Research And Application Of Service Composition Method For Cloud Manufacturing

In the cloud manufacturing system,cloud manufacturing service composition is an indispensable technology for ensuring the smooth execution of tasks.It can combine resource services with different functions according to the task process structure to complete complex tasks and meet user’s needs,thereby building composite services with comprehensive functions.According to different environments,cloud manufacturing service composition can be divided into static cloud manufacturing service composition and dynamic cloud manufacturing service composition.A lot of the existing static cloud manufacturing service composition methods used intelligent optimization algorithms(such as heuristic algorithms),which faced huge challenges in large-scale environments,such as low efficiency and low solution quality.The hybrid optimization algorithm that is able to enhance the global search ability of the algorithm and can solve the challenges mentioned above.However,the intelligent optimization algorithm has a fixed pattern and requires manual tuning of parameters,and it cannot adapt when the environment changes.Deep reinforcement learning can react and adjust immediately in dynamic environments,allowing the system to readjust to a stable state.In conclusion,we propose a static cloud manufacturing service composition algorithm based on a hybrid optimization algorithm and then present a dynamic cloud manufacturing service composition algorithm based on deep reinforcement learning.The main work of this thesis is as follows:(1)This thesis analyzes the research background,research status,relevant theories and shortcomings of cloud manufacturing service composition and introduces the shortcoming of existing algorithms.(2)This thesis proposes a static cloud manufacturing service composition method based on a hybrid optimization algorithm.First,the sub-candidate services set are composed of high-quality services selected by the candidate service set based on the skyline query.Then,the teaching stage of teaching-learning-based optimization is combined with the crisscross optimization algorithm to enhance the algorithm’s global search ability.Next,in the learning stage of the algorithm,the formula of the learning stage is changed to improve the convergence speed and stability of the algorithm.Finally,the service composition solutions are compared and sorted,and the optimal solution is output.The effectiveness of the algorithm is verified in experiments.(3)This thesis presents a deep reinforcement learning framework for dynamic cloud manufacturing service composition.First,the DQN is used as the basic framework and a dueling structure is used to build the network structure.Meanwhile,the prioritized replay mechanism is introduced to store the sample data.Then,the double estimator is used to calculate the objective function.Finally,three strategies are added in the framework to ensure that the algorithm returns to a stable state and obtains the best solution.The effectiveness of the framework is verified in experiments.(4)This thesis designs and implements the prototype system of manufacturing resource network collaborative sharing cloud platform.Based on the algorithms proposed in Chapter 3 and Chapter 4,a prototype system of manufacturing resource network collaborative sharing cloud platform is designed.This system provides users with functions such as task/service demand release,idle resource release,and service release.The system selects the optimal service composition solution for service demand users according to the service demand information and resource description information released by users.