Automatic Identification of Highway Horizontal Curves Based on BP Neural Network

Automatic alignment design is always one pursuing goal of highway geometric designers with the rapid development of modern technologies such as intellection, digitalization and automatization. Because of the fact that artificial neural network is widely used in the field of pattern recognition, an approach to fit some control points and then to determine the appropriate type of horizontal curves based on BP neural network is presented. Firstly, the curvature charts which are more intuitive and simpler compared with equations are selected to describe the horizontal curves. Then the fitting and standard curvature curves are defined. Secondly, the eigenvectors and identification vectors are structured by calculating characteristic coefficients of curvature charts. Thirdly, a three-layer BP identification model which takes the eigenvectors and identification vectors of curvature charts as input and output, respectively, is presented. Finally, two cases are given. The corresponding alignment of horizontal curve can be determined by the model which is trained by the eigenvectors and identification vectors of standard curvature chart. The result shows that the BP identification model is able to determine the type of curve.