軟體代理人程式測試方法之研究

Abstract

軟體代理人程式具有學習人類的行為模式、互相的合作以自動地解決繁雜的問題，以及透過行動力可以在網路上任意地點執行工作的優點，因此漸漸被應用在資料收集、資料過濾、資訊管理、電子商務、工作流程管理、網路管理及服務管理等各項領域之上，而現今雖然有軟體代理人測試環境被提出，但尚未有人依軟體代理人特性提出系統化測試方法，常造成軟體代理人測試不夠完整，而易於發生執行錯誤。 本論文根據軟體代理人的四種特性：(1)自動化，(2)學習能力，(3)行動力及(4)合作能力，以及軟體代理人間的兩種關係：(1)主從關係與(2)供需關係，擬定一套可完整測試軟體代理人行為與特性的測試順序之策略，並提出可避免重複測試的滾動式軟體代理人程式的測試方法。本研究並根據這套測試策略與方法製作一軟體代理人程式測試環境，此環境具有下列功能自動剖析軟體代理人程式、建立軟體代理人間的關係，及產生軟體代理人的測試順序的功能。 本研究分析軟體代理人的特性與行為模式來進行分析，找出軟體代理人特性對行為模式的影響，並歸納出軟體代理人間的主從關係與供需關係；藉由軟體代理人間相互影響的關係，可找出多軟體代理人系統的測試順序，而對單一軟體代理人程式的測試，經分析各項特性間交互影響後關係，提出必須先測試自動化特性，繼而測試行動力特性，最後測試合作能力，而學習力則必入自動化特性中測試，進而提出一套滾動式的軟體代理人程式測試方法，本研究的測試方法以IBM Aglets軟體代理人程式語言為範例，並在IBM Aglets軟體代理人執行環境開發測試工具。

Software agent programs have three advantages include learning human behaviors, automatically solving complicated problems by collaboration, and executing tasks anywhere in networks by mobility. Therefore, software agent programs gradually have been applied in domains of data collection, data filtering, information management, e-commerce, task process management, network and service administration, etc. Although there are testing environments for software agents that have been suggested, nobody has proposed a systematic testing method according to the characteristics of software agents. That usually causes an incomplete testing of software agents and makes execution errors occur frequently. This thesis is based on four characteristics of software agents: (1) Autonomy (2) Learning (3) Mobility (4) Collaboration, and two relationships between software agents: (1) master-slave (2) provider-requestor, to design a strategy that could perform a thoroughly testing on software agents behaviors and testing orders of characteristics. We also implemented a testing environment according to these testing strategies and methods. This environment has following abilities: parsing agent programs, building relationships between agents, and generating testing orders of agents. In this thesis, the characteristics of software agents and their behavior modes are analyzed to find the influences of characteristics on behavior modes and to infer the master-slave and provider-requestor relationships among agents. Testing orders could be founded by relationships between agents. After the analysis of reciprocal effects among characteristics, we proposed that in the testing on a single agent, the autonomy should be tested first, the mobility is tested then, and the collaboration is tested finally. The testing of learning is incorporated into the testing of autonomy. A rolling-style testing method of software agents is proposed. Our testing method takes the IBM Aglets software agent language as an example, and develops testing tools under the IBM Aglets execution environment.