Purpose Purpose Purpose Purpose Case study Case study Case study Case study Metamodel Metamodel Metamodel Metamodel performance Metamodel performance
Water network Application category Reference Application Location Size # Pipes in model /[area \(\mathbf{k}\mathbf{m}^{\mathbf{2}}\)] Classification by size Type Deviations from simple MLP Inputs (Number) Outputs (Number) Computational saving Accuracy
Urban drainage systems
Optimisation
(Seyedashraf et al., 2021)
Design
Bogotá, Colombia; Windsor, Canada
511 and 122
L, M
Stormwater - Real cases
Generalized regression - 2 hidden layers
SUDS characteristics: area, type, and location (20)
Boundary condition: Inflow (1)
95%
Mean error (<0.015) CC (0.99)
(W. Zhang et al., 2019) Design Urban catchment in China 182 M Stormwater* - Real case Ensemble of 100 MLPs Tank length and width (2) Flood depth (1) or peak flow (1) 80 - 90 % NSE (Between 0.66 and 0.92 depending on the rainfall scenario)
(Raei et al., 2019) Design Tehran, Iran [20 \(km^{2}\)] I Stormwater* - Real case 2 hidden layers Area sizes of the LID, Imperviousness and rainfall (3), TSS/BOD build-up (+1), TSS/BOD wash-off (+1) The volume of runoff (1) or BOD (1) or TSS (1) Not reported NSE (0.99)
(Latifi et al., 2019) Design Tehran, Iran [20 \(km^{2}\)] I Stormwater* - Real case Rainfall value, 6 build-up coefficients, 6 wash off coefficients, 6 imperviousness coefficients, and 32 values for area and type of LIDs (51) Runoff volume, BOD, TSS (3) Not reported Not mentioned
(Huang et al., 2015) Design Zhong-He district, Taiwan [20.29 \(km^{2}\)] L Stormwater* - Real case Catchment precipitation, Full pipe percentage of water flow in 3 points, the quantity and capacity of rain barrels in four regions (12) Water level/flooding at t + 1 (1) Not reported MAE (<15%) CC (>0.94 ~0.97)
Real-time (Kim & Han, 2020) Flood prediction Seoul, Korea [3.19 \(km^{2}\) *] M Stormwater* - Real case 8 hidden layers Total rainfall, Max. Rainfall in 1 - 3 hours, rainfall intensity, statistics (SD, Skewness, kurtosis), inter-event time (9) Total accumulative overflow (1) ~99% Mean relative errors between 2% - 62%
(Keum et al., 2020) Flood prediction Seoul, South Korea [7.4 \(km^{2}\)] M Stormwater* - Real case ANFIS Rainfall(t-1), Volume (t-1), Building coverage ratio Volume (t) 99%* NSE (0.959)*
(Kim et al., 2019) Flood prediction Gangnam area, Korea [7.4 \(km^{2}\)] M Stormwater* - Real case SVNARX and SOFM Accumulative rainfall Overflow at nodes (103) 98.50% NSE (0.6 - 0.94)
(She & You, 2019) Outflow prediction Tianjin, China 33 / [0.1314 \(km^{2}\)] S Real case with synthetic data Radial Basis function and NARX Rainfall intensities (6) Drainage outfall (1) Not reported SSE (0.273)
(Berkhahn et al., 2019) Flood prediction Anonymous 1224 and 299 L, I Stormwater* - Modifications of real cases 1 - 4 hidden layers Precipitation intensities every 5 minutes (24 for a 2h rain event) The maximum water level at different water cells NA RMSE (<0.35 cm)
(Chiang et al., 2010) Flood prediction Yu-Cheng, Taiwan [16.45 \(km^{2}\)] I Stormwater* - Real case RNN with 1 hidden layer, 3 neurons Registered water level and precipitation at time t (4) Water level at time t+n (1) NA NSE (>0.97), CC (>0.93), NRMSE (<0.26)
LFPB complement (Bermúdez et al., 2018) Surface flood volume estimation Ghent, Belgium 6025 / [27.50 \(km^{2}\)] L 85% Combined - Real case Ensemble of ANNs Rainfall-runoff volumes aggregated over 10 and 30 min windows and volume in the underground system of the closest storage cell (3) Presence of flooding (1) and magnitude (1) \(10^{4}\)x faster* NSE (~0.9) but variable
(Wolfs & Willems, 2017) Sewer water quantity simulation Ghent, Belgium 6025 / [27.50 \(km^{2}\)] L 85% Combined - Real case Volumes between two sub-catchments (2) Flow (1) \(10^{6}\)x faster* NSE (0.95 in average)
(Vojinovic et al., 2003) Wet weather flow prediction Catchment in Auckland, New Zealand [1.07 \(km^{2}\)] S Combined and Separated - Real case Radial Basis function Error, rainfall, model output (1 - 3) Error estimation of flow (1) NA Improvements of 15 - 26%