Research On Road Network Generation Method Of Open-pit Mine Based On Graph Network

With the rapid development of intelligent construction of open-pit mine,the research on truck dispatching systems and unmanned equipment in open-pit mine has attracted much attention.However,the progress of these intelligent researches all depend on the construction of road network models.The road network model reflects the geographic location information and topological relationship of roads in open-pit mine,and is the basis for promoting the rapid development of truck dispatching systems and unmanned equipment in open-pit mine.Therefore,generating real-time and accurate open-pit mine road network is very important for the intelligent construction of open-pit mine.The purpose of this thesis is to study the road network generation method suitable for the geographical environment of open-pit mine.For this reason,based on the massive GPS(Global Positioning System)trajectory data generated by open-pit mine trucks,this thesis discusses how to generate an open-pit mine road network that meets the actual situation.The main research contents are as follows:(1)Road network generation method of open-pit mine based on dual graph convolution networkExisting methods often only use single trajectory features.In addition,the constructed deep network can only utilize local contextual information,lacking the ability to capture global contextual information.To address these issues,this thesis proposes a Road network Generation method of open-pit mine based on Dual Graph convolution network(RGDG).Firstly,this method constructs multiple sets of trajectory features,and performs feature fusion to obtain trajectory fusion features.Then,a Road Centerline Prediction Network(RCP-Net)is designed to predict the open-pit road centerline,which extracts the deep features of the trajectory fusion features through residual encoders and models the spatial and channel global contextual information of the deep features using a dual graph convolution network,outputting deep contextual features.Next,the deep contextual features are input into the open-pit road centerline decoder to generate the road centerline probability map.Finally,the vectorized openpit mine road network is constructed using the road centerline probability map and its topological structure is refined.The experimental results show that the RGDG method has improved by 3.2% in terms of road network geometry compared to the best performing baseline method,and by 4.4% in terms of road network topology.(2)Road network generation method of open-pit mine based on ensemble RCPNet quantifying uncertaintyIt is easy to overfit when using a single RCP-Net for predicting the road centerline of open-pit mine.Additionally,due to the complexity of the RCP-Net deep model,its predictions often have uncertainty,leading to incorrect predictions for narrow open-pit mine roads and sparse trajectory roads.To address these issues,this thesis proposes a Road network Generation method of open-pit mine based on Ensemble RCP-Net quantifying Uncertainty(RGERU).Firstly,the model diversity of the ensemble RCPNet is enhanced by using a random layer sampling method to reduce the risk of overfitting.Secondly,the posterior distribution of the weight parameters of the ensemble RCP-Net model is estimated using the approximate Bayesian method.Then,posterior distribution samples are sampled to statistically calculate the prediction variance to obtain an uncertainty probability map,which quantifies the uncertainty of the ensemble RCP-Net and reflects the degree of uncertainty in predicting the road centerline of open-pit mine.Finally,the uncertainty probability map is used to construct an uncertainty joint loss function,which enables the ensemble RCP-Net to overcome the prediction limitations of uncertain roads and improve the accuracy of predicting the road centerline of open-pit mine.Experimental results show that the open-pit road network generated using the RGERU method has improved geometric properties by3.6% and topological properties by 2.9% compared to the RGDG method.(3)Design and application of open-pit road network generation systemBased on the research on the road network generation methods of open-pit mine in chapters two and three,a system for generating open-pit mine road network was designed.This system includes the following functions: GPS trajectory data preprocessing,truck trajectory feature extraction,model sample creation,open-pit mine road network generation model training,and open-pit mine road network generation testing.Ultimately,this system was integrated into the open-pit mine intelligent dispatch platform to provide support for the automatic construction of open-pit mine road network,thereby promoting the development of intelligent open-pit mine construction.This thesis has 43 diagrams,4 tables,and 101 references...