基于小波包分解与相空间重构的SSA-ELM水文时间序列预报模型  

SSA-ELM Hydrological Time Series Forecast Model Based on Wavelet Packet Decomposition and Phase Space Reconstruction

作　　者：李禄德 崔东文 LI Lude;CUI Dongwen(Wenshan Branch of Yunnan Hydrology and Water Resources Bureau,Wenshan 663000,China;Yunnan Province Wenshan Prefecture Water Affairs Bureau,Wenshan 663000,China)

机构地区：[1]云南省水文水资源局文山分局,云南文山663000 [2]云南省文山州水务局,云南文山663000

出　　处：《人民珠江》2022年第8期100-108,共9页Pearl River

摘　　要：针对水文时间序列非线性、多尺度等特征,提出基于小波包分解(WPD)和相空间重构的松鼠搜索算法(SSA)-极限学习机(ELM)预报模型,并应用于云南省上果水文站月径流和月降水量预报。首先利用WPD对径流和降水时序数据进行分解,并采用Cao方法对各子序列分量进行相空间重构;其次简要介绍SSA原理,通过各分量训练样本构建目标函数,利用SSA对目标函数进行寻优,并与鲸鱼优化算法(WOA)、灰狼优化(GWO)算法、粒子群优化(PSO)算法的寻优结果进行比较;最后利用SSA、WOA、GWO、PSO算法寻优获得的ELM输入层权值和隐含层偏值建立SSA-ELM、WOA-ELM、GWO-ELM、PSO-ELM模型及未经优化的ELM模型对各子序列分量进行预报,将预报结果加和重构得到最终预报结果。结果表明:SSA对各分量目标函数的寻优效果优于WOA、GWO、PSO算法,具有更好的寻优精度。SSA-ELM模型对月径流、月降水量预报的平均相对误差、平均绝对误差、圴方根误差、预报合格率分别为5.32%和3.84%、0.078 m3/s和0.169 mm、0.103 m3/s和0.209 mm、97.5%和95.8%,预报精度优于WOA-ELM等其他模型。Considering the nonlinear and multi-scale characteristics of hydrological time series,this paper proposes a squirrel search algorithm(SSA)-extreme learning machine(ELM)forecasting model based on wavelet packet decomposition(WPD)and phase space reconstruction.It is then applied to the Shangguo Hydrological Station in Yunnan Province for monthly runoff and precipitation forecasting.Specifically,WPD is performed to decompose the runoff and precipitation time series data,and the Cao method is applied to reconstruct the phase space of each subseries component.Then,the principle of SSA is outlined,and objective functions are constructed through the training samples of each component.The objective functions are optimized by SSA,and the results are compared with the optimization results of the whale optimization algorithm(WOA),the gray wolf optimization(GWO)algorithm,and the particle swarm optimization(PSO)algorithm.Finally,the weight of the ELM input layer and the hidden layer bias obtained by optimization based on SSA,WOA,GWO algorithm,and PSO algorithm,respectively,are utilized to build SSA-ELM,WOA-ELM,GWO-ELM,and PSO-ELM models,which,in addition to the unoptimized ELM models,are applied to forecast each subseries component,and the forecast results are summed and reconstructed to obtain the final forecasting results.The results show that SSA outperforms WOA,GWO algorithm,and PSO algorithm in optimizing the objective functions of each component and that it offers better optimization accuracy.The mean relative error,mean absolute error,mean square root error,and forecast pass rate of the proposed SSA-ELM model for monthly runoff and monthly precipitation forecast are 5.32%and 3.84%,0.078 m3/s and 0.169 mm,0.103 m3/s and 0.209 mm,97.5%and 95.8%,respectively,indicating that its forecasting accuracy is higher than that of other models such as the WOA-ELM model.

关 键 词：水文预报 小波包分解 相重间重构 松鼠搜索算法 极限学习机 

分 类 号：P338[天文地球—水文科学]