Research On Functional Area Identification Of Open-pit Mine Based On Graph Embedding

With the development of society,China has put forward higher requirements for intelligent mining of coal resources.In order to promote the intelligent construction of open-pit mines and improve mining efficiency,various open-pit mines have successively launched online scheduling management platforms.In these platforms,correctly identifying the functional areas of open-pit mines can quickly and reasonably schedule trucks,thereby improving the work efficiency of mining area trucks.Therefore,research on the identification of functional areas in open-pit mines is of great significance for improving mining efficiency in mining areas.This thesis focuses on the identification of functional areas in open-pit mining areas.The main research contents are as follows:(1)Open-pit Mine Functional Area Identification Based on Deep Region EmbeddingExtracting hidden features from trajectory data is one of the long-term challenges in functional area research.In open-pit mining areas,efficient identification of functional areas is crucial for the rational planning and management of open-pit mining scheduling systems.However,due to the complexity and variability of functional areas in open-pit mines,truck trajectory data is relatively sparse,and existing methods for identifying functional areas are not suitable for open-pit mining environments.To address this issue,this thesis proposes an Open-pit mine functional area identification method based on Deep Region Embedding(ODRE).This method first divides the open pit mine into uniform grids;Secondly,construct a trajectory grid using the Global Positioning System(GPS)trajectory data of open-pit mining trucks;Furthermore,in order to effectively obtain the relationships in grid space and the local and global structures of trajectory grid graphs,the structured deep network embedding model SDNE(Structural Deep Network Embedding)is used to learn the embedding representation of grid nodes.It can effectively capture local and global information of the network by utilizing first-order and second-order similarity,achieving grid spatial region embedding;Then,the grid temporal features are obtained through bidirectional LSTM(Long Short Term Memory)to represent the temporal variation characteristics of the open-pit mining area;Finally,the embedded representation of the obtained grid nodes and grid sequence features are fused based on attention mechanism to solve the problem of uneven distribution of different feature weights,and the final feature fusion results are trained for identification.The experiment conducted on the open-pit mine dataset shows that the identification accuracy of this method is 6.47% higher than that of UFCG.(2)Open-pit Mine Functional Area Identification Based on Temporal Dynamic Region EmbeddingWith the change of resources and environment in the open pit mining area,the original functional area will also be adjusted accordingly,which will lead to the change of interactive information between related areas.In order to capture the changes in regional interaction features in this temporal dimension,this thesis proposes an Openpit mine functional area identification method based on Temporal Dynamic Region Embedding(OTDRE).This method first extracts feature information from the grid after evenly dividing the area,constructs a dynamic time series diagram structure of the open-pit mine at different time intervals,and achieves spatiotemporal relationship modeling of the functional areas of the open-pit mine;Furthermore,implement a dynamic learning network model that combines dynamic graph embedding and gated loop units,input dynamic time series graph structure to train the model,and achieve potential feature output of nonlinear spatiotemporal relationships in the open-pit mining functional area;Then,establish a multi feature time series for a single grid region,and obtain grid temporal features through bidirectional LSTM.Then,for the feature fusion of dynamic time series,adaptive spatiotemporal feature fusion is used,and the fusion results are finally trained for identification.The experiment conducted on the open-pit mine dataset shows that the identification accuracy of this method is 3.85% higher than that of ODRE.Prove that OTDRE can better capture temporal changes in frequently changing functional areas.(3)Prototype and Application of Functional Area Identification in Open-pit MineThis thesis develops a prototype system based on the method research in Chapter2 and Chapter 3,introduces the overall structure design and detailed functional module design of the system,and presents a visual interface for identifying results,providing support for the intelligent scheduling platform of open-pit mines.The prototype system is a systematic integration and display of the research on the identification of functional areas in open-pit mines based on graph embedding.Based on the actual production situation of open-pit mines,a functional area identification system for open-pit mines has been designed and built using existing technologies,involving both system development and deep learning.This thesis implements the collection,transmission,storage,and preprocessing modules of the open-pit mine truck trajectory dataset,as well as the interactive training and interface display of the model.Finally,the model is applied to the open-pit mine scheduling platform to effectively enhance the robustness of the open-pit mine production scheduling platform,thus improving the production efficiency of the open-pit mine.In this thesis,there are 43 figures,7 tables and 101 references.