Test System for Mapping Interdependencies of Critical Infrastructures for Intelligent Management in Smart Cities

. The critical infrastructures such as power distribution networks, water networks, transportation and telecommunication networks that are settled within the area of a city produce a large amount of data from applications such as AMI, SCADA, Renewable Energy Management System, Asset Management System, Weather data, etc. To convert these massive data into useful information, visualization is an effective solution. Visualization is an established methodology in scientific community and it has been used in many fields because of its strong capability in large data management and information display. However, its applications in a holistic view of critical infrastructures mapping at a city level is a missing link or still in infancy stages for some specific infra-structures (e.g. water and power networks). In this paper, a technique to visualize critical infrastructure data by using a system that consists of GIS (Geographic Information System) for buffer spatial analysis and Google Earth for display is proposed. The goal of this work is to design, model and validate a benchmark system that is capable to visualize and map as well as to prepare the next inter-linking phase of modeling interdependencies of several critical infrastructures such as distribution power networks, water networks, transportation networks and communication networks. The tool that will be used in achieving this goal is the open source QGIS. Further we aim to provide the grounds for a theoretical framework that can capture the interdependencies between critical infra-structures using techniques from graph theory, machine learning, econometric science and operation research. This framework is the first step in developing fundamental mechanisms for resilient management of critical infrastructures for the safe operation of smart cities.


Introduction
Cities may be seen as critical living habitats due to the concentration of critical infrastructures and people on a relatively small territory.Critical infrastructures are multidimensional systems of systems with highly complex inter and intra-dependencies of collections of technologies, processes, and people.U.S. Patriot Act defines the critical infrastructures as "the systems and assets, whether physical or virtual, so vital to [the state] that the incapacity or distraction of such systems and assets would have a debilitating impact on security, national economic security, national public health or safety, or any combination of those matters" (Sec.1016(e)) [1].A similar definition is given also by the European Commission: "critical infrastructures are those physical facilities and networks, which, if disrupted, would significantly affect the health, safety, security or economic well-being of citizens" (COM(2004) 702 final).Failure in one component of a system of systems such as the critical infrastructures within a city may have cascading effects not only to the network to which the component is part of, but also to the other dependent and interdependent networks.To exemplify, a failure of a power transformer serving an area within a city may also lead to water supply shortage in the same area or even a larger part of a city/village, e.g.due to the interruption of a pump operation or at least creating low pressure conditions in the water distribution networks (WDN).To be noted that WDN depend heavily on power to operate their pumping system that deliver water for consumption.Interdependency modeling became key field of study since early 2000s, with a number of modeling approaches proposed over the years, which can be summarized in several works [2][3][4][5][6][7].Some of the most popular interdependency modeling approaches is the inoperability input-output methods, agent-based modeling approaches and network based approaches.The inoperability input-output model (IIM) can estimate at a holistic level the inoperability (i.e. the % of multi-function) of infrastructures using the dependency coefficients (also known as Leontief coefficients) [8].Agent-based approaches consider CIs as complex adaptive systems and represent their components as agents, with interdependencies usually emerging from the interactions of agents [9,10].Network-based approaches generally assume that each CIS consists of a set of components (usually represented as nodes) forming a network, and any existing interdependency is represented as a type of relationship between nodes belonging to different networks [11,12].There are also several other approaches, few are based on various probabilistic methodologies (e.g., petri nets, stochastic activity networks and Bayesian networks) [13][14][15], some others are considering multi-layer modeling approaches, where infrastructures are seen at different layers and interactions between them are considered at different levels of granularity [16], and also some empirical approaches that analyze CIs interdependencies according to historical accident or disaster data and expert experience [17].From our research we believe hybrid systems and their open hybrid automata are also great for modeling infrastructures interdependencies [18].Their main advantage is that they can model components from any Critical Infrastructures (CI).Thus, they can be seen as a common modeling framework.Moreover, open hybrid automata allows the development of models at various levels of abstraction, i.e. very detailed models with many variables or quite simple ones, depending on the modeling objectives and the available data.
It is becoming widely accepted that technology forms the bone for empowering governments and city administrators with new ways of serving their citizens.Terrorist attacks and cyber intrusion actions on public or private owned assets, raised the concern of many governments to take actions in treating all information related to critical infrastructures as classified [19].Without real data about these systems, the research community must develop virtual but realistic test systems such that they can test and validate the performance and applicability of their methods.Several test systems were developed over the years as benchmarks to test models or applications related networks (either for steady state models or dynamic) but they were mainly designed for one specific network such as power systems [20], water networks [21] or telecommunication networks [22].
The concept of "virtual city" emerged from computer gaming domain.Data visualization techniques are used to enable decision makers to see analytics in a pictorial manner such that they can grasp difficult concepts or identify new patterns.Thus, in recent years we are seeing a high interest in software developments for urban planning and smart e-governance of cities based on the integration of "virtual cities" , data analytics geographically stamped and augmented reality [23], [24].The state of the art with respect to integration of data for critical infrastructures in these virtual cities is however limited if non-existent.A library of two virtual cities, basically focused only on the urban topology and water distribution networks was developed by a group of researchers at Texas A&M University [25,26].To be noted however that only the small virtual city, called Micropolis, has been actually detailed in terms of two (only) critical infrastructures (water and power distribution systems), while the second city has details only with respect to geographical, urban building assets and demographic data.The electrical distribution layer of Micropolis is documented with respect to several components and single-line diagram connections in [19], however the corresponding GIS files for this layer are still not publically available.Another missing aspect is in the design itself of the power distribution network of Micropolis with respect to the presence of renewable power generation, as well as universality of the design in a city distribution power network.
In this work we took the challenge in designing the power distribution system layer as well as the cellular telecommunication layer of Micropolis.This is an ongoing work and in the coming future the authors plan to develop more layers of several other infrastructures such as transportation, gas, health system etc.The second aim of this work is to use realistic design methods, in sync with the current trends of developments toward smart cities (e.g.including advanced technologies for power distribution grids, water distribution networks and advanced telecommunication networks), that lead to a comprehensive test system, fully integrated in a powerful visualization tool such as QGIS.It is the authors hope that this work will serve the research community with the right tool to test and validate their solutions related to critical infrastructures within a smart city context as well as providing an integrated middleware for their data analytics visualization and communication of research results.
The reminder of the paper is structured as follows: first a description on the virtual city called "Micropolis" is provided, then we dig into the theoretical background in developing generic test systems for critical networks such as water distribution networks, power systems, telecommunication networks, then we detail the design methodologies and tools used in developing each sub-system layer and at the end we conclude the work and provide hints for further research and applications in which the developed test system can be used.

Micropolis
The geographical layout of the virtual city of Micropolis was developed by a group of researchers at Texas A&M University.The design follows the development pattern of a small city of 5000 inhabitants over a period of 130 years at the beginning of the 19 th century.The city covers a relatively small area, of about 3.6 km 2 , has one main water source and one water stream.The terrain is almost flat, with very little elevation differences between one part and another of the town.There are three major destinations for the stock of buildings within Micropolis, such as: residential (single house and multiple apartment buildings), commercial and industrial.The number of inhabitants was randomly selected per type of building and according to their volume/surface occupied.
The legacy data available at the beginning of this work was the city's buildings, land use map, road and railway systems, as well as the water distribution system layers in GIS format.For the detailed description of the design of the water distribution network of Micropolis, as well as the main assumptions of this design the reader may consult [25].
In Table 1 bellow, we summarize the most important characteristics of Micropolis that impact the design of the three infrastructures we developed and are presenting in this paper.

Water distribution network layer
In this section we present the extension, we have made to the water distribution layer already available in order to enhance our study from the critical infrastructure point of view.Due to their vital role, water systems are considered among the critical infrastructures, along with power and telecommunications systems.Water systems are interdepended to other critical infrastructures, since power is needed to operate pumps, and telecommunication is required to facilitate the operation of the SCADA system.In addition, if an event occurs, such as a pipe burst, it may affect power supply and transportation in the area.
Water quality sensors may be used for monitoring the system for hydraulic events, such as leakages, as well as for quality events which may be caused due to accidents or even malicious attacks.In general, sensors which monitor water quality are important for improving the system's security.Typically, however, utilities have a small number of water quality sensors to install in a large-scale network, due to their highcosts.Deciding the best locations to install these sensors, is a challenging research task.

The quality sensor placement problem
The problem of deciding where to install water quality sensors within distribution networks, for enhancing their monitoring and security capability, has been widely investigated within the last decade by the water research community [27].In most works, sensor placement has been addressed as an optimization problem which aims to choose a finite subset of nodes out of the set of all the network nodes where it is feasible to install sensors, by minimizing a set of objectives (e.g.risk) with respect to certain impact metrics (e.g. the number of people infected) [28], [29].Various challenges have been identified in research, which affect the sensor placement solutions: the uncertainties in the model and the water demands, the impact metrics and the risk objectives selection, the contamination scenarios selection, the sensor measurement uncertainties, the response time delays, the solution methodology and its computational efficiency [30][31][32][33].The state-of-the-art in application software is the TEVA-SPOT, which is available under an open-source software license [34].The lack of platform which could be used for scientific experiments and benchmarking has motivated KIOS in developing and releasing an open-source software, the "Sensor PLACEment Toolkit" (S-PLACE), implemented in Matlab, for computing sensor placement solutions, based on the mathematical framework proposed in [29].Figure 1 depicts the modular software architecture.The software has been designed to be user-friendly, both for the professional as well as the academic community, making it easy to evaluate solutions under various scenarios.In addition, its GUI provides an intuitive way of interfacing with the software and with the water distribution network.The software architecture is modular, and each module can be accessed independently through stand-alone functions.Furthermore, the S-PLACE is extendible, as it allows to add, modify or remove methods and network elements, in accordance to the research objectives.For instance, new risk functions can be programmed and used in optimization, new scenario selection algorithms can be evaluated, or new nodes and pipes can be added to the network.The use of the software is illustrated on several benchmark networks which capture different types of real network topologies, such as looped and branched networks.The software GUI is depicted in Figure 2

Micropolis EPANET Model
Water distribution networks can be modelled using the EPANET standard and simulated using the EPANET libraries.In general, water distribution networks can be modelled as multiple layers in a GIS (pipe, junction, tank, reservoir, pump and valve layer) and through transformation they can be integrated into an EPANET model.A visual description of the Micropolis EPANET model is provided in Figure 33.Each node in the graph has specific parameters, such as a base demand (i.e. the average daily water consumption), and an associated demand pattern which captures the variation of water consumption throughout the day.The EPANET model is comprised of

Quality Sensor Placement for Micropolis
To solve the quality sensor placement problem for Micropolis, a large number of contamination scenarios was simulated and analyzed.In specific, 1915 random scenarios where produced (i.e injection of contaminant with 1 mg/L concentration at a single node within the first 24 hours), with a simulation time of 48 hours.No uncertainty was considered in the system parameters.For measuring impact, the Contaminated Water Consumption Volume metric was considered.It calculates how much contaminated water is used for consumer demands until it is detected by some sensors.The S-PLACE system searched for the optimal location of two sensors with respect to two risk objectives, which minimize the average and the maximum impact.An evolutionary multi-objective optimization method was used to identify the Paretofront solutions, which are provided in Table 2.

Concepts of system design for power distribution networks
Distribution power networks were traditionally designed only to deliver power to loads, thus unidirectional power flow rules applied.Hence, increasing distributed generation (DG) penetration from photovoltaic (PV) systems, small wind turbines or even backyard thermal generators is causing profound changes for Distribution System Operators (DSOs) in planning, operation and maintenance of distribution networks.Thus, planning a distribution network with renewable power generation is in itself a challenging task due to little design tools available as well as little state of the art practice in doing so.Besides the technical aspects with respect to modeling and availability of "on the shelf analysis tools", there is also the regulatory apparatus that also impacts the power distribution network planning methods in unbundled electricity markets.Thus, our design methodology relies on the following assumptions: • From the DG possibilities, we choose that only PV systems are integrated at the LV side of the distribution network of Micropolis.• 30% of randomly selected houses from the total buildings asset of Micropolis have PVs installed on top of their roof.This is indeed a close to reality scenario taking into account the European Climate change targets, as well as the evolution of PV installations up to now.The random location of PV systems was based on the fact that currently there is no actual restriction with respect to the location of the PV installation, as long as the National Grid Codes are respected.• Small residential buildings are supplied with single-phase circuit distribution lines, while the medium size residential and commercial buildings are supplied by treephase circuit distribution lines.• The MV system was modeled as a voltage source with impedance specified by the short circuit currents.
The model of the distribution power network of Micropolis followed the approach of the design of a typical European Low Voltage feeder test system, as documented by the IEEE Test Feeders Working Group [35].Thus, in our design the low voltage power distribution network of Micropolis is a radial distribution feeder with a base frequency of 50 Hz.The feeder is connected to the medium voltage (MV) system through a transformer at substation.The transformer steps the voltage down from 11 kV to 416 V.The main feeder and laterals are at the voltage level of 416 V.The oneline diagram of the test feeder is shown in Figure 4.
The starting point in planning and sizing the power distribution network of Micropolis is to foresee the equipment to be connected and the resulting total power demand.Thus, the building destination is the major parameter to be taken into account for the choice of distribution equipment and wiring.Table 3 below summarizes the various areas of land use and buildings destinations and their impact on the requirements for the electric distribution grid and equipment.

Busbar trunking systems
There are three basic configurations that can be adopted for power distribution grids, such as radial networks, ringed networks and meshed networks.The simplest and widely adopted in small towns and rural areas is the radial network configuration.This configuration was used in the design of the power distribution network of Micropolis.For the estimation of the power demand of each type of loads in Micropolis we have used the data provided in [36].Besides the average estimation of the necessary power demand of each load one may take into account a set of calibration factors according to the building destination, such as: calibration factor for the building placement (kplc), calibration factor for the room structure (kstruct), calibration factor for the level of comfort (kcomf), calibration factor for the air conditioning options (kclim), calibration factor for the technical characteristics (ktech), calibration factor for building management (kBA/TBM) .Thus, the total estimated power demand for a specific building (end-load) calculates as, Where   is the estimated power demand of a specific type of building (land use),   is the average estimated power demand, and   is the total calibration coeficient, calculated as,   = (  +   +   +   +  ℎ +    )/6.
According to the total estimated power demand, one may calculate the sizing of the main feeds as well as the sizing of the substation transformer.For detailed calculations, one may refer to [36].

Concepts of system design for cellular communication networks
A wireless communication network [37] is designed under the system-level idea of the "cellular concept"; instead of using a single, high power transmitter, a large number of low power transmitters are utilized, each one covering a small portion of the service area.In this way, all the available channels are assigned to a small number of neighboring base stations.These neighboring stations are assigned with different groups of channels, so as to minimize interference.The result is that the available channels may be reused many times.The geographic area served by a certain base station is called cell.The design process of selecting and allocating channels to the base stations is called frequency planning.The actual radio coverage of a cell is called "footprint" and it is determined from field measurements or propagation models.

Channel Assignment strategies
Channel assignment can be either fixed or dynamic.For the case of fixed channel assignment, a predetermined set of channels is assigned to each cell.If all the channels of a cell are occupied, an upcoming request for call is not established.One ap-proach that is applied to overcome this problem is the borrowing strategy, where a cell borrows channels from neighboring cells so as to establish a request for a new call, if its own channels are completely occupied.This procedure is supervised by the Mobile Switching Center (MSC).For the case of dynamic channel assignment, channels are not permanently allocated to cells.Each time a call request is made, the corresponding base station requests a channel from the MSC, and the latter corresponds to this request according to a certain algorithm.Dynamic channel assignment leads to decreased percentage of blocked cells, compared to the fixed case.

Handoff Operation
During a call, it is possible that the user is moving, thus shifting to a different cell.In this case, the MSC transfers the call to a new channel belonging to a new base station (handoff operation).Processing handoffs is an important task in any cellular radio system.Many handoff strategies prioritize handoff requests over call initiation requests when allocating unused channels in a cell site.Handoffs must be performed successfully and as infrequently as possible, and be imperceptible to the users.In order to meet these requirements, system designers must specify an optimum signal level at which to initiate a handoff.Once a particular level is specified as the minimum usable signal for acceptable voice quality at the base station receiver, a slightly stronger signal level is used as threshold at which a handoff is made.

Interference
Interference is the major limiting factor of the performance of a cellular radio system.Sources of it can be another mobile in the same cell, an established call in a neighboring cell, other base stations, or other, non-cellular systems.Interference can negatively influence the quality of a call in progress.It can also cause problems to the control channels, leading to missed and blocked calls.1) Co-channel Interference: Frequency reuse implies that in a given coverage area there are several cells that use the same set of frequencies.These cells are called cochannel cells and the interference between signals from these cells is called cochannel interference.The latter cannot be cancelled by simple increasing the carrier power of a transmitter.This is because an increase in carrier transmit power increases the interference to neighboring channel cells.To reduce co-channel interference, cochannel cells must be physically separated by a minimum distance to provide sufficient isolation due to propagation.
2) Adjacent channel Interference: Interference resulting from signals which are adjacent in frequency to the desired signal is called adjacent channel interference.Adjacent channel interference results from imperfect receiver filters which allow nearby frequencies to leak into the passband.It can be minimized through careful filtering and channel assignments.Since each channel is given only a fraction of the available channels, a cell need not be assigned channels which are all adjacent in frequency.By keeping the frequency separation between each channel in a given cell as large as possible, the adjacent channel interference can be considerably reduced.

Trunking and Grade of Service
Cellular radio systems rely on trunking to accommodate a large number of users in a limited radio spectrum.The concept of trunking allows a large number of users to share the relatively small number of channels in a cell by providing access to each user, on demand, from a set of available channels.In a trunked radio system, each user is allocated a channel on a per call basis, and upon termination of the call, the previously occupied channel is immediately returned to the set of available channels.Trunking exploits the statistical behavior of users so that a fixed number of channels may accommodate a large, random user community.There is a trade-off between the number of available channels and the possibility that a particular user will not be able to find an available channel during the peak calling time.The Grade Of Service (GOS) is a measure of the ability of a user to access a trunked system during the busiest hour.The busy hour is based upon customer demand during a certain period of time.The GOS is a benchmark used to define the desired performance of a particular trunked system by specifying a desired likelihood of a user obtaining channel access given a specific number of channels available in the system.GOS is typically given as the likelihood that a call is blocked, or the likelihood of a call experiencing a delay greater than a certain queuing time.

Techniques for Improving Coverage and Capacity 1) Cell Splitting:
Cell splitting is the process of subdividing a congested cell into smaller cells, each with its own base station and a corresponding reduction in antenna height and transmitter power.Cell splitting increases the capacity of a cellular system by increasing the number of times that channels are reused.These smaller cells are called microcells.
2) Sectoring: Another way to increase coverage is to replace the (single) omnidirectional antenna at the base station by several directional antennas, each radiating within a certain sector.By using directional antennas, a given cell will receive interference and transmit with only a fraction of the available co-channel cells.When sectoring is applied, the channels used in a particular cell are broken down into sectored groups and are used only within a particular sector.
3) Range Extension Using Repeaters: Radio transmitters, called repeaters, are often used to provide range extension.They simultaneously send and receive signals from a serving base station.They work using over-the-air signals.Therefore, they can be installed everywhere.Upon receiving signals from a base station forward link, the repeater amplifies and reradiates the base station signals to the specific coverage region.
There are several algorithms used in the literature for base station placement in cellular communication networks.Some indicative algorithms are the following: • k-Center Algorithms [38]: The k-center problem refers to the case of finding a subset of nodes in the network such that the maximum distance from any other node to the closest one of the selected nodes, is minimised.Algorithms that solve this problem can be utilised so as to allocate the base stations int eh cellular network in a way that every point in the network will be close enough to at least one base station, so as to be able to be served by the latter.

•
Genetic Algorithms [39]: These are search heuristics that imitate the process of natural selection.They belong to the larger class of evolutionary algorithms (EA), which generate solutions to optimization problems using techniques inspired by natural evolution, such as inheritance, mutation, selection and crossover.

•
Simulated Annealing [40]: It is a metaheuristic to approximate global optimization in a large search space.

•
Tabu Search [41]: It is a metaheuristic search method employing local search methods used for mathematical optimization.It constitutes an enhancement of the Neighbourhood Search.

Figure 5 QGIS layer of Cellular Network: Black antennas are the micro cells; coverage area of the microcells is the circle in light green; pico cell antennas are small red, and their corresponding coverage areas are the orange circles
In our implementation we have applied the k-Center Algorithm.We are assuming the following: Micropolis being a small city (near the size of a village), we have chosen to use only two types of cellular in our cellular communication network: micro and pico cells.The reference area to place micro-cell antennas (up to 1300 m coverage radius) is the entire territory of Micropolis.Thus, we first created heat maps with QGIS to determine the areas where pico-cells shall be installed (coverage area up to 100m in radius), and then we apply the k-mean algorithm to place the two types of antennas.The implementation of the algorithm was done in Matlab, and the results (x,y axis of each antenna placement point) were then parsed to QGIS for actual positioning, as it is presented in Figure 5.

Framework for interdependencies modeling
The normal operation of the three Micropolis infrastructure systems (water, power and communication) heavily depends on the various interdependencies that exist among them.For instance, the water and communication systems receive electricity from the power system to feed the water pumps and the cell antennas, respectively, while the power and water systems rely on the communication system for data transmission for their monitoring and control.If interdependencies break due to failure in one system, there will be a disruption to the normal operation of the other systems, thus is important to study them and develop models to analyze their cascading effects.
To facilitate the identification, understanding, and analysis of interdependencies, we usually classify them based on their characteristics into the following four principal types [28]: (a) physical, if the operations of one infrastructure depends on the physical output(s) of the other and vice versa, (b) cyber, if there is information/signal transmission between different infrastructures, (c) geographic, if components of different infrastructures are in close spatial proximity, and (d) logical, due to any other mechanism (e.g.various policy, legal, or regulatory regimes) that can link logically two or more infrastructures.Interdependencies are bidirectional relationships while dependencies are unidirectional relationships.When multiple infrastructures are connected as "system of systems" and their individual components are considered, interdependencies are the result of multiple dependencies between the components of different infrastructures.These are often called external dependencies and they can be the same type as interdependencies i.e., physical, cyber, geographic, or logical, but just unidirectional.There are also internal dependencies, which are the connections between components inside the same infrastructure, for instance the connection inside the water infrastructure between a pump and a tank is an internal dependency.
For the city of Micropolis the main components of the three systems and their internal and external dependencies are shown in Figure 6.Specifically, the following components are considered: for the power system (1) the substation that supplies power to Micropolis and (2) the SCADA that remotely monitors and controls the substation; for the communication system (3) the network that consists of a number of microcell antennas; and lastly for the water system (4) the tank that supplies water to Micropolis and (5) the pump to which the tank is dependent for water, thus there is an internal dependency between the two.As external dependencies among the components three types are considered: (i) the physical dependencies of the pump and the network on the substation for power supply; (ii) the cyber dependencies of all the components on the network, i.e., the communication links to the substation, the SCADA, the pump and the tank; and (iii) the logical dependencies such as the power and water demands for the substation and the tank, respectively.Although there are geographic interdependencies between all the components, due to the close spatial proximity between them (Micropolis is a small town), we don't consider them in this work since they usually have an effect during explosions, floats, earthquakes etc. and given the size of Micropolis such events would affect all the components at once without any cascading effects between them.In the next subsections the open hybrid automata models for the main components of Micropolis are presented and then the composition model is derived based on the dependencies among them (see Figure 6).The open hybrid automata models for the power substation, SCADA, are similar with those presented in the work of the second author of this work, [18], with minor modifications mostly to the variables names to make them less generic and more relevant to each infrastructure component.Descriptions for each model variables are summarized in Table 4.

PDN Components -Substation and SCADA
The open hybrid automaton model for the Power Substation is shown in Figure 7 and is the same as the one in [18].The model consists of two discrete states that represent the behavior of the substation i.e., Supply Power or Switch Off.When in Supply Power state the output   =   (power supply to Micropolis equals to the power demand) and when in the Switch Off state the output   = 0, also   , that denotes the power measurements, changes accordingly.The transitions between the two discrete are represented by the guards.For instance, the transition from the Supply Power to the Switch Off state will occur either if its requested remotely from the SCADA (  = 1) or due to safety reasons when the power demand is larger than the limit of the substation (  ≥   ).More detailed explanation for the model can be found in [18].The open hybrid automaton model for the SCADA is shown in Figure 8.The model is also the same as the one in [18] and represents the SCADA behavior according to the power measurements input (  ) and the operator input (  ).The SCADA model will be at the Close state if the substation is in the Supply Power state, representing the state of the switches at the substation, or at the Open state if the substation is in the Switch Off state or the operator decides to remotely cut the power for maintenance.In case of fault at the network it will go to the Conn.Down state.The discrete transitions are determined by the two inputs (  ,   ) and also by the continuous state    that acts as a timer while in the Close and Open states.

CN Component -Network
The open hybrid automaton model for the Network is shown in Figure 9.The model is an extension of the work [18] of one of the authors with the addition of an extra signal input ( 3 ) that represents the communication between the water tank and the pump.The model consists by three discrete states that denote the behavior of the network.In the Healthy state network operates normally providing communication services to Micropolis.In case of a power cut, where the power is not enough to feed the antennas (  <   ) they will switch to UPS, thus the model transitions to the UPS usage state.However, the UPS can last for certain amount of time and this is what the continuous state    is counting.Once it reaches the limit (   ≥   ) the network will stop operating and this is represented in the model by transitioning to the Net Down state.Transition to the Net Down state is also triggered in case of a technical fault in the network which is denoted with the input   .More detailed explanation for the model can be found in [18].

Composition Model
The five open hybrid automata models are composed together as shown in Fig. xx7, creating the composition model, a larger model that includes the various dependencies between the components.As elaborated above, there are physical, cyber and logical external dependencies between the components and internal dependency between the tank and the pump, that are all depicted in detail in Figure 12.In the composition the component models run in parallel with the output of one model to become the input to the other, developing various feedback loops between them, which represent infrastructure interdependencies.For example, both the power substation and the SCADA use the network to transmit power measurements and control signals between them, as depicted in Figure 12 with the necessary connections.The network is also depended on the substation model for power.These dependency connections create feedback loop between these three models, which subsequently form interdependency between the power and the communication infrastructure.This allows the study of cascading effects between the various components by running simulation scenarios with the composition model as is presented in the next section.

Simulation Results and discussions
The components models and their composition model, as described in the previous sections, were implemented in Matlab's Simulink software with the purpose to study the cascading effects between the different infrastructure components due to (inter)dependencies.To achieve this, the composition model can be simulated for a number of scenarios where different components set to fail at a specific times and then observe how the state of other components changes.Thus, Figure 13 and Figure 14 show the results of one such simulation scenario.Specifically, Figure 13 shows the simulation scenario timeline where, at the top, is the duration of the induced event at specific component, and at the bottom the consequences of that event to other components.Figure 14 shows for each component how the discrete states evolve and some of the inputs/outputs or states of the model, during the simulation scenario.

Conclusions
Critical infrastructure analysis with application to smart cities management is an emerging field of study with tremendous importance for the entire society.Test systems for such analysis are in need, especially due to the lack of public data available in the case of several real critical infrastructures such as power distribution systems, telecommunication systems, and water distribution systems.This work presented the design and development of such a test-bed system under QGIS platform, with the scope of make it available for the research community to serve in analysis to interdependencies and impact studies related to natural or manmade disasters or for developing strategies for preparedness in such events.We have also presented a specific application of analysis using the developed test system.

Figure 1 Figure 2
Figure 1 Software Architecture of S-PLACE

Figure 3 EPANET
Figure 3 EPANET Model of Micropolis.Reservoirs correspond to green squares, tanks to cyan stars and pumps to purple triangles.Filled circles correspond to junctions which link pipes or are locations where water is consumed.Two monitoring stations are assumed in the locations indicated with red filled circles.

Figure 4
Figure 4 Power Distribution network of Micropolis: the red towers are the power substations; light orange line is the medium voltage distribution line; brown lines are the distribution feeders; and yellow diamonds are the PV systems installed.

Figure 6
Figure 6 Overview of Micropolis main infrastructures components and their dependencies.To model the main components of Micropolis and the dependencies between them in order to study possible cascading effects we use open hybrid automata as proposed in[18] which the reader is encourage to review for more insides.The open hybrid automata are used as models for hybrid systems, i.e., systems with both continuous and discrete behaviors such as critical infrastructures, for instance in a water system the start and stop of a pump is discrete while the flow of water is continuous.With open hybrid automata is possible to model the behavior of each critical infrastructure component at the necessary level of abstraction, i.e., how the component will behave in case it fails or if there is no power supply or communication available.The dependencies among the components on the other hand, are represented by the connections between the inputs and the outputs of the various open hybrid automata models.Formally, connecting the inputs and the outputs of two or more open hybrid automata is called composition, and the result is another open hybrid automaton bigger and more complex that has the remaining inputs and outputs.The new open hybrid automaton is often referred as the composition model and consist with all the open hybrid automata running in parallel.The composition model is the one that can be simulated for various scenarios where component(s) set to fail at specific time(s) and observe the cascading effects to the other components.

Figure 7
Figure 7 Power substation open hybrid automaton model.

Figure 8
Figure 8 Power SCADA open hybrid automaton model.

Figure 9
Figure 9 Network open hybrid automaton model WDN Components -Tank and PumpThe tank receives and stores water from the pump and supplies Micropolis based on the demand.This has to be controlled accordingly since too much water from the pump can overflow the tank while less water can drain it.The tank open hybrid automaton model is shown in Figure10.The continuous state   denotes the volume of the tank that changes according to ̇ =   −   , where   denotes the water supply rate from the pump, and   the water demand rate of consumers.The discrete states of the model are determined by the tank volume.For instance, the model will be at the "Healthy" state while 0 <   ≤where   is the tank's maximum volume, and it will transition to either the Drained state if   ≤ 0 or to the Overflow state if   >   .The model will return back to the "Healthy" state if the water supply becomes larger than the demand   >   and is in the Drained state, or if the water demand becomes larger than the supply   >   and is in the Overflow state.The single output of the model   denotes the tank's volume measurement depending on the state, and is transmitted to the pump through the network so that the pump can start and stop as descripted next.

Figure 10
Figure 10 Tank open hybrid automaton model.The pump supplies Micropolis tank with water by receiving and comparing the tank volume measurement with some threshold, if the tank volume goes below the threshold value then the pump will start given that the substation provides the necessary power.There are also restrictions to the pump operation such as maximum working period and minimum resting period.The pump open hybrid automaton model is shown in Figure11.The model has three discrete states: (i) the Pump Off when the

Figure 11
Figure 11 Pump open hybrid automaton model.

Figure 13
Figure 13 Simulation scenario timeline with induced events/faults at the top and their subsequent consequences at the bottom.

Figure 14
Figure 14 Plots showing plots from each component associated with the simulation scenario timeline.From the results is clear that the composition model can represent cascading effects due to dependencies.For instance, when the Power Substation due to overload Switch Off at 05:58 the Network immediately switches to UPS use and the Pump remains without power.Also a Network fault between 15:13-16:52 causes the SCADA to immediately lose communication and then at 15:34 the Tank to overflow since the Pump does not receive measurements for the tank volume to stop once the tank is full.Finally, when at 20:13 the SCADA operator remotely switch off the Power Substation due to maintenance, at first the same consequence events during the overload at 05:58 occur, then since the substation remains off for more than an hour the UPS at the Network depletes (at this scenario UPS is set to hold the network operational for 1 hour of power cut) leaving Micropolis without communication services and the SCADA operator unable to remotely control the substation.

Table 2
Pareto Solutions for the 2-sensor placement problem

Table 3
Land use and its impact on the electric grid equipment

Table 4
Models variables descriptions