Machine Learning-based VOC Distribution Pattern And Exposure Risk Prediction In Vehicles

In the context of motor vehicles becoming a consumer product in common households,the health risks of in-vehicle air pollution should be taken seriously by the majority of drivers and passengers.There are limitations in the current research on the safety of in-vehicle air,so it is important to assess the risk of inhalation exposure to in-vehicle pollution,accurately predict pollutant concentrations,and identify hazards in advance for the health of drivers and passengers.As a new technology,machine learning is widely used in various scenarios and also provides new ideas for predicting in-vehicle air.In order to predict the exposure risk and concentration of VOCs in the car,based on the above paper,we did the following two parts:first,the experimental part.Firstly,toluene and formaldehyde in the car were selected as the main targets of this study,after which CO₂ was used as the tracer gas to calculate the number of air changes in the car,and then phenol reagent spectrophotometry was used to measure the concentration of formaldehyde in the car,and a standard curve for the determination of formaldehyde was produced.The detection rate of formaldehyde and toluene in the car was 100%,and comparing the above two kinds of pollutants in the car can get the concentration of toluene in the car is greater than the concentration of formaldehyde in the car,and none of them exceeded the prescribed limit.The exposure risk of in-vehicle pollutants was evaluated using Monte Carlo simulation,and the results showed that the non-carcinogenic risk of in-vehicle formaldehyde and toluene were both less than 1,while the carcinogenic risk of formaldehyde was in the range of 1×10^-6-1×10^-4;the carcinogenic risk of women was greater than that of men.It was concluded from the sensitivity analysis that exposure time was the largest factor affecting the non-carcinogenic risk;the carcinogenic potency factor of pollutants in the car contributed the most to the carcinogenic risk,followed by pollutant concentration.After that,based on BP neural network,a regression prediction model with the then outside temperature,inside temperature,inside humidity,outside wind speed and inside light intensity data as input information and pollutant concentration as output information was established.Using the detected 84 sets of data,75%of them are used as training data and 25%as validation set.Using Sigmiod as the activation function,the analysis of the prediction results showed that the MAE of the BP neural network in the prediction results ranged from0.00937-0.02463 and 0.00223-0.00329.The accuracy of the toluene prediction model was improved by about 33%and the accuracy of the formaldehyde prediction model was improved by about 41%by using the genetic algorithm to optimize the BP neural network compared with the accuracy of the model before optimization.Using the gray correlation method,the correlation coefficients of the five influencing factors on the pollutants in the vehicle were calculated,and the results showed that the temperature in the vehicle had the greatest influence on the pollutants.Finally,an autoregressive sliding average prediction model（ARMA）for future and past data points of in-vehicle pollutants was established,and the differentially processed time series data were used as input data for calculation after smoothness analysis.The best value of the model was found using the AIC criterion,and the prediction models with the values of p and q taken as 11 and 13,respectively,were established,and the fit to the estimated data was 96.12%.After that,the untrained data were called out of the model for actual prediction,and the root mean square error of the quantified result model was 3.841×10^-6,and the prediction target accuracy of the training data reached 98.11%,which verified that the model could better achieve the prediction of pollutant concentration in the vehicle in a short time.