Water traffic accident cause prediction method, device, equipment and medium

Abstract

The invention relates to a water traffic accident cause prediction method, device and equipment and a medium, and belongs to the technical field of water traffic safety, and the water traffic accident cause prediction method comprises the steps: constructing a water traffic accident cause body based on obtained water traffic accident investigation report information; the water traffic accident cause ontology is used as a knowledge framework, a constructed graph neural network is trained based on the water traffic accident investigation report information, the graph neural network comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises a plurality of graph convolutional network units and a random inactivation unit; the to-be-predicted water traffic accident data is predicted based on the fully trained graph neural network, the water traffic accident cause prediction result is obtained, and the accuracy and efficiency of water traffic accident cause analysis and prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of water traffic safety, and in particular to a method, device, equipment and medium for predicting causes of water traffic accidents. Background Art

[0002] Water traffic accidents often cause serious casualties, property losses and environmental pollution. Water traffic accident investigation is an important way to identify accident responsibilities and improve navigation rules. However, traditional water traffic accident investigations mainly follow the steps of evidence collection, evidence analysis, finding out the cause of the accident and comprehensive analysis. Since it is necessary to analyze and investigate evidence from multiple factors such as crew behavior, degree of damage to the ship, cargo conditions and natural environment, and complete a comprehensive integrated evaluation after analyzing them one by one, the time required for accident investigation is prolonged. The traditional system integration investigation and processing form will also have an adverse effect on the final comprehensive analysis.

[0003] At present, the research field of causal analysis of water traffic accidents mainly focuses on extracting causal factors from accident investigation reports. This process usually involves establishing mathematical models to identify the causes of accidents or the severity of accidents, thereby achieving the prevention of water traffic accidents.

[0004] However, the reliance on the extracted causes and analysis of the severity of accidents often results in a lag problem, which makes it impossible to respond quickly and effectively to potential accident risks. In addition, the current method has certain limitations in universality. The information extracted for a specific accident may be difficult to use effectively in other similar situations. This limitation makes it difficult to accurately reflect the causes and patterns of different types of water traffic accidents, which in turn affects the effectiveness of accident prevention strategies. Summary of the invention

[0005] In view of this, it is necessary to provide a method, device, equipment and medium for predicting the causes of water traffic accidents to solve the technical problems of lag and limitations in the analysis of the causes of water traffic accidents.

[0006] In order to solve the above problems, the present invention provides a method for predicting the cause of water traffic accidents, comprising: Construct the water traffic accident cause ontology based on the acquired water traffic accident investigation report information; Taking the water traffic accident cause ontology as the knowledge framework, the constructed graph neural network is trained based on the water traffic accident investigation report information, wherein the graph neural network includes an input layer, a hidden layer and an output layer, and the hidden layer includes a plurality of graph convolutional network units and a random inactivation unit; Based on the well-trained graph neural network, the water traffic accident data to be predicted is predicted to obtain the prediction results of the causes of water traffic accidents.

[0007] In a possible implementation, the water traffic accident cause ontology is constructed based on the acquired water traffic accident investigation report information, including: Acquire water traffic accident investigation report information, and determine the water traffic accident cause field and the ontology of the water traffic accident cause field based on the water traffic accident investigation report information and expert experience; Determine the key terms of the ontology according to expert experience and rule requirements, and obtain a term list based on the key terms; Analyzing and defining the term list to determine the classes of the ontology and the class hierarchy, wherein the hierarchy includes primary concepts, secondary concepts, and tertiary concepts; Based on the classes and class hierarchical structure of the ontology, the relationship attributes and data attributes of the classes are defined to construct a water traffic accident cause ontology, wherein the water traffic accident cause ontology includes a head entity, a relationship, and a tail entity.

[0008] In a possible implementation, the water traffic accident cause ontology is used as a knowledge framework, and the constructed graph neural network is trained based on the water traffic accident investigation report information, including: Using the water traffic accident cause ontology as a knowledge framework, classifying the water traffic accident investigation report information to determine accident characteristics and underlying concepts of accident cause factors; According to the classification of the secondary concepts, the underlying concepts of the accident characteristics and accident causative factors are input into the input layer, and the underlying concepts of the accident characteristics and accident causative factors are converted through the input layer to generate graph data, wherein the structure of the graph data includes node features, edge information, and edge weights; Inputting the graph data into a hidden layer, performing a graph convolution operation on the graph data through a graph convolution network unit of the hidden layer to extract structural information of the graph, and performing a discard operation on neurons through a random inactivation unit of the hidden layer to obtain a feature vector, wherein the graph convolution network unit includes a ReLU activation function; The feature vector is input into the output layer, and the feature vector is mapped into a specific category space through the output layer to generate a predicted value.

[0009] In a possible implementation, the edge weight is an average value of the mutual information between each accident causative factor and the accident feature.

[0010] In a possible implementation, the calculation formula of the graph convolutional network unit is: , , in, is the feature vector matrix of all nodes in each layer, For the The feature vector matrix of all nodes in the layer, is the index of the layer in the graph neural network, is a nonlinear activation function, is the adjacency matrix of the graph, is the enhanced adjacency matrix, is the degree matrix of the enhanced adjacency matrix, is the weight matrix, is the identity matrix; The calculation formula of the random deactivation unit is: , , , , in, is the random mask of the random deactivation unit, is the index of the random variable, is a Bernoulli distribution, is the retention probability of the randomly deactivated unit, For the The node characteristics of the layer, For the Layer The bias term of the node, For the Layer The linear transformation result of nodes is: is the activation function, For the Layer The final output result of each node is is a vector of independent Bernoulli random variables, For the The output of the layer, is the refined output; The calculation formula of the mutual information is: , in, is a random variable With random variables The mutual information between is the set of possible values ​​of the first random variable, is the set of possible values ​​of the second random variable, for and The joint probability distribution function of for The marginal probability distribution function of for The marginal probability distribution function of For a random variable The integral variable, For a random variable The integral variable of .

[0011] In a possible implementation, the water traffic accident cause ontology is used as a knowledge framework, and the constructed graph neural network is trained based on the water traffic accident investigation report information, further comprising: The performance of the graph neural network is evaluated using the average ranking, top three hit rate, and top ten hit rate scores as indicators.

[0012] In a possible implementation, the well-trained graph neural network is used to predict the water traffic accident data to be predicted, and the prediction result of the cause of the water traffic accident is obtained, including: Determine accident features based on the water traffic accident data to be predicted, input the accident features into the well-trained graph neural network, and obtain a prediction score of the cause of the accident; The predicted scores of the accident causes are sorted to obtain accident causes with preset rankings.

[0013] On the other hand, the present invention also provides a device for predicting causes of water traffic accidents, comprising: A cause ontology construction module is used to construct a water traffic accident cause ontology based on the acquired water traffic accident investigation report information; A training module, used to train a constructed graph neural network based on the water traffic accident cause ontology as a knowledge framework and the water traffic accident investigation report information, wherein the graph neural network includes an input layer, a hidden layer and an output layer, and the hidden layer includes a plurality of graph convolutional network units and a random inactivation unit; The prediction module is used to predict the water traffic accident data to be predicted based on the fully trained graph neural network to obtain the prediction results of the causes of water traffic accidents.

[0014] In another aspect, the present invention further provides an electronic device, comprising: a processor and a memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the steps in the method for predicting the cause of water traffic accidents as described above are implemented.

[0015] On the other hand, the present invention also provides a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the method for predicting the causes of water traffic accidents as described above.

[0016] The beneficial effects of the present invention are as follows: a water traffic accident cause ontology is constructed based on the acquired water traffic accident investigation report information, and a knowledge framework of entities and relationships is obtained by constructing the water traffic accident cause ontology, which clearly displays the classification and hierarchical relationship of accident characteristics and accident causes, can reflect the causes and laws of different types of water traffic accidents, solves the problem of limitations in the analysis of water traffic accident causes, establishes a logical framework for graph neural networks to perform causal inference on accidents, trains the constructed graph neural network based on the water traffic accident investigation report information, predicts the water traffic accident data to be predicted based on the trained graph neural network, obtains the water traffic accident cause prediction results, and can predict the causes of water ship traffic accidents with high precision and efficiency, saving time and resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flow chart of an embodiment of a method for predicting causes of water traffic accidents provided by the present invention; Figure 2 A flow chart of constructing a water traffic accident cause ontology for the water traffic accident cause prediction method provided by the present invention; Figure 3 A schematic diagram of the structure of the water traffic accident cause body of the water traffic accident cause prediction method provided by the present invention; Figure 4 A schematic diagram of a visualization result of a water traffic accident cause ontology of the water traffic accident cause prediction method provided by the present invention; Figure 5 A schematic diagram of accident cause prediction results of the method for predicting causes of water traffic accidents provided by the present invention; Figure 6 A schematic diagram of the structure of an embodiment of a device for predicting causes of water traffic accidents provided by the present invention; Figure 7 A schematic structural diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0018] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0019] The present invention discloses a method, device, equipment and medium for predicting the cause of water traffic accidents, which can be used in a computer. The method, equipment or computer-readable storage medium involved in the present invention can be integrated with the above-mentioned equipment or can be relatively independent.

[0020] A specific embodiment of the present invention discloses a method for predicting the cause of water traffic accidents, which can be executed by a computer, specifically by one or more processors of the computer. Figure 1 As shown in Figure 2, the prediction methods for the causes of water traffic accidents include: S101, constructing a water traffic accident cause ontology based on the acquired water traffic accident investigation report information; S102, taking the water traffic accident cause ontology as the knowledge framework, and training the constructed graph neural network based on the water traffic accident investigation report information, wherein the graph neural network includes an input layer, a hidden layer, and an output layer, and the hidden layer includes multiple graph convolutional network units and random inactivation units; S103. Predict the water traffic accident data to be predicted based on the well-trained graph neural network to obtain the prediction results of the causes of water traffic accidents.

[0021] Among them, the construction of the ontology of the causes of water traffic accidents provides a standardized language and structure, clarifies the definitions of entities and relationships, strengthens the understanding of data semantics, and improves data processing and analysis capabilities. The ontology of the causes of water traffic accidents is established through the combination of entities and relationships, forming a triple framework (knowledge framework) consisting of head entities, relationships, and tail entities. The relationship between accident causes and accident characteristics is analyzed through graph neural networks (improved graph neural networks), providing valuable reference for accident investigations. The graph neural network includes an input layer, a hidden layer, and an output layer. The hidden layer includes multiple graph convolutional network (GCN) units and random dropout (Dropout) units.

[0022] Compared with the prior art, the method for predicting the cause of water traffic accidents provided in this embodiment constructs a water traffic accident cause ontology based on the acquired water traffic accident investigation report information; by constructing the water traffic accident cause ontology, a knowledge framework of entities and relationships is obtained, which clearly displays the classification and hierarchical relationship of accident characteristics and accident causes, can reflect the causes and laws of different types of water traffic accidents, and establishes a logical framework for graph neural networks to perform causal inference on accidents. The water traffic accident cause ontology is used as the knowledge framework, and the constructed graph neural network is trained based on the water traffic accident investigation report information, wherein the graph neural network includes an input layer, a hidden layer and an output layer, and the hidden layer includes multiple graph convolutional network units and random inactivation units; based on the well-trained graph neural network, the water traffic accident data to be predicted is predicted to obtain the water traffic accident cause prediction results, which can predict the causes of water ship traffic accidents with high accuracy and efficiency, saving time and resources.

[0023] In some embodiments, in step S101, a water traffic accident cause ontology is constructed based on the acquired water traffic accident investigation report information. The prediction of the cause of water traffic accidents requires the construction of a knowledge framework containing entities and relationships based on data analysis. Following the steps of ontology construction, entities and relationships are constructed based on the information in the water traffic accident investigation report of professional domain knowledge. OWL is used as the ontology description language, and the protégé tool is used to generate the ontology. The flowchart of constructing the water traffic accident cause ontology is shown in FIG. Figure 2 ,like Figure 2 As shown in the figure, the specific steps of constructing the ontology of causes of water traffic accidents include: S201, obtaining water traffic accident investigation report information, and determining the water traffic accident cause field and the ontology of the water traffic accident cause field based on the water traffic accident investigation report information and expert experience; S202, determining the key terms of the ontology according to expert experience and rule requirements, and obtaining a term list based on the key terms; S203, analyzing and defining the term list to determine the classes of the ontology and the hierarchical structure of the classes, wherein the hierarchical structure includes primary concepts, secondary concepts, and tertiary concepts; S204. Based on the classes and class hierarchical structure of the ontology, the relationship attributes and data attributes of the classes are defined to construct a water traffic accident cause ontology, wherein the water traffic accident cause ontology includes a head entity, a relationship, and a tail entity.

[0024] In some embodiments, in step S201, water traffic accident investigation report information is obtained, and the field of water traffic accident causes and the ontology of the field of water traffic accident causes are determined based on the water traffic accident investigation report information and expert experience, that is, the scope and field of the ontology are determined, and the professional field of the entity is clearly constructed to define the concepts in the field of water traffic accidents.

[0025] In some embodiments, in step S202, the key terms of the ontology are determined based on expert experience and rule requirements, a term list is obtained based on the key terms, the key terms in the water traffic accident causes ontology are listed, and a list of terms including the water traffic accident causes ontology is listed.

[0026] In some embodiments, in step S203, the term list is analyzed and defined to determine the classes of the ontology and the hierarchy of the classes, wherein the hierarchy includes first-level concepts, second-level concepts, and third-level concepts, and the ontology of the causes of water traffic accidents is analyzed and defined starting from the largest class to construct a complete hierarchical system of the ontology of the causes of water traffic accidents, that is, the hierarchy of entities is defined, and a total of five first-level concepts, six second-level concepts, and fifty-seven third-level concepts are constructed.

[0027] In some embodiments, in step S204, the relationship attributes and data attributes of the class are defined, the data attributes and object attributes of the accident characteristics and causes are defined, and the attributes are constructed from different facets to build a water traffic accident cause ontology, where facets refer to the multi-dimensional attributes of things, that is, the data types of data attributes.

[0028] By determining the overall planning of the domain ontology, the ontology is constructed, and some concepts in the field of ship collision are extracted and processed. First, the professional field of the entity is clearly constructed to define the concepts in the field of water traffic accidents. Secondly, the important concepts in the entity are extracted and the hierarchical structure of the entity is defined. A total of five first-level concepts, six second-level concepts and fifty-seven third-level concepts are constructed. After that, the relationship between entities is defined to complete the mutual connection between them and complete the construction of the ontology of the causes of water traffic accidents. The ontology of the causes of water traffic accidents is a triple framework composed of head entity, relationship and tail entity. The triple framework comprehensively processes the unstructured text in the field of water traffic accidents and meets the requirements for predicting the causes of water traffic accidents. Finally, the visualization of the ontology of the causes of water traffic accidents is realized. For the structural diagram of the ontology of the causes of water traffic accidents, please refer to Figure 3 ,like Figure 3As shown in the figure, the five first-level concepts of the cause ontology are: accident waters, accident type, accident level, accident time and accident cause, and the six second-level concepts are: human factors, ship factors, management factors, environmental factors, time period and accident quarter. The fifty-seven third-level concepts at least include general accidents, major accidents and extremely serious accidents corresponding to the accident levels, and port waters, fishing areas, waterways, offshore waters, other waters, etc. corresponding to the accident waters. The five first-level concepts, six second-level concepts and fifty-seven third-level concepts are all classes of the water traffic accident cause ontology; please refer to the schematic diagram of the visualization result of the water traffic accident cause ontology. Figure 4 ,like Figure 4 As shown in the figure, the visualization results show the relationship between five first-level concepts, six second-level concepts and some fifty-seven third-level concepts. The ontology visualization clearly shows the classification and hierarchical relationship of accident characteristics and accident causes. By realizing ontology visualization, a logical framework is established for the subsequent use of graph neural networks for causal inference of accidents.

[0029] In some embodiments, in step S102, the water traffic accident cause ontology is used as the knowledge framework, and the constructed graph neural network is trained based on the water traffic accident investigation report information, wherein the graph neural network includes an input layer, a hidden layer and an output layer, and the hidden layer includes multiple graph convolutional network units and random inactivation units, that is, it includes three layers of continuous GCN units and Dropout units; in marine ship traffic accidents, human factors account for a large part of the causes of the accidents, and the accident investigation process is also affected by the subjective judgment of investigators. In order to achieve more objective and accurate cause prediction, the graph neural network adopts a multi-layer graph convolutional network (Graph Convolutional Network, GCN) architecture to enhance feature extraction capabilities, smooth local noise and human factors, add activation functions and Dropout layers to prevent overfitting and improve the generalization ability of the model, introduce mutual information to adjust the edge weights of the graph convolution, thereby strengthening the influence of important features and reducing the influence of human factors on the model results. In probability theory and information theory, the mutual information of two random variables is a measure of the mutual dependence between variables. Mutual information is not limited to real-valued random variables. It is more general and determines the joint distribution. The product of the decomposed marginal distribution degree of similarity; During the training process of the graph neural network, firstly, the water traffic accident cause ontology is used as the knowledge framework to classify the water traffic accident investigation report information to determine the accident characteristics and the underlying concepts of the accident causative factors, that is, the accident characteristics and the underlying concepts of the accident causes are extracted from the investigation report, and the underlying concepts correspond to the fifty-seven third-level concepts of the cause ontology; according to the classification of the second-level concepts, the underlying concepts of the accident characteristics and the accident causes are input to the input layer, that is, each time the input is made, only one third-level concept corresponding to each second-level concept is input, and the underlying concepts of the accident characteristics and the accident causes are converted through the input layer to generate a graph data structure, wherein the graph data structure includes node features, edge features, Information and edge weights; the input layer converts the original data features into initial representations suitable for model processing, which are then used for learning and reasoning in the next layers. The input layer combines mutual information to calculate the correlation between features and causal factors, and affects the learning and reasoning process of the graph neural network by assigning edge weights. The edge weight is the average value of the mutual information between each accident causal factor and the accident feature, that is, the average mutual information. In the process of calculating the edge weight, the average mutual information between each accident causal factor and the accident feature is used. This value is used as the edge weight, which reflects the strength of the association between the accident causal factors, thereby improving the reliability of the results; the calculation formula for the mutual information is: , in, is a random variable With random variables The mutual information between and is a continuous random variable, is the set of possible values ​​of the first random variable, is the set of possible values ​​of the second random variable, for and The joint probability distribution function of for The marginal probability distribution function of for The marginal probability distribution function of For a random variable The integral variable, For a random variable The integral variable of ; Two discrete random variables and The calculation formula of mutual information between is: , in, is the set of possible values ​​of the first random variable, is the set of possible values ​​of the second random variable; Secondly, the graph data is input into the hidden layer, and the graph convolution operation is performed on the graph data through the GCN unit of the hidden layer to extract the structural information of the graph, and the neurons are discarded through the Dropout unit of the hidden layer to obtain the feature vector. Among them, the GCN unit includes the ReLU activation function; GCN is a graph-based neural network that analyzes and predicts by extracting the attributes of nodes and edges. Each layer of GCN includes a linear transformation, followed by a ReLU activation function to capture the complex relationship and feature representation between nodes. The calculation formula of its GCN unit is: , , in, is the feature vector matrix of all nodes in each layer, For the The feature vector matrix of all nodes in the layer, It is the index of the layer in the graph neural network. In the graph neural network, there are usually multiple layers, and each layer performs a graph convolution operation to capture the structural information of the graph by propagating and updating node features. is a nonlinear activation function, is the adjacency matrix of the graph, which represents the connection relationship between the nodes in the graph. is an enhanced adjacency matrix, which means that self-loops are added to the adjacency matrix. is the degree matrix of the enhanced adjacency matrix, is the weight matrix, which is used to learn the parameter matrix of node feature transformation. is the identity matrix, with diagonal elements equal to 1 and all off-diagonal elements equal to 0; The Dropout unit reduces overfitting and improves the generalization ability of the model by randomly discarding some neurons in each training iteration. The Dropout unit is applied to the node features of the hidden layer. In the graph convolution operation of each layer, Dropout randomly discards a certain proportion of neurons to reduce the model's dependence on certain specific node features, thereby improving the generalization ability. The graph neural network contains L hidden layers, index represents each hidden layer, Indicates The input vector of the hidden layer is Indicates The output vector of the hidden layer is and Respectively represent The weights and biases of the hidden layers are added, and the Dropout unit is added to the hidden layer. The calculation formula of the Dropout unit is: , , , , in, is the random mask of Dropout, is the index of the random variable, is a Bernoulli distribution, is the retention probability of Dropout, For the The node characteristics of the layer, For the Layer The bias term of the node, For the Layer The linear transformation result of nodes is: is the activation function, For the Layer The final output result of the nodes, for any layer, is a vector of independent Bernoulli random variables, and the probability of each variable being 1 is , For the The output of the layer, The refined output is used as the input of the subsequent layers, and this process is applied to each layer. During training, the derivative of the loss function is back-propagated through the sub-network, and during testing, the weights are scaled to maintain the performance of the network; Finally, the feature vector is input into the output layer, which maps the feature vector to a specific category space and generates a prediction value. The output layer consists of a graph convolutional network (GCN), whose input feature dimension corresponds to the output dimension of the hidden layer, which is 3, and the output feature dimension corresponds to the length of the causal pool. The output layer maps the features extracted from the hidden layer to a specific category space and generates a prediction value for each category, thereby completing the causal prediction.

[0030] When training the graph neural network, the hyperparameters of the graph neural network are set. For the hyperparameters set, please refer to Table 1. Table 1

[0031] The learning rate in the hyperparameter controls the speed at which the parameters of the graph neural network model are updated in each iteration. A higher value will result in greater changes in the parameters during the update process. The number of iterations determines the total number of training cycles. More iterations will result in more training rounds for the model. The embedding dimension refers to the dimension of the entity and relationship vectors. A higher dimension enhances the model's ability to represent nodes and relationships. The number of neighbors considered when processing graph data refers to the number of adjacent entities considered. A larger value improves the model's ability to capture the graph structure. Set the parameters of the graph neural network. For the parameters set, please refer to Table 2. Table 2

[0032] In order to ensure that the model outputs the most representative features, based on empirical selection and cross-validation, the initial dimension of the hidden layer is set to 128, which helps the model fully learn and represent the complex relationships in the input data. The dimension is then gradually reduced to 64 and further to 32. This reduction helps eliminate redundant features and avoid overfitting. In addition, the ReLU activation function is added to the hidden layer for nonlinear transformation, thereby increasing the complexity of the model fitting function. In order to prevent overfitting caused by the superposition of multiple GCN units, the Dropout unit is added to the graph neural network. The Dropout unit randomly discards a part of neurons during the training phase to improve the generalization ability of the model. In addition, the Dropout unit forces the graph neural network to learn from redundant feature representations, enabling it to process new data more effectively. In the output layer, the dimension is set to the length of the causal pool to ensure that the model can output the predicted scores of all accident causal factors.

[0033] The performance of graph neural network is evaluated by using average ranking, top three hit rate and top ten hit rate scores as indicators. )、Top 3 hit rate( ) and top 10 hit rate ( ) score as an indicator to evaluate the performance of the graph neural network. The average ranking is to calculate the ranking of the actual target in the test sample in the predicted results, and then calculate the average ranking of all test samples to evaluate the overall ranking performance of the graph neural network. The lower the average ranking value, the more accurate the graph neural network is in predicting the position of positive samples. Indicators such as the top 3 hit rate and the top 10 hit rate are used to evaluate the actual effectiveness of the model. The top 3 hit rate measures the frequency of the correct result appearing in the top 3 recommendations, and the top 10 hit rate measures the frequency of the correct result appearing in the top 10 recommendations. The higher the value of these indicators, the higher the effectiveness of the graph neural network. The calculation formula of the indicators is: , , , in, is the average ranking, is the top three hit rate, which means the proportion of the actual target appearing in the top three of the prediction results. The hit rate ratio of all test samples is calculated. is the top ten hit rate, which calculates the proportion of the actual target appearing in the top 10 of the predicted results. For the The ranking of the actual target of the samples in the prediction results, It is an indicator function, which takes the value of 1 when the condition is met, otherwise it takes the value of 0.

[0034] In some embodiments, in step S103, the water traffic accident data to be predicted is predicted based on the well-trained graph neural network to obtain the prediction results of the causes of the water traffic accidents, the accident characteristics are determined based on the water traffic accident data to be predicted, and the accident characteristics are input into the well-trained graph neural network to obtain the prediction scores of the causes of the accidents; the prediction scores of the causes of the accidents are sorted to obtain the causes of the accidents with preset rankings, that is, the top six causes of the accidents are obtained; in order to facilitate the data evaluation of the graph neural network, the text information in the accident data is converted into numerical representation, and the text information is converted into numerical representation, please refer to Table 3, Table 3

[0035] The accident features are input into the well-trained graph neural network, which will output the top six accident causes corresponding to each set of features. For a schematic diagram of the accident cause prediction results, please refer to Figure 5 ,like Figure 5 As shown, the left nodes represent the IDs corresponding to the characteristics of water traffic accidents. By entering these IDs according to the feature type, the causes of the accidents can be obtained and sorted from high to low by score. The right nodes represent the causes of the accidents, and the numbers 1 to 6 on the arrows represent the top six causes of the accidents.

[0036] In some embodiments, the graph neural network is compared and evaluated, and the evaluation index values ​​of each model in the prediction of the cause of water vessel traffic accidents are listed in the table. Due to the influence of professional terminology in water traffic and accident investigation, traditional graph neural networks often cannot achieve the best results. This problem is particularly prominent when processing professional texts. , and The proposed improved graph neural network performs best in these three indicators and can better complete the task of predicting the causes of marine ship traffic accidents. The scores of each model indicator are shown in Table 4. Table 4

[0037] Common graph neural network models perform poorly in processing professional terms because they have difficulty capturing the relationships in the field of causes of water traffic accidents. The score is 4.73% higher than that of Our Model (improved graph neural network), indicating that the GCN model more frequently recommends factors that are irrelevant to the actual cause, which makes the maritime department spend more time to exclude irrelevant factors during the investigation process. In addition, compared with Our Model, the GCN model and The scores decreased by 6.92% and 3.09% respectively, which means that in practical applications, the GCN model is less accurate in recommending the top three or top ten causes, and additional time is needed to determine the real cause of the accident.

[0038] Improved graph neural network , and It performs best in all aspects and can predict the causes of water traffic accidents with high accuracy and efficiency. It is better at handling professional terms and the recommended causes are more relevant and accurate. With the improved performance of the model, the cause of the accident can be determined more quickly, thus saving time and resources, which is crucial to preventing similar accidents and ensuring the safety of water transportation.

[0039] In order to better implement the method for predicting the cause of a water traffic accident in the embodiment of the present invention, based on the method for predicting the cause of a water traffic accident, correspondingly, Figure 6 As shown, the embodiment of the present invention further provides a device for predicting the cause of a water traffic accident. The device 600 for predicting the cause of a water traffic accident includes: The cause ontology construction module 601 is used to construct the cause ontology of water traffic accidents based on the acquired water traffic accident investigation report information; A training module 602 is used to train the constructed graph neural network based on the water traffic accident cause ontology as a knowledge framework and the water traffic accident investigation report information, wherein the graph neural network includes an input layer, a hidden layer and an output layer, and the hidden layer includes a plurality of graph convolutional network units and a random inactivation unit; The prediction module 603 is used to predict the water traffic accident data to be predicted based on the well-trained graph neural network to obtain the prediction results of the causes of the water traffic accidents.

[0040] like Figure 7 As shown, the present invention also provides an electronic device 700 , which can be a computing device such as a mobile terminal, a desktop computer, a notebook, a palmtop computer, a server, etc. The electronic device 700 includes a processor 701 , a memory 702 , and a display 703 . Figure 7Only some components of the electronic device 700 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0041] In some embodiments, the memory 702 may be an internal storage unit of the electronic device 700, such as a hard disk or memory of the electronic device 700. In other embodiments, the memory 702 may also be an external storage device of the electronic device 700, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 700. Further, the memory 702 may also include both an internal storage unit of the electronic device 700 and an external storage device. The memory 702 is used to store application software and various types of data installed in the electronic device 700, such as program codes installed in the electronic device 700. The memory 702 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a water traffic accident cause prediction program is stored on the memory 702, and the water traffic accident cause prediction program can be executed by the processor 701, thereby realizing the water traffic accident cause prediction method of each embodiment of the present invention.

[0042] In some embodiments, the processor 701 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 702, such as a method for predicting the cause of water traffic accidents.

[0043] In some embodiments, the display 703 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 703 is used to display the identification information of the water traffic accident cause prediction program and to display a visual user interface. The components 701-703 of the electronic device 700 communicate with each other through a system bus.

[0044] In some embodiments, when the processor 701 executes the water traffic accident cause prediction program in the memory 702, the various steps in the water traffic accident cause prediction method described in the above embodiments are implemented. Since the water traffic accident cause prediction method has been described in detail above, it will not be repeated here.

[0045] Accordingly, the present invention also provides a computer-readable storage medium, which is used to store computer-readable programs or instructions. When the program or instructions are executed by a processor, it can implement the steps or functions of the water traffic accident cause prediction method provided by the above-mentioned method embodiments.

[0046] In summary, the method, device, equipment and medium for predicting the causes of water traffic accidents provided by the present invention construct a water traffic accident cause ontology based on the acquired water traffic accident investigation report information; use the water traffic accident cause ontology as the knowledge framework, and train the constructed graph neural network based on the water traffic accident investigation report information, wherein the graph neural network includes an input layer, a hidden layer and an output layer, and the hidden layer includes multiple graph convolutional network units and random inactivation units; predict the water traffic accident data to be predicted based on the well-trained graph neural network, and obtain the water traffic accident cause prediction results, thereby improving the accuracy and efficiency of water traffic accident cause analysis and prediction.

[0047] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0048] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting the cause of a water traffic accident, characterized in that: include: Construct the water traffic accident cause ontology based on the acquired water traffic accident investigation report information; Taking the water traffic accident cause ontology as the knowledge framework, the constructed graph neural network is trained based on the water traffic accident investigation report information, wherein the graph neural network includes an input layer, a hidden layer and an output layer, and the hidden layer includes a plurality of graph convolutional network units and a random inactivation unit; Based on the well-trained graph neural network, the water traffic accident data to be predicted is predicted to obtain the prediction results of the causes of water traffic accidents.

2. The method for predicting causes of water traffic accidents according to claim 1, characterized in that: The water traffic accident cause ontology is constructed based on the acquired water traffic accident investigation report information, including: Acquire water traffic accident investigation report information, and determine the water traffic accident cause field and the ontology of the water traffic accident cause field based on the water traffic accident investigation report information and expert experience; Determine the key terms of the ontology according to expert experience and rule requirements, and obtain a term list based on the key terms; Analyzing and defining the term list to determine the classes of the ontology and the class hierarchy, wherein the hierarchy includes primary concepts, secondary concepts, and tertiary concepts; Based on the classes and class hierarchical structure of the ontology, the relationship attributes and data attributes of the classes are defined to construct a water traffic accident cause ontology, wherein the water traffic accident cause ontology includes a head entity, a relationship, and a tail entity.

3. The method for predicting causes of water traffic accidents according to claim 2, characterized in that: The method of taking the water traffic accident cause ontology as the knowledge framework and training the constructed graph neural network based on the water traffic accident investigation report information includes: Using the water traffic accident cause ontology as a knowledge framework, classifying the water traffic accident investigation report information to determine accident characteristics and underlying concepts of accident cause factors; According to the classification of the secondary concepts, the underlying concepts of the accident characteristics and accident causative factors are input into the input layer, and the underlying concepts of the accident characteristics and accident causative factors are converted through the input layer to generate graph data, wherein the structure of the graph data includes node features, edge information, and edge weights; Inputting the graph data into a hidden layer, performing a graph convolution operation on the graph data through a graph convolution network unit of the hidden layer to extract structural information of the graph, and performing a discard operation on neurons through a random inactivation unit of the hidden layer to obtain a feature vector, wherein the graph convolution network unit includes a ReLU activation function; The feature vector is input into the output layer, and the feature vector is mapped into a specific category space through the output layer to generate a predicted value.

4. The method for predicting causes of water traffic accidents according to claim 3, characterized in that: The edge weight is the average value of the mutual information between each accident causative factor and the accident feature.

5. The method for predicting causes of water traffic accidents according to claim 4, characterized in that: The calculation formula of the graph convolutional network unit is: ， ， in, is the feature vector matrix of all nodes in each layer, For the The feature vector matrix of all nodes in the layer, is the index of the layer in the graph neural network, is a nonlinear activation function, is the adjacency matrix of the graph, is the enhanced adjacency matrix, is the degree matrix of the enhanced adjacency matrix, is the weight matrix, is the identity matrix; The calculation formula of the random deactivation unit is: ， ， ， ， in, is the random mask of the random deactivation unit, is the index of the random variable, is a Bernoulli distribution, is the retention probability of the randomly deactivated unit, For the The node characteristics of the layer, For the Layer The bias term of the node, For the Layer The linear transformation result of nodes is: is the activation function, For the Layer The final output result of each node is is a vector of independent Bernoulli random variables, For the The output of the layer, is the refined output; The calculation formula of the mutual information is: ， in, is a random variable With random variables The mutual information between is the set of possible values ​​of the first random variable, is the set of possible values ​​of the second random variable, for and The joint probability distribution function of for The marginal probability distribution function of for The marginal probability distribution function of For a random variable The integral variable, For a random variable The integral variable of .

6. The method for predicting causes of water traffic accidents according to claim 3, characterized in that: The method of taking the water traffic accident cause ontology as the knowledge framework and training the constructed graph neural network based on the water traffic accident investigation report information also includes: The performance of the graph neural network is evaluated using the average ranking, top three hit rate, and top ten hit rate scores as indicators.

7. The method for predicting causes of water traffic accidents according to claim 3, characterized in that: The water traffic accident data to be predicted are predicted based on the well-trained graph neural network to obtain the prediction results of the causes of water traffic accidents, including: Determine accident features based on the water traffic accident data to be predicted, input the accident features into the well-trained graph neural network, and obtain a prediction score of the cause of the accident; The predicted scores of the accident causes are sorted to obtain accident causes with preset rankings.

8. A device for predicting causes of water traffic accidents, characterized in that: include: A cause ontology construction module is used to construct a water traffic accident cause ontology based on the acquired water traffic accident investigation report information; A training module, used to train a constructed graph neural network based on the water traffic accident cause ontology as a knowledge framework and the water traffic accident investigation report information, wherein the graph neural network includes an input layer, a hidden layer and an output layer, and the hidden layer includes a plurality of graph convolutional network units and a random inactivation unit; The prediction module is used to predict the water traffic accident data to be predicted based on the fully trained graph neural network to obtain the prediction results of the causes of water traffic accidents.

9. An electronic device, characterized in that: including memory and processor; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the steps in the method for predicting the cause of water traffic accidents as described in any one of claims 1-7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the method for predicting the cause of water traffic accidents as described in any one of claims 1-7.

Images