More accurate prediction of the complex velocity field in sewers based on uncertainty analysis using extreme learning machine technique

Abstract

ABSTRACT In this study, we present a novel procedure to simulate the velocity distribution of a narrow sewer channel using a combination of a simple and accurate machine learning tool, namely, extreme learning machine (ELM). We also examined the uncertainty associated with the model in simulating the more complex velocity–depth profiles in narrow sewer channels using the 95 percent predicted uncertainties (95PPU%) and d-factor indices. The data for estimating the velocity distribution were measured at a municipal sewer pipe in France. We measured the velocity distribution in a cross section for 10 different depths with width-to-depth ratios ranging from 1.75 to 2.44. The new ELM model has the 95PPU% and the d-factor of 53.25% and 0.36, respectively, in the training dataset and 84.93% and 1.9 in the test dataset. The error analysis confirmed that ELM model outperforms all existing models for predicting velocity profile in narrow sewers. The proposed method is compared with existing methods, and it appears that in addition to being simpler and having no limitations, this model exhibits better performance as well (root mean square error (RMSE) = 0.04 and mean absolute percentage error (MAPE) = 3.56%).