Equivalent Roughness in Nature Channel

• Main Authors: Ching-Chun Huang, 黃靖純
• Other Authors: Yin-Sung Hsu
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/yu8xnw

碩士 === 逢甲大學 === 水利工程與資源保育研究所 === 98 === Manning’s coefficient (n-value) is needed in hydraulic or hydrologic analysis that serves as the basis of river planning and management. In order to obtain proper n-values, back-calculation from Manning’s formula is generally adapted. However, hydraulic parameters that are needed in the formula are not easy to get without actual field survey. This study investigates if the approach developed by Cowan (1956), which detects the n-values by summing contributing factors and the value of each contributing factor can be determined according to field descriptions instead of actual surveys, can alternately be used to estimate the n-values in natural channels. This approach applied the back-propagation neural network (BPN) algorithm on Manning’s coefficients and Cowan’s roughness factors determined from natural streams in Illinois. Correlation of each roughness factor with Manning’s coefficient is determined through stepwise regression and factors that have high correlations are grouped as possible inputs. Among different groupings, two categories are selected based on judgment and 15 models have been tested. At present, the best results are derived from model No. 8 (ANN8), with 9 input factors: surfacial material (n0), surface irregularity (n1), variation in shape and size of the channel cross section (n2), obstruction (n3), adjustment for vegetation (n4), correction factor for meandering channels (m5), discharge (Q), wetted perimeter (P), and velocity (V). The algorithm employs 3 neurons in 1 hidden layer, learning rate as 0.01, output layer transfer function is linear, hidden layer transfer function is tangent sigmoid function. With the 9 input factors this algorithm gives the best estimate of Manning’s n value in natural channels.