Discharge forecasting using an Online Sequential Extreme Learning Machine (OS-ELM) model: A case study in Neckar River, Germany

详细信息    查看全文

• 作者：Basant Yadav^a ; Sudheer Ch^b ; ^{sudheer108@gmail.com" class="auth_mail" title="E-mail the corresponding author} ; Shashi Mathur^a ; Jan Adamowski^c
• 关键词：Flood forecasting ; Extreme Learning Machine ; Artificial intelligence techniques
• 刊名：Measurement
• 出版年：2016
• 出版时间：October 2016
• 年：2016
• 卷：92
• 期：Complete
• 页码：433-445
• 全文大小：2153 K

文摘

Flood forecasting in natural rivers is a complicated procedure because of uncertainties involved in the behaviour of the flood wave movement. This leads to complex problems in hydrological modelling which have been widely solved by soft computing techniques. In real time flood forecasting, data generation is continuous and hence there is a need to update the developed mapping equation frequently which increases the computational burden. In short term flood forecasting where the accuracy of flood peak value and time to peak are critical, frequent model updating is unavoidable. In this paper, we studied a new technique: Online Sequential Extreme Learning Machine (OS-ELM) which is capable of updating the model equation based on new data entry without much increase in computational cost. The OS-ELM was explored for use in flood forecasting on the Neckar River, Germany. The reach was characterized by significant lateral flow that affected the flood wave formation. Hourly data from 1999–2000 at the upstream section of Rottweil were used to forecast flooding at the Oberndorf downstream site with a lead time of 1–6 h. Model performance was assessed by using three evaluation measures: the coefficient of determination (R²), the Nash-Sutcliffe efficiency coefficient (NS) and the root mean squared error (RMSE). The performance of the OS-ELM was comparable to those of other widely used Artificial Intelligence (AI) techniques like support vector machines (SVM), Artificial Neural Networks (ANN) and Genetic Programming (GP). The frequent updating of the model in OS-ELM gave a closer reproduction of flood events and peak values with minimum error compared to SVM, ANN and GP.