Research On Clustering Analysis Of Port Operation Vehicle Trajectory And Abnormality Detection Method

The scale and throughput of the ports along the middle and upper reaches of the Yangtze River are relatively small,and the loading and unloading of cargoes and transshipment mainly rely on inland vehicles to complete.After a long-term summary of the operation situation by port dispatchers,the inland river ports have relatively complex roads,diverse yard environments,diverse operation types and 24-hour nonstop work,as well as some human factors that lead to abnormal operations in the process of cargo transfer by operation vehicles,which are inseparable from the real intention of drivers.How to detect the abnormal behavior of operating vehicles according to their historical trajectories and provide a guarantee for safe cargo transfer is an indispensable part of building a smart inland port.To address the above problems,this thesis determines whether the operating process of vehicles is correct by detecting abnormalities in the trajectory data of operating vehicles in inland waterway ports.The research contents of this thesis are as follows.(1)According to the characteristics of trajectory data of inland river ports,a trajectory data cleaning algorithm of the safety zone is designed.As the vehiclemounted terminal of the port mainly collects the trajectory of vehicles for civilian GPS,its trajectory accuracy is low and there is low-quality repetitive data,and the roads in the port are relatively close and the trajectory distribution is uneven,so data cleansing is needed.In this thesis,we first extract the road model in the port based on the historical trajectory data,get the representative points that can constitute the road,and get the smooth road frame by smoothing processing and connectivity processing;based on the road frame points,we propose a trajectory noise identification algorithm based on the safety zone,to separate the local noise points from the normal trajectory points.(2)To describe the trajectory features more comprehensively,a self-encoder trajectory feature fusion model is constructed.Due to the limitation of port operation space,the trajectory features of operating vehicles often have great spatial similarity,and it is difficult to find abnormalities.Therefore,to weaken the spatial attributes of operating vehicle trajectories and discover their potential depth sub-hidden features,a self-encoder-based trajectory feature fusion model is proposed to comprehensively describe the trajectory features of operating vehicles in operation.First,the original trajectory dataset is divided into a collection of trajectory segments based on open perspectives,and the trajectory segments are used as the basic unit of study.Next,the surface features of the operational vehicle trajectory segments are extracted,highdimensional features are obtained using the self-encoder,and finally the surface and high-dimensional features are stitched and input to the self-encoder to fuse them into a fused feature sequence.This solves the current problem that the detection of vehicle trajectory anomalies is too dependent on the density and spatio-temporal features,which makes the detection of high-dimensional,composite,and hidden trajectory features difficult.(3)The Optics Clustering Algorithm-Based Trajectment Feature Sequences,OPTICS-FFS,which designed a trajectory fusion characteristic sequence..The purpose of this algorithm is mainly used to detect whether the trajectory of the operating vehicle is abnormal,to achieve the purpose of supervising cargo safety.This algorithm is based on OPTICS(Ordering Points to Identify the Clustering Structure,OPTICS)algorithm,and the cosine similarity between fused feature sequences is used as the measure of distance,and the distance metric matrix between fused feature sequences is calculated by the clustering result set derived from the OPTICS-FFS algorithm to measure the degree of similarity of each fused feature sequence in the same category,and it is judged as an abnormal trajectory segment if it exceeds the threshold(fused feature sequence is the trajectory segment(The fusion feature sequence is the feature representation of the trajectory segment);according to whether the length of the abnormal trajectory segment in the whole trip exceeds the threshold value to determine whether it is an abnormal trajectory.(4)Experimental verification and analysis of the relevant models and algorithms in this thesis.The results show that the trajectory data cleaning algorithm based on safety zone can effectively remove low-quality and duplicate trajectory data and compress the size of trajectory data set;the trajectory feature fusion model based on self-encoder can effectively fade the spatial attributes of trajectory in extracting the deep trajectory features of operation vehicles.Based on the above methods and models,OPTICS-FFS algorithm and other trajectory anomaly detection algorithms(LCS,ATDRNN,i Forest.)under the comparison,OPTICS-FFS has better performance in detecting trajectory anomalies of operating vehicles.The research work in this thesis can effectively detect the abnormal trajectories of port operation vehicles,which is of great practical significance and value for improving port cargo safety and building intelligent modern ports.