基于改进神经网络的板形识别方法

发布时间：2018-04-23 19:25

本文选题：板形识别 + 混沌免疫遗传算法　；参考：《济南大学》2017年硕士论文

【摘要】：随着经济迅速发展和科学技术的不断进步,社会的各行业对板带钢的需求也日益增长,因此,对板带钢的质量要求也越来越高。板形作为检测板带钢质量的重要指标,如何科学地解决板形的问题也早已成为国内外专家研究的重要课题。本文以人工智能理论中神经网络作为研究基础,选择冷轧板带钢的板形缺陷识别作为研究课题。本文通过对冷轧带钢板形缺陷模式的分析,提出了一种基于Elman神经网络的板形识别方法。首先,本文详细分析了Elman神经网络的优缺点,对于该神经网络训练速度慢、存在早熟现象的缺陷,提出利用遗传算法对网络进行优化改进。本文进而分析了遗传算法的性能特点,发现遗传算法求解多峰非线性问题时,容易陷入局部极值和出现早熟收敛现象。为此,本文将人工免疫思想、自适应机制和混沌优化思想引入算法的进化过程,对种群个体进行免疫选择、自适应交叉、混沌变异,从而增强算法的全局搜索能力,提高算法搜索精度,实现对遗传算法的改进,进而提出了混沌免疫遗传优化算法。然后,本文采用此算法对Elman神经网络的初始权值和阈值参数进行优化,建立了一种改进Elman网络的板形缺陷识别模型,并利用若干组测试样本数据对该模型方法进行仿真实验。通过比较该模型和BP、Elman以及GA-Elman网络的仿真识别效果,证明了用混沌免疫遗传算法优化的Elman网络板形缺陷识别方法,解决了网络收敛速度慢、易早熟的问题,相比单纯采用遗传算法优化的网络模型具有更高的识别精度,识别速度更快,因此,该方法也对板形进行实时控制具有重大意义。
[Abstract]:With the rapid development of economy and the continuous progress of science and technology, the demand for strip steel in various sectors of society is also increasing day by day, therefore, the quality requirement of strip steel is becoming higher and higher. As an important index to measure the quality of plate and strip, how to solve the problem of shape scientifically has already become an important subject of experts at home and abroad. In this paper, the neural network in artificial intelligence theory is used as the research foundation, and the shape defect identification of cold rolled strip is selected as the research topic. In this paper, a shape recognition method based on Elman neural network is proposed by analyzing the shape defect pattern of cold rolled strip. Firstly, this paper analyzes the advantages and disadvantages of Elman neural network in detail. In view of its slow training speed and premature phenomenon, the genetic algorithm is used to optimize and improve the neural network. In this paper, the performance characteristics of genetic algorithm are analyzed. It is found that genetic algorithm is prone to fall into local extremum and premature convergence when solving multi-peak nonlinear problems. In this paper, the idea of artificial immune, adaptive mechanism and chaos optimization are introduced into the evolutionary process of the algorithm, and the immune selection, adaptive crossover and chaos mutation of the individual population are carried out, so as to enhance the global search ability of the algorithm. The search accuracy of the algorithm is improved and the genetic algorithm is improved. A chaotic immune genetic optimization algorithm is proposed. Then, this algorithm is used to optimize the initial weights and threshold parameters of Elman neural network, and a shape defect recognition model of improved Elman neural network is established, and some test samples are used to simulate the model. By comparing the simulation results of the model with that of BP Elman and GA-Elman neural networks, it is proved that the Elman network shape defect recognition method optimized by chaos immune genetic algorithm can solve the problem of slow convergence and premature convergence of the network. Compared with the network model optimized by genetic algorithm, this method has higher recognition accuracy and faster recognition speed. Therefore, this method also has great significance for real-time shape control.
【学位授予单位】：济南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG335;TP183

【参考文献】