The Research And Implementation Of Anomalous Trajectory Detection System Based On Rnn

With the progress of Positioning System technology and the popularization of relevent equipment,increasing amounts of location data in coordinate series form have been recorded.One of the most common is urban public transportation data,such as bus trajectory,taxi trajectory or pedestrain travel records.These location data are all time series.At the same time,due to the progress in Statistical Machine Learning techonology,using the Recurrent Neural Network(RNN)to construct a anomalous trajectory detection model will caputrue more worthy information.Anomalous trajectory is modeled by taxi coordinate data,so anomalous trajectory detection is a time series data analysis question.Not only constructed by Recurrent Neural Network,this model but also combined with parallel computing technology based on Compute Unified Device Architecture(CUDA)to support the online learning.In addition,this paper uses real taxi trajectory data sets for experiments,and achieved the best results.This article mainly includes the following research:(1)The research and implementation of the anomalous trajectory detection model based on Recurrent Neural Network combined with spatial-temporal law.This paper will structure the coordinate point data based on the global positioning system and transform it into a sequence of discrete trajectory points.This paper introduces a specific Recurrent Neural Network model,and uses time-series data embedding to enhance the ability of represents learning.The model combines the Attention Mechanism to enhance the weight of the point where the trajectory changes drastically and improve the representation ability of the model.The study of the temporal and spatial law of the trajectory enhances the performance of the model,and achieves the state-of-the-art performance on the real data set.(2)Hierarchical clustering algorithm for trajectory updating based on parallel computing using Graphics Processing Unit(GPU):The traditional hierarchical clustering algorithm for trajectory updating has problems of high computational time complexity.Since the road network topology will keep changing in real scenario,the model requires abilities such as online learning and rapid iteration.The parallel computing hierarchical clustering algorithm for trajectory updating based on GPU can solve the problem of slow trajectory similarity calculation.This model has the ability for online learning.Quickly iterate ability can adapt to complex and changing urban traffic conditions and are more in line with industrial applications.(3)Visualized anomalous trajectory detection system:This paper builds a visual anomalous trajectory detection system based on the proposed model and trajectory updating algorithm.Since the trajectory data is mainly composed of sequential coordinate sequences,it has the problems of small storage capacity of a single data instance,high proportion of data transmission delay,and high similarity calculation density.Based on the above problems,this article uses Leaflet’s GIS map service to support the construction of a B/S visualization system.The server needs to support the CUDA framework,Python and C++development environment.