基于多维多步长LSTM网络的区域AQI预测研究

Research on regional AQI prediction based on multi-dimensional and multi-step LSTM network

现有空气质量预测方法未同时考虑前期空气质量和气象影响,不能有效体现空气质量的时空依赖性。研究采集2017—2020年重庆市历史气象信息和历史空气质量数据,通过皮尔逊相关性分析,筛选重要的输入变量,构建基于多维多步长LSTM的重庆市空气质量预测模型,并与其他预测模型进行对比分析。结果表明:重庆市AQI与PM_(2.5)、PM_(10)有非常强的正相关性,与能见度、平均温度、湿度、降水量呈较强的负相关性;前7 d的空气质量状态和气象状态对空气质量预测有显著影响,能更准确的预测AQI值。预测损失均方根误差为12.206,平均绝对误差为9.430。

The existing air quality prediction methods do not consider the influence of previous air quality state and meteorological state at the same time,and can not reflect the spatial and temporal dependence of air quality effectively.The historical meteorological elements and historical air quality index(AQI)of Chongqing during 2017 to 2020 were collected.Through Pearson correlation analysis of characteristic variables,important input variables were screened out.The air quality prediction model of Chongqing based on multi-dimensional and multi-step LSTM was established and compared with other prediction models.The results show that AQI in Chongqing has a very strong positive correlation with PM_(2.5) and PM_(10),and a strong negative correlation with visibility,average temperature,humidity and precipitation.The air quality state and meteorological state in the first 7 days have a significant impact on the air quality,and can predict the AQI value more accurately.The root mean square error of loss prediction is 12.206,and the mean absolute error is 9.430.