Air Pollution Anomaly Detection And Source Tracing In Industrial Parks

Industrial parks are densely populated with various types of enterprises and numerous air pollution sources.The air pollutants they generate and emit have a serious impact on the atmosphere environment,threatening people’s property safety and life health.Therefore,it is crucial to detect the abnormal air pollution situation and trace the pollution sources in industrial parks.Based on the ambient air quality monitoring network in industrial parks,this study focuses on air pollution anomaly detection and pollution source tracing methods for the scenarios such as excessive emissions and gas leakage in industrial parks.Air pollution anomaly detection is the first step of air pollution control in industrial parks,which can monitor the abnormal air pollution status in the parks and is the trigger condition for subsequent air pollution source tracing.In the actual pollution control of industrial parks,it is usually judged by whether the measurement of a certain environmental sensor in the park exceeds the environmental standard to determine whether an abnormal situation has occurred.However,in this method,the relationship between the emissions and the ambient sensors is not taken into account,and using uniform environmental standards hardly guarantees that all excessive emissions can be detectable.Furthermore,this univariate detection method does not take into account the correlation between the monitoring stations.Therefore,this thesis proposes an air pollution anomaly detection method based on multivariate statistical analysis,using principal component analysis(PCA)to extract the hidden spatial relationships between historical environmental monitoring data.Then,two monitoring indicators T2 and Q are utilized to monitor the pollution status in the industrial park,and the thresholds of these monitoring indicators are trained using the historical measurements.Simulation experiments show that the method can detect the abnormal pollution conditions caused by excess emissions in the park in real time with a high detection rate.The success rate of anomaly detection in some scenarios is close to 100%.The average anomaly detection success rate is 72.35%,while the average false alarm rate is only 3.7%.Air pollution source tracing is the second step of air pollution control in industrial parks.In this thesis,the source-receptor relationship in industrial parks is first analyzed,and the theory of traceability is proposed to explain when the source parameters(source emission rates,source location,etc.)can be uniquely traced.In industrial parks,the number of ambient concentration sensors is usually far smaller than that of pollution sources,which does not satisfy the traceability condition.Therefore,the pollution sources are untraceable.For the daily pollution whose source location is known,it focuses on real-time estimation of the pollution source strength.Based on the proposed traceability theory,this thesis proposes a real-time source strength estimation method based on a time augment approach,which transforms the untraceable problem into an approximately traceable estimation problem.The results of the simulation experiments show that the proposed method enables the source strength to be estimated for about 88%scenarios,and the solution has a high confidence level.The estimation error of source strength is less than 6%,which solves the untraceable problem brought by sparse sensing in industrial parks.To further reduce the proportion under untraceable cases of the proposed source tracing method,this thesis designs an optimized objective function based on the traceability theory to optimize the sensor layout,to maximize the difference of the coefficient matrix caused by the small variation of wind direction in the dominant wind direction.The results show that the weighted success rate of air pollution episode detection increases from 65.82%to 84.33%,and the percentage of untraceable situations decreases from 12.37%to 4.21%.For sudden gas leakage in the park,mobile sensing devices(e.g.,UAVs,etc.)are more advantageous in the rapid trace of gas leak scenarios due to their flexibility and other characteristics.This thesis proposes a hybrid autonomous source searching approach,named as regression-enhanced Entrotaxis,which incorporates existing Entrotaxis algorithm with the regression methods.In the proposed algorithm,only the position of leak source is inferred by particle filtering,while the strength is estimated by the least squares based on the historical measurements.Through this way,the searching space in particle filtering is converted from 3dimensions(x,y,q)to 2-dimensions(x,y),therefore speeding up the computation significantly.The results of Monte Carlo simulations demonstrate that fewer particles in our method could be used to reach a specific success rate,saving time in the search process.The computation time of this method is nearly 5 times faster than the Entrotaxis algorithm.Besides,the proposed strategy relies on less prior information regarding source strength,and the allowable initial source strength range is at least 2 times larger than that of the Entrotaxis algorithm.In addition,the search area that the proposed method can adapt to is nearly 3 times larger.