基于小波包分解-非洲秃鹫优化算法-深度极限学习机的水文预报模型及其应用  

Hydrological Forecasting Model Based on Wavelet Packet Decomposition-African Vulture Optimization Algorithm-Deep Extreme Learning Machine and Its Application

作　　者：王忠义 崔东文 WANG Zhong-yi;CUI Dong-wen(Beijing Hengsheng Hongda Road Investment Co.,Ltd.,Beijing 100070,China;Yunnan Province Wenshan Water Bureau,Wenshan 663000,China)

机构地区：[1]北京恒盛宏大道路投资有限公司,北京100070 [2]云南省文山州水务局,云南文山663000

出　　处：《水电能源科学》2022年第8期26-31,共6页Water Resources and Power

摘　　要：针对水文序列多尺度、非线性、非平稳性特点,将小波包分解(WPD)与非洲秃鹫优化算法(AVOA)—深度极限学习机(DELM)相结合,建立WPD-AVOA-DELM组合预测模型,利用AVOA优化DELM输入层权值、隐含层偏值或隐含层神经元数,分别建立WPD-AVOA-DELM^(1)(优化DELM输入层权值、隐含层偏值)、WPD-AVOA-DELM^(2)(优化DELM隐含层神经元数)模型,利用WPD将云南省革雷站水文序列分解为8个子序列分量,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,WPD-AVOA-DELM^(1)、WPD-AVOA-DELM^(2)模型对革雷站月径流、月降水量、年径流、年降水量预报的平均相对误差分别为1.86%、8.82%、0.79%、0.52%和1.97%、8.30%、0.92%、0.71%,预报精度远高于单一模型AVOA-DELM^(1)、AVOA-DELM^(2)、AVOA-SVM,高于组合模型WPD-AVOA-SVM,具有更好的预报效果。可见WPD能科学降低水文序列的复杂性,提高预报精度;AVOA能有效优化DELM关键参数,提高DELM网络性能。Aiming at the multi-scale,non-linear,and non-stationary characteristics of the hydrological sequence,the wavelet packet decomposition(WPD)was combined with the African vulture optimization algorithm(AVOA)and deep extreme learning machine(DELM)to establish a WPD-AVOA-DELM combined forecasting model.The WPD was used to decompose the hydrological sequence of Gelei station in Yunnan Province into 8 sub-sequence components.The AVOA was used to optimize the DELM’s input layer weight,hidden layer bias or the number of hidden layer neurons,and establish WPD-AVOA-DELM^(1)(Optimizing the DELM input layer weights,hidden layer bias),WPD-AVOA-DELM^(2)(optimizing the number of neurons in the DELM hidden layer)models to predict each sub-sequence component,and add the prediction results to reconstruct the final prediction values.The results show that the average relative errors of the WPD-AVOA-DELM^(1) and WPD-AVOA-DELM^(2) models for the forecasting monthly runoff,monthly precipitation,annual runoff and annual precipitation at Gelei Station are 1.86%,8.82%,0.79%,0.52%,1.97%,8.30%,0.92%,0.71%.The forecast accuracy is much higher than the single model AVOA-DELM^(1),AVOA-DELM^(2),AVOA-SVM,and higher than the combined model WPD-AVOA-SVM,with better forecasting effect.Thus,the WPD can scientifically reduce the complexity of hydrological sequences and improve forecast accuracy as well as the AVOA can effectively optimizes the key DELM parameters and improves the DELM network performance.

关 键 词：水文预报 小波包分解 非洲秃鹫优化算法 深度极限学习机 

分 类 号：TV124[水利工程—水文学及水资源] P338[天文地球—水文科学]