Research On Prediction Method Of Pollution Concentration Of Mobile Pollution Sources

The mobile pollution source refers to the mobile pollution source whose position changes with time and space,and mainly includes a large amount of harmful exhaust gas discharged from the motor vehicle,engineering vehicle,mobile construction machinery,ship and aircraft during the working process.Air pollution,especially ultra-fine particulate matter and volatile organic compounds(VOCs)in mobile pollution sources,has become one of the most prominent environmental problems in China.Therefore,predicting the concentration of atmospheric pollutants in advance is the basis for strengthening the prevention and control of atmospheric pollution and achieving comprehensive environmental management,which is of great significance to people's daily health and government decision-making.Based on the study of the temporal and spatial characteristics and prediction methods of air pollutant concentrations at home and abroad,the methods of neural network,extreme learning machine and deep learning are studied in depth.Different prediction conditions are explored to establish multiple concentration prediction models to improve the accuracy of concentration prediction.The evolution trend of pollutant concentration was systematically analyzed and experimentally studied.(1)Due to the inherent complexity of contaminant concentration sequences,it is difficult to describe the evolutionary trends and accurate predictions of contaminant concentrations.In order to grasp the time distribution law of air pollutant concentration and realize the accuracy and real-time of prediction,this paper proposes self-adaptive neuro-fuzzy weighted extreme learning machine(ANFIS-WELM)based on the regularization strategy of extreme learning machine(ELM).Improve the performance of a neuro-fuzzy learning machine.The proposed ANFIS-WELM has the advantages of reducing randomness,reducing computational complexity and better generalization.By using the fuzzy membership function combined with the explicit knowledge representation,the parameters of the ANFIS-WELM fuzzy layer are randomly selected,and the constraint optimization problem is solved in the weighted extreme learning machine regularization strategy to determine the parameters of the model neural layer.Experiments show that compared with GA-BPNN,SVR,ELM,WELM,ANFIS,and R-ELANFIS models,the proposed method achieves a good balance between prediction accuracy and real-time performance,and online monitoring of self-developed mobile pollution sources.The engineering application of the algorithm package is completed on the data center software system to achieve multi-scale prediction of pollutant concentration on the time axis.(2)Existing research methods fail to effectively extract the temporal and spatial characteristics of air pollutant concentration data,fail to effectively simulate long-term dependence,and most ignore the spatial correlation analysis of atmospheric impact factors.This paper proposes spatiotemporal attention convolution long short-term memory network model(Attention-CNN-LSTM),which analyzes the historical air pollutant concentration of the current station and nearest neighbors,and analyzes the Granger causality between the nearest neighbors.Relationship and design a hyperparametric Gaussian vector weight function to determine the spatial autocorrelation variable as part of the input feature,and then use the convolutional neural network(CNN)to extract the temporal and spatial features of the data used by the LSTM network,while the attention mechanism is respectively used to weight feature maps and channels to enhance the effectiveness of features,in addition to automatically extracting inherent useful features in historical air pollutant data,and incorporating meteorological data and timestamp data into the proposed model to improve the model predict performance.Experiments show that the proposed method has good stability and predictive performance,and is superior to ARIMA,SVR,LSTM,CNN and CNN-LSTM models.