Comparison of Levenberg Marquardt and Conjugate Gradient Descent Optimization Methods for Simulation of Streamflow Using Artificial Neural Network

Thabo Michael Bafitlhile, Zhijia Li, Qiaoling Li

College of Hydrology and Water Resources, Hohai University, Nanjing 210098, P. R. China.

Abstract: Prediction of absolute extreme flood peak discharge is a crucial research topic for hydrologists because it is essential in developing the best management practices, addressing water-related issues like flood warning, mitigation schemes planning, management and operation of water resources development projects, etc. The primary purpose of this study was to develop Artificial Neural Network Model (ANN) that can accurately predict Changhua streamflow using hourly data for flood events that occurred between 07/04/1998 to 16/04/2010. Zhejiang province is one of the areas in eastern of China that is prone to severe weather, including heavy rain, thunderstorms, and hail. Since 2011 Zhejiang province has continuously been hit by torrential rain which has left many deaths, loss of property and direct economic loss. Therefore, since Qingshandian reservoir function as a power generator and as flood control system, prediction of the downstream flow of Changhua River is vital for improving the management of the reservoir. Rainfall data from seven stations were used as inputs to the ANN model, and streamflow data were used as the desired outputs of the ANN model. ANN is one of the artificial intelligence method attempting to copy the human brain functioning. It acquires knowledge through a learning process that involves the shifting of connection weight and changing bias parameters to determine the optimal network. Levenberg Marquardt Algorithm (LMA) and Conjugate Gradient Descent (CGD) optimization methods were used to train ANN. The performance of the two algorithms was measured using Residual Standard Error (RSE), R squared, Nash–Sutcliffe Efficiency (NSE) and Pearson’s Product Method (PPM). The overall results show that CGD method is the best method for simulation of Changhua streamflow as compared to LMA.

Keywords: Artificial neural network, Conjugate Gradient Descent, Levenberg Marquardt, Streamflow simulation.

Pages: 217 – 237 | Full PDF Paper