Ship behavior reasoning method, electronic equipment, storage medium and program product

Abstract

The invention provides a ship behavior reasoning method, electronic equipment, a storage medium and a program product, and relates to the technical field of marine traffic monitoring and ship behavior analysis. The method comprises the following steps: acquiring navigation state data of a target ship in real time; determining abnormal navigation state data according to the navigation state data; determining an observation sequence vector based on the abnormal navigation state data; the observation sequence vector is input into a dynamic Bayesian network, a hidden behavior reasoning result of the target ship is obtained based on a probabilistic reasoning algorithm, and the dynamic Bayesian network is used for capturing time sequence correlation between the navigation motion state of the target ship and hidden behaviors. Through the ship behavior reasoning method and device, the accuracy of ship behavior reasoning is improved, false alarm or missing alarm is reduced, and therefore the effectiveness of maritime affair monitoring and the reliability of decision making are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of maritime traffic monitoring and ship behavior analysis, and in particular to a ship behavior reasoning method, an electronic device, a storage medium and a program product. BACKGROUND

[0002] Currently, the field of maritime monitoring is facing increasingly complex challenges. Real-time collection and analysis of ship tracks and motion behavior have become key to ensuring maritime safety and maintaining maritime order. With the increasing frequency of marine activities, accurately identifying normal navigation patterns and abnormal behavior of ships is of paramount importance for maritime law enforcement, safety warnings and resource management.

[0003] Currently, ship behavior reasoning is mainly achieved through rule-based anomaly detection methods. This method presets a series of rules to identify explicit violations or dangerous behaviors such as abnormal heading, sudden stop, detour, etc. By comparing real-time collected ship motion data with preset rules, it aims to quickly identify and alert potential anomalies. However, in some scenarios, the above anomaly detection method has low accuracy in ship behavior reasoning, which is prone to false positives or false negatives, thereby affecting the effectiveness of maritime monitoring and the reliability of decision-making. SUMMARY

[0004] The present application provides a ship behavior reasoning method, an electronic device, a storage medium and a program product to improve the accuracy of ship behavior reasoning, reduce false positives or false negatives, and thus improve the effectiveness of maritime monitoring and the reliability of decision-making.

[0005] In a first aspect, the present application provides a ship behavior reasoning method, comprising:

[0006] real-time acquisition of the navigation state data of the target ship;

[0007] determining abnormal navigation state data according to the navigation state data;

[0008] determining an observation sequence vector based on the abnormal navigation state data;

[0009] inputting the observation sequence vector into a dynamic Bayesian network to obtain a hidden behavior reasoning result of the target ship based on a probability reasoning algorithm, the dynamic Bayesian network being used to capture the time sequence correlation between the navigation motion state and the hidden behavior of the target ship.

[0010] In one possible implementation, the dynamic Bayesian network is constructed by the following way:

[0011] Based on the temporal dependencies between adjacent time slices of a ship, a cross-temporal topology is established between the ship's motion state nodes and hidden behavior nodes. The motion state nodes represent the ship's state within a specific time slice, and the hidden behavior nodes represent the actions or events that occur to the ship within a specific time slice.

[0012] A two-time-slice time series model is adopted to determine the conditional probability distribution of hidden behavior nodes in the cross-time series topology, so as to complete the construction of the first dynamic Bayesian network. The conditional probability distribution represents the probability of the occurrence of hidden behavior nodes in the current time slice given the motion state nodes and / or hidden behavior nodes in the previous time slice.

[0013] Initialize the conditional probability distribution of each hidden behavior node in the first dynamic Bayesian network to obtain the second dynamic Bayesian network;

[0014] Based on the prior of structural sparsity, the parameters of the second dynamic Bayesian network are adjusted to obtain the dynamic Bayesian network, which adopts a multi-level structure with temporal dependencies.

[0015] In one possible implementation, the conditional probability distribution of each hidden behavior node in the first dynamic Bayesian network is initialized to obtain the second dynamic Bayesian network, including:

[0016] Based on the expectation-maximization algorithm, the conditional probability distribution values ​​of each hidden behavior node in the first dynamic Bayesian network are determined.

[0017] Based on the conditional probability distribution values, each hidden behavior node is initialized to obtain the second dynamic Bayesian network.

[0018] In one possible implementation, determining the observation sequence vector based on abnormal navigation state data includes:

[0019] Extract motion features from abnormal navigation status data;

[0020] Based on motion characteristics, an observation sequence vector is generated.

[0021] In one possible implementation, determining abnormal navigation status data based on navigation status data includes:

[0022] Determine the time interval between adjacent navigation status data in the navigation status data;

[0023] Based on time intervals, anomaly detection is performed on the navigation status data to obtain the first abnormal navigation status data.

[0024] Kalman filtering is applied to the navigation status data to generate predicted state values;

[0025] Based on the predicted state value, anomaly determination is performed on the navigation state data to obtain the second abnormal navigation state data;

[0026] Based on shore-based radar echoes and satellite remote sensing data, anomalies are determined in the navigation status data to obtain the third abnormal navigation status data.

[0027] Based on the first abnormal navigation status data, the second abnormal navigation status data, and the third abnormal navigation status data, the abnormal navigation status data is determined.

[0028] In one possible implementation, the observation sequence vector is input into a dynamic Bayesian network, and the hidden behavior of the target vessel is inferred based on a probabilistic inference algorithm, including:

[0029] The observation sequence vector is input into a dynamic Bayesian network, and the forward-backward algorithm is used to obtain the posterior probability of the hidden behavior of the target ship. The forward-backward algorithm includes: recursion, Recursion, normalized posterior probability calculation and in Recursion and / or A multiple hypothesis generation mechanism based on particle filtering is introduced into the recursion;

[0030] The reasoning result for the hidden behavior is determined based on the posterior probability.

[0031] In one possible implementation, it also includes:

[0032] Based on the hidden behavior inference results and the observation sequence vector, determine whether the target vessel has undergone the set changes;

[0033] If the settings change, the Bayesian structure learning algorithm is invoked to reconstruct the cross-temporal topology of the dynamic Bayesian network based on the changes in settings.

[0034] Based on the observed sequence vector, the conditional probability distribution of each hidden behavior node in the reconstructed dynamic Bayesian network is incrementally adjusted.

[0035] Secondly, this application provides a ship behavior reasoning device, comprising:

[0036] The acquisition module is used to acquire the navigation status data of the target vessel in real time.

[0037] The determination module is used to determine abnormal navigation status data based on navigation status data; and to determine the observation sequence vector based on the abnormal navigation status data.

[0038] The inference module is used to input the observation sequence vector into the dynamic Bayesian network. Based on the probabilistic inference algorithm, the inference result of the hidden behavior of the target ship is obtained. The dynamic Bayesian network is used to capture the temporal correlation between the navigation motion state of the target ship and the hidden behavior.

[0039] In one possible implementation, the dynamic Bayesian network is constructed in the following way:

[0040] Based on the temporal dependencies between adjacent time slices of a ship, a cross-temporal topology is established between the ship's motion state nodes and hidden behavior nodes. The motion state nodes represent the ship's state within a specific time slice, and the hidden behavior nodes represent the actions or events that occur to the ship within a specific time slice.

[0041] A two-time-slice time series model is adopted to determine the conditional probability distribution of hidden behavior nodes in the cross-time series topology, so as to complete the construction of the first dynamic Bayesian network. The conditional probability distribution represents the probability of the occurrence of hidden behavior nodes in the current time slice given the motion state nodes and / or hidden behavior nodes in the previous time slice.

[0042] Initialize the conditional probability distribution of each hidden behavior node in the first dynamic Bayesian network to obtain the second dynamic Bayesian network;

[0043] Based on the prior of structural sparsity, the parameters of the second dynamic Bayesian network are adjusted to obtain the dynamic Bayesian network, which adopts a multi-level structure with temporal dependencies.

[0044] In one possible implementation, the determining module is specifically used for:

[0045] Based on the expectation-maximization algorithm, the conditional probability distribution values ​​of each hidden behavior node in the first dynamic Bayesian network are determined.

[0046] Based on the conditional probability distribution values, each hidden behavior node is initialized to obtain the second dynamic Bayesian network.

[0047] In one possible implementation, the determining module is specifically used for:

[0048] Extract motion features from abnormal navigation status data;

[0049] Based on motion characteristics, an observation sequence vector is generated.

[0050] In one possible implementation, the determining module is specifically used for:

[0051] Determine the time interval between adjacent navigation status data in the navigation status data;

[0052] Based on time intervals, anomaly detection is performed on the navigation status data to obtain the first abnormal navigation status data.

[0053] Kalman filtering is applied to the navigation status data to generate predicted state values;

[0054] Based on the predicted state value, anomaly determination is performed on the navigation state data to obtain the second abnormal navigation state data;

[0055] Based on shore-based radar echoes and satellite remote sensing data, anomalies are determined in the navigation status data to obtain the third abnormal navigation status data.

[0056] Based on the first abnormal navigation status data, the second abnormal navigation status data, and the third abnormal navigation status data, the abnormal navigation status data is determined.

[0057] In one possible implementation, the inference module is specifically used for:

[0058] The observation sequence vector is input into a dynamic Bayesian network, and the forward-backward algorithm is used to obtain the posterior probability of the hidden behavior of the target ship. The forward-backward algorithm includes: recursion, Recursion, normalized posterior probability calculation and in Recursion and / or A multiple hypothesis generation mechanism based on particle filtering is introduced into the recursion;

[0059] The reasoning result for the hidden behavior is determined based on the posterior probability.

[0060] In one possible implementation, the ship behavior reasoning device further includes an adjustment module, which is specifically used for:

[0061] Based on the hidden behavior inference results and the observation sequence vector, determine whether the target vessel has undergone the set changes;

[0062] If the settings change, the Bayesian structure learning algorithm is invoked to reconstruct the cross-temporal topology of the dynamic Bayesian network based on the changes in settings.

[0063] Based on the observed sequence vector, the conditional probability distribution of each hidden behavior node in the reconstructed dynamic Bayesian network is incrementally adjusted.

[0064] Thirdly, this application provides an electronic device, including: a memory and a processor;

[0065] The memory stores the instructions that the computer executes;

[0066] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0067] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the first aspect and / or various possible embodiments of the first aspect.

[0068] Fifthly, this application provides a computer program product, including a computer program that, when executed, implements the first aspect and / or various possible implementations of the first aspect.

[0069] This application provides a ship behavior reasoning method, electronic device, storage medium, and program product, relating to the field of maritime traffic monitoring and ship behavior analysis technology. The method includes: real-time acquisition of navigation status data of a target ship; determination of abnormal navigation status data based on the navigation status data; determination of an observation sequence vector based on the abnormal navigation status data; inputting the observation sequence vector into a dynamic Bayesian network; and obtaining the hidden behavior reasoning result of the target ship based on a probabilistic reasoning algorithm. The dynamic Bayesian network is used to capture the temporal correlation between the target ship's navigation motion state and hidden behavior. This application provides accurate input for the dynamic Bayesian network's reasoning by acquiring the target ship's navigation status data in real time, determining abnormal navigation status data based on the navigation status data, and then determining the observation sequence vector based on the abnormal navigation status data. The observation sequence vector is input into the dynamic Bayesian network, which captures the temporal correlation between the target ship's navigation motion state and hidden behavior. Then, a probabilistic reasoning algorithm is used to obtain the hidden behavior reasoning result of the target ship, thereby improving the accuracy of ship behavior reasoning, reducing false alarms or missed alarms, and thus enhancing the effectiveness of maritime monitoring and the reliability of decision-making. Attached Figure Description

[0070] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0071] Figure 1 A flowchart illustrating the ship behavior reasoning method provided in the embodiments of this application. Figure One ;

[0072] Figure 2 A flowchart illustrating the ship behavior reasoning method provided in the embodiments of this application. Figure Two ;

[0073] Figure 3 This is a schematic diagram of the structure of the ship behavior reasoning device provided in the embodiments of this application;

[0074] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0075] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0076] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0077] Currently, maritime monitoring primarily relies on multi-source data, including Automatic Identification System (AIS) signals, shore-based radar echoes, and satellite remote sensing, to collect and analyze ship tracks and motion behaviors in real time. Common methods include rule-based anomaly detection, static classification models, Hidden Markov Models, or static Bayesian networks to probabilistically infer ship behavior, identifying obvious violations or dangerous behaviors such as abnormal headings or sudden stops and detours. These methods, when AIS signals are complete and environmental information is sufficient, can effectively support maritime law enforcement and safety early warning.

[0078] However, the marine environment is complex and ever-changing, and AIS signals are often lost due to malfunctions, malicious shutdowns, or obstructions, resulting in discontinuous and incomplete trajectory data. Furthermore, models based on Hidden Markov Models or Static Bayesian Networks often ignore the temporal dependence of ship behavior and the dynamic changes in the environment, making it difficult to accurately characterize the evolution of ship behavior over long time scales.

[0079] In addition, existing methods rely heavily on a large amount of historical labeled data, and their performance drops significantly when data is scarce or labels are incomplete, making it difficult to meet the needs of maritime monitoring for real-time performance, online update capabilities, and high-reliability inference.

[0080] To address the aforementioned issues, this application provides a ship behavior reasoning method. This method determines abnormal navigation status data based on real-time acquired navigation status data of the target ship. Then, based on the abnormal navigation status data, it determines an observation sequence vector. Finally, the observation sequence vector is input into a dynamic Bayesian network, and based on a probabilistic reasoning algorithm, the hidden behavior reasoning result of the target ship is obtained.

[0081] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0082] Figure 1 A flowchart illustrating the ship behavior reasoning method provided in the embodiments of this application. Figure One ,like Figure 1 As shown, the method includes:

[0083] S101. Acquire real-time navigation status data of the target vessel.

[0084] In this step, it can be understood that in order to perform behavioral reasoning about a ship, it is necessary to obtain the target ship's navigation status data in real time. The specific method for obtaining the target ship's navigation status data can be selected according to the actual situation.

[0085] In one implementation, the navigation status data of the target vessel is obtained from the Automatic Identification System (AIS), that is, the AIS signal is obtained from the AIS. An AIS is a system that automatically transmits and receives dynamic and static information about a vessel via very high frequency radio waves.

[0086] Automatic Identification Systems (AIS) allow for real-time exchange of critical data, including vessel position, speed, course, name, MMSI code, port of destination, and estimated time of arrival, between vessels and between vessels and shore stations. These signals are designed to enhance maritime safety by providing detailed information about surrounding vessels to aid in collision avoidance and are an indispensable source of fundamental data in modern maritime surveillance.

[0087] S102. Determine abnormal navigation status data based on navigation status data.

[0088] Furthermore, navigation status data may be interrupted or missing due to equipment malfunctions, human intervention, or signal obstruction. For example, AIS signal data may be interrupted or missing due to equipment malfunctions, human intervention, or signal obstruction. Such data interruption or lack will directly lead to uncertainty and incompleteness in the data input of existing ship behavior reasoning methods, thereby severely weakening the ability to accurately judge and identify anomalies in ship behavior, and consequently affecting the effectiveness of maritime law enforcement and safety early warning.

[0089] To address the aforementioned issues, this application proposes a method to determine abnormal navigation status data of a target vessel even when navigation status data is incomplete, based on real-time acquired navigation status data. The core objective of this approach is to proactively identify anomalies under conditions of data uncertainty, thereby improving the robustness and accuracy of abnormal behavior detection, reducing over-reliance on data integrity, and ultimately enabling more timely early warning and intervention. This also lays the foundation for subsequent, more complex behavioral reasoning.

[0090] S103. Determine the observation sequence vector based on abnormal navigation status data.

[0091] In this step, it can be understood that after determining the abnormal navigation state data, it is necessary to determine the input of the dynamic Bayesian network based on the abnormal navigation state data, that is, to determine the observation sequence vector.

[0092] As an example, determining the observation sequence vector based on abnormal navigation state data includes: extracting motion features from the abnormal navigation state data; and generating the observation sequence vector based on these motion features. In this example, it can be understood that the generation of the observation sequence vector is based on the abnormal navigation state data. Specifically, to generate the observation sequence vector, it is necessary to first extract the motion features from the abnormal navigation state data, such as the speed and rate of change of heading. Then, based on the extracted motion features, the operating mode and intent of the target vessel are captured, enabling the generation of high-quality observation sequence vectors and significantly improving the integrity and robustness of the dynamic Bayesian network input.

[0093] S104. Input the observation sequence vector into the dynamic Bayesian network. Based on the probabilistic inference algorithm, obtain the inference result of the hidden behavior of the target ship. The dynamic Bayesian network is used to capture the temporal correlation between the navigation motion state and the hidden behavior of the target ship.

[0094] The observation sequence vector generated by S103 is input into a dynamic Bayesian network to perform inference and judgment on the hidden behavior of the target ship. Specifically, the observation sequence vector is input into the dynamic Bayesian network, which captures the temporal correlation between the target ship's navigation motion state and its hidden behavior. Then, based on the captured temporal correlation, a probabilistic inference algorithm is used to obtain the inference result of the target ship's hidden behavior.

[0095] This application embodiment acquires the navigation status data of the target vessel in real time, determines abnormal navigation status data based on the navigation status data, and then determines the observation sequence vector based on the abnormal navigation status data. This provides accurate input for the inference of the dynamic Bayesian network. The observation sequence vector is input into the dynamic Bayesian network, which captures the temporal correlation between the navigation motion state of the target vessel and its hidden behavior. Then, a probabilistic inference algorithm is used to obtain the inference result of the hidden behavior of the target vessel, thereby improving the accuracy of vessel behavior inference, reducing false alarms or missed alarms, and thus improving the effectiveness of maritime monitoring and the reliability of decision-making.

[0096] Based on the above embodiments, in some examples, the dynamic Bayesian network described in S104 is constructed in the following way: A cross-temporal topology is established between the ship's motion state nodes and hidden behavior nodes according to the temporal dependencies between adjacent time slices. The motion state nodes represent the ship's state within a specific time slice, and the hidden behavior nodes represent the actions or events that occur to the ship within a specific time slice. A two-time-slice temporal model is used to determine the conditional probability distribution of the hidden behavior nodes in the cross-temporal topology to complete the construction of the first dynamic Bayesian network. The conditional probability distribution represents the probability of the hidden behavior node occurring in the current time slice given the motion state nodes and / or hidden behavior nodes of the previous time slice. The conditional probability distribution of each hidden behavior node in the first dynamic Bayesian network is initialized to obtain the second dynamic Bayesian network. Based on the prior of structural sparsity, the parameters of the second dynamic Bayesian network are adjusted to obtain the dynamic Bayesian network, which adopts a multi-level structure with temporal dependencies.

[0097] In the above example, to construct a dynamic Bayesian network, a cross-temporal topology needs to be built based on the ship's motion state nodes, hidden behavior nodes, and the temporal dependencies between adjacent time slices. Next, a two-time-slice temporal series model is used to determine the conditional probability distribution of the hidden behavior nodes in the cross-temporal topology, completing the construction of the first dynamic Bayesian network. The first dynamic Bayesian network uses a two-time-slice temporal series model to define the conditional probability distribution of the hidden behavior nodes; the formula for the first dynamic Bayesian network is: .

[0098] in, For a moment Hidden behavioral states For a moment The ship's motion state, This refers to the hidden behavior state of the previous time slice.

[0099] By using the aforementioned cross-temporal modeling of the first dynamic Bayesian network, the constructed first dynamic Bayesian network can capture the temporal correlation between the ship's motion state and hidden behavior, thereby significantly improving the inference accuracy of ship hidden behavior, such as significantly improving the inference accuracy of illegal berthing and evasion of supervision.

[0100] Next, the conditional probability distributions of each hidden behavior node in the first dynamic Bayesian network constructed above are initialized to obtain the second dynamic Bayesian network. This means that the conditional probability distributions of each hidden behavior node are different between the first dynamic Bayesian network and the second dynamic Bayesian network.

[0101] Optionally, initializing the conditional probability distribution of each hidden behavior node in the first dynamic Bayesian network to obtain the second dynamic Bayesian network includes: determining the conditional probability distribution value of each hidden behavior node in the first dynamic Bayesian network based on the expectation-maximization algorithm; and initializing each hidden behavior node according to the conditional probability distribution value to obtain the second dynamic Bayesian network.

[0102] The conditional probability distribution values ​​of each hidden behavior node in the second dynamic Bayesian network are determined based on the expectation-maximization algorithm. Specifically, firstly, the conditional probability distribution values ​​of each hidden behavior node in the first dynamic Bayesian network are estimated based on the expectation-maximization algorithm, and then the corresponding hidden behavior nodes are initialized according to the estimated conditional probability distribution values.

[0103] The embodiment of this application estimates the conditional probability distribution of each hidden behavior node in the first dynamic Bayesian network using the expectation-maximization algorithm. This is because traditional parameter initialization methods are prone to overfitting or unstable convergence, and are difficult to adapt to parameter changes in dynamic environments.

[0104] After obtaining the second dynamic Bayesian network, a structural sparsity prior is introduced to adjust the parameters of the second dynamic Bayesian network, and the second dynamic Bayesian network with adjusted parameters is determined to be the dynamic Bayesian network described in S104.

[0105] To introduce structural sparsity priors into dynamic Bayesian networks, the core lies in modifying the optimization objective. Instead of simply minimizing data fitting errors, such as prediction accuracy, a structural penalty term is added to this error term; this penalty term is the structural sparsity regularization term.

[0106] In other words, a structural sparsity regularization term is added to the loss function of the dynamic Bayesian network to introduce a structural sparsity prior. It should be noted that introducing a structural sparsity prior into the dynamic Bayesian network can promote the stable convergence of the network parameters.

[0107] Optionally, in the aforementioned expectation-maximization algorithm, introducing this structural sparsity prior means adding a penalty term, such as based on L1 regularization or a Laplace prior, while maximizing the likelihood function. This penalty term penalizes non-zero parameters. The purpose of this is to: prevent the parameters of the dynamic Bayesian network from becoming excessively large or oscillating during training, guiding the dynamic Bayesian network towards a more stable and general solution, i.e., promoting stable convergence of the dynamic Bayesian network; "close" unimportant and redundant dependencies, i.e., promoting automatic feature selection by the dynamic Bayesian network; and improve the generalization ability of the dynamic Bayesian network to unknown data by reducing the number of effective parameters, making it more robust to the complex and ever-changing maritime environment.

[0108] In summary, the structural sparsity prior in this application embodiment is a regularization technique in the parameter learning process, used to optimize and constrain the parameter estimation results of the Conditional Probability Table (CPT) to ensure that the learned dynamic Bayesian network parameters are concise, stable, and interpretable.

[0109] In summary, this application's embodiments, based on dynamic Bayesian networks, achieve efficient ship hidden behavior inference by modeling ship motion state nodes, hidden behavior nodes, and temporal dependencies. Furthermore, the expectation-maximization algorithm is employed to optimize the initial conditional probability distribution, and structural sparsity priors are introduced, effectively improving the stability of the dynamic Bayesian network parameters and ensuring its adaptability in dynamic environments.

[0110] Furthermore, in this embodiment, a dynamic Bayesian network is used to optimize and update the conditional probability distribution in real time using the expectation-maximization algorithm, so as to improve the inference accuracy and ensure the stability of the dynamic Bayesian network. The dynamic Bayesian network adopts a multi-level structure with temporal dependencies, which can dynamically adjust the conditional probability distribution of hidden behavior nodes to adapt to complex navigation environments and ship motion patterns.

[0111] Based on the above embodiments, in some examples, S102 describes determining abnormal navigation status data based on navigation status data, including: determining the time interval between adjacent navigation status data; determining anomalies in the navigation status data based on the time interval to obtain first abnormal navigation status data; performing Kalman filtering on the navigation status data to generate predicted status values; determining anomalies in the navigation status data based on the predicted status values ​​to obtain second abnormal navigation status data; determining anomalies in the navigation status data based on shore-based radar echoes and satellite remote sensing data to obtain third abnormal navigation status data; and determining abnormal navigation status data based on the first abnormal navigation status data, the second abnormal navigation status data, and the third abnormal navigation status data.

[0112] In this example, it can be understood that the abnormal navigation status data is determined by detecting the time interval of the navigation status data, Kalman filter residual analysis, and confidence fusion of multi-source observation data.

[0113] Specifically, the system receives navigation status data and determines the time interval between adjacent abnormal navigation status data. Then, based on the deviation of the time interval from the preset normal time interval range, it performs a preliminary anomaly judgment on the navigation status data to obtain the first abnormal navigation status data.

[0114] The navigation status data is processed by Kalman filtering to obtain the predicted status value; the Kalman filter residual between the actual observed value of the navigation status data and the above predicted status value is calculated, and it is determined whether the Kalman filter residual exceeds the preset statistical threshold, so as to realize the deep anomaly verification of the navigation status data and obtain the second abnormal navigation status data.

[0115] Shore-based radar echoes and satellite remote sensing data are used as auxiliary observation sources corresponding to the navigation status data. The confidence level of these auxiliary observation sources is determined; this confidence level is a quantitative indicator reflecting the reliability, accuracy, and / or timeliness of any auxiliary observation source data. Subsequently, the consistency between the navigation status data and the shore-based radar echoes and / or satellite remote sensing data is evaluated.

[0116] Based on the consistency assessment results and the confidence levels of the aforementioned auxiliary observation source data, anomaly detection is performed on the navigation status data. Specifically, if there is a significant conflict between the navigation status data and auxiliary observation source data with high confidence, the navigation status data is further confirmed as anomalous. Conversely, even if the navigation status data shows a slight anomaly in a single detection, its potential anomaly level may be reduced if it remains consistent with high-confidence auxiliary observation source data, thereby avoiding false alarms.

[0117] Based on the determined confidence level, anomalies in the navigation status data are determined, resulting in the third abnormal navigation status data.

[0118] This application embodiment, through the time interval between adjacent navigation status data in the navigation status data, Kalman filter residual analysis, and confidence fusion of multi-source auxiliary observation data, can comprehensively utilize information from multiple sources to achieve more accurate and reliable anomaly identification in the navigation status data.

[0119] Based on the above embodiments, S104 describes inputting the observation sequence vector into a dynamic Bayesian network and inferring the hidden behavior of the target ship based on a probabilistic inference algorithm, including: inputting the observation sequence vector into the dynamic Bayesian network and using a forward-backward algorithm to obtain the posterior probability of the hidden behavior of the target ship. The forward-backward algorithm includes: recursion, Recursion, normalized posterior probability calculation and in Recursion and / or A multiple hypothesis generation mechanism based on particle filtering is introduced into the recursion; the reasoning result of hidden behavior is determined based on the posterior probability.

[0120] It is understood that the above embodiments use the forward-backward algorithm as an example of probabilistic inference to explain how to obtain the posterior probability of the hidden behavior of the target ship. The introduction of the forward-backward algorithm enables the probability of the hidden behavior node in any time slice to be accurately calculated, thereby improving the reliability and accuracy of the inference results.

[0121] Furthermore, the forward-backward algorithm through recursive process and / or The recursive process dynamically adjusts the parameters of the dynamic Bayesian network to optimize the state transition probabilities and observation models between time slices, thereby further improving inference accuracy. Optimizing the state transition probabilities and observation models is a process by which the dynamic Bayesian network, through continuous experience accumulation, corrects its understanding of behavioral habits and the meaning of cues, thus enabling it to make accurate inferences about hidden behaviors in the future.

[0122] Here, the state transition probability refers to the conditional probability distribution of the hidden behavior node in the time dimension, describing how the hidden behavior changes over time. The observation model is a component of the dynamic Bayesian network within any time slice, describing the conditional dependency between the hidden behavior node and the observation sequence vector. It is a specific set of conditional probability distributions in the Bayesian network. In other words, the observation model characterizes the spatial dependency of the hidden behavior node, describing how the hidden behavior manifests as observable features. Specifically, the observation model is manifested as follows: This defines how hidden behavior nodes affect motion state nodes within the same time slice.

[0123] Dynamic Bayesian networks utilize the forward-backward algorithm to infer the most likely hidden behavior sequence from observed clues and based on the understanding of the target ship's behavior habits, thus determining the hidden behavior inference result.

[0124] Furthermore, in complex navigation environments, the forward-backward algorithm also... recursive process and / or A multiple hypothesis generation mechanism based on particle filtering is introduced in the recursive process to handle the highly uncertain ship behavior in complex navigation environments, thereby improving the inference accuracy of the system under incomplete observation.

[0125] After determining the posterior probability, the inference result of the hidden behavior needs to be determined based on the posterior probability. For example, by selecting the hidden behavior with the highest posterior probability at any time slice, or by inferring the globally optimal hidden behavior sequence through the Viterbi algorithm, accurate inference of the hidden behavior of the target ship can be achieved.

[0126] In some examples of the above embodiments, the ship behavior reasoning method provided in this application further includes: determining whether the target ship has undergone a predetermined change based on the hidden behavior reasoning result and the observation sequence vector; if a predetermined change occurs, invoking a Bayesian structure learning algorithm to reconstruct the cross-temporal topology of the dynamic Bayesian network based on the predetermined change; and incrementally adjusting the conditional probability distribution of each hidden behavior node in the reconstructed dynamic Bayesian network based on the observation sequence vector. The predetermined change can be set according to specific circumstances, for example, the predetermined change can be set as a significant change in the ship's surrounding environment or behavior pattern.

[0127] The above example describes the automatic reconstruction of a dynamic Bayesian network through online network structure adaptation after determining the hidden behavior inference result. The automatic reconstruction of the dynamic Bayesian network through online network structure adaptation includes the following steps:

[0128] 1. Based on the latest hidden behavior inference results and observation sequence vectors, that is, based on the latest hidden behavior inference results and observation results, detect significant changes in the surrounding environment or behavior patterns of the target vessel.

[0129] In practical applications, AIS signals from ships, shore-based radar echoes, and satellite remote sensing data are acquired in real time. Each data source generates a new set of observation data through an acquisition cycle, with the data size within any time slice being 30 seconds. Specifically, it is assumed that the AIS signals received in the current time slice contain the following data: speed, position, heading, etc.

[0130] Taking AIS signals as an example, the format of each data record is as follows:

[0131]

[0132] By comparing data from the current time slice with data from the previous time slice, the rate of change of parameters such as speed, position, and heading is calculated. For example, if the ship's speed in the current time slice is 15.2 knots, while it was 16.4 knots in the previous time slice, the speed change is -1.2 knots. If this change is within a set threshold range (e.g., within 2 knots), it is considered to be insignificant; otherwise, it is considered to indicate a significant change in environmental or behavioral patterns. If multiple data sources (AIS signals, radar echoes, satellite remote sensing) show similar changes in the current period, then a significant change in environmental or behavioral patterns is considered to have occurred.

[0133] 2. Invoke Bayesian structure learning to reconstruct the topology of the dynamic Bayesian network based on the above significant changes.

[0134] Once a significant change is detected, the Bayesian structure learning algorithm is triggered to re-evaluate the topology of the dynamic Bayesian network based on the latest hidden behavior inference results and observation sequence vectors. The specific operation includes the following steps:

[0135] Change detection: Based on current data, such as a change in speed exceeding 2 knots or a change in position exceeding 500 meters, identify significant changes in ship behavior patterns.

[0136] Structure learning: Bayesian structure learning algorithms analyze the changed data and, combining the results of forward and backward inference, automatically adjust the connections between nodes in the dynamic Bayesian network. For example, if a sudden increase in ship speed is related to environmental factors, such as wind speed or tidal currents, it may be necessary to add nodes related to these environmental factors and connect them in the network structure.

[0137] Suppose the ship's behavior pattern changes from normal cruising mode to emergency obstacle avoidance mode. The Bayesian structure learning algorithm will identify new influencing factors, such as the appearance of obstacles and obstacle avoidance strategies, and update the network topology by adding relevant nodes, such as obstacle status nodes, to the network and redefining the corresponding edge connections.

[0138] 3. The conditional probability distribution parameters are incrementally adjusted by using the latest observation sequence vectors to reconstruct the dynamic Bayesian network. That is, based on the observation sequence vectors, the conditional probability distribution of each hidden behavior node in the reconstructed dynamic Bayesian network is incrementally adjusted.

[0139] After network topology reconstruction, the conditional probability distribution parameters are incrementally adjusted based on new observation data. Specifically, assuming the ship's speed changes from 10 knots to 20 knots, the conditional probability distribution related to speed in the network will be adjusted accordingly. The update of the conditional probability distribution is based on a weighted average of existing observation data and previous inference results, and is performed incrementally in the following way:

[0140] Set the current node The conditional probability distribution of (hidden behavior nodes) is: ,in This refers to current observational data (such as ship speed, heading, etc.). Using Bayes' theorem and the observational data... Combine and update the conditional probability distribution:

[0141] .

[0142] Incremental update: The conditional probability distribution at each update of the observed data. The data will be adjusted incrementally based on new observations. For example, assuming the conditional probability at the previous time step was... New data Adjustments will be made incrementally, for example, by combining previous signals with current new data. .

[0143] This incremental approach enables rapid adaptation to environmental changes, avoids the computational burden of full updates, and thus improves inference efficiency and real-time response capabilities.

[0144] The embodiments of this application can automatically reconstruct the topology of a dynamic Bayesian network when the ship's surrounding environment or behavior patterns change significantly. Based on the topology reconstruction and incremental adjustment of conditional probability distribution parameters using a Bayesian structure learning algorithm, it can efficiently adapt to complex navigation environments and changes in ship behavior patterns.

[0145] Furthermore, when automatically reconstructing dynamic Bayesian networks using online network structure adaptation, the Bayesian structure learning algorithm can be replaced with a graph neural network. This is because the Bayesian structure learning algorithm has high computational complexity, making it difficult to adapt to large-scale ship behavior reasoning. In contrast, graph neural networks can efficiently capture the complex dependencies between nodes in dynamic Bayesian networks through graph convolution mechanisms, thereby improving the accuracy and efficiency of topology reconstruction and enabling them to adapt to large-scale ship behavior reasoning.

[0146] Furthermore, when incrementally adjusting the conditional probability distribution of each hidden behavior node in the reconstructed dynamic Bayesian network, a meta-learning strategy can be introduced. Meta-learning, by learning from a small number of samples, can quickly adjust the conditional probability distribution of each hidden behavior node in the reconstructed dynamic Bayesian network, thereby improving the generalization ability of the dynamic Bayesian network to unknown behaviors.

[0147] Next, taking navigation status data as an example of AIS signals, we will explain how to use the ship behavior reasoning method provided in the embodiments of this application. Figure 2 A flowchart illustrating the ship behavior reasoning method provided in the embodiments of this application.Figure Two .like Figure 2 As shown, the method includes the following steps:

[0148] S1. Constructing a dynamic Bayesian network topology: Based on the temporal dependencies of the ship's motion state nodes, hidden behavior nodes, and adjacent time slices, construct a cross-temporal dynamic Bayesian network topology and define an initial conditional probability distribution for each node.

[0149] S2. Initialization of conditional probability distribution estimation: The expectation-maximization algorithm is used to estimate and initialize the initial conditional probability distribution, while introducing structural sparsity priors to promote stable convergence of network parameters.

[0150] S3. Real-time data acquisition and abnormal segment identification: AIS signals are acquired in real time, and combined with shore-based radar echoes and satellite remote sensing data to identify missing or abnormal segments of AIS signals, extract the motion characteristics of missing or abnormal segments of AIS signals, and generate observation sequence vectors based on the motion characteristics.

[0151] S4. Forward-backward algorithm inference: Input the observation sequence vector into the dynamic Bayesian network and use the forward-backward algorithm to calculate the posterior probability of the hidden behavior node in each time slice.

[0152] S5. Online network structure adaptation: Automatic reconstruction of dynamic Bayesian networks through online network structure adaptation.

[0153] It should be noted that the observation model described in the above embodiments is initialized in S1 and in S2. The observation model serves as a bridge between the output observation sequence vector in S3 and the input hidden behavior in S4. During the inference process in S4, the forward-backward algorithm heavily relies on the observation model. If an observation sequence vector is highly unlikely to be generated by a certain hidden behavior (i.e., ... If the probability of the hidden behavior is very low, then the probability of that behavior will be lowered. Therefore, the forward-backward algorithm can dynamically adjust the parameters of the dynamic Bayesian network through the α-recursive process and / or the β-recursive process to optimize the state transition probabilities and observation models between time slices, that is, to finely adjust the parameters using a large amount of actual observation data. The parameters of this conditional probability distribution. For example, the system may initially assume that anomalous behavior is always high speed, but actual data shows that it sometimes uses medium speed to evade detection. The forward-backward algorithm will then adjust the parameters of the observation model accordingly, so that "medium speed" also has a non-zero probability in anomalous states, thereby making the dynamic Bayesian network more accurate.

[0154] In summary, the ship behavior reasoning method provided in this application can solve the problem of difficulty in inferring hidden behaviors when data is incomplete or abnormal.

[0155] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0156] Figure 3 This is a schematic diagram of the structure of the ship behavior reasoning device provided in the embodiments of this application, as shown below. Figure 3 As shown, the ship behavior reasoning device 300 provided in this embodiment includes:

[0157] The acquisition module 301 is used to acquire the navigation status data of the target vessel in real time;

[0158] The determination module 302 is used to determine abnormal navigation status data based on navigation status data; and to determine the observation sequence vector based on the abnormal navigation status data.

[0159] The inference module 303 is used to input the observation sequence vector into the dynamic Bayesian network and obtain the inference result of the hidden behavior of the target ship based on the probabilistic inference algorithm. The dynamic Bayesian network is used to capture the temporal correlation between the navigation motion state of the target ship and the hidden behavior.

[0160] In one possible implementation, the dynamic Bayesian network is constructed in the following way:

[0161] Based on the temporal dependencies between adjacent time slices of a ship, a cross-temporal topology is established between the ship's motion state nodes and hidden behavior nodes. The motion state nodes represent the ship's state within a specific time slice, and the hidden behavior nodes represent the actions or events that occur to the ship within a specific time slice.

[0162] A two-time-slice time series model is adopted to determine the conditional probability distribution of hidden behavior nodes in the cross-time series topology, so as to complete the construction of the first dynamic Bayesian network. The conditional probability distribution represents the probability of the occurrence of hidden behavior nodes in the current time slice given the motion state nodes and / or hidden behavior nodes in the previous time slice.

[0163] Initialize the conditional probability distribution of each hidden behavior node in the first dynamic Bayesian network to obtain the second dynamic Bayesian network;

[0164] Based on the prior of structural sparsity, the parameters of the second dynamic Bayesian network are adjusted to obtain the dynamic Bayesian network, which adopts a multi-level structure with temporal dependencies.

[0165] In one possible implementation, the determining module 302 is specifically used for:

[0166] Based on the expectation-maximization algorithm, the conditional probability distribution values ​​of each hidden behavior node in the first dynamic Bayesian network are determined.

[0167] Based on the conditional probability distribution values, each hidden behavior node is initialized to obtain the second dynamic Bayesian network.

[0168] In one possible implementation, the determining module 302 is specifically used for:

[0169] Extract motion features from abnormal navigation status data;

[0170] Based on motion characteristics, an observation sequence vector is generated.

[0171] In one possible implementation, the determining module 302 is specifically used for:

[0172] Determine the time interval between adjacent navigation status data in the navigation status data;

[0173] Based on time intervals, anomaly detection is performed on the navigation status data to obtain the first abnormal navigation status data.

[0174] Kalman filtering is applied to the navigation status data to generate predicted state values;

[0175] Based on the predicted state value, anomaly determination is performed on the navigation state data to obtain the second abnormal navigation state data;

[0176] Based on shore-based radar echoes and satellite remote sensing data, anomalies are determined in the navigation status data to obtain the third abnormal navigation status data.

[0177] Based on the first abnormal navigation status data, the second abnormal navigation status data, and the third abnormal navigation status data, the abnormal navigation status data is determined.

[0178] In one possible implementation, the inference module 303 is specifically used for:

[0179] The observation sequence vector is input into a dynamic Bayesian network, and the forward-backward algorithm is used to obtain the posterior probability of the hidden behavior of the target ship. The forward-backward algorithm includes: recursion, Recursion, normalized posterior probability calculation and in Recursion and / or A multiple hypothesis generation mechanism based on particle filtering is introduced into the recursion;

[0180] The reasoning result for the hidden behavior is determined based on the posterior probability.

[0181] In one possible implementation, the ship behavior reasoning device further includes an adjustment module (not shown), which is specifically used for:

[0182] Based on the hidden behavior inference results and the observation sequence vector, determine whether the target vessel has undergone the set changes;

[0183] If the settings change, the Bayesian structure learning algorithm is invoked to reconstruct the cross-temporal topology of the dynamic Bayesian network based on the changes in settings.

[0184] Based on the observed sequence vector, the conditional probability distribution of each hidden behavior node in the reconstructed dynamic Bayesian network is incrementally adjusted.

[0185] The ship behavior reasoning device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0186] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. For example, a processing module can be a separate processing element, or it can be integrated into an integrated circuit within the above device. Alternatively, it can be stored as program code in the device's memory, and its functions can be called and executed by a processing element. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0187] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented by calling program code through a processing element, that processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together to implement a System-On-a-Chip (SOC).

[0188] Figure 4This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 400 provided in this application embodiment may include: a processor 401, and a memory 402 communicatively connected to the processor, wherein:

[0189] The memory stores the instructions that the computer executes;

[0190] The processor executes computer execution instructions stored in memory to implement the method described in the foregoing method embodiments.

[0191] It should be understood that processor 401 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the application can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor. Memory 402 may include high-speed random access memory (RAM), and may also include non-volatile memory (NVM), such as at least one disk storage device, or a USB flash drive, external hard drive, read-only memory, disk, or optical disc, etc.

[0192] Optionally, the electronic device 400 may also include a communication interface 403. In specific implementations, if the communication interface 403, memory 402, and processor 401 are implemented independently, they can be interconnected via a bus to complete communication. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc., but this does not imply that there is only one bus or one type of bus.

[0193] Optionally, in a specific implementation, if the communication interface 403, memory 402 and processor 401 are integrated on a single integrated circuit, then the communication interface 403, memory 402 and processor 401 can communicate through an internal interface.

[0194] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the methods described in any of the foregoing embodiments.

[0195] It is understood that the computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0196] An exemplary computer-readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the computer-readable storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can reside in an ASIC. Alternatively, the processor and the computer-readable storage medium can exist as discrete components in an electronic device.

[0197] The integrated modules implemented as software functional modules described above can be stored in a computer-readable storage medium. These software functional modules, stored in a computer-readable storage medium, include several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this application.

[0198] This application also provides a computer program product, including a computer program that, when executed, implements the method described in any of the foregoing embodiments.

[0199] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0200] It should be further noted that although the steps in the flowchart are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0201] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as these combinations of technical features do not contradict each other, they should be considered within the scope of this specification.

[0202] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0203] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for reasoning about ship behavior, characterized in that, include: Real-time acquisition of navigation status data of the target vessel; Based on the navigation status data, abnormal navigation status data is determined; Based on the abnormal navigation status data, determine the observation sequence vector; The observed sequence vector is input into a dynamic Bayesian network, and the hidden behavior inference result of the target ship is obtained based on the probabilistic inference algorithm. The dynamic Bayesian network is used to capture the temporal correlation between the navigation motion state and the hidden behavior of the target ship.

2. The method according to claim 1, characterized in that, The dynamic Bayesian network is constructed in the following way: Based on the temporal dependencies between adjacent time slices of the ship, a cross-temporal topology is established between the ship's motion state nodes and hidden behavior nodes. The motion state nodes represent the ship's state within a specific time slice, and the hidden behavior nodes represent the actions or events that occur to the ship within a specific time slice. A two-time-slice time-series model is used to determine the conditional probability distribution of hidden behavior nodes in the cross-time-series topology, so as to complete the construction of the first dynamic Bayesian network. The conditional probability distribution represents the probability of a hidden behavior node occurring in the current time slice given the motion state node and / or hidden behavior node in the previous time slice. Initialize the conditional probability distribution of each hidden behavior node in the first dynamic Bayesian network to obtain the second dynamic Bayesian network; Based on the prior of structural sparsity, the parameters of the second dynamic Bayesian network are adjusted to obtain the dynamic Bayesian network, which adopts a multi-level structure with temporal dependencies.

3. The method according to claim 2, characterized in that, The process of initializing the conditional probability distribution of each hidden behavior node in the first dynamic Bayesian network to obtain the second dynamic Bayesian network includes: Based on the expectation-maximization algorithm, the conditional probability distribution values ​​of each hidden behavior node in the first dynamic Bayesian network are determined. Based on the conditional probability distribution values, each hidden behavior node is initialized to obtain the second dynamic Bayesian network.

4. The method according to any one of claims 1 to 3, characterized in that, The step of determining the observation sequence vector based on the abnormal navigation state data includes: Extract the motion features from the abnormal navigation state data; Based on the motion characteristics, an observation sequence vector is generated.

5. The method according to any one of claims 1 to 3, characterized in that, The step of determining abnormal navigation status data based on the navigation status data includes: Determine the time interval between adjacent navigation status data in the navigation status data; Based on the time interval, the navigation status data is anomaly determined to obtain first abnormal navigation status data; The navigation status data is processed by Kalman filtering to generate predicted status values; Based on the predicted state value, the navigation state data is anomaly determined to obtain second abnormal navigation state data; Based on shore-based radar echoes and satellite remote sensing data, anomaly determination is performed on the navigation status data to obtain third abnormal navigation status data; The abnormal navigation status data is determined based on the first abnormal navigation status data, the second abnormal navigation status data, and the third abnormal navigation status data.

6. The method according to any one of claims 1 to 3, characterized in that, The step of inputting the observed sequence vector into a dynamic Bayesian network and inferring the hidden behavior of the target vessel based on a probabilistic inference algorithm includes: The observation sequence vector is input into the dynamic Bayesian network, and the forward-backward algorithm is used to obtain the posterior probability of the hidden behavior of the target ship. The forward-backward algorithm includes: α recursion, β recursion, normalized posterior probability calculation, and the introduction of a particle filter-based multiple hypothesis generation mechanism in α recursion and / or β recursion. The reasoning result of the hidden behavior is determined based on the posterior probability.

7. The method according to any one of claims 1 to 3, characterized in that, Also includes: Based on the hidden behavior inference results and the observation sequence vector, determine whether the target vessel has undergone a predetermined change; If a setting change occurs, the Bayesian structure learning algorithm is invoked to reconstruct the cross-temporal topology of the dynamic Bayesian network based on the setting change. Based on the observed sequence vector, the conditional probability distribution of each hidden behavior node in the reconstructed dynamic Bayesian network is incrementally adjusted.

8. A ship behavior reasoning device, characterized in that, include: The acquisition module is used to acquire the navigation status data of the target vessel in real time. The determination module is configured to determine abnormal navigation status data of the target vessel based on the navigation status data; and to determine an observation sequence vector based on the abnormal navigation status data. The inference module is used to input the observation sequence vector into the dynamic Bayesian network and obtain the inference result of the hidden behavior of the target ship based on the probabilistic inference algorithm. The dynamic Bayesian network is used to capture the temporal correlation between the navigation motion state and the hidden behavior of the target ship.

9. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed, are used to implement the method as described in any one of claims 1-7.

Images