An intelligent beacon interaction method and system based on space-time perception and causal reasoning

Abstract

The application discloses an intelligent navigation mark interaction method and system based on space-time perception and causal reasoning, which comprises the following steps: acquiring ship state data and hydrological data; performing space-time alignment on the ship state data and the hydrological data, and performing feature fusion based on a cross-scale attention mechanism; performing reasoning on the fused features based on a causal structure learning algorithm for incomplete observation data to obtain a preliminary decision scheme; performing decision enhancement on the preliminary decision scheme to obtain an enhanced situation representation; and finally deciding a scheme based on the enhanced situation representation. The application uses the cross-scale attention mechanism to perform feature fusion on the ship state data and the hydrological data, and uses the causal structure learning algorithm to reason the decision scheme, and then uses a dynamic memory enhancement mechanism to optimize the decision, so that the multi-scale space-time data can be effectively processed, reliable decisions can be provided through the causal reasoning mechanism, and the decision quality can be improved through the dynamic memory enhancement mechanism.

Description

Technical Field

[0001] This invention relates to the field of smart shipping, specifically to an intelligent navigation mark interaction method and system based on spatiotemporal perception and causal reasoning. Background Technology

[0002] As a core infrastructure for maintaining maritime traffic safety, the level of intelligence of navigation aids directly determines the operational efficiency and safety assurance capabilities of the shipping system. However, existing intelligent navigation aid systems face multiple technical bottlenecks when dealing with the complex and dynamic scenarios of modern waterways. Modern waterways are essentially multi-layered, multi-scale spatiotemporal coupled systems, encompassing dynamic processes across different time dimensions, including second-level ship maneuvering response, minute-level traffic flow evolution, and hourly tidal channel changes.

[0003] Traditional navigation aid systems can predict waterway traffic conditions based on neural networks or convolutional networks. However, they can typically only extract feature patterns at a single scale, making it difficult to establish the intrinsic correlation between instantaneous collision avoidance actions and long-term traffic conditions. For example, in the high-density waters of the lower Yangtze River, the micro-avoidance behavior of a single vessel can trigger regional congestion through a cascading effect. However, existing systems lack the ability to model cross-scale coupling and cannot accurately predict such chain reactions.

[0004] Furthermore, existing technologies rely excessively on statistical correlations rather than causal mechanisms, which can easily lead to decision-making errors. For example, the system may identify a statistical correlation between "foggy weather" and "ship deceleration," but it cannot understand that "reduced visibility" is the root cause of deceleration. Moreover, when the training data does not cover new scenarios, the system is also unable to produce accurate judgments. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent navigation mark interaction method and system based on spatiotemporal perception and causal reasoning. It utilizes a cross-scale attention mechanism to fuse features of ship status data and hydrological data, and infers decision-making schemes based on causal structure learning algorithms. Furthermore, it optimizes decisions through a dynamic memory enhancement mechanism. This approach can effectively process multi-scale spatiotemporal data, provide reliable decisions through causal reasoning, and improve decision quality through dynamic memory enhancement.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] On the one hand, an intelligent navigation beacon interaction method based on spatiotemporal perception and causal reasoning is provided, which includes the following steps:

[0008] Real-time acquisition of ship status data and hydrological data;

[0009] The acquired ship status data and hydrological data are spatiotemporally aligned, and feature fusion is performed on the spatiotemporally aligned ship status data and hydrological data based on a cross-scale attention mechanism to obtain fused features;

[0010] The fusion features are inferred based on the causal structure learning algorithm for incomplete observation data to obtain a preliminary decision scheme corresponding to the fusion features;

[0011] The initial decision-making scheme is enhanced to obtain an enhanced situational representation;

[0012] And, the final decision-making scheme based on the enhanced situational representation.

[0013] On the other hand, an intelligent navigation beacon interaction system based on spatiotemporal perception and causal reasoning is also provided, which includes:

[0014] The sensor module is used to acquire real-time ship status data and hydrological data;

[0015] The feature fusion module is used to perform spatiotemporal alignment on the acquired ship status data and hydrological data, and to perform feature fusion on the spatiotemporally aligned ship status data and hydrological data based on a cross-scale attention mechanism to obtain fused features.

[0016] The model reasoning module uses a causal structure learning algorithm for incomplete observation data to reason about the fused features in order to obtain a preliminary decision scheme corresponding to the fused features.

[0017] A dynamic memory enhancement module is used to enhance the initial decision-making scheme to obtain an enhanced situational representation.

[0018] And a decision fusion module, which outputs a final decision scheme based on the enhanced situational representation.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] This invention utilizes a cross-scale attention mechanism to fuse features of ship status data and hydrological data, and infers decision-making schemes based on a causal structure learning algorithm. Furthermore, it optimizes decisions through a dynamic memory enhancement mechanism. This effectively integrates multi-scale spatiotemporal information, explores causal mechanisms, and utilizes historical experience, thereby solving the problems of insufficient multi-scale modeling, lack of causal mechanisms, and lack of experiential knowledge inheritance in existing technologies. It can effectively process multi-scale spatiotemporal data, provide reliable decisions through causal reasoning mechanisms, and improve decision quality through dynamic memory enhancement mechanisms. Attached Figure Description

[0021] Figure 1This is a flowchart illustrating the steps of the intelligent navigation beacon interaction method based on spatiotemporal perception and causal reasoning in this invention.

[0022] Figure 2 This is a schematic diagram of the intelligent navigation beacon interaction system based on spatiotemporal perception and causal reasoning in this invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1:

[0025] like Figure 1 As shown, this embodiment provides an intelligent navigation beacon interaction method based on spatiotemporal perception and causal reasoning, which includes the following steps:

[0026] S1. Based on sensor module 1, real-time acquisition of ship status data and hydrological data is performed. Sensor module 1 includes radar, AIS system, camera, etc. The ship status data includes ship position, speed, and heading, which can be collected by navigation equipment and motion sensors mounted on the ship itself. For example, the ship's position information can be obtained through a Global Positioning System (GPS) receiver, the ship's speed through a log, and the ship's heading through an electronic compass or gyroscope. Simultaneously, the hydrological data includes water flow rate, velocity, and water level, which can be collected by monitoring equipment deployed at specific locations in the waterway. For example, the water flow velocity can be obtained through a simple current meter, the water level through a water level sensor, and the water flow rate through a flow meter.

[0027] Radar, AIS system, and camera, as specific components of sensor module 1, can significantly improve the comprehensiveness, accuracy, and reliability of ship status and hydrological data acquisition. For example, radar can provide high-precision ship motion parameters (such as position, speed, and heading), ensuring accurate perception of the ship's micro-dynamics; the AIS system provides standardized information such as the ship's identity, heading, and speed, enabling effective identification and tracking of the ship's macro-state, and can also identify other surrounding ships to construct complete traffic flow information; the camera supplements the intuitive perception of hydrological data (such as water flow direction and water surface conditions) with visual information, and can assist in identifying obstacles or abnormal situations in complex environments. This collaborative work of multi-source heterogeneous sensors overcomes the limitations of single sensors in data acquisition. For example, radar can still work effectively in adverse weather conditions, the AIS system provides standardized information, and the camera provides rich visual details.

[0028] Therefore, sensor module 1 can provide more complete, accurate and reliable raw data input for the subsequent feature fusion module 2, so that feature fusion module 2 can perform spatiotemporal alignment and feature fusion based on more comprehensive information, and finally provide high-quality input for model inference module 3 and dynamic memory enhancement module 4, which significantly improves the decision accuracy and reliability of intelligent navigation mark interaction method in complex shipping environment and effectively avoids the risk of misjudgment caused by incomplete or inaccurate data.

[0029] S2. The acquired ship status data and hydrological data are spatiotemporally aligned through feature fusion module 2, and feature fusion is performed on the spatiotemporally aligned ship status data and hydrological data based on cross-scale attention mechanism to obtain fused features.

[0030] Specifically, the spatiotemporal alignment can be achieved through timestamp matching and data resampling techniques. For example, data collected by different sensors at different frequencies can be unified onto a common time axis, or data loss can be handled through linear interpolation.

[0031] Furthermore, after completing the spatiotemporal alignment, the feature fusion module 2 performs feature fusion on the ship status data and hydrological data. For example, the ship status data and hydrological data can be spliced ​​into a feature vector by feature splicing. Furthermore, feature fusion can be performed based on a cross-scale attention mechanism and a preset weight allocation rule to obtain fused features containing pre-integrated ship and environmental information.

[0032] S3, Model Inference Module 3 uses a causal structure learning algorithm (such as FRITL algorithm) for incomplete observation data to infer the fused features in order to obtain a preliminary decision scheme corresponding to the fused features:

[0033] For example, in this embodiment, a causal graph model is first constructed, in which the nodes of the causal graph model represent different variables (such as ship behavior, hydrological conditions, etc.), and the edges represent causal relationships;

[0034] When ship data and / or hydrological data are incomplete, missing data can be estimated and causal structures can be learned based on methods such as maximum likelihood estimation or Bayesian inference.

[0035] Finally, the model reasoning module 3 analyzes the current fusion features based on the learned causal relationships, thereby generating preliminary decision-making schemes to reflect possible action suggestions under the current situation;

[0036] Therefore, this embodiment can identify causal relationships between variables from datasets that may have missing or incomplete data, and generate preliminary decision-making schemes based on the learned causal relationships;

[0037] S4. The preliminary decision-making scheme generated by the model reasoning module 3 is enhanced by the dynamic memory enhancement module 4 to obtain the enhanced situational representation h. augmented ;

[0038] The decision enhancement in this embodiment aims to optimize and improve the initial decision-making scheme by introducing historical experience or knowledge. Specifically, the decision enhancement includes the following steps:

[0039] S41. Encode historical interaction events into memory units m i (Memory Cell), and m i = {e i ,c i ,a i ,r i ,t i}, where e i ∈R d Let represent the episode embedding vector of the i-th historical interaction event. In this embodiment, the episode embedding vector e i The fused features are used to represent the core situational information of the event; c i Indicates category information; a i This represents the decision (Action) made in response to historical interaction events; r i This represents the reward for making this decision, which can be a success / failure label or a continuous value rating; t i Represents a timestamp;

[0040] The historical interaction events refer to the adoption of corresponding historical decision-making schemes for historical ship status data and historical hydrological data.

[0041] Repeat the above steps to obtain a number of memory units and construct a dynamic memory bank M={m1,m2,...,m...} containing the number of memory units. N}, where N is the total number of memory units in the dynamic memory bank M;

[0042] Therefore, this embodiment utilizes the memory unit m i Complex historical interaction events are encapsulated into reusable knowledge blocks, which can be used as reference experience to improve the quality of decision-making.

[0043] S42. Obtain the current interactive event and the current memory unit m based on the following formula. i similarity s i And the similarity s i Convert to attention weight α i :

[0044]

[0045]

[0046] Among them, e current τ is the context embedding vector for the current interaction event; τ is a temperature parameter used to control the sharpness of the weight distribution. The smaller τ is, the sharper the distribution, and the more the system focuses on the memory units m with higher similarity. i N represents the total number of memory units; s j For the current interaction event and the j-th memory unit m j Similarity;

[0047] S43. Construct the following empirical vector P according to the following formula;

[0048]

[0049] S44. Based on the following formula, the empirical vector P and the original situation representation h are fused to obtain the enhanced situation representation h. augmented :

[0050]

[0051] Where [h;P] represents the vector concatenation operation; W f b f These are the learnable weight matrix and bias vector, respectively; h is the original state representation (i.e., the preliminary decision scheme).

[0052] S45. Update the dynamic memory bank M, wherein the update includes one or more of the following: adding memory units, merging memory units, and memory forgetting processing;

[0053] The memory unit fusion includes the following steps:

[0054] The i-th memory unit m is processed according to the following formula. i The context embedding vector e i and the j-th memory unit m j The context embedding vector e j Perform fusion to obtain the fused scenario embedding vector e merged :

[0055] e merged =β*e i +(1-β)*e j

[0056] Where β is the fusion weight, which can be determined based on the i-th memory unit m. i The j-th memory unit m j The reward r is set accordingly, with memory units that receive higher rewards having greater weight; meanwhile, the i-th memory unit m i The j-th memory unit m j The similarity is greater than or equal to 0.8;

[0057] The memory forgetting process includes the following steps:

[0058] The memory utility value U of a memory unit is calculated based on the following formula. i :

[0059] U i =λ freq ·freq(m i )+λ recency ·recency(m i )+λ reward ·r i

[0060] Where freq(m i ) represents the memory unit m i access frequency; recency(m i ) represents the memory unit m i Recent visit time rating; r i For memory unit m i The reward; λ freq , λ recency , λ reward All are weighting coefficients;

[0061] And, when the memory utility value U i When the preset conditions are not met, the memory unit m i Eliminate;

[0062] Therefore, this step employs a content-based retrieval mechanism, which can quickly retrieve semantically similar past scenarios, actions taken, and their final results from the memory based on the current real-time situation. These retrieved "experiences" are used to enhance the current decision-making process, enabling the system to learn from past successes, avoid repeating failures, and achieve the reuse and accumulation of "experience," driving continuous system evolution.

[0063] S5. The enhanced situational representation h augmented Input the decision fusion module 5 and output the final decision scheme, which specifically includes the following steps:

[0064] S51, The enhanced situational representation h augmented Input Decision Fusion Module 5:

[0065] S52. The decision fusion module 5 obtains prediction results at different time scales based on different prediction models. For example, in this embodiment, the decision fusion module 5 is equipped with a first prediction model, a second prediction model, and a third prediction model. The first prediction model includes a physical dynamics model or a lightweight neural network (such as a Kalman filter network or a one-dimensional temporal convolutional network). The second prediction model includes a recurrent neural network based on an encoder-decoder structure (such as LSTM-ED (LSTM encoder-decoder) or a temporal Transformer network). The third prediction model includes a graph neural network or a multi-task learning model that combines external knowledge (such as a waterway chart or weather).

[0066] Therefore, the first prediction model, the second prediction model, and the third prediction model respectively correspond to the enhanced situational representation h. augmented Processing is performed to obtain the first prediction result P. S Second prediction result I m And the third prediction result R l ;

[0067] Wherein, the first prediction result P S Including the first time period ΔT S Ship motion state parameters within (e.g., within 30 seconds) The ship motion state parameters include one or more of the following: position, speed, and heading;

[0068] The second prediction result I m Including the second time period ΔT m Probability distribution of a ship's intentions within (e.g., within 5 minutes) I m and / or navigation trajectory T m The intent includes one or more of the following: going straight, following another, changing lanes, turning, or parking.

[0069] The third prediction result R l Including the third time period ΔT l Risk heatmap and / or probability of occurrence of key events (such as estimated arrival time of the vessel, potential conflict points where accidents (such as collisions) occur) within 1 hour;

[0070] S53. Based on the following formula, an uncertainty measure is assigned to the first prediction result, the second prediction result, and the third prediction result:

[0071]

[0072] Where, σ total 2 The total uncertainty is σ. epistemic 2 σ aleatoric 2 These are epistemic uncertainty caused by insufficient model knowledge and aleatoric uncertainty caused by data noise.

[0073] Furthermore, σ epistemic The standard deviation of cognitive uncertainty can be obtained through methods such as Monte Carlo Dropout (MCDropout) or ensemble learning, as in this embodiment. , where {P (K)} represents the prediction result of the Kth random forward propagation, where K is the number of random forward propagations, and Var(·) represents the variance calculation; σ aleatoric The standard deviation, representing random uncertainty, can be learned from the predicted mean and variance directly output by the model, as in this embodiment. , The predicted mean of the model output. f(·) represents the prediction variance of the model output, f(·) represents the prediction model, and x represents the input feature.

[0074] S54. The first prediction result, the second prediction result, and the third prediction result, which are respectively assigned uncertainty measures, are compared with the causal mechanism of the causal graph to obtain the corresponding consistency check results; the causal graph is used to represent the causal influence relationship between entities in the shipping scenario;

[0075] S55. Based on the following formula, perform uncertainty weighting calculation and fusion prediction to obtain the fusion prediction result:

[0076]

[0077]

[0078] Among them, w i The fusion weight is the weight of the i-th prediction result (i.e., any one of the first, second, and third prediction results). This fusion weight can be adjusted according to the consistency check result corresponding to the prediction result. For example, if the current prediction result is in serious conflict with the causal mechanism, its weight will be reduced. , P represents the uncertainty variance of the i-th and j-th prediction results; fuse For fusion prediction results; , These are the second and third prediction results, respectively, after scale alignment and causal constraint adjustment; w s w m w l These are the fusion weights of the first, second, and third prediction results, respectively, which are processed by the aforementioned w. i Obtain the calculation formula;

[0079] S56. Generate a final decision scheme based on the fusion prediction results, and perform objective optimization on the final decision scheme based on the constrained optimization model; wherein, the constrained optimization model can be an objective function, and the objective optimization includes: minimizing the risk of ship accidents (such as collisions, groundings, etc.), and / or minimizing the ship's sailing time, and / or minimizing the ship's fuel consumption or violent maneuvering.

[0080] In addition, step S6, output module 6 generates a decision report based on the final decision scheme for achieving the target optimization. The decision report includes one or more of the following: decision basis, causal chain, experience reference, uncertainty explanation, etc.

[0081] As can be seen from the above, this embodiment acquires ship status and hydrological data in real time, uses a cross-scale attention mechanism for feature fusion, and uses a causal structure learning algorithm to reason about decision schemes. Then, it optimizes decisions through a dynamic memory enhancement mechanism. This can effectively integrate multi-scale spatiotemporal information, explore causal mechanisms, and utilize historical experience, thereby solving the problems of insufficient multi-scale modeling, lack of causal mechanisms, and lack of inheritance of experience knowledge in the prior art. It has the ability to effectively process multi-scale spatiotemporal data, provide reliable decisions through causal reasoning mechanisms, and improve decision quality by using dynamic memory enhancement mechanisms.

[0082] Example 2:

[0083] This embodiment provides an intelligent navigation beacon interaction system based on spatiotemporal perception and causal reasoning, which can implement the intelligent navigation beacon interaction method described in Embodiment 1, such as... Figure 2 As shown, the intelligent navigation beacon interaction system includes:

[0084] Sensor module 1 is used to acquire ship status data and hydrological data in real time, and its process is the same as step S1.

[0085] Feature fusion module 2 is used to perform spatiotemporal alignment on the acquired ship status data and hydrological data, and to perform feature fusion on the spatiotemporally aligned ship status data and hydrological data based on a cross-scale attention mechanism to obtain fused features. The process is the same as step S2.

[0086] Model reasoning module 3, which uses a causal structure learning algorithm (such as FRITL algorithm) for incomplete observation data to reason about the fused features in order to obtain a preliminary decision scheme corresponding to the fused features, is the same as step S3.

[0087] The dynamic memory enhancement module 4 is used to enhance the decision-making scheme of the preliminary decision-making scheme to obtain the enhanced situational representation. Its process is the same as step S4.

[0088] The decision fusion module 5 outputs the final decision scheme based on the enhanced situational representation, and its process is the same as step S5.

[0089] And output module 6, which generates a decision report based on the final decision scheme that achieves the target optimization.

[0090] In summary, this invention utilizes a cross-scale attention mechanism to fuse features of ship status data and hydrological data, and infers decision-making schemes based on a causal structure learning algorithm. Furthermore, it optimizes decisions through a dynamic memory enhancement mechanism. This effectively integrates multi-scale spatiotemporal information, explores causal mechanisms, and leverages historical experience, thereby addressing the problems of insufficient multi-scale modeling, lack of causal mechanisms, and lack of experiential knowledge inheritance in existing technologies. It can effectively process multi-scale spatiotemporal data, provide reliable decisions through causal reasoning mechanisms, and improve decision quality through dynamic memory enhancement mechanisms.

[0091] It should be noted that, in this document, terms such as "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0092] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart navigation beacon interaction method based on spatiotemporal perception and causal reasoning, characterized in that, Includes the following steps: Real-time acquisition of ship status data and hydrological data; The acquired ship status data and hydrological data are spatiotemporally aligned, and feature fusion is performed on the spatiotemporally aligned ship status data and hydrological data based on a cross-scale attention mechanism to obtain fused features; The fusion features are inferred based on the causal structure learning algorithm for incomplete observation data to obtain a preliminary decision scheme corresponding to the fusion features; The initial decision-making scheme is enhanced to obtain an enhanced situational representation; And, the final decision-making scheme based on the enhanced situational representation.

2. The intelligent navigation beacon interaction method as described in claim 1, characterized in that, The decision enhancement process for the initial decision-making scheme to obtain an enhanced situational awareness includes the following steps: Encode historical interaction events into memory units m i And m i = {e i ,c i ,a i ,r i ,t i }, where e i Indicates the i-th historical intersection Context embedding vectors for mutual events; c i Indicates category information; a i Indicates the decisions made in response to historical interaction events; r i Indicates the reward for adopting this decision; t i Represents a timestamp; The current interaction event and the current memory unit m are obtained based on the following formula. i similarity s i And the similarity s i Convert to attention weight α i : Among them, e current The context embedding vector for the current interaction event; τ is the temperature parameter; N is the total number of memory units; s j For the current interaction event and the j-th memory unit m j Similarity; Construct the following empirical vector P according to the following formula; The empirical vector P and the original situation representation h are fused based on the following formula to obtain the enhanced situation representation h. augmented : Where [h;P] represents the vector concatenation operation; W f b f ...

3. The intelligent navigation beacon interaction method as described in claim 2, characterized in that, To enhance the preliminary decision-making scheme and obtain an enhanced situational awareness, the following steps are also included: The dynamic memory bank is updated, and the update includes one or more of the following: adding memory units, merging memory units, and handling memory forgetting.

4. The intelligent navigation beacon interaction method as described in claim 3, characterized in that, The memory cell fusion includes the following steps: The i-th memory unit m is processed according to the following formula. i The context embedding vector e i and the j-th memory unit m j The context embedding vector e j Perform fusion to obtain the fused scenario embedding vector e merged : And merged =β*e i +(1-β)*e j Where β is the fusion weight; and the i-th memory unit m i The j-th memory unit m j The similarity is greater than or equal to 0.

8.

5. The intelligent navigation beacon interaction method as described in claim 3, characterized in that, The memory forgetting process includes the following steps: The memory utility value U of a memory unit is calculated based on the following formula. i : U i =λ freq ·freq(m i )+λ recency ·recency(m i )+λ reward ·r i Where freq(m i ) represents the memory unit m i access frequency; recency(m i ) represents the memory unit m i Recent visit time rating; r i For memory unit m i The reward; λ freq , λ recency , λ reward All are weighting coefficients; And, when the memory utility value U i When the preset conditions are not met, the memory unit m i Eliminate them.

6. The intelligent navigation beacon interaction method as described in claim 1, characterized in that, The final decision-making scheme based on the enhanced situational awareness includes the following steps: Based on different prediction models, prediction results at different time scales are obtained. The prediction models include a first prediction model, a second prediction model, and a third prediction model. The prediction results at different time scales include the first prediction result, the second prediction result, and the third prediction result. The uncertainty weighting calculation and fusion prediction are performed based on the following formula to obtain the fusion prediction result: Among them, w i Let be the fusion weight for the i-th prediction result; , P represents the uncertainty variance of the i-th and j-th prediction results; fuse For fusion prediction results; , These are the second and third prediction results, respectively, after scale alignment and causal constraint adjustment; w s w m w l These are the fusion weights for the first prediction result, the second prediction result, and the third prediction result, respectively. Furthermore, a final decision scheme is generated based on the fusion prediction results.

7. The intelligent navigation beacon interaction method as described in claim 6, characterized in that, The first prediction result includes the first time period ΔT S The ship's motion state parameters within the time period; the second prediction result includes the second time period ΔT m The probability distribution of the vessel's intentions and / or its navigation trajectory; the third prediction result includes a third time period ΔT. l Risk heatmap and / or probability of occurrence of key events within the area.

8. The intelligent navigation beacon interaction method as described in claim 6, characterized in that, The final decision-making scheme based on the enhanced situational awareness also includes the following steps: The fusion weight w is adjusted based on the consistency check results. i Adjustments are made, wherein the consistency check result is determined according to the following steps: The uncertainty measure is assigned to the first, second, and third prediction results based on the following formula: Where, σ total 2 The total uncertainty is σ. epistemic 2 σ aleatoric 2 These are respectively cognitive uncertainty and accidental uncertainty; The first, second, and third prediction results, each assigned an uncertainty measure, are compared with the causal mechanism of the causal graph to obtain the corresponding consistency check results.

9. The intelligent navigation beacon interaction method as described in claim 1, characterized in that, The intelligent navigation beacon interaction method further includes the following steps: generating a decision report based on the final decision scheme that has completed the target optimization, wherein the target optimization is completed based on a constraint-based optimization model.

10. An intelligent navigational aid interaction system based on spatiotemporal perception and causal reasoning, characterized in that, include: The sensor module is used to acquire real-time ship status data and hydrological data; The feature fusion module is used to perform spatiotemporal alignment on the acquired ship status data and hydrological data, and to perform feature fusion on the spatiotemporally aligned ship status data and hydrological data based on a cross-scale attention mechanism to obtain fused features. The model reasoning module uses a causal structure learning algorithm for incomplete observation data to reason about the fused features in order to obtain a preliminary decision scheme corresponding to the fused features. A dynamic memory enhancement module is used to enhance the initial decision-making scheme to obtain an enhanced situational representation. And a decision fusion module, which outputs a final decision scheme based on the enhanced situational representation.

Images