【摘要】：多主體仿真已經(jīng)被廣泛使用于行為金融學(xué),一些基于ABM建模的仿真系統(tǒng)已經(jīng)被提出。但是傳統(tǒng)股票市場仿真系統(tǒng)運(yùn)行在單機(jī)系統(tǒng),仿真中主體(Agent)的數(shù)量受限于單機(jī)的計(jì)算能力,而金融研究人員為了獲得更有效的數(shù)據(jù)需要的Agent數(shù)量越來越多,為了獲得更智能的Agent需要仿真的輪次(round)越來越多。本文專注于大規(guī)模Agent、通訊系統(tǒng)和仿真調(diào)度提出了一個(gè)適用于股票市場的分布式的多主體仿真系統(tǒng)PSSPAM,包括通訊系統(tǒng)、Agent模塊、市場中心模塊和用戶界面四個(gè)松耦合模塊。Agent基于可以自定義的網(wǎng)絡(luò)拓?fù)浣Y(jié)構(gòu),通過通訊系統(tǒng)模塊和其他Agent以及市場中心進(jìn)行通信。本文提出了一個(gè)分布式round線程調(diào)度機(jī)制,對(duì)較大數(shù)量的Agent線程進(jìn)行調(diào)度。為了提高分布式環(huán)境中帶有可演化社交網(wǎng)絡(luò)模型的PSSPAM仿真系統(tǒng),節(jié)點(diǎn)間的負(fù)載均衡和節(jié)點(diǎn)間通訊總量都需要被考慮。本文提出一種處理調(diào)度方法LBMIC,在考慮節(jié)點(diǎn)間負(fù)載的不均衡性低于給定閾值并且節(jié)點(diǎn)間通信量盡量少的情況下將Agent劃分并部署到不同的就按節(jié)點(diǎn)。LBMIC將上述問題轉(zhuǎn)化為圖劃分問題并使用多層圖劃分算法來實(shí)現(xiàn)Agent的調(diào)度。當(dāng)社交網(wǎng)絡(luò)發(fā)生演化時(shí),LBMIC通過遷移Agent來動(dòng)態(tài)調(diào)整初始劃分。本文進(jìn)行了一些實(shí)驗(yàn)來驗(yàn)證PSSPAM仿真平臺(tái)的有效性可擴(kuò)展性。同時(shí),實(shí)驗(yàn)還表明無論是LBMIC的初始劃分還是網(wǎng)絡(luò)演化時(shí)的動(dòng)態(tài)調(diào)整,都能顯著提高通訊密集型仿真的效率。最后,本文還提出一種面向分布式多主體股票市場仿真的規(guī)則驅(qū)動(dòng)編程模型,以規(guī)則組合定義股票市場模型中的市場、主體和主體間交互網(wǎng)絡(luò),設(shè)計(jì)并實(shí)現(xiàn)了相應(yīng)的插件式體系結(jié)構(gòu)支持規(guī)則定制。
[Abstract]:Multi-agent simulation has been widely used in behavioral finance, and some simulation systems based on ABM modeling have been proposed. However, the traditional stock market simulation system runs in a single computer system, in which the number of principal (Agent) is limited by the computing ability of a single machine. However, more and more Agent are needed by financial researchers in order to obtain more effective data. In order to obtain a more intelligent Agent, more and more simulation rounds of (round) are needed. This paper focuses on large-scale Agent, communication system and simulation scheduling. A distributed multi-agent simulation system, PSSPAM, which is suitable for stock market, including communication system, Agent module, is proposed in this paper. Based on the customizable network topology, agent communicates with other Agent and market center modules through the communication system module, which is based on four loose coupling modules: the market center module and the user interface module, which is based on the customizable network topology. In this paper, a distributed round thread scheduling mechanism is proposed to schedule a large number of Agent threads. In order to improve the PSSPAM simulation system with evolutive social network model in distributed environment, the load balancing between nodes and the total amount of communication between nodes need to be considered. In this paper, a processing and scheduling method, LBMIC, is proposed. When considering that the load imbalance between nodes is lower than a given threshold and the traffic between nodes is minimal, the Agent is divided and deployed to different nodes. LBMIC converts the above problem into a graph partition problem and uses multi-layer graph. Partition algorithm to realize the scheduling of Agent. As social networks evolve, LBMIC dynamically adjusts the initial partition by migrating Agent. In this paper, some experiments are carried out to verify the validity and extensibility of the PSSPAM simulation platform. At the same time, the experiment also shows that both the initial partition of LBMIC and the dynamic adjustment of network evolution can significantly improve the efficiency of communication-intensive simulation. Finally, this paper proposes a rule-driven programming model for distributed multi-agent stock market simulation, and defines the market, the interaction network between agents and agents in the stock market model by rule combination. Design and implement the corresponding plug-in architecture support rule customization.
【學(xué)位授予單位】：天津大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2016
【分類號(hào)】：F830.91;TP391.9

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 高寶俊;宣慧玉;李璐;;一個(gè)基于Agent的股票市場仿真模型的Swarm實(shí)現(xiàn)[J];系統(tǒng)仿真學(xué)報(bào);2006年04期

，

本文編號(hào)：2433417