基于选择性神经网络集成的Web服务可信性预测技术研究

发布时间：2018-08-20 14:14

【摘要】：随着服务计算相关技术的不断普及,Web服务作为依托于互联网之上的重要软件资源而被广泛应用。在实际应用场景中,Web服务的可信性成为人们在对Web服务进行选择、推荐时所需要考虑的重要目标。为此,对Web服务的可信性进行有效的评估与预测便成为服务应用过程中的重要问题。通常,服务质量(quality of service,QoS)是Web服务可信性的一个直观且重要的体现,因此通过对QoS的综合分析并在此基础上实现对服务可信性的预测就显得尤为必要。本文试图从机器学习的角度来解决Web服务可信性的预测问题,所提出的基于选择性神经网络集成的Web服务可信性预测是集合了BP神经网络、选择性集成学习、粒子群优化算法等技术为一体的解决方案。在该方案中,将已知可信性等级的Web服务的QoS数据信息对BP网络进行训练学习,生成多个候选神经网络,随后采用粒子群优化策略对候选神经网络的集成权重进行优化搜索,按照搜索得到的最优集成权重方案实现神经网络的选择性集成。在技术实现过程中,基于集成权重编码的不同,本文提出了两种集成模式算法,即PSO-SEN算法和QPSO-SEN算法。通过针对公开数据的实验对比分析,验证基于选择性神经网络集成的Web服务可信性预测方法的可行性和有效性以及参数对算法的影响。实验结果表明,与其他的典型方法相比,该技术在预测准确度上具有明显的优势,且该技术对分类器集成方式、种群大小、分类器隐含层节点数敏感度较低,具有良好的鲁棒性。Web服务的可信性预测有助于用户在对众多功能等价的Web服务开展服务选择时提供可信性特征的指导,以便做出科学合理的服务选择决策。此外,在融合神经网络、集成学习和智能搜索等技术构建混合式预测算法的研究也会对类似管理决策问题提供一定的参考。
[Abstract]:With the popularity of service computing technology, Web services have been widely used as an important software resource based on the Internet. In practical application scenarios, the credibility of Web services has become an important goal that people need to consider when choosing and recommending Web services. Quality of service (QoS) is usually an intuitive and important embodiment of Web service trustworthiness. Therefore, it is necessary to predict the service trustworthiness based on the comprehensive analysis of QoS. In order to solve the problem of predicting Web service trustworthiness, the proposed Web service trustworthiness prediction based on selective neural network ensemble is a solution which integrates BP neural network, selective ensemble learning, particle swarm optimization and other technologies. In this scheme, the QoS data information of Web service with known trustworthiness level is used to BP. The network is trained and trained to generate several candidate neural networks, and then the ensemble weights of the candidate neural networks are optimized by particle swarm optimization strategy. The ensemble weights of the neural networks are optimized according to the optimal ensemble weights. Two integration pattern algorithms, PSO-SEN algorithm and QPSO-SEN algorithm, are proposed to verify the feasibility and effectiveness of the selective neural network ensemble-based Web service credibility prediction method and the influence of parameters on the algorithm by comparing the experimental results with other typical methods. The technology has obvious advantages in prediction accuracy, and it has low sensitivity to classifier integration mode, population size, and the number of hidden nodes in the classifier. It has good robustness. In addition, the research on hybrid prediction algorithm based on neural network, ensemble learning and intelligent search will provide some reference for similar management decision-making problems.
【学位授予单位】：江西财经大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;TP393.09

【参考文献】