基于CNN满意度模型的云制造服务组合优化算法

Optimization algorithm for cloud manufacturing service portfolio based on CNN satisfaction model

本文研究云制造服务组合问题,针对传统云制造服务组合采用单一的线性加权模型,不能动态地反映用户满意度的需求,提出一种基于卷积神经网络(CNN)的满意度动态模型,从而构建基于动态模型和改进遗传算法的云制造服务组合的优化算法。构造非对称的多尺度卷积神经网络,使用历史数据训练CNN,实现权值的动态调节,从而构建满足用户不同需求的云制造服务满意度的动态模型。对遗传算法采取了动态的精英策略并且引入排挤机制和快速非支配排序。结果表明,得到的云制造服务组合用户满意度相比线性加权模型提高了17.33%,相比BP神经网络构建的非线性模型提高了9.34%,均方误差(MSE)降至0.00226。在云制造服务平台中验证所提模型构造方法和算法策略的可行性和有效性,在高维度模型中收敛速度仍然很快,克服了以往使用固定权值的线性组合的弊端,使得满意度得到提高,同时具有较高的稳定性。

This paper investigates the problem of cloud manufacturing service portfolio and proposes a dynamic model of satisfaction based on Convolutional Neural Network(CNN)to build an optimization algorithm for cloud manufacturing service portfolio based on dynamic model and improved genetic algorithm.An asymmetric multi-scale Convolutional Neural Network is constructed and the CNN is trained using historical data to achieve dynamic adjustment of weights,so as to construct a dynamic model of cloud manufacturing service satisfaction that meets different needs of users.A dynamic elite strategy is adopted for the genetic algorithm and a crowding mechanism and fast non-dominated ranking are introduced.The results show that the obtained user satisfaction of the cloud manufacturing service portfolio is 17.33%higher than that of the linear weighted model and 9.34%higher than that of the non-linear model constructed by BP neural network,and the Mean Square Error(MSE)is reduced to 0.00226.The feasibility and effectiveness of the proposed model construction method and algorithm strategy are verified in the cloud manufacturing service platform,and the convergence speed is still fast in the high dimensional model.The proposed model construction method and algorithm strategy are validated in a cloud manufacturing service platform,and the convergence speed is still fast in high-dimensional models,which overcomes the drawbacks of previous linear combinations using fixed weights,resulting in improved satisfaction while having high stability.