Uncertain demand prediction for guaranteed automated vehicle fleet performance

Abstract: Mobility-on-demand (MoD) services offer a convenient and efficient transportation option, using technology to replace traditional modes. However, the flexibility of MoD services also presents challenges in controlling the system. One of the major issues is supply-demand imbalance, caused by uneven stochastic travel demand. To address this, it is crucial to predict the network behavior and proactively adapt to future travel demand. In this thesis, we present a stochastic model predictive controller (SMPC) that accounts for uncertainties in travel demand predictions. Our method make use of Gaussian Process Regression (GPR) to estimate passenger travel demand and predict time patterns with uncertainty bounds. The SMPC integrates these demand predictions into a receding horizon MoD optimization and uses a probabilistic constraining method with a user-defined confidence interval to guarantee constraint satisfaction. This result in a Chance Constrained Model Predictive Control (CCMPC) solution. Our approach has two benefits: incorporating travel demand uncertainty into the MoD optimization and the ability to relax the solution into a simpler Mixed-Integer Linear Program (MILP). Our simulation results demonstrate that this method reduces median customer wait time by 4% compared to using only the mean prediction from GPR. By adjusting the confidence bound, near-optimal performance can be achieved.