【摘要】 针对面向高可扩展性、复杂性和异构性服务环境的Web服务组合难以进行优化的问题,提出一种利用长短期记忆单元改进深度Q神经网络(Long Short-Term Memory Deep Q-Network,LSTM-DQN)的Web服务组合优化方法。利用Markov对Web服务组合优化问题进行建模,分析各变量参数之间的关系,同时加入强化学习的组合优化模型,简化了组合优化过程;引入记忆单元对深度Q网络算法进行优化,提出LSTM-DQN方法,提升了DQN算法的全局寻优能力;将LSTM-DQN应用于大规模服务环境下的Web服务组合优化问题,对所建立的马尔可夫决策(Markov Decision Process,MDP)模型进行优化,以提升Web服务组合的处理效率。实验结果表明,该方法相对于传统方法在大规模服务环境下对Web服务组合优化所消耗时间更短,服务组合成功率更高,具有更强的处理能力和处理效率。更多还原

【Abstract】 In order to solve the problem that it is difficult to optimize Web service composition for high scalability, complexity and heterogeneous service environment, an improved deep Q neural network(LSTM-DQN) based on long-term and short-term memory units is proposed to optimize the composition of Web services. Markov was used to model the Web service composition optimization problem, and the relationship between the parameters of each variable was analyzed. The combinatorial optimization model of reinforcement learning was added to simplify the combinatorial optimization process. The memory unit was introduced to optimize the deep Q network algorithm, and the LSTM-DQN method was proposed to improve the global optimization ability of DQN algorithm. The LSTM-DQN was applied to the optimization of Web service composition in large-scale service environment, and the established Markov decision process(MDP) model was optimized to improve the processing efficiency of Web service composition. The experimental results show that the proposed method consumes shorter time, has higher success rate of service composition, and has stronger processing ability and efficiency than the traditional method in large-scale service environment.更多还原