Research On AIS Track Fitting Based On The Broad Learning System

At present,the main way of maritime traffic monitoring by the maritime department is still mainly manual monitoring,which is not only time-consuming and laborious,but also lacks pertinence.Especially in some busy ports,manual monitoring alone cannot meet the demand of port security.In order to realize real-time tracking of ship navigation behavior and make effective prediction,this thesis proposes a fitting algorithm based on the BLS(Broad Learning System,BLS),which is used to establish the AIS track fitting model.As an important part of navigation technology,prediction has important practical significance for improving the safety of navigation.In this thesis,ship AIS data is taken as the research object and relevant research is carried out from the perspective of track fitting.The main research work can be summarized as the following four aspects.(1)This thesis introduces the database processing of 72GB AIS raw data,then extracts the AIS data in Dalian Port waters from the database,and performs data cleaning and data conversion pretreatment to determine the research object of this thesis,which laid the data foundation for the subsequent research.Based on the concept of first-order backward difference and differential first-order backward difference,the construction method of time series prediction data is introduced,which lays the foundation for the subsequent track fitting experiment.(2)The broad learning algorithm and its model structure are introduced in detail,and the specific mathematical derivation process of the broad learning algorithm is expounded.The track fitting method based on broad learning is also demonstrated.(3)In the case of an AIS track fitting experiment,the track fitting based on the LM(Linear model,LM)model is complicated,while the BLS model is simple and can obtain the fitted values of longitude and latitude in one time in terms of complexity;in terms of fitting accuracy,the longitude and latitude RMSE values of BLS are both Less than the LM model,the BLS has a higher fitting accuracy;in terms of running time,the BLS model runs faster than the LM model.(4)In order to further improve the fitting accuracy of the first AIS track fitting,a piecewise fitting method based on K-means clustering is proposed.In the track fitting of K-means clustering,comparative experiments are carried out mainly in three aspects:longitude,latitude and latitude and longitude.For each aspect,based on the original data,the first-order backward difference data and the first-order differential backward difference data are obtained by using the time series construction method.Based on LM model and BLS model,experiments are carried out separately.The intra-cluster variance,RMSE and operation time are taken as evaluation indexes,and the following conclusions are drawn.In terms of variance within the cluster,based on the results of the track fitting comparative experiments,the three track fitting experiments showed the same pattern.As the number of clustering increases,the value of the variance based on the original data,the first-order backward differential data,and the differential first-order backward differential data decreases continuously,indicating that the AIS data is clustered after the introduction of the K-means clustering algorithm.The lower degree of dispersion,the higher the tightness,and the better lack of dispersion of the original data.In terms of RMSE,based on the comparative experimental results,the three track fitting experiments show the same rule.With the number of clusters increases,the RMSE values of LM model and BLS model decrease continuously,which indicates that the fitting accuracy of BLS model is higher and higher.At the same time,we can also see that the RMSE values of BLS model are always smaller than LM model,which indicates that the fitting accuracy of BLS model is higher than LM model.In addition,based on the experimental results of the original data,first-order backward difference data and differential first-order backward difference data are compared in two ways.It can be seen that the effect of track fitting based on first-order backward difference data and differential first-order backward difference data is better than that of original data.Among them,the effect of track fitting based on differential first-order backward difference data is the best and the original data fitting accuracy is improved.In terms of running time,the overall law of presentation is that when the number of clusters is 1,the running time of the LM model is longer than that of the BLS model.When the number of clusters is greater than 1 or 2,the running time of the BLS model is longer than that of the LM model.Long,from the overall point of view,the running time of the two methods is basically the same,but the BLS model is more stable.In summary,this thesis proposes a trajectory fitting algorithm based on the broad learning algorithm.Compared with the traditional fitting method,the algorithm has a large improvement in fitting accuracy,which lays a foundation for the later establishment of the AIS trajectory prediction mode theoretical basis.At the same time,this provides a new forecasting idea for the maritime department to realize the monitoring of maritime traffic,which has certain application value and has important practical significance for improving the navigation safety of ships.