Research On Trajectory Mining Algorithms Based On Deep Learning

With the popularity of intelligent mobile terminals and the development of location-aware technology,more and more mobile object trajectory data are captured.Trajectory data are typical time-series data,containing rich spatio-temporal information,such as travel path,time and location.These data provide valuable data support for many fields such as urban planning and environmental protection,and also drive the rapid development of trajectory mining algorithms.Maritime navigation trajectory is a trajectory data with significant application value,and the mining of maritime navigation trajectory data has strong practical value and important theoretical significance.Due to the existence of navigation uncertainty and acquisition bias,the mining of maritime trajectory data is more challenging.The efficiency and safety of maritime traffic can be improved by mining the navigational laws of maritime trajectories.In addition,maritime trajectory data can be used for marine resource development,maritime disaster warning and marine environmental protection,which have wide application prospects.In this paper,we explore the background,current situation,main contents and challenges of trajectory mining,and introduce in detail the deep learning-based trajectory mining algorithms,especially in the application of maritime trajectory mining.We focus on the distribution pattern of fishery resources in space and time,and provide effective support for fisheries supervision by studying three aspects of fishing vessel AIS(Automatic Identification System for Ships)trajectory data,including trajectory segmentation,operation point extraction and operation hotspot prediction.The main technical works of this paper are as follows:First,this paper proposes a segmentation algorithm for fishing vessel trajectories,aiming to analyze the behavior patterns and fishing strategies of fishing vessels.The algorithm maps the trajectory data into a complex network and clusters them using a community detection algorithm to obtain segmentation points,achieve trajectory segmentation,and ensure the continuity of the same trajectory segment in time and space.Compared with existing algorithms,the algorithm is more applicable to fishing vessel AIS trajectory data and can better deal with equipment failure and poor signal situations.In addition,the algorithm has better applicability and robustness,and can cope with the complex problems in practical situations,which has practical value.Then,this paper proposes a fishing vessel operation point extraction algorithm based on segmentation algorithm,which aims to extract the point information of fishing vessel operation from the AIS trajectory data of fishing vessels in order to monitor and regulate the fishing behavior.The algorithm integrates the problem of vessel motion patterns and can extract fishing vessel operation points more accurately.Specifically,the algorithm uses a voyage extraction algorithm to filter port mooring points,uses complex networks and community detection algorithms to map fishing vessel trajectories and divide communities,and finally uses the DBSCAN algorithm to density cluster communities to obtain real fishing vessel fishing operation points.With this algorithm,the fishing vessel operation points can be extracted more accurately and provide powerful support for fisheries resource management and monitoring.Finally,this paper proposes a spatio-temporal Transformer fishing vessel operation hotspot prediction model based on honeycomb grid,which aims to effectively deploy fishery resources and safeguard the marine ecological environment.The model is trained on the fishing vessel operation data obtained by the job point extraction algorithm proposed in this article,resulting in more accurate prediction of future operation hotspots.Unlike most existing hotspot prediction algorithms that use a square grid to divide the study area,the model uses a honeycomb grid to divide the study area,which can adapt to the directional freedom of ocean motion patterns.Meanwhile,the model takes into account the importance of marine hotspots subject to local spatial and temporal dependence,and therefore uses local foveal mapping as the data sample unit and reduces the feature dimensionality.Finally,in order to capture the spatio-temporal dependence of fishing vessel operation hotspots,this model uses the spatio-temporal Transformer model to accurately predict the future fishing vessel operation hotspots.With this model,the operational hotspots of future fishing vessels can be more accurately predicted,thus realizing the effective deployment of fishery resources and the protection of marine ecological environment.