Research On Meteorological Prediction And Source Parameter Inversion Of Typical Sudden Air Pollution Accidents

In recent years,some typical sudden air pollution accidents（SAPAs;leakage,fire,explosion,respectively）have occurred frequently,which seriously damaged ecological-environment and threated the safety of human life and property.It was important to accurately predict the local scale key meteorological elements,identify unknown accident source parameters（emissions,location）,judge the development and change characteristics of typical emergencies（accident development,pollution emission/diffusion law）,the dispersion of pollution and risk assessment to solve the above problems and realized rapid emergency warning and scientific decision-making and disposal of emergencies.Accurate prediction of local scale wind speed and direction is the basis of accurate prediction of pollutant transport path.Based on the historical meteorological data,the local scale short-term wind speed and direction were predicted by using long short-term memory neural network（LSTM）.The results showed that the wind speed prediction performance of univariable single step LSTM prediction model is better（R=0.86;RMSE=0.47m/s;MAE=0.3m/s）;considering the temporal and spatial correlation between different meteorological elements,the prediction performance of the multivariable single step LSTM model was better than univariable LSTM model,especially for the wind direction,the multivariable model had the better performance for capturing the sudden changes of wind speed and wind direction time series data（R=0.71;RMSE=31.68°;MAE=22.14°）.For the multi-step LSTM prediction model,the distribution probability of absolute error gradually decreased with the increase of wind speed,and was concentrated in the range of 0 m/s～1 m/s,with an average absolute error of 0.31 m/s;in terms of the wind direction,the distribution probability of absolute error also gradually decreased with the increase of wind speed,with an average absolute error of 32.71°.It could be seen that the multistep LSTM model have the poorer performance for capture the sudden changes of wind direction data.Source parameter inversion technique includes three important parts:optimization technique,atmospheric diffusion model and optimization cost function.From the perspective of optimization technique,based on the field experimental data,the source parameter inversion performance of six bio-inspired optimization algorithms was compared in detail.Seeker algorithm（SOA）has the best inversion performance（score<0.0661）and the best computational efficiency（T=1.92s）;when the population size was small,the robustness of SOA was significantly better than other algorithms;based on synthetic data of different signal to noise ratio（SNR）,SOA could maintain a strong anti-noise ability in small population size condition.Under different atmospheric stabilities conditions,Differential Evolution（DE）and Genetic Algorithm（GA）were more suitable for source inversion under unstable conditions,while SOA algorithm was more suitable for source inversion under neutral or stable conditions.From the perspective of atmospheric diffusion model,based on Prairie grass experiment,the forward simulation performance and the inversion performance of three dispersion parameter schemes were systematically evaluated from the perspectives of peak concentration,overall concentration,inversion accuracy and robustness.In terms of forward simulation performance,from the perspective of fractional deviation（FB）and normalized mean square deviation（NMSE）,the overall performance of the National Standard Recommended scheme was the best（peak concentration:FB=0.34,NMSE=1.03;overall concentration:FB=0.36,NMSE=2.14）.According to FAC2,Venkatram scheme performed the best in overall concentration（FAC2=0.44）,In terms of inversion performance,Venkatram scheme has the best accuracy（only source strength is unknown:ARD=（20.97±9.92）%;source strength and location are unknown:ARD=（32.51±11.04）%,AD=（12.86±5.19）m）.The robustness of BRRIGS scheme is the best.In comprehensive view,FAC2 might be the key factor to improve the accuracy of inversion results.From the perspective of cost function,the inversion performance differences of eight typical cost functions were systematically evaluated from the perspectives of accuracy,robustness and overestimation rate.It was found that the inversion performance of different cost functions showed significant difference.In terms of single unknown source parameter inversion（only for source strength（Q））,the cost function based on logarithm transformation scheme had the highest overestimation rate（79.4%）,and that based on the sum of deviation squares scheme exhibited the best accuracy(P_ARD<50%=82.3%,ARD=（35.3±9.1）%);while no significant difference（CV<0.01）was found in the robustness of different cost functions.In terms of three unknown source parameter inversion（source strength（Q）and location（x,y））,the cost function based on normalized root mean square error scheme had the highest overestimation rate（98.5%）for source strength,the logarithm transformation cost function performed the best seen from the perspective of accuracy and robustness(P_ARD<50%=91.1%,ARD=（48.4±9.8）%;CV=0.01);the cost function based on the sum of deviation squares scheme had the highest accuracy for the source location（AD=（36.12±11.39）m）,while the logarithm transformation cost function showed the best robustness（CV=0.0018）.In terms of four unknown source parameters inversion（source strength（Q）and location（x,y,z））,the cost function based on normalized root mean squared error exhibited the highest overestimation rate（98.5%）for source strength,and the logarithm transformation scheme got the best accuracy and robustness（PARD<50%=61.7%,ARD=（55.2±16.5）%;CV=0.03）;the correlation coefficient scheme performed the best in accuracy and robustness for source location（AD=（34.37±10.72）m;CV=0.011）.Generally speaking,the logarithm transformation cost function had the most stable inversion performance with the increase of the unknown source parameters needed to be inversed.This paper further evaluated the uncertainty of source intensity inversion results of different typical objective functions under different downwind distances.The results showed that the uncertainty of source strength inversion results of cost functions have an obvious upward trend with the increasing of downwind distance.If the monitoring concentration data with the same downwind distance was used,the uncertainty of source strength inversion results with different cost functions was obviously different.In the case of single parameter,the fractional deviation cost function performed better near the source（50m）;when using the concentration monitoring data with the downwind distance of 100m,200m and 400m for inversion,the logarithmic transformation cost function performed the best;when using the concentration monitoring data with the downwind distance of 800m,the composite cost function performs the best.In the case of multiple parameters,the fractional deviation cost function performed better near the source（50m）;when the remote monitoring concentration data was used for inversion（100m,200m,400m and800m）,the logarithmic transformation cost function had the best performance.Taking the SF₆ diffusion tracer test in typical urban terrain around Yanshan Petrochemical Plant in Fangshan District of Beijing on July 18,2019 as the research case,the applicability of local scale key meteorological elements prediction technique,source parameter inversion technique and model simulation were verified and evaluated.From the perspective of meteorological prediction,LSTM model was better than WRF model in the prediction performance of wind speed and wind direction during the test period;from the perspective of source parameter inversion,the source parameter inversion technique based on bio-inspired optimization algorithm was better than transfer coefficient method;from the perspective of pollutant diffusion range,The range of pollutant diffusion is more consistent with the actual situation by using the wind speed and wind direction predicted by LSTM model as the driving field.The case study results showed that the local scale key meteorological elements prediction technique and source parameter inversion technology proposed in this study could be used for the analysis and emergency response of sudden air pollution accidents.