基于MDP的服务不确定性自适应决策方法

A MDP-Based Adaptive Decision-Making Method:Dealing with Run-Time Service Uncertainties

服务系统运行中因环境变化而面临各种不确定性事件,导致期望价值无法完全实现.有多种应对不确定性的策略,不同的策略导致服务最终成功执行的概率不同,所导致的质量变化(时间延迟、成本溢出等)也不同,需要寻求最优的应对策略,使不确定性造成的损失和应对策略所付出的代价最小.对不确定性事件和相应的应对策略进行了分类,根据流程结构定义服务执行中的各种状态,采用不确定性触发关系图(UTG)刻画状态与决策动作之间的关系,进而采用Markov决策过程(MDP)进行最优策略的选择.通过仿真实验验证了方法的有效性,并给出了影响决策效果的若干因素.

The execution of a service system is faced with a variety of uncertainties which lead to the expected values cannot be fully implemented and delivered. There are several strategies to handle these uncertainties but different ones will result in different consequences, e. g. , the pos- sibility of a final successful execution, the time delay and cost overflow. It is necessary to pursue the optimized one based on the dynamic execution of services, so that the loss of uncertainties and the cost paid to the selected strategy are both minimized. Service uncertainties and the corre- sponding decision strategies are classified, then based on the states of service execution, Uncer- tainty Triggering Graph （UTG） is proposed to describe the triggering relations between uncertain states and decision strategies. Next, a Markov Decision Process （MDP） based algorithm is presented to help select the global optimized strategy. Simulation experiments validate the effectiveness of our model and method, and some key factors that impact the decision effects are briefly discussed.