AIS Ship Trajectory Clustering Based On Convolutional Auto-encoder

With the development of big data and artificial intelligence in recent years,how to use large-scale data for data mining is a hot research issue.In recent years,with the deepening of economic globalization cooperation,90% of world trade is inseparable from maritime transport.This has led to a rapid increase in maritime trade and transportation activities.The increasingly complex maritime transportation environment makes it difficult to apply traditional manual monitoring of maritime activities to this task.Therefore,it is urgent to use artificial intelligence technology to assist mariners in monitoring and managing maritime vessels.The main research directions of navigation safety include: ship trajectory anomaly detection,ship path planning,ship position prediction and ship collision detection.Trajectory clustering is the basis and the key to solve the above problems.The motion of ship trajectory model for mining the traditional trajectory clustering method usually need,depending on the type of data volume and trajectory calculation complexity,noise and other factors,to select the measurement method of the space-time path.However,the choice of optimal similarity measure formula requires a great deal of prior knowledge and extensive experiments,which can lead to waste of computing resources and time.Therefore,this paper proposes a ship AIS trajectory motion pattern extraction algorithm based on convolutional autoencoder,which does not need the traditional spacetime trajectory similarity measurement method.The multi-feature fusion self-encoder proposed in this paper is used to extract the low-dimensional potential representation of ship track's position feature,velocity feature and heading feature respectively and carry out feature fusion.Then,the ship track is analyzed by clustering.It solves the problems of the traditional trajectory similarity measurement methods(i.e.,they are vulnerable to noise and require a lot of time and computing resources).The experimental results show that the algorithm has good clustering performance while maintaining the main motion characteristics of ship tracks.The time series relation of track points can be maintained and the deviation of distance calculation of track points can be reduced.Finally,the trained algorithm model is deployed in the Web platform,and the state-of-the-art front and rear end technologies are used to integrate the platform to realize the ship trajectory clustering system based on convolutional autoencoder,which can provide users with API interface,fully meet their needs and facilitate their use.