Developing artificial neural networks to estimate real-time onboard bus ride comfort

Abstract

The ride comfort of bus passengers is a critical factor that is recognised to attract greater ridership towards a sustainable public transport system. However, it is challenging to estimate bus passenger comfort onboard while travelling due to the complex non-linear interaction among various factors. A practicable method to collect real-time comfort ratings by passengers is also not readily available. This study developed an artificial neural network (ANN) model with three layers to precisely estimate real-time ride comfort of bus passengers. The inputs are vehicle-related parameters (speed, acceleration and jerk), passenger-related features (posture, location, facing, gender, age, weight and height), ride comfort index in ISO 2631-1997 (vibration dose value and maximum transient vibration value), and output is passenger rating (collected from a specialised mobile application). The ANN model provided a satisfactory performance and good correlation between inputs and output with an average MSE = 0.03 and R-value = 0.83, respectively. Sensitivity analysis was also conducted to quantify the relative contribution of each variable in the ANN model, revealing similar contributions among all influencing factors in the range of 4–6%. On average, passenger-related factors contribute slightly higher than vehicle-related factors to the ride comfort estimation based on the connection weight approach. The development of ANN model which can precisely estimate bus ride comfort is important as a considerable amount of machine learning and artificial intelligence are utilised to guide autonomous bus (AB). The present findings can help AB designers and engineers in improving AB technology to achieve a higher level of passengers’ onboard comfort.