A method for detecting abnormal behavior of a ship based on AIS big data

Abstract

The application discloses a ship abnormal behavior detection method based on AIS big data and relates to the technical field of ship abnormal detection, and comprises the following steps: using a direction sequence to perform point-by-point direction comparison on a position change sequence and a heading change sequence, judging whether the position change sequence and the heading change sequence change in opposite directions and are offset in a fusion process, marking a section meeting the opposite change, and generating a reverse coupling indicator; performing amplitude and direction coupling operation on the position change sequence and the heading change sequence in the marked section according to the reverse coupling indicator, forming an offset strength parameter, and determining a characteristic offset degree according to the offset strength parameter. The application solves the problem that abnormal information is covered when the position change feature and the heading change feature change in opposite directions and are offset, realizes weight adaptive adjustment based on the offset degree, and thus improves the accuracy and reliability of abnormal behavior detection.

Description

Technical Field

[0001] This invention relates to the field of ship anomaly detection technology, specifically to a method for detecting abnormal ship behavior based on AIS big data. Background Technology

[0002] Ship anomaly behavior detection based on AIS big data refers to a type of information processing method that uses the massive spatiotemporal dynamic data continuously broadcast by the Automatic Identification System (AIS) to analyze the motion state and behavior patterns of ships during navigation through data processing and behavior modeling techniques, thereby identifying abnormal behaviors that deviate from normal navigation patterns. Its core lies in transforming raw AIS data (including ship position, speed, heading, timestamps, and identity information, etc.) into structured trajectory data that can be used for analysis, constructing a "normal behavior model" based on historical behavior or established rules, and then achieving anomaly identification through comparative analysis of real-time or offline data. Existing technologies typically begin by collecting AIS data via shore-based base stations or satellite systems and preprocessing it, including data cleaning (removing noise and erroneous data), missing data completion, and trajectory reconstruction. Subsequently, feature extraction is performed on the processed trajectory data to construct multi-dimensional behavioral features such as speed change rate, steering angle, dwell time, and course deviation. Based on this, rule-based methods (such as threshold judgment), statistical analysis methods, or machine learning models are used to establish normal ship behavior patterns. Finally, the real-time or historical behavioral characteristics of the ship to be detected are input into the aforementioned model for evaluation. Anomaly scoring or deviation degree is used to determine whether abnormal behavior exists, and alarm information is output. The entire process typically includes key steps such as data acquisition, data preprocessing, feature construction, behavioral modeling, and anomaly detection, thereby achieving automated and large-scale detection of abnormal ship behavior.

[0003] The existing technology has the following shortcomings: In the process of detecting abnormal ship behavior based on AIS big data, when the behavioral features calculated from position changes and those calculated from heading changes in AIS data show opposite trends in numerical change and cancel each other out during feature fusion, the overall behavioral feature value may be weakened or even stabilized. This is because, under certain complex maneuvering conditions, changes in spatial position and heading adjustments exhibit opposite directions in mathematical expression. Existing methods typically use linear weighting or averaging for feature fusion, without modeling the interaction between features, resulting in direct superposition and feature cancellation during fusion. Due to this cancellation effect, features that could originally reflect abnormal behavior are masked. Existing technologies cannot adjust the abnormal behavior judgment weights based on the degree of feature cancellation when position change features and heading change features in AIS data show opposite trends and cancel each other out during fusion. This results in abnormal behavior features not being able to effectively participate in the judgment process, leading to abnormal behavior being misjudged as normal behavior or not being identified at all, ultimately affecting the accuracy and reliability of the abnormal ship behavior detection results.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for detecting abnormal ship behavior based on AIS big data, so as to solve the problems in the background art mentioned above.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting abnormal ship behavior based on AIS big data, specifically including the following steps: S1. Simultaneously extract position change information and heading change information from the ship's AIS data, construct position change sequence and heading change sequence, and introduce a unified behavior domain benchmark for numerical adjustment to generate a direction sequence; S2. Use the direction sequence to perform point-by-point direction comparison between the position change sequence and the heading change sequence, determine whether the position change sequence and the heading change sequence are opposite and cancel each other out during the fusion process, and calibrate the sections that meet the opposite change to generate a reverse coupling indicator. S3. Perform amplitude and direction coupling operations on the position change sequence and heading change sequence within the calibration section according to the reverse coupling indication to form the cancellation strength parameter, determine the characteristic cancellation degree based on the cancellation strength parameter, and construct the cancellation compensation sequence to retain the cancelled change information. S4. Based on the degree of feature cancellation and the cancellation compensation sequence, the position change sequence and heading change sequence are weighted and compensated and superimposed to form a reconstructed feature sequence. The reconstructed feature sequence is then coupled to generate the abnormal evaluation coefficient corresponding to the abnormal behavior of the ship. S5. Combine the feature cancellation degree and the anomaly evaluation coefficient to construct a weight evolution sequence, dynamically update the weights, and continuously correct the coupling operation of the feature sequence accordingly.

[0007] Preferably, S1 is as follows: Ship AIS data is sorted in chronological order, and position change information and heading change information corresponding to each time point are extracted. The position change information and heading change information are synchronously matched using a time index. A position change sequence is constructed by calculating the position difference between adjacent time points, and a heading change sequence is constructed by calculating the heading difference between adjacent time points. A unified behavioral domain benchmark is introduced, and numerical mapping processing is performed on the position change sequence and the heading change sequence. The corresponding values ​​of the position change sequence and the heading change sequence are limited to a unified numerical range. The numerical data is then normalized by scaling and range transformation to form the normalized position change sequence and heading change sequence. For the adjusted position change sequence and heading change sequence, the positive and negative attributes of the corresponding changes at each time point are extracted, and the positive and negative attributes are converted into direction indicators. The direction indicators are then arranged in chronological order to generate a direction sequence.

[0008] Preferably, S2 specifically includes the following steps: S201. Use the direction sequence to perform point-by-point direction comparison between the position change sequence and the heading change sequence. Extract the direction identifiers in the direction sequence corresponding to each time point in chronological order, and compare the direction identifiers corresponding to the position change sequence with the direction identifiers corresponding to the heading change sequence one by one to form a direction comparison result sequence. S202. Based on the direction comparison result sequence, determine whether the position change sequence and the heading change sequence are opposite and cancel each other out during the fusion process. Filter the time points in the direction comparison result sequence where the direction identifiers have opposite values, and synchronously compare the position change sequence values ​​and heading change sequence values ​​corresponding to the filtered time points. When the position change sequence values ​​and heading change sequence values ​​are opposite in their numerical change trends and the corresponding values ​​have a mutual reduction relationship in the superposition calculation, it is determined that they cancel each other out during the fusion process, thus forming a set of time points where they are opposite and cancel each other out during the fusion process. S203. For the set of time points that change in opposite directions and cancel each other out during the fusion process, the time continuity is used to divide the time into segments and determine the segment boundaries. Each segment is calibrated and a reverse coupling indicator is constructed based on the correspondence between the direction identifier and the position change sequence and the heading change sequence within each segment. The reverse coupling indicator sequence is then arranged in chronological order.

[0009] Preferably, S203 is as follows: For the set of time points that show opposite changes and cancel each other out during the fusion process, the time points are traversed in the order of time index, the time interval between adjacent time points is calculated, the time interval is compared with a fixed time interval benchmark, when the time interval is equal to the fixed time interval benchmark, it is assigned to the same segment, when the time interval is greater than the fixed time interval benchmark, it is used as the segment division position, thereby completing the segment division and determining the boundary of each segment. After the segment division is completed and the segment boundaries are determined, each segment is calibrated. Segment identifiers are constructed based on the segment start time index and end time index. The corresponding direction identifiers, position change sequence values ​​and heading change sequence values ​​are extracted in each segment to form a set of corresponding relationships within the segment. Based on the set of correspondences within each segment, a coupling association operation is performed on the correspondence between the direction indicator and the position change sequence and the heading change sequence. The value state of the direction indicator is synchronously mapped with the value of the position change sequence and the value of the heading change sequence to form a segment-level reverse coupling indicator. The reverse coupling indicators corresponding to each segment are arranged in chronological order to form a reverse coupling indicator sequence.

[0010] Preferably, S3 specifically includes the following steps: S301. Extract the reverse coupling indicator, position change sequence value and heading change sequence value corresponding to each time point in the calibration section according to the time sequence. Adjust the direction of the position change sequence value and heading change sequence value according to the value state of the reverse coupling indicator. Perform amplitude matching processing on the adjusted value to form the amplitude and direction coupling calculation result of the corresponding time point. S302. Based on the distribution of amplitude and direction coupling calculation results within the calibration section, the amplitude and direction coupling calculation results at each time point are continuously accumulated. The accumulated results are divided into segments according to the variation range within the calibration section. The accumulated results corresponding to each segment are used as the cancellation strength parameter. The characteristic cancellation degree is determined in combination with the corresponding range of each segment. S303. Combine the degree of feature cancellation to perform compensation and reconstruction processing on the position change sequence and heading change sequence within the calibration section. Extract the reduction part corresponding to each time point from the amplitude and direction coupling calculation results, and perform reverse recovery processing on the reduction part according to the cancellation intensity parameter. The recovered change amount is superimposed on the position change sequence and heading change sequence respectively to form a cancellation compensation sequence.

[0011] Preferably, S302 is as follows: Read the amplitude and direction coupling calculation results corresponding to each time point in the calibration section in chronological order. Take the amplitude and direction coupling calculation result of the first time point as the starting accumulated value, and then merge the amplitude and direction coupling calculation results of subsequent time points into the starting accumulated value to form the continuous accumulated result corresponding to each time point. The intervals are divided around the change range of the continuous accumulation result within the calibration section. The starting value and ending value of the interval are extracted according to the numerical change order of the continuous accumulation result. The continuous accumulation result is then segmented and assigned according to the starting value and ending value of the interval, forming multiple segments corresponding to the accumulation result range. The continuous cumulative results within the range corresponding to each segment are aggregated into segments, and the aggregated results of each segment are used as the offset strength parameter. According to the correspondence between the range of each segment and the offset strength parameter, the offset strength parameter within each segment is sorted. Based on the sorting position, the offset strength parameter is divided into three continuous intervals, where the offset strength parameter in the first interval corresponds to weak offset, the offset strength parameter in the middle interval corresponds to medium offset, and the offset strength parameter in the last interval corresponds to strong offset, thus completing the determination of the feature offset degree.

[0012] Preferably, S4 specifically includes the following steps: S401. Based on the degree of feature cancellation and the cancellation compensation sequence, the position change sequence and the heading change sequence are weighted and compensated and superimposed. The degree of feature cancellation corresponding to each time point is extracted in time order, and the position change sequence and the heading change sequence are assigned corresponding weights according to the degree of feature cancellation. At the same time, the cancellation compensation sequence is superimposed on the position change sequence and the heading change sequence according to the time point to form the weighted sequence data. S402. Based on the sequence data after weight redistribution, the position change sequence and heading change sequence corresponding to the same time point are combined and processed accordingly. The combination results of each time point are arranged in chronological order to form the reconstructed feature sequence. S403. Perform coupling operation processing on the reconstructed feature sequence, perform numerical correlation operation on the reconstructed feature sequence at each time point in chronological order, and perform intra-segment integration processing on the coupling operation results of all time points. Use the integrated numerical results as the anomaly assessment coefficient corresponding to the abnormal behavior of the ship.

[0013] Preferably, S403 is as follows: Extract the reconstructed feature sequences corresponding to each time point in chronological order, perform numerical pairing on the position change sequence values ​​and heading change sequence values ​​within the same time point, and use the paired numerical combinations within each time point as the input for numerical association operations. Numerical correlation operations are performed around the paired numerical combinations at each time point. The numerical values ​​of the position change sequence and the heading change sequence are coupled point by point to form the coupling operation results corresponding to each time point, and the coupling operation results are arranged in chronological order to form a set of coupling operation results. The coupling operation results set is integrated within a segment. The coupling operation results at each time point are assigned to a segment according to the segment boundary. The coupling operation results within each segment are continuously aggregated. The aggregated results corresponding to each segment are sequentially summarized to form the anomaly assessment coefficient corresponding to the abnormal behavior of the ship.

[0014] Preferably, S5 is as follows: Using the time index as the correlation benchmark, the feature cancellation degree corresponding to each time point is paired with the anomaly evaluation coefficient. A numerical correlation relationship between the feature cancellation degree and the anomaly evaluation coefficient is established in the pairing results. Based on this numerical correlation relationship, the weight value corresponding to each time point is determined. The weight values ​​of all time points are arranged in chronological order to construct a weight evolution sequence. By focusing on the weight changes in the weight evolution sequence, the weights of the position change sequence and the heading change sequence are updated point by point, so that the weight values ​​at each time point participate in the numerical correlation process of the reconstructed feature sequence. Furthermore, the weight change trend of the previous time point is introduced into the subsequent time point, and the reconstructed feature sequence is subjected to progressive correction coupling operation processing to achieve continuous correction of the feature sequence coupling operation.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention, through simultaneous extraction, unified organization, direction comparison, reverse coupling identification, cancellation strength quantification, compensation reconstruction, weight reallocation, and dynamic evolution update of position and heading change information from ship AIS data, can accurately identify feature cancellation relationships when position and heading change features exhibit opposite changes and cancel each other out during the fusion process. It further determines the degree of feature cancellation and reintroduces the canceled change information into subsequent calculations via a cancellation compensation sequence. Simultaneously, it adjusts the weights of the position and heading change sequences based on the degree of feature cancellation, thus avoiding the problems of weakened, masked, or even excluded abnormal features due to linear weighting or average fusion in existing technologies. Furthermore, by constructing anomaly evaluation coefficients and weight evolution sequences, the abnormal behavior determination not only reflects the feature coupling state at the current time point but also continuously corrects the coupling calculation process based on the changing trends of previous and subsequent time points. Therefore, it improves the ability to identify complex maneuvering states, concealed abnormal states, and feature cancellation states during ship abnormal behavior detection, thereby enhancing the accuracy, stability, and continuity of abnormal behavior detection results. It also strengthens the adaptability of the technical solution to multi-dimensional behavioral feature fusion analysis in AIS big data scenarios. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0018] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0019] This invention provides, for example Figure 1 The method for detecting abnormal ship behavior based on AIS big data, as shown, specifically includes the following steps: S1. Simultaneously extract position change information and heading change information from the ship's AIS data, construct position change sequence and heading change sequence, and introduce a unified behavior domain benchmark for numerical adjustment to generate a direction sequence; In this embodiment, S1 specifically refers to: Ship AIS data is sorted in chronological order, and position change information and heading change information corresponding to each time point are extracted. The position change information and heading change information are synchronously matched using a time index. A position change sequence is constructed by calculating the position difference between adjacent time points, and a heading change sequence is constructed by calculating the heading difference between adjacent time points. When processing ship AIS data, the original data records are first sorted in ascending order according to the time field, ensuring that each data entry has a clear temporal sequence. Then, the geographical location parameters and heading parameters for each corresponding moment are extracted from the sorted data. The geographical location parameters can be represented by a combination of longitude and latitude, while the heading parameters reflect changes in the ship's orientation. Based on this, a time index is constructed using chronological numbering, and the time index is used to match the position and heading information at the same moment, thus achieving synchronization. After matching, for data from two consecutive time points, the change in position is calculated. This can be achieved by combining the differences in longitude and latitude between adjacent moments to calculate the displacement. The change in position, for example, if the position is point A at a certain moment and point B at the next moment, the difference between the two points represents the magnitude of spatial movement. At the same time, the difference in heading angles at adjacent time points is calculated to form the heading change, for example, if the heading changes from one angle to another, the angle difference reflects the degree of turning. By performing this process sequentially on all adjacent time points, the continuously calculated position changes are arranged in chronological order to form a position change sequence, and the corresponding heading changes are arranged in the same order to form a heading change sequence. This provides a continuous change basis for subsequent behavioral analysis. The reason for this approach is that it transforms discrete state information into continuous change characteristics, which is beneficial for characterizing the ship's motion trend and behavioral dynamics.

[0020] Ship AIS data refers to a dynamic set of navigation data generated through periodic broadcasts from automatic identification equipment, including time information, position coordinates, and heading angle. Position change information refers to changes in geographical location between adjacent time points, typically reflected by changes in longitude and latitude to indicate spatial displacement. Heading change information refers to changes in the ship's direction of travel between consecutive time points, used to characterize turning behavior. A time index is a sequence marker that identifies data records according to chronological order, used to establish correspondences between data. Synchronous matching refers to using the same time index to create a one-to-one correspondence between position information and heading information at the same time point, ensuring data consistency. The position difference between adjacent time points refers to the amount of spatial position change between two consecutive time points, used to describe the magnitude of movement. A position change sequence is a set of consecutive position differences arranged in chronological order, used to reflect the trend of trajectory changes. The heading difference between adjacent time points refers to the amount of change in heading angle between two consecutive time points, used to describe the magnitude of turning. A heading change sequence is a set of heading differences arranged in chronological order, used to reflect the process of direction change. Through the coordinated construction of these elements, the description of ship motion can be transformed from a static description to a dynamic description.

[0021] A unified behavioral domain benchmark is introduced, and numerical mapping processing is performed on the position change sequence and the heading change sequence. The corresponding values ​​of the position change sequence and the heading change sequence are limited to a unified numerical range. The numerical data is then normalized by scaling and range transformation to form the normalized position change sequence and heading change sequence. When processing position change sequences and heading change sequences in a unified manner, a unified behavioral domain benchmark can be constructed first to describe the value range of the two types of changes under the same reference system. For example, the overall distribution range of position change and heading change can be statistically analyzed, and common upper and lower boundaries can be selected as mapping references. Subsequently, numerical mapping processing is performed on the position change sequences and heading change sequences, mapping the original values ​​to a unified numerical range according to this reference relationship. For example, position change and heading change, which originally had large differences in value range, can be compressed into the same range through linear scaling. During the mapping process, the numerical values ​​are scaled proportionally. Amplitude adjustments are made to make data of different dimensions comparable. At the same time, interval transformations are combined to reposition the original distribution so that the two types of sequences express the trend of change within the same scale. For example, larger positional changes are compressed to a smaller range, while smaller heading changes are expanded. By continuously performing scaling and interval transformations, the positional change sequences and heading change sequences are numerically normalized to form normalized positional change sequences and heading change sequences at a unified scale. The reason for this treatment is to eliminate the influence of different dimensions and numerical amplitudes, so that subsequent comparisons and coupling analyses of the trend of change have a consistent numerical basis.

[0022] A unified behavioral domain benchmark is a reference standard used to uniformly scale position change sequences and heading change sequences. Its purpose is to eliminate differences in units and numerical ranges between different types of behavioral characteristics, enabling position change and heading change information to be compared, integrated, and coupled for analysis on the same numerical basis. The unified behavioral domain benchmark can be established based on the overall distribution of position and heading changes in historical AIS data, mapping the corresponding values ​​of position and heading change sequences to the same numerical range, allowing behavioral characteristics from different sources to express change trends at a unified scale. The unified behavioral domain benchmark is not a specific spatial or geographical region, but rather a data reference standard used to describe the unified expression range of behavioral characteristics. Its role is to ensure that position and heading change sequences have a consistent data representation basis in subsequent direction comparisons, cancellation determinations, and coupling operations.

[0023] A unified behavioral domain benchmark refers to a unified reference standard used to constrain the value range of different types of change data, ensuring that position changes and heading changes are expressed in the same numerical system. Numerical mapping processing refers to the process of transforming the original data to a target numerical range according to a preset mapping relationship based on the unified behavioral domain benchmark. A unified numerical range refers to the common value range of the position change sequence and heading change sequence after mapping, used to achieve comparability between different types of data. Scaling and interval transformation are key means to achieve mapping processing, where scaling is used to adjust the numerical amplitude and interval transformation is used to adjust the numerical distribution position, so that the data as a whole falls into a unified range. Numerical normalization refers to converting data from different sources into a data set with a unified scale and expression form through the above processing. The normalized position change sequence and heading change sequence refer to the sequence results after completing the unified numerical interval mapping, which can reflect their respective change trends under the same numerical standard, providing a consistent data foundation for subsequent directional relationship judgment and feature fusion.

[0024] For the adjusted position change sequence and heading change sequence, the positive and negative attributes of the corresponding changes at each time point are extracted, and the positive and negative attributes are converted into direction indicators. The direction indicators are then arranged in chronological order to generate a direction sequence.

[0025] After numerical normalization of the position change sequence and heading change sequence, the sign of the change at each time point can be determined, i.e., whether the value is positive or negative, thereby extracting the trend information. In practice, the normalized sequence data can be read point by point, and the sign of the change at each time point can be determined. When the value is greater than zero, it is determined to be a positive change, and when the value is less than zero, it is determined to be a negative change. Different sign states are converted into pre-defined direction labels. For example, positive change is mapped to one direction label, and negative change is mapped to another direction label. Then, the direction labels corresponding to all time points are arranged in chronological order to form a continuous direction sequence. For example, if the position change sequence shows continuous positive change and the heading change sequence shows continuous negative change within a certain period, a set of direction label sequences is generated to represent this trend difference. Through this processing, the original numerical change can be transformed into a discrete representation at the direction level, thereby highlighting the consistency or opposition of the change trend and providing a direct basis for subsequent identification of opposite change relationships.

[0026] The positive or negative attribute of the change at each time point refers to the sign characteristic of the change value at each time point in the normalized sequence, used to reflect the direction of change rather than the magnitude of change; the direction identifier is a symbolic mark formed by discretizing the positive and negative attributes, used to uniformly describe the direction of change, for example, a positive change corresponds to one identifier, and a negative change corresponds to another identifier; the direction sequence is a set of direction identifiers arranged in chronological order, used to describe the continuous evolution of the direction of change over time; by converting continuous numerical changes into direction identifiers and serializing them, the relationship between change trends can be directly characterized without relying on the specific numerical magnitude, thus providing a clear data expression basis for subsequent judgment on whether different sequences show opposite changes.

[0027] S2. Use the direction sequence to perform point-by-point direction comparison between the position change sequence and the heading change sequence, determine whether the position change sequence and the heading change sequence are opposite and cancel each other out during the fusion process, and calibrate the sections that meet the opposite change to generate a reverse coupling indicator. In this embodiment, S2 specifically includes the following steps: S201. Use the direction sequence to perform point-by-point direction comparison between the position change sequence and the heading change sequence. Extract the direction identifiers in the direction sequence corresponding to each time point in chronological order, and compare the direction identifiers corresponding to the position change sequence with the direction identifiers corresponding to the heading change sequence one by one to form a direction comparison result sequence. When performing point-by-point directional comparisons between position change sequences and heading change sequences using directional sequences, the directional identifiers of the position change sequence and heading change sequence at each time point can be read sequentially. The two directional identifiers at the same time point are then placed in the same comparison unit for synchronous comparison. Specifically, a sequential traversal approach can be used, extracting the directional identifier at each time point in turn. If the position change sequence corresponds to a positive identifier at a certain time point, while the heading change sequence corresponds to a negative identifier at the same time point, then that time point is recorded as a comparison result with opposite directions. If the two directional identifiers are the same, then it is recorded as a comparison result with the same directions. For example, in three consecutive time points, if the directional identifiers corresponding to the position change sequence are positive, positive, and negative, and the directional identifiers corresponding to the heading change sequence are negative, positive, and positive, then the directional comparison results at the three time points can be recorded as opposite, consistent, and opposite, respectively. Through this one-to-one point-by-point comparison, the directional information originally scattered in the two sequences can be converted into a unified directional relationship sequence, facilitating subsequent determination of whether the position change sequence and heading change sequence exhibit continuous opposite changes and whether they meet the prerequisite for cancellation.

[0028] Point-by-point direction comparison refers to the process of comparing the direction attributes of position change sequences and heading change sequences at the same time position, using a single time point as the comparison unit. The core is to maintain consistency between the comparison granularity and the time granularity. The direction identifier in the direction sequence corresponding to each time point is a discrete marker used to represent the direction of change at each time point, usually used to distinguish between positive and negative changes. One-to-one correspondence comparison means that the direction identifiers in the position change sequence and the direction identifiers in the heading change sequence must be compared in pairs under the same time index. Cross-time point comparisons are not allowed, otherwise the time synchronization relationship will be disrupted. The direction comparison result sequence is the set of results after arranging the direction relationship results formed after each time point comparison in chronological order. It is used to record whether the position change sequence and the heading change sequence are in the same direction or opposite direction over the entire time range. By forming the direction comparison result sequence, the correspondence between the two original direction sequences can be further structured, providing direct input for subsequent screening of opposite change time points, division of opposite change segments, and construction of reverse coupling indicators.

[0029] S202. Based on the direction comparison result sequence, determine whether the position change sequence and the heading change sequence are opposite and cancel each other out during the fusion process. Filter the time points in the direction comparison result sequence where the direction identifiers have opposite values, and synchronously compare the position change sequence values ​​and heading change sequence values ​​corresponding to the filtered time points. When the position change sequence values ​​and heading change sequence values ​​are opposite in their numerical change trends and the corresponding values ​​have a mutual reduction relationship in the superposition calculation, it is determined that they cancel each other out during the fusion process, thus forming a set of time points where they are opposite and cancel each other out during the fusion process. When determining whether the position change sequence and the heading change sequence show opposite changes and cancel each other out during the fusion process, we can first filter out the time points in the direction comparison result sequence where the direction indicators have opposite values. That is, the time points at the same time position where one sequence shows a positive change and the other shows a negative change. After completing the filtering, we then synchronously compare the position change sequence values ​​and heading change sequence values ​​corresponding to these time points. Specifically, we can read the value changes of the two sequences point by point according to the same time index, first determine whether the changes of the two values ​​are opposite, and then further determine whether the two values ​​cancel each other out during the fusion calculation. The question is whether mutual cancellation will occur. For example, if the position change sequence value shows an upward change while the heading change sequence value shows a downward change, and after the two are merged, the increment of one value in the overall result will be canceled out by the decrement of the other value. In this case, the time point can be considered to simultaneously satisfy both the conditions of opposite changes and cancellation. Similarly, if in multiple consecutive time points the position change sequence values ​​increase sequentially while the heading change sequence values ​​decrease sequentially, and the overall value amplitude is significantly reduced after being superimposed at the corresponding time points, these time points can be categorized as time points showing opposite changes and cancellation during the fusion process. The reason for this processing order is that simply relying on directional indicators only shows that the two sequences are in opposite directions, but it cannot directly indicate that cancellation will definitely occur during fusion. Only by combining the position change sequence values ​​with the heading change sequence values ​​synchronously and further analyzing the reduction relationship after superposition can we distinguish between opposite directions and true cancellation.

[0030] The time points where the direction identifiers have opposite values ​​refer to the time points where, under the same time index, the direction identifiers corresponding to the position change sequence and the direction identifiers corresponding to the heading change sequence represent opposite directions. These time points are candidate objects for subsequent judgment. The position change sequence values ​​and heading change sequence values ​​refer to the numerical content used to characterize the magnitude of spatial position change and heading change at each time point, respectively, to reflect the specific change state of the two sequences at the same moment. Synchronous comparison refers to strictly comparing the values ​​of the two sequences at the same time point according to the same time index, rather than comparing across time points. The opposite trend in numerical change refers to one sequence showing an increasing trend or a positive trend in consecutive time points, while the other sequence shows a decreasing trend or a reverse trend at the corresponding time point. The mutual reduction relationship between corresponding values ​​in the superposition calculation means that after the two sequence values ​​enter the same fusion calculation, the boosting effect of one sequence on the comprehensive result is partially offset by the suppressing effect of the other sequence on the comprehensive result, thereby weakening the comprehensive change magnitude. The time point set refers to the data set formed by collecting all time points that simultaneously meet the opposite direction and the cancellation judgment condition during the fusion process in chronological order, which is used for subsequent segment division and construction of reverse coupling indicators.

[0031] S203. For the set of time points that change in opposite directions and cancel each other out during the fusion process, the time continuity is used to divide the time into segments and determine the segment boundaries. Each segment is calibrated and a reverse coupling indicator is constructed based on the correspondence between the direction identifier and the position change sequence and the heading change sequence within each segment. The reverse coupling indicator sequence is then arranged in chronological order.

[0032] In this embodiment, S203 specifically refers to: For the set of time points that show opposite changes and cancel each other out during the fusion process, the time points are traversed in the order of time index, the time interval between adjacent time points is calculated, the time interval is compared with a fixed time interval benchmark, when the time interval is equal to the fixed time interval benchmark, it is assigned to the same segment, when the time interval is greater than the fixed time interval benchmark, it is used as the segment division position, thereby completing the segment division and determining the boundary of each segment. When dividing a set of time points that exhibit opposite changes and cancel each other out during the fusion process into segments, one can first iterate through all time points in time index order, pairing each time point with its next adjacent time point and reading the time index values ​​corresponding to the two time points. The time interval between adjacent time points is determined by the difference between the two time index values. Then, a fixed time interval benchmark is introduced, comparing each time interval with the fixed time interval benchmark. When the time interval between a certain time point and the next time point matches the fixed time interval benchmark, it indicates that the two time points maintain a continuous relationship in time and should be assigned to the same segment. When the time interval between a certain time point and the next time point is greater than the fixed time interval benchmark, it indicates that the temporal continuity is broken, and this is determined as the segment division position. The previous time point is used as the end boundary of the current segment, and the next time point is used as the start boundary of the next segment. For example, if a group of time points, arranged by time index, corresponds to consecutive... When sampling time, if the first few time intervals are consistent with the fixed time interval benchmark, these time points are classified as the same segment. When a subsequent time interval is significantly longer than the fixed time interval benchmark, a new segment is created at that point. This allows the original set of time points to be divided into multiple continuous segments. Traversing according to the time index order means strictly checking the connection status between time points according to their chronological order. The time interval between adjacent time points refers to the interval length between two consecutive time points on the time axis. The fixed time interval benchmark is a unified reference value used to determine the continuity of time. The same segment refers to a set of time points that are continuous in time and have not experienced any interval jumps. The segment division position refers to the division point corresponding to the interruption of time continuity. The boundary of each segment refers to the start and end positions of each continuous set of time points. Through this process, discrete cancellation time points can be organized into continuous time periods, which facilitates the independent calibration and analysis of each continuous cancellation behavior.

[0033] After the segment division is completed and the segment boundaries are determined, each segment is calibrated. Segment identifiers are constructed based on the segment start time index and end time index. The corresponding direction identifiers, position change sequence values ​​and heading change sequence values ​​are extracted in each segment to form a set of corresponding relationships within the segment. When calibrating each segment after segment division and boundary determination, the start and end positions of each segment on the timeline can be read sequentially. Segment identifiers can be constructed using the segment start and end time indices as boundary information. For example, the start and end time indices of a segment can be combined in a fixed order to form a unique identifier for that segment, distinguishing different segments. After constructing the segment identifiers, the direction identifier, position change sequence value, and heading change sequence value consistent with the segment's time range can be extracted point-by-point within the segment. Using the same time index as the connection basis, the direction identifier, position change sequence value, and heading change sequence value at the same time point are bound to form a set of correspondences within the segment. For example, if a segment covers a set of consecutive time points, one direction identifier, one position change sequence value, and one heading change sequence value corresponding to each time point can be combined into a corresponding record. All corresponding records are then compiled in chronological order to form the set of correspondences for that segment. The reason for using this processing method is that segmentation only completes the separation of time ranges. However, when performing reverse coupling analysis, it is necessary to clarify the correspondence between different types of data within each segment at the same time position. Therefore, it is necessary to use calibration processing to transform each segment from a simple time segment into an information unit with boundary information, sequence information, and correspondence. Among them, calibration processing refers to the process of assigning clear identification to each segment and supplementing the related data within the segment. The segment start time index and end time index refer to the time position markers that define the start and end points of the segment. The segment identifier refers to the segment identity marker formed based on the start and end boundaries. The direction identifier refers to the discrete marker that represents the change in direction at each time point. The position change sequence value and heading change sequence value refer to the corresponding spatial position change and heading change at each time point, respectively. The correspondence set within the segment refers to the data set formed by corresponding the direction identifier, position change sequence value, and heading change sequence value point by point within the same segment according to the same time index.

[0034] Based on the set of correspondences within each segment, a coupling association operation is performed on the correspondence between the direction indicator and the position change sequence and the heading change sequence. The value state of the direction indicator is synchronously mapped with the value of the position change sequence and the value of the heading change sequence to form a segment-level reverse coupling indicator. The reverse coupling indicators corresponding to each segment are arranged in chronological order to form a reverse coupling indicator sequence.

[0035] When performing coupling association operations based on the set of correspondences within each segment, the direction identifier, position change sequence value, and heading change sequence value can be read point by point in each segment according to time sequence, and these three can be used as association inputs at the same time and position for synchronous processing. The value of the direction identifier is used to characterize the correspondence between the current position change and heading change at the direction level, while the position change sequence value and heading change sequence value are used to characterize the magnitude and strength of the two types of changes at the same time point. Based on this, coupling association operations can be performed on the correspondence between the direction identifier and the two types of sequence values ​​according to the processing methods of point-by-point pairing within the segment, cumulative association within the segment, or hierarchical statistics within the segment. Specifically, this can be understood as follows: first, distinguish opposite direction states based on the direction identifier; then, synchronously map this state to the position change sequence value and heading change sequence value at the same time point, binding the direction relationship and the numerical relationship within the same association unit; subsequently, continuously aggregate all association units within the same segment to form a segment-level reverse coupling indicator that can characterize the degree of reverse change association within that segment. For example, within a certain segment, if multiple... If the corresponding position change sequence values ​​and heading change sequence values ​​consistently maintain a canceling relationship in terms of numerical amplitude, then by continuously aggregating the correlation results of these time points, a segment-level reverse coupling indicator reflecting the overall reverse coupling state of the segment can be formed. Then, by arranging the segment-level reverse coupling indicators corresponding to each segment according to the chronological order of the segments on the time axis, a reverse coupling indicator sequence can be formed, used to reflect the continuous evolution of the reverse coupling state within different time segments. Here, coupling correlation operation refers to the operation of jointly calculating directional and numerical relationships in the same processing process; the value state of the direction identifier refers to the direction category corresponding to the direction identifier at a specific time point; the position change sequence values ​​and heading change sequence values ​​refer to the position change and heading change amounts corresponding to the same time point, respectively; synchronous mapping refers to binding the direction identifier to the two types of values ​​one-to-one according to the same time position; the segment-level reverse coupling indicator refers to the representation result of the reverse coupling relationship formed for a single segment; and the reverse coupling indicator sequence refers to the set of multiple segment-level reverse coupling indicators arranged in chronological order.

[0036] The reverse coupling indicator is a marker used to characterize whether a position change sequence and a heading change sequence have opposite directions at the same time point and exhibit mutual attenuation during the fusion process. After point-by-point directional comparison of the position change and heading change sequences, if the position change direction and heading change direction are opposite at the same time point, and the corresponding values ​​weaken each other during the fusion calculation, then that time point is identified as having a reverse coupling relationship, and a corresponding reverse coupling indicator is generated. The reverse coupling indicator is used to represent the reverse correlation state between position change features and heading change features, and is used for subsequent amplitude matching processing, attenuation strength analysis, and compensation reconstruction processing. The reverse coupling indicator can correspond to a single time point or a continuous time segment; essentially, it describes the reverse change relationship and coupling state between AIS behavioral features.

[0037] S3. Perform amplitude and direction coupling operations on the position change sequence and heading change sequence within the calibration section according to the reverse coupling indication to form the cancellation strength parameter, determine the characteristic cancellation degree based on the cancellation strength parameter, and construct the cancellation compensation sequence to retain the cancelled change information. In this embodiment, S3 specifically includes the following steps: S301. Extract the reverse coupling indicator, position change sequence value and heading change sequence value corresponding to each time point in the calibration section according to the time sequence. Adjust the direction of the position change sequence value and heading change sequence value according to the value state of the reverse coupling indicator. Perform amplitude matching processing on the adjusted value to form the amplitude and direction coupling calculation result of the corresponding time point. When extracting the reverse coupling indicator, position change sequence value, and heading change sequence value corresponding to each time point within the calibration section in chronological order, the three types of data can be read point by point within the calibration section using the time index as the sorting basis, and the three types of data at each time point can be grouped into the same associated record. Then, based on the value state of the reverse coupling indicator, the position change sequence value and heading change sequence value are adjusted in direction. Specifically, when the reverse coupling indicator corresponds to a reverse coupling state, one type of value in the position change sequence value and the heading change sequence value is converted in the opposite direction, so that the two types of values ​​enter the same comparison direction system. When the reverse coupling indicator corresponds to a non-reverse coupling state, the original direction attribute is maintained. After completing the direction adjustment, amplitude matching processing is performed on the adjusted values, specifically by comparing the position change sequence value and the heading change sequence value point by point. The amplitude relationship involves constraining the side with the larger amplitude and matching the side with the smaller amplitude to form a comparable amplitude combination between the two sequences at the same time point. For example, at a certain time point, the position change sequence shows a large positive change, while the heading change sequence shows a small negative change, and the corresponding negative coupling indicator at that time point is in a negative coupling state. In this case, the heading change sequence value is first converted to a comparison direction consistent with the position change sequence value, and then the matching process is completed based on the difference in amplitude between the two, thus forming the amplitude and direction coupling calculation result at that time point. The reason for adopting this processing order is that the position change sequence value and the heading change sequence value originally belong to different change dimensions. It is difficult to directly determine the cancellation relationship based on the original values ​​alone. Only by first unifying the directional expression through direction adjustment and then unifying the amplitude comparison basis through amplitude matching can a coupling result that can be used to quantify the cancellation strength be formed.

[0038] The value state of the reverse coupling indicator refers to the state category of the reverse coupling indicator at a certain point in time, used to characterize whether a reverse coupling relationship exists at that point in time and the state to which the reverse coupling relationship belongs; direction adjustment refers to the unified processing of the direction attributes of the position change sequence values ​​and the heading change sequence values ​​based on the value state of the reverse coupling indicator, so that the two sequences enter subsequent calculations under the same direction reference; amplitude matching processing refers to the correspondence processing of the amplitude relationship between the position change sequence values ​​and the heading change sequence values ​​at the same point in time after the direction adjustment is completed, so that the two types of values ​​form a pairing relationship under the same comparison scale; the amplitude and direction coupling operation result refers to the joint operation result formed after comprehensively considering the direction relationship and the amplitude relationship at the same point in time, used to characterize the cancellation association state between the position change sequence values ​​and the heading change sequence values; among them, the value state of the reverse coupling indicator determines the execution condition of the direction adjustment, the direction adjustment determines the comparison premise of the amplitude matching processing, and the amplitude matching processing further determines the specific numerical content of the amplitude and direction coupling operation result, so these contents constitute a continuous and connected processing relationship.

[0039] S302. Based on the distribution of amplitude and direction coupling calculation results within the calibration section, the amplitude and direction coupling calculation results at each time point are continuously accumulated. The accumulated results are divided into segments according to the variation range within the calibration section. The accumulated results corresponding to each segment are used as the cancellation strength parameter. The characteristic cancellation degree is determined in combination with the corresponding range of each segment. S303. Combine the degree of feature cancellation to perform compensation and reconstruction processing on the position change sequence and heading change sequence within the calibration section. Extract the reduction part corresponding to each time point from the amplitude and direction coupling calculation results, and perform reverse recovery processing on the reduction part according to the cancellation intensity parameter. The recovered change amount is superimposed on the position change sequence and heading change sequence respectively to form a cancellation compensation sequence.

[0040] When performing compensation and reconstruction processing on the position change sequence and heading change sequence within the calibration section based on the degree of feature cancellation, the amplitude and direction coupling calculation results can be read point by point in the calibration section according to the time sequence. The mutually subtracted numerical portions can be identified from the coupling results at each time point and used as the subtracted portion for that time point. Then, a reverse recovery process is performed on the subtracted portion based on the cancellation strength parameter. Specifically, when the cancellation strength parameter corresponds to a higher cancellation level, a larger recovery amplitude is assigned to the subtracted portion; when the cancellation strength parameter corresponds to a lower cancellation level, a smaller recovery amplitude is assigned. This allows the information subtracted at different time points to be re-compensated according to the strength of the cancellation. After completing the reverse recovery process, the recovered changes are then superimposed onto the position change sequence. In the sequence of position and heading changes, compensation results are formed that correspond one-to-one with the original change sequences, ultimately constituting a compensation-offset sequence. For example, at a certain point in time, the position change sequence and the heading change sequence significantly reduce each other during fusion, resulting in the effective change amplitude in the coupled result being lower than the original change amplitude. In this case, the reduced portion is first separated from the coupled result, and then the reduced portion is replenished according to the corresponding compensation strength parameter. Subsequently, it is superimposed back onto the position change sequence and the heading change sequence, so that the change information at that point in time re-enters the subsequent judgment process. The reason for adopting this processing method is that simply identifying the compensation relationship only indicates that the information has been weakened, while the compensation and reconstruction processing can restore the reduced change content back into the sequence, so that the subsequent judgment of abnormal behavior no longer depends solely on the weakened coupling result.

[0041] Compensation and reconstruction processing refers to the process of replenishing and reconstructing the weakened change information during the coupling process based on the offsetting relationship. The core is to reintegrate the reduced change portion into the original change sequence. The reduced portion at each time point refers to the amount of change reduced by mutual offsetting after the position change sequence and heading change sequence are coupled at each time point, which is used to reflect the degree of information loss at a single point. Reverse recovery processing refers to the process of reverse replenishing the reduced portion based on the offsetting strength parameter, so that the reduced portion is converted back into a change amount that can participate in subsequent calculations according to the offsetting strength. The recovered change amount refers to the replenished value formed after the reverse recovery processing, which is used to replenish the lost information in the position change sequence and heading change sequence. The offsetting compensation sequence refers to the sequence result formed by superimposing the recovered change amount onto the position change sequence and heading change sequence respectively, which is used to characterize the continuous change state after compensation and reconstruction.

[0042] The compensation sequence is a sequence of compensated results formed after restoring the weakened change information during the fusion of position change and heading change sequences. After identifying the cancellation relationship between position change and heading change features, the weakened portion of the change information due to mutual reduction is extracted from the amplitude and direction coupling calculation results. This weakened portion is then restored based on the corresponding cancellation strength parameter. The restored change is then superimposed onto the position change and heading change sequences to form the compensation sequence. The compensation sequence is used to retain the anomalous change information that was weakened or masked during feature fusion, allowing subsequent anomalous behavior detection processes to continue utilizing the reduced behavioral features for anomaly determination.

[0043] There is a corresponding correlation between the offsetting strength parameter and the offsetting compensation sequence. The offsetting strength parameter characterizes the degree to which the position change sequence and the heading change sequence mutually reduce each other during the fusion process, while the offsetting compensation sequence is the compensation result sequence formed by restoring the weakened change information during the fusion process based on the offsetting strength parameter. Specifically, the offsetting strength parameter is formed by continuously accumulating, segmenting, and analyzing the offsetting degree of the amplitude and direction coupling calculation results. Then, based on the feature offsetting degree corresponding to different offsetting strength parameters, the weakened change parts in the position change sequence and heading change sequence due to mutual offsetting are restored, and the restored change amount is re-superimposed on the corresponding sequence, thus forming the offsetting compensation sequence. Therefore, the offsetting strength parameter reflects the degree of offsetting between features, and the offsetting compensation sequence is used to restore the reduced abnormal change information based on the offsetting degree. Together, they realize the identification and compensation of the offsetting relationship of AIS behavioral features during the fusion process.

[0044] In this embodiment, S302 specifically refers to: Read the amplitude and direction coupling calculation results corresponding to each time point in the calibration section in chronological order. Take the amplitude and direction coupling calculation result of the first time point as the starting accumulated value, and then merge the amplitude and direction coupling calculation results of subsequent time points into the starting accumulated value to form the continuous accumulated result corresponding to each time point. When reading the amplitude and direction coupling calculation results corresponding to each time point within the calibration segment in chronological order, the time points can be arranged according to the time index first, and the corresponding amplitude and direction coupling calculation results can be extracted point by point. The amplitude and direction coupling calculation result of the first time point is used as the initial accumulated value. Subsequently, starting from subsequent time points, the amplitude and direction coupling calculation results of the current time point are successively merged into the existing accumulated value, so that the accumulated result is gradually updated over time, thus forming a continuous accumulated result at each time point. For example, in multiple consecutive time points, the first time point forms the initial accumulated value, the second time point merges its own result to form a new accumulated value, and the third time point... The current result is then added to the previous accumulated value, and so on, to obtain a complete sequence of continuous accumulated results. This process allows for the continuous integration of coupling changes at discrete time points, providing a foundation for subsequent segmentation and the construction of offsetting strength parameters. The initial accumulated value is the amplitude and direction coupling calculation result corresponding to the first time point within the calibration segment. Incorporating the initial accumulated value in sequence means adding the amplitude and direction coupling calculation results of subsequent time points to the existing accumulated results in chronological order. The continuous accumulated result corresponding to each time point refers to the cumulative value formed from the first time point to the current time point, which is used to reflect the continuous accumulation state of coupling changes within the calibration segment.

[0045] The intervals are divided around the change range of the continuous accumulation result within the calibration section. The starting value and ending value of the interval are extracted according to the numerical change order of the continuous accumulation result. The continuous accumulation result is then segmented and assigned according to the starting value and ending value of the interval, forming multiple segments corresponding to the accumulation result range. When dividing the data into intervals based on the variation range of continuously accumulated results within a calibrated segment, the continuously accumulated results at each time point can be extracted in chronological order. Then, all data can be sorted according to the numerical patterns of the continuously accumulated results to determine the overall variation range. Next, the starting position is selected from the sorted values ​​as the starting value of the interval, and the corresponding ending value is determined based on subsequent numerical changes, thus forming multiple continuous intervals. After the intervals are determined, each continuously accumulated result is matched with the starting and ending values ​​of each interval to assign each continuously accumulated result to its corresponding interval, completing the segmentation and forming multiple ranges of accumulated results corresponding to each segment. For example, when the continuously accumulated results show an increasing trend... Under varying conditions, the data can be divided into low-value, medium-value, and high-value intervals, and the continuous cumulative results corresponding to each time point can be assigned to the corresponding intervals. This process can divide the continuously changing values ​​into different range levels, providing a segmented basis for the subsequent construction of offset strength parameters. The variation range of the continuous cumulative results within the calibration segment refers to the overall numerical range formed by the continuous cumulative results within the time range. Interval division is the segmentation of this range. The starting and ending values ​​of the intervals are used to define the boundaries of each interval. Segment assignment refers to assigning the continuous cumulative results to the corresponding intervals according to their numerical ranges. The cumulative result range corresponding to multiple segments refers to the set of continuous cumulative results contained within each interval.

[0046] The continuous cumulative results within the range corresponding to each segment are aggregated into segments, and the aggregated results of each segment are used as the offset strength parameter. According to the correspondence between the range of each segment and the offset strength parameter, the offset strength parameter within each segment is sorted. Based on the sorting position, the offset strength parameter is divided into three continuous intervals, where the offset strength parameter in the first interval corresponds to weak offset, the offset strength parameter in the middle interval corresponds to medium offset, and the offset strength parameter in the last interval corresponds to strong offset, thus completing the determination of the feature offset degree.

[0047] When aggregating consecutive cumulative results within the range of cumulative results corresponding to each segment, one can first extract all consecutive cumulative results already assigned to that segment in chronological order within each segment. Then, using methods such as same-direction accumulation, interval summation, or interval balanced accumulation, the consecutive cumulative results within the same segment are integrated into a single segment aggregation result, which is then used as the offsetting strength parameter corresponding to that segment. For example, if a segment contains several consecutive cumulative results that collectively reflect the offsetting accumulation state within the same numerical level, summing these results forms an offsetting strength parameter that represents the overall offsetting level of that segment. After all segments have completed segment aggregation, the offsetting strength parameter is then used to further refine the offsetting strength parameter across all segments. The elimination intensity parameters are sorted according to their numerical values, and then divided into three consecutive intervals: the first, middle, and last intervals. The elimination intensity parameters in the first interval correspond to weak elimination, those in the middle interval correspond to moderate elimination, and those in the last interval correspond to strong elimination. For example, a smaller elimination intensity parameter falling into the first interval indicates weak elimination accumulation within the corresponding segment, a medium elimination intensity parameter falling into the middle interval indicates intermediate elimination accumulation, and a larger elimination intensity parameter falling into the last interval indicates strong elimination accumulation. The reason for this processing is that although continuous accumulation results can reflect the local accumulation state, only through segment aggregation and sorting can the scattered accumulated values ​​be transformed into clear elimination levels.

[0048] The cancellation strength parameter is a data parameter used to characterize the degree of mutual attenuation between position change sequences and heading change sequences during the fusion process. By continuously accumulating, dividing, and aggregating the amplitude and direction coupling calculation results at each time point within the calibration segment, the overall cancellation state between position change features and heading change features in different segments can be obtained, and the corresponding cancellation strength parameter is formed based on this overall cancellation state. The magnitude of the cancellation strength parameter reflects the degree of mutual cancellation between position change features and heading change features. A higher cancellation degree indicates a significant information attenuation relationship between the two features during the fusion process; a lower cancellation degree indicates a weaker mutual attenuation effect between the two features. The cancellation strength parameter is used for subsequent feature cancellation degree classification and cancellation compensation sequence construction.

[0049] Segment aggregation refers to the process of integrating all continuous cumulative results within the same segment. Its core is compressing multiple discrete cumulative values ​​within a segment into a unified result. The segment aggregation result is a representative value formed after integration, used to characterize the overall cumulative state within that segment. The offsetting strength parameter is a parameter derived from the segment aggregation result, used to characterize the strength of offsetting between positional and heading changes within a segment. The correspondence between the range of each segment and the offsetting strength parameter refers to the attribution relationship between the numerical range of each segment and the offsetting strength parameter formed after aggregation. The offsetting strength parameter of the first segment in the sorting refers to the set of parameters with values ​​at the beginning of the sorting, corresponding to weak offsetting and indicating a low level of offsetting. The offsetting strength parameter of the middle segment in the sorting refers to the set of parameters with values ​​in the middle position, corresponding to moderate offsetting and indicating an intermediate level of offsetting. The offsetting strength parameter of the last segment in the sorting refers to the set of parameters with values ​​at the end of the sorting, corresponding to strong offsetting and indicating a high level of offsetting. Through this continuous processing from segment aggregation to parameter sorting and then to offsetting level division, the offsetting states of different segments can be converted into structured results with clear hierarchical meanings.

[0050] S4. Based on the degree of feature cancellation and the cancellation compensation sequence, the position change sequence and heading change sequence are weighted and compensated and superimposed to form a reconstructed feature sequence. The reconstructed feature sequence is then coupled to generate the abnormal evaluation coefficient corresponding to the abnormal behavior of the ship. In this embodiment, S4 specifically includes the following steps: S401. Based on the degree of feature cancellation and the cancellation compensation sequence, the position change sequence and the heading change sequence are weighted and compensated and superimposed. The degree of feature cancellation corresponding to each time point is extracted in time order, and the position change sequence and the heading change sequence are assigned corresponding weights according to the degree of feature cancellation. At the same time, the cancellation compensation sequence is superimposed on the position change sequence and the heading change sequence according to the time point to form the weighted sequence data. When redistributing and compensating the position change sequence and heading change sequence based on the degree of feature cancellation and the compensation sequence, the degree of feature cancellation at each time point can be read sequentially in chronological order, and the degree of feature cancellation at different time points can be used as the basis for weight allocation. In specific implementation, a correspondence between the degree of feature cancellation and the weight values ​​can be established in advance. For example, when a certain time point corresponds to a weak degree of cancellation, a set of basic weights can be assigned to the position change sequence and the heading change sequence; when a certain time point corresponds to a moderate degree of cancellation, another set of weights can be assigned to the two types of sequences; when a certain time point corresponds to a strong degree of cancellation, a different weight combination than the previous two types can be assigned. After the weighting is completed, the compensation sequence values ​​corresponding to the same time point are read, and the compensation sequence values ​​are superimposed respectively. The offset compensation sequence is added to the position change sequence and the heading change sequence, so that the original sequence forms a new numerical state after compensation. For example, at a certain time point, the feature offset degree is divided into strong offset degree. Then, the position change sequence can be assigned a higher participation weight, and the heading change sequence can be assigned a corresponding matching weight. At the same time, the offset compensation sequence value at that time point is merged into the position change sequence value and the heading change sequence value respectively, forming the sequence value after redistribution and compensation superposition. The reason for adopting this processing order is that the feature offset degree reflects the offset strength at different time points, while the offset compensation sequence reflects the recovery result of the reduced information. Only by applying both to the position change sequence and the heading change sequence at the same time can the reconstructed feature sequence subsequently constructed contain both offset strength information and compensation recovery information.

[0051] Weight redistribution and compensation superposition processing refers to the joint operation of simultaneously assigning weights and superimposing compensation values ​​in the same processing step. Weight redistribution adjusts the participation ratio of position change sequences and heading change sequences at the current time point, while compensation superposition incorporates the recovered values ​​from the compensation sequence into the two original sequences. Assigning corresponding weights means configuring weight values ​​matching the current compensation level for the position change and heading change sequences based on the degree of feature compensation at each time point, ensuring differences in the numerical contributions of the two sequences at different compensation levels. The weight-redistributed sequence data refers to the new sequence result formed after the position change and heading change sequences have undergone weight adjustment and the compensation sequence values ​​have been superimposed. This result no longer only retains the original change information but also includes the numerical content corrected by the compensation relationship. From a processing perspective, the degree of feature compensation determines the specific value of the assigned weight, which in turn determines the numerical proportion of the position change and heading change sequences at the current time point. The compensation sequence further alters the numerical content of the two sequences, ultimately forming the weight-redistributed sequence data, which is used for subsequent coupling operations on the reconstructed feature sequences.

[0052] S402. Based on the sequence data after weight redistribution, the position change sequence and heading change sequence corresponding to the same time point are combined and processed accordingly. The combination results of each time point are arranged in chronological order to form the reconstructed feature sequence. Based on the weighted sequence data, when combining the position change sequence and heading change sequence corresponding to the same time point, the two types of sequence values, after weight adjustment and compensation superposition, can be read point by point in chronological order. Using the same time index as the basis, the position change sequence values ​​and heading change sequence values ​​at the same time point are paired and combined into the same set of joint data. In practice, point-by-point binding, sequential splicing, or binary pairing can be used to ensure that each time point forms a unique combination result. For example, if the position change sequence value at a certain time point reflects the contribution of spatial movement after weight adjustment, and the heading change sequence value reflects the contribution of directional change after compensation superposition, then the two are combined into the joint feature of that time point. The records are then arranged sequentially according to time to form a reconstructed feature sequence. The reason for adopting this processing method is that although the position change sequence and heading change sequence have undergone weight redistribution and compensation superposition in the previous stage, they still exist as two independent sequences. Only through corresponding combination processing can the spatial change information and heading change information at the same time point be unified into the same sequence structure, which is convenient for subsequent coupling operations. The corresponding combination processing refers to the process of pairing the position change sequence and heading change sequence according to the same time index. The reconstructed feature sequence is a continuous sequence formed by arranging the corresponding combination results of all time points in chronological order, which is used to characterize the joint behavior change process after weight correction and compensation recovery.

[0053] S403. Perform coupling operation processing on the reconstructed feature sequence, perform numerical correlation operation on the reconstructed feature sequence at each time point in chronological order, and perform intra-segment integration processing on the coupling operation results of all time points. Use the integrated numerical results as the anomaly assessment coefficient corresponding to the abnormal behavior of the ship.

[0054] In this embodiment, S403 specifically refers to: Extract the reconstructed feature sequences corresponding to each time point in chronological order, perform numerical pairing on the position change sequence values ​​and heading change sequence values ​​within the same time point, and use the paired numerical combinations within each time point as the input for numerical association operations. When extracting the reconstructed feature sequences corresponding to each time point in chronological order, the position change sequence values ​​and heading change sequence values ​​corresponding to each time point can be read point by point according to the time index. Then, at the same time position, numerical pairing processing is performed on the two types of values, that is, binding the position change sequence values ​​and heading change sequence values ​​within the same time point into a joint value. In specific implementation, a fixed-order pairing processing method can be adopted, placing the position change sequence values ​​first and the heading change sequence values ​​last, thereby ensuring that the combination structure remains consistent across all time points. For example, at a certain time point, the position change sequence values ​​reflect the spatial change state at that moment, and the heading change sequence values ​​reflect the directional change state at that moment; in this case, the two are combined into a pair of values, which serves as the current... The numerical correlation calculations are input from previous time points, and the same process is repeated for subsequent time points, thus forming an input set arranged in chronological order. The reason for adopting this processing method is that although the reconstructed feature sequence has completed the previous processing, if the position change sequence values ​​and heading change sequence values ​​are not uniformly paired within the same time point, the subsequent numerical correlation calculations will lack clear input units, making it difficult to characterize the joint relationship between the two types of changes at the same time and position. Here, numerical pairing refers to the process of combining the position change sequence values ​​and heading change sequence values ​​within the same time point into a unified numerical unit according to a fixed correspondence. The numerical correlation calculation input refers to the joint numerical set formed after numerical pairing, which is used for subsequent point-by-point coupling calculations and intra-segment integration processing.

[0055] Numerical correlation operations are performed around the paired numerical combinations at each time point. The numerical values ​​of the position change sequence and the heading change sequence are coupled point by point to form the coupling operation results corresponding to each time point, and the coupling operation results are arranged in chronological order to form a set of coupling operation results. When performing numerical correlation operations on paired numerical combinations at each time point, the position change sequence and heading change sequence values ​​that have been paired can be read point by point in chronological order, and the two values ​​at the same time point can be treated as a joint calculation unit. Then, point-by-point coupling calculations are performed on each joint calculation unit. Specifically, synchronous incorporation, corresponding comparison, or joint accumulation can be used to place the position change sequence values ​​and heading change sequence values ​​into the same calculation process to form a single-point coupling result corresponding to the current time point. For example, if the position change sequence values ​​at a certain time point reflect spatial displacement changes and the heading change sequence values ​​reflect directional changes, then the two are jointly calculated at the same time position to form the coupling operation result corresponding to that time point. The same process is then repeated for all time points, and all single-point coupling results are arranged in chronological order to form a coupling... The reason for adopting this processing method is that numerical pairing only completes the corresponding binding of two types of values ​​at the same time point, and does not form a result that can directly represent the joint change relationship. Only through numerical correlation operation and point-by-point coupling calculation can the position change sequence value and the heading change sequence value be transformed into a coupling result that can reflect the comprehensive change state at the same time point. Among them, numerical correlation operation refers to the calculation process of performing joint processing on the position change sequence value and the heading change sequence value around the same time point. Point-by-point coupling calculation refers to the processing form of performing joint calculation separately for each time point. The coupling operation result corresponding to each time point refers to the result value formed after completing the joint calculation within a single time point. The coupling operation result set refers to the result set formed by arranging the coupling operation results corresponding to all time points in chronological order, which is used for subsequent integration processing within the segment.

[0056] The coupling operation results set is integrated within a segment. The coupling operation results at each time point are assigned to a segment according to the segment boundary. The coupling operation results within each segment are continuously aggregated. The aggregated results corresponding to each segment are sequentially summarized to form the anomaly assessment coefficient corresponding to the abnormal behavior of the ship.

[0057] When performing intra-segment integration processing on the set of coupling operation results, the coupling operation results at each time point in the set can be mapped to the corresponding segment one by one according to the determined segment boundaries, thus completing the segment assignment. Then, within each segment, the coupling operation results belonging to that segment are continuously aggregated according to time order. This can be achieved through point-by-point accumulation, sequential merging, or intra-segment continuous integration, integrating the coupling operation results from multiple time points within the same segment into a single segment-level aggregated result. For example, if a segment contains multiple consecutive time points, the coupling operation results from these time points are first assigned to the same segment, and then integrated sequentially according to time order to form a single segment-level result for that segment. After continuous aggregation is completed for all segments, the aggregation results corresponding to each segment are sequentially summarized according to the segment's arrangement on the time axis, forming a result that can characterize the entire ship's behavior. The comprehensive result of the changing state is used as the anomaly assessment coefficient corresponding to abnormal ship behavior. The reason for adopting this processing method is that the coupling operation result at a single time point only reflects the joint change state at a local moment, while segment assignment can put the discrete results back into their respective time periods, continuous aggregation can reflect the cumulative change within each segment, and sequential summarization further integrates the change states of multiple segments into a unified judgment result. Among them, the segment integration processing refers to the process of sorting and merging the coupling operation results on a segment basis, the segment assignment refers to assigning the coupling operation result at each time point to the corresponding segment, the continuous aggregation refers to the continuous integration of multiple coupling operation results in chronological order within a single segment, the sequential summarization refers to the overall aggregation of the aggregation results of each segment in chronological order, and the anomaly assessment coefficient corresponding to abnormal ship behavior is the comprehensive assessment value formed by summarizing the aggregation results of all segments.

[0058] S5. Combine the feature cancellation degree and the anomaly evaluation coefficient to construct a weight evolution sequence, dynamically update the weights, and continuously correct the coupling operation of the feature sequence accordingly.

[0059] In this embodiment, S5 specifically refers to: Using the time index as the correlation benchmark, the feature cancellation degree corresponding to each time point is paired with the anomaly evaluation coefficient. A numerical correlation relationship between the feature cancellation degree and the anomaly evaluation coefficient is established in the pairing results. Based on this numerical correlation relationship, the weight value corresponding to each time point is determined. The weight values ​​of all time points are arranged in chronological order to construct a weight evolution sequence. Using time index as the correlation benchmark, when pairing the feature cancellation degree and anomaly evaluation coefficient corresponding to each time point, we can first read the feature cancellation degree and anomaly evaluation coefficient corresponding to each time point in chronological order, and use the same time point as the unique corresponding position to bind the two types of data into the same pairing unit. After completing the pairing, we then establish a numerical correlation relationship between the feature cancellation degree and the anomaly evaluation coefficient around each pairing unit. Specifically, we can use corresponding mapping, hierarchical association, or combined assignment. For example, when a time point corresponds to a strong cancellation degree and the anomaly evaluation coefficient is at a high level, we assign a higher weight value to that time point; when a time point corresponds to a weak cancellation degree and the anomaly evaluation coefficient is at a low level, we assign a higher weight value to that time point. When the time is horizontal, a lower weight value is assigned to that time point. For example, if, across multiple consecutive time points, the degree of feature cancellation changes from moderate to strong cancellation, and the anomaly evaluation coefficient increases simultaneously, then the corresponding weight value is adjusted to a higher level. After determining the weight values ​​for all time points, the weight values ​​for all time points are arranged in chronological order to form a weight evolution sequence. The reason for using this approach is that the degree of feature cancellation reflects the strength of the cancellation relationship, and the anomaly evaluation coefficient reflects the current level of anomaly. Only by associating the two at the same time point can a basis for weight adjustment that changes over time be formed, enabling the subsequent coupling operation correction process to have continuous temporal evolution characteristics.

[0060] The time index refers to the time position marker used to identify the chronological order of each time point, ensuring accurate correspondence between the feature cancellation degree and the anomaly evaluation coefficient at the same time point; the feature cancellation degree refers to the strength level of mutual cancellation between the position change sequence and the heading change sequence during the fusion process, used to characterize the degree of cancellation relationship; the anomaly evaluation coefficient refers to the comprehensive evaluation value formed around the abnormal behavior of the ship, used to characterize the magnitude of the abnormal state; corresponding pairing refers to combining the feature cancellation degree and the anomaly evaluation coefficient at the same time point into the same data unit based on the same time index; numerical correlation refers to establishing the corresponding numerical relationship between the feature cancellation degree and the anomaly evaluation coefficient in the pairing results, used to support the determination of weight values; weight value refers to the weight value configured for each time point, used to subsequently adjust the participation degree of the position change sequence and the heading change sequence; the weight evolution sequence refers to the sequence result formed by arranging the weight values ​​corresponding to all time points in chronological order, used to characterize the continuous change process of weights on the time axis.

[0061] By focusing on the weight changes in the weight evolution sequence, the weights of the position change sequence and the heading change sequence are updated point by point, so that the weight values ​​at each time point participate in the numerical correlation process of the reconstructed feature sequence. Furthermore, the weight change trend of the previous time point is introduced into the subsequent time point, and the reconstructed feature sequence is subjected to progressive correction coupling operation processing to achieve continuous correction of the feature sequence coupling operation.

[0062] When performing point-by-point updates on the weights of position change sequences and heading change sequences based on weight value changes in the weight evolution sequence, we can first read the weight values ​​corresponding to each time point in the weight evolution sequence in chronological order, and then map the weight value of each time point to the reconstructed feature sequence at the same time point. In practice, the position change sequence value and heading change sequence value at each time point can be used as the objects to be updated. Then, the weight value corresponding to the current time point is introduced into the numerical association process, so that the joint value of the current position is adjusted synchronously with the weight changes. Subsequently, when processing subsequent time points, not only the weight value of the current time point is used, but the trend of weight value changes from previous time points is also introduced to progressively correct the coupling operation at the current time point. For example, if the weight value continuously increases within a certain continuous time period, the coupling operation at subsequent time points will be superimposed with the correction effect corresponding to the previous upward trend on the current weight. Conversely, if the weight value at previous time points continuously decreases, the coupling operation at subsequent time points will incorporate the corresponding downward correction process. For example, if the weight value at a certain time point is low, but the weight values ​​at several subsequent time points continuously increase, the reconstructed feature sequences at these subsequent time points will gradually increase the participation of the corresponding values ​​during the coupling operation. The reason for adopting this approach is that using the weight value at the current time point alone can only reflect the instantaneous state, while introducing the trend of weight value changes at previous time points can make the weight update process continuous, allowing the coupling operation to be continuously corrected as time progresses.

[0063] The weight value change in the weight evolution sequence refers to the change process of the weight values ​​at each time point in the weight evolution sequence along the time sequence, used to characterize the dynamic evolution state of the weights on the time axis; the point-by-point update processing of the weights of the position change sequence and the heading change sequence means that at each time point, according to the corresponding weight value, the proportion of the values ​​of the position change sequence and the heading change sequence participating in subsequent calculations is updated; the numerical association process of the weight values ​​at each time point with the reconstructed feature sequence means that the weight value of the current time point is directly introduced into the calculation of the reconstructed feature sequence at that time point, so that the weight values... The value becomes part of the numerical association; the trend of weight changes at previous time points refers to the direction and magnitude of the changes in weight values ​​at multiple time points before the current time point in chronological order; the progressive correction of coupling operation processing refers to the process of gradually correcting the coupling calculation by combining the current weight value with the trend of weight changes at previous time points when performing coupling operation at the current time point; the continuous correction of coupling operation of feature sequence refers to the continuous introduction of new weight update results and historical change trends into the coupling operation at subsequent time points as time progresses, so that the coupling operation is always in a state of continuous adjustment.

[0064] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means (e.g., infrared, wireless, microwave, etc.). A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0065] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0066] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0067] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0068] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0069] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0070] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting abnormal behavior of a ship based on AIS big data, characterized in that, Specifically, the following steps are included: S1. Simultaneously extract position change information and heading change information from the ship's AIS data, construct position change sequence and heading change sequence, and introduce a unified behavior domain benchmark for numerical adjustment to generate a direction sequence; S2. Use the direction sequence to perform point-by-point direction comparison between the position change sequence and the heading change sequence, determine whether the position change sequence and the heading change sequence are opposite and cancel each other out during the fusion process, and calibrate the sections that meet the opposite change to generate a reverse coupling indicator. S3. Perform amplitude and direction coupling operations on the position change sequence and heading change sequence within the calibration section according to the reverse coupling indication to form the cancellation strength parameter, determine the characteristic cancellation degree based on the cancellation strength parameter, and construct the cancellation compensation sequence to retain the cancelled change information. S4. Based on the degree of feature cancellation and the cancellation compensation sequence, the position change sequence and heading change sequence are weighted and compensated and superimposed to form a reconstructed feature sequence. The reconstructed feature sequence is then coupled to generate the abnormal evaluation coefficient corresponding to the abnormal behavior of the ship. S5. Combine the feature cancellation degree and the anomaly evaluation coefficient to construct a weight evolution sequence, dynamically update the weights, and continuously correct the coupling operation of the feature sequence accordingly.

2. The method for detecting abnormal ship behavior based on AIS big data according to claim 1, characterized in that, S1 specifically refers to: Ship AIS data is sorted in chronological order, and position change information and heading change information corresponding to each time point are extracted. The position change information and heading change information are synchronously matched using a time index. A position change sequence is constructed by calculating the position difference between adjacent time points, and a heading change sequence is constructed by calculating the heading difference between adjacent time points. A unified behavioral domain benchmark is introduced, and numerical mapping processing is performed on the position change sequence and the heading change sequence. The corresponding values ​​of the position change sequence and the heading change sequence are limited to a unified numerical range. The numerical data is then normalized by scaling and range transformation to form the normalized position change sequence and heading change sequence. For the adjusted position change sequence and heading change sequence, the positive and negative attributes of the corresponding changes at each time point are extracted, and the positive and negative attributes are converted into direction indicators. The direction indicators are then arranged in chronological order to generate a direction sequence.

3. The method for detecting abnormal ship behavior based on AIS big data according to claim 1, characterized in that, S2 specifically includes the following steps: S201. Use the direction sequence to perform point-by-point direction comparison between the position change sequence and the heading change sequence. Extract the direction identifiers in the direction sequence corresponding to each time point in chronological order, and compare the direction identifiers corresponding to the position change sequence with the direction identifiers corresponding to the heading change sequence one by one to form a direction comparison result sequence. S202. Based on the direction comparison result sequence, determine whether the position change sequence and the heading change sequence are opposite and cancel each other out during the fusion process. Filter the time points in the direction comparison result sequence where the direction identifiers have opposite values, and synchronously compare the position change sequence values ​​and heading change sequence values ​​corresponding to the filtered time points. When the position change sequence values ​​and heading change sequence values ​​are opposite in their numerical change trends and the corresponding values ​​have a mutual reduction relationship in the superposition calculation, it is determined that they cancel each other out during the fusion process, thus forming a set of time points where they are opposite and cancel each other out during the fusion process. S203. For the set of time points that change in opposite directions and cancel each other out during the fusion process, the time continuity is used to divide the time into segments and determine the segment boundaries. Each segment is calibrated and a reverse coupling indicator is constructed based on the correspondence between the direction identifier and the position change sequence and the heading change sequence within each segment. The reverse coupling indicator sequence is then arranged in chronological order.

4. The method for detecting abnormal ship behavior based on AIS big data according to claim 3, characterized in that, S203 specifically refers to: For the set of time points that show opposite changes and cancel each other out during the fusion process, the time points are traversed in the order of time index, the time interval between adjacent time points is calculated, the time interval is compared with a fixed time interval benchmark, when the time interval is equal to the fixed time interval benchmark, it is assigned to the same segment, when the time interval is greater than the fixed time interval benchmark, it is used as the segment division position, thereby completing the segment division and determining the boundary of each segment. After the segment division is completed and the segment boundaries are determined, each segment is calibrated. Segment identifiers are constructed based on the segment start time index and end time index. The corresponding direction identifiers, position change sequence values ​​and heading change sequence values ​​are extracted in each segment to form a set of corresponding relationships within the segment. Based on the set of correspondences within each segment, a coupling association operation is performed on the correspondence between the direction indicator and the position change sequence and the heading change sequence. The value state of the direction indicator is synchronously mapped with the value of the position change sequence and the value of the heading change sequence to form a segment-level reverse coupling indicator. The reverse coupling indicators corresponding to each segment are arranged in chronological order to form a reverse coupling indicator sequence.

5. The method for detecting abnormal ship behavior based on AIS big data according to claim 1, characterized in that, S3 specifically includes the following steps: S301. Extract the reverse coupling indicator, position change sequence value and heading change sequence value corresponding to each time point in the calibration section according to the time sequence. Adjust the direction of the position change sequence value and heading change sequence value according to the value state of the reverse coupling indicator. Perform amplitude matching processing on the adjusted value to form the amplitude and direction coupling calculation result of the corresponding time point. S302. Based on the distribution of amplitude and direction coupling calculation results within the calibration section, the amplitude and direction coupling calculation results at each time point are continuously accumulated. The accumulated results are divided into segments according to the variation range within the calibration section. The accumulated results corresponding to each segment are used as the cancellation strength parameter. The characteristic cancellation degree is determined in combination with the corresponding range of each segment. S303. Combine the degree of feature cancellation to perform compensation and reconstruction processing on the position change sequence and heading change sequence within the calibration section. Extract the reduction part corresponding to each time point from the amplitude and direction coupling calculation results, and perform reverse recovery processing on the reduction part according to the cancellation intensity parameter. The recovered change amount is superimposed on the position change sequence and heading change sequence respectively to form a cancellation compensation sequence.

6. The method for detecting abnormal ship behavior based on AIS big data according to claim 5, characterized in that, S302 specifically refers to: Read the amplitude and direction coupling calculation results corresponding to each time point in the calibration section in chronological order. Take the amplitude and direction coupling calculation result of the first time point as the starting accumulated value, and then merge the amplitude and direction coupling calculation results of subsequent time points into the starting accumulated value to form the continuous accumulated result corresponding to each time point. The intervals are divided around the change range of the continuous accumulation result within the calibration section. The starting value and ending value of the interval are extracted according to the numerical change order of the continuous accumulation result. The continuous accumulation result is then segmented and assigned according to the starting value and ending value of the interval, forming multiple segments corresponding to the accumulation result range. The continuous cumulative results within the range corresponding to each segment are aggregated into segments, and the aggregated results of each segment are used as the offset strength parameter. According to the correspondence between the range of each segment and the offset strength parameter, the offset strength parameter within each segment is sorted. Based on the sorting position, the offset strength parameter is divided into three continuous intervals, where the offset strength parameter in the first interval corresponds to weak offset, the offset strength parameter in the middle interval corresponds to medium offset, and the offset strength parameter in the last interval corresponds to strong offset, thus completing the determination of the feature offset degree.

7. The method for detecting abnormal ship behavior based on AIS big data according to claim 1, characterized in that, S4 specifically includes the following steps: S401. Based on the degree of feature cancellation and the cancellation compensation sequence, the position change sequence and the heading change sequence are weighted and compensated and superimposed. The degree of feature cancellation corresponding to each time point is extracted in time order, and the position change sequence and the heading change sequence are assigned corresponding weights according to the degree of feature cancellation. At the same time, the cancellation compensation sequence is superimposed on the position change sequence and the heading change sequence according to the time point to form the weighted sequence data. S402. Based on the sequence data after weight redistribution, the position change sequence and heading change sequence corresponding to the same time point are combined and processed accordingly. The combination results of each time point are arranged in chronological order to form the reconstructed feature sequence. S403. Perform coupling operation processing on the reconstructed feature sequence, perform numerical correlation operation on the reconstructed feature sequence at each time point in chronological order, and perform intra-segment integration processing on the coupling operation results of all time points. Use the integrated numerical results as the anomaly assessment coefficient corresponding to the abnormal behavior of the ship.

8. The method for detecting abnormal ship behavior based on AIS big data according to claim 7, characterized in that, S403 specifically refers to: Extract the reconstructed feature sequences corresponding to each time point in chronological order, perform numerical pairing on the position change sequence values ​​and heading change sequence values ​​within the same time point, and use the paired numerical combinations within each time point as the input for numerical association operations. Numerical correlation operations are performed around the paired numerical combinations at each time point. The numerical values ​​of the position change sequence and the heading change sequence are coupled point by point to form the coupling operation results corresponding to each time point, and the coupling operation results are arranged in chronological order to form a set of coupling operation results. The coupling operation results set is integrated within a segment. The coupling operation results at each time point are assigned to a segment according to the segment boundary. The coupling operation results within each segment are continuously aggregated. The aggregated results corresponding to each segment are sequentially summarized to form the anomaly assessment coefficient corresponding to the abnormal behavior of the ship.

9. The method for detecting abnormal ship behavior based on AIS big data according to claim 1, characterized in that, S5 specifically refers to: Using the time index as the correlation benchmark, the feature cancellation degree corresponding to each time point is paired with the anomaly evaluation coefficient. A numerical correlation relationship between the feature cancellation degree and the anomaly evaluation coefficient is established in the pairing results. Based on this numerical correlation relationship, the weight value corresponding to each time point is determined. The weight values ​​of all time points are arranged in chronological order to construct a weight evolution sequence. By focusing on the weight changes in the weight evolution sequence, the weights of the position change sequence and the heading change sequence are updated point by point, so that the weight values ​​at each time point participate in the numerical correlation process of the reconstructed feature sequence. Furthermore, the weight change trend of the previous time point is introduced into the subsequent time point, and the reconstructed feature sequence is subjected to progressive correction coupling operation processing to achieve continuous correction of the feature sequence coupling operation.

Images