基于改进的WM算法和SVM在混凝土组分反预测中的应用

发布时间：2018-01-22 18:46

本文关键词： 模糊规则提取 WM算法支持向量机最大相对误差最小　出处：《华侨大学》2017年硕士论文　论文类型：学位论文

【摘要】：混凝土强度、流和坍落度是建筑工程和土木工程中混凝土质量控制的研究重点。由于不同地区温度、湿度等外界因素干扰,以及混凝土中各组分间复杂的物理、化学反应,使得混凝土组分反预测更加复杂。因此,解决混凝土组分反预测有着重要意义。模糊规则可以通过模拟人的思考方式,将专家知识转化成模糊规则的形式,可以得到不错的效果,然而模糊规则提取的好坏决定了模糊系统的预测能力。目前常用的模糊规则提取算法是Wang-Mendel算法(WM算法),该算法是由Wang和Mendel提出的,可以较好解决实际工程应用中存在的非线性、高维和时变性问题。但是WM算法在性能上,如预测精度、运行效率、鲁棒性和完备性还有改进空间。另外,支持向量机(Support Vector Machine,SVM)对于非线性、小样本等问题有着不错的效果,但是还有改进的空间。因此,为了解决混凝土组分反预测问题,本文展开了以下研究:(1)基于聚类算法改进WM算法的预测精度、运行效率、鲁棒性和完备性。本文引入快速搜索和密度峰发现聚类算法(Clustering by Fast Search and Find of Density Peaks,FSFDP)对数据进行预处理,去除噪声数据,从而改进算法的鲁棒性和预测精度。通过使用样本之间的关系信息,可以对模糊规则库中缺失的规则进行预测,从而保证算法的完备性。另外,在数据规模较大、数据属性和模糊区间个数较多的时候,使用FSFDP算法中的聚类中心点提取模糊规则,可以大大减少模糊规则数量,从而提高算法的效率。并使用实验来验证算法的性能。(2)基于粒子群优化算法(Particle Swarm Optimization,PSO)的最大相对误差最小SVM算法。使用相对误差最小改进SVM的最大间隔约束条件,使得改进的算法更符合实际工程中的应用;使用PSO最小化最大相对误差对改进的SVM进行参数优化,得到性能更佳的模型。并使用混凝土坍落度试验数据集验证算法的可行性。(3)基于相对误差支持向量机改进的WM算法。由于WM算法的模糊规则后件使用的是集合形式,为了增强模糊规则的拟合能力,本文使用改进的支持向量机作为模糊规则的后件,从而提高算法的性能。最后,使用混凝土强度数据集对算法进行验证。
[Abstract]:The strength, flow and slump of concrete are the key points of concrete quality control in construction and civil engineering. Because of the interference of temperature, humidity and other external factors in different areas, and the complex physics of each component in concrete. Chemical reaction makes the back prediction of concrete component more complicated. Therefore, it is very important to solve the inverse prediction of concrete component. Fuzzy rules can be used to simulate human thinking. The expert knowledge can be transformed into the form of fuzzy rules, and good results can be obtained. However, the quality of fuzzy rule extraction determines the prediction ability of fuzzy system. At present, the commonly used fuzzy rule extraction algorithm is Wang-Mendel algorithm. The proposed algorithm is proposed by Wang and Mendel, which can solve the nonlinear, high dimensional and time-varying problems in practical engineering applications, but the WM algorithm has good performance, such as prediction accuracy. There is also room for improvement in efficiency, robustness and completeness. In addition, support vector machine support Vector machine is nonlinear. Small samples and other problems have good results, but there is room for improvement. Therefore, in order to solve the problem of back prediction of concrete composition. In this paper, the following research is carried out: (1) improve the prediction accuracy and running efficiency of WM algorithm based on clustering algorithm. Robustness and completeness. This paper introduces a fast search and density peak discovery clustering algorithm (. Clustering by Fast Search and Find of Density Peaks. FSFDP is used to preprocess the data to remove the noise data, so as to improve the robustness and prediction accuracy of the algorithm. The missing rules in the fuzzy rule base can be predicted to ensure the completeness of the algorithm. In addition, when the data scale is large, the number of data attributes and fuzzy intervals is large. The number of fuzzy rules can be greatly reduced by extracting fuzzy rules by using clustering center points in FSFDP algorithm. In order to improve the efficiency of the algorithm, and use experiments to verify the performance of the algorithm. 2) Particle Swarm Optimization based on particle swarm optimization algorithm. The maximum relative error minimum (SVM) algorithm is used to improve the maximum interval constraint condition of SVM, which makes the improved algorithm more suitable for practical engineering applications. PSO is used to minimize the maximum relative error to optimize the parameters of the improved SVM. Get better performance models. Use concrete slump test data set to verify the feasibility of the algorithm. The improved WM algorithm based on relative error support vector machine. In order to enhance the fitting ability of fuzzy rules, the improved support vector machine (SVM) is used to improve the performance of the algorithm. Finally, the concrete strength data set is used to verify the algorithm.
【学位授予单位】：华侨大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;TU528

【参考文献】