As one of the most applicable flow diversion structures, side weirs are utilized for adjusting and measuring the flow in the floodplain. In practice, in order to facilitate the design, the cross section of main channels is constructed trapezoidal. In this paper, a modern evolutionary artificial intelligence approach entitled "Generalized Structure Group Method of Data Handling (GSGMDH)" is used for the first time for approximating and modeling the discharge coefficient of side weirs placed upon trapezoidal main channels. Compared to the group method of data handling (GMDH) classical method, GSGMDH has more flexibility and higher ability in estimating different phenomena, because nodes existing in the hidden layer can make connection with non-adjacent layers. Initially, all parameters influencing the discharge coefficient of side weirs installed on trapezoidal canals are identified. Then, seven GSGMDH models with various architectures are developed using the mentioned parameters. For training the artificial intelligence models, 70% of the experimental data are implemented and the remaining 30% are used for testing them. The superior model is introduced through the analysis of the modeling results. The superior model simulates the discharge coefficient values with acceptable accuracy. For example, the values of the correlation coefficient (R), Scatter Index (SI) and the Nash–Sutcliffe efficiency coefficient (NSC) are calculated equal to 0.987, 0.025 and 0.987, respectively, in the testing mode. Based on the sensitivity analysis results, the flow Froude number (Fr) and the ratio of the height of the side weir crest to the flow depth at the weir upstream (W/y₁) are introduced as the most influencing input parameters. Subsequently, the results of the superior GSGMDH model are compared with the classical GMDH model and it is revealed that the GSGMDH has a greater performance. For instance, about 19% of the GMDH model results have errors more than 10%, while this is about 2% for those of the GSGMDH. After that, an uncertainty analysis is carried out for these models to exhibit that the GSGMDH has an underestimated performance. Moreover, the comparison of the superior model results with the previous studies confirms the superiority of the GSGMDH over the earlier artificial intelligence models. Lastly, a formula is put forward for the superior GSGMDH model for evaluating the effects of the input parameters on the changing pattern of the objective parameter through the conduction of a partial derivative sensitivity analysis (PDSA).

作为最适用的分流结构之一，侧堰用于调节和测量洪泛区的流量。实际上，为了便于设计，主通道的横截面构造为梯形。在本文中，首次使用一种名为“数据处理的通用结构组方法”（GSGMDH）的现代进化人工智能方法来近似和建模梯形主通道上的侧堰的流量系数。与经典的数据处理组方法（GMDH）相比，GSGMDH具有更大的灵活性和更高的估计不同现象的能力，因为隐藏层中存在的节点可以与非相邻层建立连接。原来，确定了影响梯形渠中安装的侧堰排放系数的所有参数。然后，使用上述参数开发了具有不同架构的七个GSGMDH模型。为了训练人工智能模型，实施了70％的实验数据，其余30％用于测试它们。通过对建模结果的分析来引入上级模型。上级模型以可接受的精度模拟排放系数值。例如，相关系数的值（70％的实验数据已实现，其余30％用于测试它们。通过对建模结果的分析来引入上级模型。上级模型以可接受的精度模拟排放系数值。例如，相关系数的值（70％的实验数据已实现，其余30％用于测试它们。通过对建模结果的分析来引入上级模型。上级模型以可接受的精度模拟排放系数值。例如，相关系数的值（在测试模式下，R），散射指数（SI）和纳什-苏特克利夫效率系数（NSC）分别等于0.987、0.025和0.987。基于灵敏度分析结果，流量弗洛德数（Fr）和侧堰顶高与上游堰的流深之比（W / y ₁）作为最有影响力的输入参数。随后，将高级GSGMDH模型的结果与经典GMDH模型进行比较，结果表明GSGMDH具有更好的性能。例如，约GMDH模型结果的19％的误差超过10％，而GSGMDH的误差约为2％。之后，对这些模型进行不确定性分析，以显示GSGMDH的性能被低估了。此外，优越模型结果与先前研究的比较证实了GSGMDH优于早期人工智能模型。最后，