Accurate prediction of shear stress distribution along the boundary in an open channel is the key to solving numerous critical engineering problems such as flood control, sediment transport, riverbank protection, and others. Similarly, the estimation of flow discharge in flood conditions is also challenging for engineers and scientists. The flow structure in compound channels becomes complicated due to the transfer of momentum between the deep main channel and the adjoining floodplains, which affects the distribution of shear force and flow rate across the width. Percentage sharing of shear force at floodplain (%S_fp) is dependent on the non-dimensional parameters like width ratio of the channel \((\alpha )\), relative depth \((\beta )\), sinuosity \((s)\), longitudinal channel bed slope \((S_{{\text{o}}} ),\) meander belt width ratio \((\omega )\), and differential roughness \((\gamma )\). In this paper, various artificial intelligence approaches such as multivariate adaptive regression spline (MARS), group method of data handling Neural Network (GMDH-NN), and gene-expression programming (GEP) are adopted to construct model equations for determining %S_fp for meandering compound channels with relative roughness. The influence of each parameter used in the model for predicting the %S_fp is also analyzed through sensitivity analysis. Statistical indices are employed to assess the performance of these models. Validation of the developed %S_fp model is performed for the experimental observations by conventional analytical models; to verify their effectiveness. Results indicate that the proposed GMDH-NN model predicted the %S_fp satisfactorily with the coefficient of determination (R²) of 0.98 and 0.97 and mean absolute percentage error (MAPE) of 0.05% and 0.04% for training and testing dataset, respectively as compared to GEP and MARS. The developed model is also validated with various sinuous channels having sinuosity 1.343, 1.91 and 2.06.

准确预测明渠边界沿线的剪应力分布是解决防洪、输沙、河岸保护等众多关键工程问题的关键。同样，洪水条件下的流量估计对工程师和科学家来说也具有挑战性。由于深部主河道与相邻洪泛区之间的动量传递，复合河道中的水流结构变得复杂，这影响了剪切力和流速在宽度上的分布。洪泛区剪切力的百分比共享 (% S _fp ) 取决于无量纲参数，例如通道的宽度比\((\alpha )\)、相对深度\((\beta )\), 曲率\((s)\) , 纵向河床坡度\((S_{{\text{o}}} ),\)曲流带宽度比\((\omega )\) , 和差异粗糙度\(( \gamma )\)。本文采用多元自适应回归样条（MARS）、数据处理神经网络组法（GMDH-NN）、基因表达编程（GEP）等多种人工智能方法构建模型方程来确定% S _fp用于弯曲具有相对粗糙度的复合通道。模型中用于预测 % S _fp的每个参数的影响也通过敏感性分析进行分析。统计指标用于评估这些模型的性能。已开发的 % S _fp模型的验证是通过常规分析模型对实验观察进行的；以验证其有效性。结果表明，所提出的 GMDH-NN 模型以 0.98 和 0.97的决定系数 ( R ² ) 以及训练和测试数据集的平均绝对百分比误差 ( MAPE ) 分别为 0.05% 和 0.04%令人满意地预测了 % S _fp，如下所示与 GEP 和 MARS 相比。开发的模型还通过具有曲率 1.343、1.91 和 2.06 的各种曲折通道进行了验证。