A multi-agent paradigm as structuring principle for planning support systems

https://doi.org/10.1016/j.compenvurbsys.2007.03.001Get rights and content

Abstract

Although urban plans may appear simple once completed, they are the outcomes of a highly complex decision-making process. This complexity makes it difficult to develop Planning Support Systems (PSS) that are comprehensive, flexible, intelligent and, yet, understandable to users. Great improvements are within reach, however, when observing the advances being made with regard to component-based software engineering and development and the role that multi-agent technology can play in this respect. This paper intends to bring this to mind and introduces a conceptual framework for PSS to describe how using a multi-agent paradigm as structuring principle could further both the intelligence and flexibility of such systems in terms of operation, functionality, usability, extensibility and adaptability. The framework is structured as an agent organization consisting of three types of agents: ‘interface agents’ to improve the user–system interaction; ‘tool agents’ to support the use and management of models; and ‘domain agents’ to provide access to specialized knowledge. As the agents are linked to specific system components, data flows in the system take the form of agent communication, while the interaction between agents enables a smooth integration of system functionality. Furthermore, the consistent definition of agents as human-like specialists facilitates users’ understandability.

Introduction

Land use plans are the main instrument in Dutch urban planning, providing both quantitative and qualitative regulations for the types of permissible land uses in a demarcated area. They can be developed in various degrees of detail, ranging from plans that provide a general land use allocation for zones to plans that give additional regulations about the internal design of these zones. Although land use plans may have a simple appearance once they are completed, their development takes place by means of a decision-making process that is full of complexities (Ferrand, 1996, Yeh and Shi, 1999). As their plans affect the quality of life, planners are obliged to practice well-informed decision-making, which requires them to have access to analytic tools of various kinds and knowledge regarding multiple disciplines. Planning support systems (PSS) are primarily focussed on addressing the former aspect, being organized as collections of tools (e.g., Batty, 1995, Geertman and Stillwell, 2004, Yeh, 1999). Although knowledge is often incorporated, it is typically fragmented and concealed in the mechanisms of these tools. In order to effectively support the development of land use plans and urban plans in general, it is essential that systems explicitly provide access to both tools and knowledge and offer assistance in using these resources. Moreover, they have to do so by means of a most understandable organization that is flexible in terms of extensibility (to incorporate knowledge and tools that will result from future developments) and adaptability (to match typical characteristics of cases and/or user groups).

In the area of planning support, multi-agent technology is commonly known as a methodological instrument to develop micro-simulation tools that allow planners to visualize, analyse and forecast collective phenomena emerging from the interaction of individual behaviours of agents that represent consumers who conduct activities, such as shopping (Koch, 2000), recreation (Bishop & Gimblett, 2000) or combinations of daily activities. This understanding of the technology – i.e., agents being submerged into a virtual urban environment to interact with each other and their environment – has resulted in many interesting and promising tools (e.g., Parker, Manson, Janssen, Hoffmann, & Deadman, 2003) to predict and evaluate the effects of different policy scenarios and plan alternatives. Hence, multi-agent technology has found a valuable application in planning support.

However, academics in the area of software engineering are unanimous in placing the technology in a much broader perspective due to the fact that it offers generic tools, techniques, and metaphors to design and implement systems that are open, complex and/or ubiquitous (Jennings & Wooldridge, 1998). Hence, multi-agent technology may also have more to offer to PSS development than only facilitating micro-simulation. First, it can ease the development and management of PSS through modularity at different levels, i.e., from interconnecting system components to composing the knowledge and behaviour of individual agents. Second, it provides the desired opportunities for intelligent planning support through concepts of autonomy and flexibility of agents. Third, its inherent anthropomorphy can greatly contribute to system usability (Cuena & Ossowski, 1999).

The solution that multi-agent technology provides to the ever-increasing demand for more advanced and more intelligent systems is, basically, a fundamental change of the abstraction used to design and develop these systems. According to Jennings (2001), the most powerful abstractions are those that minimize the semantic gap between the units of analysis that are intuitively used to conceptualize a problem and the constructs present in the solution paradigm. In comparison to the logical or mechanical approaches traditionally used, multi-agent technology adopts concepts from sciences such as organizational and social science to assimilate system behaviour to that of a human organization or a society (Zambonelli & Van Dyke Parunak, 2003). The characterization in terms of agents has proven to be a most natural abstraction to many real world problems, having convinced researchers and developers in a wide variety of domains (e.g., Jennings et al., 1998, Nwana, 1996) of the great potential of multi-agent solutions.

Involving a wide variety of actors who are involved at different stages and try to achieve different objectives related to different scales, planning decision-making is complex by nature. In order to effectively support planners, PSS need to be intelligent, flexible and understandable. In this paper, a conceptual multi-agent framework is presented that could meet these requirements. In the next section, we provide an agent-based view on the urban system – i.e., the general object of study in planning – and describe how the process of urban plan development can be considered as a process in which (groups) of agents interact with each other and their environment. This is followed by the introduction of the conceptual framework of a PSS named Multi-Agent System for supporting the Quest for Urban Excellence (

) that intends to exploit the versatile potential of multi-agent technology for supporting the development of land use plans. Then, the various types of agents operating in this system are further discussed in terms of their particular tasks and capabilities. After giving more detailed attention to the operation of agents who are part of the system’s ‘knowledge’ component, a brief illustration addresses a prototype application that was developed to demonstrate how multi-agent concepts can be used to generate alternative plans. The paper closes with conclusions and a discussion.

Section snippets

Urban systems and multi-agents

In an exemplary effort to define a suitable and comprehensive PSS structure, Hopkins (1999) suggested building on elements of both geographic modelling and planning according to an object-oriented approach and, as such, to distinguish actors, activities, flows, investments, facilities, regulations, rights, issues, forces, opportunities and constraints. This apt suggestion acknowledges the fact that the urban environment is the result of the actions and reactions of various parties that attempt

system framework

In the literature (Bishop, 1998, Klosterman, 1999), the ideal system architecture of PSS is believed to consist of three internal components – referred to as ‘information’, ‘models’, and ‘visualization’ – that are made accessible through a user interface (Fig. 1a). This architecture has two major drawbacks, however. First, the successful application of PSS following this architecture highly depends on the users’ ability to oversee the complete functionality of the system and to comprehend how

Agent typology

The conceptual framework of

distinguishes three types of agents (Fig. 1b), each dedicated to serving a specific purpose in system operation. As the agents are linked to specific system components, data flows in the system take the form of agent communication, while the interaction between agent types enables a smooth integration of system functionality. Furthermore, the consistent definition of agents as human-like specialists in a particular field improves users’ understandability.

knowledge component

In comparison with contemporary PSS, the framework of

incorporates an additional ‘knowledge’ component (Fig. 1b) that interconnects the system to the plan development process by means of a group of domain agents (Section 4.3) that assist and advice the user in a most recognizable way as these agents are specified to closely resemble the human actors involved in planning practice. In practical terms, the interconnection between system and process implies that the domain agents are working

Illustration: agents to support alternative plan generation

Given the advocated necessity of adding a ‘knowledge’ component to the

framework, and the concurrent need to demonstrate the suitability and potential of multi-agent concepts for planning support in ways that add up to the current mainstream of agent-based micro-simulation tools, one of the core parts of the framework is the operation of domain agents within the ‘knowledge’ component. Accordingly, a model was specified that describes how domain managers – agents specialized in particular land

Conclusions and discussion

Rationally, the location-sensitive nature of planning could be regarded as the key factor that conflicts with the notion of a generally applicable framework and, thus, only permits the development of one-off applications. To consider this as an accomplished fact, however, is not in line with the advances being made with regard to component-based software engineering and development and the role that multi-agent technology can play in this respect.

This paper has made an attempt to bring this to

References (35)

  • O. Devisch et al.

    Towards a generic multi-agent engine for the simulation of spatial behavioural processes

  • Ferrand, N. (1996). Modelling and supporting multi-actor patial planning using multi-agents systems. In Proceedings of...
  • J. Hendler

    Making sense out of agents

    IEEE Intelligent Systems

    (1999)
  • L.D. Hopkins

    Structure of a planning support system for urban development

    Environment and Planning B: Planning and Design

    (1999)
  • N.R. Jennings

    An agent-based approach for building complex software systems

    Communications of the ACM

    (2001)
  • N.R. Jennings et al.

    A roadmap of agent research and development

    Autonomous Agents and Multi-Agent Systems

    (1998)
  • N.R. Jennings et al.

    Applications of intelligent agents

  • Cited by (26)

    • The Urban CoBuilder – A mobile augmented reality tool for crowd-sourced simulation of emergent urban development patterns: Requirements, prototyping and assessment

      2018, Computers, Environment and Urban Systems
      Citation Excerpt :

      Another modelling approach, agent-based models (ABM), allows for an even better understanding of emergent urban forms, processes or systems through simulation of the “individual actions of diverse agents, measuring the resulting system behavior and outcomes over time” (Crooks et al., 2008, p. 417). ABMs can also be applied to increase the intelligence and flexibility of Planning Support Systems (Saarloos, Arentze, Borgers, & Timmermans, 2008). An agent can be described as an autonomously acting object (Portugali, 2012) existing in, and interacting with, an environment that contains other agents, resources, etc. (Batty, 2009).

    • Socially rational agents in spatial land use planning: A heuristic proposal based negotiation mechanism

      2016, Computers, Environment and Urban Systems
      Citation Excerpt :

      This plan generation phase is part of their prepared Planning Support Systems (PSS). In that context, three types of agents are distinguished: a group of interface agents to monitor user's behavior, a group of tool agents to help users or other agents with selecting and using specific tools and a group of domain agents representing the variety of land use experts involved in planning (Saarloos, Arentze, Borgers, & Timmermans, 2008). As an extension of their work, Arentze, Borgers, Ma, and Timmermans (2010) and Ma, Arentze, Borgers, and Timmermans (2007) addressed the problem of how to incorporate the uncertainty about the future spatial distribution of competitive and synergetic land uses.

    • Introducing a novel model of belief-desire-intention agent for urban Land use planning

      2013, Engineering Applications of Artificial Intelligence
      Citation Excerpt :

      So MAS model is a suitable technique for understanding and simulating them. The works of Valbuena et al. (2008), Saarloos et al. (2008), Tian et al. (2011) are samples of such agent usage in land use planning. Valbuena et al. (2008) described an empirical method that defined an agent typology for an entire region.

    View all citing articles on Scopus
    1

    Present address: Transportation Engineering Laboratory, Graduate School for International Development and Cooperation, Hiroshima University, 1-5-1 Kagamiyama, Higashi-Hiroshima 739-8529, Japan.

    View full text