Application of novel hybrid machine learning techniques for particle Froude number estimation in sewer pipes

The hydraulic capacity of the channel is significantly impacted by the deposition of sediment in sewers and urban drainage systems. Sediment deposition affects a channel's hydraulic capacity in urban drainage and sewage systems. To decrease the effects of this continuous deposition of silt particles, sewer systems frequently have a self-cleaning device to keep the channel bottom clear from sedimentation. Therefore, accurate particle Froude number (Fr) prediction is essential for sewage system design. This study looked at three datasets from the literature that covered a wide range of volumetric sediment concentration (C_v), dimensionless grain size of particles (D_gr), sediment median size (d), hydraulic radius (R), and pipe friction factor for the condition of non-deposition without deposited bed. We employed Kstar, M5P, and random forest (RF) models as standalone models as well as additive regression (AR) models as hybrid machine learning (ML) models for the prediction of Fr. In all, we looked at six ML methods: Kstar, AR-Kstar, M5P, AR-M5P, RF, and AR-RF. Several performance metrics, including mean absolute error (MAE), Nash–Sutcliffe efficiency (NSE), root-mean-square error (RMSE), Pearson correlation coefficient (R), etc., have been used to assess the performance of suggested models. In comparison to standalone ML models and empirical equations, hybrid ML models perform better. For the prediction of particle Froude number (Fr) in sewage system design under the condition of non-deposition without deposited bed, AR-Kstar (MAE = 0.435, NSE = 0.922, and RMSE = 0.623, and R² = 0.923) performed the best, followed by AR-RF, Kstar, RF, AR-M5P, and M5P.