A shipping risk assessment method

Abstract

The application relates to a shipping risk assessment method, comprising the following steps: S1, establishing a semantic network of entities related to shipping and the relationship between the entities, and determining a shipping basic information knowledge graph; S2, acquiring multi-source data related to shipping risks, and extracting risk entities and risk information from the multi-source data; S3, expanding the shipping basic information knowledge graph according to the risk entities and the risk information, constructing a risk correlation knowledge graph, and acquiring the relationship between different risk factors and ship risks; S4, clustering the risk factors, constructing a risk identification model according to the clustering results, quantitatively evaluating the risks, obtaining a shipping risk level according to a shipping risk reasoning rule, and directly outputting the shipping risk level if the shipping risk level is low risk or high risk, otherwise, turning to S5; and S5, based on the quantitative evaluation results, weighted average calculation of the multi-factor risks is carried out to obtain a risk index, and the shipping risk level of the ship is determined according to the risk index. Compared with the prior art, the application has the advantages of high precision and reliability.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of shipping risk assessment, and in particular to a shipping risk assessment method. BACKGROUND

[0002] With the rapid development and wide application of big data technology and artificial intelligence, the shipping industry is entering a new era of intelligent transformation. In the complex environment of global trade, shipping companies face a variety of risks, including meteorological disasters, navigation safety, market fluctuations, etc. In order to improve the efficiency of risk management, data-driven risk identification and assessment methods have gradually attracted attention. As an advanced technical means, knowledge graph can integrate, analyze and reason a large amount of heterogeneous data in a structured and semantic way, providing intelligent risk identification and prediction support for the shipping industry.

[0003] In the shipping industry, the complexity and interrelation of risk factors are very strong, and the traditional risk management method usually relies on experience and pre-defined rules, which is difficult to handle large-scale and dynamic changing risk information. This leads to the increasing demand for accuracy and real-time of risk identification and assessment. By constructing a knowledge graph-based system, different sources of risk data can be effectively integrated to identify potential risk relationships, making the complex risk network clearer and providing strong support for decision-making.

[0004] In recent years, shipping risk identification and assessment methods mainly rely on data statistics and model analysis, usually relying on manually set rules and algorithms. These traditional methods usually require a large number of preprocessing steps, including data cleaning, feature extraction, risk classification, etc., and when dealing with variable shipping environments (such as weather changes, market fluctuations, operation errors, etc.), they often show certain limitations. Common risk identification methods include Bayesian networks, risk scorecards, etc., but their versatility is poor and they are difficult to adapt to complex scenarios and variable risk factors.

[0005] Because traditional methods rely heavily on experience rules, they are limited by subjective judgment and specific situations, and are not effective in dealing with different types of risks (such as ship accidents, port delays, natural disasters), especially when faced with complex shipping environments, traditional methods often cannot quickly and accurately identify potential risks. In addition, the diversity and dynamic nature of shipping data (such as real-time weather data, ship trajectory data, market fluctuation information) make it difficult for risk identification based on fixed rules to cope with, resulting in low accuracy and timeliness of risk assessment. SUMMARY

[0006] The purpose of the present application is to overcome the defects of the prior art and provide a precise and reliable shipping risk assessment method.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A shipping risk assessment method includes:

[0009] S1. Establish a semantic network of shipping-related entities and relationships between them, and determine a knowledge graph of basic shipping information;

[0010] S2. Obtain multi-source data related to shipping risks, and extract risk entities and risk information from the multi-source data;

[0011] S3. Expand the shipping basic information knowledge graph based on the extracted risk entities and risk information, construct a risk association knowledge graph, and obtain the relationship between different risk factors and ship risks;

[0012] S4. Clustering algorithms are used to cluster risk factors, and a risk identification model is constructed based on the clustering results to quantitatively assess the risks. The shipping risk level is obtained according to the shipping risk reasoning rules. If the shipping risk level is low risk or high risk, it is directly output; otherwise, go to S5.

[0013] S5. Based on the quantitative assessment results of S4, the risk index is calculated by weighted average of multiple risk factors, and the shipping risk level of the ship is determined according to the risk index.

[0014] Preferably, in step S1, a semantic network is established concerning shipping-related entities and the relationships between them, and a knowledge graph of basic shipping information is determined. Specifically, this includes:

[0015] Define shipping-related entities E = {E1, E2, ..., E...} n The entities include ship entities, port entities, shipping route entities, meteorological entities, and crew entities;

[0016] Determine entity E i Attributes in, For entity E i The mth i One attribute;

[0017] Construct a mapping network between entities, for entity E i With entity E j The mapping relationship is represented as R(E) i E j )={R1, R2, ..., R p}, where R p For entity E i With entity E j The p-th mapping result between them;

[0018] The resource description framework is used to describe entities and the relationships between them, where entity E i With entity E j The resource description framework relationship network structure between them is represented as RDFs = (E i R, E j );

[0019] Construct a knowledge graph of basic shipping information G = (E, A) E A G W E R E ), where E is the set of entity nodes, A E ={A(E1), A(E2), ..., A(E n A is the set of attributes for all entity nodes. G =[a ij ], (e i e j )∈E indicates that entity node i and entity node j are associated, W E R represents the strength of the entity relationship. E It is a relational type matrix.

[0020] Preferably, the multi-source data related to shipping risks includes meteorological data, shipping accident reports, vessel information, and route information, wherein the shipping accident reports contain human factors.

[0021] Preferably, step S2 involves acquiring multi-source data related to shipping risks and extracting risk entities and risk information from the multi-source data, specifically including:

[0022] S21. Match the multi-source data based on the timestamp attribute and geographic coordinate attribute, and convert the multi-source data into a unified representation. Among them, U k This is a unified representation of the k-th data item, where t is the timestamp, and x and y are the geographic coordinates. For the Mth data after unifying the kth data k One attribute, M k The number of attributes in the k-th data item;

[0023] S22. Use a trained conditional random field model to identify risk entities and risk information in shipping accident reports.

[0024] Preferably, in step S22, a pre-trained Conditional Random Field (CRF) model is used to identify risk entities and risk information in shipping accident reports. The pre-training process of the CRF model specifically includes:

[0025] The named entity recognition technology is used to label each word in the shipping accident report, and the label is a set entity category;

[0026] The obtained text data T is converted into a feature vector F={f1, f2,..., f s} where f s represents the s-th feature;

[0027] The conditional random field model is trained using the labeled data set, and the mathematical expression of the conditional random field model is:

[0028]

[0029] where λ k is the weight of the k-th feature, f k is the k-th feature, u is the number of words in the input sequence x, and Z(x) is the normalization factor.

[0030] Preferably, in S3, the shipping basic information knowledge graph is expanded according to the extracted risk entity and risk information, a risk correlation knowledge graph is constructed, and the relationship between different risk factors and ship risk is obtained, including the following sub-steps:

[0031] S31, according to the extracted risk entity and risk information, the shipping basic information knowledge graph is expanded, and a risk correlation knowledge graph is constructed;

[0032] S32, the risk factors affecting the safety of ship navigation under different weather, sea area state and crew situation are extracted, and a random variable set {X1, X2,..., X q} representing the shipping risk factors is established, where X q corresponds to the q-th shipping risk factor, and the shipping risk factors include wind speed, wave height, ship type, channel condition, sea area depth, crew experience, crew working time, crew fatigue and operation error rate;

[0033] S33, the influence degree of different weather factors, sea area state and human factors on navigation safety is obtained using Bayesian network, and the risk level with the maximum posterior probability is selected as the final predicted risk level;

[0034] S34, the Bayesian network risk assessment result obtained by S33 is used as the input for querying and reasoning the relationship between different factors under complex weather and sea area state, and SPARQL is used to query the relationship between multiple factors and ship risk under different weather and sea area state, wherein the multiple factors include ship age, ship type and channel congestion degree.

[0035] Preferably, in S4, a clustering algorithm is used to cluster the risk factors, including the following sub-steps:

[0036] S401, feature extraction is performed by using One-hot coding, and different shipping risk factors are converted into feature vectors;

[0037] S402, the incomplete multi-view clustering algorithm based on self-representation is used to cluster the shipping risk factors, and the potential correlation between the data under different views is captured, and the clustered risk feature vectors C={C1, C2,..., C r} are outputted, wherein C r is the rth feature group after clustering, indicating a feature cluster closely related to the risk under certain conditions;

[0038] The self-incomplete multi-view clustering algorithm based on self-representation uses the subspace information shared by multiple views to learn a consistency matrix S to save the effective consistency information of multiple views, and assigns different weights w v to the views, and the objective function expression is:

[0039]

[0040]

[0041] wherein, X v is the matrix of the vth view, d v is the feature dimension of the vth view, N is the number of samples, represents the known data matrix of the vth view, is the extraction matrix of the vth view, H v is the representation matrix of the vth view.

[0042] Preferably, the risk identification model is constructed according to the clustering results in the S4, the risk is quantitatively evaluated, the shipping risk level is obtained according to the shipping risk reasoning rules, and the specific steps include the following sub-steps:

[0043] S411, a risk identification model is constructed, the risk feature vectors C obtained by clustering and the consistency matrix S are taken as input data, and the risk quantitative evaluation results are outputted;

[0044] S412, the shipping risk is reasoned by using the knowledge reasoning to construct the shipping risk reasoning rules, and the shipping risk reasoning rules include weather condition reasoning sub-rules, ship state reasoning sub-rules, environmental condition reasoning sub-rules and human factor reasoning sub-rules;

[0045] S413, according to the correlation between the shipping risk factors, the determination result of each reasoning sub-rule is comprehensively reasoned and determined, and the multi-factor superimposed shipping risk level reasoning result is obtained, if the reasoning result of each reasoning sub-rule is low risk, it is determined that the shipping risk level is low risk and directly output, if the reasoning result of any reasoning sub-rule is high risk, it is determined that the shipping risk level is high risk and directly output, and other cases are converted to S5.

[0046] Preferably, the shipping risk reasoning rules in S412 specifically include:

[0047] The weather condition reasoning sub-rule includes:

[0048] If the wind speed > A1 km / h and the wave height > B meters and the route is open sea, it is determined as high risk;

[0049] If the wind speed ≤ A1 km / h and the wave height ≤ B meters and the route is a coastal route, it is determined as medium risk; and

[0050] If the wind speed ≤ A2 km / h and the weather is sunny and the route is an inland river route, it is determined as low risk;

[0051] The ship state reasoning sub-rule includes:

[0052] If the ship type is a small ship and the load > C% and the wind speed > A3 km / h, it is determined as high risk;

[0053] If the ship type is a large cargo ship and sails in a safe channel and the wind speed ≤ A3 km / h, it is determined as low risk; and

[0054] If the ship type is a large cargo ship and the wind speed > A1 km / h and sails in open sea, it is determined as medium risk;

[0055] The environmental condition reasoning sub-rule includes:

[0056] If the sea area is a shallow water area and the current speed > D m / s and the visibility < E1 km, it is determined as high risk;

[0057] If the route is a safe channel and the visibility > E2 km, it is determined as low risk; and

[0058] If the route is a busy channel and the visibility < E3 km, it is determined as medium risk;

[0059] And the human factor reasoning sub-rule includes:

[0060] If the crew working time > F1 hours and the crew fatigue is high and the ship type is a small ship, it is determined as high risk;

[0061] if the crew experience > G years and the crew fatigue is low and the operation failure rate < H1%, it is determined to be low risk; and

[0062] if the crew work duration is < F2 hours and the operation failure rate > H2% and the weather condition is high risk, it is determined to be medium risk;

[0063] wherein A1, A2, A3, B, C, D, E1, E2, E3, F1, F2, G, H1 and H2 are set constants, and satisfy A1 > A3 > A2, E2 > E1 > E3, F1 > F2, H2 > H1.

[0064] Preferably, based on the quantitative evaluation results of S4 in S5, the risk index is calculated by weighted average of multi-factor risk, and the navigation risk level of the ship is determined according to the risk index, and the risk index a calculation expression is:

[0065]

[0066] wherein a i is the corresponding risk quantitative evaluation result of the i th shipping risk factor, quantifying high risk as 9, medium risk as 5, and low risk as 2, w i is the weight of the i th shipping risk factor, obtained using a regression model, and

[0067] For the risk index a, if a [0, 3) is determined to be low risk, if a [3, 7) is determined to be medium risk, and if a [7, 10], it is high risk.

[0068] Compared with the prior art, the present application has the following beneficial effects:

[0069] (1) The present application defines the attributes and relationships of shipping related entities, constructs a knowledge graph to integrate shipping multi-source data including meteorological data, shipping accident reports, ship information, route information and human factors, realizes the effective association between shipping multi-source data, can comprehensively reflect various factors affecting ship navigation safety, greatly improves the understanding and evaluation ability of shipping risk, and ensures that the decision is more scientific.

[0070] (2) The present application constructs a risk identification model according to the risk factor clustering results, quantitatively evaluates the shipping risk level using shipping risk reasoning rules, and combines knowledge graph and machine learning, which can deeply capture the complex relationship between shipping risk factors, such as associating weather conditions, navigation environment, crew conditions and historical accident data, providing context support, and significantly improving the accuracy of shipping risk prediction.

[0071] (3) This invention uses a knowledge graph-based reasoning mechanism to perform complex risk reasoning operations based on established entity relationships, automatically inferring risk associations under specific conditions, such as the risk exposure levels of different ship types under specific weather conditions. This provides a practical method for discovering potential risks and adds important basis for the scientific and systematic nature of shipping decision-making process, ensuring that shipping management is more efficient and safer.

[0072] (4) The self-representation-based incomplete multi-perspective clustering algorithm is used to cluster shipping risk factors, capture the potential correlation between data from different perspectives, improve the mining and utilization of information in existing shipping data, and further improve the accuracy of subsequent shipping assessment results. Attached Figure Description

[0073] Figure 1 This is a flowchart of the method of the present invention;

[0074] Figure 2 This is a schematic diagram illustrating the process of shipping risk prediction and quantitative assessment.

[0075] Figure 3 This is a diagram illustrating the risk assessment process. Detailed Implementation

[0076] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0077] Example

[0078] like Figure 1 As shown in the figure, this embodiment provides a shipping risk assessment method that can effectively optimize the bottleneck problems of shipping risk information dispersion and risk identification and assessment accuracy. The method includes the following steps:

[0079] S1. Establish a semantic network of shipping-related entities and relationships between them, and determine a knowledge graph of basic shipping information;

[0080] S2. Obtain multi-source data related to shipping risks, and extract risk entities and risk information from the multi-source data;

[0081] S3. Expand the shipping basic information knowledge graph based on the extracted risk entities and risk information, construct a risk association knowledge graph, and obtain the relationship between different risk factors and ship risks;

[0082] S4. Clustering algorithms are used to cluster risk factors, and a risk identification model is constructed based on the clustering results to quantitatively assess the risks. The shipping risk level is obtained according to the shipping risk reasoning rules. If the shipping risk level is low risk or high risk, it is directly output; otherwise, go to S5.

[0083] S5. Based on the quantitative assessment results of S4, the risk index is calculated by weighted average of multiple risk factors, and the shipping risk level of the ship is determined according to the risk index.

[0084] The method of this embodiment will now be described in detail.

[0085] S1. Establish a semantic network of entities related to shipping and the relationships between them, and determine a knowledge graph of basic shipping information, specifically including:

[0086] S11. Define shipping-related entities E = {E1, E2, ..., E...} n The entities include ship entities, port entities, shipping route entities, weather entities, and crew entities; for example, E1 = ship, E2 = port, E3 = shipping route, E4 = weather, E5 = crew, etc.

[0087] S12, Determine entity E i Attributes in, For entity E i The mth i One attribute;

[0088] For example, A (ship) = {deadweight, length, ...}, A (port) = {throughput, geographical location, ...}, A (crew) = {crew experience years, crew working hours, ...}, etc.

[0089] S13. Construct a mapping network between entities, for entity E i With entity E j The mapping relationship is represented as R(E) i E j )={R1, R2, ..., R p}, where R p For entity E i With entity E j The p-th mapping result between them.

[0090] For terms such as "ship-berthing-port", "ship-navigation-route", and "route-affected-weather", there are R(ship, port) = berthing, R(ship, route) = navigation, and R(route, weather) = affected.

[0091] S14. Use the Resource Description Framework (RDF) to describe entities and the relationships between them, where entity Ei With entity E j The resource description framework relationship network structure between them is represented as RDFs = (E i R, E j ).

[0092] S15. Construct a knowledge graph of basic shipping information G = (E, A) E A G W E R E ), where E is the set of entity nodes, A E ={A(E1), A(E2), ..., A(E n A is the set of attributes for all entity nodes. G =|a ij ], (e i e j )∈E indicates that entity node i and entity node j are associated, W E R represents the strength of the entity relationship. E It is a relational type matrix.

[0093] S2. Acquire multi-source data related to shipping risks, extract risk entities and risk information from the multi-source data, and perform preprocessing. Multi-source data related to shipping risks includes meteorological data, shipping accident reports, vessel information, and route information, etc. Shipping accident reports contain human factors. The specific steps include the following:

[0094] Before processing multi-source data, missing value imputation and outlier handling are performed, and the format is standardized.

[0095] S21. Match the multi-source data based on the timestamp attribute and geographic coordinate attribute, and convert the multi-source data into a unified representation. Among them, U k This is a unified representation of the k-th data item, where t is the timestamp, and x and y are the geographic coordinates. For the Mth data after unifying the kth data k One attribute, M k The number of attributes in the k-th data item;

[0096] S22. Use command entity recognition technology to extract shipping risk information from shipping accident reports, and use a trained conditional random field model to identify important risk entities and risk information such as ship names, ports, and weather conditions in shipping accident reports. For example, identify important entities such as ship names, ports, and weather conditions in shipping reports.

[0097] The pre-training process of the Conditional Random Field model is as follows:

[0098] The named entity recognition technology NER is used to label each word in the shipping accident report with a label, wherein the label is a set entity category (such as "ship state", "port condition", "weather condition" or "human factor");

[0099] The obtained text data T is converted into a feature vector F = {f1, f2,..., f s}, wherein f s represents the s-th feature, and the features usually include the word itself, the word form, the prefix and suffix of the word, whether it is capitalized, the number feature, the context word, etc.

[0100] The obtained text data T is converted into a feature vector F = {f1, f2,..., f s}, wherein f s represents the s-th feature.

[0101] The conditional random field model is trained using the labeled data set, and the mathematical expression of the conditional random field model is:

[0102]

[0103] wherein λ k is the weight of the k-th feature, f k is the k-th feature, u is the number of words in the input sequence x, and Z(x) is a normalization factor to ensure that the sum of the probabilities of all possible label sequences is 1.

[0104] S3, according to the extracted risk entity and risk information, the shipping basic information knowledge graph is expanded, the risk association knowledge graph is constructed, and the relationship between different risk factors and ship risk is obtained, such as Figure 2 as shown, specifically including the following sub-steps:

[0105] S31, according to the extracted risk entity and risk information, the shipping basic information knowledge graph is expanded, the risk association knowledge graph KG2 is constructed, and KG2 = KG1∪E ner ∪R risk , KG1 is the shipping basic information knowledge graph, E ner is the set of risk entities identified from the shipping accident report, and R risk is the set of relationships between the risk entities identified from the shipping accident report;

[0106] S32, the risk factors affecting the safety of ship navigation under different weather, sea area state and crew conditions are extracted, and a set of random variables {X1, X2,..., X q} representing shipping risk factors is established, wherein X qCorresponding to the qth shipping risk factor, the shipping risk factor includes wind speed, wave height, ship type, channel condition, sea depth, crew experience, crew working time, crew fatigue and operation failure rate, for example, including:

[0107] X1 = wind speed (such as: {low, medium, high})

[0108] X2 = wave height (such as: {low, medium, high})

[0109] X3 = ship type (such as: {small ship, medium ship, large ship})

[0110] X4 = crew experience (such as: {junior, intermediate, senior})

[0111] X5 = crew working time (such as: {short, moderate, long})

[0112] X6 = crew fatigue (such as: {low, medium, high})

[0113] …

[0114] X q = other risk factors (such as channel condition, sea depth, etc.)

[0115] S33, use Bayesian network to obtain the influence degree of different weather factors, sea state and human factors on navigation safety, select the risk level risk with maximum posterior probability * As the final predicted risk level, the calculation expression is:

[0116]

[0117] Where P(risk) is the prior probability, P(X i |risk) is the conditional probability of each risk factor given the risk level risk;

[0118] S34, use the Bayesian network risk assessment result obtained by S33 as the input for querying and reasoning the relationship between different factors under complex weather and sea state, and use SPARQL to query the relationship between multiple factors (such as ship age, ship type and channel congestion degree, etc.) and ship risk under different weather and sea state.

[0119] S4, use clustering algorithm to cluster risk factors, and construct risk identification model according to the clustering result to quantitatively evaluate the risk, get the shipping risk level according to the shipping risk reasoning rule, if the shipping risk level is low risk or high risk, directly output, and form a risk assessment report, otherwise go to S5, including:

[0120] Use clustering algorithm to cluster risk factors, including the following sub-steps:

[0121] S401, feature extraction is performed using One-hot encoding, and different shipping risk factors (such as weather conditions, ship type, sailing time, crew working hours, etc.) are converted into feature vectors X = [x1, x2,..., x n ];

[0122] S402, similar shipping risk factors are clustered using an incomplete multi-view clustering algorithm based on self-representation, capturing the potential relevance between data under different views, and outputting the clustered risk feature vector C = {C1, C2,..., C r} where C r is the rth feature group after clustering, representing a feature cluster closely related to risk under certain conditions.

[0123] Due to the complexity of the association between shipping risk factors, such as different weather conditions that may affect different types of ships. In the basic missing case, the spectral clustering algorithm is added with a self-representation model This model represents or reconstructs data with its own linear combination, which can effectively capture the potential relevance between data under different views. For example, under different sea areas and weather conditions, certain types of ships may be more susceptible to risk.

[0124] In this embodiment, the incomplete multi-view clustering algorithm IMCS based on self-representation uses subspace information shared by multiple views to learn a consistency matrix S to save effective consistency information of multiple views, and assigns different weights w v to the views, and the objective function expression is:

[0125]

[0126]

[0127] where X v is the matrix of the vth view, d v is the feature dimension of the vth view, N is the number of samples, represents the known data matrix of the vth view, is the extraction matrix of the vth view, H v is the representation matrix of the vth view.

[0128] According to the clustering results, a risk identification model is constructed to quantitatively evaluate the risk, and the shipping risk level is obtained according to the shipping risk reasoning rules, which specifically includes the following sub-steps:

[0129] S411, a risk identification model is constructed, and the risk feature vector C obtained by clustering and the consistency matrix S are used as input data, and the risk quantitative evaluation result is output;

[0130] The feature vectors and risk labels (low risk, medium risk, and high risk) in the clustering result are used as training data to train the risk identification model.

[0131] S412, using knowledge reasoning technology to construct shipping risk reasoning rules based on weather conditions, navigation environment, and crew conditions, etc. to reason the shipping risk, the shipping risk reasoning rules including weather condition reasoning sub-rules, ship state reasoning sub-rules, environmental condition reasoning sub-rules, and human factor reasoning sub-rules;

[0132] 1) Weather condition reasoning sub-rules, such as wind speed, wave height, visibility, etc. meteorological condition factors, comprehensive evaluation according to different conditions and route types, including:

[0133] If the wind speed > A1 km / h (in this embodiment, A1 = 50) and the wave height > B meters (in this embodiment, b = 5) and the route is open sea, it is determined as high risk;

[0134] If the wind speed ≤ A1 km / h (in this embodiment, A1 = 50) and the wave height ≤ B meters (in this embodiment, b = 5) and the route is a coastal route, it is determined as medium risk; and

[0135] If the wind speed ≤ A2 km / h (in this embodiment, A2 = 20) and the weather is sunny and the route is an inland river route, it is determined as low risk;

[0136] 2) Ship state reasoning sub-rules, such as ship load, ship type, navigation speed, etc. factors, combined with weather conditions and navigation environment for comprehensive evaluation, including:

[0137] If the ship type is a small ship and the load > C% (in this embodiment, C = 90) and the wind speed > A3 km / h (in this embodiment, A3 = 30), it is determined as high risk;

[0138] If the ship type is a large cargo ship and the navigation is in a safe channel and the wind speed ≤ A3 km / h (in this embodiment, A3 = 30), it is determined as low risk; and

[0139] If the ship type is a large cargo ship and the wind speed > A1 km / h (in this embodiment, A1 = 50) and the navigation is in open sea, it is determined as medium risk;

[0140] 3) Environmental condition reasoning sub-rules, such as sea area characteristics, channel conditions, etc., including:

[0141] If the sea area is a shallow water area and the current speed > D m / s (in this embodiment, D = 3) and the visibility < E1 km (in this embodiment, E1 = 2), it is determined as high risk;

[0142] If the route is a safe waterway and the visibility is > E2 km (E2 = 5 in this embodiment), it is determined to be low risk; and

[0143] If the route is a busy waterway and the visibility is < E3 km (E3 = 1 in this embodiment), it is determined to be medium risk;

[0144] 4) Human factor reasoning sub-rules, such as crew working hours, experience, fatigue level, operation error rate and other factors, combined with ship state and weather conditions for evaluation, including:

[0145] If the crew working hours (daily working hours) > F1 hours (F1 = 12 in this embodiment) and the crew fatigue level is high and the ship type is a small ship, it is determined to be high risk;

[0146] If the crew experience > G years (G = 10 in this embodiment) and the crew fatigue level is low and the operation error rate < H1% (H1 = 5 in this embodiment), it is determined to be low risk; and

[0147] If the crew working hours are ≤ F2 hours (F2 = 8 in this embodiment) and the operation error rate > H2% (H2 = 10 in this embodiment) and the weather condition is high risk, it is determined to be medium risk.

[0148] S413, according to the correlation between the shipping risk factors, the determination results of each reasoning sub-rule are comprehensively reasoned and determined, and the multi-factor superimposed shipping risk level reasoning result is obtained, as shown in the following table: Figure 3 If the reasoning results of each reasoning sub-rule are all low risk, it is determined that the shipping risk level is low risk and is directly output, if the reasoning result of any reasoning sub-rule is high risk, it is determined that the shipping risk level is high risk and is directly output, and other cases are converted to S5.

[0149] S5, based on the quantitative evaluation results of S4, the risk index is calculated by weighted average of multi-factor risk, and the shipping risk level of the ship is determined according to the risk index.

[0150] Based on the quantitative evaluation results of S4, the risk index is calculated by weighted average of multi-factor risk, and the shipping risk level of the ship is determined according to the risk index, and the risk index a calculation expression is:

[0151]

[0152] Wherein, a i is the corresponding risk quantitative evaluation result of the i th shipping risk factor, the high risk is quantified as 9, the medium risk is quantified as 5, and the low risk is quantified as 2, w i is the weight of the i th shipping risk factor, which is obtained by using a regression model, and

[0153] For the risk index a, if a∈[0, 3) is determined as low risk, if a∈[3, 7) is determined as medium risk, and if a∈[7, 10] is determined as high risk.

[0154] In this embodiment, the risk index is recalculated every certain time, the risk level is updated according to the new risk index, and the risk condition is continuously monitored.

[0155] The above merely provides a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements shall be encompassed within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method of shipping risk assessment, characterized by, The method comprises the following steps: S1, a semantic network of entities related to shipping and the relationship between entities is established, and a basic information knowledge graph of shipping is determined; S2, multi-source data related to shipping risks is obtained, and risk entities and risk information are extracted from the multi-source data; The multi-source data related to shipping risks comprises meteorological data, shipping accident reports, ship information and route information, wherein the shipping accident reports contain human factors; S3, the basic information knowledge graph of shipping is extended according to the extracted risk entities and risk information, a risk correlation knowledge graph is constructed, and the relationship between different risk factors and ship risks is obtained; S4, a clustering algorithm is used to cluster the risk factors, and a risk identification model is constructed according to the clustering results to quantitatively evaluate the risks, and the shipping risk level is obtained according to the shipping risk reasoning rules, if the shipping risk level is low risk or high risk, it is directly output, otherwise, it is converted to S5; S5, based on the quantitative evaluation results of S4, the risk index is calculated by weighted average of multi-factor risks, and the shipping risk level of the ship is determined according to the risk index; In S4, the clustering algorithm is used to cluster the risk factors, and the following sub-steps are included: S401, One-hot encoding is used for feature extraction, and different shipping risk factors are converted into feature vectors; S402, using an incomplete multi-view clustering algorithm based on self-representation to cluster shipping risk factors, capturing the potential correlation between data under different perspectives, and outputting a clustered risk feature vector wherein is the th feature group after clustering, representing a feature cluster closely related to the risk under certain conditions; The self-based incomplete multi-view clustering algorithm uses the subspace information shared by multiple views to learn a consistency matrix S to save the effective consistency information of multiple views, and assigns different weights to the views , and the target function expression is wherein, is a matrix of the th view, is a feature dimension of the th view, is a number of samples, denotes a known data matrix of the th view, is an extraction matrix of the th view, is a representation matrix of the th view; In S4, the risk identification model is constructed according to the clustering results to quantitatively evaluate the risks, and the shipping risk level is obtained according to the shipping risk reasoning rules, which includes the following sub-steps: S411, a risk identification model is constructed, the risk feature vectors C obtained by clustering and the consistency matrix S are taken as input data, and the risk quantitative evaluation results are output; S412, knowledge reasoning is used to construct shipping risk reasoning rules to reason the shipping risks, wherein the shipping risk reasoning rules comprise weather condition reasoning sub-rules, ship state reasoning sub-rules, environmental condition reasoning sub-rules and human factor reasoning sub-rules; S413, according to the correlation between the shipping risk factors, the determination results of each reasoning sub-rule are comprehensively reasoned and determined, the multi-factor superimposed shipping risk level reasoning results are obtained, if the reasoning results of each reasoning sub-rule are low risk, it is determined that the shipping risk level is low risk and directly output, if the reasoning result of any reasoning sub-rule is high risk, it is determined that the shipping risk level is high risk and directly output, otherwise, it is converted to S5.

2. The method of claim 1, wherein, In S1, the semantic network of entities related to shipping and the relationship between entities is established, and the basic information knowledge graph of shipping is determined, which includes the following steps: Defining shipping-related entities , the entities including a ship entity, a port entity, a route entity, a weather entity, and a crew entity; determining attributes of an entity wherein, is a first attribute of the entity ​​​ Construct a mapping network between entities, for entities With entity The mapping relationship is represented as ,in, For entities With entity Between the first One mapping result; An entity and relationships between entities are described using a resource description framework, wherein the entity is represented as a resource description framework relationship network structure between the entity ; Constructing shipping basic information knowledge graph wherein, is a set of entity nodes, is a set of attributes of all entity nodes, , , represents an entity node is associated with an entity node , represents the strength of the entity relationship, is a relationship type matrix.

3. The method of claim 1, wherein, In S2, the multi-source data related to shipping risks is obtained, and the risk entities and risk information are extracted from the multi-source data, which includes the following steps: S21, according to the timestamp attribute and the geographic coordinate attribute, matching the multi-source data, and converting the multi-source data into a unified representation wherein, is a first unified representation of the data, is a timestamp, and is a geographic coordinate, is a first unified representation of the data, is a first attribute of the unified data, is a number of attributes in the data. S22, a trained conditional random field model is used to identify the risk entities and risk information in the shipping accident reports.

4. The method of claim 1, wherein, In S22, the trained conditional random field model is used to identify the risk entities and risk information in the shipping accident reports, and the pre-training process of the conditional random field model is as follows: A named entity recognition technology is used to label each word in the shipping accident report, wherein the label is a set of entity categories; The obtained text data T is converted into a feature vector wherein, represents the th feature; The conditional random field model is trained by using the labeled data set, and a mathematical expression of the conditional random field model is as follows: in, For the first The weights of each feature, For the first One characteristic, For the input sequence Number of words in Chinese It is the normalization factor.

5. The method of claim 1, wherein, The S3 extends the shipping basic information knowledge graph according to the extracted risk entity and risk information, constructs a risk correlation knowledge graph, and obtains the relationship between different risk factors and ship risks, including the following sub-steps: S31, extending the shipping basic information knowledge graph according to the extracted risk entity and risk information, and constructing a risk correlation knowledge graph; S32, extracting risk factors affecting the safety of ship navigation under different weather, sea state and crew conditions, and establishing a set of random variables representing shipping risk factors wherein corresponding to the th shipping risk factor, the shipping risk factor includes wind speed, wave height, ship type, channel condition, sea depth, crew experience, crew working time, crew fatigue and operation failure rate; S33, using a Bayesian network to obtain the influence degree of different weather factors, sea area states and human factors on navigation safety, selecting a risk level with maximum posterior probability as a final predicted risk level; S34, using the Bayesian network risk assessment result obtained in S33 as an input for query and reasoning of the relationship between different factors under complex weather and sea area states, and using SPARQL to query the relationship between multiple factors and ship risks under different weather and sea area states, wherein the multiple factors include ship age, ship type and channel congestion degree.

6. The method of claim 1, wherein, The shipping risk reasoning rules in the S412 specifically include: A weather condition reasoning sub-rule, including: If the wind speed > A1 km / h and the wave height > B meters and the route is an open sea area, it is determined as high risk; If the wind speed ≤ A1 km / h and the wave height ≤ B meters and the route is a coastal route, it is determined as medium risk; and If the wind speed ≤ A2 km / h and the weather is sunny and the route is an inland waterway, it is determined as low risk; A ship state reasoning sub-rule, including: If the ship type is a small ship and the load > C % and the wind speed > A3 km / h, it is determined as high risk; If the ship type is a large cargo ship and the navigation is in a safe channel and the wind speed ≤ A3 km / h, it is determined as low risk; and If the ship type is a large cargo ship and the wind speed > A1 km / h and the navigation is in an open sea area, it is determined as medium risk; An environmental condition reasoning sub-rule, including: If the sea area is a shallow water area and the current speed > D m / s and the visibility < E1 km, it is determined as high risk; If the route is a safe channel and the visibility > E2 km, it is determined as low risk; and If the route is a busy channel and the visibility < E3 km, it is determined as medium risk; And a human factor reasoning sub-rule, including: If the crew working hours > F1 hours and the crew fatigue is high and the ship type is a small ship, it is determined as high risk; If the crew experience > G years and the crew fatigue is low and the operation failure rate < H1 %, it is determined as low risk; and If the crew working hours ≤ F2 hours and the operation failure rate > H2 % and the weather condition is high risk, it is determined as medium risk; wherein, A1, A2, A3, B, C, D, E1, E2, E3, F1, F2, G, H1 and H2 are constants set and satisfy , , , .

7. The method of claim 1, wherein, The risk index is calculated by weighting and averaging the multi-factor risks based on the quantitative evaluation results of S4 in S5, and the navigation risk level of the ship is determined according to the risk index The calculation expression is: in, For the first The risk quantification assessment results for each shipping risk factor are as follows: high risk is quantified as 9, medium risk as 5, and low risk as 2. For the first The weights of each shipping risk factor were obtained using a regression model, and ; For the risk index , if then it is determined to be low risk, if then it is determined to be medium risk, and if then it is high risk.

Images