Amirkabir University of Technology Amirkabir Journal of Civil Engineering 2588-297X 49 2 2017 07 23 Performance Improvement of Biological BOD in Rivers based on De-noising Comparison Wavelet-ANN Conjunction, GP, ANN and MLR Methods (Case Study:Karaj Dam Outlet Station) Performance Improvement of Biological BOD in Rivers based on De-noising Comparison Wavelet-ANN Conjunction, GP, ANN and MLR Methods (Case Study:Karaj Dam Outlet Station) 273 284 710 10.22060/ceej.2016.710 FA Taher Rajaee Civil Engineering Department, University of Qom, Qom, Iran Hamideh Jafari Civil Engineering Department, University of Qom, Qom, Iran Roghaye Rahimi Civil Engineering Department, University of Qom, Qom, Iran Journal Article 2014 05 27 This study considered artificial neural network (ANN), multi-linear regression (MLR), Genetic <br />Programming (GP) and wavelet analysis and ANN combination (WANN), models for monthly water <br />biological oxygen demand (BOD) in station Karaj Dam outlet and investigates the effects of data <br />preprocessing on model performance using discrete wavelet. For this purpose, In the first proposed <br />model, observed time series of BOD were decomposed into several subtime series at different scales by <br />discrete wavelet transform. Then these subtime series were imposed as inputs to the ANN method. In <br />the second proposed model, observed time series of BOD were decomposed at ten scales by wavelet <br />analysis. Then, total effective time series BOD were imposed as inputs to the neural network model for <br />prediction of BOD in one month ahead. Results showed that the wavelet neural network models <br />performance was better in prediction rather than the neural network and multilinear regression <br />models. The wavelet analysis model produced reasonable predictions for the extreme values. This <br />model dropped the mean absolute percentage error for the MLR, GP, ANN and the first hybrid <br />models from 1.87, 0.91, 0.65 and 0.46 respectively, to 0.44 and increased the Nash-Sutcliffe model <br />efficiency coefficient from 0.23, 0.53, 0.73 and 0.81 to 0.83. This study considered artificial neural network (ANN), multi-linear regression (MLR), Genetic <br />Programming (GP) and wavelet analysis and ANN combination (WANN), models for monthly water <br />biological oxygen demand (BOD) in station Karaj Dam outlet and investigates the effects of data <br />preprocessing on model performance using discrete wavelet. For this purpose, In the first proposed <br />model, observed time series of BOD were decomposed into several subtime series at different scales by <br />discrete wavelet transform. Then these subtime series were imposed as inputs to the ANN method. In <br />the second proposed model, observed time series of BOD were decomposed at ten scales by wavelet <br />analysis. Then, total effective time series BOD were imposed as inputs to the neural network model for <br />prediction of BOD in one month ahead. Results showed that the wavelet neural network models <br />performance was better in prediction rather than the neural network and multilinear regression <br />models. The wavelet analysis model produced reasonable predictions for the extreme values. This <br />model dropped the mean absolute percentage error for the MLR, GP, ANN and the first hybrid <br />models from 1.87, 0.91, 0.65 and 0.46 respectively, to 0.44 and increased the Nash-Sutcliffe model <br />efficiency coefficient from 0.23, 0.53, 0.73 and 0.81 to 0.83. https://ceej.aut.ac.ir/article_710_e70611883d2760c8bbafb4acb29e3446.pdf