Premium
Comparison of ANN models for predicting water quality in distribution systems
Author(s) -
D'Souza Celia D.,
Kumar M.S. Mohan
Publication year - 2010
Publication title -
journal ‐ american water works association
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.466
H-Index - 74
eISSN - 1551-8833
pISSN - 0003-150X
DOI - 10.1002/j.1551-8833.2010.tb10152.x
Subject(s) - residual , water quality , artificial neural network , calibration , water supply , approximation error , environmental science , correlation coefficient , biomass (ecology) , levenberg–marquardt algorithm , computer science , data mining , statistics , environmental engineering , machine learning , mathematics , algorithm , ecology , oceanography , biology , geology
Residual chlorine and biomass concentrations are two important indicators of water quality in water distribution systems. To assist in the supply of safe drinking water to consumers, several process‐based models have been developed for predicting chlorine residual attenuation and bacterial regrowth in distribution networks. However, calibration of these models requires extensive and accurate data regarding numerous water quality parameters. In this research, an alternative modeling procedure that incorporates an artificial neural network was used to predict temporal chlorine residual and biomass concentrations at different nodes for five water distribution systems. The authors considered three types of algorithms based on feed‐forward neural networks: resilient back propagation, Levenberg‐Marquardt, and general regression. The models developed were tested for unseen data, and comparisons were made on the basis of the mean absolute error and coefficient of correlation.