Supporting Intelligent Control Design of Rail Infrastructures

Transcription

1 Supporting Intelligent Control Design of Rail Infrastructures Elisangela Mieko Kanacilo Alexander Verbraeck Delft University of Technology Faculty of Technology Policy and Management Systems Engineering Section P.O. Box 5015, 2628BX, Delft, The Netherlands {elisangelak, Abstract Currently, rail control designers are looking for innovative types of control strategies in order to make rail infrastructures more robust, more reliable, and more efficient. Control designers are focusing on more flexible control strategies in order to keep the system performance stable when non planned situations occurs. Applying new control strategies to rail infrastructures is difficult, because strategies cannot be fully assessed for their quality before implementation. Existing testing tools normally only assess the strategy itself and do not consider its impact on the whole system, nor do they assess the impact of the system on the strategy. This issue was confirmed during an inductive case study performed at a Dutch transportation company. To tackle this problem, a simulation based environment is proposed to better assess control strategies for rail infrastructures. Keywords: Control design, rail infrastructure, simulation, modeling, decision support system. 1 Introduction Several trends are causing infrastructure control designers to look for new types of control strategies [1]. Examples are increasing competition within sectors, infrastructures becoming more interdependent (for example, trams depending on energy to drive), infrastructures being more connected to each other (e.g. tramlines sharing space with cars and buses), and users demanding more reliable and efficient transport services. These changes can for instance be noticed in the transport sector. Rail infrastructures were chosen as the domain to be analyzed in this study. Apart from changes in the market that influence the way the system has to be controlled and thereby influence its performance, rail infrastructure can be easily disturbed by many external factors, such as bad weather conditions, accidents, and equipment breakdown, which impact the quality of the service provided (delaying passenger or goods transportation). These disturbances cannot be avoided, but the infrastructure can be designed and controlled in such a way that it has more flexibility to deal with these factors. In this research, we do not focus on the re-design of the rail infrastructure itself. After a rail system has been built, the physical part is hardly ever changed, because these changes would involve many stakeholders (local authorities, company managers, etc.), they are expensive, and they might disrupt the service itself. Introducing flexibility through the control of rail systems is a more adequate solution and this is the focus of our study. Because of the dynamic 1 environment of rail infrastructures intelligent control strategies are becoming necessary [2] to improve the system efficiency, reliability and robustness. The word intelligent here relates to the definition of Intelligent System, which means a system that acts appropriately in an uncertain environment, where an appropriate action is that which increases the probability of success [3]. Introducing new control strategies in rail systems is not an easy task. Because of their distributed nature, rail infrastructures have many distributed actors (local control officers) whose control decisions can influence the performance of a strategy taken by another actor. In this dynamic and multi actor environment it is difficult to assess the consequences that a set of strategies will have on the system considering the interactions among multiple actors. A more integrated testing environment becomes necessary in order to evaluate the effects of a control action taking into account the whole system configuration (local versus global optimization). Simulation models used to support the assessment of control decisions are normally implemented based on set of requirements of a specific project and therefore, reutilization of models is not always possible [4]. Models need to be built from scratch for each project, making the 1 [5] distinguish between first order dynamics the day-today changes, and second order dynamics, the changes in the system over time.

2 design phase of models too long and expensive. Common off-the-shelf simulation packages present interoperability problems. As described in [6] wrappers used to link one model to another make simulation environments unstable and hard to maintain, because they normally have code of their own in addition to the model logic. A component based solution seems to be appropriate to solve this problem [6]. In this paper, we propose a component based simulation environment to support the assessment of rail control strategies in a multi actor and distributed system. This simulation environment can be used to support the decision-making processes during the design and operation phases of rail infrastructure control. This document is structured as follows: Section 2 is a description of an inductive case study performed at a Dutch transportation company, Section 3 is a summary of available theories about simulation and support control tools; in Section 4 the proposed simulation environment is described; and we draw conclusions in Section 5. Section 6 mentions the next research challenges. 2 HTM Case Study HTM Personenvervoer NV is a Dutch transportation company that offers collective passengers transport services (with busses and trams) for the city of The Hague and the surrounding region. The case was focused on understanding how the control of their rail network is designed and operated. HTM is extending its infrastructure to introduce light rail 2 transport services. For that reason, new tracks are being built and some of the new tracks will be connected to existing ones. This innovation requires control designers to formulate new control strategies for the new area and a special attention needs to be given to the area where new tracks join the existing network, because the traffic flow of light rail trams will influence the timetable of other tram lines. In addition, infrastructure capacity needs to be planned considering that in some parts, tracks will be shared between the two types of trams. The safety system for light rail vehicles is different from the one used in traditional trams, therefore in the areas where tracks will be shared by the two types of trams, different control strategies need to be applied to address the safety requirements of both kinds of vehicles. The safety measurements (e.g. how fast trams can drive in a curve without offering risk to passengers) are considered part of 2 Light rail transport uses vehicles that combine features of both trams and trains. Light rail vehicles are lighter than trains and faster than trams, and are normally used to connect more distant locations. the control design because they influence the throughput of trams, and thereby the timetable. In this case, one of the problems faced is that infrastructure capacity is mainly planned based on controller s experiences and on the analysis of historical data. The maximum throughput of trams per hour for the shared tracks was estimated based on the analysis of the number of trams driving on existing tracks. The company uses monitoring tools, like TRIPTAPT 3 during operation where journey data are gathered and stored. These are the data which control designers use to make adjustments to the applied control strategies, when necessary. Allowing control strategies to be adjusted after commissioning gives a certain flexibility to the system, but these changes should be anticipated as much as possible, because having a poor system performance can result in high economic costs for the organization and inconvenience for passengers. Therefore, control strategies should be better assessed and for this case, a more precise estimation of infrastructure capacity should be given. Making a more precise estimation is a big challenge, because the control policies applied in the area were not completely assessed as there was no readily available technological support for detailed infrastructure capacity analysis. Deciding on the capacity of a new infrastructure based on historical data of an existing system might lead to wrong conclusions, as control policies might differ from one area to another, and as they influence the traffic flow of which the historical data has been used as input for the analysis. Examples of decisions that should be made for the HTM infrastructure are whether or not to give priority to a certain tram line at crossings, the speed limit to satisfy the safety requirements, which is different per type of vehicle, the distance of the traffic light sensors, considering the physical characteristics of each type of vehicle (acceleration and deceleration rate) and many others. Therefore, for each scenario, specific conditions and control policies should be considered in order to have a more precise assessment. With all these factors to be taken into account when designing rail systems control, one can conclude that it is a quite complex scenario to analyze and using only human expertise and analysis of historical data is not enough. In addition, having all these different system situations in a physical experiment is not desirable. Transport infrastructures, in general, are considered critical infrastructures [7], which mean society (people and other organizations) relies on transport services of good quality to perform their daily tasks. Therefore, experimenting in the real setting should definitely be avoided. 3

3 The lack of technological support for the assessment of control decisions regarding the planning of both infrastructure capacity and tramlines schedules was identified as a major challenge. We argue that simulation would help in providing control designers and operational controllers with a better understanding of the new system configuration and to test (combinations of) different strategies, after which a choice can be made for the most appropriate combinations of control strategies. 3 Simulation to Support Control Design As stated in [8], simulation is the process of designing a model of a real system and conducting experiments with this model for the purpose either of understanding the behavior of the system or of evaluating various strategies for the operation of the system. With this definition, we argue that simulation is a very suitable method to gain insight in the complex environment of rail infrastructure and to support the decision-making process during the control design and its testing for effectiveness. Indeed simulation has been widely used to support the control design of logistics systems [9], [10] and transportation systems [11], [12]. The history of simulation shows that in early years, simulation models were built as single use models [13]. Models were developed in a very detailed way and reutilization was not always possible, making the design phase long and expensive. In the last decade, the simulation industry is focusing on extending the useful life of models, by developing generic components that can be used on an ongoing basis [4]. Models can be used throughout the system life cycle [14]. Therefore, to extend the life cycle of models given the current market scenario [13], three basic requirements need to be provided: simulation models need to be generic enough to be used in different phases of the system life cycle (design, testing and operation). Models can be generic within a specific domain (rail, in this case); simulation models need to interoperate with other applications (data gathering systems, databases, control systems, geographical information systems, etc.); simulation models have to be accessible from different locations to support decision making in more complex and multi actor environments. To satisfy these three requirements, we argue that a component based solution would form the basis of a generic solution, and it would facilitate the interoperability between simulation model components and other systems. Component can be defined [6] as the implementation of a building block in a software environment. A building block [15] is a self-contained, interoperable, reusable and replaceable unit, encapsulating its internal structure and provides usable services to its environment through precisely defined interfaces. 4 Simulation Library to Support the Design of Rail Systems Control In the HTM project, we developed a library of components to simulate the behavior of rail-based infrastructure elements and their control logic. Our simulation environment has been developed in Java and it is based on DSOL Distributed Simulation Object Library [16], an open source 4, Java based simulation suite that allows for: object oriented modeling; distributed modeling and the combination of multiple formalisms in a single model. Using the simulation services provided by DSOL we created an architecture consisting of the following modules: - A library of rail infrastructure components: an abstraction of the basic elements of a rail infrastructure such as tracks of different shapes, traffic lights, speed signs, different types of sensors, stations and many other rail infrastructure elements. These components are generic and loosely coupled so that they can be plugged to each other or unplugged without interfering with the rest of the model, which is a problem that is often visible when components are linked in a tightly coupled way. Component behavior can be based on different formalisms (continuous or discrete; event scheduling or process interaction) while still being interoperable. For instance, the tram behavior is modeled based on a set of differential equations and the behavior of traffic lights is discrete and based on event scheduling. - A control library: This is a sub-library of the infrastructure library. Here, the different types of control logic of rail systems elements are implemented. The elements behavior can be modeled in different formalisms, and the use of encapsulation provided by object oriented modeling allows them to be interoperable. It is worth to highlight that only the control logic is embedded in the object structure. Values that would make the object specific for one company or for a certain scenario needs to be specified by the user. This is how intelligence can be incorporated in the design (see Section 5). Control designers can play with different values and choose 4 Available at

4 the one(s) that performs better, such as location of sensors, speed limits, etc. - An XML Parser: users specify the scenario they want to test by inputting data in an XML file. An XML parser is provided to check whether the data entered by the user are correct and in the right format to instantiate the model. Control can also be specified in form of XML definitions, like specifying that a certain tram line will have priority over other tram lines at a specific crossing, setting the speed limits at a certain area of the network, and others. Figure 1 shows a piece of an XML file where a tram vehicle is defined. The parser will check the validity of the entered data, and in case of violating the carefully specified definitions, the parser will give an error and will not build the scenario. Components developed so far were applied at HTM to support control designers to plan the capacity of a new part of the infrastructure, as described in section 2. In this first stage, models were only validated by experts, and not yet with measured data. Because these tracks do not exist yet, the company does not yet have the real data to be used in the validation process. In a second project, we will apply the same models to test the timetable of a complete, existing tram line. To model a whole tram line, we identified additional requirements, which led to the development of new components, and as a result we are currently extending the components library. - An Animator: To support users to have a better insight in the scenario being tested, an animation window is available. Animation can be combined with GIS (Geographical Information Systems) drawings, like a city map to be used as background. These drawings help the user to quicker identify the analyzed section (based on the surrounding buildings and streets) and also helps in verifying whether the scenario built is defined in the correct location. See in Figure 2 an example of the animation window. Figure 2: scenario animation with a map city as background Figure 1: example of a vehicle definition Figure 3 describes the relation among all these modules and how the testing environment works. First, the user specifies a scenario through an XML file. Before loading the model, the parser checks if all necessary data are entered and if they are in the correct format to build the desired scenario. When the model has been created, it is ready to be simulated using the services provided by DSOL. During the simulation run, graphics and statistics are gathered and are displayed to the user to support him/her in the process of deciding which strategy will be more appropriate for a certain system configuration. 5 Intelligent Control Design Using the words mentioned in the definition of intelligence given by [3] (see introduction), one can see intelligence as the flexibility given to users (control designers) to experiment with the attributes specification of control elements and find the appropriate value(s) which will increase the probability of success. Seeing from the control designer s perspective, success, in this case, refers to the achievement of a more robust, reliable and efficient rail system. Taking a speed sign as an example, users can define a number of attributes, for instance the location at which a speed sign will be placed, the speed limit vehicles can reach at a certain area and the text that will appear for the driver (according to conventions used in the company), which might be a symbolic display different from the actual speed limit in km/h.

5 components can support intelligent control design of rail infrastructures. Figure 3: linking all the modules A simple example experiment placing a speed sign before a station with the limit of 50km/h, showed that this limit is too high for a tram to enter a station. Looking at the animation of this scenario, users could see that, based on the physical characteristics of the trams (acceleration and deceleration rate), vehicles driving at a speed of 50km/h were not able to brake in time to stop at the correct place to collect passengers. Different values were tested and 25km/h showed to be an appropriate one. In addition, the limit of 50km/h does not satisfy the safety requirements of the company, because stations are areas where there is a high concentration of passengers and drivers must slow down. 25km/h was confirmed to be an acceptable limit. When testing the infrastructure capacity, control designers can play with the control policies applied to the area being analyzed. For example, the crossing showed in Figure 2 corresponds to the place where new tracks (forming a curve) join the existing network (straight tracks). Designers can experiment with different numbers of vehicles and analyze the results through animation and graphs generated during the simulation run. By the highlighted circles in the graphs shown in Figure 4, users can see that the tram named RR-LV.1, which is a light rail tram, has to stop for some time before the crossing until the tracks are released. As light rail vehicles are expected to serve a bigger demand of passengers, this delay might not be desirable. The decision of whether giving priority or not for light rail vehicles at crossings can be tested and designers can analyze how much delay this will represent to the traffic of existing tram lines. Variances on the number of vehicles in combination with types of priorities can be easily tested and the statistics generated during the simulation run can be used to support this decision. The proposed simulation library can support many other types of control decisions as control designers can immediately see the consequences of each control policy applied. By this case, we concluded that our simulation Figure 4: speed x progression (left) and speed x time (right) graphs 6 Conclusions In this paper, we argue that a simulation based environment can support the development of new control systems of a rail infrastructure by providing a testing environment for rail control strategies. Using a component based solution, we can solve the problems encountered with existing simulation based tools and environments. Developing generic (but specific to the rail infrastructure domain) and loosely coupled components can increase the interoperability among components and between components and external systems, creating a more flexible environment for testing different system configurations. By describing a case, we showed how a simulation library of components can incorporate intelligence in the control design. A next step will be to enable these components to run in distributed locations (different computers distributed geographically) in order to be used in the multi actor environment of rail infrastructure design. 7 Future Research Applying this simulation environment to support the design of rail system control will add value to the infrastructure management, as we believe that the system performance can be improved by designing a more effective control. But this is not enough. In reality, many disturbances occur and they might influence the overall system behavior and as a result the developed strategies need to be adjusted to the current system conditions. For that reason, our simulation environment will be linked to the real infrastructure in the future to support rail control during operation as well. First, it will have only monitoring purposes. We will study how to enable the link to the physical infrastructure in order to make comparisons between the planned and observed behavior. If a difference larger than the accepted value is obtained, considering the

6 relevant performance indicators for the infrastructure, a warning will be given to the controller to take a decision. Later, we intend to use the simulation environment to support controllers to take a decision when a non-desired situation has been detected. For this part of the future research, we will build on the theories of Just-In-Time simulation cloning [17] and distributed simulation [18]. Acknowledgements The authors would like to acknowledge HTM Personenvervoer NV for the support on this research. This research project has been funded by the BSIK-NGI program. References [1] Weijnen, M. P. C., Bouwmans, I. (2004), Innovating Infrastructures: Dealing with Complexities in Networked Systems, Keynote lecture at the IEEE International Conference on Systems, Man and Cybernetics, October, 2004, The Hague, The Netherlands. [2] Verwater-Lukso, Z., Herder, P. M. (2005), Intelligent Infrastructures the first step towards Next Generation Infrastructure Systems, in Proceedings of 2005 IEEE International Conference on Networking, Sensing and Control, Tucson, Arizona. [3] Meystel, A. M., Albus, J. S. (2002), Intelligent Systems: Architecture, Design and Control, Willey Series on Intelligent Systems, John Wiley & Sons, Inc., New York. [4] Hicks, D. A. (1998), Simulation Market Forces Can t be Ignored, In IIE Solutions, May 1998, [5] Ramackers, G. J., Verrijn-Stuart, A. A. (1991), First and Second Order Dynamics in Information Systems International Working Conference on Dynamic Modelling of Information Systems, H. G. Sol and K.M. van Hee (eds), Noordwijkerhout, The Netherlands [6] Verbraeck, A. (2004). Component-based Distributed Simulations. The Way Forward? In Proceeding of 18 th Workshop on Parallel and Distributed Simulation (PADS 04), eds. S. Kawada, , ACM Press, New York, USA. [7] Herder, P. M., Thissen, W. A., (2003), Critical Infrastructures: A New and Challenging Research Field, in the book Critical Infrastructures: State of Art in Research and Application, eds. Thissen, W. A. and Herder, P. M., Kluwer Academic Publishers, The Netherlands. [8] Shannon, R. E., (1975), Systems Simulation: the art and the science, Prentice-Hall, New Jersey. [9] Ebben, M. (2001), Logistic Control in Automated Transportation Networks, doctoral dissertation, Twente University Press, Enschede, The Netherlands. [10] Meer, R. van der. (2000), Operational control of internal transport, doctoral dissertation, TRAIL, Delft, The Netherlands [11] Jansen, M.F.W.H.A. (2001), Designing electronic intermediaries: an agent-based approach for designing interorganizational coordination mechanisms, doctoral dissertation, Delft University of Technology, Delft, The Netherlands. [12] Manivannan, M. S. (1998), Simulation of logistics and transportation systems, Handbook of Simulation: principles, methodology, advances, applications and practice, J. Banks(ed.),Wiley & Sons. [13] Harrell, C. R.; Hicks, D. A. (1998), Simulation Software Component Architecture for simulation-based enterprise applications, in Proceedings of the 1998 Winter Simulation Conference, D.J. Medeiros, E.F.Watson, J.S. Carson and M. S. Manivannan (eds), [14] Saanen, Y. A., (2004), An approach for designing robotized marine container terminals, doctoral dissertation, Systems Engineering Group, Delft University of Technology, Delft. [15] Verbraeck, A., Dahanayake, A. N. W. (eds.) (2001), Building blocks for Effective Telematics Application Development and Evaluation, Delft University of Technology, [16] Jacobs, P. H. M., Langs, N. A., Verbraeck, A. (2002); D-SOL: A Distributed Java Based Discrete Event Simulation Architecture, in Proceedings of the 2002 Winter Simulation Conference, E. Yucesan, C.-H. Chen, J. L. Snowdon, and J. M. Charnes (eds.), [17] Hybinette, M. (2004), Just in Time Cloning, In Proceeding of 18 th Workshop on Parallel and Distributed Simulation (PADS 04), eds. S. Kawada, 45-51, ACM Press, New York, USA. [18] Fujimoto, R. (1999); Parallel and Distributed Simulation ; in Proceedings of the 1999 Winter Simulation Conference; Ed. P. A. Farrington, H. B. Nembhard, D. T. Sturrock, and G. W. Evans,