Methods for Risk Reduction: Modelling Users’ Updating Utilities in Urban Transport Networks

Abstract

The paper deals with transportation system models (TSMs) on behalf of methods for risk reduction in urban areas. As far as urban transport networks are concerned, further development of models is necessary in order to capture the potentialities of emerging information and communication technologies (e-ICTs) in providing valuable information about the evolution of a transportation network during an evacuation. A new learning process structure has been proposed to take into account how the path costs (disutilities) in ordinary and emergency conditions will be perceived and updated by the users. The data provided by emerging ICT allow for the incorporation of dynamics inside the network model, concerning the update of information provided by the internet of things and big data.

1. Introduction

Methods for risk reduction in urban areas by means of evacuation planning received great impulse since the early 2000s after the catastrophic events of 9/11 and Hurricane Katrina [1,2,3]. Today, the new frontiers that have been opened by emerging information and communication technologies (e-ICTs) require further development of research on models to support evacuation planning in order to identify their benefits and costs.

Evacuation planning consists of generating an optimal scenario in terms of configuration of the transport network in order to reduce the exposure component of risk [4]. A decisive role in the planning phase is played by transport system models (TSMs), which allow for the assessment and the optimization of evacuation procedures. In particular, TSMs facilitate the computation of evacuation times given the size of the population to evacuate from an urban area and the computation of the amount of travel demand evacuated given a time interval before the calamity approaches and starts to produce its effects. The outputs produced by TSMs are, in turn, the input for the computation of the variables quantifying the exposure component.

The methods for risk reduction were extensively developed inside the research project SICURO at the Università Mediterranea di Reggio Calabria (Italy). The production of scientific publications was rich; hereby, the ones concerning the progress of the TSM with regard to risk conditions are noted. The general framework for the analysis of an urban road transportation system in emergency conditions is presented in [5]. The framework was tested in three different urban areas, assuming that calamitous events occur and that there is a finite time window between the instant when the calamitous event is notified and the instant when the effects of the event on the population are tangible. Different models composing the TSM have been specified, calibrated, and validated according to data observed during some real evacuations of urban areas. The models and the decision support systems developed refer to travel demand [6] and transport network modelling [7]. The problem of path design for emergency vehicles (e.g., ambulances) travelling on the road network in emergency conditions (e.g., when a calamitous event such as a hurricane is approaching) to rescue weak users, such as older and disabled people, is studied in [8]. Marcianò et al. [9] present a signal-setting optimization procedure where the travel demand is assigned to a congested road transport network. The procedure is validated on a real portion of a small town where the population must be evacuated due to an approaching calamitous event. Di Gangi [10] and Di Gangi et al. [11] present a mesoscopic extension of a dynamic traffic assignment model to assess the effects of an approaching calamitous event on a transport network. The objective was to determine the quantitative indicators for estimating the exposure component of risk. Urban planning could support the definition of specific actions for pursuing risk reduction according to the four emergency management phases, which are prevention, preparedness, response, and recovery. Rindone and Panuccio [12] focus on the preparedness phase and the transport planning process at the urban level. Russo et al. [13] focus on non-material training activities in order to improve capabilities connected to evacuation planning and implementation, while Russo and Rindone [14,15] present an advancement regarding the analysis of the planning process according to the logical framework approach of urban systems in emergency conditions.

The possibilities facilitated by emerging ICT, offering historical and real-time information, can be capitalized by introducing modifications in the traditional structure of the TSMs. The problem is emphasized by the fact that, generally, the road networks under ordinary conditions may significantly differ from those encountered in emergency conditions.

The major challenge for transport planners and managers of the road network in emergency conditions is related to the quantification and updating of the costs related to the links and the nodes of the network. The two main barriers are represented by the scarcity of information about the current conditions of the road network and difficulty in predicting the cause-effect mechanism generated by each calamitous event on the road network.

The process that describes the transport users’ knowledge about the network is called a learning process in the literature [16,17], and it explains how transport users forecast the path costs (or disutilities) of the transport network today, from previous experience of the network and/or from previous forecasts. The consolidated equations of the learning process need adjustments to accommodate the variability in network conditions arising during evacuations from urban areas.

From the considerations reported above, the following general research questions emerge:

• How is it possible to modify the current structures of the learning process inside the TSMs to incorporate the potential benefits provided by e-ICT?
• Is it possible to specify, calibrate, and validate a user’s updating utility model able to capture real-time information consistently with the extension of time windows available for people evacuation?

The paper focuses on transport network modelling in emergency conditions on behalf of the TSM framework. The research contribution of the paper concerns the proposal of a formulation, where the presence of e-ICT allows for the specification of a learning process of path costs. The learning process allows for the capture of the dynamics of the transportation network in emerging conditions (e.g., during the evacuation). In particular, it explains how transport users (e.g., evacuees) forecast the costs of the transport network that they will experience during evacuation based on previous experience of the network and/or from existing information systems (e.g., e-ICTs). The learning process could be fed by information provided by emerging ICTs, such as the internet of things (IoT) and big data (BD).

According to the above research contribution, the remaining part of the paper is articulated in line with the steps of the adopted research methodology (Figure 1).

The first step (Section 2) concerns the introduction of the problem of evacuation planning and the description of the structure of transport system models (TSMs) in terms of transport network, travel demand, and assignment sub-models, with a focus on the transport network models and on the traditional approach of the learning process.

The second step (Section 3) reports a classification of emerging ICTs that allow for the definition and implementation of dynamic solutions inside the three modelling components of the TSMs, with a focus on applications finalized for risk reduction. Furthermore, the section reports three characteristics of the information that are necessary to support the estimation of users’ updating utility models.

The third step (Section 4) presents a new structure of the learning process as a dynamic extension of the supply sub-model consistent with the availability of inter-periodic information provided by e-ICTs.

The last step (perspectives) reports the future activities of the ongoing research concerning two main elements: (1) the execution of an experimental test on a portion of an urban area, where a calamitous approaching event is notified and some people need to reach a safe area from a public building, and (2) the calibration of the updating utility model specified in the present paper by means of information provided by a (simulated) information system to evacuees.

It is worth noting that the boxes in Figure 1 associated with the first three sections are drafted with a continuous line, meaning that they are steps of the research methodology presented in this paper, while the box associated with the perspective section is drafted with a dotted line, meaning that it is a future step of the research methodology.

2. Evacuation Planning and Transport System Models (TSMs)

2.1. Evacuation Planning

Evacuation planning, which is part of the broader disaster management cycle, consists of generating an optimal emergency scenario to reduce the exposure component of risk (Figure 2).

Exposure may be reduced by means of evacuation measures, which allow people and goods to be rescued from an area where a calamitous event is supposed to produce its effects. According to objectives and constraints defined by decision-makers, several emergency scenarios may be defined in terms of measures regarding transport infrastructures and services. The effects of these measures can be observed (once they are realized) by means of e-ICT tools, which provide observed data about transportation systems. They may be forecasted by means of transport system models (TSMs), which provide forecasted data about transport systems. TSMs requires as input the socio-economic data, the transport infrastructures and services, and the data available by (emerging) ICT and provide as output flows and performances of the elements of the network (links and nodes). The combination of ICT and TSMs form the so-called intelligent transportation system (ITS), which, in the presence of emerging ICT, could be named advanced ITS. The observed and forecasted data are inputs for the exposure and risk models in order to estimate the level of risk corresponding to the defined emergency scenario, in relation to a given level of occurrence and vulnerability. The level of risk could be used as feed-back to define further emergency scenarios until an optimal one is obtained according to the defined objectives.

2.2. Structure of TSMs

TSMs simulate a transport system, in which transport supply and travel demand interact. TSMs are composed by three interacting sub-models, as depicted in Figure 3.

The travel demand models predict the demand patterns that are the results of user decisions influenced by the performances of transport infrastructures and services. Such models can be classified as behavioral or non-behavioral. Within the behavioral framework [16,18,19], these models may be stochastic or deterministic. This segmentation is necessary to consider whether the (dis)utility associated with each user’s decision is treated as a random or deterministic variable. The transport supply model evaluates the costs (or disutilities) that users incur when they use the transport infrastructures and services. The most prevalent method is the topological model, characterized by a network structure comprising links, nodes, and associated cost functions such as time–flow relationships or fundamental diagrams [16,20]. The supply–demand interaction (or assignment) model assesses the costs and flows resulting from the interaction between users’ decisions and the performances of transport infrastructures and services. This model generally refers to the topological–behavioral framework. The prevailing demand–supply interaction models can be categorized as reported in [16,17]. These models may be static or dynamic, according to their capacity to simulate the transport system:

• stationary scenarios, when it is reasonable to accept that the travel demand and/or the network capacity are the objects of slight fluctuations around average values inside a reference time period, or
• dynamic process scenarios, characterized by relevant fluctuations in travel demand and/or network capacity inside a reference time period.

Static models are further categorized into the following:

• free-flowing models, based on the network loading approach, or
• equilibrium-based models, based on the equilibrium approaches such as “user equilibrium” and “system optimum” [16,17].

Conversely, dynamic models can be segmented into two distinct classes, each one capturing different temporal dynamics of the transport system, as reported in [17]:

• within-day models, which incorporate the temporal evolution of the transport system inside a designated day or inside a reference time period (e.g., morning peak period), and
• day-to-day models, which describe the evolution of the transport system from a given day (or a reference time period) to subsequent days.

2.3. Network Model and Learning Process

This section reports the general formulation of a congested network model in static conditions and the classic structure of the learning process.

The static network model is composed of the following two equations [16,17]. The first one is the path costs versus link costs consistency equation:

g = Δ^T c (f)

(1)

with

• g as the path costs vector;
• Δ as the link–path incidence matrix;
• c() as the additive link cost functions vector; and
• f as the link flows vector.

Equation (1) puts in relation the path costs on a transport network with the additive link costs, which depend on link flows (congested network), through the transposition of the link–path incidence matrix reporting the links that compose each path. An example of additive cost is the travel time; in this case, the travel time of the path may be obtained as the sum of the travel times of the links that compose the path.

Equation (1) may be expressed in scalar form as follows:

g_k = ∑_h β ∙ x_nk ∀ k

(2)

where x_hk is the attribute h of path k (e.g., travel time, monetary cost).

Path cost variable may have non-additive components. Examples of non-additive costs are, for instance, some structures of highway tolls and transit fares. The non-additive path costs in Equation (1) may be taken in consideration by adding a vector of non-additive cost, g^na:

g = Δ^T c (f) + g^na

(3)

The second one is the path flows versus link flows consistency equation:

f = Δ h

(4)

with h as the path flows vector.

Equation (4) puts the link flows in relation to the path flows of the transport network, through the link–path incidence matrix. According to Equation (4), the flow passing through every section of a link (the stationary condition holds) is obtained as the sum of the flows of the paths that the link belongs to. The path flow variable is intrinsically additive in Equation (4).

The specification of a network model according to an inter-periodic dynamic pattern requires an extension of the static Equations (1) and (4).

The classic approach of the learning process may be simulated with the (dis)utility updating model in ordinary conditions, which allows for obtaining the forecasted/computed costs.

In the case of path costs, the day-to-day pattern allows the vector of forecasted (or computed) path costs at “day” [y], g^fo[y], to be expressed as follows [16,17]:

g^fo[y]= γ(β, g^ex[y − 1], g^fo[y − 1])

(5)

where

• γ() is the implicit updating function, which may be differently specified;
• β is the vector of parameters to be calibrated;
• g^ex[y − 1] is the vector of experienced path costs at day [y − 1] obtained from Equation (1); and
• g^fo[y − 1] is the vector of forecasted (or computed) path costs at day [y − 1] obtained from Equation (1).

The implicit updating function, γ(), may be specified in different ways [13]:

• “yesterday” filter, where the forecasted path costs at day [y] are assumed to be equal to the experienced path costs at day [y − 1] (yesterday);
• moving average filter, where the forecasted path costs at day [y] depend on the experienced path costs at day [y − 1], [y − 2], [y − 3], …, [y − n]; and
• exponential filter, where the forecasted path costs are defined by means of a strict convex combination of forecasted and experienced costs at day [y − 1].

3. Emerging ICT and Internet of Things

3.1. Classification of Emerging ICT

Emerging forms of information and communication technologies (ICTs) have been developed in the last decade. They have quickly spread in many areas of economy and society (e.g., manufacturing, mobility, social media, etc.), influencing the behavior of people at individual and collective levels. Some applications are proposed in the fields of city logistics [21,22,23] and of ports and maritime transport ([24]).

In broad terms, the different emerging technologies present in the market may be classified into four groups:

• Internet of things (IoT) is a “framework that leverages on the availability of heterogeneous devices and interconnection solutions, as well as augmented physical objects providing a shared information base on global scale, to support the design of applications involving at the same virtual level both people and representations of objects” [25];
• Big data (BD) enrich information derived from traditional data in terms of volume and variety. ICT collects, manages, and processes big data within a tolerable elapsed time, in order to improve the representation of traffic phenomena and their modelling;
• Artificial intelligence (AI) aims to improving the prediction of and decision-making related to transport operations, as well as to automatic repetitive manual transactions;
• Blockchain (BC), or “distributed ledger,” is a “technology for keeping a chronologically ordered list of transactions (a ledger) on multiple, independent stakeholders (called nodes). Updates of the list should be identical on each of the nodes and each node checks the validity of the transactions before updating its list” [24,26,27].

The 5G (or 6G) communication network is not considered because it is transversal to all other technologies.

The above classes of emerging ICTs are able to provide new opportunities to define and implement integrated and dynamic solutions to optimize the different components of the TSM. It is worth noting that these emerging ICTs are mature today, and they are, generally, part of the stand-alone digital platforms available for transport users and/or decision-makers (e.g., transport system managers). They may be considered emerging, as their implementation inside the intelligent transport systems (ITSs) is still in fieri.

Among the emerging ICTs classified above, IoT and BD may be considered the ones with the greatest impact on path choice [28]. According to their features, IoT allows real-time communication between vehicles (vehicle-to-vehicle communication) and between vehicles and infrastructure (vehicle-to-infrastructure communications) [29], and BD are able to provide an exhaustive and detailed representation of historical patterns of network costs.

3.2. IoT-Based Applications in Risk Reduction

This section recalls some relevant papers regarding IoT implementations related to the topic of risk reduction.

The first group of publications present state-of-the-art, open challenges and research trends on IoT-enabled systems for disaster management. Ray et al. [30] present a systematic survey on IoT-based disaster management issues, highlighting the key protocols. The authors provide a classification of market-ready IoT-enabled infrastructures that aid disaster management, and, finally, they suggest the key challenges and future directions in IoT-based disaster management systems. Bail et al. [31] present a literature review about IoT in disaster management and applications of more specific devices or technologies. The authors identified four main areas of applications of IoT: knowledge about the disaster; assistance to victims; supplies management; and environmental management.

The second group of publications focus on more specific IoT applications. Among them, some papers are recalled as follows. Garg et al. [32] propose the use of cloud computing in order to address these challenges with almost unlimited capacity for computation, storage, and networking and to offer computational modelling of natural hazards. The authors present a conceptual cloud-based framework for modelling and managing natural hazards using the cloud in combination with IoT and edge computing. Shah et al. [33] propose the architecture of a disaster-resilient smart city through the integration of IoT and BD technologies. The implementation of the architecture requires the execution of several steps: big-data harvesting, aggregation, pre-processing, and analytics and service platforms. A number of datasets about buildings, pollution, vehicular traffic, and social media are used to validate the architecture in generating alerts for a fire in a building, increments of pollution level, and emergency evacuation paths and to collect information about natural disasters, such as earthquakes and tsunamis. The evaluation of the system efficiency is measured in terms of processing time and throughput. Dachyar et al. [34] aim to generate a new logistics inventory business process in the shortest response time as an implementation of a rapid disaster relief distribution process by utilizing IoT potential. After a disaster occurs, coordination of inventory management for logistic relief can be poor, and, in this context, IoT could provide tangible support in sharing, receiving, and analyzing data to meet urgent business requirements. Sharma et al. [35] describe the role of IoT in the management of different kinds of disasters with a comparison among some solutions that are available in the market. The authors show some examples of applications of IoT such as for an early-warning system for fire detection and earthquake detection. The study directs the stakeholders in the use of IoT technology to secure smart cities’ infrastructures, to manage disaster, and to reduce risks. Niu et al. [36] analyze the average speed variation of vehicular flow on links of real-world networks and estimates the link attributes using crowd-sourced data. The authors define a resilience measurement in terms of the threshold of vehicle speed, where the road link recovers from a disruptive event. The results indicate that link graph-based metrics and attributes have a high impact on network resilience. A general framework for disaster emergency management under the combination of IoT with artificial intelligence technologies is proposed in [37]. The general framework mainly contains six core components: (1) sensor and monitoring devices, containing various categories of sensors to sample the states of real-world objects; (2) raw-data storage, whose primary function is to connect and manage large numbers of sensors or monitoring devices; (3) data index and geo-data server, where the former is responsible to create an index to improve the efficiency of data queries and the latter is responsible to provide the road network information of the city; (4) data querying, which includes a set of operators to query the sensor data and to conduct statistical analysis; (5) an integrated application, which contains various top-level applications; and (6) the user interface. The experiments are conducted on a portion of the road network where cars can travel freely along the 6th Ring Road in Beijing (China). Astarita et al. [38] present an empirical analysis of functional requirements for the management of emergencies through a system platform based on emerging ICT. The system platform is composed of a decision support system, a mobile application, and a web platform. According to the authors, the application of this system platform could bring significant advantages to the management of emergency situations in urban areas, when supported by ICT.

3.3. Information

This section presents an overview of information about the transport system that may be available in emergency conditions.

It is worth recalling that the information has a double usage [39,40]. The first concerns the information gathered by public authorities and road network agencies for transport network management and control. The second concerns the information disseminated to transport users to support their trip choices.

Information gathered may belong to two categories.

Traffic information, such as vehicular flows, densities, speeds, travel times, etc., is endogenous to one of the three components of the transport system (network, demand, demand–network interaction), and it may be obtained as output of the TSM framework.

Event information is exogenous to the transport system and is not output of the TSM framework. Event information, as considered in this paper, concerns the information related to the unexpected or calamitous event that happened or that is approaching and that determines, or may determine in the short-term, an impact on the transport system. There is a variety of calamitous events that may affect in different ways the transport system, which may be classified by the time when the calamitous event reaches the urban area; the time when it starts its effects; the time when it ceases its effects.

Information disseminated from a transport agency in ordinary conditions, or from civil protection in emergency conditions, to road users may be classified by different criteria. The information may be static when it is disseminated via fixed signs present along the road network or dynamic when it is disseminated via messages arriving to users though different media. The information may also be descriptive, indicating or suggesting trip solutions (e.g., path, speed) or providing some warnings about approaching events or events that have already happened, or prescriptive, reporting an obligation such as speed reduction or path diversion.

4. Proposed Structure of Updating Utility

The new opportunities offered by e-ICT in providing historical, real-time, and forecasted information can be exploited by introducing some modifications in the traditional structure of the learning process, specified in Equation (5).

In general, the knowledge of the users about the conditions of the road network in ordinary conditions may be very different from their knowledge about the network they face in emergency conditions.

A common specification of the updating utility model in ordinary conditions, in Equation (5), is the exponential filter, which allows for the forecast of the generic attribute h associated with path k at “day” y, x_hk^fo[y], as follows:

x_hk^fo[y] = ζ x_hk^exp[y − 1] + (1 – ζ) x_hk^fo[y − 1]

(6)

where

• x_hk^exp[y − 1] is the attribute of path k experienced at day [y − 1];
• x_hk^fo[y − 1] is the attribute of path k forecasted (or computed) at day [y − 1];
• ζ ∈ [0, 1] is the weight given by the user to the experienced attribute.

The model of Equations (5) and (6) could be modified to take into account the variability of the network conditions that arise when evacuation of people from a portion of an urban area is necessary due to an approaching calamitous event. The problem in emergency conditions concerns the quantification and updating of costs at the links and nodes of the transport network. Two elements make the quantification and updating problematic: (a) the poor and discontinuous information about the evolution of the conditions of the road network and (b) the complexity connected to the prediction of the effects of every single calamitous event on the road network.

A specification of the updating utility model in dynamic conditions by means of the exponential filter was recently proposed in [21,22,28], where the generic attribute x_hk^fo is forecasted by considering the inter-periodic process along two reference time periods.

The first is the wider time period, which has similar characteristics to the “day” y defined in the traditional approach, where an average value of the attribute x_hk is quantified according to the extension of the day (e.g., if an evacuation of the population from a portion of coastal urban area due to an approaching tsunami lasts 16 h, the “day” is assumed equal to y = 16 h).

The second is a shorter period, which is defined as “time window” t, where an average value of the attribute x_hk is quantified according to the limited extension of the window (e.g., the extensions of the windows during the evacuation considered in the previous point may be set to one hour; therefore, t = 1 h).

The availability of data about path costs provided by e-ICT offers further possibilities to forecast the attribute x_hk^fo[t, y] by means of an updating utility model that incorporates the dynamics related to the two time periods defined above.

The first time period is the day y for a given time window $\overline{\mathrm{t}}$:

$${{\mathrm{x}}_{\mathrm{hk}}}^{\mathrm{fo}}[\overline{\mathrm{t}}, \mathrm{y}]=\varphi ({{\mathrm{x}}_{\mathrm{hk}}}^{\exp }[\mathrm{y}-1], {{\mathrm{x}}_{\mathrm{hk}}}^{\mathrm{fo}}[\mathrm{y}-1])\hspace{1em}\hspace{1em}\mathrm{for} \mathrm{t}=\overline{\mathrm{t}}$$

(7)

where

• ϕ() is an implicit function that may be differently specified according to [17];
• x_hk^exp[y − 1] is the value of the attribute experienced by the user at day [y − 1] for a given time window $\overline{\mathrm{t}}$; and
• x_hk^fo[y − 1] is the value of the attribute forecasted by the user at day [y − 1] for a given time window $\overline{\mathrm{t}}$.

The second time period is the time window t for a given day $\overline{\mathrm{y}}$:

$${{\mathrm{x}}_{\mathrm{hk}}}^{\mathrm{fo}}[\mathrm{t}, \overline{\mathrm{y}}]=\varphi \left({{\mathrm{x}}_{\mathrm{hk}}}^{\exp }\right[\mathrm{t}-1, \overline{\mathrm{y}}], {{\mathrm{x}}_{\mathrm{hk}}}^{\mathrm{fo}}[\mathrm{t}-1, \overline{\mathrm{y}}\left]\right)\hspace{1em}\hspace{1em}\mathrm{for} \mathrm{y}=\overline{\mathrm{y}}$$

(8)

where

• ϕ() is an implicit function that may be differently specified according to [17];
• x_hk^exp[t − 1] is the value of the attribute experienced by the user at time window [t − 1] for a given day $\overline{\mathrm{y}}$; and
• x_hk^fo[t − 1] is the value of the attribute forecasted by the user at day [t − 1] for a given time window $\overline{\mathrm{y}}$.

At this stage, it is possible to make the following assumptions in relation to the two time periods defined above.

IoT could provide information consistent with the “time window” t; in other words, IoT could provide real-time information about the path costs of the network. Therefore, it is acceptable to associate the knowledge of the attribute x_hk^fo[t] forecasted by the users at t with the real-time knowledge of the attribute x_h^IoT[t − 1] coming from IoT at time window t − 1:

x_hk^fo[t] = x_hk^IoT[t − 1]

(9)

BD could provide information consistent with the “day” y with regard to the historical path costs of the network. Therefore, it is acceptable to associate the knowledge of the attribute x_hk^exp[y] experienced by the users at day y with the historical knowledge of the attribute x_hk^BD[y − 1] coming from big data (BD) at day y − 1:

x_hk^exp[y] = x_hk^BD[y − 1]

(10)

According to Equations (9) and (10), Equations (7) and (8) can be updated as follows:

x_hk^fo[t, y] = ϕ(x_hk^BD[y − 1], x_hk^IoT[t − 1])

(11)

where

• x_hk^BD[y − 1] is the value of the historical attribute related to path k obtained through BD at day [y − 1];
• x_hk^IoT[t − 1] is the value of the real-time attribute related to path k obtained through IoT at time [t − 1].

The most common attribute associated with the path cost is the path travel time of a vehicle traveling from an origin (e.g., the home place where a household lives) to a destination (e.g., the safe area that must be reached from the household for rescue).

With regard to the process along the day y, the information about path travel times is provided from BD in relation to two elements.

• The values of travel times experienced in ordinary conditions if no evacuation exercises or training courses (training) have been carried out by the population of an urban area.
• The values of travel times experienced in emergency conditions during evacuation exercises or learned during training courses (training) carried out by the population. If the exercise and/or the training is repeated more times, it is possible to take into consideration the additional data by introducing a dependency of travel time into the previous hypothetical “days” [y − 2], [y − 3], [y − 4], which are the ones when the evacuation exercises took place [41].

5. Research Perspectives

The paper focuses on the supply (network) component of the TSM framework. A modelling extension of consolidated static network models was operated in order to capture two elements, responding to the two research questions formulated in the introduction:

• the potentialities of emerging ICTs in providing valuable information according to predefined time periods and
• the inter-periodic dynamics of the transportation network during an evacuation in terms of both topology (e.g., link closure, contraflow) and capacity of the links and nodes generated by endogenous (traffic) and exogenous (event) causes.

A new learning process structure has been proposed to take into account how the path costs (disutilities) in emergency conditions may be perceived and updated by the users. The proposed structure of the learning process allows for the incorporation inside the network model of the new potentialities offered by the emerging ICTs, such as IoT and BD, to update inter-periodic information.

The paper presents some intermediate results concerning the specification of an ongoing research project titled “RISK: Recovery Increasing by Social Knowledge,” composed of different research lines. Among them, it is worth recalling the methods for training and exercises to pursue the planned evacuation [41].

The steps of the ongoing research methodology are described in Figure 1, where the last step concerns the execution of a planned experimental test on a portion of a real network to verify how IoT and BD information could be canalized to estimate and update the attributes inside the updating utility model formalized in this paper. Assuming that a calamitous approaching event could be notified and that some people must reach a safe area (destination) from a public building (origin), the road network could offer two main alternatives to evacuees: the first alternative could be characterized by an urban path and the second alternative by a motorway path. The two paths are connected to each other by means of some junctions. Considering that the network conditions and the information accessible to evacuees could change consistently within the time windows t, as previously described, two network scenarios could be defined:

• In the first scenario, no real-time information is provided to evacuees, and historical, or BD, information is available to estimate the attributes inside the updating utility.
• In the second scenario, real-time information about potential damage of road infrastructures is available to evacuees at one or more junctions, and “IoT” information may be used to estimate the attributes inside the updating utility model.

The implementation of the two abovementioned network scenarios in a lab context could allow for the calibration and validation of the updating utility model specified in the present paper by means of real-time information being provided to evacuees.

The findings of the research have potential applicability both for researchers working on the development of TSM frameworks and for urban transport planners and managers. In the research field, the information about the current network conditions perceived by the evacuees estimated by means of the proposed updating utility model may be the input for the dynamic demand models simulating the updated choices in terms of paths undertaken by users along the network during an evacuation. In the technical field, the estimation of the current network conditions perceived by the evacuees provided by the proposed updating utility model may support both public and private decision-makers in the development of evacuation plans.