**1. Introduction**

Knowledge of user flows on transportation systems is crucial for implementing control and managemen<sup>t</sup> policies. In this context, monitoring systems assume a central role and are widely used in many road and rail networks: they are one of the most important (and necessary) components of Intelligent Transport Systems (ITSs).

Monitoring systems are based on sensors (or detectors) that measure some characteristics of flows and transmit them to a control room, where the relevant data are used for implementing control and managemen<sup>t</sup> strategies. For instance, knowing traffic flows on a road network is useful for implementing flow-responsive traffic-signal systems, while the user loads on public transport vehicles can be used for real-time scheduling/rescheduling tasks.

Despite their usefulness, monitoring systems are not always provided on road and rail networks or, sometimes, implemented systems do not have enough sensors to collect the data required for control and managemen<sup>t</sup> strategies. Indeed, the costs of these systems require significant investments from public administration or public transport firms which are often inadmissible.

In this paper, we focus on metro lines, where access is controlled by turnstiles which can count the passengers entering each station with or without identifying their direction (in most of the existing stations, except terminals, passengers can board trains in different directions). These data can be easily collected every few minutes (e.g., fifteen-minute intervals) without installing new sensors. The objective

of this paper is to propose a method based on Artificial Neural Networks (ANNs) for estimating the number of passengers on each segmen<sup>t</sup> of a metro line using data obtained from turnstiles. Indeed, on-board data are rarely available in real time because they require infrared scanners, or alternatively weight-based, image-based or photocell-based sensors that are seldom installed on coaches.

The paper is organised as follows: Section 2 provides the background; Section 3 describes the problem to solve and the ANN approach; the method adopted for generating the data used in numerical tests is reported in Section 4; Section 5 describes the case study and the results; Section 6 concludes and identifies some research prospects.
