海水位预测误差分析及系统

发布时间：2018-06-02 15:49

本文选题：海水位波动 + 海水位波动预测　；参考：《华东师范大学》2017年硕士论文

【摘要】：随着全球变暖不断加剧,海水位的上升问题逐渐引起了人们的关注。大量研究基于长期海水位的趋势分析,相比之下,短时海水位波动的研究相对较少。本文实现了对短时海水位时间序列的预测,并对预测结果进行了分析,设计了海水位预测系统。本文的主要亮点有:一是使用了线性模型——自回归模型和非线性模型——支持向量机和BP神经网络对海水位时间序列进行了预测。结果表明,支持向量机和BP神经网络在样本长度较小时表现出了比自回归模型更为精确的预测结果,使用非线性模型大大缩短了预测所需的样本长度,并在增大预测步长时依然保持着低于线性模型的预测误差。实验证明,支持向量机和BP神经网络的应用将大大节约海水位时间序列预测成本;二是得到了海水位时间序列预测误差与预测步长的关系曲线,并用幂函数拟合得到具体的函数表达式。本文使用了支持向量机、BP神经网络和自回归模型分别验证了预测误差与预测步长的定量关系,对今后工作中预测步长的选取和预测误差的估计有着借鉴意义;三是利用MALTAB GUI平台设计了海水位时间序列预测系统。该系统具有海水位时间序列的预测功能,可以计算出分别使用三类预测模型时的预测误差,并自动生成预测误差与预测步长关系曲线、预测误差与样本长度关系曲线,产生定量关系式。用户使用该系统可以全面了解预测模型,深入理解海水位时间序列的预测结果。
[Abstract]:With the increasing of global warming, the problem of rising seawater level has attracted more and more attention. A large number of studies are based on the trend analysis of long term seawater level. In this paper, the prediction of short-term seawater potential time series is realized, and the prediction results are analyzed, and a seawater level prediction system is designed. The main highlights of this paper are as follows: first, the potential time series of seawater are predicted by using linear model-autoregressive model and nonlinear model-support vector machine (SVM) and BP neural network. The results show that support vector machine and BP neural network show more accurate prediction results than the autoregressive model when the sample length is small, and the nonlinear model can greatly shorten the sample length required for prediction. The prediction error is still lower than that of the linear model when the prediction step is increased. Experimental results show that the application of support vector machine and BP neural network will greatly reduce the prediction cost of seawater potential time series, and the relationship curve between prediction error and prediction step size of seawater potential time series is obtained. The specific function expression is obtained by power function fitting. In this paper, support vector machine BP neural network and autoregressive model are used to verify the quantitative relationship between prediction error and prediction step size, which is useful for the selection of prediction step size and estimation of prediction error in future work. The third is the design of seawater potential time series prediction system based on MALTAB GUI platform. The system has the function of predicting the time series of seawater potential. It can calculate the prediction errors when using three kinds of prediction models, and automatically generate the curve of relation between prediction error and prediction step size, and the relationship curve between prediction error and sample length. A quantitative relationship is produced. With the system, the prediction model can be fully understood and the prediction results of seawater potential time series can be deeply understood.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O211.61

【参考文献】