Research And Implementation Of A Service Composition Method Based On Graph Neural Network

With the development of cloud service technology and computer technology,more and more Web services are published on the network and each Web service provides a simple function.However,the functional requirements of users are becoming more and more complex,so it is necessary to use the service composition method to combine multiple Web services and deliver them to users as a whole.At present,there are a large number of Web services in the network.Among them,web services with the same function usually have different quality of service(QoS).Service composition system needs to make the formed composite service as close to the highest QoS as possible when selecting services for composition,which is an NP-hard problem.Today,there have been a lot of related researches in the field of service computing,but when faced with large-scale service composition problems,the existing methods have low efficiency.Therefore,when performing service composition,how to find a composite service with with near the highest service quality more quickly is still an urgent problem to be solved.Targeting the above problems,this paper proposes a hybrid service composition optimization algorithm based on graph neural network and genetic algorithm.The algorithm first models the service composition problem as a link prediction problem in a heterogeneous graph,and then builds a multi-branch heterogeneous graph neural network model based on graph neural network and attention mechanism,and further trains the model through supervised learning.When a new service composition problem comes,the algorithm first predict the probability of each service being selected by the trained model,then generate an initial population based on the predicted probability,and use the genetic algorithm for further optimization,so as to efficiently obtain a composite service with higher service quality.The research content of this paper is divided into the following aspects:(1)Define a QoS-based service composition optimization problem.This paper first introduces five common QoS attributes of single Web services in detail,and then defines the workflow structure of composite service.Finally,the calculation method of each QoS attribute of composite service is defined,and the service composition optimization problem is defined from the mathematical point of view.(2)Design and implement a hybrid service composition optimization algorithm based on graph neural network and genetic algorithm,called GNN＿GA algorithm.Firstly,the service composition problem is constructed as a heterogeneous graph link prediction problem,and the heterogeneous graph neural network layer based on graph neural network and attention mechanism is used to extract the deep semantic information of nodes in the graph,and then the multi-branch prediction network is used to adaptively use the node features of different size receptive fields.The model is trained using supervised learning so that the model can predict the probability of each service being selected in a service composition problem.Then,based on this predicted probability,the genetic algorithm is used to optimize the composite service efficiently.(3)Three kinds of experiments are designed to verify the effectiveness,efficiency and stability of the GNN＿GA algorithm.This paper selects multi-agent ant colony algorithm(MAACS),multi-group genetic algorithm(P＿MPGA),hybrid algorithm combining elite evolution strategy and Harris Hawk optimization(EESHHO),hybrid algorithm combining branch-and-bound skyline calculation and Q-learning(BBSFWS),service composition algorithm based on dueling Deep Q-Network with priority replay(PD＿DQN),and service composition algorithm based on pre-training and deep reinforcement learning(PPDRL)as contrasting algorithms.The QoS of the composite service and the algorithm running time are used as two indicators to measure the performance of these algorithms.This paper conducts experiments and analyses based on two datasets and three service composition problems with different scales.The experimental results show that,compared with the other six comparison algorithms,the GNN＿GA algorithm achieves an average improvement of12.5% in the QoS of the obtained composite service,and an average improvement of 58.3%in the algorithm running time.It is proved that the GNN＿GA algorithm is effective,efficient and stable.