軟體代理人開發工具之設計

Abstract

軟體代理人是單一獨立程式，可不需外部的操控即可自主執行以達成目標任務，並可藉由代理人間之訊息傳遞合作達成共同的目標 [2]。由於代理人具有自主性，移動性，合作性，及學習性等獨特特性，因此已被專注應用於多執行緒的大型分散式系統，如資料搜集、網路管理、網路交易等等系統 [13][17]。為加速代理人程式之開發，已有許多軟體代理人開發工具問世 [1][8][18]，然而這些工具均著重於代理人程式間類別的發展，缺乏代理人系統架構分析到代理人程式執行驗証的整套工具。因此本研究提出一套漸進式發展的代理人系統開發工具，以支援軟體代理人系統之設計、製作及驗証。

在軟體代理人吺計階段，此工具可支援使用者定義軟體代理人系統的系統架構，包括代理人的目標、工作、代理人間關係、行動性、及學習性；在製作階段，此工具自動依各代理人特性與關係，自動提供樣桯式供設計人選擇填入，以節省程式撰寫時間；在驗証階段，本工具支援使用者定義代理人系統之佈局與各代理人啓動之順序，以驗証代理人系統程式之正確性。

本工具架構於IBM分司 Aglets 軟體代理人平台系統上 [1][19]，利用此工具確可彌補 Aglets 系統在系統架構與系統驗証能力之不足，並可提升代理人程式發展效率與品質。

Software agent is a program that has its own goal and execution thread. Software agent can act as a human-being that has the ability to work against any trouble and cooperate with other agents to achieve its goal. Software agent has four characteristics: autonomous, collaboration, learning human behaviors, and executing tasks anywhere in network by mobility. Therefore, software agent programs have been applied to developing large-scale multi-thread distributed system, such as data collection, network management, and network transaction. To accelerate the development of software agent-based system, there have been some development tools. However, these agent development tools just focused on coding level by providing agent classes to support user to write agent source code. There are no agent development tools to give an overall support from agent system analysis, agent behavior design, to agent system execution and validation. This thesis has design an agent system development tool that supports incremental development process and help user to analysis, design, and validate agent programs.

At agent system design phase, our tool can support users to define agent system architecture, including the goal of agent, the tasks of agent, the relations between agents, mobility of agent, and learning ability of agent. At design and implementation phase, our tool will provide proper template code for user to apply. At validation phase, our tool supports user to define agent system deployment model and execute agent programs in order to validate the execution results.

Our tool, which is based on IBM Aglets System, provides more services including architecture analysis tool and system validation tool. Users are able to increase agent system development performance and quality with our tool.