基于机器学习的车内VOC分布规律和暴露风险预测
Distribution law and exposure risk prediction of VOC in vehicles based on machine learning
摘要：
对车内VOC进行了实验采样检测,发现车内的主要污染物为甲苯和甲醛。利用蒙特卡罗模拟,评估了车内甲苯和甲醛的暴露风险。在此基础上,采用遗传算法优化后的BP神经网络(GA-BP神经网络)建立了车内污染物浓度预测模型,提高了预测准确性和稳定性。结果表明:根据蒙特卡罗模拟结果,车内甲醛和甲苯的非致癌风险危害商数均小于1,女性的致癌风险大于男性;GA-BP神经网络模型对污染物浓度的预测值与实测值吻合良好,甲苯和甲醛预测模型的精度比未优化前分别提升了33%和41%左右。通过灰色关联度分析了5个影响因子对车内污染物浓度的关联度,结果显示影响程度排序为车内温度、室外温度、车内湿度、光照强度和室外风速。
Abstract：
Experimental sampling and testing of VOC in the vehicle is carried out, and it is found that the main pollutants in the vehicle are toluene and formaldehyde. The Monte Carlo simulation is used to assess the exposure risk of toluene and formaldehyde in the vehicle. On this basis, a BP neural network optimized by the genetic algorithm (GA-BP neural network) is used to establish a prediction model for the concentration of pollutants in the vehicle, which improves the prediction accuracy and stability. The results show that according to the Monte Carlo simulation results, the non-carcinogenic risk hazard quotient of formaldehyde and toluene in the vehicle is less than 1, and the carcinogenic risk is greater in women than in men. The predicted values of the GA-BP neural network model on the concentration of pollutants are in good agreement with the measured values. The accuracy of the toluene and formaldehyde prediction models is about 33% and 41% respectively higher than that of the model before optimization. The gray correlation degree is used to analyse the correlation degree of five influence factors on the concentration of pollutants in the vehicle. The results show that the influence degree is ranked as temperature in the vehicle, outdoor temperature, humidity in the vehicle, light intensity and outdoor wind speed.
Keywords:volatile organic compound(VOC); exposure risk assessment; vehicle; pollutant concentration; carcinogenic risk; machine learning; GA-BP neural network