ESTUDO COMPARATIVO ENTRE OS MODELOS STORM WATER MANAGEMENT E REDES NEURAIS ARTIFICIAIS NA AVALIAÇÃO DO DESEMPENHO DE TRINCHEIRA DE INFILTRAÇÃO

Abstract

Stormwater infiltration systems applied to the lots gain importance as compensatory practices of runoff management. In this research two models were compared for the performance evaluation of an experimental infiltration trench built in Campinas, São Paulo, Brazil. In order to do so, the Storm Water Management Model 5.1 (SWMM)was compared to an Artificial Neural Networks (ANN) model developed through experimental area monitoring data. Two design rainfalls were tested: one of 52 mm/h and 30 min of duration, and other of 192 mm/h and 8 min of duration, whose volumes were discharged into infiltration trench, with cross section of 0,6 x 0,6 m and length of 5,0 m, filled with crushed stone. The hydraulic profiles simulated by SWMM 5.1 and ANN model were compared with experimental data. The ANN model was better adjusted to the observed performance data, resulting in design volumes between 0.5% and 0.75% higher than those observed for design rains. The SWMM model presented values between 32.7% and 41.1% higher than those observed. In terms of constructive cost, the ANNsized system would result in savings of 24% to 28.8% higher than SWMM-sized systems.