A method for controlling the risk of navigation of a ship

Abstract

The present application belongs to the technical field of ship navigation risk management, and particularly relates to a ship navigation risk control method. The method comprises the following steps: obtaining a ship standard data file through a ship navigation service system; determining corresponding elements under an accident category; sorting comprehensive weights and influence weights to form a risk accident element graph model; extracting historical navigation accident data according to the element types in the risk accident element graph model; establishing a graph form sample database; classifying and training the navigation graph form samples by using a multi-layer network classification model; obtaining a typical feature category; and matching a risk accident feature of a ship to be analyzed for risk control. The present application is used for analyzing the feature of each element in the historical data of ship navigation risk accidents, completing the classification and identification of the risk accident features of different types of ships in the navigation process, and classifying and predicting the risk accident of a ship to be analyzed by using the same, so as to determine the navigation accident feature of the ship, and to take targeted accident prevention measures.

Description

Technical Field

[0001] This invention belongs to the field of ship navigation risk management technology, and in particular relates to a method for controlling ship navigation risks. Background Technology

[0002] Ship navigation accidents often result in significant personal and property losses. The analysis and control of ship navigation accidents often involve numerous influencing factors, including but not limited to the ship's own condition, weather conditions, and waterway conditions. Each navigation accident is also often involved in the combined effects of multiple factors. Currently, the risk control of ship navigation accidents mainly relies on ship operators to analyze and judge based on past experience combined with the current ship condition and route condition, supplemented by various standardized risk control procedures. Summary of the Invention

[0003] The purpose of this invention is to provide a method for intelligent and automated ship navigation risk analysis, so as to identify the key points of ship risk control and select appropriate control schemes.

[0004] To achieve the above objectives, the present invention adopts the following technical solution.

[0005] A method for controlling ship navigation risks includes the following steps:

[0006] Step 1: Obtain navigation accident data information during the ship's navigation process through the ship navigation service system, formulate standard data files according to the needs of navigation risk influencing factors, and summarize the attribute data of various ship navigation service systems through the standard data files;

[0007] Step 2: Determine the relevant elements under the categories of accident caused by improper operation, ship-related problems, and route-related problems; specifically:

[0008] Historical data on navigation accidents caused by improper ship operation were extracted and organized to obtain the element sequence {X1,X2,X3,...X} corresponding to the improper operation accidents. i ,...X I}, where X i It refers to the i-th element, and I refers to the total number of samples of the elements;

[0009] For a certain type of operational error accident Y j Analyze each accident caused by improper operation Y j The cause of the accident was improper operation during extraction. j The occurrence of the element sequence {X jk1 ,X jk2 ,...X jki ,...X jkI};

[0010] Where X jki This refers to the accident Y that caused the kth operation error. j When it occurs, element X i The indicated value, X jki =1 indicates that the accident Y is caused by improper operation in the kth operation. j When it occurs, element X exists. i The impact of X jki =0 indicates that the accident Y is due to improper operation during the kth operation. j Element X did not exist when it occurred. i The impact;

[0011] Get all feature sequences, delete X jki The element = 0 is then removed, and duplicate element sequences are eliminated to obtain Y, which caused the operational mishap. j Element combination sequence T j ={T j1 ,T j2 ...T jm ...T jM}; where T jm Y indicates an accident caused by improper operation. j The m-th element set,

[0012] Extract a risk incident sample set B from historical data that is affected by any combination of factors. j Based on expert scoring systems or statistical analysis, statistical T jm Combination of medium-term factors on risk incident sample set B j The influence of the medium sample is the lowest among all factor combinations, represented by the sample size b. jm.min Calculate T jm Influence coefficient under corresponding element combination Where b j Refers to sample set B j Total number of samples in the middle sample; calculation of accident Y due to improper operation. j The comprehensive weight of the combination of factors involved

[0013] Based on the above steps, calculate the ship problem accident O. p The combination of elements of ship problems O pq Corresponding factor combination influence coefficient and element combination sequence O p ={O p1 O p2 ...O pq ...O pQ The overall weight of} Calculate the route problem accident U v The combination of elements of the route problem U vsCorresponding factor combination influence coefficient and element combination sequence U v ={U v1 U v2 ...U vs ...U vS The overall weight of}

[0014] Among them O pq This refers to the q-th combination of ship problem elements corresponding to the p-th type of ship problem accident, b p This refers to the accident sample set B in historical data that is affected by any combination of ship-related problem elements. p Total number of samples, b pq.min It is sample set B p Combination of elements of Chinese shipbuilding issues O pq The impact level is the lowest among all combinations of ship-related factors; where U vs This refers to the combination of route problem elements in the s-th category corresponding to the v-th type of route problem accident, b v This refers to the accident sample set B in historical data that is affected by any combination of factors related to flight routes. v Total number of samples, b vs.min It is sample set B v Combination of elements of the problem in the middle route U vs The degree of impact is the lowest among all combinations of route problem elements in terms of sample size;

[0015] Step 3: Sort the comprehensive weight and impact weight respectively. For each type of ship, sort the improper operation accidents, ship problem accidents and route problem accidents in order of weight from large to small. At the same time, screen and sort the corresponding combination elements within the improper operation accidents, ship problem accidents and route problem accidents. Select several combination elements with large impact weights from each type of accident in order of large to small to form the risk accident element model of each type of ship.

[0016] Step 4: Based on the element types in the risk accident element model of various types of ships, extract the historical navigation accident data of the corresponding ships; for each ship navigation accident sample, extract the element index value y. i Then normalize to obtain the normalized value y. i ;

[0017]

[0018] Where y i ′ .max It is the factor indicator y i The maximum value of y′ i ′ .min It is the factor indicator y iThe minimum value of ′;

[0019] Based on the order of the risk and accident element diagrams for the corresponding ship type, the element sequence Y for each ship is obtained after sorting. w ={y w1 ,y w2 ,y w3 ,...y wl ,...y wL}; where y wl It refers to the index value of the l-th element in the risk accident element model of the w-th ship, and L refers to the total number of elements in the risk accident element model;

[0020] The element sequence Y w Unified conversion to graphical samples Organize all chart samples for each type of ship to create a database of nautical chart samples for that type of ship. Where H×W represents the dimensions of the graphic;

[0021] Step 5: Use a multi-layer network classification model to classify and train the navigation chart-style samples; the multi-layer network classification model is based on the ResNet34 multi-branch network and includes a feature extraction module and a feature fusion module;

[0022] The input to the feature extraction module is a graph-based sample. The ResNet34 network generates output images with different dilation rates through multiple dilated convolutional blocks. The output images are then processed by the Inception network to extract their features.

[0023] The feature fusion module uses a multi-level feature weighting fusion network to perform feature weighting fusion, which includes dilating convolution, normalizing, linear rectified activation and adaptive average pooling of feature maps of output images with different dilation rates to output fused features;

[0024] Graph-based sample database The dataset is divided into a training dataset and a validation dataset. The multilayer network classification model is trained using the training dataset, and the parameters of the multilayer network classification model are optimized and adjusted using the validation dataset to obtain the final multilayer network classification model.

[0025] Step 6: Input the samples from the graph-based sample database into the multi-layer network classification model after training to obtain several types of classification results. This yields a typical feature category representing a certain type of ship's navigation accident. The navigation accident characteristics of ships with similar basic features to the classification results are consistent with this typical feature category.

[0026] By collecting information on past navigation accidents of the vessel to be analyzed and inputting it into a classification network for classification, the typical characteristics of its navigation accidents can be obtained. By matching the risk accident characteristics of similar vessels with the same typical characteristics on the expected route of the vessel to be analyzed, the navigation risk accidents of the vessel to be analyzed can be predicted, and risk management can be carried out based on the prediction results.

[0027] Further improvements or preferred implementation steps for the aforementioned ship navigation risk control methods, wherein the ship navigation service system includes various ship navigation service systems such as VTS system, GPS-CDMA, and AIS.

[0028] For further improvements or preferred implementation steps of the aforementioned ship navigation risk control method, the standard data file includes data blocks with the following data indicators:

[0029] Code index data block: Used to store the code sequence number data of each ship, serving as a unique identification code for each ship;

[0030] Type Index Data Block: Used to store the type information of each vessel. The vessel types include at least small vessels, medium vessels, large vessels, unmanned vessels, and special vessels. Among them, small vessels refer to vessels with an overall length of less than or equal to 20m, medium vessels refer to vessels with an overall length of more than 20m but less than or equal to 100m, and large vessels refer to vessels with an overall length of more than 100m. Unmanned vessels refer to vessels that mainly rely on autonomous driving technology to complete navigation, and special vessels refer to vessels with specific purposes such as engineering vessels, scientific research vessels, and rescue vessels.

[0031] Speed ​​data block: Used to store the speed information of each ship during navigation;

[0032] Heading indicator data block: Used to store the route data information of each ship;

[0033] Lifecycle Indicator Data Block: Used to store the average lifecycle of each vessel and its lifecycle stage information;

[0034] Water depth index data block: used to store water depth data information of the ship at each navigation node or segment;

[0035] Resistance Index Data Block: Used to store the ship's navigation resistance index, which refers to the ratio of the resistance generated by the ship's current course due to surface wind force and underwater fluid force during navigation at various navigation segments or nodes to the ship's power.

[0036] Visibility index data block: used to store the visibility range data of ships at various navigation nodes or sections;

[0037] Unmanned Vessel Index Data Block: Used to store the number and percentage of unmanned vessels on each current shipping route;

[0038] Route Obstacle Index Data Block: Used to store obstacle distribution data on each flight segment;

[0039] Risk Vessel Index Data Block: Used to store the proportion of high-risk vessels on each route. High-risk vessels refer to the proportion of vessels that would cause serious consequences in the event of a dangerous situation, or vessels that have a high risk of being overtaken in the current segment of the voyage.

[0040] For further improvements or preferred implementation steps of the aforementioned ship navigation risk control method, the standard data file is saved using a unified CSV text file.

[0041] Further improvements or preferred implementation steps to the aforementioned ship navigation risk control method include, in step 1, cleaning and completion processing of the original data, specifically:

[0042] A1. Since ship navigation data is continuous in time, for single missing data, it is filled by difference and average processing of the time before and after the sampling point. For continuous missing data, it is obtained by fitting data within the sampling period or by averaging historical sampling data in the same period.

[0043] A2. For the original data after filling in the information, its quality needs to be analyzed and judged to ensure that the provided data has sufficient accuracy. Specifically, this includes:

[0044] Verify the consistency of all data, ensuring that the data has the required type, magnitude, format, and units. Inconsistent data should be checked and analyzed, and errors should be marked. This includes verifying whether numerical indicators such as speed have a relatively consistent magnitude and whether code data such as dates, models, and serial numbers have a consistent text format.

[0045] Perform fitting analysis on multiple samples of a single indicator to determine whether the indicator data exceeds the normal and reasonable range. Data that exceeds the normal and reasonable range is incorrectly labeled, including: numerical data exceeding the normal upper limit; code data exceeding the normal code range, etc.

[0046] A3. For the preprocessed raw data, we need to collect complete data for each voyage of a single vessel. Since the vessel has records in different navigation service systems, a large amount of duplicate data will be generated during the initial data collection process. These duplicate data exhibit random fluctuations within a certain range at their respective nodes. This fluctuation is due to differences in monitoring methods, monitoring time nodes, and data storage accuracy among various navigation service systems. Only one set of duplicate data is needed as the raw data for analysis. In order to retain the basic data characteristics of these data, clustering analysis algorithms are used to compress the data. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of a multi-layer network classification model. Detailed Implementation

[0048] The present invention will be described in detail below with reference to specific embodiments.

[0049] The ship navigation risk control method of this application is mainly used to provide a method that can classify and identify the risk accident characteristics of different types of ships during navigation by analyzing the characteristics of various elements in the historical data of ship navigation risk accidents, and thereby classify and predict the navigation risk accidents of the ships to be analyzed, determine their navigation accident characteristics, so as to take targeted accident prevention measures.

[0050] It mainly includes the following steps:

[0051] Step 1: Obtaining Basic Navigation Elements

[0052] Current ship navigation service systems have established a comprehensive ship navigation information and data management process by combining audio-visual communication measures and radar communication. Through various ship navigation service systems such as VTS, GPS-CDMA, and AIS, navigation accident data information during ship navigation can be obtained. As the navigation risk control requirements to be considered in this application, it is necessary to extract the corresponding element information from the data information provided by the above-mentioned ship navigation service systems for subsequent use and processing.

[0053] Since various ship navigation service systems usually have their own independent file data formats, in order to facilitate unified processing, standard data files should first be drafted according to the needs of navigation risk influencing factors. The attribute data in various ship navigation service systems should be summarized through the standard data files as the original data required in this application for extracting various basic navigation elements.

[0054] Specifically, to meet the most basic requirements, the standard data file in this application should include at least a data block with the following data indicators:

[0055] Code index data block: Used to store the code sequence number data of each ship, serving as a unique identification code for each ship;

[0056] Type Index Data Block: Used to store the type information of each vessel. The vessel types include at least small vessels, medium vessels, large vessels, unmanned vessels, and special vessels. Among them, small vessels refer to vessels with an overall length of less than or equal to 20m, medium vessels refer to vessels with an overall length of more than 20m but less than or equal to 100m, and large vessels refer to vessels with an overall length of more than 100m. Unmanned vessels refer to vessels that mainly rely on autonomous driving technology to complete navigation, and special vessels refer to vessels with specific purposes such as engineering vessels, scientific research vessels, and rescue vessels.

[0057] Speed ​​data block: Used to store the speed information of each ship during navigation;

[0058] Heading indicator data block: Used to store the route data information of each ship;

[0059] Lifecycle Indicator Data Block: Used to store the average lifecycle of each vessel and its lifecycle stage information;

[0060] Water depth index data block: used to store water depth data information of the ship at each navigation node or segment;

[0061] Resistance Index Data Block: Used to store the ship's navigation resistance index, which refers to the ratio of the resistance generated by the ship's current course due to surface wind force and underwater fluid force during navigation at various navigation segments or nodes to the ship's power.

[0062] Visibility index data block: used to store the visibility range data of ships at various navigation nodes or sections;

[0063] Unmanned Vessel Index Data Block: Used to store the current number and percentage of unmanned vessels on each route;

[0064] Route Obstacle Index Data Block: Used to store obstacle distribution data on each flight segment;

[0065] Risk Vessel Index Data Block: Used to store the proportion of high-risk vessels on each route. High-risk vessels refer to the proportion of vessels that will cause serious consequences in the event of a dangerous situation, or vessels that have a high risk of being overtaken in the current segment of the voyage.

[0066] The basic data of various ship navigation service systems are usually stored in various text data tables. In the process of processing and integrating the above raw data, we consider converting and saving it as a unified CSV text file so that it can be read and processed by various data analysis and processing software, simplifying the processing difficulty of raw data and improving the efficiency of data screening and analysis.

[0067] The basic data of various ship navigation service systems includes a large amount of common data that has been stored and processed. However, due to factors such as data accuracy, recording time, and deviations in detection location, there may be differences in the data for the same indicators. Furthermore, for various reasons, there may be instances where specific node or time-related data is missing. Therefore, this application also includes cleaning and completion processing of the original data. Specifically:

[0068] A1. Since ship navigation data is continuous in time, for single missing data, it can be filled by difference and average processing of the time before and after the sampling point. For continuous missing data, it can be obtained by fitting data within the sampling period or by averaging historical sampling data in the same period.

[0069] A2. For the original data after filling in the information, its quality needs to be analyzed and judged to ensure that the provided data has sufficient accuracy. Specifically, this includes:

[0070] Verify the consistency of all data, ensuring that the data has the required type, magnitude, format, and units. Inconsistent data should be checked and analyzed, and errors should be marked. This includes verifying whether numerical indicators such as speed have a relatively consistent magnitude and whether code data such as dates, models, and serial numbers have a consistent text format.

[0071] Perform fitting analysis on multiple samples of a single indicator to determine whether the indicator data exceeds the normal and reasonable range. Data that exceeds the normal and reasonable range is incorrectly labeled, including: numerical data exceeding the normal upper limit; code data exceeding the normal code range, etc.

[0072] A3. After preprocessing, the raw data needs to collect complete data for each voyage of a single vessel. Since the vessel may have records in different navigation service systems, a large amount of duplicate data will be generated during the initial data collection process. These duplicate data exhibit random fluctuations within a certain range at their respective nodes. This fluctuation is due to differences in monitoring methods, monitoring time nodes, and data storage accuracy among various navigation service systems. Only one set of duplicate data is needed as the raw data for analysis. To preserve the basic data characteristics of these data, clustering analysis algorithms are used to compress the data.

[0073] Although ship navigation accidents are considered low-probability events in navigation databases, they are diverse, including collisions between ships, collisions with obstacles, grounding and capsizing, fires and explosions, windstorms, sinking, flooding, and other accidents. Analyzing the causes of these accidents can be categorized as either due to improper ship operation or problems with the ship or its route. Based on the fault tree analysis principle, a fault tree with a ship navigation accident as the final event can be divided into accidents caused by improper operation, accidents caused by ship problems, and accidents caused by route problems.

[0074] Various basic elements of navigation can cause various risks in ship navigation and serve as fundamental factors leading to accidents. For example, accidents caused by improper ship operation are mainly affected by improper or erroneous operational behaviors during the ship operation process. By analyzing the characteristics of various failures and the correlation between ship operation behaviors, we can identify the corresponding elements of ship operation accidents. Common elements mainly include: ship speed, ship acceleration, ship course, angular velocity, and draft, etc. The meanings of each element are shown in Table 1.

[0075] Table 1. Elements and the Meaning of Improper Operation

[0076]

[0077] Taking 62 ship accidents caused by improper operation that occurred on a certain type of ship in the past three years in a certain waterway as an example, the operation data were extracted based on the aforementioned factors to obtain Table 2. At the same time, 62 samples were randomly selected from the same type of ship that were passing normally, and their operation data were extracted to obtain Table 3.

[0078] Table 2. Element data corresponding to operational errors.

[0079]

[0080] Table 3 Normal Element Data

[0081]

[0082] Comparative analysis reveals that, compared to data from normal operations, the range of variation for speed, acceleration, and heading deviation is significantly wider in cases of operational misconduct. Furthermore, the variance and standard deviation data show that the fluctuations in these elements also exceed those of normal operations, indicating that during operational misconduct incidents, the speed and the amount of maneuvering involved are greater. Compared to normal operations, the range of variation and volatility for elements such as heading deviation and variable angular velocity are also relatively higher in cases of operational misconduct, suggesting that operators tend to be more aggressive in controlling these elements, resulting in greater maneuvering and more frequent operations. These operational characteristics are directly related to the operator's habits and operational practices.

[0083] Further statistical analysis revealed that operators with similar operating habits are more likely to experience the same or similar navigation accidents in the same position and under the same conditions when operating ships. By extracting various elements from the raw data, the navigational risks of operators of the same type when operating corresponding ships to perform similar tasks can be assessed, thereby enabling risk planning in advance.

[0084] Generally speaking, a single inappropriate factor does not necessarily lead to a risk incident. It is usually the simultaneous or consecutive occurrence of multiple inappropriate operations that leads to a risk incident. In addition, the superposition of multiple factors leads to the occurrence of an operational misconduct incident. In order to analyze the probability of an incident, it is necessary to determine the importance of each factor to the risk incident.

[0085] Historical data on navigation accidents caused by improper ship operation were extracted and organized to obtain the element sequence {X1,X2,X3,...X} corresponding to the improper operation accidents. i ,...X I}, where X i It refers to the i-th element, and I refers to the total number of samples of the elements;

[0086] For a certain type of operational error accident Y j Analyze each accident caused by improper operation Y j The cause of the accident was improper operation during extraction. j The occurrence of the element sequence {X jk1 ,X jk2 ,...X jki ,...X jkI};

[0087] Where X jki This refers to the accident Y that caused the kth operation error. j When it occurs, element X i The indicated value, X jki =1 indicates that the accident Y is caused by improper operation in the kth operation. j When it occurs, element X exists. i The impact of X jki =0 indicates that the accident Y is due to improper operation during the kth operation. j Element X did not exist when it occurred. i The impact;

[0088] Get all feature sequences, delete X jki The element = 0 is then removed, and duplicate element sequences are eliminated to obtain Y, which caused the operational mishap. j Element combination sequence T j ={T j1 ,T j2 ...T jm ...T jM}; where T jm Y indicates an accident caused by improper operation. j The m-th element set,

[0089] Extract a risk incident sample set B from historical data that is affected by any combination of factors. j Based on expert scoring systems or statistical analysis, statistical T jm Combination of medium-term factors on risk incident sample set B j The influence of the medium sample is the lowest among all factor combinations, represented by the sample size b. jm.min Calculate T jm Influence coefficient under corresponding element combination Where b j Refers to sample set B j Total number of samples in the middle sample; calculation of accident Y due to improper operation. j The comprehensive weight of the combination of factors involved

[0090] Based on the above steps, calculate the ship problem accident O. p The combination of elements of ship problems O pq Corresponding factor combination influence coefficient and element combination sequence O p ={O p1 O p2 ...O pq ...O pQ The overall weight of} Calculate the route problem accident U v The combination of elements of the route problem U vs Corresponding factor combination influence coefficient and element combination sequence U v ={U v1 U v2 ...U vs ...U vS The overall weight of}

[0091] Among them O pq This refers to the q-th combination of ship problem elements corresponding to the p-th type of ship problem accident, b p This refers to the accident sample set B in historical data that is affected by any combination of ship-related problem elements. p Total number of samples, b pq.min It is sample set B p Combination of elements of Chinese shipbuilding issues O pq The impact level is the lowest among all combinations of ship-related factors; where U vs This refers to the combination of route problem elements in the s-th category corresponding to the v-th type of route problem accident, bv This refers to the accident sample set B in historical data that is affected by any combination of factors related to flight routes. v Total number of samples, b vs.min It is sample set B v Combination of elements of the problem in the middle route U vs The degree of impact is the lowest among all combinations of route problem elements in terms of sample size;

[0092] Based on the aforementioned steps, determine the comprehensive weights of different types of ship misoperation accidents, ship problem accidents, and route problem accidents, as well as the influence weights of the combination of elements corresponding to misoperation accidents, ship problem accidents, and route problem accidents;

[0093] The overall weight and the impact weight are sorted separately. For each type of ship, the accidents of improper operation, ship problems, and route problems are sorted in descending order of weight. At the same time, the corresponding combination elements within the accidents of improper operation, ship problems, and route problems are screened and sorted. From each type of accident, several combination elements with larger impact weights are selected in descending order to form the risk accident element model of each type of ship.

[0094] Based on the element types in the risk and accident element model of various types of ships, extract the historical navigation accident data of the corresponding ships; for each ship navigation accident sample, extract the element indicator value y. i Then normalize to obtain the normalized value y. i ;

[0095]

[0096] Where y i ′ .max It is the factor indicator y i The maximum value of y′ i ′ .min It is the factor indicator y i The minimum value of ′;

[0097] Based on the order of the risk and accident element diagrams for the corresponding ship type, the element sequence Y for each ship is obtained after sorting. w ={y w1 ,y w2 ,y w3 ,...y wl ,...y wL}; where y wl It refers to the index value of the l-th element in the risk accident element model of the w-th ship, and L refers to the total number of elements in the risk accident element model;

[0098] The element sequence Y w Unified conversion to graphical samples Organize all chart samples for each type of ship to create a database of nautical chart samples for that type of ship. Where H×W is the size of the graphic; the graphic sample contains the feature elements of each ship navigation accident, and all graphic samples of the same type of ship contain the navigation risk accident features of each type of ship. In order to obtain response features, a navigation risk accident identification scheme for different ships is established, and a multi-layer network classification model is used to classify and train the navigation graphic samples.

[0099] The multi-layer network classification model of this application is as follows: Figure 1 As shown, the multi-layer network classification model is based on the ResNet34 multi-branch network and includes a feature extraction module and a feature fusion module.

[0100] The input to the feature extraction module is a graph-based sample. The ResNet34 network generates output images with different dilation rates through multiple dilated convolutional blocks. The output images are then processed by the Inception network to extract their features.

[0101] The feature fusion module uses a multi-level feature weighting fusion network to perform feature weighting fusion, which includes dilating convolution, normalizing, linear rectified activation and adaptive average pooling of feature maps of output images with different dilation rates to output fused features;

[0102] Graph-based sample database The dataset is divided into a training dataset and a validation dataset. The multilayer network classification model is trained using the training dataset, and the parameters of the multilayer network classification model are optimized and adjusted using the validation dataset to obtain the final multilayer network classification model.

[0103] Inputting samples from the graph-based sample database into the trained multi-layer network classification model yields several types of classification results, resulting in typical feature categories representing navigation accidents of a certain type of ship. The navigation accident characteristics of ships with similar basic features to the classification results are consistent with these typical feature categories.

[0104] By collecting information on past navigation accidents of the vessel to be analyzed and inputting it into a classification network for classification, the typical characteristics of its navigation accidents can be obtained. By matching the risk accident characteristics of similar vessels with the same typical characteristics on the expected route of the vessel to be analyzed, the navigation risk accidents that the vessel to be analyzed may cause can be predicted, and targeted measures can be taken to avoid the occurrence of navigation risk accidents.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for controlling ship navigation risks, characterized in that, Includes the following steps: Step 1: Obtain navigation accident data information during the ship's navigation process through the ship navigation service system, formulate standard data files according to the needs of navigation risk influencing factors, and summarize the attribute data of various ship navigation service systems through the standard data files; Step 2: Determine the relevant elements under the categories of accident caused by improper operation, ship-related problems, and route-related problems; specifically: Historical data on navigation accidents caused by improper ship operation were extracted and organized to obtain the element sequence corresponding to the improper operation accidents. ,in It refers to the first One element, This refers to the total number of samples for each element; For a certain type of operational error Analyze each accident caused by improper operation. The reason for the accident was improper operation during extraction. The occurrence of element sequence ; in This refers to the cause of the first Improper operation accidents Elements of occurrence The indicated value, Indicates the first Improper operation accidents Elements present when it occurs The impact, Indicates the first Improper operation accidents There were no elements when it occurred. The impact; Get all feature sequences and delete. The elements are then removed, and duplicate element sequences are eliminated to obtain the sequence that caused the operational mishap. Element combination sequence ;in Indicates an accident caused by improper operation. The A set of elements, ; Extract a sample set of risk events affected by arbitrary combinations of factors from historical data. Based on expert scoring systems or statistical analysis, statistics Combination of medium-term factors on risk and accident sample sets The influence of the medium sample is the lowest among all factor combinations. ,calculate Influence coefficient under corresponding element combination ;in Refers to sample set Total number of samples in the middle sample; accidental calculation due to improper operation. The comprehensive weight of the combination of factors involved ; Based on the above steps, calculate ship problem accidents. Combination of elements of ship problems Corresponding factor combination influence coefficient and element combination sequence Overall weight ; Calculate flight path problems and accidents Combination of elements of the route problem Corresponding factor combination influence coefficient and element combination sequence Overall weight ; in It refers to the first The first type of ship-related accident Combination of elements of ship-related problems This refers to a set of accident samples from historical data that are affected by any combination of ship-related problem factors. Total number of samples It is a sample set Combination of elements of Chinese shipbuilding issues The impact level represents the lowest sample size among all combinations of ship-related factors; among which It refers to the first The first type of route problem accident The combination of factors related to the China-Europe route problem This refers to a set of accident samples in historical data that are affected by any combination of factors related to flight routes. Total number of samples It is a sample set Combination of problem elements in the Chinese route The degree of impact is the lowest among all combinations of route problem elements in terms of sample size; Step 3: Sort the comprehensive weight and impact weight respectively. For each type of ship, sort the improper operation accidents, ship problem accidents and route problem accidents in order of weight from large to small. At the same time, screen and sort the corresponding combination elements within the improper operation accidents, ship problem accidents and route problem accidents. Select several combination elements with large impact weights from each type of accident in order of large to small to form the risk accident element model of each type of ship. Step 4: Based on the element types in the risk accident element model of various types of ships, extract the historical navigation accident data of the corresponding ships; for each ship navigation accident sample, extract the element indicator values. Then normalize to obtain the normalized value. ; ； in It is a factor indicator The maximum value, It is a factor indicator The minimum value; Based on the order of the risk and accident element diagrams for the corresponding ship type, the element sequence for each ship is obtained after sorting. ;in It refers to the first In the risk accident element model of a ship, the first The index values ​​of each element, This refers to the total number of elements in the risk and incident element model; convert feature sequence Unified conversion to graphical samples ; Compile all nautical chart samples for each type of ship to create a database of nautical chart samples for that type of ship. ,in The dimensions of the graphic; Step 5: Use a multi-layer network classification model to classify and train the navigation chart-style samples; The multi-layer network classification model is based on the ResNet34 multi-branch network and includes a feature extraction module and a feature fusion module. The input to the feature extraction module is a graph-based sample. The ResNet34 network generates output images with different dilation rates through multiple dilated convolutional blocks. The output images are then processed by the Inception network to extract their features. The feature fusion module uses a multi-level feature weighting fusion network to perform feature weighting fusion, which includes dilating convolution, normalizing, linear rectified activation and adaptive average pooling of feature maps of output images with different dilation rates to output fused features; Graph-based sample database The dataset is divided into a training dataset and a validation dataset. The multilayer network classification model is trained using the training dataset, and the parameters of the multilayer network classification model are optimized and adjusted using the validation dataset to obtain the final multilayer network classification model. Step 6: Input the samples from the graph-based sample database into the multi-layer network classification model after training to obtain several types of classification results. This yields a typical feature category representing a certain type of ship's navigation accident. The navigation accident characteristics of ships with similar feature elements to the classification results are consistent with this typical feature category. By collecting information on past navigation accidents of the vessel to be analyzed and inputting it into a classification network for classification, the typical characteristics of its navigation accidents can be obtained. By matching the risk accident characteristics of similar vessels with the same typical characteristics on the expected route of the vessel to be analyzed, the navigation risk accidents of the vessel to be analyzed can be predicted, and risk management can be carried out based on the prediction results.

2. The method for controlling ship navigation risks according to claim 1, characterized in that, The ship navigation service system includes various ship navigation service systems such as VTS, GPS-CDMA, and AIS.

3. The method for controlling ship navigation risks according to claim 1, characterized in that, The standard data file includes data blocks with the following data metrics: Code index data block: Used to store the code sequence number data of each ship, serving as a unique identification code for each ship; Type Index Data Block: Used to store the type information of each vessel. The vessel types include at least small vessels, medium vessels, large vessels, unmanned vessels, and special vessels. Among them, small vessels refer to vessels with an overall length of less than or equal to 20m, medium vessels refer to vessels with an overall length of more than 20m but less than or equal to 100m, and large vessels refer to vessels with an overall length of more than 100m. Unmanned vessels refer to vessels that navigate based on autonomous driving technology, and special vessels refer to engineering vessels, scientific research vessels, and rescue vessels for specific purposes. Speed ​​data block: Used to store the speed information of each ship during navigation; Heading indicator data block: Used to store the route data information of each ship; Lifecycle Indicator Data Block: Used to store the average lifecycle of each vessel and its lifecycle stage information; Water depth index data block: used to store water depth data information of the ship at each navigation node or segment; Resistance Index Data Block: Used to store the ship's navigation resistance index, which refers to the ratio of the resistance generated by the ship's current course due to surface wind force and underwater fluid force during navigation at various navigation segments or nodes to the ship's power. Visibility index data block: used to store the visibility range data of ships at various navigation nodes or sections; Unmanned Vessel Index Data Block: Used to store the number and percentage of unmanned vessels on each current shipping route; Route Obstacle Index Data Block: Used to store obstacle distribution data on each flight segment; Risk Vessel Index Data Block: Used to store the proportion of high-risk vessels on each route. High-risk vessels refer to the proportion of vessels that would cause serious consequences in the event of a dangerous situation, or vessels that have a high risk of being overtaken in the current segment of the voyage.

4. The method for controlling ship navigation risks according to claim 1, characterized in that, The standard data files are saved using a unified CSV text file.

5. The method for controlling ship navigation risks according to claim 1, characterized in that, Step 1 also includes cleaning and completing the original data, specifically: A1. Since ship navigation data is continuous in time, for single missing data, it is filled by difference and average processing of the time before and after the sampling point. For continuous missing data, it is obtained by fitting data within the sampling period or by averaging historical sampling data in the same period. A2. For the original data after filling in the information, its quality needs to be analyzed and judged to ensure that the provided data has sufficient accuracy. Specifically, this includes: Verify the consistency of all data, ensuring that the data has the required type, magnitude, format, and units. Inconsistent data should be checked and analyzed, and erroneous data should be marked. This includes verifying whether speed numerical indicators have a relatively consistent magnitude and whether date, model, and serial number code data have a consistent text format. Perform fitting analysis on multiple samples of a single indicator to determine whether the indicator data exceeds the normal and reasonable range. Data that exceeds the normal and reasonable range is incorrectly labeled, including: numerical data exceeding the normal upper limit; and code data whose codes exceed the normal code range. A3. For the preprocessed raw data, we need to collect complete data for each voyage of a single vessel. Since the vessel has records in different navigation service systems, a large amount of duplicate data will be generated during the initial data collection process. These duplicate data exhibit random fluctuations within a certain range at their respective nodes. This fluctuation is due to differences in monitoring methods, monitoring time nodes, and data storage accuracy among various navigation service systems. Only one set of duplicate data is needed as the raw data for analysis. In order to preserve the data characteristics of these data, clustering analysis algorithms are used to compress the data.

Images