Small scale multi-agent systems can be developed just by defining agents and specifying interactions among them. This approach starts to fail if the number of agents increases, exponentially increasing the complexity of interactions, because every pair of agents may interact each other. To solve this problem, the concepts of architecture and organization are introduced into the multi-agent systems. Additionally, these concepts are used for agents to create teams and cooperate in problem solving. Two best known multi-agent system architectures are holonic multi-agent systems (Fischer et al, 2003) and multi-multi-agent systems (Nimis & Stockheim, 2004). The idea of holonic multi-agentsystems is that autonomy of agents is reduced and agents are merged into holons, which appear to outside as a single entity (Fischer et al, 2003). The term “holon” (Greek word “holos” has meaning “whole” and suffix ”-on” denotes “part”) is adopted from biological system research done by A. Koestler (1967). Holon is a self-organizing structure which consists of substructures and is a part of larger superstructure. In terms of multi-agent systems holon or holonic agent is an agent that consists of other agents named subholons.
In holonic multi-agent systems agents form a hierarchical structure, i.e., each holon can be a part of a higher level holon and consist of lower level holons. It allows adapting the system to the structure of the domain. Hierarchy makes holons suitable for task and result sharing. If the holon has a task assigned to it, the task can be decomposed into some subtasks that are assigned to subholons, which can decompose them into the next level subtasks and so on. If the agent receives a task that it is not able to accomplish it can also find other agents to create a holon with, to accomplish the task together (Fischer et al, 2003).
The autonomy of agents that form a holon is usually reduced by giving one agent (called head or head agent) the privilege to do resource and task allocation in the holon. It can have partial or total control over other agents. Agents that are parts of the holon, but are not head agents are called body of the holon (Gerber et al, 1999). To outside the holon appears as a single entity represented by the head of the holon. The body agents do not communicate outside the holon. So, holons have an interface (head) and they can be developed separately as modules of traditional software engineering. Holons also make change implementation easier, because changes of an agent in one holon directly affect only agents from the same holon. Lavendelis & Grundspenkis (2008) have concluded that ITSs comply well with the main criteria of holonic domains defined by the authors of the holonic approach (Gerber et al, 1999), namely operator abstraction, hierarchical structure, partial decomposability and cooperative system. Thus the holonic multi-agent systems are suitable for ITSs.
The second well-known multi-agent architecture - multi-multi-agent system has been developed inside the Agent.Enterprise methodology (Nimis & Stockheim, 2004). The main goal of the project is integration of several multi-agent systems. It is similar to the holonic multi-agent systems in the sense that both architectures propose to create systems that consist of subsystems - holons and multi-agent systems, respectively. Multi-multi-agent systems offer the concept of the gateway agent that accomplishes routing and message conversion between different message formats used in different multi-agent systems. Interactions among agents of various multi-agent systems are enabled. Still, comparing holonic multi-agent systems and multi-multi-agent systems one may conclude that in the context of ITSs there are significant advantages of holonic multi-agent systems. The main of them is that heads of holons unlike the gateway agents can accomplish not only mediator tasks, but also many other tasks. It allows implementing great part of the intelligent mechanisms into the heads of holons. One more important advantage is that holons can be dynamically changed and allow to build open systems. For other advantages of holonic multi-agent systems see (Lavendelis & Grundspenkis, 2011), where it is concluded that the holonic multi agent systems are more suitable for the ITSs and should be used to develop a specific agent based ITS architecture. The next subsection describes such architecture.
- 1378 reads