A NEURAL-NETWORK-ALGORITHMS-BASED MODEL FOR ASSESSING THE QUANTITY AND CONCENTRATION OF FINE EMISSIONS OF HARMFUL SUBSTANCES FROM ROAD TRANSPORT

Abstract

The transport sector is one of the main sources of emissions of harmful substances into the atmosphere in large densely populated cities. Solid particles (PM) from car exhaust gases have a strong impact on human health. To analyze exhaust emissions generated by vehicle flows, it is necessary to use specialized emission models, since it is impossible to equip most of the roads with special devices for measuring those. The emission models currently used may be inadequate in relation to the vehicle design under study (for example, hybrid vehicles) or may be inaccurate due to the macro-scaling being used. This article presents a model of continuous monitoring of suspended PM2.5 particles from vehicles in real time under the current state of road traffic and meteorological conditions. This model is based on the training of the YOLOv4 convolutional neural network, which provides continuous collection and filling of the database of big data on traffic parameters, taking into account the characteristics of the transport infrastructure. Based on big data, the concentration of suspended substances in the atmosphere and their dispersion are calculated.