High-emitter identification model establishment using weighted extreme learning machine and active sampling

Abstract

High-emitting vehicles cause disproportionate air pollutants, thus making the identification and control of high-emitters a critical issue to reduce air pollution. On-road emission remote sensing (OERS), which can measure the emission of passing vehicles without interfering the normal driving, is an ideal means to identify the on-road high-emitting vehicles. Since the remote sensing measurements only reflect the instantaneous emission status, there is no doubt that OERS-output pollutant concentrations are related to the operation conditions and ambient environments at the measuring moment. Therefore, in order to identify on-road high-emitters effectively and accurately, a high-emitter identification model considering the relationship between OERS-output pollutant concentrations and their influence factors (such as passing speed, passing acceleration, wind speed, wind direction, temperature, etc.) should be established. In this paper, the way to establish the high-emitter identification model by machine learning is investigated. Because of the imbalanced distribution characteristic of emitter dataset, the weighted extreme learning machine is adopted as the identification model. Meanwhile, to enable an efficient establishment of the identification model, the active sampling that considers the dataset imbalance is introduced to select valuable samples to be labeled. The experimental results show that the high-emitter identification model establishment method based on weighted extreme learning machine can reduce the identification error for high-emitters significantly. Additionally, the active sampling can select valuable samples and improve the identification performance through the model update.