KR101902997B1 - Automatic identification system for anomaly operation status of ship using unsupervised learning method and method thereof - Google Patents

Abstract

The present invention relates to an automatic identification system for the abnormal operation status of a vessel using an unsupervised learning method and a method thereof. Especially, the automatic identification system receives the track data of a radar and a vessel′s automatic identification device (AIS) in a vessel traffic service system (VTS) to perform unsupervised learning, automatically determines the operation status of the vessel operating a control area by using the learning result, identifies an abnormal operation ship, and displays the result. The automatic identification system includes a vessel traffic database, an unsupervised learning machine, an abnormal status checking device and a vessel traffic control information display device.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic identification system for a ship,

[0001] The present invention relates to a system and method for automatically detecting an abnormal operation status of a ship using a non-guidance learning method, and more particularly, to a system and method for detecting an abnormal operation status of a ship by using an automatic navigation system (AIS) A vessel using an unoccupied learning method for automatically identifying the operation status of a ship operating the control zone using the learning result and identifying the abnormal operation ship and displaying the result, The present invention relates to an automatic abnormality detection system and method thereof.

In general, the Vessel Traffic Service System is a system that provides traffic services to enhance port safety or harbor operation efficiency in terms of traffic congestion, increase in dangerous cargo, and potential environmental pollution . The maritime traffic control system provides information on the maritime situation and marine traffic situation to the ship in a timely manner in the control area, so that it can help the maritime decision making process on the ship. The maritime traffic control system was first installed in Korea in Pohang Port in 1993 and has been installed in 14 trade ports and coastal areas of JinDo.

On the other hand, the abnormal ship refers to an abnormal navigation ship, such as a smuggler, escape vessel, accident vessel, non-powered vessel, and unregulated vessel. Above such behavior, the ship generally has a serious effect on the ship which normally operates by repeating departure and return of the route in the form of a gaiter, by a sudden change of line speed, by turning around, or by suddenly changing the route .

Korean Patent Registration No. 1281673 proposes a fuzzy reasoning algorithm for abnormal navigation identification that can automatically identify a malfunctioning ship so that identification of the malfunctioning ship can be performed accurately and precisely without expertise or proficiency, There is disclosed a new type of intelligent navigation operation abnormal vessel identification method in which the identification work of the abnormal ship can be effectively supplemented.

Such a conventional technique is a method for calculating an abnormal state of an input item by defining an abnormal state as an interval function and summing the calculated abnormal state to calculate a degree of behavior abnormality. The problem that the magnitude of the amount of change and the cumulative amount of change must be defined in advance for each item , And the definition of this section has an adverse effect on the performance of the entire system.

It is therefore an object of the present invention to provide a fuzzy inference method of a conventional art in which a pattern is learned using a large amount of data without adjusting the parameters of the system, It is possible to classify the state and anomalous state, and it is possible to identify the abnormal state of the ship more accurately than the map learning method which learns by using the data composed of input-output pairs. System and method therefor.

In order to achieve the above object, an automatic abnormality detection system of a ship using a non-beauxiliary learning method according to an embodiment of the present invention is provided with an automatic ship identification system (AIS) and a radar snapshot data in a marine traffic control system (VTS) A vessel traffic database configured to receive and database; A non-navigation learning unit configured to receive a set of navigation data in the marine traffic database and to learn non-marginality; An abnormality state identifier configured to automatically determine an operation state of a ship operating a control area using the result of the non-feature learning by the non-position learning device to identify an abnormally operated ship; And a vessel traffic control information display device configured to display an abnormally operated vessel identification result made by the abnormal state identifier.

In the system for automatically identifying an abnormal operation status of a ship using the non-guidance learning method according to the above embodiment, the marine traffic database is the marine data of the marine automatic identification device. The marine traffic database includes information on the ship name, type of ship, The position information, the speed information of the ship with respect to time, the ship's rolling information according to the time, and the loaded cargo information, and it is the radar data of the radar, the position information of the timepiece according to time, And information on the rolling of the following objects.

In the automatic abnormality navigation system for a ship using the non-ambiguity learning method according to the above embodiment, the non-ambiguity learning device comprises: a data normalization unit configured to receive and normalize the navigation data of the ship automatic identification device; A navigation state learning unit configured to learn a neuron using the learning data normalized by the data normalization unit and generate an encoder and a decoder using a model in which learning is completed; A navigation state classifying unit configured to classify the learning data into a predetermined number of groups using the encoder; And a navigation state sequence learning unit configured to generate a navigation state sequence for each ship using the learning data classified by the navigation state classification unit and to learn the generated navigation state sequence as learning data.

In the automatic abnormality navigation system of a ship using the non-ambiguity learning method according to the above embodiment, the non-ambiguity learning apparatus can use an auto-encoder model which is a neural network model using learning data composed of input-input pairs.

In the automatic abnormality detection system for a ship using the non-tracking learning method according to the above embodiment, the auto encoder model has four input layers, eight hidden layers 1, four hidden layers 2, eight hidden layers 3 , And a neuron with four output layers, and the input and output data can be configured in four dimensions (latitude, longitude, incidence, velocity).

In the automatic abnormality detection system for a ship using the non-tracking learning method according to the present invention, the navigation state sequence learning unit may use an LSTM auto-encoder model for time series data .

In the system for automatically identifying an abnormal operation status of a ship using the non-guidance learning method according to the above embodiment, the ship traffic control information display device displays, on an electronic chart of the monitor, Orange: Warning, Red: Above).

In order to achieve the above object, the present invention provides a method of automatically identifying abnormal operation status of a ship by using a non-guidance learning method according to another embodiment of the present invention, in which a ship navigation database receives navigation data of an automatic ship identification device and a radar in a marine traffic control system Database; The non-bore learning machine receives the set of navigation data from the vessel traffic database and learns the bidirectionality; Identifying an abnormality-operating vessel by automatically determining an operation state of the vessel operating the control area using the result of the unidirection-learned by the abnormality state learning step; And displaying the abnormal operation vessel identification result made by the abnormal condition identifier in the vessel traffic control information display apparatus.

The method of automatically identifying an abnormal operation status of a ship using the non-affinity learning method according to another embodiment of the present invention includes the steps of: normalizing the data normalization unit by receiving the snapshot data of the ship automatic identification device; Learning the neuron using the learning data normalized by the normalizing step by the operating state learning unit, and generating an encoder and a decoder using the learning completed model; Classifying the training data into a set number of groups using the encoder; And a step of the navigation state sequence learning unit generating a navigation state sequence for each ship using the learning data classified in the classification step and using the generated sequence as a learning data.

According to the system for automatically identifying abnormal operation status of a ship and the method using the non-guidance learning method according to the embodiment of the present invention, the ship navigation database receives the navigation data of the ship automatic identification device and the radar in the marine traffic control system, The non-bore learning machine inputs a set of wake-up data in the ship traffic database and learns non-bore degree, and the abnormal state identifier uses the result of non-bore learning by the non-bore level learning step to operate the vessel And the vessel traffic control information display device displays the abnormal operation vessel identification result made by the abnormal condition identifier, so that the system parameters such as the conventional fuzzy inference method Even if you do not adjust one by one, It is possible to classify the steady state and the abnormal state of the ship by learning the pattern using the tool and it has an excellent effect of accurately identifying the abnormal condition of the ship as compared with the learning method of learning using the data composed of the input- have.

That is, according to the present invention, an unoccupied learning method using a set of snapshot data composed only of input data is used to identify an abnormally operated ship. Therefore, even if the system parameters are not adjusted one by one like the conventional fuzzy inference method, It is possible to classify the steady state and the abnormal state of the ship by learning the pattern by using the data and it is possible to accurately identify the abnormal state of the ship as compared with the learning method of learning using the data composed of the input- have.

1 is a block diagram of an automatic abnormality detection system for a ship using a non-guidance learning method according to an embodiment of the present invention.
2 is a detailed block diagram of the non-illustrated learning apparatus of FIG.
3 is a configuration diagram of an auto-encoder model used in the flight state learning unit of FIG.
4 is a configuration diagram of an LSTM auto-encoder model used in the flight state sequence learning unit of FIG.
5 is a flow chart showing a method of automatically identifying an abnormal operation state of a ship using a non-background learning method according to an embodiment of the present invention.
FIG. 6 is a detailed flowchart of S200 of FIG.

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

FIG. 1 is a block diagram of an automatic abnormality detection system for a ship using a non-beauxiliary learning method according to an embodiment of the present invention, and FIG. 2 is a detailed block diagram of the non-beating learning apparatus of FIG.

As shown in FIGS. 1 and 2, the system for automatically detecting an abnormal operation status of a ship using the non-guidance learning method according to an embodiment of the present invention includes a ship traffic database 100, a non-shadow learning device 200, An identifier (300) and a vessel traffic control information display device (400).

The Vessel Traffic Database 100 plays a role of receiving the ship's automatic identification device (AIS) and the radar data from the VTS in a database.

As shown in Table 1 below, the ship traffic database 100 stores the ship name, the type of ship, the specification of the ship, the position information of the ship with respect to time, The information of the ship's velocity along the time, the information of the ship's ship along time, the information of the radar, the position information of the timetable according to time, the speed information of the timetable according to time, .

[Table 1]

On the other hand, a ship not equipped with an automatic identification device (AIS) can grasp flight information of a ship only through radar.

The navigation data of the ship automatic identification device (AIS) and the learner are received every second by the marine traffic control system (VTS), and this information is accumulated in the marine traffic database 100 and used as learning data.

The unavailability learning device 200 receives a set of navigation data as learning data in the marine traffic database 100 and learns the non-navigation learning.

Since the non-mapping learning device 200 is difficult to generate learning data composed of input-output pairs due to the characteristics of marine traffic data, it uses an auto-encoder model which is a neural network model using learning data composed of input-input pairs. The auto-encoder model can learn the model using only input data consisting of a set of navigation data, and can generalize the navigation pattern of the target port.

2, the non-illustrated learning apparatus 200 includes a data normalization unit 210, a flight state learning unit 220, a flight state classifying unit 230, and a flight state sequence learning unit 240 .

The data normalization unit 210 receives the snapshot data (i.e., the position (latitude and longitude), the slope, and the speed information of the ship) of the automatic ship identification system (AIS) in the ship traffic database 100, Interpolation is performed with a period (about 10 seconds) and normalized to a real value between 0 and 1 using the maximum value / minimum value of each item based on the interpolated data as shown in the following Equation (1).

[Equation 1]

는 정규화 된 입력이고,

는 임의의 학습 데이터이며,

는 전체 학습 데이터 세트를 나타낸다)(here,

Is a normalized input,

Is arbitrary learning data,

Represents the entire learning data set)

The navigation state learning unit 220 learns the neuron using the learning data normalized by the data normalization unit 210, and generates an encoder and a decoder using the model in which the learning is completed.

The navigation state learning unit 220 is an auto-encoder model that uses learning data of input-input pairs.

The auto-encoder model includes four input layers, eight hidden layers 1, four hidden layers 2, eight hidden layers 3, and four output layers as input and output data Is composed of four dimensions (latitude, longitude, penetration, speed).

The learning is performed until the data of the input layer and the output layer are equal to each other. The learning completed model generates an encoder for projecting the input data into the four-dimensional data and a decoder for restoring the encoder. do.

The navigation state classifying unit 230 classifies the learning data into a set number of groups using an encoder.

The navigation state classifying unit 230 projects the learning data on the four-dimensional space using the encoder of the learned model.

The auto-encoder model has the advantage of grouping data with similar meaning and generalizing the snapshot data.

Accordingly, the projected result can be divided into arbitrary groups through a clustering algorithm such as k-means. In the embodiment of the present invention, the projected results are classified into 20 status groups (the number of status groups is not important, Determined by the appropriate number that can be distinguished).

As a result of classification, all learning data are divided into 20 states.

The navigation state sequence learning unit 240 generates a navigation state sequence for each ship using the learning data classified by the navigation state classification unit 230, and learns it by using it as learning data.

The navigation state sequence learning unit 240 uses an LSTM auto-encoder model for time-series data, as shown in FIG.

The learning data used in the flight state sequence learning unit 240 learns a flight state sequence for a total of 60 seconds using six consecutive state sequences interpolated in units of 10 seconds.

The time range (60 seconds) of the state sequence can be adjusted by expanding or reducing the input and output of the model.

The abnormal condition identifier 300 automatically identifies the operation status of the ship operating the control area by using the result of the non-feature learning by the non-feature learning device 200, thereby identifying the abnormal operation ship.

In the abnormal condition identifier 300, the abnormal operation status of the ship includes both 1) untrained navigation status and 2) untrained navigation status.

1) takes the difference between the input value and the output value of the learned flight state learning model as anomality, and it can be exemplified as operating in a low water depth area or stopping in a route.

2) represents the difference between the input value and the output value of the operating state sequence learning model as the outliers.

The difference between the input value and the output value can be obtained by using the euclidean distance value between the two vectors or by using the cosine similarity value.

The final ship's abnormal operation value is calculated by summing 1) value and 2) value.

The vessel traffic control information display device 400 displays the abnormal operation vessel identification result made by the abnormal state identifier 300.

The vessel traffic control information display device 400 displays the abnormal state of each ship on the electronic chart of the monitor in a stepwise color (for example, green: safe, orange: warning, red: ideal) do.

Hereinafter, a description will be made of a method of automatically identifying an abnormal operation state of a ship using the automatic abnormality detection system of a ship using the non-guidance learning method according to an embodiment of the present invention constructed as described above.

FIG. 5 is a flow chart showing a method of automatically identifying an abnormal operation state of a ship using a non-guidance learning method according to an embodiment of the present invention, FIG. 6 is a detailed flowchart of S200 of FIG. 5, ).

First, the ship traffic database receives the navigation data of the ship automatic identification device (AIS) and the radar in the VTS (100), and forms the database (S100).

Then, the non-illustrated learning machine 200 receives the set of the navigation data from the marine traffic database 100 and learns the non-learning data (S200).

Step S200 will be described in more detail with reference to FIG.

First, the data normalization unit 210 classifies the ship position (latitude, longitude), slip and speed information among the information of the snapshot data of the AIS, according to the ship, interpolates it in a set period, And normalized to a real number value between 0 and 1 using the maximum value and the minimum value of each item based on the calculated values (S210).

Then, the navigation state learning unit 220 learns each of the neurons using the learning data normalized by the step S210, and the learning is performed until the data of the input layer and the output layer are equal to the set number, The completed model generates an encoder for projecting the input data into four-dimensional data and a decoder for reconstructing the encoder (S220).

Thereafter, the flight state classifying unit 230 projects the learning data onto the four-dimensional space using the encoder, and classifies the projected results into the set number of groups using the clustering algorithm (S230).

Next, the navigation state sequence learning unit 240 generates a navigation state sequence for each ship using the learning data classified in step S230, and learns the navigation state sequence as learning data (S240).

In step S300, the abnormal state identifier 300 automatically determines the operating state of the ship operating the control area using the result of the non-grid learning in step S200, thereby identifying the abnormally operated vessel.

In step S400, the vessel traffic control information display device 400 displays the abnormally operated vessel identification result made by the abnormal state identifier 300. [

According to the system for automatically identifying abnormal operation status of a ship and the method using the non-guidance learning method according to an embodiment of the present invention, the ship navigation database receives the navigation data of the ship automatic identification device and the radar in the marine traffic control system, The non-bore learning machine inputs a set of wake-up data in the ship traffic database and learns non-bore degree, and the abnormal state identifier uses the result of non-bore learning by the non-bore level learning step to operate the vessel And the vessel traffic control information display device displays the abnormal operation vessel identification result made by the abnormal condition identifier, so that the system parameters such as the conventional fuzzy inference method Large amounts of data Can be used to classify the ship's steady state and anomalous state by learning the pattern and it is possible to accurately identify the anomalous state of the ship as compared with the MAP learning method using data composed of input-output pairs.

That is, according to the present invention, an unoccupied learning method using a set of snapshot data composed only of input data is used to identify an abnormally operated ship. Therefore, even if the system parameters are not adjusted one by one like the conventional fuzzy inference method, It is possible to classify the steady state and the abnormal state of the ship by learning the pattern using the data, and it is possible to accurately identify the abnormal state of the ship as compared with the map learning method which learns using the data composed of the input-output pairs.

Although the best mode has been shown and described in the drawings and specification, certain terminology has been used for the purpose of describing the embodiments of the invention and is not intended to be limiting or to limit the scope of the invention described in the claims. It is not. Therefore, those skilled in the art will appreciate that various modifications and equivalent embodiments may be made without departing from the scope of the present invention. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

100: Vessel traffic database
200: Non-Bidirectional Learning Machine
210: Data normalization unit
220: operating state learning section
230:
240: Operation status sequence learning unit
300: Abnormal state identifier
400: Ship traffic control information display device

Claims (9)

A ship traffic database configured to receive and record database of the ship automatic identification device (AIS) and the radar data of the radar in the VTS;
A non-navigation learning unit configured to receive a set of navigation data in the marine traffic database and to learn non-marginality;
An abnormality state identifier configured to automatically determine an operation state of a ship operating a control area using the result of the non-feature learning by the non-position learning device to identify an abnormally operated ship; And
And a vessel traffic control information display device configured to display an abnormally operated vessel identification result made by the abnormal state identifier,
The non-
A data normalization unit configured to receive and normalize the snapshot data of the ship automatic identification device;
A navigation state learning unit configured to learn a neuron using the learning data normalized by the data normalization unit and to generate an encoder and a decoder using a model in which learning is completed;
A navigation state classifying unit configured to classify the learning data into a predetermined number of groups using the encoder; And
And a navigation state sequence learning unit configured to generate a navigation state sequence for each ship by using the learning data classified by the navigation state classification unit and to learn it by using it as learning data, Automatic identification system. The method according to claim 1,
The vessel traffic database
The data of the ship's automatic identification device includes the ship's name, the type of ship, the specification of the ship, the position information of the ship with respect to time, the speed information of the ship with respect to time,
A system for automatically detecting an abnormal operation state of a ship by using a non-geometry learning method, which includes position information of a timeline along time, speed information of a timeline along time, and creep information of a timber along with time. delete The method according to claim 1,
The non-guidance learning apparatus is an automatic abnormality detection system for a ship using a non-background learning method using an auto-encoder model, which is a neural network model using learning data composed of input-input pairs. 5. The method of claim 4,
The auto-encoder model
Four input layers, eight hidden layers 1, four hidden layers 2, eight hidden layers 3, and four output layers,
The input and output data consists of four dimensions (latitude, longitude, slope, speed). The method according to claim 1,
The navigation state sequence learning unit
An automatic identification system of abnormal operation status of ship using non-grid learning method using LSTM auto-encoder model for time series data. The method according to claim 1,
The ship traffic control information display device
An automatic identification system of abnormal operation status of a ship by using the non-grid learning method, in which the abnormal condition of each ship on the monitor's electronic chart is displayed in a stepwise color (green: safe, orange: warning, red: ideal). A method for automatically identifying an abnormal operation status of a ship using an automatic abnormal operation status identification system of a ship using the non-background determination method as set forth in claim 1,
The ship traffic database receiving the ship automatic identification device and the radar data of the radar in the marine traffic control system,
The non-bore learning machine receives the set of navigation data from the vessel traffic database and learns the bidirectionality;
Identifying an abnormality-operating vessel by automatically determining an operation state of the vessel operating the control area using the result of the unidirection-learned by the abnormality state learning step; And
And displaying the vessel operation result identification result made by the abnormality condition identifier by the vessel traffic control information display apparatus. 9. The method of claim 8,
The non-background learning step
The data normalization unit receives and normalizes the snapshot data of the ship automatic identification device;
Learning the neuron using the learning data normalized by the normalizing step by the operating state learning unit, and generating an encoder and a decoder using the learning completed model;
Classifying the training data into a set number of groups using the encoder; And
Wherein the navigation state sequence learning unit generates a navigation state sequence for each ship using the learning data classified in the classification step and uses the generated sequence as learning data to learn the abnormal state of the ship.

Images