基于协同进化神经网络集成的控制图模式识别技术研究

发布时间：2019-06-18 13:51

【摘要】：随着生产技术的进步,消费者的需求日益提高。这种需求不仅意味着需求量的增加,更有需求品质的提升。质量管理是现代化工业生产提高市场竞争优势的一个重要方法。在现代化工业生产过程中,稳定的工艺流程是影响产品质量的一个重要因素。统计过程控制中的质量控制图,常被用来监控产品质量的稳定性。然而传统的控制图已不再适应现代化大生产的需求。借助于先进的计算机信息处理技术,把人工智能技术应用工业过程控制中去,实现工业过程中质量控制的实时性、准确性是当前国内外专家学者研究的方向之一。本文总结了在现代化工业生产过程中,关于质量管理领域的控制图模式识别的国内外研究现状和发展趋势,介绍了统计过程控制的基本概念和质量控制图的基本原理,对本文涉及的控制图的判定原则、神经网络的基本理论及其泛化和集成理论、协同进化等相关理论进行了阐述和分析,给本文研究的开展提供了理论支撑。通过分析目前质量控制图模式识别方法中存在的不足和缺陷,结合人工神经网络在处理复杂分类问题方面的特点,利用协同进化的思想提出了一种神经网络集成的设计和训练的方法。通过对神经网络集成泛化误差的分析,将神经网络学习算法和协同进化算法相结合,用个体网络的相关度度量网络集成的误差从而实现个体网络的差异性,个体神经网络的结构在学习过程中自动确定,保持了个体网络的准确性,通过构造方法自动确定神经网络集成的结构,提高了集成学习系统的稳定性和泛化能力。最后利用蒙特卡罗质量特征数据模拟方法生成与实际生产过程相似的质量特征序列,运用MATLAB2012a对控制图6种基本模式识别网络进行编程训练,仿真结果表明所训练的神经网络集成CNNE模型具有很强的识别能力其性能明显优于BP网络和RBF网络等单个的神经网络分类方法,也优于Bagging和Adaboost等传统的集成方法。
[Abstract]:With the progress of production technology, the demand of consumers is increasing day by day. This demand not only means the increase of demand, but also the improvement of demand quality. Quality management is an important method to improve the competitive advantage of modern industrial production. In the process of modern industrial production, stable technological process is an important factor affecting product quality. The quality control chart in statistical process control is often used to monitor the stability of product quality. However, the traditional control chart is no longer suitable for the needs of modern mass production. With the help of advanced computer information processing technology, artificial intelligence technology is applied to industrial process control to realize the real-time quality control in industrial process. Accuracy is one of the current research directions of experts and scholars at home and abroad. This paper summarizes the research status and development trend of control chart pattern recognition in the field of quality management at home and abroad in the process of modern industrial production, introduces the basic concept of statistical process control and the basic principle of quality control chart, and expounds and analyzes the decision principle of control chart, the basic theory of neural network and its generalization and integration theory, co-evolution and so on. It provides theoretical support for the development of this paper. By analyzing the shortcomings and defects of the current quality control chart pattern recognition methods, combined with the characteristics of artificial neural network in dealing with complex classification problems, a neural network integration design and training method is proposed by using the idea of co-evolution. Through the analysis of the generalization error of neural network integration, the neural network learning algorithm and co-evolution algorithm are combined, and the correlation degree of individual network is used to measure the error of network integration so as to realize the difference of individual network. The structure of individual neural network is determined automatically in the learning process, which maintains the accuracy of individual network, and the structure of neural network integration is determined automatically by construction method. The stability and generalization ability of the integrated learning system are improved. Finally, Monte Carlo quality feature data simulation method is used to generate quality feature sequences similar to the actual production process, and MATLAB2012a is used to program and train six basic pattern recognition networks in control chart. The simulation results show that the trained neural network integrated CNNE model has strong recognition ability, and its performance is obviously better than that of BP network and RBF network, as well as the traditional integration methods such as Bagging and Adaboost.
【学位授予单位】：中北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183

【参考文献】