Abstract: Throughout the last decades, several simulation models have been proposed in an attempt to reproduce aircraft evacuation scenarios and provide an alternative to real certification trials. Furthermore, simulation models have been proven to be a useful tool when designing new aircraft enclosures. Among these, the airEXODUS model has seen widespread application and validation, being successful in predicting past certification trials and examining issues related to aircraft enclosure layout. This work introduces EvacSafeX, a multi-agent model centered around proper representation of human behaviour in aircraft evacuation scenarios. EvacSafeX finds its inspiration in complex systems and in the airEXODUS model, seeking to make complex interactions and behaviour emergence from individual modeling of human passengers. The model also takes novel approaches to represent human behaviour and passengers movement along the aircraft cabin. EvacSafeX adopts a perception-action approach to represent agent's capabilities, following a behavioral model defined by a rule list brain. Passengers are also characterised by both physical and psychological attributes identified as the most relevant in evacuation scenarios. In addition to the many components already included in the model, its generic architecture allows other ones to be easily incorporated. The components implemented beforehand in EvacSafeX prototype were verified through a set of validation experiments. At first sight, it was observed a significant sensitivity of some passenger's personal attributes, representative of their influence in real life cases. Furthermore, the proposed model demonstrated a high flexibility and diversity in the representation of passenger's behaviour, leading to an emergence of several different phenomena observed in real evacuation scenarios. Finally, promising results were obtained in an attempt to reproduce real certification demonstrations results and other experiments conducted with state of the art models.

Abstract: Railways are a key infrastructure for any modern country, so that their state of development has even been used as a significant indicator of a country's economic advancement. Moreover, their importance has been growing in the last decades either because of the growing Railway Traffic and to governments investments, aiming at exploiting railways means to reduce CO2 emissions and hence global warming. To the present day, many extreme events (i.e. major disruptions and large delays compromising the correct functioning of the system) occurs on a daily basis. However these phenomena have been approached, so far, from a transportation engineering point of view while a general theoretical understanding is still lacking. A better comprehension of these critical situation from a theoretical point of view could be undoubtedly useful in order to improve traffic handling policies. In this work we move toward this comprehension by proposing a model about train dynamics on railways network aiming to unveil how delays spawn and spread among the network. Inspired by models for epidemic spreading, we model the diffusion of delays among train as the diffusion of a contagion among a population of moving individuals. We built and tested our model using two large dataset about Italian and German railway traffic, collected using APIs intended to give passengers information about the trains, the state of the service and train delays. The model reproduces adequately delays dynamics in both systems, meaning that it captures the underlying key factors. In particular, our model predicts that the insurgence of clusters of stations with large delays is not due to external factors, but mainly to the interaction between different trains. Also, through our model is capable to give a quantitative account of the difference between the two considered railway systems in terms of probability of contagion and delays dynamics.

Abstract: Recent studies of human mobility largely focus on displacements patterns. Power-law fits of empirical long-tailed distributions of distances have been associated to scale-free super-diffusive random walks called Levy flights. However, drawing conclusions about a complex system from a fit, without any further knowledge of the underlying dynamics, might lead to erroneous interpretations. We show on a dataset describing the trajectories of 780,000 private vehicles in Italy, that the Levy flight model cannot explain the behavior of travel-times and speeds. We therefore introduce a new class of accelerated random walks, validated by empirical observations, where the velocity changes due to acceleration kicks at random times. Combining this mechanism with an exponentially decaying distribution of travel-times leads to a short-tailed distribution of distances which could indeed be mistaken with a truncated power-law. These results illustrate the limits of purely descriptive models and provide a mechanistic view of human mobility.

Abstract: This paper studies the coevolution of opinion dynamics in growing networks with attachment rule that depends on the opinion updating process. We propose that individuals choose to link with others according to the Hegselmann-Krause opinion dynamics model; each individual forms its neighborhood with others whose opinions are close to its own in an interval minor to some confidence level. Since individuals hold an opinion value in the continuous interval [0,1], then for a new agent on the network, the neighborhood will depend not only on her confidence level but also on her initial opinion value. We analyze the network structure when the initial opinion value is selected with: i) an uniform probability, ii) a probability as a function of the degree of the new agent, and iii) a probability as a function of the cluster coefficient of the new agent. Since the confidence value and the initial opinion selection influence the network structure, we then present a method to approximate the degree distribution and the number of cluster based on these two variables. In order to complete the coevolution analysis, we also study the convergence of opinions. When a new agent is added to the network, the opinion update for all individual process could take place immediately (all agents change their opinion as the average of their neighborhood) or could present a delay (agents change their opinion when they detect a variation in at least one of their neighbor opinion value). We then demonstrate that the convergence is affected by the confidence level and the initial opinion value selection, but the convergence time depends on when the updating process occurs.

Abstract: Variations in seasonal influenza epidemic initiation, timing, and magnitude yield highly variable illness data that can help researchers to understand the spatial spread of influenza. For the United States, predictable spatial patterns will contribute to more accurate predictive models for ascertaining when influenza infection will occur and to understand long distance connections. In order to evaluate the interdependence of cases we propose the use of a transfer entropy model (TE) that measures the amount of information transfer from one variable to the other; yet, in this context the number of cases ‘’transferred’’ from one region to another. TE is a non-parametric and non-linear model that offers an alternative measure of effective connectivity based on information theory, more powerful than Granger causality or assumption-based dynamic causal models. More precisely, TE is quantifying causal networks between time series where node/variable distance, node connectivity, and link weights are related to variable undirected statistical closeness, dependence, and directional entropy reduction. Furthermore, transfer entropy is an asymmetric measure that conveys directional information. Considering TE on CDC data it results that Northeast and Northwest US are the most influential nodes in the network. Conversely, Midwest and Southwest regions are strongly affected by other regions. There are long-distance connections between Northeast and Midwest, and between Mid-Atlantic and Southwest regions. Some pairs regions that are very far from each other (~1500-3000km) still show significant correlation with each other (r=0.45-0.65) that emphasizes the importance to assess effective connections rather than geographical connections. The results allow us to conclude that long-distance effects are relevant in the dispersion of influenza cases and to infer locally generated cases. The TE model can be useful in analyzing any other complex disease where interactions among sub-systems/regions are expected to be non-linear and where minimal a priori knowledge is available.