Ship anchor ground abnormal behavior detection method and system

Abstract

The invention provides a ship anchoring ground abnormal behavior detection method and system, and the method comprises the steps: carrying out the preprocessing of ship navigation data, screening and segmenting of anchoring state data, coordinate system conversion, carrying out the clustering processing of the processed data through employing a DBSCAN clustering algorithm and an MLESAC algorithm, and carrying out the circle fitting and noise point recognition. And marking abnormal data points in the ship anchoring process by combining the noise points and a circle fitting result, and classifying the data abnormal points and outputting a result. When a ship is in an anchoring state, once dragging occurs, the ship deviates from an original arc-shaped area, track points present low-density noise distribution, noise in anchoring track data can be effectively processed by adopting the DBSCAN algorithm, the identification of dragging track data has remarkable advantages, the MLESAC algorithm is subjected to multiple times of iterative optimization, and the optimal circle center and radius are calibrated and fitted, so that the noise in the anchoring track data can be effectively processed. The accuracy is high, the advantage of processing complex data is obvious, and the abnormal behaviors of ship anchoring ground anchoring are accurately recognized and classified.

Description

Technical Field

[0001] The present invention relates to the technical field of ship behavior recognition, and in particular to a method and system for detecting abnormal behavior of a ship at anchor. Background Art

[0002] Anchoring is one of the most basic operating states in the shipping process, which allows ships to be fixed in port anchorages or specific areas for loading and unloading of goods, shelter from the wind, rest and replenishment, etc. As an important part of global trade and supply chain, the safety, efficiency and intelligence level of the shipping industry have an important impact on economic development and strategic position. With the development of intelligent and automated technologies, many countries and regions around the world have issued smart shipping development strategies to promote the development of ships towards intelligence, automation and information sharing.

[0003] In the context of intelligent shipping, ship behavior recognition technology has become one of the key technologies, especially for the recognition of abnormal behavior of ships at anchor (such as dragging anchor). Dragging anchor refers to the phenomenon that a ship loses its anchoring stability and begins to drift when it is anchored, which may lead to major safety accidents such as property loss and casualties. Identifying and detecting abnormal behavior of ships at anchor, especially dragging anchor, is the basis for ensuring the safe operation of ships. The traditional dragging anchor detection method relies on the analysis of parameters such as the speed and draft of the ship, and is inferred by the mathematical model of the anchor chain length and the berth water depth. Although this method has a certain theoretical basis, it has the problem of insufficient accuracy in actual operation, is prone to large errors, and cannot effectively guarantee the accurate recognition of the ship's anchoring state. Therefore, how to accurately identify the abnormal behavior of ships anchoring at anchor has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In order to solve the technical problem of accurately classifying and identifying abnormal behaviors of ships anchoring in anchorages, the present invention proposes a method and system for detecting abnormal behaviors of ships anchoring in anchorages, which processes historical data of ships and solves the problem that traditional dragging anchor detection is difficult to implement accurately and is prone to large errors.

[0005] The present invention provides the following specific solutions:

[0006] In a first aspect, a method for detecting abnormal behavior of a ship at an anchorage includes:

[0007] S1. Data acquisition and preprocessing: Obtain the navigation data of the ship from the database, clean, align and interpolate the navigation data according to the original time series to obtain the data to be processed; the navigation data includes basic information of the ship, longitude and latitude coordinates and navigation status of the ship;

[0008] S2. Filtering the anchoring state data and segmenting: traversing the data to be processed, filtering out the data of the ship in the anchoring state, and segmenting the anchoring state data according to the time series to obtain anchoring segments; each segment of the anchoring segment represents an anchoring state period of the ship;

[0009] S3. Coordinate conversion: Calculate the average latitude and longitude coordinates of the ship during an anchoring period, determine the UTM area and hemisphere of the ship in the current anchoring section based on the average latitude and longitude coordinates, and convert the latitude and longitude coordinates into UTM coordinates through a converter;

[0010] S4. Clustering of anchoring status data: DBSCAN clustering is performed on the UTM coordinate data of each anchoring segment obtained based on the conversion of S3, the UTM coordinate data are grouped based on the neighborhood density to obtain clustering results, and each UTM coordinate of the anchoring segment is marked as a specific cluster or the first noise point;

[0011] S5. Circle fitting and noise point identification: Use the MLESAC algorithm to fit the clustering results to establish an optimal model, calculate the optimal center and radius, and mark the points outside the fitting radius as the second noise point;

[0012] S6. Classify the noise points in combination with the noise points and the circle fitting results: merge the first noise point and the second noise point into noise points, set an outlier classification rule, classify the noise points according to the outlier classification rule, determine whether the radius of the fitting circle exceeds the threshold, obtain the first type of outliers, and set them as the first priority; and further determine the anchoring state of the first type of outliers based on the noise points and the DBSCAN clustering results, whether they belong to the second type of outliers; if they belong to the second type of outliers, determine the anchoring state based on the distance between the noise points, and further classify the second type of outliers to obtain the second priority, the third priority, and the fourth priority;

[0013] S7. Result output: The classification and priority classification results of the abnormal points are output, and the track abnormality classification of each anchoring section is distinguished and drawn with different colors and shapes.

[0014] Preferably, the cleaning, alignment and interpolation preprocessing of the navigation data according to the original time series in S1 includes:

[0015] Preprocess the navigation data to ensure that the time intervals of the data points are consistent and aligned with the time axis; delete duplicate data; interpolate if there are missing values; and ensure that the data columns have the correct data type;

[0016] Ship={a1,a2,...,a n} (1)

[0017]

[0018]

[0019] Where Ship is a given list of ships, is the longitude and latitude coordinates of the i-th ship at time point t, is the navigation status of the i-th ship at time point t, where Indicates that the vessel is at anchor.

[0020] Preferably, the specific steps of S2 are:

[0021] S21: traverse the data to be processed and filter out The data is the anchor status data;

[0022] S22: Set storage list: Set n storage lists D for storing data current , where current = 1, 2, ... n, N is the number of data points in the storage list; Formula (5) is regarded as the beginning of a new anchoring section, and current starts from 1;

[0023]

[0024] Among them, D current To store the list, is the navigation status of the i-th ship at time point t, is the longitude and latitude coordinates of the i-th ship at time point t;

[0025] S23: Complete a mooring segment screening: set a temporary storage list N non , used to record the points in the non-anchored state. If 5 consecutive points are in the non-anchored state and the current storage list D current If there are no less than 20 points in the mooring section, the screening of one mooring section is completed, that is,

[0026] if N non ≥5 and |D current |≥20,D current ←D current [1:N-5] (6)

[0027] Among them, N non Stores a list of non-anchored points, D current is the storage list, N is the number of data points in the storage list;

[0028] S24: Traverse all data: store current D current To D, use the next D currentContinue to traverse the remaining data until the screening is completed, and finally obtain the segmented processing list of the anchor data, that is, the anchor segment set D.

[0029] D={D1,D2,...,D n} (7)

[0030] Where D is the mooring segment set.

[0031] Preferably, the specific steps of S3 are:

[0032] S31: Calculate the average longitude and latitude of each anchoring segment: traverse the anchoring segment set D and calculate the current anchoring segment D i The average value of the longitude and latitude of all longitude and latitude coordinates in , as shown in equations (8) and (9);

[0033]

[0034] Where N is the number of coordinate points, lon j is the longitude of point j, lat j is the latitude of point j, mean_lon is the mean longitude, and mean_lat is the mean latitude;

[0035] S32: Determine the UTM region and the hemisphere to which it belongs: using the average value of the longitude and latitude, determine the UTM region by formula (10), and determine the hemisphere to which it belongs by formula (11);

[0036]

[0037] Among them, utm_zone is the UTM zone, and hemisphere is the hemisphere to which it belongs;

[0038] S33: Coordinate conversion: After determining the UTM area and the hemisphere to which it belongs, set the projection parameters of the corresponding UTM coordinate system and use the converter to convert the current anchor segment D i All longitude and latitude coordinates in are converted to UTM coordinates, as shown in formula (12):

[0039] M′=(x,y)=Transformer.transform(M(lon,lat)) (12)

[0040] Among them, M(lon,lat) is the original longitude and latitude coordinates, and M′ is the converted UTM coordinates;

[0041] S34: After traversing all anchoring segments: calculate the average latitude and longitude of all latitude and longitude coordinates of the next anchoring segment, determine the UTM region and the hemisphere to which it belongs based on the average latitude and longitude, and use a converter to convert all latitude and longitude coordinates of the current anchoring segment into UTM coordinates until the traversal is completed, and obtain each anchoring segment Di UTM coordinate data.

[0042] Preferably, the specific steps of S4 include:

[0043] S41: DBSCAN clustering of UTM coordinate data: DBSCAN clustering is performed on the UTM coordinate data of each anchoring segment, and the neighborhood radius parameter eps and the minimum sample number min_sample are set, as shown in formula (13):

[0044] db=DBSCAN(eps=100,min_sample=8).fit(M′) (13)

[0045] Among them, eps is the neighborhood radius parameter, min_sample is the minimum number of samples;

[0046] S42: Determine the core point: If there is a point P, and the number of UTM coordinates in the neighborhood with a radius of eps and centered at point P is greater than or equal to the minimum number of samples min_sample, then point P is a core point; as shown in formula (14):

[0047] P = core point, if N eps (P)≥min_sample(14)

[0048] Among them, eps is the neighborhood radius parameter, min_sample is the minimum number of samples, N eps (P) is the number of points in the neighborhood;

[0049] S43: Coordinate point judgment: other points in the neighborhood of the core point are boundary points and stored as a specific cluster. Points outside the neighborhood are marked as the first noise points, stored as the first noise points, and marked with cluster labels, as shown in formula (15-16).

[0050]

[0051] H={H1,H2,...,H n},cluster(H i )=-1 (16)

[0052] Among them, L is the set of specific clusters, H is the set of the first noise points, unique_clusters i is the cluster label set, M′(unique_clusters i ) is a coordinate array marked with cluster labels, cluster(H i ) is the cluster label of the noise point.

[0053] Preferably, the specific steps of S5 are:

[0054] S51: traverse the specific cluster set L;

[0055] S52: Determine the center and radius of the circle: From the current specific cluster L i In the UTM coordinates, randomly select at least three sample points to calculate the parameters of the circle, namely the center (xc, yc) and the radius r;

[0056] S53: Calculate the current specific cluster L i All other points (x i ,y i ) to the center of the circle, as shown in formula (17),

[0057]

[0058] Among them, R i is the distance from point i to the center of the circle, (x i ,y i ) is any point other than the center of the circle, (xc,yc) is the center of the circle;

[0059] S54: Calculate residual value: Calculate the residual value of the distance from all other points to the center of the circle and the radius r, as shown in formula (18),

[0060] residual i =|r i -r| (18)

[0061] Among them, residual i is the residual value, r i is the distance from point i to the center of the circle, and r is the radius;

[0062] S55: Determine the coordinate point: set a residual value threshold residual_threshold, and among all other points, the points whose residual values ​​are less than the threshold are marked as inner points, otherwise they are marked as outer points;

[0063] S56: Determine the best fitting circle and mark the second noise point: Repeat steps S52-S55 until the preset number of trials is reached, select the model with the most inner points among all model results, and output the best fitting radius R and the center (XC, YC) in the model, mark the outer point as the second noise point, and store it as G i ,

[0064] G i =(x i ,y i ), if residual i ≥ residual_threshold (19)

[0065] Among them, residuali is the residual value, residual_threshold is the threshold of the residual value;

[0066] S57: Execute the next specific cluster until the traversal is completed, and obtain the set G of the second noise point:

[0067] G={G1,G2,...,G n} (20)

[0068] Wherein, G is the set of second noise points.

[0069] Preferably, S6 specifically includes:

[0070] S61: Merge noise points: store the first noise points and the second noise points as a noise point set F, as shown in formula (21):

[0071] F=G∪H (21)

[0072] Where F is the set of noise points, G is the set of second noise points, and H is the set of first noise points;

[0073] S62: Setting a first threshold, and traversing the noise point set F;

[0074] S63: First priority classification: for the specific cluster L to which the current noise point belongs i In the above example, if the radius of a fitted circle is greater than the first threshold, then this part of the noise points does not conform to the normal anchoring form and is marked as the first type of abnormal points;

[0075] If there is a noise point F j , from the center of the fitting circle to which it belongs (XC i ,YC i ) is greater than 1.2 times the optimal fit radius R i , then the noise point is marked as the second type of outlier point I; as shown in formula (22),

[0076]

[0077] Among them, (x j ,y j ) is the coordinate of noise point j, (XC i ,YC i ) is the center of the fitting circle, R i is the optimal fitting radius;

[0078] S64: Second priority classification: For the second type of abnormal point I, if there is a noise point located at the starting point of the specific cluster to which it belongs, then the noise point and the continuous noise points retrieved backward are marked as entering anchoring; if there is a noise point located at the ending point of the specific cluster to which it belongs, then the noise point and the continuous noise points retrieved forward are marked as ending anchoring or changing behavior after moving anchor;

[0079] S65: Third priority classification: For the second type of abnormal point I, if there is a single noise point I a Or two noise points I a ,I b , the average distance from two adjacent normal points M and N is greater than 5 times the distance between M and N, then the noise point deviates greatly from the specific cluster to which it belongs, which is marked as data drift, that is,

[0080]

[0081] Among them, I a ,I b are two noise points, M, N are I a ,I b The points before and after, I a M.I a N.I. a M.I b N and MN both represent the distance between two corresponding points;

[0082] S66: Fourth priority classification: For the second type of abnormal point I, if a single noise point or two normal data points before and after a continuous noise point segment belong to different specific clusters, then this part of the noise points is marked as anchored and re-anchored, as shown in formula (24-25).

[0083]

[0084] Among them, (x m-1 ,y m-1 ) is a specific cluster L a The coordinates of point m-1 in the middle, R a For a specific cluster L a The fitting radius, (XC a ,YC a ) is the center of the fitting circle, (x n+1 ,y n+1 ) is a specific cluster L b Coordinates of point n+1, R b For a specific cluster L b The fitting radius, (XC b ,YC b ) is the center of the fitting circle, M′ m-1 , M′ n+1are two normal data points before and after the noise point paragraph;

[0085] S67: Noise points that do not conform to the above classification are uniformly marked as other anchoring states.

[0086] In the second aspect, a ship anchorage abnormal behavior detection system includes: a data acquisition and processing module, a data screening and segmentation module, a coordinate conversion module, a clustering module, a fitting module, a noise classification module and a result output module;

[0087] The data acquisition and processing module includes: a data acquisition submodule for acquiring the navigation data of the ship from the database; a data preprocessing submodule for cleaning, aligning and interpolating the navigation data according to the original time series to obtain the data to be processed; the navigation data includes basic information of the ship, longitude and latitude coordinates and navigation status of the ship;

[0088] The data screening and segmentation module includes: a screening submodule for traversing the data to be processed and screening out the data of the ship in the anchored state, and a segmentation submodule for segmenting the anchored state data according to the time series to obtain the anchored segments; each segment of the anchored segment represents an anchored state period of the ship;

[0089] The coordinate conversion module is used to calculate the average value of the longitude and latitude coordinates of the ship during an anchoring state period, determine the UTM area and the hemisphere to which the ship belongs in the current anchoring section based on the average value of the longitude and latitude coordinates, and convert the longitude and latitude coordinates into UTM coordinates through a converter;

[0090] The clustering module is used to perform DBSCAN clustering on the UTM coordinate data of each anchoring segment obtained by the coordinate conversion module, group the UTM coordinate data based on neighborhood density to obtain clustering results, and mark each UTM coordinate of the anchoring segment as a specific cluster or a first noise point;

[0091] The fitting module uses the MLESAC algorithm to fit the clustering results to establish an optimal model, calculate the optimal center and radius, and mark the points outside the fitting radius as second noise points;

[0092] The noise point classification module includes: merging the first noise point and the second noise point as noise points, and setting a classification rule submodule for abnormal point classification rules, wherein the abnormal point classification rules include: classifying the noise points according to the abnormal point classification rules, judging whether the radius of the fitting circle exceeds a threshold, obtaining a first type of abnormal point, and setting it as a first priority; and further judging the anchoring state of the first type of abnormal point based on the noise point and the DBSCAN clustering result, whether it belongs to the second type of abnormal point; if it belongs to the second type of abnormal point, then judging the anchoring state based on the distance between the noise points, further classifying the second type of abnormal point, and obtaining a second priority, a third priority, and a fourth priority;

[0093] The result output module includes: a result output submodule for outputting the classification and priority classification results of the abnormal points, and a trajectory abnormality classification display submodule for distinguishing and drawing the trajectory abnormality classification of each anchoring section with different colors and shapes.

[0094] Beneficial effects of the present invention:

[0095] The present invention provides a method and system for detecting abnormal behavior of ship anchorage, which preprocesses, screens and segments ship navigation data, performs coordinate transformation on the processed data, processes the ship navigation data using DBSCAN clustering algorithm and MLESAC algorithm, and marks abnormal data points during the ship anchoring process. When a ship is at anchor, its normal trajectory points are usually densely distributed on a circular or arc-shaped path. Once the ship drags anchor, its speed will increase significantly, and the trajectory will shift, causing it to deviate from the original circular or arc-shaped area. At this time, the trajectory points present a low-density noise distribution. The DBSCAN algorithm can effectively process the noise in the ship's anchoring trajectory data, automatically identify data features, and does not require a preset number of clusters. It has significant advantages in identifying dragging anchor trajectory data; secondly, the MLESAC algorithm has undergone multiple iterations of optimization, calibrated and fitted the optimal center and radius, with high accuracy, and can adaptively mark inner and outer points according to data features, which has obvious advantages in processing complex data; it does not rely on static tests or sensor data, and realizes abnormal behavior detection at ship anchorages through analysis of ship historical data. The algorithm is highly robust, lower cost, and easier to operate; the output results are clear and intuitive, and can clearly distinguish abnormal behaviors of different feature classifications during the visualization process, so as to accurately identify and classify abnormal behaviors of ships anchoring at anchorages.

[0096] In particular, the use of different colors and shapes to distinguish and draw the trajectory abnormality classification of each anchoring section can clearly and intuitively discover abnormal behavior of ships at anchorages, shorten the reaction time of observers, and effectively improve the safety of shipping. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1The present invention provides a flow chart of a method for detecting abnormal behavior of a ship anchorage.

[0098] Figure 2 A schematic diagram of the abnormal point classification logic in a method for detecting abnormal behavior at a ship anchorage provided by the present invention.

[0099] Figure 3 A schematic diagram of "anchoring morphology abnormality" in a method for detecting abnormal behavior of a ship anchorage provided by the present invention.

[0100] Figure 4 A schematic diagram of "entering anchorage" and "changing behavior after ending anchorage or dragging anchor" in a method for detecting abnormal behavior of a ship anchorage provided by the present invention.

[0101] Figure 5 A schematic diagram of "data drift" in a method for detecting abnormal behavior in a ship anchorage provided by the present invention.

[0102] Figure 6 A schematic diagram of "re-anchoring after dragging anchor" in a method for detecting abnormal behavior of a ship anchorage provided by the present invention.

[0103] Figure 7 This is a framework diagram of a ship anchorage abnormal behavior detection system provided by the present invention. DETAILED DESCRIPTION

[0104] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0105] Figure 1 The flowchart of the abnormal behavior detection method of a ship anchorage provided by the present invention is as follows. Figure 1 As shown, a method for detecting abnormal behavior of a ship at anchorage includes:

[0106] S1. Data acquisition and preprocessing: Obtain the ship's navigation data from the database, clean, align and interpolate the navigation data according to the original time series to obtain the data to be processed; the navigation data includes basic ship information, longitude and latitude coordinates and ship navigation status.

[0107] Specifically, the cleaning, alignment and interpolation preprocessing of the navigation data according to the original time series includes:

[0108] Preprocess the navigation data to ensure that the time intervals of the data points are consistent and aligned with the time axis; delete duplicate data; interpolate if there are missing values; and ensure that the data columns have the correct data type;

[0109] Ship={a1,a2,...,a n} (1)

[0110]

[0111] Where Ship is a given list of ships, is the longitude and latitude coordinates of the i-th ship at time point t, is the navigation status of the i-th ship at time point t, where Indicates that the vessel is at anchor.

[0112] S2. Filtering and segmenting the anchoring state data: traversing the data to be processed, filtering out the data of the ship in the anchoring state, and segmenting the anchoring state data according to the time series to obtain anchoring segments; each segment of the anchoring segment represents an anchoring state period of the ship.

[0113] Specifically, the above S2 specific steps are:

[0114] S21: traverse the data to be processed and filter out The data is the anchor status data;

[0115] S22: Set storage list: Set n storage lists D for storing data current , where current = 1, 2, ... n, N is the number of data points in the storage list; Formula (5) is regarded as the beginning of a new anchoring section, and current starts from 1;

[0116]

[0117] Among them, D current To store the list, is the navigation status of the i-th ship at time point t, is the longitude and latitude coordinates of the i-th ship at time point t;

[0118] S23: Complete a mooring segment screening: set a temporary storage list N non , used to record the points in the non-anchored state. If 5 consecutive points are in the non-anchored state and the current storage list D current If there are no less than 20 points in the mooring section, the screening of one mooring section is completed, that is,

[0119] if N non ≥5 and |D current |≥20,D current ←D current [1:N-5] (6)

[0120] Among them, N non Stores a list of non-anchored points, D current is the storage list, N is the number of data points in the storage list;

[0121] S24: Traverse all data: store current D current To D, use the next D current Continue to traverse the remaining data until the screening is completed, and finally obtain the segmented processing list of the anchor data, that is, the anchor segment set D.

[0122] D={D1,D2,...,D n} (7)

[0123] Where D is the mooring segment set.

[0124] S3. Coordinate system conversion: Calculate the average values ​​of the longitude and latitude coordinates of the ship during an anchoring period respectively, determine the UTM area and hemisphere to which the ship belongs in the current anchoring section based on the average values ​​of the longitude and latitude coordinates, and convert the longitude and latitude coordinates into UTM coordinates through a converter.

[0125] Specifically, the above S3 specific steps are:

[0126] S31: Calculate the average longitude and latitude of each anchoring segment: traverse the anchoring segment set D and calculate the current anchoring segment D i The average value of the longitude and latitude of all longitude and latitude coordinates in , as shown in equations (8) and (9);

[0127]

[0128] Where N is the number of coordinate points, lon j is the longitude of point j, lat j is the latitude of point j, mean_lon is the mean longitude, and mean_lat is the mean latitude;

[0129] S32: Determine the UTM region and the hemisphere to which it belongs: using the average value of the longitude and latitude, determine the UTM region by formula (10), and determine the hemisphere to which it belongs by formula (11);

[0130]

[0131] Among them, utm_zone is the UTM zone, and hemisphere is the hemisphere to which it belongs;

[0132] S33: Coordinate conversion: After determining the UTM area and the hemisphere to which it belongs, set the projection parameters of the corresponding UTM coordinate system and use the converter to convert the current anchor segment D i All longitude and latitude coordinates in are converted to UTM coordinates, as shown in formula (12):

[0133] M′=(x,y)=Transformer.transform(M(lon,lat)) (12)

[0134] Among them, M(lon,lat) is the original longitude and latitude coordinates, and M′ is the converted UTM coordinates;

[0135] S34: After traversing all anchoring segments: calculate the average latitude and longitude of all latitude and longitude coordinates of the next anchoring segment, determine the UTM region and the hemisphere to which it belongs based on the average latitude and longitude, and use a converter to convert all latitude and longitude coordinates of the current anchoring segment into UTM coordinates until the traversal is completed, and obtain each anchoring segment D i UTM coordinate data.

[0136] S4. Clustering of anchoring status data: DBSCAN clustering is performed on the UTM coordinate data of each anchoring segment obtained based on the S3 conversion, the UTM coordinate data are grouped based on the neighborhood density to obtain clustering results, and each UTM coordinate of the anchoring segment is marked as a specific cluster or the first noise point.

[0137] Specifically, the above S4 specific steps include:

[0138] S41: DBSCAN clustering of UTM coordinate data: DBSCAN clustering is performed on the UTM coordinate data of each anchoring segment, and the neighborhood radius parameter eps and the minimum sample number min_sample are set, as shown in formula (13):

[0139] db=DBSCAN(eps=100,min_sample=8).fit(M′) (13)

[0140] Among them, eps is the neighborhood radius parameter, min_sample is the minimum number of samples;

[0141] S42: Determine the core point: If there is a point P, and the number of UTM coordinates in the neighborhood with a radius of eps and centered at point P is greater than or equal to the minimum number of samples min_sample, then point P is a core point; as shown in formula (14):

[0142] P = core point, if N eps (P)≥min_sample(14)

[0143] Among them, eps is the neighborhood radius parameter, min_sample is the minimum number of samples, N eps (P) is the number of points in the neighborhood;

[0144] S43: Coordinate point judgment: other points in the neighborhood of the core point are boundary points and stored as a specific cluster. Points outside the neighborhood are marked as the first noise points, stored as the first noise points, and marked with cluster labels, as shown in formula (15-16).

[0145]

[0146] H={H1,H2,...,H n},cluster(H i )=-1 (16)

[0147] Among them, L is the set of specific clusters, H is the set of the first noise points, unique_clusters i is the clustering label set, M′(unique_clusters i ) is a coordinate array marked with cluster labels, cluster(H i ) is the cluster label of the noise point.

[0148] S5. Circle fitting and noise point identification: Use the MLESAC algorithm to fit the clustering results to establish the optimal model, calculate the optimal circle center and radius, and mark the points outside the fitting radius as the second noise point.

[0149] Specifically, the above S5 specific steps are:

[0150] S51: traverse the specific cluster set L;

[0151] S52: Determine the center and radius of the circle: From the current specific cluster L i In the UTM coordinates, randomly select at least three sample points to calculate the parameters of the circle, namely the center (xc, yc) and the radius r;

[0152] S53: Calculate the current specific cluster L i All other points (x i ,y i ) to the center of the circle, as shown in formula (17),

[0153]

[0154] Among them, R i is the distance from point i to the center of the circle, (x i ,y i ) is any point other than the center of the circle, (xc,yc) is the center of the circle;

[0155] S54: Calculate residual value: Calculate the residual value of the distance from all other points to the center of the circle and the radius r, as shown in formula (18),

[0156] residual i =|r i -r| (18)

[0157] Among them, residual i is the residual value, r i is the distance from point i to the center of the circle, and r is the radius;

[0158] S55: Determine the coordinate point: set a residual value threshold residual_threshold, and among all other points, the points whose residual values ​​are less than the threshold are marked as inner points, otherwise they are marked as outer points;

[0159] S56: Determine the best fitting circle and mark the second noise point: Repeat steps S52-S55 until the preset number of trials is reached, select the model with the most inner points among all model results, and output the best fitting radius R and the center (XC, YC) in the model, mark the outer point as the second noise point, and store it as G i ,

[0160] G i =(x i ,y i ), if residual i ≥ residual_threshold (19)

[0161] Among them, residual i is the residual value, residual_threshold is the threshold of the residual value;

[0162] S57: Execute the next specific cluster until the traversal is completed, and obtain the set G of the second noise point:

[0163] G={G1,G2,...,G n} (20)

[0164] Wherein, G is the set of second noise points.

[0165] S6. Classify the noise points in combination with the noise points and the circle fitting results: merge the first noise points and the second noise points into noise points, set the outlier point classification rules, classify the noise points according to the outlier point classification rules, judge whether the radius of the fitting circle exceeds the threshold, obtain the first type of outliers, and set them as the first priority; and further judge the anchoring state of the first type of outliers based on the noise points and the DBSCAN clustering results, whether they belong to the second type of outliers; if they belong to the second type of outliers, judge the anchoring state based on the distance between the noise points, and further classify the second type of outliers to obtain the second priority, the third priority, and the fourth priority.

[0166] Figure 2 The present invention provides a method for detecting abnormal behavior at anchorages of ships. Figure 2 As shown, the logical steps of the above S6 for classifying noise points by combining noise points and circle fitting results are as follows:

[0167] S61: Merge noise points: store the first noise points and the second noise points as a noise point set F, as shown in formula (21):

[0168] F=G∪H (21)

[0169] Among them, F is the set of noise points, G is the set of second noise points, and H is the set of first noise points.

[0170] S62: Set a first threshold and traverse the noise point set F.

[0171] S63: First priority classification: for the specific cluster L to which the current noise point belongs i If the radius of a fitted circle is greater than the first threshold, then this part of the noise points does not conform to the normal anchoring form and is marked as the first type of abnormal points.

[0172] If there is a noise point F j , from the center of the fitting circle to which it belongs (XC i ,YC i ) is greater than 1.2 times the optimal fit radius R i , then the noise point is marked as the second type of outlier point I; as shown in formula (22),

[0173]

[0174] Among them, (x j ,y j ) is the coordinate of the noise point j, (XC i ,YC i ) is the center of the fitting circle, R i is the best fitting radius.

[0175] Through the above steps, the noise points are classified into the first type of abnormal points and the second type of abnormal points. The following steps continue to classify the second type of abnormal points.

[0176] S64: Second priority classification: For the second type of abnormal point I, if there is a noise point located at the starting point of the specific cluster to which it belongs, then the noise point and the continuous noise points retrieved backward are marked as entering anchoring; if there is a noise point located at the ending point of the specific cluster to which it belongs, then the noise point and the continuous noise points retrieved forward are marked as ending anchoring or changing behavior after moving anchor.

[0177] For example, Figure 4As shown in the figure, the noise point marked with an orange triangle is located at the starting position of the specific cluster to which it belongs, and the noise point and the continuous noise points retrieved backward are all in the anchoring state, so this part of the noise points is marked as entering anchoring; the noise point marked with a blue inverted triangle is located at the end point of the specific cluster to which it belongs, so the noise point and the continuous noise points retrieved forward are all marked as ending anchoring or changing behavior after moving anchor.

[0178] S65: Third priority classification: For the second type of abnormal point I, if there is a single noise point I a Or two noise points I a ,I b , the average distance from two adjacent normal points M and N is greater than 5 times the distance between M and N, then the noise point deviates greatly from the specific cluster to which it belongs, which is marked as data drift, that is,

[0179]

[0180] Among them, I a ,I b are two noise points, M, N are I a ,I b The points before and after, I a M.I a N.I. a M.I b N and MN both represent the distance between two corresponding points.

[0181] For example, Figure 5 As shown, the average distance between the noise point marked by the five-pointed star and the two normal points is much greater than 5 times the distance between the two normal points, so the noise point is marked as data drift.

[0182] S66: Fourth priority classification: For the second type of abnormal point I, if a single noise point or two normal data points before and after a continuous noise point segment belong to different specific clusters, then this part of the noise points is marked as anchored and re-anchored, as shown in formula (24-25).

[0183]

[0184]

[0185] Among them, (x m-1 ,y m-1 ) is a specific cluster L a The coordinates of point m-1 in the middle, R a For a specific cluster L a The fitting radius, (XC a ,YC a ) is the center of the fitting circle, (x n+1 ,y n+1 ) is a specific cluster Lb Coordinates of point n+1, R b For a specific cluster L b The fitting radius, (XC b ,YC b ) is the center of the fitting circle, M′ m-1 , M′ n+1 are two normal data points before and after the noise point paragraph.

[0186] For example, Figure 6 As shown, the normal points before the noise point paragraph marked by the cross are located in a specific cluster, while the normal points after the paragraph are located in another specific cluster, so this part of the noise points is marked as re-anchored after dragging the anchor.

[0187] S67: Noise points that do not conform to the above classification are uniformly marked as other anchoring states.

[0188] S7. Result output: The classification and priority classification results of the abnormal points are output, and the track abnormality classification of each anchoring section is distinguished and drawn with different colors and shapes.

[0189] like Figure 3-6 As shown in the figure, the trajectory anomaly classification of the anchoring section is drawn and output with different colors and shapes, which can clearly and intuitively discover and distinguish the abnormal behavior of ships in the anchorage.

[0190] Figure 7 This is a framework diagram of a ship anchorage abnormal behavior detection system provided by the present invention. Figure 7 As shown, a ship anchorage abnormal behavior detection system includes: a data acquisition and processing module, a data screening and segmentation module, a coordinate conversion module, a clustering module, a fitting module, a noise classification module and a result output module;

[0191] The data acquisition and processing module includes: a data acquisition submodule for acquiring the navigation data of the ship from the database; a data preprocessing submodule for cleaning, aligning and interpolating the navigation data according to the original time series to obtain the data to be processed; the navigation data includes basic information of the ship, longitude and latitude coordinates and navigation status of the ship;

[0192] The data screening and segmentation module includes: a screening submodule for traversing the data to be processed and screening out the data of the ship in the anchored state, and a segmentation submodule for segmenting the anchored state data according to the time series to obtain the anchored segments; each segment of the anchored segment represents an anchored state period of the ship;

[0193] The coordinate conversion module is used to calculate the average value of the longitude and latitude coordinates of the ship during an anchoring state period, determine the UTM area and the hemisphere to which the ship belongs in the current anchoring section based on the average value of the longitude and latitude coordinates, and convert the longitude and latitude coordinates into UTM coordinates through a converter;

[0194] The clustering module is used to perform DBSCAN clustering on the UTM coordinate data of each anchoring segment obtained by the coordinate conversion module, group the UTM coordinate data based on neighborhood density to obtain clustering results, and mark each UTM coordinate of the anchoring segment as a specific cluster or a first noise point;

[0195] The fitting module uses the MLESAC algorithm to fit the clustering results to establish an optimal model, calculate the optimal center and radius, and mark the points outside the fitting radius as second noise points;

[0196] The noise point classification module includes: merging the first noise point and the second noise point as noise points, and setting a classification rule submodule for abnormal point classification rules, wherein the abnormal point classification rules include: classifying the noise points according to the abnormal point classification rules, judging whether the radius of the fitting circle exceeds a threshold, obtaining a first type of abnormal point, and setting it as a first priority; and further judging the anchoring state of the first type of abnormal point based on the noise point and the DBSCAN clustering result, whether it belongs to the second type of abnormal point; if it belongs to the second type of abnormal point, then judging the anchoring state based on the distance between the noise points, further classifying the second type of abnormal point, and obtaining a second priority, a third priority, and a fourth priority;

[0197] The result output module includes: a result output submodule for outputting the classification and priority classification results of the abnormal points, and a trajectory abnormality classification display submodule for distinguishing and drawing the trajectory abnormality classification of each anchoring section with different colors and shapes.

[0198] It should be noted that the above-described specific implementations can enable those skilled in the art to more fully understand the invention, but do not limit the invention in any way. Therefore, although this specification has described the invention in detail with reference to the drawings and embodiments, those skilled in the art should understand that the invention can still be modified or replaced by equivalents. In short, all technical solutions and improvements that do not deviate from the spirit and scope of the invention should be included in the protection scope of the patent for the invention.

Claims

1. A method for detecting abnormal behavior of a ship anchorage, characterized in that: include: S1. Data acquisition and preprocessing: Obtain the navigation data of the ship from the database, clean, align and interpolate the navigation data according to the original time series to obtain the data to be processed; the navigation data includes basic information of the ship, longitude and latitude coordinates and navigation status of the ship; S2. Filtering the anchoring state data and segmenting: traversing the data to be processed, filtering out the data of the ship in the anchoring state, and segmenting the anchoring state data according to the time series to obtain anchoring segments; each segment of the anchoring segment represents an anchoring state period of the ship; S3. Coordinate conversion: Calculate the average latitude and longitude coordinates of the ship during an anchoring period, determine the UTM area and hemisphere of the ship in the current anchoring section based on the average latitude and longitude coordinates, and convert the latitude and longitude coordinates into UTM coordinates through a converter; S4. Clustering of anchoring status data: DBSCAN clustering is performed on the UTM coordinate data of each anchoring segment obtained based on the conversion of S3, the UTM coordinate data are grouped based on the neighborhood density to obtain clustering results, and each UTM coordinate of the anchoring segment is marked as a specific cluster or the first noise point; S5. Circle fitting and noise point identification: Use the MLESAC algorithm to fit the clustering results to establish an optimal model, calculate the optimal center and radius, and mark the points outside the fitting radius as the second noise point; S6. Classify the noise points in combination with the noise points and the circle fitting results: merge the first noise point and the second noise point into noise points, set an outlier classification rule, classify the noise points according to the outlier classification rule, determine whether the radius of the fitting circle exceeds the threshold, obtain the first type of outliers, and set them as the first priority; and further determine the anchoring state of the first type of outliers based on the noise points and the DBSCAN clustering results, whether they belong to the second type of outliers; if they belong to the second type of outliers, determine the anchoring state based on the distance between the noise points, and further classify the second type of outliers to obtain the second priority, the third priority, and the fourth priority; S7. Result output: The classification and priority classification results of the abnormal points are output, and the track abnormality classification of each anchoring section is distinguished and drawn with different colors and shapes.

2. The method for detecting abnormal behavior of a ship anchorage according to claim 1, characterized in that: The navigation data is cleaned, aligned and interpolated according to the original time series as described in S1, including: Preprocess the navigation data to ensure that the time intervals of the data points are consistent and aligned with the time axis; delete duplicate data; interpolate if there are missing values; and ensure that the data columns have the correct data type; Ship={a1,a2,...,a n } (1) Where Ship is a given list of ships, is the longitude and latitude coordinates of the i-th ship at time point t, is the navigation status of the i-th ship at time point t, where Indicates that the vessel is at anchor.

3. The method for detecting abnormal behavior of a ship anchorage according to claim 2, characterized in that: The specific steps of S2 are: S21: traverse the data to be processed and filter out The data is the anchor status data; S22: Set storage list: Set n storage lists D for storing data current , where current = 1, 2, ... n, N is the number of data points in the storage list; Formula (5) is regarded as the beginning of a new anchoring section, and current starts from 1; Among them, D current To store the list, is the navigation status of the i-th ship at time point t, is the longitude and latitude coordinates of the i-th ship at time point t; S23: Complete a mooring segment screening: set a temporary storage list N non , used to record the points in the non-anchored state. If 5 consecutive points are in the non-anchored state and the current storage list D current If there are no less than 20 points in the mooring section, the screening of one mooring section is completed, that is, if N non ≥5 and |D current |≥20,D current ←D current [1:N-5] (6) Among them, N non Stores a list of non-anchored points, D current is the storage list, N is the number of data points in the storage list; S24: Traverse all data: store current D current To D, use the next D current Continue to traverse the remaining data until the screening is completed, and finally obtain the segmented processing list of the anchor data, that is, the anchor segment set D. D={D1,D2,...,D n } (7) Where D is the mooring segment set.

4. The method for detecting abnormal behavior of a ship anchorage according to claim 3, characterized in that: The specific steps of S3 are: S31: Calculate the average longitude and latitude of each anchoring segment: traverse the anchoring segment set D and calculate the current anchoring segment D i The average value of the longitude and latitude of all longitude and latitude coordinates in , as shown in equations (8) and (9); Where N is the number of coordinate points, lon j is the longitude of point j, lat j is the latitude of point j, mean_lon is the mean longitude, and mean_lat is the mean latitude; S32: Determine the UTM region and the hemisphere to which it belongs: using the average value of the longitude and latitude, determine the UTM region by formula (10), and determine the hemisphere to which it belongs by formula (11); Among them, utm_zone is the UTM zone, and hemisphere is the hemisphere to which it belongs; S33: Coordinate conversion: After determining the UTM area and the hemisphere to which it belongs, set the projection parameters of the corresponding UTM coordinate system and use the converter to convert the current anchor segment D i All longitude and latitude coordinates in are converted to UTM coordinates, as shown in formula (12): M ′ =(x,y)=Transformer.transform(M(lon,lat)) (12) Among them, M(lon,lat) is the original longitude and latitude coordinates, M ′ is the converted UTM coordinate; S34: After traversing all anchoring segments: calculate the average latitude and longitude of all latitude and longitude coordinates of the next anchoring segment, determine the UTM region and the hemisphere to which it belongs based on the average latitude and longitude, and use a converter to convert all latitude and longitude coordinates of the current anchoring segment into UTM coordinates until the traversal is completed, and obtain each anchoring segment D i UTM coordinate data.

5. The method for detecting abnormal behavior of a ship anchorage according to claim 4, characterized in that: The specific steps of S4 include: S41: DBSCAN clustering of UTM coordinate data: DBSCAN clustering is performed on the UTM coordinate data of each anchoring segment, and the neighborhood radius parameter eps and the minimum sample number min_sample are set, as shown in formula (13): db=DBSCAN(eps=100,min_sample=8).fit(M ′ ) (13) Among them, eps is the neighborhood radius parameter, min_sample is the minimum number of samples; S42: Determine the core point: If there is a point P, and the number of UTM coordinates in the neighborhood with a radius of eps and centered at point P is greater than or equal to the minimum number of samples min_sample, then point P is a core point; as shown in formula (14): P = core point, if N eps (P)≥min_sample (14) Among them, eps is the neighborhood radius parameter, min_sample is the minimum number of samples, N eps (P) is the number of points in the neighborhood; S43: Coordinate point judgment: other points in the neighborhood of the core point are boundary points and stored as a specific cluster. Points outside the neighborhood are marked as the first noise points, stored as the first noise points, and marked with cluster labels, as shown in formula (15-16). L={L 1 ,L 2 ,...,L n },M′(unique_clusters i )∈L i (15) unique_clusters i ={c∈db.labels i |c≠-1} H={H1,H2,...,H n },cluster(H i )=-1 (16) Among them, L is the set of specific clusters, H is the set of the first noise points, unique_clusters i is the cluster label set, M ′ (unique_clusters i ) is a coordinate array marked with cluster labels, cluster(H i ) is the cluster label of the noise point.

6. The method for detecting abnormal behavior of a ship anchorage according to claim 5, characterized in that: The specific steps of S5 are: S51: traverse the specific cluster set L; S52: Determine the center and radius of the circle: From the current specific cluster L i In the UTM coordinates, randomly select at least three sample points to calculate the parameters of the circle, namely the center (xc, yc) and the radius r; S53: Calculate the current specific cluster L i All other points (x i ,y i ) to the center of the circle, as shown in formula (17), Among them, R i is the distance from point i to the center of the circle, (x i ,y i ) is any point other than the center of the circle, (xc,yc) is the center of the circle; S54: Calculate residual value: Calculate the residual value of the distance from all other points to the center of the circle and the radius r, as shown in formula (18), residual i =|r i -r| (18) Among them, residual i is the residual value, r i is the distance from point i to the center of the circle, and r is the radius; S55: Determine the coordinate point: set a residual value threshold residual_threshold, and among all other points, the points whose residual values ​​are less than the threshold are marked as inner points, otherwise they are marked as outer points; S56: Determine the best fitting circle and mark the second noise point: Repeat steps S52-S55 until the preset number of trials is reached, select the model with the most inner points among all model results, and output the best fitting radius R and the center (XC, YC) in the model, mark the outer point as the second noise point, and store it as G i , G i =(x i ,y i ),if residual i ≥residual_threshold (19) Among them, residual i is the residual value, residual_threshold is the threshold of the residual value; S57: Execute the next specific cluster until the traversal is completed, and obtain the set G of the second noise point: G={G1,G2,...,G n } (20) Wherein, G is the set of second noise points.

7. The method for detecting abnormal behavior of a ship anchorage according to claim 6, characterized in that: S6 specifically includes: S61: Merge noise points: store the first noise points and the second noise points as a noise point set F, as shown in formula (21): F=G∪H (21) Where F is the set of noise points, G is the set of second noise points, and H is the set of first noise points; S62: Setting a first threshold, and traversing the noise point set F; S63: First priority classification: for the specific cluster L to which the current noise point belongs i In the above example, if the radius of a fitted circle is greater than the first threshold, then this part of the noise points does not conform to the normal anchoring form and is marked as the first type of abnormal points; If there is a noise point F j , from the center of the fitting circle to which it belongs (XC i ,YC i ) is greater than 1.2 times the optimal fit radius R i , then the noise point is marked as the second type of outlier point I; as shown in formula (22), Among them, (x j ,y j ) is the coordinate of noise point j, (XC i ,YC i ) is the center of the fitting circle, R i is the optimal fitting radius; S64: Second priority classification: For the second type of abnormal point I, if there is a noise point located at the starting point of the specific cluster to which it belongs, then the noise point and the continuous noise points retrieved backward are marked as entering anchoring; if there is a noise point located at the ending point of the specific cluster to which it belongs, then the noise point and the continuous noise points retrieved forward are marked as ending anchoring or changing behavior after moving anchor; S65: Third priority classification: For the second type of abnormal point I, if there is a single noise point I a Or two noise points I a ,I b , the average distance from two adjacent normal points M and N is greater than 5 times the distance between M and N, then the noise point deviates greatly from the specific cluster to which it belongs, which is marked as data drift, that is, Among them, I a ,I b are two noise points, M, N are I a ,I b The points before and after, I a M.I a N.I. a M.I b N and MN both represent the distance between two corresponding points; S66: Fourth priority classification: For the second type of abnormal point I, if a single noise point or two normal data points before and after a continuous noise point segment belong to different specific clusters, then this part of the noise points is marked as anchored and re-anchored, as shown in formula (24-25). Among them, (x m-1 ,y m-1 ) is a specific cluster L a The coordinates of point m-1 in the middle, R a For a specific cluster L a The fitting radius, (XC a ,YC a ) is the center of the fitting circle, (x n+1 ,y n+1 ) is a specific cluster L b The coordinates of point n+1, R b For a specific cluster L b The fitting radius, (XC b ,YC b ) is the center of the fitting circle, M ′ m-1 、M ′ n+1 are two normal data points before and after the noise point paragraph; S67: Noise points that do not conform to the above classification are uniformly marked as other anchoring states.

8. A ship anchorage abnormal behavior detection system using the method according to any one of claims 1 to 7, comprising: Data acquisition and processing module, data screening and segmentation module, coordinate conversion module, clustering module, fitting module, noise classification module and result output module; The data acquisition and processing module includes: a data acquisition submodule for acquiring the navigation data of the ship from the database; a data preprocessing submodule for cleaning, aligning and interpolating the navigation data according to the original time series to obtain the data to be processed; the navigation data includes basic information of the ship, longitude and latitude coordinates and navigation status of the ship; The data screening and segmentation module includes: a screening submodule for traversing the data to be processed and screening out the data of the ship in the anchored state, and a segmentation submodule for segmenting the anchored state data according to the time series to obtain the anchored segments; each segment of the anchored segment represents an anchored state period of the ship; The coordinate conversion module is used to calculate the average value of the longitude and latitude coordinates of the ship during an anchoring state period, determine the UTM area and the hemisphere to which the ship belongs in the current anchoring section based on the average value of the longitude and latitude coordinates, and convert the longitude and latitude coordinates into UTM coordinates through a converter; The clustering module is used to perform DBSCAN clustering on the UTM coordinate data of each anchoring segment obtained by the coordinate conversion module, group the UTM coordinate data based on neighborhood density to obtain clustering results, and mark each UTM coordinate of the anchoring segment as a specific cluster or a first noise point; The fitting module uses the MLESAC algorithm to fit the clustering results to establish an optimal model, calculate the optimal center and radius, and mark the points outside the fitting radius as second noise points; The noise point classification module includes: merging the first noise point and the second noise point as noise points, and setting a classification rule submodule for abnormal point classification rules, wherein the abnormal point classification rules include: classifying the noise points according to the abnormal point classification rules, judging whether the radius of the fitting circle exceeds a threshold, obtaining a first type of abnormal point, and setting it as a first priority; and further judging the anchoring state of the first type of abnormal point based on the noise point and the DBSCAN clustering result, whether it belongs to the second type of abnormal point; if it belongs to the second type of abnormal point, then judging the anchoring state based on the distance between the noise points, further classifying the second type of abnormal point, and obtaining a second priority, a third priority, and a fourth priority; The result output module includes: a result output submodule for outputting the classification and priority classification results of the abnormal points, and a trajectory abnormality classification display submodule for distinguishing and drawing the trajectory abnormality classification of each anchoring section with different colors and shapes.

Images