基于EEMD去噪和果蝇支持向量机的变形预测方法研究

发布时间：2019-01-27 22:39

【摘要】：变形监测是采集变形体的变形信息的技术方法，对变形信息进行处理分析是变形监测的最终目的。变形数据的预处理可以有效地去除数据中的误差，有利于下一步的变形分析和预报结果精度的提高。由于变形体的变形具有非线性、模糊性和不确定性等特点，变形预测的传统精确数学模型结果与实际情况相差较大。支持向量机是由Vapnik等人于上世纪90年代基于统计学习理论提出的新的机器学习方法。它能够寻求有限样本数据的最优解，并且比经验风险原理的神经网络学习算法具有更强的理论依据和更好的泛化性能。支持向量机预测模型的参数决定了样本训练误差和对预测样本的推广性，然而目前尚没有完备的理论和方法解决这一问题，只能通过实例仿真以及算法优化。果蝇优化算法是根据果蝇寻觅食物的特性，嗅觉记忆与视觉记忆协同作用下表现出来的群智能。它对于参数的优化选择有着很好的效果，能够做到全局寻优。本文首先利用集成经验模态分解方法分离出变形数据中的高频噪声信号，并针对高频噪声也含有有用信号的问题，，对其进行阈值量化处理，保留噪声中所含有用信号，完成变形数据预处理工作。然后针对支持向量机参数的选择这一开放性问题，也是实际应用支持向量机预测模型成功的关键问题，利用果蝇算法进行优化选择，结合工程实例，证明果蝇优化算法简化了支持向量机参数选择，避免了实际工程支持向量机预测应用中超参数选择的盲目性。
[Abstract]:Deformation monitoring is a technical method for collecting deformation information of deformable bodies. Processing and analysis of deformation information is the ultimate purpose of deformation monitoring. The preprocessing of deformation data can effectively remove the errors in the data, and it is helpful to improve the accuracy of deformation analysis and prediction results in the next step. Because the deformation of deformable body is nonlinear, fuzzy and uncertain, the results of traditional accurate mathematical model for deformation prediction are quite different from the actual situation. Support vector machine (SVM) is a new machine learning method proposed by Vapnik et al in 1990s based on statistical learning theory. It can find the optimal solution of the finite sample data and has stronger theoretical basis and better generalization performance than the neural network learning algorithm based on empirical risk principle. The parameters of support vector machine (SVM) prediction model determine the sample training error and the generalization of prediction samples. However, there is no complete theory and method to solve this problem, which can only be solved by example simulation and algorithm optimization. Drosophila optimization algorithm is based on the characteristics of searching for food, olfactory memory and visual memory under the synergistic action of swarm intelligence. It has a good effect on the optimization of parameters, and can achieve global optimization. In this paper, the high frequency noise signals in the deformation data are separated by the method of integrated empirical mode decomposition. To solve the problem that the high frequency noise also contains useful signals, the threshold quantization is carried out to retain the signals contained in the noise. Finish the preprocessing of deformation data. Then, aiming at the open problem of parameter selection of support vector machine, which is also the key problem in practical application of support vector machine prediction model, we use Drosophila algorithm to optimize selection, and combine with engineering example. It is proved that the optimization algorithm of Drosophila simplifies the parameter selection of support vector machine and avoids the blindness of super-parameter selection in the application of practical engineering support vector machine prediction.
【学位授予单位】：辽宁工程技术大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP18;P22

【参考文献】