Study Of Horizontal Alignment Indicator Recommended Method Based On Machine Learning

The optimization of the horizontal alignment has been developed for many years.Since the curve indicators are often determined mechanically,the results of the optimization are difficult to meet the goals of smooth lines and balanced design indicators.Horizontal alignment design software from the initial auxiliary design software to the current BIM design software is limited to the thinking of machine-assisted human drawing,without human-computer interaction.Based on design example data,the thesis builds a horizontal alignment indicators recommendation model with the help of machine learning algorithms,and completes the development of horizontal alignment indicators recommendation software on Civil3 D.The thesis collected a large number of expressway design cases under the karst landscape in Guangxi.Analyze the statistical law of indicators from linear elements of straight line,circular curve,and transition curve,including the intersection spacing,straight line length,decflection angle,radius of circular curve,transition curve length,and flat curve length.The relationship between the indicators is analyzed,including the declination and the radius of the circular curve,the interval between the intersection points,the length of the transition curve and the length of the circular curve,and the value of transition curve A and the radius of the circular curve.The statistical law of the ratio of the declination angle and the radius of the circular curve in the S-shaped curve and the reverse curve is analyzed.The thesis analyzes the recommended model of the linear indicator of the route plane,and proposes the CBR recommendation method suitable for the single intersection design and the double intersection design process,the geometric of curve combination recommendation model,the basic curve indicator recommendation model,the S-curve indicator recommendation model and the reverse curve indicator Recommended model.In the geometric of curve combination prediction task,two strategies,deterministic recommendation and non-deterministic recommendation,are proposed,and a linear combination judgment model of the recommended linear combination type and the set probability of the predicted linear combination is constructed using a machine learning algorithm.In the basic curve index prediction task,two strategies of recommended radius range and recommended radius value are proposed,and a machine learning algorithm is used to complete the construction of a radius value recommendation model.In the S-curve and reverse curve index recommendation tasks,a machine learning algorithm and two linear constraints are used to construct an S-curve and reverse curve index recommendation model.The thesis divides the planar linear index recommendation system into three modules based on Civil3 D to complete the development based on the requirements of planar linear data analysis,machine learning model construction and planar linear index recommendation.The planar linear index recommendation system includes a data analysis module that provides data statistical analysis and correlation analysis functions,a machine learning model construction module that provides the index recommendation model construction function,and a planar linear index recommendation module that provides planar assisted design.The thesis studies the method of recommending plane linear index based on machine learning,and develops the software of plane linear index recommendation,which can significantly improve the degree of information support in the interactive design process and lay the foundation for the optimization of practical design.