| Depth of the river erosion around the piers is an important basis to determine the depth of the foundation piers. Too much erosion is one of the main reasons for bridges damage coursed by water, so that whether the scouring design is correct or not is directly related to the use of the bridge's safety. Many domestic and foreign scholars and scientific workers have carried a large number of piers scour indoor model tests,field observations and numerical studies. Besides, they put forward to many empirical formulas and semi-empirical semi-theoretical formulas.This paper summarized the achievements in studying pier scour. The factors affecting pier local score are very complex, some of which reveal randomly. The status and effects of various factors are different and some of them are interaction. However, the existing theoretical formula does not reflect the interaction between these factors. Based on the recognizing the complexity of the factors effecting riverbed scour around the piers, this paper try to introduce artificial neural network theory and the fuzzy neural network theory to study the piers'local scour.The basic tenets of artificial neural network, fuzzy theory and fuzzy neural network are mainly introduced. Hypothesis testing method is proposed to determine the factors effecting the pier local scour depth. And BP neural network model and fuzzy neural network model are built to forecast the pier local scour depth. Beside, these two models'predicted results are compared.Compared with some BP algorithms, it's found that LMBP algorithm has lower accuracy in single forecast and higher forecast risk but has higher training efficiency and stable training frequency. While BP adaptive algorithms introducing momentum factor reduces the probability of net concussion and it has higher accuracy and training speed, but training times fluctuate large. It's recommended that using LMBP algorithm to determine the network structure and adaptive algorithms BP to forecast.Fuzzy neural network , which is a fusion of fuzzy theory and neural network technology, greatly broadens the neural network's scope and capacity to process information.Fuzzy neural network built in this paper has higher mapping accuracy. It not only considers the impact of the bridge pier scour local factors indicators in forecasting the local scour depth with varying degrees of fuzziness but also gets the relationship between indicators factors and predicted target. It has clear network structure and every layer has a clear physical meaning. What's more, according to experience, initial value can be set up. All the above are what the neural network can't have.By comparing, it can be found that the prediction accuracy of neural network is closer to that of fuzzy neural network model. But the convergence rate of fuzzy neural network model is faster than BP neural network model's. Through this paper's study, it's known that both artificial neural network and fuzzy neural network model can be used to forecast the pier local scour and offer a kind of new method to calculate the pier local scour depth. |
