The complex nature of decisive situations often implies the existence of a formalization process which can compromise decision-making criteria. In fact, when managing a transport network, as for managing any process, decision-makers come upon major difficulties in making decisions due to the conflicts between the criteria that they need to take into account and to modeling. As a result, it is necessary for the decision-makers to benefit from a modeling approach which acknowledges these criteria as well as a strategy which can reflect their compromise.
Several decision-making problems in the field of urban transport pose great difficulty. In the majority of cases, it is not possible to fully meet all the objectives by simply outlining a set of decisions. The challenge lies in the fact of modeling and identifying uncertain knowledge, as well as the parameters and the criteria which are critical for decision-making. For a public transport customer, the arriving at a destination often requires changing vehicles several times at different transport hubs. These connections are sometimes made in non-optimal conditions due to unforeseen incidents causing buses to be delayed or ahead of time. These uncertain and incomplete data can lead to considerable difficulties for regulating traffic.
The development of a decision is a task of great importance in an uncertain environment such as transportation (traffic regulation, connection management). This is an area in which theory has not introduced any model, mainly due to the great diversity of practical cases and to the lack of efficient tools for solving the problem.
In the pages that follow, a model is proposed for building a set of decisions which engage a variety of transport actors (the engineering office, engineers, regulators, experts, etc.). On the one hand, this model takes into account a set of uncertain data which concern the fuzzy characterization of the operating resources (e.g. drivers, buses, information), and on the other hand, a set of rules and methods which may be of use for the decision-making process, based on expertise which has been acquired in the field of transport. The model is essentially based on the application of fuzzy logic and on the integration of a corpus of theoretical and practical knowledge developed by the heads of department of heterogeneous areas (statistics, forecast, opinion, etc.). The model begins by acquiring knowledge regarding different human, material and informational resources. This type of knowledge is modeled and processed through fuzzy decision-making, which offers decision sets inspired in the expertise of regulators, as well as other actors involved in the transport network.
In collective urban transport systems, information is often considered as a set of elements, which are not only interrelated among each other, but also interfere with their environment. Knowledge about these systems is sometimes fraught with uncertainty and inaccuracies. These imperfections in knowledge are of different kinds, something which may further disrupt the control and regulation of these systems. Therefore, it becomes important to make it possible to express and model inaccurate information by using the new modeling theories dedicated to uncertain environments.
The final goal of their search is to improve the connections of the urban public transport network: this aims to help regulators master regulation, to ensure that connections are made in the best possible conditions, to avoid the excessive waiting of customers, to improve the quality of service and to increase the appeal of public transport. The site for carrying out this project is Montbéliard, whose operator is CTPM (Compagnie des Transports du Pays de Montbéliard). Such an aim is reached by developing a model which can provide solutions and answers to decision support problems related to the connections and the traffic of the transport network.
This chapter’s contribution is related to methodological aspects, mainly the modeling of uncertain information, as well as developing a decision support model based on the expertise of simulation engineers, regulators or operations managers. As regards information modeling, the chapter will tackle the problems related to the acquisition of information, and will have the opportunity of taking an in-depth look at the analysis of consequences and of developing the suitable criteria to accompany decisions, which are all important phases of the decision-making process. As regards information modeling to ensure reliability and to obtain results untainted by inaccuracies, the theory of fuzzy subsets has been suggested as a rational theory for modeling and manipulating uncertain knowledge, offering a point of equilibrium between its flexibility of use and its illustrative power. This theory enables the construction of simple and effective models for decision support problems. In this view, and due to this aspect, the quality of the service offered to the customer will certainly improve. In that respect, some important concepts in the field of urban transport are defined.
An urban transport network may have a defined architecture strategy depending on the company responsible for its management, as is the case of the network of Montbéliard city, based on the hub and spoke principle, due to geographical reasons [HAY 02]. This structure was developed in the United States by airlines. It is an air network structure that systematically reduces the traffic at the airports acting as pivots. Montbéliard’s network is organized around two large hubs from which all the city bus lines depart. Other companies have adopted a more classical architecture (gridded, circular, etc.).
As mentioned above, the Montbéliard’s transport network is organized around two hubs (Figure 7.1):
All the lines leave from these two hubs, covering the city of Montbéliard, and these are connected by the most important network line that has a very high frequency: “Diam”. There are also other secondary lines of average frequency enabling users to go to the other places in the city.
The structure of this network makes it possible to concentrate traffic towards these hubs, which represent strategic obligatory points for moving within the territory. The quality of the network depends on the quality of these hubs.
Using public means of transport is nowadays a major phenomenon at all levels, either at the daily commute to work, for intra-urban transport or other types. For a transit customer, the arrival at destination often requires changing vehicle several times in the connecting platforms, and it is at this stage of the journey that the quality of the transport network is truly perceptible to the user. This notion of connection becomes crucial when we want to modify or optimize a transport network to best meet customer demand or material constraints. However, research has not made a significant contribution in this area in the face of the evolving demands of the transport customer. These customers, who demand a high quality in terms of comfort and safety, have difficulty in complying with the queuing discipline. This difficulty prompted operators to connect lines together by setting up connection points and ensuring faster travel. Connection platforms are hubs for the exchange of passengers. This can be achieved by pooling a stretch of line or stations.
The quality of service perceived by the public transport user depends, on the one hand, on the true-to-life quality of the service delivered, and on the way users interpret such reality on the other hand. The relationship between the real and the perceived is interesting because it characterizes the relationship between the provider and the users. Considering the customer’s demands and temporal aspects, the main elements affecting the customers during their journey connections are:
The ideal is to provide transit users with reliable information to manage their movements and avoid delays in case of disturbances.
The operation of an urban collective transport network essentially comprises two major parts. After elaborating a production program which is materialized in the preparation of a timetable (TT), the regulation, which is the most difficult part, has to be done: this implies an adaptation of the program to the real conditions which is translated by a real-time schedule controlled by the regulatory strategies adopted by the regulators and the experts of the urban transport network. More precisely, it represents the process of real-time updating of the schedule to the operating conditions. Figure 7.2 shows the characteristics of this regulation.
The task of regulation is complex. It has four parts:
Following these different tasks that the regulator fulfills, we become aware of the complexity of regulation, especially when many incidents occur simultaneously. Hence, the need for a decision support model which is both robust and user-friendly for modeling uncertain incident information, and for analyzing and suggesting control strategies, while respecting service quality requirements and operational constraints [SOU 00, HAY 03]. The decision support model should:
Despite the clearly multidimensional aspect of real decision-making problems, traditional models and operational research techniques have long been based on the recognition of a single criterion supposed to account for the essential relevant aspects of the problem under consideration. When the problem is obviously multidimensional, pursuing this trend implicitly reduces the evaluation of consequences to a single unit. The potential heterogeneity of this set of consequences can then lead the analyst to mix and reduce diverse criteria such as duration and other physical quantities, the complexity degree of procedures, probabilistic risks or dysfunction levels to complex formulas.
One of the contributions in this scientific approach to decision support is to recover this type of inconvenience by applying the model introduced in Figure 7.3. This generally takes into account all the decision criteria simultaneously through the use of fuzzy decision rules and by making fuzzy inferences.
Another essential point of the proposed approach is to overcome the uncertainty and inaccuracy of key criteria in decision support by modeling and formalizing these through the theory of fuzzy subsets.
The approach is translated by process interaction constituting the suggested model. Processes such as knowledge modeling through the concept of fuzzy subsets, their processing, the development of traffic regulatory strategies and their analysis acknowledging the expertise of regulators and operational managers of the transport network will be developed later in this chapter.
In order to meet the operational requirements and constraints and the goals of service quality, the model needs to integrate the following functions:
The first concern during this research process even before the modeling formalism was how to elicit the knowledge of experts, who were in this case transport regulators and traffic operational managers in general. An endeavor demanding so much preciseness, such as knowledge acquisition, may take weeks or even months.
Furthermore, given the complexity of expert knowledge (it sometimes becomes difficult to make their mental processes explicit), acquired knowledge may be inaccurate, incomplete or even inconsistent.
The proposed approach to knowledge acquisition is based on information gathering techniques via the methods mentioned in the second chapter, such as interviews with experts and regulation operational managers. Following the multiple collaboration sessions held with them on several occasions, and the working days spent with them, and after the classification of this knowledge, criteria involved in the development of traffic control strategies were selected, among which are:
Seeking to grant the reliability and robustness of the fuzzy decision support model for knowledge acquisition, a knowledge structuring and filtration process should be carried out in the following way:
In order to structure the complex and heterogeneous mass of information concerning potential actions, it is necessary to identify the various points of view (aspects, characteristics or attributes regarding possible decisions) which may play a role in reasoning and establishing judgments. Thanks to this, it is possible to describe and to model the consequences of the implementation of potential actions, in relation to all the points of view deemed relevant.
Once formalized, the consequences of the actions do not always make it possible to directly translate the preferred judgment. Thus, taking into account one attribute for reasoning will only become effective when the decision-maker and/or all the actors concerned have agreed to hierarchize it according to relevant criteria.
There is a more general alternative for modeling preferences for a particular attribute. For this, it is necessary to build a set of attributes having a value function within the interval [0, 1] reflecting the degree to which a given consequence corresponds to the aspirations of the individuals involved.
Thanks to such techniques, it is possible to build a criterion, that is, a preference model related to a consequence or a set of consequences, which can be grouped along the same axis of synthetic meaning clearly perceived by the different actors. Formally, let us call a criterion any f(x) function defined on A, with a certain value within set E equipped with the relation of total order ≥E, making it possible to compare performance, depending on the following relation:
Rf is a binary relation defined on A. The element E written as f(a) is classically called performance of a on criterion f.
For the purposes of regulation (set of maneuvers), the proposed approach resorts to a broad palette of more or less subjective criteria. The criteria listed below are the result of several collaborative sessions with regulators by applying the aforementioned knowledge acquisition techniques. Important criteria are, for example:
All these criteria have to be taken into account in the decision support system, despite the fact that some may seem empirical. The proposed approach will also take into account the following criteria:
Some system processors require the intervention of human interlocutors, either observers or potential users. These interlocutors introduce subjective descriptions, which imply the presence of the five human senses. In this case, granting reliability and obtaining results which are untainted by inaccuracies is a difficult task, and demands the use of specific knowledge modeling approaches.
The practical accomplishment of a decision support system necessarily implies knowledge acquisition. The simultaneous processing of this knowledge, which may be uncertain, can be achieved in a simpler manner using fuzzy subset theory.
This theory is also applied when:
In other cases, fuzzy subset theory is used because it is easier to implement than traditional methods, and yields comparable results.
Criteria modeling through fuzzy subset theory is done through fuzzification that the interface establishes between the physical world and the data used by the model, playing the role of a digital/linguistic converter. It models and represents information through membership functions, in this case, by converting the available features in the operating system into linguistic terms defined by membership functions.
This process includes different phases:
Figure 7.5 gives an example of a fuzzification process for the advance–delay of the connecting buses. Being ahead of time/delayed is defined as the difference between the actual time of passage and the theoretical time. This is represented by fuzzy subsets and the following linguistic terms: very late, late, ahead of time.
Following the various collaborative sessions with the operational managers of the transport network, as well as with the regulators, and counting upon the expertise elicited during the knowledge acquisition phase, it is possible to define the time frames concerning the state of each of the connecting buses. These frames represent the time intervals or fuzzy subsets used in fuzzy inference through decision rules. When the time frame of one bus moves from one fuzzy subset to another, this results in a new situation and, as a result, fuzzy inference may result in a new type of decision. Thus, the subset “VL” reflects a situation in which the delay of the bus is excessive, and the subset “L” reflects a situation in which the delay of the bus is average, etc.
Another example is the process of fuzzification of the passenger flow, shown in Figure 7.6. The number of travelers is represented by the following fuzzy subsets and linguistic terms: negligible, medium, important. This is the number of travelers who are planning to make a connection.
The number of travelers is also a key parameter for regulation strategies. In the case where there are no customers planning to make a connection, the control action is not essential, but if it is not performed, it may cause problems in another point of the network. In this view, fuzzy subsets represent the value ranges which were key to changes in the regulation strategy. Thus, the large fuzzy subset reflects a situation in which the number of passengers is considerable, resulting in a very particular regulation strategy through the decision rules used in the fuzzy inference.
The process of fuzzification for the time of day is illustrated in Figure 7.7. The time of day reflecting traffic conditions is represented by the following fuzzy subsets and linguistic terms: off-peak hour, rush hour, team change-off time, late evening.
The time of day is a parameter reflecting traffic conditions. This information is as important as the one that was just identified in order to decide on the nature of the regulatory action. During the rush hour the traffic conditions are quite different from those at 3 pm, for example. For this, based on the expertise of regulators, fuzzy subsets were defined so as to model the information which influences the way in which human operators manage and regulate the transport network. At around noon, for example, the managers plan changing drivers, that is, the “team change-off”; this timing is modeled by using the fuzzy subset “CT”: team change-off time.
Figure 7.8 illustrates the fuzzification of proposals on the slowdown of buses. The proposals regarding the status of the bus refer to slowdown values for the regulation of traffic in the transport network when an incident disrupts the network. These are modeled by fuzzification through membership functions.
The generation of decisions for regulating traffic is the most important step in the approach. In fact, it reflects the automation of the regulators’ thinking process, reasoning and action, by drawing on their expertise and taking into account all the criteria which are decisive in regulatory strategies. It is the result of the fuzzy inference process after having synthetized the expertise obtained during the consultation with all the experts, in order to round it up and complete it in a rigorous and efficient way. The rules are expressed as follows:
The slowdown always refers to the most advanced bus, all the rules are grouped in matrices in Table 7.1.
Table 7.1. Decision rule table
Arriving bus | Departing bus | Passenger flow | Timing of the day | Slowdown | ||
IF | VL | VL | Negligible | RH | THEN | Negligible |
VL | VL | Negligible | LE | Negligible | ||
VL | VL | Negligible | OP | Negligible | ||
VL | VL | Average | RH | Negligible | ||
VL | VL | Average | LE | Negligible | ||
VL | VL | Average | OP | Negligible | ||
VL | VL | Important | RH | Negligible | ||
VL | VL | Important | LE | Negligible | ||
VL | VL | Important | OP | Negligible | ||
VL | L | Negligible | RH | Negligible | ||
VL | L | Negligible | LE | Negligible | ||
VL | L | Negligible | OP | Negligible | ||
VL | L | Average | RH | Weak | ||
VL | L | Average | LE | Average | ||
VL | L | Average | OP | Weak | ||
VL | L | Important | RH | Average | ||
VL | L | Important | LE | Average | ||
VL | L | Important | OP | Weak | ||
VL | A | Negligible | RH | Weak | ||
VL | A | Negligible | LE | Weak | ||
VL | A | Negligible | OP | Weak | ||
VL | A | Average | RH | Important | ||
VL | A | Average | LE | Important | ||
VL | A | Average | OP | Average | ||
VL | A | Important | RH | Important | ||
VL | A | Important | LE | Important | ||
VL | A | Important | OP | Important | ||
L | VL | Negligible | RH | Negligible | ||
L | VL | Negligible | LE | Negligible | ||
L | VL | Negligible | OP | Negligible | ||
L | VL | Average | RH | Weak | ||
L | VL | Average | LE | Average | ||
L | VL | Average | OP | Weak | ||
L | VL | Important | RH | Average | ||
L | VL | Important | LE | Average | ||
L | VL | Important | OP | Weak | ||
L | L | Negligible | RH | Negligible | ||
L | L | Negligible | LE | Negligible | ||
L | L | Negligible | OP | Negligible | ||
L | L | Average | RH | Weak | ||
L | L | Average | LE | Weak | ||
L | L | Average | OP | Weak | ||
L | L | Important | RH | Weak | ||
L | L | Important | LE | Weak | ||
L | L | Important | OP | Weak | ||
L | A | Negligible | RH | Weak | ||
L | A | Negligible | LE | Weak | ||
L | A | Negligible | OP | Weak | ||
L | A | Average | RH | Weak | ||
L | A | Average | LE | Average | ||
L | A | Average | OP | Weak | ||
L | A | Important | RH | Average | ||
L | A | Important | LE | Average | ||
L | A | Important | OP | Average | ||
A | VL | Negligible | RH | Weak | ||
A | VL | Negligible | LE | Weak | ||
A | VL | Negligible | OP | Weak | ||
A | VL | Average | RH | Important | ||
A | VL | Average | LE | Important | ||
A | VL | Average | OP | Important | ||
A | VL | Important | RH | Important | ||
A | VL | Important | LE | Important | ||
A | VL | Important | OP | Important | ||
A | L | Negligible | RH | Weak | ||
A | L | Negligible | LE | Weak | ||
A | L | Negligible | OP | Weak | ||
A | L | Average | RH | Average | ||
A | L | Average | LE | Average | ||
A | L | Average | OP | Weak | ||
A | L | Important | RH | Average | ||
A | L | Important | LE | Average | ||
A | L | Important | OP | Weak | ||
A | A | Negligible | RH | Weak | ||
A | A | Negligible | LE | Weak | ||
A | A | Negligible | OP | Weak | ||
A | A | Average | RH | Weak | ||
A | A | Average | LE | Weak | ||
A | A | Average | OP | Weak | ||
A | A | Important | RH | Weak | ||
A | A | Important | LE | Weak | ||
A | A | Important | OP | Weak |
The defuzzification process performs the opposite action to fuzzification, playing the role of a linguistic/digital converter. It turns the information in the form of subsets from the universe of state descriptions, deriving from the inference process, into numerical variables (Figure 7.9).
The defuzzification method used for the different membership functions of the chosen criteria represents the center of gravity method. This is the most accurate method, and is often used in similar cases to that of the regulation of urban transport traffic. However, it requires an integral calculation. The advantage of this method is that it takes into account all the available information and offers a very satisfying result. The formula is given in the following form (Figure 7.10):
In the decision support system for the regulation of connecting urban transport traffic, the discourse universe and the slowing down or waiting time value for connecting buses are the result of the difference between the theoretical time and the practical time.
Four examples are now introduced. For the first one, only one rule is activated, two are activated for the second one, four for the third and 16 for the last one, representing one of the most complex occurrences that may appear.
For example no. 1, we have the following vector:
The rule used in Figure 7.11 is: “if the arriving bus is late, and the departing bus is ahead of time, and the number of passengers is average, and the timing of day is the rush hour, then the slowdown of the departing bus is average”
For example no. 2, we have the following vector:
The rules used in Figure 7.12 are as follows:
This value, determined by defuzzification, represents the slowdown or waiting value of the bus ahead of time compared to the other connecting buses.
For example no. 3, we have the following vector:
The rules used in Figure 7.13 are as follows:
Finally, in example no. 4, we have the following vector:
As examples, here are some rule illustrations in Figure 7.14:
Given the large number of scenarios and of inferences the system may encounter, a classification of the regulation strategies into types of decisions is proposed according to the value found after defuzzification.
This decision type refers to “slowing down of x minutes”, and is possible when the slowdown value is low. In this case, the slowdown should not interfere with the network operation and the regulation takes place in the best conditions.
In this type of decision, “slowing down of x minutes and waiting at the connecting point”, the slowdown becomes more or less important and may hinder network operation. This explains the need for this type of slowdown when there is heavy traffic and waiting at the connecting node.
From the moment the slowdown becomes very important, neither the slowdown nor the wait are possible. In this case, the type of decision will embrace several regulation strategies.
Substitution refers to the action of substituting one of the cars for the intended service. This maneuver involves using an available car (parked, re-entering, delayed) to replace the one originally planned and which became unavailable (breakdown, late arrival, etc.). The machinist chooses a parked car or picks up a car on its way back, or a delayed car, and sets out to the place where the incident was reported. The change-off service will park its car on arrival at the terminus or relay it to the incoming service, or to the delayed service. Figure 7.15 is an example of this type of maneuver being two stations: Temple and Acropole.
Car 01 blocked in station Z will be late for the connection at 12:40. Car 05 is moved in order to ensure this departure and replace Car 01. The machinist of Car 01 will park the car on arrival at Temple.
Transshipment is the transfer of passengers from one car to another. It is planned when substitution is not possible due to a lack of available means. This maneuver makes it possible to relieve a car from its charge. It is essential in the case of car unavailability on a line. Furthermore, it makes other maneuvers such as turning around possible. In order to carry out a transshipment, it is necessary to have at least two grouped cars. The number of grouped cars, the load of each of them, the number of passengers to be removed in the section after the transshipment point are all factors which determine the feasibility of the maneuver. The need for a car near the transshipment points (load in the opposite direction, gaps to be filled) and the constraints of the staff management (team change-off to be granted) determine the relevance of this decision.
For instance, in Figure 7.16, car no. 1 is very late due to a breakdown. The decision is to transship the passengers of car no. 1 to car no. 2.
This maneuver refers to postponing the departure of a bus to another stopping point. It makes it possible to re-establish the regularity of the traffic and the absorption of an irregular charge: the car is then delayed at the meeting point in question, that is, the time required for the late bus to reach the stop and the passengers to climb into the car. For example, in Figure 7.17, bus line 3 has fallen behind and will therefore not be able to reach the Temple connection node with bus line 2. However, the two lines 3 and 2 will meet again at Acropolis and will be able to ensure the connection.
After detecting the incidents of transport connections in related to the operation feedback, the DSS carries out the process of problem-solving and infers which decisions or regulation strategies can be suggested. This proposal is materialized in the maneuvers that each of the connecting bus drivers must perform. These maneuvers seek to coordinate the buses at the connection stage, acknowledging various sources of information, such as incident reports (mechanical breakdown, traffic jams, etc.), or the creation of connections, which are simulated in order to assess their impact on the connection, and therefore make it possible to decide on the validity of the decision.
Although they take into account all the experimental and practical theoretical knowledge of the experts of the transport networks and regulators, decisions or suggested strategies are subject to validation by regulators in the sense that this is a decision support tool. They are evaluated on the basis of significant criteria such as:
These criteria, as others that the operator recommends, enable regulators to decide on the validity of these regulatory strategies and to classify them in view of defining the best solutions for traffic regulation. For this, after the proposal takes place, an immediate and automatic simulation should be implemented in order to reveal the impact of these strategies on the network in real time. In the case where the suggested strategies do not fulfill the best conditions for traffic regulation or do not completely satisfy the requirements set by the managers, after a technical or material incident has taken place, regulators may suggest other strategies that will be integrated into the knowledge base for use in future configurations.
Once validated, the regulatory strategies are stored for future uses and implemented on the transport network. Throughout their application a simulation must be performed in order to measure and study the impact of the regulation on the network in a theoretical way and to be able to make a comparison with practical study cases. These measurements and evaluations become the subject of statistics for the future perspectives of operators, particularly in relation to:
In this chapter, a decision support model based on fuzzy subset theory was proposed. Different stages were presented: from the acquisition and processing of knowledge to the elaboration of decisions. This model resorts to human expertise as a source of knowledge in view of helping decision-makers to develop strategies and decisions, which explains and justifies the foundation of this model on the theory of fuzzy subsets.
First, the model functioning focused on the way in which knowledge acquisition and knowledge processing take place. Then, explanations detailed its modeling through membership functions, and showed how strategies and decisions are developed through fuzzy inference, and how the DSS offers results to decision-makers so as to validate them.
The concept of decision support as envisioned through the model introduced meets the constraints and prospects of the operation, as well as the goals of those responsible for the decision framework. Another essential point of this model is that it helps overcome the uncertainty and inaccuracy stemming from expert descriptions and ensures reliability.
[HAY 02] HAYAT S., MAOUCHE S., DEKOKERE S. et al., Élaboration et mise au point d’un système d’aide à la décision (SAD) pour la gestion du réseau de transport collectif de Montbéliard, PREDIT Final Project Report, INRETS/RT-02-714-FR, 2002.
[HAY 03] HAYAT S., OULD SIDI M.M., “Towards fuzzy aid decision-making system of the Valenciennes transport network connections”, The International Conference on Information Reuse and Integration, Las Vegas, United States, pp. 27–29, October 2003.
[SOU 00] SOULHI A., HAYAT S., HAMMADI S. et al., “Fuzzy decision-making in the traffic regulation of the bus networks”, World Automation Congress WAC’2000, pp. 11–16, United States, June 2000.
Chapter written by Saïd HAYAT and Saïd Moh AHMAED.
3.15.168.214