Probability mass functions forecasting of occupants’ sensation votes under the effects of temperature, illuminance, and sound level based on ANN

Abstract

Occupants' individual sensation has been usually quantified under the stimulation of single environmental factor and presented by a mean sensation vote in previous literatures. However, individual factor sensation was affected by multiple environmental factors and personal sensation vote will probably deviate considerably from a mean vote. Hence, there is an attempt to adopt Probability Mass Function (PMF) as an evaluation method for presenting the distributions of occupants' sensation vote under the simultaneous stimulation of multiple environmental parameters and predict such distributions. Our investigation firstly conducted a controlled laboratory test and collected occupants' sensation votes on thermal, acoustic, and visual environments. Subsequently, we developed, optimized, and trained three artificial neural networks (ANNs) using the collected data for predicting PMFs of occupants' 7-point sensation votes i.e., thermal sensation vote, acoustic sensation vote, and visual sensation vote, under the simultaneously stimulus of temperature, illuminance , and sound level. Finally, the ANN models were generalized to predict the PMFs of thermal, visual, and acoustic sensation vote under different environments within typical ranges of environmental factors. Results indicated that the PMFs predicted by the trained ANN models show a good agreement with the collected vote proportion distributions. Thereafter, by comparing the expected values of sensation votes under different defined conditions, it confirmed the existence of crossed and interaction effects of multiple environmental factors on human comfort. The combinations of environmental parameters that lead to a neutral sensation were illustrated as a space with small thicknesses under three-dimensional coordinate system. The neutral air temperature, illumination, and sound level are between 23 °C–24.7 °C, 580 Lux~650 Lux, and 41.5 dB–43.5 dB, respectively. Moreover, the effect directions and effect sizes of temperature, illuminance, and sound level on occupants’ sensation were numerically expressed and discussed by variable control. Suggestions on environment control strategy and further research were also provided. • Probability Mass Function (PMF) was adopted as a description method for occupants' sensation votes. • Three Three-Input Seven-Output ANN models were developed for predicting PMFs of occupants' 7-point sensation votes. • Effect directions and effect sizes of temperature, illuminance, and sound level on sensation were numerically expressed.