基于自组织演化的多Agent可信联盟研究

不同于现有采用博弈论的方法研究多Agent联盟技术，从联盟信任这一角度开展可信联盟研究，首先在历史合作信息的基础上构建个体信任度，在此基础上建立联盟信任度，进而形成可信联盟.通过联盟间的竞价协商来使得多个小规模的联盟合并成规模更大的联盟从而完成复杂的任务，刻画出联盟演化的自组织性并将信任贯穿于联盟的演化过程.为获得稳定的联盟，在联盟内部收益分配的机制上采用自由竞价与信任评估方式，最后在联盟收益分配结束后，通过信任修正来解决自由竞价中的欺诈行为，并同时获得数据隐私保护，使得联盟内部成员获得公平合理而又高效的分配机制.通过联盟信任，建立起分布式合作模型，可以大规模地降低计算复杂度和可控的风险收益，为动态联盟的稳定性提供有效保障.

Study of Multi-Agent Trust Coalition Based on Self-Organization Evolution

Different from the existing game theory used in multi-agent coalition, this paper studies it from coalition trust. The trust degree of individual is built on history cooperating information and then the trust degree of coalition is built above it for trusting coalition. To finish more complex task, some small coalitions unite to a large coalition by coalition trust and competitive negotiation. This way makes trust run through the total process of evolvement of coalition with describing the self-organization of evolvement. In order to get stable coalition, the free competition and trust evaluation is used for distributing income among coalition. For the effective and fair distributing mechanism, the frame behavior in free competition will be eliminated by amending trust degree with the protection of private data after distributing income. Using trust, the produce structure and evolution process of coalition is described and the stable coalition is obtained by fair income distribution with private protection. The distributed cooperation model is built and the computing complex is reduced greatly by coalition trust with controllable venture income. Trust coalition will provide an effective guarantee for dynamic coalition.