Abstract

Predicting the Suspended Sediment Load (SSL) of the river is a very significant and challenging task. Being a non-linearity in the SSL data, it requires a non-linear method to get predicated in absence of field measurements. In recent decades, the application of Artificial Neural Networks (ANNs) has been considered as a milestone for the prediction of different hydrological parameters. Therefore, the present study was performed to investigate the application of ANNs to estimate the total SSL, particularly in the absence of river discharge data. The predication of SSL was done for the Ganga River at a station situated in the city of Varanasi, India. A multilayer perceptron (MLP) ANNs with a back-propagation algorithm were used to predict the SSL. The reliability of models was tested using the performance criteria such as Normalized Root Mean Square Error (NRMSE), Theil’s U statistics and Correlation coefficient. Several input parameters like rainfall, SSL, River stage. River depth and change in river bed level were used on the basis of different cases and tested to predict the SSL. The results show that the model performed efficiently to predict the SSL in the Ganga River, even in the absence of daily discharge data of the river. The values of SSL were also predicted using the predicted previous time step values of SSL to simulate the model in-case of missing field values. The model was found efficient to predict the values up to 95% of accuracy in-case of missing field value of SSL.

中文翻译：

---

人工神经网络模型在大河悬浮泥沙负荷预测中的应用

摘要

预测河流的悬浮沉积物负荷 (SSL) 是一项非常重要且具有挑战性的任务。作为 SSL 数据中的非线性，它需要一种非线性方法来在没有现场测量的情况下进行预测。近几十年来，人工神经网络（ANNs）的应用被认为是预测不同水文参数的里程碑。因此，本研究旨在调查应用 ANN 来估计总 SSL，特别是在没有河流流量数据的情况下。SSL 的预测是在位于印度瓦拉纳西市的一个站对恒河进行的。使用具有反向传播算法的多层感知器 (MLP) ANN 来预测 SSL。. 几个输入参数，如降雨量、SSL、河流水位。在不同情况下使用河流深度和河床水位变化并进行测试以预测 SSL。结果表明，即使在没有河流每日流量数据的情况下，该模型也能有效地预测恒河中的 SSL。SSL 的值也使用 SSL 的预测的先前时间步长值来预测，以在缺少字段值的情况下模拟模型。发现该模型在 SSL 字段值缺失的情况下可以有效预测高达 95% 的准确率。