基于小波消噪与秩次集对分析的水文时间序列预测模型

Hydrologic temporal series prediction model based on wavelet de-noising and rank and set-pair analysis

对水文现象观测得到的水文时间序列,通常具有趋势性、周期性和随机性等多项特征.特别是在大尺度条件下,传统水文时间序列预测模型存在构建方法单一、多未考虑噪声影响等问题.为此,本文将小波消噪(wavelet de-noise,WD)与秩次集对分析(rank and set pair analysis,RSPA)联合使用,建立了基于小波消噪与秩次集对分析的水文时间序列预测模型(WD-RSPA模型),以充分发挥小波分析多尺度分析、消噪的独特性能和RSPA概念清晰、计算简单的优势,并克服集合元素分类标准确定的主观性.应用所建模型对黄河花园口站1998—2007年的年径流量以及郑州站2001—2009年的年降水量进行了预测,与传统模型预测结果加以对比.结果显示,在合适消噪小波函数以及集合维数下,WD-RSPA模型能够有效避免噪声对模型的影响,模型构建概念清晰、计算简单、预测结果精度较高,验证了所建模型的适用性和优越性.

Observations of hydrologic series are always with tendency, periodicity, randomness and other characteristics. Especially under large-scale conditions, there are many problems in the hydrologic series prediction, such as single construction method and without consideration of the noise. Therefore, a hydrologic series prediction model （WD-RSPA model） is developed on the combination of wavelet de-noise（WD）method and rank and set-pair analysis（RSPA）, which takes advantage of multi scale analysis and noise reduction in wavelet analysis and clear concept, simple calculation, and overcomes the subjectivity to certain the standard of set elements in RSPA. The WD-RSPA model is applied to predict the annual runoff of Huayuankou Station and the annual precipitation of Zhengzhou Station. The prediction results by WD-RSPA model are compared with results of traditional RSPA model, AR model and BP neural network model. It indicates that, with appropriate de-noising wavelet function and pair dimension, WDRSPA model can avoid the impact of noise efficiently, the concept to establish WD-RSPA model is clear, the computation is simper and the accuracy of prediction results is higher. Consequently, the applicability,dependability and advantage of WD-RSPA model is validated.