为提高水电站入库径流多步预测精度，精准调控发电用水量，基于小波包变换（WPT）数据分解方法，菲克定律优化（FLA）算法、正则化极限学习机（RELM），提出WPT-FLA-RELM预测模型，并将其应用于云南省马鹿塘水电站入库日径流多步预测。首先利用WPT将水电站入库日径流时间序列分解为1个周期项分量和3个波动项分量；然后利用FLA及遗传算法（GA）、粒子群优化（PSO）算法优化RELM输入层权值和隐含层偏差（超参数）；最后利用优化获得的最佳超参数建立WPT-FLA-RELM及WPTGA-RELM、WPT-PSO-RELM模型对入库日径流周期项及波动项分量进行预测，重构后得到最终预测结果，并构建WPT-FLA-极限学习机（ELM）、WPT-FLA-BP模型和未经优化的WPT-RELM模型作对比分析模型。

In order to improve the accuracy of multi-step prediction of inflow runoff of hydropower stations and accurately regulate the water consumption for power generation, a WPT-FLA-RELM prediction model is proposed based on WPT data decomposition method, FLA algorithm, and RELM, and applied to multi-step prediction of inflow daily runoff of Malutang Hydropower Station in Yunnan Province. Firstly, WPT is used to decompose the daily runoff time series of hydropower stations into one periodic component and three fluctuation components. Then, FLA, GA, and PSO algorithms are used to optimize the input layer weights and hidden layer deviations(hyperparameters) of RELM. Finally, the optimal hyperparameters obtained through optimization were used to establish WPT-FLA-RELM, WPT-GA-RELM, and WPT-PSO-RELM models to predict the periodic and fluctuating components of daily inflow runoff. After reconstruction, the final prediction results were obtained, and WPT-FLA ELM, WPT-FLA-BP model, and unoptimized WPT-RELM model were constructed for comparative analysis.