基于小波包分解的AJS-GMDH月径流时间序列预测研究  

Research on AJS-GMDH Monthly Runoff Time Series Forecast Based on Wavelet Packet Decomposition

作　　者：杨琼波 崔东文 YANG Qiongbo;CUI Dongwen(Yunnan Provincial Bureau of Hydrology and Water Resources Honghe Branch,Honghe 661199,Yunnan,China;Yunnan Wenshan Prefecture Water Affairs Bureau,Wenshan 663000,Yunnan,China)

机构地区：[1]云南省水文水资源局红河分局,云南红河661199 [2]云南省文山州水务局,云南文山663000

出　　处：《水力发电》2022年第6期45-51,共7页Water Power

摘　　要：为提高月径流时间序列预测精度,建立基于小波包分解(WPD)、人工水母搜索(AJS)算法、数据分组处理方法(GMDH)的WPD-AJS-GMDH月径流时间序列预测模型。采用WPD将月径流时序数据分解为若干子序列分量;选取6个典型函数在不同维度条件下对AJS算法进行仿真测试;利用AJS算法优化GMDH网络关键参数,建立WPD-AJS-GMDH模型,并构建基于支持向量机(SVM)、BP神经网络及完全集合经验模态分解(CEEMD)、小波分解(WD)的17种对比分析模型;最后利用云南省龙潭站1952年~2016年780组的月径流时间序列数据对所建立的18种模型进行检验。结果表明,在不同维度条件下,AJS算法均具有较好的寻优效果;WPD-AJS-GMDH模型预测误差均小于其他17种模型;对于月径流时序数据分解,WPD分解效果优于CEEMD、WD方法;AJS算法能有效优化GMDH网络参数,提高预测性能。In order to improve the prediction accuracy of monthly runoff time series,a WPD-AJS-GMDH monthly runoff time series prediction model based on wavelet packet decomposition(WPD),artificial jellyfish search(AJS)algorithm,and data packet processing method(GMDH)is established.Firstly,the WPD is used to decompose the monthly runoff time series data into several sub-sequence components.Secondly,six typical functions are selected to simulate and test the AJS algorithm under different dimensional conditions.Thirdly,the AJS algorithm is used to optimize the key parameters of the GMDH network and establish the WPD-AJS-GMDH model,and other 17 kinds of analysis models based on support vector machine(SVM),BP neural network,complete ensemble empirical mode decomposition(CEEMD),wavelet decomposition(WD)are built for forecast result comparison.Finally,780 sets of monthly runoff time series data of Longtan Station in Yunnan from 1952 to 2016 are used to test the 18 models.The results show that:(a)under different dimensional conditions,the AJS algorithm has a good optimization effect;(b)the prediction errors of the WPD-AJS-GMDH model are all smaller than those of the other 17 models;(c)for the decomposition of monthly runoff time series data,the WPD decomposition effect is better than the CEEMD and WD methods;and(d)the AJS algorithm can effectively optimize the GMDH network parameters and improve the prediction performance.

关 键 词：月径流预测 时间序列分解 人工水母搜索算法 数据分组处理方法 仿真测试 

分 类 号：TV121[水利工程—水文学及水资源]