and better generalization ability;and finally, influence of the policy guidance and industrial development, as well as the city residents’ livelihood on annual and seasonal changes of NO2 concentration in Tianjin was analyzed.