区域VOCs聚集态势RF-LSTM智能感知方法

Regional VOCs gathering situation RF-LSTM intelligent sensing method

为了提高VOCs质量浓度预测精度,实现VOCs聚集态势感知,采用RF-LSTM方法提出了基于浓度预测的VOCs聚集态势感知法,简称聚集态势感知法,该方法将态势感知的概念引入VOCs研究,将区域VOCs聚集态势直观展示出来。首先在区域网格划分的基础上利用距离平方反比进行空间插值,收集区域VOCs数据信息;其次利用随机森林结合长短时记忆神经网络对网格VOCs质量浓度进行预测;最后根据预测结果计算VOCs聚集态势值,并将态势感知结果可视化。以西安市某区为例进行VOCs质量浓度预测及VOCs聚集态势感知,结果表明:与RF模型、LSTM模型相比,RF-LSTM模型减少了输入变量,实现了VOCs质量浓度预测模型输入参数的优化,降低了预测模型的复杂度,提高了预测精度,得到RF-LSTM模型的平均绝对误差、均方根误差、平均绝对百分比误差分别为6.24、9.75、10.36%;VOCs聚集态势感知能够对区域VOCs聚集的发展趋势和状态进行可视化,传达了更多的信息,具有一定的实用价值。因此,该聚集态势感知方法可以为区域VOCs污染防治和预警提供决策支持。

To improve the prediction accuracy of VOCs concentration and obtain the situational awareness of regional VOCs gathering,the RF-LSTM method is used to develop a concentration prediction-based VOCs gathering situational awareness method called the gathering situational awareness method.The aggregation situation of VOCs is visually displayed.First of all,the study area is divided into grids,and then the inverse square of the distance is used for spatial interpolation and the data information of all grid VOCs in the area is collected;secondly,the random forest is used for obtaining feature importance ranking and selection,and the long-and short-term memory neural network is used for predicting the grid VOCs concentration;finally,the VOCs aggregation situation value of a single grid and the entire area is calculated according to the prediction results,and the situation awareness results are visualized with the help of ArcGIS software and situation trend diagrams.Taking a district of Xi’an as an example,the prediction of VOCs concentration and the situational awareness of VOCs aggregation is carried out.The results show that:Compared with the RF model and the LSTM model,the RF-LSTM model has the best VOCs concentration prediction accuracy,and the average absolute error of the RF-LSTM model is the root mean square error and the average absolute percentage error were 6.24,9.75,and 10.36%,which were reduced by 14.05%,11.60%,and 8.80%compared with the RF model,and reduced by 21.01%,17.10%,and 14.59%compared with the LSTM model.VOCs gathered situational awareness can visualize the development trend and status of regional VOCs gathering from the two dimensions of time and space,conveying more information and having certain practical value.Therefore,the cluster situation awareness method can provide decision support for regional VOCs pollution prevention and early warning.