基於Java的實時多代理系統之環境智能服務

Abstract

In this thesis, we proposed an extension of a popular Java-based multi agent system framework enabling agent to express their task and execution fulfilling real time requirements while also able to respond to external events without violating any real time constraint. Such a framework allows developers to code real time agent with simple programming API. The proposed framework utilizes several technologies such as RTSJ to execute agent’s real time behaviour and algorithms from real time literature including: (1) schedulability analysis based upon rate monotonic analysis, (2) task server to supervise the handling of event. With their aid, the agent can decide whether all its execution tasks will be able to meet their deadline while still serving aperiodic events yielding good average response time. Preliminary tests demonstrated the simplified workflow and feasibility of such framework with minimal time latency while handling event.

在這篇論文中，我們提出了一個基於Java的多代理人的智能體系統架構的延伸，適合用於環境智能服務。該架構足以讓它們在所執行的任務表達對實時的需求，同時也能執行非週期性事件並不影響其它對實時性有需求的任務 。該架構具有簡單的編程API，讓開發人員能夠有效地編寫實時程式。本文利用了RTSJ來實時執行以Java編寫的代理人之實時行為，其他技術包括：（1）基於速率單調分析的調度性分析，（2）任務服務器監督非週期性事件 。在上述方法的幫助下，代理人可以決定所有該執行的任務是否能夠滿足它們的最後期限，同時還能對服務非週期性事件得到很好的平均響應時間。在我們的測試中顯示了應用這種架構的簡化工作流程和可行性，並且以最小的延遲時間來處理事件。