Research On Reverse Identification Algorithm Of Pollution Sources Based On Low-Cost Sensor

Environmental pollution and occupational exposure caused by sudden industrial pollutant leakage will seriously endanger human health and environmental ecology,which must be timely and accurately monitored and controlled.Therefore,it is becoming more and more important to obtain the accurate distribution of pollution information and to locate and identify the leakage sources.To reduce the cost of accurate monitoring of pollutants in the industrial field,this paper develops a low-cost general framework of sensor anomaly detection,which includes five basic processes,and can accurately identify three typical data types in various scenarios and automatically match the combination strategy of highperformance algorithms.Specifically,the stable abrupt change data type I adopts wavelet transform combined with LOF algorithm,the irregular oscillation data type II adopts Score algorithm based on statistics,and flat trend data type III adopts LOF combined with ARIMA algorithm.Model verification based on real workshop monitoring data shows that the accuracy of anomaly detection for three typical data types can reach 93.3%,and the performance degradation caused by classification errors can be controlled within 3.52%.In this paper,eight different algorithms combination strategies under the framework are compared.Through parameter optimization and performance evaluation of the model,it is concluded that grid search is effective in determining the optimal matching of parameter combination,and can meet the abnormal detection requirements of low-cost sensors in the industrial field such as working condition change and scene migration.To realize the reverse traceability tracking of industrial workshops with the goal of large-scale regional positioning,this paper builds an abnormal leakage identification model based on sensor network.Field surveys show that the fluctuation of oil mist concentration in the machining workshop has a daily periodic pattern,and model training for a deployment location can be extended to the entire sensor network.The evaluation of the effect of trend fitting shows that the running trend of the monitoring data can be used as a basis for model training,and it is recommended to use the ARIMA model to fit the details of fluctuations.Box plot analysis can be used to predict the upper and lower limits of pollutant fluctuations in industrial workshops.The length of the sliding window and the leak criterion are two important parameters.The missed detection rate and the misjudgment rate should be weighed in conjunction with the specific requirements of leak detection to complete the model adjustment.This study carried out model verification on real workshop data sets,and the results showed that the upper and lower limits of the prediction were relaxed as the length of the sliding window increased,thereby increasing the risk of missed detection.Using a single point of data exceeding the upper limit of the prediction as a leakage criterion could easily lead to misjudgments and false alarms.To improve the accuracy of traceability and location,this research built a machine learning-based reverse identification model of fixed pollution sources.Sensors deployed near the wall can collect the concentration distribution that meets the discrimination required for source identification.In order to break the limitation of the high computational cost of the traditional forward method,eigen-orthogonal decomposition can be used to quickly obtain a large number of intermediate working conditions to achieve data set expansion.It is recommended to use the order of normalization first and then random forest feature selection to improve the accuracy and generalization ability of the model.The research shows that the incomplete enumeration method can effectively optimize the sensor deployment scheme,reduce the cost and test workload.Four sensors deployed in a specific position can achieve positioning accuracy of over 95%.In addition,the leak source determination sequence developed in this paper can effectively guide the staff to check the leaks in turn,and the efficiency of the three checks is as high as 99.91%.