Research On Autonomic Computing Oriented Automated Service Negotiation

In autonomic computing environment, autonomic elements are self-management, so no super elements schedule and command their actions uniformly. Negotiation is a necessary method which solves the problem of interest Conflict and task allocation when autonomic elements cooperate. This paper mainly applied artificial intelligent and Distributed Artificial Intelligent approaches to research automated service negotiation. The theory, architecture and decision model of automated negotiation are discussed in two aspects: external environment and internal reasoning of negotiation. The main creative outcomes are as follows:1) An architecture of multi- Participants automated service negotiation is presented. The architecture meet the following three requirements:①Flexibility: each agent can decide whether it will participate in negotiation, and which one to negotiate with. And it is able to negotiate with multiple opponents.②Distribution: Every process of negotiation instance doesn't require the involvement of a third party. The establishment and termination of single instance will not damage the whole negotiation environment.③Dynamic: Any Participant can join in or get out at any time, so negotiation instance is dynamic creation and end.2) To deal with negotiation decision model's conflict bewteen "autonomous" and "trustworthy", a two-layer decision model was presented. The upper layer is a macro instruction layer and the lower layer is micro-action layer. User can provide coarse-grained guidance to negotiation agent by writing macro instructions. So user can control the negotiation process and trust the agent. Micro-action layer decide on the specific action by trading off the micro-instruction and information of environment. This embodies the autonomy of agent.3) A micro-action layer model based on opponent's attitude was presented. In this case, the procedure of negotiation was viewed as a proposal sequence which can be transformed to multiple negotiation tracks—one negotiation track for one negotiation issue—by mapping them to a new feature space. Then the opponent's attitudes on these issues can be got by learning from the negotiating tracks. This model can make trade-offs efficiently between opponent's attitudes and self's preferences. Additionally, if the model's tradeoff functions meet certain constraints, the convergence and monotone of model can be guaranteed. Experimental results show that the model can both decrease negotiation time and increase the joint utility.4) A negotiation decision model based on learning opponent's utility function is proposed. The model learns opponent's approximate utility function from negotiation history, and then makes effective trade-off between opponent and own utility. This model avoids the classical model's assumption that some priori knowledge should be available. Our experiment shows that the proposed model is effective and efficient in the environments where information is private and prior knowledge is not available.5) Mining temporal association rules in negotiation can provide a strong support for designing negotiation strategy. The analysis based on a great deal of real negotiation procedures show that temporal association in negotiation procedure bears the characteristic of approximation. Such approximation indicates that attributes values and associated time interval of rules may fluctuate within certain bounds. To mine approximate temporal association rules in negotiation, an approach based on clustering was advanced. The approximate temporal association rules with valid bounds of attribute value and associated time interval were found by clustering all instances of rule patterns. The experiments show that the approximate temporal association rule has more flexibility and adaptability than the general association rule, and the descriptions it provdies for the rule are more compatible with the real negotiation process.