Abstract: Aiming at the QoS parameter volatility and uncertainty of service computing environment,high dimensional input of deep neural networks and reinforce learning were considered to resolve dynamic service optimization and recommendation in complex cloud environment,so an intelligence collaborative service recommendation model based on three-layer MAS architecture was built.Specifically,state transfer model of the business process was constructed from the perspective of activity-level,and the local QoS and global service collaboration was evaluated,which could solve the time-dependent problem of state transition in large-scale business process modeling.A Business Process as a service Recommendation algorithm based on time-series Evolution Deep Q-Learning (EDQL-BPR) was proposed,and the Q value update strategy based on particle swarm optimization was designed,which improved the optimization efficiency of learning Agent of deep neural network and the recommendation quality of BPaaS service,and achieved a good balance between efficiency and adaptability under dynamic environment.

Key words: service computing, service recommendation, deep learning, crowd-based cooperation

摘要： 针对服务质量参数波动性与服务计算环境的不确定性问题,将深度神经网络的高维输入与强化学习相结合,解决复杂云计算环境下的动态优化和推荐问题,实现了一个基于三层多智能体架构的群智协同服务推荐模型。从更细粒度的活动级构造服务过程的活动状态迁移模型,分析业务活动的局部QoS评价和全局服务协作度量,解决了大规模服务过程建模中状态转移的时间依赖性问题。提出了一种基于深度学习的服务推荐算法(EDQL-BPR),并设计了基于粒子群算法的Q值更新策略,提高深度神经网络的学习智能体的寻优效率,有效提高了BPaaS服务的推荐质量,实现动态环境下效率和适应性的良好平衡。

关键词: 服务计算, 服务推荐, 深度学习, 群智协同