KR102161147B1 - Apparatus and method for identifying abnormal sailing ship - Google Patents

Abstract

Disclosed are an apparatus and a method for identifying an abnormally sailing ship. The method for identifying the abnormally sailing ship comprises the steps of: collecting track data of a ship sailing in a target area; dividing the target area into a grid area of a preset size; learning the track data collected for each divided grid area and calculating the learning data for a sailing pattern of the ship; classifying and clustering the calculated learning data into a preset number of clusters in accordance with the sailing pattern; generating a probability model for each cluster by using the learning data for each cluster in response to the clustering; and identifying an abnormally sailing ship by using the probability model for each cluster generated for each grid area. According to the present invention, abnormal sailing of a ship can be identified by using the calculated learning data for each divided area.

Description

Apparatus and method for identifying abnormal sailing ship}

The present invention relates to an apparatus and method for identifying abnormal vessels.

In general, the Vessel Traffic Service System (VTS) provides transit services to improve port safety and efficiency of port operation in response to congestion of maritime traffic, increase of dangerous cargo, and potential risk of environmental pollution. It is a system to do. The maritime traffic control system can provide information on the surrounding situation and maritime traffic situation to the ship in a timely manner within the control area, so that the ship can help the navigator in the decision process.

On the other hand, a number of systems have been developed to support controllers performing control tasks using the maritime traffic control system. However, since most of the developed systems operate in a rule-base, it is impossible to determine a situation that violates the defined rules. And, it is impossible to define all maritime traffic conditions.

Accordingly, methods based on machine learning models have been developed to identify the abnormal operation state of a ship. Existing methods have difficulty in generalizing the entire target area into one model. In other words, since there are a wide variety of navigation patterns in the target area, it is difficult to classify or identify them, and when learning is performed based on past track data, a low frequency navigation pattern is likely to be identified as an abnormal flight. In addition, it is often unclear for what reason a vessel identified as operating abnormally is identified as operating abnormally.

Korean Registered Patent Publication No. 10-1947675 (2019.02.07)

The present invention divides the control area in charge of the Vessel Traffic Service System (VTS), learns track data for each divided area, calculates learning data for the operation pattern of the ship, and calculates each divided area. Abnormal operation vessel identification device and method for identifying abnormal operation of the vessel using learning data.

According to an aspect of the present invention, a method for identifying an abnormally operating vessel performed by an abnormally operating vessel identification device is disclosed.

The method for identifying an abnormal vessel according to an embodiment of the present invention includes the steps of collecting track data of a ship operating a target area, dividing the target area into a grid area of a preset size, and the divided grid area. Learning the collected track data to calculate learning data on the navigation pattern of the ship, classifying the calculated learning data into a preset number of clusters according to the navigation pattern, and clustering the calculated learning data, according to the clustering. And generating a probability model for each cluster using the training data for each cluster, and identifying an abnormally operating vessel using the probability model for each cluster generated for each grid region.

The collected track data is the track data and radar received from the Automatic Identification System (AIS) installed in each vessel operating within the control area in charge of the Vessel Traffic Service System (VTS). Includes track data acquired through.

In the step of calculating the training data for the sailing pattern of the ship, unsupervised learning is performed using an autoencoder model, which is a neural network model that generates training data composed of a pair of inputs and inputs, The autoencoder model is trained using only the input data composed of the track data to generalize the navigation pattern of the ship in the target area.

The calculating of the learning data for the navigation pattern of the ship includes: normalizing the collected track data, generating training data by unsupervised learning the normalized track data using the autoencoder model, and the Including the step of generating an encoder that projects the input data into 3D data and a decoder that restores it as the autoencoder model training is completed, the input and output of the autoencoder model The data consists of three-dimensional data of position, course and velocity.

The clustering may include generating 3D learning data by performing encoding to project the generated training data into 3D data using the generated encoder, and clustering the generated 3D training data by k-means It includes the step of clustering using an algorithm, but uses an elbow technique or a silhouette technique to find the optimal number of clusters.

In the step of generating the probability model for each cluster, a probability model of the position value, course value, or velocity value of the vessel for each grid region is generated as a probability density function (PDF), but the probability density function is Gaussian- It is estimated using the Kernel Density Estimation (KDE) method, and the cluster-specific probability model includes a probability density function of a position value, a course value, and a velocity value.

The step of identifying the abnormal vessel may include receiving a track value including a real-time position value, a course value, and a speed value of a ship operating in the target area, wherein the arbitrary ship operates in the target area. Comprising the step of extracting a grid area and comparing the inputted track value with a probability model for each group of the extracted grid area, and determining whether the arbitrary ship is operating abnormally according to the separation degree of the track value, wherein the The step of determining whether an arbitrary vessel is operating abnormally includes comparing the probability model for each cluster with the input position value, course value, and speed value to determine whether the ship is operating abnormally for each position, course, or speed of the ship. Discriminate.

According to another aspect of the present invention, an apparatus for identifying an abnormal vessel is disclosed.

An apparatus for identifying an abnormal sailing vessel according to an embodiment of the present invention includes a memory storing an instruction and a processor executing the instruction, wherein the instruction is the step of collecting track data of a vessel operating a target area, the target Dividing an area into a grid area of a preset size, learning the collected track data for each of the divided grid areas to calculate learning data for the sailing pattern of the ship, and applying the calculated learning data to a navigation pattern. According to this, the steps of classifying and clustering into a preset number of clusters, generating a probability model for each cluster using the training data for each cluster according to the clustering, and using the probability model for each cluster generated for each grid region Performs a method of identifying an abnormally operating vessel including the step of identifying an abnormally operating vessel.

An apparatus and method for identifying abnormal vessels according to an embodiment of the present invention divides a control area in charge of a Vessel Traffic Service System (VTS), and learns track data for each segmented area to provide a vessel operation pattern. It is possible to calculate the learning data for and identify the abnormal operation of the ship by using the calculated learning data for each divided area.

1 is a flowchart showing a method of identifying an abnormally operating vessel performed by an abnormally operating vessel identification apparatus according to an embodiment of the present invention.
2 to 5 are diagrams for explaining a method for identifying an abnormal vessel according to an embodiment of the present invention.
6 is a view schematically illustrating the configuration of an abnormal vessel identification device according to an embodiment of the present invention.

Singular expressions used in the present specification include plural expressions unless the context clearly indicates otherwise. In the present specification, terms such as “consisting of” or “comprising” should not be construed as necessarily including all of the various elements or various steps described in the specification, and some of the elements or some steps It may not be included, or it should be interpreted that it may further include additional elements or steps. In addition, terms such as "... unit" and "module" described in the specification mean units that process at least one function or operation, which may be implemented as hardware or software, or as a combination of hardware and software. .

Hereinafter, various embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart showing a method of identifying an abnormally operating vessel performed by an abnormally operating vessel identification apparatus according to an embodiment of the present invention, and FIGS. 2 to 5 are It is a drawing. Hereinafter, a method of identifying an abnormally operating vessel performed by the abnormally operating vessel identification apparatus according to an embodiment of the present invention will be described with reference to FIGS. 2 to 5.

In step S110, the abnormal vessel identification device collects track data.

For example, the Vessel Traffic Service System (VTS) is acquired through track data or radar received from the Automatic Identification System (AIS) installed on each ship operating within the control area in charge. You can collect and store track data. In addition, the abnormal navigation vessel identification device may collect the track data by receiving the track data collected from each ship from the maritime traffic control system. Track data of ships not equipped with an automatic ship identification device can be collected only through radar.

Here, the track data of the automatic ship identification device is the ship name, ship type, ship specifications (ship length and width), ship location information according to time, ship speed information according to time, ship course information according to time, and loading cargo. Information may be included, and the radar track data may include location information of the target over time, speed information of the target over time, and course information of the target over time.

In step S120, the abnormal navigation vessel identification device divides the target area into a grid area of a preset size. Here, the target area may be a control area in charge of the maritime traffic control system. For example, FIG. 2 shows a target area 10 divided into a grid area 15.

In step S130, the abnormal navigation vessel identification device calculates learning data on the navigation pattern of the vessel by learning the track data collected for each divided grid area.

In other words, since it is difficult to generate training data composed of pairs of input and output due to the characteristics of the track data, an autoencoder model, which is a neural network model that generates training data composed of pairs of inputs and inputs, is used in the abnormal ship identification device Unsupervised learning can be performed by using. Thus, the abnormal sailing vessel identification apparatus can generalize the operation pattern of the vessel in the target area 10 by learning the autoencoder model using only input data composed of track data.

For example, in order to unsupervised learning of track data using an autoencoder model, the apparatus for identifying an abnormal ship may first normalize the collected track data to a real value between 0 and 1. That is, the abnormal vessel identification device classifies the input track data (i.e., ship's location information, course information, and speed information) for each ship, and stores the classified track data at a predetermined period (for example, 10 seconds). Interpolation can be performed and the interpolated track data can be normalized to a real value between 0 and 1 based on the maximum and minimum values.

Subsequently, the abnormal ship identification device may generate training data by unsupervised learning the normalized track data using an autoencoder model. The input and output data of the autoencoder model may be composed of three-dimensional data of position, course, and speed. Accordingly, the autoencoder model may be composed of three input and output layers. The abnormal vessel identification device can perform unsupervised learning until the data of the input layer and the output layer become the same as possible, and an encoder that projects the input data into 3D data as the learning of the autoencoder model is completed ( encoder) and a decoder that reconstructs the same.

In step S140, the abnormal navigation vessel identification device classifies the calculated learning data into a preset number of clusters according to a navigation pattern and clusters them.

For example, the apparatus for identifying an abnormal sailing vessel generates 3D learning data by performing encoding that projects the training data calculated by using the encoder of the learned autoencoder model into 3D data, and then converts the generated 3D training data. It can be classified and clustered according to the operation pattern. At this time, the abnormal ship identification device may use a clustering algorithm such as k-means to cluster the learning data, and may use an elbow technique or a silhouette technique to find the optimal number of clusters.

In step S150, the apparatus for identifying an abnormally operating vessel generates a probability model for each cluster by using the learning data for each cluster.

For example, the abnormal ship identification device generates a probability model of the ship's position value, course value, or speed value for each grid area as a probability density function (PDF: Probability Density Function) as shown in the graph shown in FIG. can do. Here, the probability density function may be estimated using a Gaussian-Kernel Density Estimation (KDE) method.

For example, FIG. 4 shows an example of a probability model 20 of n clusters. Referring to FIG. 4, the probability model for each cluster may include a probability density function of a position value, a course value, and a speed value.

In addition, the probability model for each cluster may be generated by classifying each vessel type, vessel size, and time slot.

In step S150, the apparatus for identifying abnormal vessels identifies abnormal vessels using the probability model for each cluster generated for each grid area.

For example, when the abnormal vessel identification device receives the real-time track value (position value, course value, and speed value) of an arbitrary ship operating in the target area 10, the corresponding ship operates in the target area 10. The grid area is extracted, and the probability model for each group of the extracted grid area is compared with the input track value, and according to the separation of the track value, it is possible to determine whether or not the ship has an abnormal operation. That is, the abnormal navigation vessel identification device may determine whether the vessel has abnormal operation for each of the vessel's position, course, or speed by comparing the probability model for each group with the input vessel position value, course value, and speed value.

For example, FIG. 5 schematically illustrates a probability density function for determining whether an abnormal flight has occurred. Referring to FIG. 5, the normal range of the wake value may be determined by setting the left and right probability values such as 34.1% or 34.1% + 13.6% based on the average value μ in the probability density function. Identification sensitivity can be adjusted through such adjustment of the probability value. So, assuming that 34.1% is set as the right and left probability values as the probability value of the normal range, the input track value 30 of any ship is included in the normal range.

Information on the normal range of such a track value may be input by a user when generating a probability model for each cluster and included in the probability model for each cluster.

In addition, the probability model for each group may include user information such as restricted zone information, course limit information, minimum and maximum speed information, etc. to determine the normal range of the position value, the course value, and the speed value.

6 is a diagram schematically illustrating the configuration of an abnormal sailing vessel identification apparatus according to an embodiment of the present invention.

Referring to FIG. 6, an apparatus for identifying an abnormal vessel according to an embodiment of the present invention includes a processor 610, a memory 620, a communication unit 630, and an interface unit 640.

The processor 610 may be a CPU or semiconductor device that executes processing instructions stored in the memory 620.

The memory 620 may include various types of volatile or nonvolatile storage media. For example, the memory 620 may include ROM, RAM, or the like.

For example, the memory 620 may store instructions for performing the method of identifying an abnormally sailing vessel according to an embodiment of the present invention.

The communication unit 630 is a means for transmitting and receiving data with other devices through a communication network.

The interface unit 640 may include a network interface and a user interface for accessing a network.

Meanwhile, the constituent elements of the above-described embodiment can be easily grasped from a process point of view. That is, each component can be identified as a respective process. In addition, the process of the above-described embodiment can be easily grasped from the viewpoint of the components of the device.

In addition, the above-described technical contents may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiments, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of the program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

The above-described embodiments of the present invention are disclosed for the purpose of illustration, and those skilled in the art who have ordinary knowledge of the present invention will be able to make various modifications, changes, and additions within the spirit and scope of the present invention, and such modifications, changes and additions It should be seen as belonging to the following claims.

610: processor
620: memory
630: Ministry of Communications
640: interface unit

Claims (8)

In the abnormal vessel identification method performed by the abnormal vessel identification device,
Collecting track data of a ship operating in a target area;
Dividing the target area into grid areas of a preset size;
Calculating learning data on the navigation pattern of the ship by learning the collected track data for each of the divided grid areas;
Classifying and clustering the calculated training data into clusters of an optimal number of clusters according to a navigation pattern;
Generating a cluster-specific probability model for each vessel type, vessel size, and time slot using the clustered learning data according to the clustering;
Receiving a track value including a real-time position value, a course value, and a speed value of a ship operating in the target area;
Extracting a grid area in which the arbitrary ship operates from the target area; And
Comprising the step of comparing the inputted track value with a probability model for each ship type, ship size, and group of times in the extracted grid area, and determining whether the arbitrary ship has abnormal operation according to the separation degree of the track value,
In the generating of the cluster-specific probability model, the probability model of the position value, course value, or speed value of the vessel for each grid area for each vessel type, vessel size, and time slot is converted into a probability density function (PDF: Probability Density Function). However, the probability density function is estimated using the Gaussian-Kernel Density Estimation (KDE) method, and each of the cluster-specific probability models generated by ship type, ship size, and time period is the probability density of position value, course value, and velocity value. Contains functions,
The step of determining whether the arbitrary vessel has abnormal operation may include comparing the probability model for each cluster generated for each vessel type, vessel size, and time slot, and the input position value, course value, and speed value, and the position and course of the vessel. Or an abnormal operation vessel identification method, characterized in that to determine whether the vessel is abnormal operation for each speed.
The method of claim 1,
The collected track data is the track data and radar received from the Automatic Identification System (AIS) installed in each vessel operating within the control area in charge of the Vessel Traffic Service System (VTS). An abnormal vessel identification method comprising the track data obtained through.
The method of claim 1,
The step of calculating the learning data for the navigation pattern of the ship,
Unsupervised learning is performed using an autoencoder model, which is a neural network model that generates training data composed of pairs of inputs and inputs,
The method for identifying abnormal ships, characterized in that the autoencoder model is trained using only the input data composed of the track data to generalize the operation pattern of the ship in the target area.
The method of claim 3,
The step of calculating the learning data for the navigation pattern of the ship,
Normalizing the collected track data;
Generating training data by unsupervised learning the normalized track data using the autoencoder model; And
As the learning of the autoencoder model is completed, generating an encoder for projecting input data into 3D data and a decoder for reconstructing the same,
The input and output data of the auto-encoder model, characterized in that consisting of three-dimensional data of the position, course, and speed of abnormal vessel identification method.
The method of claim 1,
The clustering step,
Generating 3D training data by performing encoding to project the generated training data into 3D data using the generated encoder; And
Including the step of clustering the generated three-dimensional learning data using a k-means clustering algorithm,
A method of identifying an abnormal vessel, characterized in that an elbow technique or a silhouette technique is used to find the optimal number of clusters.
delete delete delete

Images