【作者】 王妮妮；

【导师】 刘晓东；

【作者基本信息】 大连理工大学 ， 控制理论与控制工程， 2013， 博士

【摘要】 径流、降水、温度等水文气象时间序列是用于研究与地球相关现象的基本数据之一自然因素与人类活动引起环境的渐进性与颠覆性改变会导致水文气象时间序列的渐进变化或跳跃变化。水文气象时间序列的分割与预测研究有助于理解自然因素或人类活动引起的人类生存环境的改变。由于在研究与实践中的重要性,对径流、降水、温度等水文气象记录的分割与预测问题已经受到广泛关注。水文气象时间序列的分割与预测问题已成为水文气象科学研究的重要课题。新兴的智能计算技术在计算效率上具有十分显著的优越性,并已广泛应用于水文气象时间序列的分割与预测研究。本文采用智能计算技术对水文气象时间序列进行分割与预测,主要研究内容包括三个部分：(1)提出用于自动模糊分割一元与多元水文气象时间序列的改进的Gath-Geva聚类算法。该算法把时间序列分割问题看成是Gath-Geva聚类,采用最小信息长度准则作为分割阶数的选择标准。改进的Gath-Geva聚类算法具有以下优点：首先,具有非监督特性,可以自动确定最优的分割阶数；其次,在经典Gath-Geva聚类中采用分量形式的期望最大化算法以避免经典的期望最大化算法中对初值敏感及需要避免收敛到参数空间的边界等缺点；另外.通过把分割阶数选择与模型参数估计过程紧密结合,可以提高算法的稳定性。(2)提出一种用于自动模糊分割多元时间序列的竞争模糊聚类算法。该算法基于新的竞争机制,能够自动确定聚类数目并有效的选择“分裂”或“合并”操作。而且该算法对聚类数目与模型参数不敏感。人工数据实验表明该算法能够检测出多元时间序列的概率特性的改变(均值改变、方差改变与变量相关结构的改变)。(3)注意到用于水文气象时间序列预测的神经网络结构通常是静态的,而静态神经网络无法有效反映系统动态的变化。本文利用序贯学习的径向基函数神经网络对水文气象时间序列进行精确实时预测。该预测模型采用滑动数据窗口作为动态观测器来调整网络结构和参数以适应水文气象时间序列的动态变化。采用这三个算法对人工和水文气象时间序列进行分割与预测仿真实验,结果验证了这三个算法对于时间序列分割与预测的有效性。更多还原

【Abstract】 Hydrometeorological time series such as streamflow, precipitation and temperature are among the basic data used to study earth-related phenomena. Evolutionary and disruptive changes in the environment caused by natural factors and human activities result in trends or jumps in hydrometeorological time series. Segmentation and prediction for hydrometeorological time series help men understand natural factors or human-induced changes in human living environment. Due to the importance in research and practice, segmentation and prediction of hydrometeorological records such as streamflow, precipitation and temperature have been received extensive attention. Segmentation and prediction for hydrometeorological time series have always been important topics in hydrometeorological sciences. The emerging intelligent computing techniques have shown their power in computational efficiency, and have been widely applied in segmentation and prediction of hydrometeorological time series. In this paper, intelligent computing techniques are conducted to segmentation and prediction of hydrometeorological time series. The main research contents of this paper include three parts:Firstly, an improved Gath-Geva clustering algorithm is proposed for automatic fuzzy segmentation of univariate and multivariate hydrometeorological time series. The algorithm considers time series segmentation problem as Gath-Geva clustering with the minimum message length criterion as segmentation order selection criterion. One characteristic of the improved Gath-Geva clustering algorithm is its unsupervised nature which can automatically determine the optimal segmentation order. Another characteristic is the application of the modified component-wise expectation maximization algorithm in Gath-Geva clustering which can avoid the drawbacks of the classical expectation maximization algorithm:the sensitivity to initialization and the need to avoid the boundary of the parameter space. The other characteristic is the improvement of numerical stability by integrating segmentation order selection into model parameter estimation procedure.Secondly, a competitive fuzzy clustering algorithm is presented for automatic fuzzy segmentation of multivariate time series. The proposed algorithm is capable of automatically choosing the clustering number and selecting the "split" or "merge" operations efficiently based on the new competitive mechanism. It is insensitive to the initial configuration of the cluster component number and model parameters. Experiments on synthetic data show that the proposed algorithm is able to handle time-varying characteristics of multivariate time series:changes in the mean:changes in the variance; and changes in the correlation structure among the variables. Thirdly, noticed that the structures of the commonly implemented neural networks for the prediction of hydrometeorological time series are static, while the static neural network cannot represent the changes of system dynamics efficiently. In this paper, the application of a sequential learning radial basis function is presented for accurate real-time prediction of hydrometeorological time series. The prediction model employs a sliding data window as dynamical observer, and tunes the structure and parameters of radial basis function neural network to adapt to the dynamical changes of hydrometeorological time series.The three algorithms have been experimentally tested on artificial and hydrometeorological time series. The obtained results show the effectiveness of the algorithm for time series segmentation and prediction.更多还原