Ship navigation risk assessment system based on multi-source heterogeneous data fusion

Abstract

The invention relates to the technical field of ship navigation risk assessment, in particular to a ship navigation risk assessment system based on multi-source heterogeneous data fusion, which comprises a multi-source data integration module, a spatial-temporal feature mapping module, a dynamic risk detection module, a linkage decision control module and a feedback optimization module. According to the method, standardized operation data is generated through multi-source data cleaning and fusion, a spatial-temporal feature distribution map is generated by using a multi-dimensional dynamic clustering algorithm, a risk index set is extracted in combination with adaptive boundary adjustment and a nonlinear optimization algorithm, and accurate path planning and real-time regulation are realized. In addition, a global sensitivity analysis framework and an early warning module are introduced into the system, and the ship navigation safety and reliability are improved. According to the method, the risk prediction accuracy can be remarkably improved, the navigation accident probability is reduced, the navigation efficiency is optimized, and safe operation of the ship is guaranteed.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent transportation and shipping safety, and specifically relates to a ship navigation risk assessment system based on multi-source heterogeneous data fusion. BACKGROUND

[0002] With the increase of ship navigation density and the change of complex water environment, the problem of ship navigation safety is increasingly prominent. The existing ship navigation risk assessment technology mainly relies on single data source or risk modeling in specific scenarios, and the ability to comprehensively reflect multi-source heterogeneous data has certain limitations. Therefore, how to realize more accurate and real-time ship navigation risk assessment through multi-source heterogeneous data fusion technology has become a problem to be solved.

[0003] After searching, a ship collision risk assessment and prediction method and device (publication date: June 13, 2023) with publication number CN115331486B realizes high-precision prediction of ship collision risk by obtaining ship AIS data of the target area, combining clustering analysis of approximate collision position and extraction of navigation trajectory. However, this technical solution mainly uses AIS data and fails to fully integrate other types of data (such as meteorological data, hydrological data, radar data, etc.), which has certain limitations in adaptability in complex water environment. In addition, in the multi-ship dense interaction scene, the risk prediction ability of dynamic obstacles is still insufficient, and it may be difficult to fully capture potential collision risks.

[0004] After searching, a ship waterway navigation risk early warning method and system (publication date: August 6, 2024) with publication number CN118314770B generates high-accuracy risk warning information by real-time acquisition of navigation parameters of the target ship, combining distance risk analysis of static obstacles and intention navigation route prediction of dynamic obstacles. However, although this technical solution has certain advantages in risk assessment of static and dynamic obstacles, its data sources are still relatively single, mainly relying on navigation environment images and AIS data, lacking deep fusion analysis of multi-source heterogeneous data (such as weather, tides, and ocean currents). In addition, this scheme has room for improvement in multi-scale risk quantization modeling, and may have certain limitations in diversified risk assessment in different water environments.

[0005] The above problems show that the existing ship navigation risk assessment technology still has room for further improvement in data fusion capability, multi-scale risk modeling, and complex environment adaptability. Therefore, the present application provides a ship navigation risk assessment system based on multi-source heterogeneous data fusion, aiming to integrate AIS data, meteorological data, hydrological data, radar data and other data sources to build a more comprehensive and accurate risk assessment model, thereby improving the accuracy, real-time performance and adaptability of ship navigation risk assessment, and meeting the needs of modern shipping industry for efficient and intelligent risk management. SUMMARY

[0006] The present application aims to provide a ship navigation risk assessment system based on multi-source heterogeneous data fusion to solve the problems raised in the background art.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0008] A ship navigation risk assessment system based on multi-source heterogeneous data fusion, the system comprising: a multi-source data integration module for obtaining a plurality of operating data sets in a target water area, and processing the operating data sets according to a preset multi-source data cleaning strategy to generate standardized operating data; a space-time feature mapping module for mapping the standardized operating data to a high-dimensional space-time feature space, generating a space-time feature distribution map based on a multi-dimensional dynamic clustering algorithm, the space-time feature distribution map containing a multi-dimensional correlation of navigation trajectory, hydrological state and meteorological parameters; a dynamic risk detection module for adjusting the space-time feature distribution map according to a preset risk identification rule, dividing real-time monitoring areas by a nonlinear optimization algorithm, and extracting a risk indicator set (including collision probability indicator, reciprocal of ship distance; meteorological risk indicator, quantified wind speed level; hydrological risk indicator, based on tidal height difference) in each area; a linkage decision control module for building a decision unit containing a dynamic path planning model, updating the dynamic path planning model with the risk indicator set, and generating a safe navigation control strategy; a feedback optimization module for optimizing the safe navigation control strategy according to a preset global sensitivity analysis framework, and outputting a ship navigation full-dimensional real-time regulation and control instruction.

[0009] Preferably, the multi-dimensional dynamic clustering algorithm-based spatiotemporal feature distribution map is generated by extracting the timestamps, ship IDs and operating parameters in the standardized operating data, constructing a high-dimensional spatiotemporal matrix, segmenting the high-dimensional spatiotemporal matrix using a sliding window segmentation algorithm to generate a time series sub-matrix set, performing spatial density analysis on the time series sub-matrix set using a density-based spatial clustering model (such as the DBSCAN algorithm, with a neighborhood radius ε of 0.5 and a minimum number of points MinPts of 5) to eliminate isolated noise points, and performing multi-level fusion on the remaining sub-matrices using a condensed hierarchical clustering algorithm (with Euclidean distance as the metric and Ward method for aggregation) to generate a feature distribution map with spatiotemporal correlation.

[0010] Preferably, the adaptive boundary adjustment includes calculating the mean and variance of key risk indicators in each region based on the historical operating data distribution of the real-time monitoring region, performing exponential smoothing correction (with a smoothing coefficient α of 0.3) on the mean and variance based on a dynamic sliding window algorithm (with a window size of 10 time units) to generate a dynamic boundary reference, and performing boundary fuzzification on the dynamic boundary reference using fuzzy logic rules (defining a triangular membership function and a rule base including 'IF risk is high THEN boundary is wide') to generate an adaptive boundary interval. Preferably, the nonlinear optimization algorithm for dividing the real-time monitoring region includes defining the objective function as the weighted maximization of navigation safety and environmental adaptability in the monitoring region, setting the constraint conditions as water topological connectivity, ship capacity limitation and meteorological parameter tolerance range, and solving the objective function by genetic algorithm to output the optimal monitoring region division scheme.

[0011] Preferably, the dynamic path planning model performs parameter updating by inputting the risk indicator set into the input layer of the dynamic path planning model, extracting the path feature vector using a graph neural network (such as the GCN model with an input layer dimension of 3), generating a priority sequence through attention mechanism weight allocation on the path feature vector, and updating the model weight using the backpropagation algorithm combined with the particle swarm optimization algorithm to adjust the control parameters of the decision unit.

[0012] Preferably, the multi-source data cleaning strategy includes identifying duplicate data segments in the operating data set and performing deduplication based on timestamp alignment rules, detecting missing data points and completing the missing data points using a spatiotemporal interpolation method (such as the inverse distance weighting interpolation method with a spatial weight coefficient of 2), and normalizing the completed data to generate standardized operating data with a mean of zero and a variance of one.

[0013] Preferably, the global sensitivity analysis framework includes: constructing a multivariate sensitivity calculation framework based on variance decomposition to quantify the influence of control commands on each operating parameter; generating a parameter perturbation sample set through random sampling to calculate the sensitivity contribution rate of each parameter; and selecting parameters with contribution rates higher than a preset threshold as core variables for optimized control.

[0014] Preferably, the multi-source data fusion optimization includes: integrating ship sensor data, meteorological data, and hydrological forecast data to construct a heterogeneous data fusion matrix; using principal component analysis to perform dimensionality reduction on the heterogeneous data fusion matrix and extracting the main feature vectors; and inputting the main feature vectors into the feedback optimization module to generate optimized control commands. Preferably, the system further includes: constructing an early warning module containing a ship health assessment function; performing correlation analysis between the real-time control commands and ship health; and triggering a preset emergency response mechanism based on the graded early warning signals to generate ship maintenance or route adjustment commands.

[0015] Preferably, the present invention further includes an electronic device, the device further including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the operation of the above-described ship navigation risk assessment system based on multi-source heterogeneous data fusion.

[0016] Compared with existing technologies, this invention has the following significant technical advantages: At the data processing and analysis level, the multi-source data integration module can acquire multiple sets of operational data within the target water area and process them through a multi-source data cleaning strategy to generate standardized operational data, greatly improving data quality. The spatiotemporal feature mapping module maps the standardized data to a high-dimensional spatiotemporal feature space, and uses a multi-dimensional dynamic clustering algorithm to generate a spatiotemporal feature distribution map containing multi-dimensional correlations between navigation trajectories, hydrological conditions, and meteorological parameters, deeply mining the potential connections between data. This not only allows for a more accurate grasp of ship navigation status but also enables early detection of abnormal signs. For example, by analyzing the correlation between navigation trajectories and meteorological parameters, it can predict the collision risk of ships in specific environments. Compared with traditional single-parameter monitoring, the accuracy of risk prediction is significantly improved, effectively reducing the probability of sudden collision accidents.

[0017] The dynamic risk detection module adaptively adjusts the boundaries of the spatiotemporal feature distribution map based on preset risk identification rules. It then uses a nonlinear optimization algorithm to divide the real-time monitoring area and extract a set of risk indicators. This adaptive boundary adjustment dynamically changes based on historical operational data of the real-time monitoring area, resulting in more accurate and sensitive risk assessment compared to traditional fixed-boundary monitoring, reducing false alarms and missed alarms. The nonlinear optimization algorithm for dividing the monitoring area comprehensively considers factors such as navigation safety and environmental adaptability, waterway topological connectivity, vessel capacity limitations, and meteorological parameter tolerance ranges, making the monitoring area division more scientific and reasonable. This improves monitoring efficiency, enables timely detection of potential safety hazards, and ensures safe navigation for vessels.

[0018] In terms of operational control, the coordinated decision-making control module constructs a decision unit containing a dynamic path planning model. It uses a set of risk indicators to update the model parameters and generate a safe navigation control strategy. Path feature vectors are extracted through a graph neural network, and a priority sequence is generated using an attention mechanism. Then, the model weights are updated using a backpropagation algorithm combined with a particle swarm optimization algorithm. This approach allows for dynamic adjustment of the control strategy based on real-time operational conditions, achieving precise control of ship navigation. During navigation, route planning can be quickly adjusted to ensure normal ship operation, reduce navigation risks, and improve navigation efficiency. Compared to traditional fixed control strategies, this effectively reduces energy waste and lowers operating costs.

[0019] The feedback optimization module, based on a pre-defined global sensitivity analysis framework, performs multi-source data fusion optimization on the safe navigation control strategy, outputting real-time control commands across all dimensions. This module integrates ship sensor data, meteorological data, and hydrological forecast data, using principal component analysis to extract key feature vectors through dimensionality reduction, making the control commands more scientific and forward-looking. Considering that meteorological data can proactively address the impact of severe weather on ship navigation, combining it with hydrological forecast data can optimize route planning, avoid navigation risk imbalances, and further improve the safety and reliability of ship navigation.

[0020] In addition, the system includes an early warning module that constructs a ship health assessment function. This function correlates real-time control commands with ship health status, outputs tiered early warning signals, and triggers an emergency response mechanism to generate ship maintenance or route adjustment commands. This allows for timely warnings of potential ship malfunctions, enabling advance maintenance planning, extending ship lifespan, and, in the event of problems, reasonable route adjustments to ensure uninterrupted navigation and improve the user's sailing experience. Attached Figure Description

[0021] Figure 1 The diagram shows the overall system structure of this invention, illustrating the logical connections between the multi-source data integration module, the spatiotemporal feature mapping module, the dynamic risk detection module, the linkage decision control module, and the feedback optimization module.

[0022] Figure 2 This is a schematic diagram of the spatiotemporal feature distribution map generation process, which describes in detail the processing from standardized operating data to high-dimensional spatiotemporal matrix construction, sliding window segmentation, spatial density analysis and hierarchical aggregation.

[0023] Figure 3 The flowchart for updating parameters in a dynamic path planning model illustrates the steps of inputting risk indicators, extracting features from a graph neural network, assigning weights using an attention mechanism, and updating weights by combining backpropagation and particle swarm optimization algorithms.

[0024] Figure 4 This is a flowchart of the global sensitivity analysis framework, including the process of constructing a heterogeneous data fusion matrix, principal component analysis for dimensionality reduction, and the optimization and control of core variables for calculating sensitivity contribution rates.

[0025] The attached diagram is labeled as follows: 1. Multi-source data integration module; 2. Spatiotemporal feature mapping module; 3. Dynamic risk detection module; 4. Linked decision control module; 5. Feedback optimization module; 6. High-dimensional spatiotemporal matrix; 7. Sliding window segmentation algorithm; 8. Dynamic path planning model; 9. Graph neural network; 10. Principal component analysis algorithm. Detailed Implementation

[0026] This invention provides a ship navigation risk assessment system based on multi-source heterogeneous data fusion, the overall structure of which is as follows: Figure 1 As shown, the system includes a multi-source data integration module 1, a spatiotemporal feature mapping module 2, a dynamic risk detection module 3, a coordinated decision-making and control module 4, and a feedback optimization module 5. These modules work together through logical connections to achieve a comprehensive assessment and real-time control of navigation risks within the target waters.

[0027] In the specific implementation process, the multi-source data integration module 1 first acquires the operational data set within the target water area. This data originates from various sources, including but not limited to ship sensor data, meteorological monitoring data, hydrological forecast data, and other relevant operational parameters. To ensure data quality, module 1 employs a pre-defined multi-source data cleaning strategy to process the raw data. The specific steps of this strategy include identifying duplicate data segments in the dataset and removing duplicates based on timestamp alignment rules; detecting missing data points and completing them using spatiotemporal interpolation; and finally, normalizing the completed data to generate standardized operational data with a mean of zero and a variance of one. This processing effectively eliminates noise and outliers, enabling subsequent analysis to be based on high-quality data.

[0028] The standardized operational data is then input into the spatiotemporal feature mapping module 2 to generate a high-dimensional spatiotemporal feature distribution map. For example... Figure 2As shown, this module first extracts timestamps, vessel IDs, and operational parameters from standardized operational data to construct a high-dimensional spatiotemporal matrix 6. Next, a sliding window segmentation algorithm 7 is used to segment the high-dimensional spatiotemporal matrix, generating a set of time-series sub-matrices. Based on this, a density-based spatial clustering model is used to perform spatial density analysis on the time-series sub-matrices, eliminating isolated noise points and retaining data points with significant spatiotemporal correlations. Finally, a hierarchical aggregation algorithm is used to perform multi-level fusion of the remaining sub-matrices, forming a spatiotemporal feature distribution map containing multi-dimensional correlations between navigation trajectories, hydrological conditions, and meteorological parameters. This process not only achieves feature mapping from low to high dimensions but also fully explores the potential connections between different data dimensions, providing a solid foundation for subsequent risk assessment.

[0029] The dynamic risk detection module 3 receives the spatiotemporal feature distribution map as input and adaptively adjusts its boundaries according to preset risk identification rules. Specifically, module 3 first calculates the mean and variance of key risk indicators in each real-time monitoring area, and then performs exponential smoothing correction on the mean and variance based on a dynamic sliding window algorithm to generate a dynamic boundary benchmark. Subsequently, it uses fuzzy logic rules to perform boundary fuzzification processing on the dynamic boundary benchmark to generate an adaptive boundary interval. After completing the boundary adjustment, module 3 uses a nonlinear optimization algorithm to divide the real-time monitoring area. This algorithm defines the objective function as the weighted maximization of navigation safety and environmental adaptability within the monitoring area, while setting constraints, including waterway topological connectivity, ship capacity limits, and meteorological parameter tolerance ranges. The objective function is solved through a genetic algorithm, and the optimal monitoring area division scheme is finally output, and a set of risk indicators is extracted from each area (including collision probability indicators, calculated as the reciprocal of ship spacing; meteorological risk indicators, quantified wind speed levels; hydrological risk indicators, based on tidal height differences). This process ensures the accuracy and sensitivity of risk detection and avoids the false alarms and false negatives that may occur in traditional fixed boundary methods.

[0030] The linkage decision control module 4 is responsible for generating a safe navigation control strategy based on the risk indicator set. Internally, module 4 constructs a dynamic path planning model 8. This model extracts path feature vectors through a graph neural network 9 and uses an attention mechanism to weight these vectors, generating a priority sequence. Then, it updates the model weights using a backpropagation algorithm combined with a particle swarm optimization algorithm, thereby adjusting the control parameters of the decision unit. In practical applications, module 4 dynamically adjusts the control strategy based on real-time operating conditions. For example, when the risk indicator set within a monitoring area indicates a high collision risk, module 4 quickly adjusts the route planning to avoid the high-risk area, ensuring the normal operation of the vessel. Furthermore, the model can comprehensively consider navigation efficiency and energy consumption, improving navigation efficiency and reducing operating costs while reducing navigation risks.

[0031] Feedback optimization module 5 further optimizes the safe navigation control strategy, outputting real-time control commands across all dimensions. For example... Figure 4 As shown, module 5 first constructs a global sensitivity analysis framework, which quantifies the impact of control commands on various operating parameters based on variance decomposition. A parameter disturbance sample set is generated through random sampling, the sensitivity contribution rate of each parameter is calculated, and parameters with contribution rates higher than a preset threshold are selected as core variables for optimized control. Subsequently, module 5 integrates ship sensor data, meteorological data, and hydrological forecast data to construct a heterogeneous data fusion matrix, and uses principal component analysis algorithm 10 to reduce the dimensionality of the matrix and extract the main feature vectors. These feature vectors are input to feedback optimization module 5 to generate optimized control commands. For example, under severe weather conditions, module 5 can adjust route planning in advance to avoid potential navigation risks; combined with hydrological forecast data, it can also optimize the ship's operating path in complex waters, further improving navigation safety and reliability.

[0032] In addition, the system includes an early warning module. This module constructs a ship health assessment function, correlates real-time control commands with ship health status, and outputs tiered early warning signals. When a tiered early warning signal triggers a preset emergency response mechanism, the system generates ship maintenance or route adjustment commands. For example, when the ship health assessment indicates a potential equipment malfunction, the early warning module will issue a maintenance prompt and recommend that the ship berth for repairs as soon as possible; if the current water environment is unsuitable for continued navigation, a route adjustment command will be triggered to guide the ship to a safe area. This design not only extends the ship's service life but also ensures uninterrupted navigation in emergencies, improving the user's navigation experience.

[0033] In terms of hardware implementation, the present invention also includes an electronic device comprising at least a processor and a memory communicatively connected thereto. The memory stores instructions executable by the processor, which, when executed, enables the operation of the aforementioned ship navigation risk assessment system based on multi-source heterogeneous data fusion. This hardware architecture ensures efficient system operation and real-time response capabilities, making it suitable for ship navigation management scenarios of various scales.

[0034] As can be seen from the above specific implementation methods, the various modules of this invention work closely together, forming a complete closed-loop system from data acquisition, cleaning, and feature extraction to risk assessment, path planning, and feedback optimization. The functionality of each module depends on the output results of the preceding module, while also providing necessary input data for subsequent modules, thereby ensuring the efficient operation of the entire system.

[0035] To enable those skilled in the art to fully understand and implement this invention, the specific implementation principle of this invention is further explained below in conjunction with a specific application scenario.

[0036] When the vessel navigation risk assessment system operating in the target waters is activated, the multi-source data integration module 1 first performs data acquisition and cleaning tasks. By accessing the operational data sets from ship sensors, meteorological monitoring stations, and the hydrological forecasting system, module 1 begins processing the raw data. For example, in a complex waters scenario, module 1 identifies duplicate records in the AIS data and performs deduplication based on timestamp alignment rules. Simultaneously, to address the issue of missing meteorological data due to signal loss, module 1 uses spatiotemporal interpolation to complete the data.

[0037] The completed data is then normalized to generate standardized operating data with a mean of zero and a variance of one, ensuring that subsequent analyses can be based on high-quality data. This process is as follows: Figure 1 As shown, the logical connections between Module 1 and other modules are illustrated. Subsequently, standardized operational data is transmitted to the spatiotemporal feature mapping module 2 to generate a high-dimensional spatiotemporal feature distribution map. Taking a navigation trajectory within a specific time period as an example, Module 2 extracts the timestamp, ship ID, and operational parameters for that time period to construct a high-dimensional spatiotemporal matrix 6. Next, the sliding window segmentation algorithm 7 segments the high-dimensional spatiotemporal matrix, generating multiple sets of time-series sub-matrices. Based on this, a density-based spatial clustering model is used to perform spatial density analysis on these sub-matrices, removing isolated noise points and retaining data points with significant spatiotemporal correlations. Finally, a hierarchical aggregation algorithm is used to perform multi-level fusion of the remaining sub-matrices, forming a spatiotemporal feature distribution map containing multi-dimensional correlations between navigation trajectories, hydrological conditions, and meteorological parameters. This process is as follows: Figure 2 As shown, the processing flow from standardized operational data to high-dimensional spatiotemporal matrix construction, sliding window segmentation, spatial density analysis, and hierarchical aggregation is described in detail.

[0038] After receiving the spatiotemporal feature distribution map, the dynamic risk detection module 3 adaptively adjusts its boundaries according to preset risk identification rules. Taking a real-time monitoring area as an example, module 3 calculates the mean and variance of key risk indicators within that area and performs exponential smoothing correction on the mean and variance based on a dynamic sliding window algorithm to generate a dynamic boundary benchmark. Subsequently, fuzzy logic rules are used to fuzzify the dynamic boundary benchmark, generating an adaptive boundary interval. After completing the boundary adjustment, module 3 uses a nonlinear optimization algorithm to divide the real-time monitoring area. This algorithm defines the objective function as the weighted maximization of navigation safety and environmental adaptability within the monitoring area, while setting constraints, including waterway topological connectivity, ship capacity limitations, and the tolerance range of meteorological parameters. The objective function is solved using a genetic algorithm, ultimately outputting the optimal monitoring area division scheme and extracting the risk indicator set for each area (including collision probability indicators, calculated as the reciprocal of ship spacing; meteorological risk indicators, quantified wind speed levels; and hydrological risk indicators, based on tidal height differences). This process enables the system to more accurately capture potential risks in complex aquatic environments.

[0039] The linkage decision control module 4 generates a safe navigation control strategy based on the extracted set of risk indicators. For example, when the set of risk indicators in a certain monitoring area shows a high collision risk, the dynamic path planning model 8 within module 4 begins operation. Model 8 extracts path feature vectors through a graph neural network 9 and assigns weights to these vectors using an attention mechanism to generate a priority sequence. Then, it updates the model weights using a backpropagation algorithm combined with a particle swarm optimization algorithm, thereby adjusting the control parameters of the decision unit. In practical applications, module 4 quickly adjusts the route planning to avoid high-risk areas and ensure the normal operation of the vessel. Furthermore, the model comprehensively considers navigation efficiency and energy consumption, improving navigation efficiency and reducing operating costs while reducing navigation risks. This process is as follows: Figure 3 As shown, this illustrates the steps of inputting risk indicators, extracting features from graph neural networks, assigning weights using the attention mechanism, and updating weights by combining the backpropagation algorithm with the particle swarm optimization algorithm.

[0040] The feedback optimization module 5 further optimizes the safe navigation control strategy, outputting real-time control commands across all dimensions. For example, under adverse weather conditions, module 5 first constructs a global sensitivity analysis framework, which quantifies the impact of control commands on various operating parameters based on variance decomposition. It generates a parameter disturbance sample set through random sampling, calculates the sensitivity contribution rate of each parameter, and selects parameters with contribution rates higher than a preset threshold as core variables for optimized control. Subsequently, module 5 integrates ship sensor data, meteorological data, and hydrological forecast data to construct a heterogeneous data fusion matrix, and uses principal component analysis algorithm 10 to reduce the dimensionality of the matrix and extract key feature vectors. These feature vectors are input to the feedback optimization module 5 to generate optimized control commands. For example, combined with hydrological forecast data, module 5 can optimize the ship's operating path in complex waters, further improving navigation safety and reliability. This process is as follows: Figure 4 As shown, the process of selecting core variables for optimization and control includes constructing a heterogeneous data fusion matrix, performing principal component analysis for dimensionality reduction, and calculating the sensitivity contribution rate.

[0041] In addition, the system includes an early warning module. This module constructs a ship health assessment function, correlates real-time control commands with the ship's health status, and outputs tiered early warning signals. For example, when the ship health assessment indicates a potential equipment malfunction, the early warning module will issue a maintenance prompt and recommend that the ship berth for repairs as soon as possible. If the current water environment is unsuitable for continued navigation, it will trigger a route adjustment command to guide the ship to a safe area. This design not only extends the ship's service life but also ensures uninterrupted navigation in emergencies, improving the user's navigation experience.

[0042] In terms of hardware implementation, the present invention also includes an electronic device comprising at least a processor and a memory communicatively connected thereto. The memory stores instructions executable by the processor, which, when executed, enables the operation of the aforementioned ship navigation risk assessment system based on multi-source heterogeneous data fusion. This hardware architecture ensures efficient system operation and real-time response capabilities, making it suitable for ship navigation management scenarios of various scales.

[0043] As can be seen from the specific application scenarios described above, the various modules of this invention work closely together, forming a complete closed-loop system from data acquisition, cleaning, and feature extraction to risk assessment, path planning, and feedback optimization. The functionality of each module depends on the output results of the preceding module, while also providing necessary input data for subsequent modules, thereby ensuring the efficient operation of the entire system.

Claims

1. A ship navigation risk assessment system (1) based on multi-source heterogeneous data fusion, characterized in that, The system comprises: a multi-source data integration module (1) for acquiring a set of operation data in a target water area range (including AIS data obtained through an API interface, meteorological data from NOAA format, and hydrological data in JSON structure), and processing the set of operation data according to a preset multi-source data cleaning strategy to generate standardized operation data; a space-time feature mapping module (2) for mapping the standardized operation data into a high-dimensional space-time feature space, generating a space-time feature distribution atlas based on a multi-dimensional dynamic clustering algorithm, the space-time feature distribution atlas containing a multi-dimensional correlation of a navigation track, a hydrological state and meteorological parameters; a dynamic risk detection module (3) for adaptively adjusting a boundary of the space-time feature distribution atlas according to a preset risk identification rule, dividing a real-time monitoring area through a nonlinear optimization algorithm, and extracting a risk index set (including a collision probability index, a reciprocal of a ship distance, a meteorological risk index, a wind speed level, and a hydrological risk index based on a tidal height difference) in each area; a linkage decision control module (4) for constructing a decision unit containing a dynamic path planning model (8), updating parameters of the dynamic path planning model (8) using the risk index set, and generating a safe navigation control strategy; and a feedback optimization module (5) for performing multi-source data fusion optimization on the safe navigation control strategy according to a preset global sensitivity analysis framework, and outputting a ship navigation full-dimensional real-time regulation and control instruction.

2. The ship navigation risk assessment system (1) based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The space-time feature distribution atlas generated based on the multi-dimensional dynamic clustering algorithm comprises: extracting a timestamp, a ship ID and an operation parameter in the standardized operation data, and constructing a high-dimensional space-time matrix (6); segmenting the high-dimensional space-time matrix (6) using a sliding window segmentation algorithm (7) to generate a time series sub-matrix set; performing spatial density analysis on the time series sub-matrix set using a density-based spatial clustering model to eliminate isolated noise points; and performing multi-level fusion on the remaining sub-matrices through a hierarchical aggregation algorithm to generate a feature distribution atlas with space-time correlation.

3. The ship navigation risk assessment system (1) based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The adaptive boundary adjustment comprises: calculating a mean value and a variance of a key risk index in each area according to historical operation data distribution of the real-time monitoring area; performing exponential smoothing correction on the mean value and the variance based on a dynamic sliding window algorithm to generate a dynamic boundary benchmark; and performing boundary fuzzification on the dynamic boundary benchmark using a fuzzy logic rule to generate an adaptive boundary interval.

4. The ship navigation risk assessment system (1) based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The nonlinear optimization algorithm for dividing the real-time monitoring area comprises: defining a target function as a weighted maximization of navigation safety and environmental adaptability in the monitoring area; setting a constraint condition as water area topological connectivity, ship capacity limitation and meteorological parameter tolerance range; and solving the target function through a genetic algorithm to output an optimal monitoring area division scheme.

5. The ship navigation risk assessment system (1) based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The dynamic path planning model (8) performs parameter updating, including: inputting the risk index set into the input layer of the dynamic path planning model (8), extracting a path feature vector using a graph neural network (9); performing weight distribution on the path feature vector through an attention mechanism to generate a priority sequence; updating model weights using a back propagation algorithm combined with a particle swarm optimization algorithm to adjust the control parameters of the decision unit.

6. The ship navigation risk assessment system (1) based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The multi-source data cleaning strategy includes: identifying duplicate data segments in the running data set and performing deduplication based on timestamp alignment rules; detecting missing data points and using a spatiotemporal interpolation method (such as inverse distance weighting interpolation method with a spatial weight coefficient of 2) to complete the missing data points; normalizing the completed data to generate standardized running data with a mean of zero and a variance of one.

7. The ship navigation risk assessment system (1) based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The global sensitivity analysis framework includes: constructing a multivariate sensitivity calculation framework based on variance decomposition method to quantify the influence degree of the control instruction on each operating parameter; generating a parameter perturbation sample set through random sampling and calculating the sensitivity contribution rate of each parameter; selecting parameters with a contribution rate higher than a preset threshold as optimization control core variables.

8. The ship navigation risk assessment system (1) based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The multi-source data fusion optimization includes: integrating ship sensor data, weather data, and hydrological prediction data to construct a heterogeneous data fusion matrix; using principal component analysis algorithm (10) to perform dimensionality reduction processing on the heterogeneous data fusion matrix to extract main feature vectors; inputting the main feature vectors into the feedback optimization module (5) to generate optimized control instructions.

9. The ship navigation risk assessment system (1) based on multi-source heterogeneous data fusion according to any one of claims 1 to 8, characterized in that, The system further includes: constructing an early warning module containing a ship health degree evaluation function, performing correlation analysis on the real-time control instruction and the ship health degree, and outputting a graded early warning signal; triggering a preset emergency response mechanism according to the graded early warning signal to generate ship maintenance or route adjustment instructions.

10. An electronic device, comprising: It includes: At least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the operation of the ship navigation risk assessment system based on multi-source heterogeneous data fusion as claimed in any one of claims 1 to 9.

Images