passengers steering trains

Post on 15-Jan-2016

50 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

DESCRIPTION

Passengers steering trains. A Multi-Actor Approach for Operations and Control in the Netherlands Railways. Niels R. Faber * , René J. Jorna * Erwin Abbink † , Ramon Lentink † & Fred van Blommestein * * University of Groningen † Netherlands Railways. Outline. Introduction - PowerPoint PPT Presentation

TRANSCRIPT

Passengers steering trainsA Multi-Actor Approach for Operations and Control in the Netherlands Railways

Niels R. Faber*, René J. Jorna*

Erwin Abbink†, Ramon Lentink† & Fred van

Blommestein*

* University of Groningen† Netherlands Railways

Outline

› Introduction› Research questions› Method› Case 1: initial MAS› Case 2: statistically simulated passengers› Discussion

MAS projects at RUG/EB/IMS › Simulatie projectmanagement bij Ministerie OCW (1990):

Gazendam› Computational Transaction Cost Economics (Tomas Klos,

2000) Jorna/Gazendam› Multi-Actor SOAR (Hans van den Broek, 2001):

Gazendam/Jorna› The Social Cognitive Actor (Martin Helmhout, 2006):

Gazendam/Jorna› Simulation of Crowd and Riot Control (Nanda Wijermans):

Jorna/Jager› Transportbesturing door Smart Agents (MAS@NS) (2007)

Helmhout, Gazendam, Jorna & Faber. MAS@NS: Transportbesturing door Smart Agents

3

Introduction

› Netherlands Railways› Train timetable specifies

planned train movements

› However:delays & disruptions occur

Introduction

› Handling delay & disruption• Dispatching task• Specific type of planning & control

› Objective:• Restore train movements according to train

timetable as quickly as possible

Introduction

› Dispatch task:• Solving a logistical puzzle

› Available means:• Train level:

• Speed of train (faster / slower)• Direction of train (reversing movement)• Action of train (activate / cancel train)

• Between trains level:• Holding trains at station for transfer of

passengers

Introduction

› Dispatch task organization:

Control center

Node coordinatio

n

Network control

Train shift control

NSRProRail

Central control

Local control

44

13

13

Introduction

› Dispatch task knowledge:• Railway network• Train timetable in controlled region• Contact information train drivers and ticket

inspectors

Introduction

› Main foci of dispatch task• Restore train timetable• Balance moving material• Balance personnel

Introduction

› But what about passengers travelling by train?• Are they considered in dispatching?• How are demands and wishes of passengers

included in dispatching solutions?• What does it mean to give passengers a

voice in dispatching?

Introduction

› Types of control• Input oriented control

•Dominant: budgets• Process oriented control

•Dominant: production planning, utilization of capacity

• Output oriented control•Dominant: costs and revenues per

product, marketing

Introduction

› Organization of planning

• Organizational units

• Costs and revenues

• Quality

Introduction

› Desire to incorporate passenger demands in dispatching• Currently, implicit consideration in handling

delays / disruptions• No formal role in decision process of

choosing dispatch solution• No coordination structure for incorporation

in dispatch task

Research questions

› What is the intended role of passengers in dispatching? (informing / deciding)

› What organizational structure supports the involvement of passengers in dispatching?

› What knowledge about passengers needs to be included in dispatching?

› What knowledge should passengers provide if they participate in dispatching?

Method

› Developments in thinking about planning and organization• System theory -> Cognition• Top-down -> Bottom-up• Logic -> Computational Mathematics• Fixed task distribution (DAI) -> learning

systems (MAS)

Method

› Human cognition• Fayol as forerunner• Simon, March• DSS movement: The user at the driving

wheel• KADS

(Henk Gazendam, 2007)

Method

› Top-down -> bottom-up• Anthony: Hierarchical model planning• Lindblom: The science of muddling through

•Planning cannot proceed until consensus is reached

•Technically acceptable solutions -> Socially acceptable solutions

• Mintzberg: The Rise and Fall of Strategic Planning•Emergent strategy

(Henk Gazendam, 2007)

Method

› Logic• Closed worldview• Problems with processing of changes and

time• Logic other than first order logic large

computational complexity

(Henk Gazendam, 2007)

Method

› Computational mathematics• Emergence (Holland)• Coherence (Thagard)• Production of complex systems (Wolfram)• Evolution (Gigerenzer, Dennett)

(Henk Gazendam, 2007)

Method

› DAI -> MAS• From fixed task distribution (DAI) to learning systems

(MAS)• DAI: Distributed Artificial Intelligence

• Actors are not autonomous• Actors only able to execute specific task• Top-down, hierarchical coordination model

(contract net protocol)• Communication limited to task execution• Fixed task distribution and fixed specialization of

actors

(Henk Gazendam, 2007)

Method

› Multi-Actor system (identified as promising technique)

› Actor• Autonomous• Communicative ability

• Ontologies• Protocols

• Task execution ability• General problem solving methods

• Recognition of task environment• Search in problem spaces (weak methods)

• Learning ability• Exploration and imitation• Optimization (neural net)• Evolutionary learning (variation and selection)

Method

› Actors can be:• People• Active computer programmes meeting certain

conditions (agents)• “An agent is a computer system, situated in

some environment, that is capable of flexible autonomous action in order to meet its design objectives” (Wooldridge, 2002, p.15)

• Organisational units represented by a human actor or computer agent

Method

› Characteristics of a MAS (Jennings et al., 1998)• Each agent has incomplete information, or

capabilities for solving the problem, thus each agent has a limited viewpoint.

• There is no global system control• Data is decentralized• Agents communicate through messages• Patterns of messages can be specified in

protocols(FIPA: Foundation for Intelligent Physical Agents)

Method

› Agent task execution:• Tasks are executed through behaviours• Types of behaviours from simple to

complex, composite• Various models to shape behaviour, e.g.:

•Utility•BDI

MethodSocial abilities

Individual abilities

Individual knowledge Social knowledge

Distributedsystems

Multi-agentsystems

Control knowledgeSolution plansWorld mapsBehaviorsresources

RoleCommitments, beliefsProtocols, primitives

Distributedproblem-solving

CooperationCommunicationInteraction

PlanningNavigation & obstacle avoidanceTask solvingSecure mechanisms, perception

Cooperative agents

Autonomous agents

(Glaser, 2002)

Method

› For Netherlands Railways: MAS used to explore organizational structuring when incorporating passengers in dispatching task

› Human actors and software agents collaborate to solve problem

Case 1: initial MAS

› Objective:• develop a MAS prototype that enables

passenger involvement in dispatching in situations of delays and disruptions of the train timetable

• prototype is a useable simulation platform for future simulations

Case 1: initial MAS

› Exploration of;• organizing for retrieval of desires and

demands from passengers

• consequences of delay scenarios

› Using real-time passenger agents

Case 1: initial MAS

› Methodology:• Prometheus Design Tool

(http://www.cmis.rmit.edu.au/agents/pdt)• UML

Case 1: initial MAS

› PDT overview

Case 1: initial MAS

Case 1: initial MAS

Case 1: initial MAS› Main agents:

• Planner• communication with dispatcher• handling disruptions and delays

• TravelManager• handles travelling information• message forwarding to/collection from passengers

• CustomerTravelCoach• aids passengers with travel plan selection

• TravelAssistent• communication with passengers

• SecurityAssistent• handling subscription of passengers

Case 1: initial MAS

› MAS platform:

Java Agent DEvelopment framework (JADE)

Case 1: initial MAS

› JADE characteristics• agents have behaviours• agents communicate through messages• protocols fix specific messages passing

patterns• ontologies available to specify valid message

content• communication between agents follows

communication act theory• BDI possible through JADEX extension

Case 1: initial MAS› Scenario:

• Disruption: tracks between Haren and Zwolle• Cause: leaves on tracks

DispatcherPlanner TravelManager

Passenger 1 Passenger n

……

TravelCoach TravelCoach

……

Case 1: initial MAS

Case 1: initial MAS

Case 1: initial MAS

Case 1: initial MAS

Case 2: statistically simulated passengers› Extension of initial MAS› Inclusion of statistical data to:

• generate more realistic payloads of trains (passengers)• shape statistically based passenger agents with

characteristics for:• travelling motive• ticket type• travelling frequency• departure-destination combinations• travel plan

Case 2: statistically simulated passengers› Objectives:• extend MAS with passive passenger agents• integrate statistical data about passenger

movements• prepare alternative disruption / delay

scenarios• coordinate passenger responses• plan for empirical testing

Case 2: statistically simulated passengers› Additional agents:

• StatisticalPassenger• simulates one passenger in a train based on statistical

data• StatisticalManager

• handles statistical data• CommunicationManager

• transforms information about delayed trains into StatisticalPassenger agents

DispatcherPlanner TravelManager

Passenger 1 Passenger n

……

TravelCoach TravelCoach

……

TravelManager

StatisticalPassengerStatisticalPassenger

CommunicationManager StatisticalManager

Statistical data

Statistical data

Case 2: statistically simulated passengers› StatisticalPassenger

• Station of departure• Station of destination• Ticket type, travel motive, travel frequency,• Travel plan (route and schedule)• Responds to messages about delays and disruptions from

TravelManager

Case 2: statistically simulated passengers› Coordination of reactions of passengers• TravelManager agent is enhanced• TravelManager needs to send one clear

message to the dispatcher (through the Planner agent)

• Collect responses from real-time passengers• Collect responses from StatisticalPassenger

agents

Case 2: statistically simulated passengers› Current status:• Statistical passengers are created• Behaviour of passengers needs to be

implemented• Behaviour of TravelManager agent for

collecting and handling responses from real-time and statistical passengers needs to be implemented

Discussion

› Focus on organization of passenger involvement in dispatching

› Communication and collaboration between human actors and software agents

› Cognition currently located in human actors

top related