水电能源科学

受水文序列非平稳性和复杂性影响，传统单一模型预测精度有限。为提高月径流预测精度，基于季节趋势分解(STL)—小波包变换(WPT)二次分解技术、多策略山猫优化算法(MSOA)/多策略耳廓狐优化(MFFO)算法和在线惯序极限学习机(OSELM),提出STL-WPT-MSOA/MFFO-OSELM模型，通过云南省南康河下游南康河水文站、勐统河下游勐大水文站月径流预测实例进行验证。首先利用STL将原始月径流序列分解为趋势分量、季节分量和残差分量，通过WPT将残差分量分解为1个高频分量和1个低频分量，划分各分量训练集和验证集，并基于训练集构建OSELM超参数优化实例目标函数；然后基于Tent混沌映射等多种策略改进山猫优化算法(SOA)和耳廓狐优化(FFO)算法，提出多策略MSOA/MFFO,利用MSOA/MFFO优化实例目标函数获得OSELM最优超参数；最后利用最优超参数建立STL-WPT-MSOA/MFFO-OSELM模型对各分量进行预测和重构，并构建12种模型作对比分析。结果表明，STL-WPT-MSOA/MFFO-OSELM融合模型预测效果最佳，能更精准地捕获原始月径流量的变化特征和规律；多种策略改进方法能有效提升MSOA/MFFO性能，获得更佳OSELM超参数；STL-WPT二次分解技术能有效地消除月径流非平稳性特征，改进月径流序列分解效果。研究方法及结果可为水文时间序列预测提供参考。

Due to the nonstationarity and complexity of hydrological series, the prediction accuracy of traditional single model is limited. To improve the accuracy of monthly runoff prediction, the STL-WPT-MSOA/MFFO-OSELM model is proposed and validated through monthly runoff prediction examples at the Nankang River and Mengda hydrological stations in Yunnan Province. Firstly, STL is used to decompose the original monthly runoff sequence into trend component, seasonal component, and residual component. WPT is used to decompose the residual component into one high-frequency component and one low-frequency component, and the training set and validation set of each component are divided. Based on the training set, the objective function of the OSELM hyperparameter optimization instance is constructed. Then, based on various strategies such as Tent chaotic mapping, the Bobcat Optimization Algorithm(SOA) and the Ear Fox Optimization(FFO) algorithm were improved, and a multi-strategy MSOA/MFFO was proposed. The objective function of the MSOA/MFFO optimization instance was used to obtain the optimal hyperparameters of OSELM. Finally, the STL-WPT-MSOA/MFFO-OSELM model was established using the optimal hyperparameters to predict and reconstruct each component, and 12 models were constructed for comparative analysis. The results show that the STL-WPT-MSA/MFFO-OSELM fusion model had the best prediction performance and could more accurately capture the changing characteristics and patterns of the original monthly runoff; Multiple strategy improvement methods can effectively enhance the performance of MSOA/MFFO and obtain better OSELM hyperparameters; The STL-WPT secondary decomposition technique can effectively eliminate the nonstationary characteristics of monthly runoff and improve the decomposition effect of monthly runoff sequences. The research methods and results provide reference for hydrological time series prediction.