Activities Identification For Stay Area Of Open-pit Mines

With the rapid development and widespread application of global positioning system,people can track the trajectory of various moving objects.At the same time,with the rise of intelligent open-pit mine,GPS devices have been installed on transportation equipment in open-pit mine.As a result,a large amount of GPS trajectory data has been accumulated which provides application potential for mining transportation vehicle travel activities.This thesis focuses on the activity recognition of stay areas in open-pit mine,and conducts research around stay area detection and activity recognition methods.The main contents are as follows:(1)Multi-granularity stay area detection based on area information of open-pit mineDue to the complex and variable environment in open-pit mine,GPS trajectory data may have problems such as missing data of trajectory sampling points and unstable sampling rate.At the same time,traditional stay area recognition methods highly rely on thresholds.In open-pit mines,a fixed threshold may not be able to identify stops generated by different activities.Therefore,this thesis proposes a Multi-Granularity Stay Area detection method based on area information of open-pit mine(MGSA).Therefore,this thesis first analyzes the spatiotemporal density of trajectories,segments the trajectories smoothly,and then calculates the density function of each segment to filter out noise points and trajectories in motion.Secondly,a multi-granularity stay area recognition method is used.Based on the identified stop trajectories,different granularities are used for hierarchical clustering of stop trajectories in different regions in the mine,and finally,stay areas are obtained.This method fully considers the operational characteristics of trucks in open-pit mines,and uses different granularities to recognize stay areas of trajectory points,which can adapt to the situation of the mine and improve the accuracy of stay area recognition.The results show that the recognition accuracy reached 81.23%,which is a 6.6% improvement compared to the baseline.(2)Activity identification of truck stay area for open-pit mine based on heterogeneous graph embedding and recurrent neural networkBase on the research of stay areas,analyzing the data and features of stay areas is a further task in trajectory data mining.In open-pit mine,stay areas are an important part of truck transportation tasks,and recognizing activities in stay areas is of great significance for supervising and improving truck operations in open-pit mine.Therefore,this thesis proposes an Activity Identification of truck stay area for open-pit mine based on Heterogeneous Graph Embedding and recurrent neural network(AIHGE).Firstly,starting from the stay areas themselves,key features are selected and calculated for training.Then,by making full use of information data in open-pit mines,a heterogeneous feature joint graph is constructed for stay areas,area information,vehicle information,etc.,enriching the data features.Next,an efficient batched heterogeneous graph sampling algorithm is used,and HGT(Heterogeneous graph Transformer)is used to embed the heterogeneous graph,obtaining feature vectors for stay areas.Meanwhile,considering the time dependence between stay areas in open-pit mines,a Bi LSTM neural network with time interval awareness mechanism(T-Bi LSTM)is used to capture time interval features.Through experimental analysis,the best recognition accuracy of this method reached 80.73%,which is a 6.05% improvement compared to the baseline.(3)Design and application of prototype system for activities identification of open-pit trucks stay areaPrototype and System Based on the research in Chapters 2 and 3,a prototype system was developed in this thesis.Firstly,the four modules of the prototype system were introduced,including data collection module,data transmission and storage module,basic function module,and production application module.Secondly,the functions of stay area detection and stay area activity recognition were introduced in detail,including data loading,parameter setting,and model selection.Finally,the prototype system can visualize the recognition results,providing an intuitive and convenient way to view the recognition results.This thesis has 50 figures,4 tables and 123 references.