Infering Air Quality from Traffic Data Using Transferable Neural Network Models

This work presents a neural network based model for inferring air quality from traffic measurements. It is important to obtain information on air quality in urban environments in order to meet legislative and policy requirements. Measurement equipment tends to be expensive to purchase and maintain. Therefore, a model based approach capable of accurate determination of pollution levels is highly beneficial. The objective of this study was to develop a neural network model to accurately infer pollution levels from existing data sources in Leicester, UK. Neural Networks are models made of several highly interconnected processing elements. These elements process information by their dynamic state response to inputs. Problems which were not solvable by traditional algorithmic approaches frequently can be solved using neural networks. This paper shows that using a simple neural network with traffic and meteorological data as inputs, the air quality can be estimated with a good level of generalisation and in near real-time. By applying these models to links rather than nodes, this methodology can directly be used to inform traffic engineers and direct traffic management decisions towards enhancing local air quality and traffic management simultaneously.