Maria Davidich

Predicting Pedestrian Flow: A Methodology and a Proof of Concept Based on Real-Life Data

Building a reliable predictive model of pedestrian motion is very challenging: Ideally, such models should be based on observations made in both controlled experiments and in real-world environments. De facto, models are rarely based on real-world observations due to the lack of available data; instead, they are largely based on intuition and, at best, literature values and laboratory experiments. Such an approach is insufficient for reliable simulations of complex real-life scenarios: For instance, our analysis of pedestrian motion under natural conditions at a major German railway station reveals that the values for free-flow velocities and the flow-density relationship differ significantly from widely used literature values. It is thus necessary to calibrate and validate the model against relevant real-life data to make it capable of reproducing and predicting real-life scenarios. In this work we aim at constructing such realistic pedestrian stream simulation. Based on the analysis of real-life data, we present a methodology that identifies key parameters and interdependencies that enable us to properly calibrate the model. The success of the approach is demonstrated for a benchmark model, a cellular automaton. We show that the proposed approach significantly improves the reliability of the simulation and hence the potential prediction accuracy. The simulation is validated by comparing the local density evolution of the measured data to that of the simulated data. We find that for our model the most sensitive parameters are: the source-target distribution of the pedestrian trajectories, the schedule of pedestrian appearances in the scenario and the mean free-flow velocity. Our results emphasize the need for real-life data extraction and analysis to enable predictive simulations.

Data Driven Monitoring of Rolling Stock Components

In the rolling stock business, the digital age marks the arrival of a new paradigm for operation, maintenance and efficiency: combining data gathered from millions of machine and infrastructure sensors with big data analytics capabilities allows to monitor entire fleets down to individual components and plan maintenance actions only when they are necessary. This manuscript presents a case study of train-door condition monitoring in which a machine learning platform is leveraged to efficiently monitor and predict anomalies.

Experiences of Application in the Automation Domain

In this chapter, we explain the application of the SPES XT modeling framework for the running example of the desalination plant. The objective is to explain the philosophy and methodology of the SPES XS modeling framework to automation domain experts based on artifacts, processes, and tools usually used today in the automation domain.

Waiting Zones for Real Life Scenarios: A Case Study Using a German Railway Station as an Example

Simulations of pedestrian dynamics aim to reproduce and predict the natural behaviour of pedestrians in different situations. In most models it is assumed that pedestrians constantly walk towards their destinations. Here we investigate the legitimacy of this assumption using data, collected during a field experiment and obtained from analysis of video recordings, at a major German railway station. Our observations suggest that a substantial proportion of people stand at certain locations for some time. In order to reproduce the observed behaviour adequately, we enhance an existing cellular automata framework with a new element to model standing persons, the so called waiting zones. Through simulations, we demonstrate how standing persons influence the overall dynamics. We also analyse how the developed model can be used for analysis of critical situations.

A Methodological Approach to Adjustment of Pedestrian Simulations to Live Scenarios: Example of a German Railway Station

Pedestrian stream simulations serve to predict the flow of a crowd. Applications range from planning safer buildings, performing risk analysis for public events to evaluating the clever placement of advertisement. The usability of a simulator depends on how well it reproduces real behavior. Unfortunately very little data from live scenarios has been available so far to compare simulations to. Calibration attempts have relied on literature values or, at best, laboratory measurements. This paper is based on live video observations at a major German railway station. We present a methodological approach to extract key data from the videos so that calibration of the simulation tool against live video observations becomes possible. The success of the approach is demonstrated by reproducing the real scenario in a benchmark simulator and comparing the simulation with the live video observations.