Environmental Decision Support: a Multi-Agent Approach 3 Benedita Malheiroy, Eug´enio Oliveira z yISEP, DEC, Rua de S. Tom´e, P-4200 Porto, Email:
[email protected] WWW URL: http://www.up.pt/˜bene zFEUP, DEEC, Rua dos Bragas, P-4099 Porto CODEX, Email:
[email protected] WWW URL: http://garfield.fe.up.pt:8001/˜eol/eco.html Portugal Abstract The environmental management domain is vast and encompasses many identifiable activities: impact assessment, planning, project evaluation, etc. In particular, this paper focusses on the modelling of the project evaluation activity. The environmental decision support system under development aims to provide assistance to project developers in the selection of adequate locations, guaranteeing the compliance with the applicable regulations and the existing development plans as well as satisfying the specified project requirements. The inherent multidisciplinarity features of this activity lead to the adoption of the Multi-Agent paradigm, and, in particular, to the modelling of the involved agencies as a community of cooperative autonomous agents, where each agency contributes with its share of problem solving to the final system’s recommendation. To achieve this behaviour the many conclusions of the individual agencies have to be justifiably accommodated: not only they may differ, but can be interdependent, complementary, irreconcilable, or simply, independent. We propose different solutions (involving both local and global consistency) to support the adequate merge of the distinct perspectives that inevitably arise during this type of decision making.
Introduction Decision making in the field of environmental management is highly complex and involves a great number of contradictory interests (social, economic, ecological, etc.). The development of adequate tools that act as decision support systems contribute to the making of sensible, justifiable, and legally correct decisions. Environmental management is a vast domain with many identifiable activities: environmental impact assessment, project evaluation, planning, etc.. In particular, this work focusses in the project evaluation activity. Every new project has to be submitted to the set of public (and private) agencies with evaluation competences in the project domain. The evaluation agencies task is to produce a final recommendation, i.e, to accept or reject the project based on the set of applicable regulations and development plans.
3 c ACM 1997 - Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.
This activity presents a number of well defined features: (i) is distributed over the group of agencies; (ii) each agency performs autonomously his share of problem solving; (iii) the existing dependency between expertise domains establishes cooperative links among the members of the group of agencies; and (iv) lacks a methodology to accommodate, whenever possible, the divergent recommendations that may arise. In face of this scenario, we chose to model the project evaluation team activity through an autonomous cooperative multiagent system with the ability to sensibly accommodate divergent opinions. The acommodation of perspectives, supported by distributed belief revision techniques, is based on the establishment of inter-agent argumentations. Finally, the integration of such a multi-agent system with Geographical Information Systems (GIS) will provide the resulting multiagent system with an adequate support for the geographical data characterising the candidate area. The resulting heterogeneous multi-agent system, composed of evaluation agents (EVA), data provider or GIS agents (GISA) and a user interface agent (UIA), acts as a decision support system for project developers, providing guidance in the selection of appropriate locations. We briefly present the developed distributed belief revision techniques, the system and agent architectures and the conclusions.
Distributed Belief Revision In a multi-agent system two perspectives regarding the consistency of the available information coexist: (i) the global system perspective – when the union of each individual agent set of propositions is consistent the system is classified as globally consistent; and (ii) the local or partial view of the system distributed components – when each agent set of propositions is internally consistent the system is referred as locally consistent. An agent’s group of propositions is divided into two sets: (i) the set of private propositions - propositions only used by this agent; and (ii) the set of shared propositions - propositions that are shared with some acquaintance. The belief revision of the private propositions is automatically performed by the local agent. The belief revision of the shared propositions is accomplished by the shared propositions’ owner agent. An
agent, upon revising the belief status of a shared proposition, immediately communicates its updated belief status to every recipient with whom it is shared. The decision of how and when to include external beliefs in an agent’s knowledge base is fundamental to the characterisation of a distributed belief revision system (see [Malheiro, 1996] for a detailed description). Our agents act in ”good faith” and exchange messages using the direct message passing mechanism, thus guaranteeing, that the information received by a recipient agent is, not only, relevant for its activity, but also, truthful from the sender’s perspective. Two policies for the local inclusion of communicated beliefs were adopted: local consistency of the shared propositions - the local beliefs prevail over the communicated beliefs, i.e., the adoption of an external belief is conditioned by the existence or absence of the belief in the agent’s local knowledge base; global consistency of the shared propositions - every communicated belief is unconditionally added to the local knowledge base of the recipient agent. Upon accepting a set of external beliefs, an agent may find itself with conflicting belief status for the same proposition. We adopted two distinct synthesis criteria to guarantee the assignment of an unique belief status to every shared proposition: the disjunctive (OR) synthesis criterion - a shared proposition is believed as long as there is some agent where it is believed; the conjunctive (AND) synthesis criterion - a shared proposition is believed, if and only if, it is believed by every agent that share the proposition. These two synthesis criteria reflect different levels of demand: in the case of the OR synthesis, the belief in a shared proposition by one of the involved agents is enough to make it believed by the system, while, in the case of the AND synthesis, only the consensus among the involved agents makes a shared proposition believed by the system. It is the user privilege to select at launch time, from the available set of agents the sub-set to be run, the synthesis criterion to be applied, and the level of consistency desired.
Architecture The adopted multi-agent architecture is based on the architectural model proposed by the Esprit ARCHON project (Wittig, 1992). The agents have a double layer architecture: the Cooperation Layer (CL) and the Intelligent System (IS) layer. While the latter contains the agent’s domain knowledge based system, the former, holds the functionalities needed for the establishment of the inter-agent cooperative actions. A need for a third layer, the Convergence Layer (CvL) occurs whenever pre-existing domain knowledge systems need to be included in the community.
The CL contains a model of the agent - the Self Model, as well as a model of its acquaintances - the Acquaintances Model. Based on these models, the CL determines when and what type of cooperative action to start, and guarantees that the data sent is relevant to the activity of the recipients through the use of a direct message passing mechanism. The CvL is responsible for translating the requests presented by the community into internally recognisable commands by the pre-existing domain knowledge systems. Depending on the type of agents different architectural options were adopted:
Evaluation Agents – Composed of CL + IS. The Intelligent System Layer is a Belief Revision System composed of two modules: the problem solver and an Assumption Based Truth Maintenance System (ATMS) (de Kleer,1986); Data Provider Agents – Composed of CL + CvL + IS. The Intelligent System Layer corresponds to the associated domain knowledge database, the Geographical Information System (GIS). The translation of the requests presented to the GIS into GIS commands is performed by the Convergence Layer (CvL); User Interface Agent – Composed of CL + IS. The role of the Intelligent System Layer is played by the user himself, by submitting new projects and simulating different requirements.
Conclusion The Environmental Decision Support System under development is a tool envisaged to provide developers with guidance in finding locations that comply with the applicable regulations. As a result, we hope to reduce drastically the time usually spent during the submission and evaluation phase, avoiding an endless period of alterations and subsequent project resubmissions. Although the system prototype is still under development, the results we have been collecting, so far, allow us to be optimistic: the choice of the DAI paradigm to model the environmental project evaluation activity along with the development of methodologies supporting conciliation of perspectives are proving to be consistent with our goal.
References J. de Kleer, ”An Assumption-based TMS”, Artificial Intelligence, 28 (2), 1986. B. Malheiro e E. Oliveira, ”Consistency and Context Management in a Multi-Agent Belief Revision Testbed”, Agent Theories, Architectures, and Languages, Eds. M. Wooldridge, J. P. Muller and M. Tambe, Springer-Verlag, Lecture Notes in Artificial Intelligence, Vol: 1037, 1996. T. Wittig (ed.), ”ARCHON: An Architecture for Cooperative Multi-Agent Systems”, Ellis Horwood, 1992.
