Prediction of air quality perception in aircraft cabin based on psychophysical model and artificial neural network (ANN)-based model

As people are the ultimate arbiters of air quality in built environments, perceived air quality (PAQ) is receiving increasing attention. Odor is often designated as the main target of PAQ regulation, but due to the complex mechanism of cross-modal human perception under multi-pollutant coupling, the accuracy of odor perception evaluation and prediction in the real environment is limited. This study obtained passengers’ evaluation of their perception of cabin air quality (CAQ) and odor intensity (OI) in commercial aircraft cabins through on-board measurement of 36 flights and 878 supporting questionnaires. Although the CAQ was generally acceptable, 25 % of passengers were not satisfied, and odor complaints (OI ≥ 3) were captured on 6 flights. The odor concentration (OC) and OI in the aircraft cabin were calculated based on the olfactory threshold and the Weber-Fechner psychophysical model, and the total OC distribution in different flight phases ranged from 28.4 to 66.1. Aldehydes (especially long-chain) were most likely to be smelled directly. Limited by the two basic assumptions that VOC interaction was non-existent and that the odor intensity was only related to VOC, the accuracy of OI calculated by the existing model was about 0.4. In order to improve the accuracy of evaluation, a new data-driven model for human perception (CAQ and OI) prediction based on a knowledge-based BP neural network was proposed, and its prediction accuracy (R 2 : 0.81–0.87) and generalization (R 2 : 0.76–0.93) were verified. The new model is able to consider the interactions among individual differences, environmental factors and VOC concentrations, thus providing a method innovation for realizing people-oriented VOC control. • Available prediction of human perception by objective parameters. • Odor assessment of aircraft cabin by both survey and on-site measurement. • Critical review of psychophysical models for total OI conversion. • ANN-based network increased the prediction accuracy of total OI by more than 80 %.