基于PLS和SVR的水质预测模型研究

发布时间：2018-05-31 02:04

本文选题：水质预测 + 支持向量回归机　；参考：《昆明理工大学》2017年硕士论文

【摘要】：水环境是全球自然生态环境中不可或缺的一部分。目前,我国愈发严峻的污染态势,使水环境的保护和治理更加被重视。水质预测是水环境研究的重要内容,是现代环境科学理论研究的重要课题之一。针对洱海流域的弥苴河水质的特点,考虑到水环境系统的复杂性,提出了基于 PLS(Partial Least Squares)和 SVR(Support Vector Regression)弥苴河水质预测模型,本文主要研究成果如下:针对主成分分析法在对水质预测模型输入变量提取主成分时,对水质预测模型输出变量的解释能力比较差而导致水质预测模型预测精度下降的问题。本文采用偏最小二乘回归法对水质预测模型输入变量进行成分提取,使得提取出的成分在最大限度的解释输入变量的同时对输出变量的解释能力也达到最大。这样,在减少水质预测模型输入变量的同时也提高了预测精度。针对BP神经网络(Back Propagation Nerual Network)对小样本数据泛化能力差,模型输出不稳定的问题。本文采用支持向量回归机对弥苴河水质数据进行非线性回归,较好的解决了小样本数据的训练学习以及预测精度的提高。针对遗传算法的寻优效率不高,容易陷入局部极小值的问题,本文提出了改进型遗传算法,动态调整交叉和变异的概率。在提高寻优能力的同时对支持向量回归机的初始参数进行优化。本文在大理州洱海水质监测站弥苴河水质数据基础上,将本文提出的模型应用于弥苴河水质预测,并和目前常用的基于PCA(Principal Component Analysis)和SVR水质预测模型以及改进型遗传算法优化BP神经网络水质预测模型进行对比。仿真实验结果表明,本文提出的水质预测模型预测结果的准确性和稳定性相比于其他两种模型都有了一定的提高。
[Abstract]:Water environment is an indispensable part of the global natural ecological environment. At present, more and more serious pollution situation makes the protection and treatment of water environment more attention. Water quality prediction is an important content of water environment research and one of the important subjects of modern environmental science theory research. Considering the complexity of water environment system, the water quality prediction model of Miju River based on PLS(Partial Least Squares and SVR(Support Vector Regression is put forward according to the water quality characteristics of Miju River in Erhai River Basin. The main research results of this paper are as follows: when the principal component analysis is used to extract the principal components from the input variables of the water quality prediction model, the interpretation ability of the output variables of the water quality prediction model is relatively poor, which leads to the deterioration of the prediction accuracy of the water quality prediction model. In this paper, the partial least square regression method is used to extract the input variables of water quality prediction model, so that the extracted components can interpret the input variables as well as the output variables to the maximum extent. In this way, the input variables of water quality prediction model are reduced and the prediction accuracy is improved. In view of the problem of poor generalization ability of BP neural network back Propagation Nerual Network) to small sample data and instability of model output. In this paper, the support vector regression machine is used for nonlinear regression of water quality data of Miju River, which solves the problem of training and learning of small sample data and the improvement of prediction accuracy. Aiming at the problem that the optimization efficiency of genetic algorithm is not high and it is easy to fall into local minimum, this paper proposes an improved genetic algorithm to dynamically adjust the probability of crossover and mutation. At the same time, the initial parameters of support vector regression machine are optimized. Based on the water quality data of Miju River from Erhai River Monitoring Station in Dali Prefecture, the model presented in this paper is applied to the water quality prediction of Miju River. The model of water quality prediction based on PCA(Principal Component analysis, SVR and improved genetic algorithm is compared with BP neural network. The simulation results show that the accuracy and stability of the proposed water quality prediction model are better than those of the other two models.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X52;TP18

【参考文献】