Estimation of Elevator Passenger Traffic Based on the Most Likely Elevator Trip Origin-Destination Matrices

We present a constraint programming formulation for the elevator trip origin-destination matrix estimation problem using a lexicographic bi-criteria optimization method where least squares minimization is applied to the measured counts and the minimum information or the maximum entropy approach to the whole matrix. An elevator trip consists of successive stops in one direction of travel with passengers inside the elevator. It can be defined as a directed network, where the nodes correspond to the stops on the trip and the arcs to the possible origins and destinations of the passengers. The goal is to estimate the most likely counts of passengers for the origin-destination pairs of every elevator trip occurring in a building that are consistent with the measured boarding and alighting counts and any prior information about the trip matrix. These counts are used to make passenger traffic forecasts which, in turn, are used in elevator dispatching to reduce uncertainties related to future passengers and therefore to improve passenger service level. Artificial test data was obtained by simulations of lunch hour traffic in a typical multi-story office building. This resulted in complex problem instances that enable robust performance and quality testing. The results show that the proposed approach outperforms previous alternatives in terms of quality, and can take an advantage of prior information. In addition, the proposed approach satisfies real time elevator group control requirements for estimating elevator trip origin-destination matrices.