0227-B4
An agent-based framework for selective management system
A.M Rasol
Abstract
The reason for focusing to agent technology is because of the = ever-growing=20 complexity of forest information systems and the increasing difficulty = to plan=20 in a dynamically changing environment. Forest planning and scheduling = problems,=20 and database technology have for a long time realized to deal with the=20 robustness by developing suitable technology, Decision Tree Learning on = very=20 large forestry data sets. In this paper an implementation model for = agents with=20 layered architecture in a tropical forest Selective Management System = based on=20 database infrastructure is proposed. The model overcomes the = non-parallel in=20 implementing the various agent processing functions together with the=20 limitations imposed on concurrent execution of agent actions. An example = is=20 provided to illustrate how the various agent functions are imposed on = concurrent=20 execution of agent action using Conflict Resolution technique.
1.0 Introduction
One of the motivations for focusing to agent technology is the = increasing=20 complexity of forest information systems and the increasing difficulty = to=20 predict and plan for dynamically changing environments. None of the = forestry=20 applications seems to use existing agent architectures and yield them as = a=20 subject of research. Three-directions of agent architectures have been = developed=20 over the past decades: reactive, deliberative, and coordinating = architectures=20 (Muller and Pischer 1993). Some research efforts have been contributed = to the=20 exploration of how hybridisation of these architectures as layered = architectures=20 (Fergusson 1992; Firby 1992).
In forestry applications, the main issue is robustness in environment = that is=20 prone to dynamically changing environments. On the other hand, is the = issue that=20 database technology has focused on by developing the concept of = information=20 theory (Shannon and Weaver, 1949). If one managed to combine databases = and=20 agents technologies, there should be a much better chance of employing = layered=20 agent architectures to forestry agent-based systems. These applications = are=20 large-scale systems that must ensure a high degree of reliability while=20 maintaining high performance by executing concurrent decision trees. = This paper=20 involves the implementation model for applying layered agent = architecture in=20 forestry database environment. The implementation of the architecture is = realized by all level of components with concurrent execution of agent = actions=20 and orthogonal implementations of agent functions in a decision trees = (DT). In=20 DT, all the reliable rules for parallel execution are defined. = Contingency=20 actions, in case of failure of some of the components of the plans, are = also=20 defined within the DT. A Decision Tree Model (DTM) will be employed to = represent=20 DT to ensure the degree of high concurrency in agent actions are = increased=20 before the contingency action is committed to the database = infrastructure.=20 Another advantage is that the addition of new functions to the agents = does not=20 require detail knowledge of original functions due to the capturing of = plans=20 together with their active state within the same DT.
The remainder of the paper is organized as follows. Section 2 = presents the=20 proposed intelligent agent layered architecture, the role of each layers = in the=20 aspects of agency, and describes the detail architecture components. = Section 3=20 contains a case study in implementing a DTM typically to Selective = Management=20 System based on database infrastructure. Section 4 and section 5, gives = the=20 future works and conclusion respectively.
2.0 Intelligent Agent Architecture
Here we present an intelligent agent architecture in an attempt to = answer=20 some of the shortcomings of existing forest information systems. Our=20 architecture consists of: reactive layer, deliberative layer, social = layer,=20 knowledge base module, problem and goal module, and planning and = scheduling=20 module, forestry database, sensors, coordinator and communication = interface.=20
In the following sections we will look closely at these components = with the=20 assistant of figure 2.1.
Figure 2.1: Agent Layered Architecture
![](3D"http://www.fao.org/DOCREP/ARTICLE/WFC/XII/0227-B4-Rasol-1.gif")
3.0 Implementation Model
The agent's perceptions (sensors) are translated into reading the = database=20 objects, while its actions are translated into performing write = operations to=20 these objects. Reading and writing are done within normal set of rules = having=20 atomic, consistent, and permanent properties (Ensor and Gabbe 1988). The = agent's=20 Knowledge Base (KB) is stored in a database at each agent layer. User = Interfaces=20 of the relevant information available in underlying layers are = materialized. The=20 belief generator of each layer acts as an interpreter and problem solver = for=20 automated reasoning uses its interface to update the agent's beliefs. = These=20 beliefs are stored in the knowledge base database of the corresponding = layer.=20
The PG part of the agent contains the Problem identification (P) and = Goal=20 activation (G) modules. In light of the given goals at each layer, a = goal is=20 activated based on a chosen domain problem stored at this layer. The = higher the=20 layer is, the more complex are its plans and the more difficult it gets = in=20 representing them. Plans are to be modeled as DTM as will be shown later = in=20 Section 4.0. A goal can be activated only by the problem identification = module=20 of the corresponding layer of the agent or another agent in the case of = the=20 social layer.
In the context of the activation mechanisms the problem receives = incompetence=20 messages from the Planning and Scheduling (PS) module of the underlying = layer of=20 the architecture, whereas in the implementation model events that mainly = influence the problem identification module are belief updates. The = particular=20 activation mechanism of the problem identification depends on the = database=20 infrastructure. An active database infrastructure is recommendable, = which=20 triggers the problem module with each new situation. Other kind of = events to be=20 monitored by the problem module is events that are the result of the = activated=20 goal on the same layer: the success/fail of the actions is returned from = the PS=20 module. In the social layer, some events coming from other agents = through the=20 communication interface. The PS module receives the success/fail results = and=20 other coordination messages of the agents incorporated in = jointplans in=20 which several agent collaborate in order to achieve a common goal = (Lavesque et=20 al. 1990). In the PS part of the agent, activated goals are realized by = means of=20 activities. In the next section, these activities are defined using a = decision=20 tree model.
4.0 Decision Tree Model
Overall, the activities in the decision tree model are reflected by a = number=20 of decision nodes. Every decision node member is committed to achieving = some=20 goal while also believing the other decision node members are pursuing = that=20 goal. The decision node members exchange messages to update their mutual = beliefs=20 and synchronize their behaviour, establishing what is called joint = commitment.=20 In general, a decision node member cannot commit unless all its = decisions are=20 terminated. However, if one of the nodes fails, its parent nodes does = not have=20 to abort. It has the choice between: ignoring the unessential condition = (node),=20 reinstating the sub-node, activating another compensating sub-node , or=20 abort.
The sub-node may commit or abort independently and appear atomic to = other=20 sub-node. Therefore, executing compensating node whose role is to = mitigate the=20 permanent effects of the original sub-node must compensate the other = sub-node.=20 By defining these sub-nodes, the goal of integrating the agent actions = together=20 with their active state within a DT.
Given this decision tree model, the procedures are managed as = follows. The=20 database infrastructure schedules the sub-nodes at each agent layer. The = scheduler is responsible for identifying conflicting nodes and = serializing them=20 using locks specified in a Conflict Resolution (Martial 1992). Other=20 responsibilities of the scheduler include possible deadlock detection = and=20 resolution. The generated schedule is then passed to the Execution = level, which=20 performs the agent actions on the forestry knowledge database in an = efficient=20 and reliable way.
At the lowest layer, the sub-nodes are simple DT framework. = Operations=20 involved are database read/write. At higher layers, operations become = more=20 abstract. This allows further exploitation of the semantics of the = operations by=20 commuting their execution on the basis of the predefined Conflict=20 Resolution. The great advantage of this approach is the direct = transformation of=20 conflicts between various agent behaviors to a conflict of decision = nodes on=20 data, which is automatically handled by the database scheduler and = require no=20 direct and explicit analysis on a combinatorial basis.
When coordinating autonomous agents in order to enable their Social = layers to=20 satisfy their collaborative goals. That's why coordination at the social = layer=20 of one agent cannot initiate a sub-node at another agent. Another reason = is that=20 agents can be deployed in a heterogeneous environment, in which each = agent=20 operates on a different database infrastructure. The role of the Problem = Identification and Goal Activation module is to coordinate between them = using a=20 protocol similar to the 2Phase Commit protocol (Ensor and Gabbe = 1988). It=20 uses the Communication Interface module, which is responsible for the = exchange=20 of messages between agents in conversation languages such as KQML (Finin = et al.=20 1994) and FIPA 97 (FIPA 1998).
5.0 Case Study
In this illustrative case study, it is required to build an agent = that=20 determines the forest treatment types and status of forest stand in a = selected=20 forest. The aim of this example is to demonstrate the ease of = development of the=20 agent in various functions. By using DTM, the degree of concurrency is = largely=20 increased; thus increasing the agent flexibility in executing its plans. = The=20 overhead of coordinating these concurrent actions is undertaken by the=20 underlying database infrastructure. Moreover, due to the definition of = the agent=20 plans in the form of coordination, the plan is self-contained and = includes the=20 compensating actions, needed in case of failure of one or more the = sub-nodes.=20 This allows existing conflicts to be implicitly solved, avoiding a later = explicit solution from the user.
The forest stand (S) has a current inventory data. The agent's plan = is to=20 process the inventory data allowing the best solution strategy for the = user and,=20 at the same time, recommending the user's needs. These needs include the = following: (a)interpretation of forest inventory data, (b)needs for = selective=20 forest management system, and (c)collaborative forest planning = operations
In this implementation, the fulfilling needs (a) will at first = initiated and=20 then expanding the system for further needs. From the DT, the = independence in=20 developing the various functions can be easily concluded. Forestry = database=20 consists of four basic objects: the forest stand object (F), the tree = number=20 (N), the tree diameter classes object (D), and tree species (S). A set = of events=20 that occur is the category for D classes >45cm, +30 - 45cm, +15 - = 30cm) and=20 the tree species types (Dipertocarps and Non-Dipterocarps).
Whereas the N and DBH are updated fairly regularly by interval tasks, = the F=20 and S remains locked for a whole period. This non-shared lock is then = released=20 for a certain period in which the new value to be locked is specified, = and the=20 lock is renewed for another period. For an outline of all operations see = Figure=20 5.1. To implement (a) only the agent Reactive Layer is needed. The agent = has a=20 view of the Forest Resource (FR) i.e. forest inventory data having set = of=20 variables such as: tree number, tree species, and tree diameter. Its = problem=20 identification is simple: to analyse the FR using other software. = Launching a=20 simple orthogonal task T1 to update the F inventory data activates the=20 corresponding goal, namely, a forest inventory analysis assignment. = Another=20 task, also launched from the agent Reactive Layer, is the classification = assignments consisting the total number of trees per hectare, total = volume per=20 hectare according to tree diameter classes and tree species types.
Figure 5.1: Domain Problem Definitions
![](3D"http://www.fao.org/DOCREP/ARTICLE/WFC/XII/0227-B4-Rasol-2.gif")
This plan, T2, determines the data set required for the needs of = practicing=20 selective management system of the Forest Stand. Problem (b) involve = more=20 deliberation. The place to implement it is at the agent Reasoning Layer, = where=20 the agent tries to fulfill more complicated goals. Usually, the user = will=20 observe the required management options and must be coordinated with = getting the=20 necessary variables from the existing FR. So, at this layer, the agent = watches=20 the type of expertise to get involved. When a good deal is observed by = the=20 situation recognition module, the goal "selection of management options" = is=20 activated and a DT is executed by initiating the root problem T3 at the=20 deliberative layer to decide the necessary assignments. This root = problem=20 creates three sub-problems at the lower level: T31, T32, and T33. T31 = updates=20 the F (T31 is of the same type as T1). For example, T32 allocates the = list of=20 silviculture treatments by triggering the number of trees according to = their=20 conditions (i) diameter class >45cm, number of trees ≥ 25, = number of=20 Dominance trees ≥25, number of vine free trees ≥25, (ii) = diameter class +30 -=20 45cm, number of trees ≥32, number of Dominance trees ≥32, = number of no-climber=20 trees ≥32, (iii) ) diameter class +15 - 30cm, number of trees = ≥ 100, number of=20 Dominance trees ≥100, number of no-climber trees ≥100. T33 = waits until the=20 release of the lock on the FR to make the final decision. Under the DT = these=20 sub-problems (T31, T32, and T33) can run concurrently with each other or = with=20 other problems. All the designer has to do is to define the appropriate = behavior=20 in case of failure of these problems according to the rules mentioned in = Section=20 4.0 (e.g defining compensation node). Then, it is up to the database=20 infrastructure to learn and execute these DT, resolve possible = conflicts, and=20 handle exceptions resulting from the failure of one the sub-problems. = Figure 5.2=20 shows a semantically concurrent execution of these DT, expressed as a = partial=20 order between them. The graph proceeds from left to right and the edges = mark the=20 order between conflicting pairs of sub-problems.
Figure 5.2 A Semantically Concurrent = Execution
![](3D"http://www.fao.org/DOCREP/ARTICLE/WFC/XII/0227-B4-Rasol-3.gif")
The agent Social Layer is needed for satisfying collaborative needs = (c). In=20 this case, determining forest planning operations based on the forest = stand=20 inventoried area. At this layer, the agent has a view of the Forest = Stand, and=20 the rule set from the cutting regimes lists, silviculture treatments, = and forest=20 stand status observed by the Problem Identification. Also depending on = who is to=20 take the initiative, it observes activating messages from the other = agent(s).=20 When the opportunity comes, DT, similar to T3, are instantiated at the = Reasoning=20 Layer of both agents. Although, in principle, a single parent for both=20 sub-problems is sufficient (see Figure 5.3), this is not allowed due to = the=20 autonomous nature of agents. A model having a parent (root) problem at = the=20 Social Layer of each agent (see Figure 5.4) and having a coordinator=20 coordinating between these two root problems seems to be an appropriate=20 solution.
Figure 5.3: Single node (one root) for = nonautonomous=20 agents
![](3D"http://www.fao.org/DOCREP/ARTICLE/WFC/XII/0227-B4-Rasol-4.gif")
Figure 5.4: Loosely coupled node (two roots) for = autonomous=20 agents
![](3D"http://www.fao.org/DOCREP/ARTICLE/WFC/XII/0227-B4-Rasol-5.gif")
6.0 Future Works
The implementation model is a first step toward a full representation = of=20 agents layered architectures in terms of activities based on DT. The aim = is to=20 make use of the knowledge of decision semantics to increase degree of=20 concurrency. This DTM opens the way for another aspects in their use, = namely the=20 representation of agent plans. By capturing the plans together with = their active=20 state within the same structure, they form a powerful tool for detecting = and=20 resolving conflicts in deliberative layer. However, further refinement = and=20 fine-tuning has to be done to the DTM to better serve its new purpose as = a=20 generic representation tool for the various planning strategies. The = questions=20 on what kind of planning strategies can be represented as DT, or = conversely, how=20 these can be adjusted to meet particular strategies.
7.0 Conclusion
In this paper, an attempt is made to complement agent technology with = forestry database technology to increase system robustness. The decision = trees=20 model provides an independent implementation of agent operation. = Representing=20 agent plans in terms of DTM opens the door for a large degree of = concurrency=20 between agent actions, thus simplifying agent plans and at the same time = without=20 having the developer worrying about coordinating them. Additionally, a = clearly=20 defined DTM within a semantically concurrent execution provides a neat = design=20 and avoids much of the undesired negative interactions between the = various agent=20 actions.
References
Ensor, J.R., and J.D. Gabbe, 1988. Transactional blackboards. = Readings in=20 distributed artificial intelligence. Eds. Bond, A.H and Gasser, L. = Morgan=20 Kaufmann, California.
Fergusson, I.A., 1992. Touring Machines: An Architecture for Dynamic, = Rational, Mobile Agents. Ph.D Thesis, Computer Laboratory, University of = Cambridge, United Kingdom.
Firby, R.J., 1992. Building Symboliuc Primitives with Continous = Control=20 Routines. In.J.Hendler, editor. Proceeding of the 1 st = International=20 Conference Intelligence Planning Systems. Morgan Kaufmann.
Finin, T., R. Fritzon, D. Mckay, and R. McEntire, 1994. Draft = specification=20 of the KQML AgentCommunication Language. The DARPA Knowledge = Sharing=20 Initiative External Interfaces Working Group, Baltimore.
FIPA, 1998. Agent Communication Language. Foundation for Intelligent = Physical=20 Agents, Geneva, 1998. http://www.fipa.org/.
Lavesque, H.J., P.R. Cohen, and J.H.T. Nunes, 1990. On acting = together.=20 Proceedings of the 8 th National Conference on Artificial=20 Intelligence (AAAI-90), 94-99. Boston, MA:AAAI Press.
Martial, F. 1992. Coordinating plans of autonomous agents. Lecture = notes in=20 computer science 610, Springer Berlin.
Muller, J.P., and Pischel, M., 1993. The agent architecture inteRRaP: = Concept=20 and application. Research Report RR- 93-26, Deutsches Forschungszentrum = fur=20 Kunstliche Intelligenz, Kaiserslautern, Germany.
Shannon, C.E., and W. Weaver, 1949. Mathematical Theory of = Communication,=20 Urbana, IL:University of Illinois Press.
1 The author is a Research Officer with the = Forest=20 Research Institute Malaysia and currently doing his PhD as a Research = Student,=20 Centre of Computing Intelligence, De Monfort University, Leicester, = United=20 Kingdom.
Centre of Computing Intelligent
De Monfort University, Leicester, =
LE 1=20
9BH, UK
44-01162 577939
email: rmanaf@dmu.ac.uk