Research On Deep Learning Based On Point Cloud Data Of Metro For Line And Slope Fine-Tuning

The deviation information between the completed tunnel and the originally designed tunnel is very important for the safety adjustment of metro lines.However,there is no clear mathematical formula that can be used to accurately describe and measure the deviation.At present,the mainstream approach is to measure the invasion value of each section with the same interval manually and then sum up these values to get the deviation.This method has several disadvantages,such as time-consuming,large error and high cost.To solve these problems,this paper proposes two novel deviation representation methods based on the deep neural network,which can learn the internal relationship between the parameters of the designed tunnel and the invasion values based on the point cloud data,and then predict the parameters that can make the sum of the invasion values minimum.These parameters can be used to assist the safety adjustment of metro lines.The first solution applies the traditional neural network learning model to the optimization of subway line adjustment and slope adjustment.A one-dimensional vector is used to represent the horizontal and vertical parameters of the design line.This vector can determine a design line and then combine the point cloud big data to calculate the invasion limit of this design line.Perturbation of any component in the one-dimensional vector according to the requirements of the design specification can form multiple sets of data sets containing the design line parameters of the threshold.Use this data set to train a neural network so that the network learns the functional relationship between the design line and the tunnel deviation to obtain the optimal line and slope adjustment plan.Experiments show that the final result of scheme one is obviously better than traditional manual adjustment,but the neural network takes a lot of time during the training process,and it is impossible to achieve real-time feedback.Therefore,this paper further proposes the method of randomly weighted neural network.Optimize the subway line and slope adjustment plan.The random weighted neural network uses a non-iterative method of updating the weights of the pseudo-inverse,which greatly reduces the training time of the network and provides a solution for realtime optimization of subway lines.Finally,the experimental results on a data set collected from a real subway project show that the proposed method can quickly obtain the appropriate adjustment scheme of the lines and slopes with only a small amount of computer memory resources.