Channel passage feasibility prediction method and related device
By combining Bayesian networks and graph neural networks, multi-source monitoring data was acquired and features were extracted, solving the problem of accuracy in predicting the feasibility of waterway passage in the reservoir area and realizing the dynamic feature acquisition and accurate prediction of waterway passage risks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN UNIV OF TECH
- Filing Date
- 2026-03-31
- Publication Date
- 2026-07-24
Smart Images
Figure CN122452833A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent transportation technology, and in particular to a method and related apparatus for predicting the feasibility of waterway passage. Background Technology
[0002] The navigation conditions of vessels in the reservoir area are an important factor affecting the safety of navigation in the reservoir area.
[0003] The navigation risks in reservoir waterways exhibit significant time-varying and nonlinear characteristics. On the one hand, influenced by reservoir scheduling and changes in water inflow, hydrological factors such as flow rate, water level, and flow velocity may fluctuate rapidly over short timescales. On the other hand, the operating rhythm of locks, the density of upstream and downstream vessels, and vessel maneuvering behavior show obvious time dependence and feedback effects. These factors are coupled with each other, making navigation risks not a static superposition result, but a dynamic process that continuously evolves along the time axis and may accumulate and amplify. In this context, traditional Bayesian networks, due to their fixed graph structure and static conditional probability tables, struggle to describe the propagation path of risks across multiple time steps and cannot support continuous prediction of future navigation conditions.
[0004] This demonstrates that existing technologies cannot accurately predict the feasibility of navigation in complex reservoir areas. Summary of the Invention
[0005] In view of this, it is necessary to provide a method and related apparatus for predicting the feasibility of navigation in waterways, so as to solve the problem that existing technologies cannot accurately predict the feasibility of navigation in complex reservoir areas.
[0006] To address the aforementioned problems, in a first aspect, the present invention provides a method for predicting the feasibility of waterway passage, comprising: Acquire multi-source monitoring data of the target channel for the current target time period, and construct a spatiotemporal map sequence of the target channel based on the multi-source monitoring data and a Bayesian network; A pre-defined graph neural network is used to extract spatial dependency features from the spatiotemporal graph sequence to determine the spatial feature embedding matrix of the target channel; The spatial feature embedding matrix is used as the input to a pre-defined waterway traffic feasibility prediction model to predict the traffic feasibility of the target waterway and obtain the prediction results.
[0007] In one possible implementation, multi-source monitoring data of the target channel for the current target time period is acquired, including: Acquire AIS navigation data, hydrological monitoring data, navigation management and scheduling data, and meteorological monitoring data for the target waterway during the current target time period; Time synchronization, outlier removal, missing value interpolation, and feature standardization were performed on AIS navigation data, hydrological monitoring data, navigation management and scheduling data, and meteorological monitoring data to obtain a multi-source monitoring data matrix of continuous time series of the target waterway.
[0008] In one possible implementation, a spatiotemporal map sequence of the target route is constructed based on multi-source monitoring data combined with a Bayesian network, including: A static adjacency matrix of multi-source monitoring data at the same time is constructed based on a predefined node system of Bayesian network and a multi-source monitoring data matrix. A time-extended adjacency matrix is constructed based on the temporal relationship of the static adjacency matrix of multi-source monitoring data at different times and the state propagation relationship of nodes at adjacent times; By performing block diagonalization and splicing on the temporally extended adjacency matrix, a spatiotemporal graph sequence of the target channel is obtained.
[0009] In one possible implementation, a pre-defined graph neural network is used to extract spatial dependency features from the spatiotemporal graph sequence to determine the spatial feature embedding matrix of the target channel, including: The temporally extended adjacency matrix in the spatiotemporal graph sequence is subjected to self-connection addition and symmetric normalization to obtain the normalized adjacency matrix; Feature propagation is performed using a multi-layer stacked graph convolutional network to obtain the spatial feature embedding matrix of the target channel.
[0010] In one possible implementation, the spatial feature embedding matrix is used as input to a pre-defined waterway navigability prediction model to predict the navigability of the target waterway, yielding prediction results including: The spatial feature embedding matrix is reorganized into a node center time series, and a sinusoidal position code is added to obtain a time embedding sequence; The dependency strength at different times in the temporal embedding sequence is calculated using a multi-head self-attention mechanism; The dependency strength is nonlinearly mapped through a feedforward network, and after layer normalization and residual connection, the temporal features are obtained. By using a prediction head to predict temporal features, the future feasibility prediction results of the target waterway are obtained.
[0011] In one possible implementation, the pre-defined waterway passage feasibility prediction model, during training, includes: Using the Bayesian network inference results as the teacher signal, a hybrid loss function including prediction loss, imitation loss and regularization loss is constructed to jointly optimize the waterway passage feasibility prediction model.
[0012] In one possible implementation, the prediction loss is:
[0013] in, The number of samples in one training iteration; This represents the actual risk status of the i-th sample; To predict the probability that the i-th sample is at risk for the model; The imitation loss is:
[0014] in, For Bayesian networks to the first i The inference probability of each sample category; For the BGS-T model, the first i The inference probability of each sample category; Indexed by category; The regularization loss is:
[0015] in, The L2 regularization coefficient; These are the learnable parameters of the model; The square of the L2 norm; For structural constraint coefficients; The learned adjacency matrix; The prior adjacency matrix; It is the square of the Frobenius norm.
[0016] Secondly, the present invention also provides a waterway navigability feasibility prediction device, comprising: The data acquisition module is used to acquire multi-source monitoring data of the target channel during the current target time period, and to construct a spatiotemporal map sequence of the target channel based on the multi-source monitoring data and a Bayesian network. The feature extraction module is used to extract spatial dependency features from the spatiotemporal graph sequence using a preset graph neural network to determine the spatial feature embedding matrix of the target channel; The prediction module is used to use the spatial feature embedding matrix as input to a preset waterway traffic feasibility prediction model to predict the traffic feasibility of the target waterway and obtain the prediction results.
[0017] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, Memory, used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the waterway navigability feasibility prediction method of any of the above implementations.
[0018] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps in the waterway passage feasibility prediction method described above.
[0019] The beneficial effects of this invention are as follows: The waterway traffic feasibility prediction method provided by this invention acquires multi-source monitoring data of the target waterway during the current target time period, and constructs a spatiotemporal map sequence of the target waterway based on the multi-source monitoring data and a Bayesian network. Unlike the fixed graph structure and static conditional probability of traditional Bayesian networks, combining multi-source monitoring data of the target waterway for a time period with a Bayesian network can obtain a spatiotemporal sequence map based on a Bayesian network structure to characterize the dynamic spatiotemporal features of the waterway, realizing the acquisition of dynamic features based on the Bayesian network structure. A preset graph neural network is used to extract spatial dependency features from the spatiotemporal map sequence to determine the spatial feature embedding matrix of the target waterway. Spatial dependency and temporal dynamic features are captured respectively, realizing the organic integration of causal reasoning and data-driven modeling. The spatial feature embedding matrix is used as the input of a preset waterway traffic feasibility prediction model to predict the traffic feasibility of the target waterway and obtain the prediction result. Using the preset waterway traffic feasibility prediction model to predict waterway traffic feasibility can improve the prediction accuracy. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 A flowchart illustrating a method for predicting the feasibility of waterway passage provided in an embodiment of the present invention; Figure 2 A flowchart illustrating a multi-source monitoring data acquisition method provided in an embodiment of the present invention; Figure 3 A flowchart illustrating a spatiotemporal sequence graph construction method provided in an embodiment of the present invention; Figure 4 A Bayesian network node hierarchy diagram provided for an embodiment of the present invention; Figure 5 A flowchart illustrating a spatial dependency feature extraction method provided in an embodiment of the present invention; Figure 6 A flowchart illustrating an implementation method of S103 provided in an embodiment of the present invention; Figure 7A model training principle block diagram provided in an embodiment of the present invention; Figure 8 This is a principle block diagram of a waterway traffic feasibility prediction method provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of a waterway traffic feasibility prediction device provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0022] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form 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 intended to limit the scope of the present invention.
[0023] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.
[0024] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.
[0025] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0026] A specific embodiment of the present invention, such as Figure 1 As shown, a method for predicting the feasibility of waterway passage is disclosed, including: S101: Obtain multi-source monitoring data of the target channel for the current target time period, and construct a spatiotemporal map sequence of the target channel based on the multi-source monitoring data and a Bayesian network.
[0027] In this embodiment of the invention, the provided method for predicting the feasibility of waterway passage can be used to predict the feasibility of ship passage in dam reservoir waterways. The current target time period refers to a preset time period before the current moment, such as 12 hours before the current moment. The target waterway refers to the waterway that the ship needs to pass through. Multi-source monitoring data includes waterway data from multiple sources, used to characterize the risk factors of ships passing through the waterway. Specifically, the current target time period can be divided into multiple time steps. For each time step, multi-source monitoring data of the target waterway is collected once to obtain continuous multi-source monitoring data in the time series. A causal structure of navigation risk is constructed based on a Bayesian network. A node system including environmental conditions, ship status, traffic complexity, and target node modules is defined. The directed acyclic graph of the Bayesian network is mapped to the structural prior of a graph neural network to construct a spatiotemporal graph sequence of the target waterway.
[0028] S102, a preset graph neural network is used to extract spatial dependency features from the spatiotemporal graph sequence to determine the spatial feature embedding matrix of the target channel.
[0029] In this embodiment of the invention, the spatiotemporal graph sequence is input into a graph neural network spatial encoder. Using a Bayesian adjacency matrix as the structural prior, spatial dependency features are extracted through multi-layer graph convolution and residual propagation mechanisms to obtain a spatial feature embedding matrix. The specific construction method of the spatial feature embedding matrix will be described in detail later in this invention.
[0030] S103, the spatial feature embedding matrix is used as the input of the preset waterway traffic feasibility prediction model to predict the traffic feasibility of the target waterway and obtain the prediction result.
[0031] In this embodiment of the invention, the preset waterway passage feasibility prediction model is a pre-constructed deep learning model with a self-attention mechanism. Spatial features are embedded into the matrix input model's time encoder. Temporal dependency features are captured through temporal embedding, positional encoding, and multi-head self-attention mechanism to obtain a temporal feature representation. Then, a prediction head is used to predict the temporal feature representation to obtain the prediction result.
[0032] The waterway traffic feasibility prediction method provided by this invention acquires multi-source monitoring data of the target waterway during the current target time period, and constructs a spatiotemporal graph sequence of the target waterway based on the multi-source monitoring data and a Bayesian network. Unlike the fixed graph structure and static conditional probability of traditional Bayesian networks, this method combines multi-source monitoring data of the target waterway over a time period with a Bayesian network to obtain a spatiotemporal sequence graph based on a Bayesian network structure to characterize the dynamic spatiotemporal features of the waterway. This achieves dynamic feature acquisition based on the Bayesian network structure. A preset graph neural network is used to extract spatial dependency features from the spatiotemporal graph sequence to determine the spatial feature embedding matrix of the target waterway. By capturing both spatial dependency and temporal dynamic features, the method achieves an organic integration of causal reasoning and data-driven modeling. The spatial feature embedding matrix is used as the input of a preset waterway traffic feasibility prediction model to predict the traffic feasibility of the target waterway, obtaining the prediction result. The preset waterway traffic feasibility prediction model is used to predict waterway traffic feasibility, improving prediction accuracy.
[0033] In some possible embodiments of the present invention, such as Figure 2 As shown, multi-source monitoring data of the target channel for the current target time period is obtained, including: S201, acquire AIS navigation data, hydrological monitoring data, navigation management and scheduling data and meteorological monitoring data of the target waterway during the current target time period; S202 involves time synchronization, outlier removal, missing value interpolation, and feature standardization of AIS navigation data, hydrological monitoring data, navigation management and scheduling data, and meteorological monitoring data to obtain a multi-source monitoring data matrix of continuous time series of the target waterway.
[0034] In this embodiment of the invention, the multi-source monitoring data includes AIS navigation data, hydrological monitoring data, navigation management and scheduling data, and meteorological monitoring data. The multi-source monitoring data undergoes preprocessing, including time synchronization, outlier removal, missing value interpolation, and feature standardization, to obtain a continuous time series monitoring data matrix. Specifically, with a 12-hour target time period and a time step of half an hour, 24 sets of multi-source monitoring data are acquired. For these 24 sets of multi-source monitoring data, time synchronization is performed first, then missing value interpolation is performed, and finally, a 3D model is used. Outlier removal criteria were applied, and feature standardization was performed using the min-max normalization method to obtain a multi-source monitoring data matrix of the target waterway's continuous time series.
[0035] In some possible embodiments of the present invention, such as Figure 3 As shown, a spatiotemporal map sequence of the target channel is constructed based on multi-source monitoring data combined with a Bayesian network, including: S301, Based on the predefined node system of Bayesian network and multi-source monitoring data matrix, construct the static adjacency matrix of multi-source monitoring data at the same time; S302, Construct a time-extended adjacency matrix based on the temporal relationship of the static adjacency matrix of multi-source monitoring data at different times and the state propagation relationship of nodes at adjacent times; S303, block diagonalization and splicing of the temporally extended adjacency matrix to obtain the spatiotemporal graph sequence of the target channel.
[0036] In embodiments of the present invention, such as Figure 4 As shown, a Bayesian network node system is defined, including 11 observation nodes, 6 intermediate nodes, and 1 target node. The observation nodes correspond to the mapping features of multi-source monitoring data, and the target node represents the navigation feasibility status of ships. A static adjacency matrix is constructed based on the causal relationships between nodes, and the elements of the adjacency matrix represent the conditional dependencies between nodes. A time dimension is introduced to extend it into a dynamic Bayesian graph, and a time-extended adjacency matrix is constructed to describe the state propagation relationship between nodes at adjacent time points. The dynamic adjacency matrix is block-diagonalized and concatenated to obtain a spatiotemporal adjacency matrix containing spatial causal relationships and temporal propagation dependencies, generating a spatiotemporal graph sequence. Furthermore, the Bayesian network node system is divided into the following modules: Environmental Conditions Module: including nodes for flow conditions, ship power, full load displacement, maximum load, navigation status, and load-to-power ratio, representing the matching relationship between hydrological conditions and ship performance; Ship Status Module: including nodes for navigation time and hazardous materials attributes, reflecting ship operating characteristics and potential risks; Traffic Complexity Module: including nodes for the number of ships going up and down, the relative distance between up and down, and the traffic complexity of up and down, representing the intensity of waterway traffic interaction; Target Node: Ship navigation status, with values of permissible or not permissible.
[0037] Specifically, a causal structure for navigation risks is constructed based on Bayesian networks. A node system including modules for environmental conditions, ship status, and traffic complexity is defined. An adjacency matrix is constructed and extended to a spatiotemporal adjacency structure. The input nodes of the Bayesian network are defined as follows:
[0038] in, This represents the state of node i at time t. The node is divided into three main modules: (1) Environmental conditions module: includes six nodes: flow conditions, ship power, full load displacement, maximum load, navigation status and load power ratio, which are used to characterize the matching relationship between hydrological conditions and ship performance; (2) Ship Status Module: Includes two nodes: sailing time and dangerous goods attributes, reflecting the ship's operational characteristics and potential risks; (3) Traffic complexity module: includes six nodes, including the number of ships going up and down, the relative distance between the two sides and their complexity, to characterize the traffic interaction intensity of the waterway.
[0039] Furthermore, based on the causal relationships between nodes, an 18×18 static adjacency matrix is constructed.
[0040] The zero-order matrix indicates that node i has a causal dependency on node j, otherwise it is 0.
[0041] Expanding to a spatiotemporal adjacency structure, introducing a time dimension A, and constructing a time dependency matrix. :
[0042] in, This represents the time dependency matrix, used to describe the state propagation relationship between nodes at adjacent time points.
[0043] Suppose that the temporal dependencies of a node mainly exist between itself and its causal neighbors, therefore we define:
[0044] To achieve cross-temporal feature propagation, the dynamic adjacency matrix A is block-diagonalized and concatenated to obtain a complete spatiotemporal adjacency structure:
[0045] in, It describes the joint structure of spatial causality and temporal propagation dependence, providing a unified graph prior constraint for the dynamic modeling of subsequent graph neural networks and models.
[0046] The embodiments of the present invention capture spatial dependence and temporal dynamic features respectively, realizing the organic integration of causal reasoning and data-driven modeling.
[0047] In some possible embodiments of the present invention, such as Figure 5 As shown, a pre-defined graph neural network is used to extract spatial dependency features from the spatiotemporal graph sequence to determine the spatial feature embedding matrix of the target channel, including: S501, perform self-connection addition and symmetric normalization on the temporal extended adjacency matrix in the spatiotemporal graph sequence to obtain the normalized adjacency matrix; S502 uses a multi-layer stacked graph convolutional network for feature propagation to obtain the spatial feature embedding matrix of the target channel.
[0048] In this embodiment of the invention, the spatial feature embedding matrix is input into the Transformer temporal encoder, and temporal dependent features are captured through temporal embedding, positional encoding and multi-head self-attention mechanism to obtain a temporal feature representation.
[0049] Specifically, the Bayesian adjacency matrix is subjected to self-joining and symmetric normalization to obtain the normalized adjacency matrix:
[0050] in, This is the normalized adjacency matrix; This is the original adjacency matrix; It is the identity matrix; Let be a degree matrix, and its diagonal elements be . .
[0051] Feature propagation is performed using a multi-layered stacked graph convolutional network. l Layer node representation is calculated using the formula:
[0052] in, For the first The node feature matrix of the layer; For the first The node feature matrix of the layer, if Then it is the original input feature. ; This is the normalized adjacency matrix; For the first Layer-learnable weight matrix; For the first Layered bias vector; For activation functions; This is a residual term used to mitigate the degradation problem during deep network training.
[0053] After propagation through a preset number of layers, the output spatial feature embedding matrix is:
[0054] in, This is the node embedding matrix encoded by the graph neural network; For the first Layer output, The number of convolutional layers in the graph; It is the set of real numbers; The number of nodes; Hide dimensions for nodes.
[0055] In some possible embodiments of the present invention, such as Figure 6As shown, the spatial feature embedding matrix is used as the input to a pre-defined waterway navigability prediction model to predict the navigability of the target waterway, and the prediction results are obtained, including: S601, the spatial feature embedding matrix is reorganized into a node center time series, and a sinusoidal position code is added to obtain a time embedding sequence; S602 calculates the dependency strength at different times in a temporal embedding sequence through a multi-head self-attention mechanism; S603 uses a feedforward network to perform a nonlinear mapping of dependency strength, and after layer normalization and residual connection, it obtains time series features; S604 uses a prediction head to predict temporal features and obtains the prediction results of the future passability of the target waterway.
[0056] In this embodiment of the invention, the spatial feature embedding matrix is reorganized into a node center time series, and a sinusoidal positional encoding is added to obtain a time embedding sequence; the dependency strength at different time steps is calculated using a multi-head self-attention mechanism. l Multi-head attention output is calculated using the following formula:
[0057] in, The input sequence matrix; For the first attention head, For the number of heads; This is for outputting the projection matrix.
[0058]
[0059] in:
[0060] in, For the first The unique learnable linear transformation parameters for each component are the query, key, and value projection matrices. These are the query, key, and value matrices, respectively. For scaling factor dimensions; This is the normalization function.
[0061] Nonlinear mapping is achieved through a feedforward network. After layer normalization and residual connection, the output time-series feature representation is obtained. The calculation process is as follows:
[0062] in, For the first Layer input; This is an intermediate hidden state; For layer normalization function; This is a multi-head self-attention mechanism.
[0063]
[0064] in:
[0065] in, This is the final output of the first layer; It is a feedforward neural network used to enhance the nonlinear fitting ability of the model; This is the first layer weight matrix; This is the first layer bias vector; This is the weight matrix for the second layer; This is the second layer bias vector; This is the activation function.
[0066] go through After stacking the layers, the final timing-coded output is obtained:
[0067] in, This is the final output of the time encoder; For the first layer Transformer Output; The number of nodes; For the hidden layer dimension.
[0068] An auxiliary mimic head (BNSoft Head) is introduced outside the main prediction branch. This head receives the node embeddings output by the Transformer. Mapped to the probability space via a multi-layer perceptron (MLP):
[0069] in, To mimic the probability distribution of decision-making; It is a normalized exponential function; For nodes eigenvectors; Output for time encoder; To mimic the task of a multilayer perceptron, , This is the first layer weight matrix. This is the first layer bias vector. This is the weight matrix for the second layer. This is the second layer bias vector. This is the activation function.
[0070] In some possible embodiments of the present invention, the preset waterway traffic feasibility prediction model includes the following during training: Using the Bayesian network inference results as the teacher signal, a hybrid loss function including prediction loss, imitation loss and regularization loss is constructed to jointly optimize the waterway passage feasibility prediction model.
[0071] In this embodiment of the invention, the probability distribution output by Softmax and the inference result of BN are aligned in the same label space (i.e., "allowed / not allowed"), thereby achieving logical consistency constraints between the teacher model (BN) and the student model (GNN-Transformer).
[0072] Furthermore, the predicted loss is:
[0073] in, The number of samples in one training iteration; This represents the actual risk status of the i-th sample; The model predicts the probability that the i-th sample is at risk.
[0074] The imitation loss is:
[0075] in, Let be the inference probability of the Bayesian network for the i-th sample category; Let be the inference probability of the BGS-T model for the i-th sample category; For category indexes, such as (Safety), (risk).
[0076] The regularization loss is:
[0077] in, The L2 regularization coefficient; These are the learnable parameters of the model; The square of the L2 norm; For structural constraint coefficients; The learned adjacency matrix; The prior adjacency matrix; It is the square of the Frobenius norm.
[0078] In embodiments of the present invention, such as Figure 7 As shown, in order to enhance the interpretability and robustness of the model while maintaining prediction accuracy, a hybrid loss function consisting of three parts was designed:
[0079] in, Total loss; To predict losses; Losses due to imitation; This is the regularization loss; These are the weighting coefficients.
[0080] The main task loss (predicted damage) uses a binary classification cross-entropy function:
[0081] in, The number of samples in one training iteration; This represents the actual risk status of the i-th sample; The model predicts the probability that the i-th sample is at risk.
[0082] The imitation loss employs the knowledge distillation loss function, and the KL divergence is used to measure the difference between the BN inference distribution and the model imitation distribution.
[0083] This encourages the model's output probability to remain consistent with the causal judgment results of Batch Normalization (BN), ensuring that the model does not deviate from the expert knowledge system during the dynamic learning process.
[0084] Furthermore, to prevent overfitting and maintain graph structure stability, the regularized loss function introduces dual constraints on parameters and structure:
[0085] in, The L2 regularization coefficient; These are the learnable parameters of the model; The square of the L2 norm; For structural constraint coefficients; The learned adjacency matrix; The prior adjacency matrix; It is the square of the Frobenius norm.
[0086] In this embodiment of the invention, the F1 score and the area under the characteristic curve (AUC) are selected as evaluation metrics:
[0087]
[0088]
[0089] in, Precision rate (the proportion of events predicted as risk that are actually risky). Recall rate (the proportion of all actual risk events correctly identified by the model); This refers to a situation where the model predicts "risk" and the actual situation is "risk"; a situation where the model predicts "risk" and the actual situation is "safe"; a situation where the model predicts "safe" and the actual situation is "safe"; and a situation where the model predicts "safe" and the actual situation is "risk".
[0090] When a model trades off between precision and recall, the F1 score can serve as a comprehensive indicator reflecting the overall performance of the model; the AUC score measures the model's ability to distinguish between navigable and non-navigable samples, with a value range of [0,1]; the closer the AUC is to 1, the stronger the model's discriminative ability; AUC is particularly suitable for binary classification problems with class imbalance and can be used as a core indicator for evaluating the overall predictive performance of different models.
[0091] in, same ; This represents the proportion of actual safety incidents that are mistakenly reported as risks.
[0092] In the embodiments of the present invention, it was verified that Bayesian causal structure, GNN spatial coding, Transformer temporal coding and BN-SoftHead module all make important contributions and can achieve the best prediction effect through their synergistic effect.
[0093] In summary, as Figure 8 The diagram shown is a complete block diagram of the waterway navigation feasibility prediction method provided in this embodiment of the invention. It involves acquiring multi-source monitoring data of the Three Gorges Reservoir area waterway, preprocessing the multi-source monitoring data to obtain a continuous time-series monitoring data matrix, constructing a causal structure of navigation risk based on a Bayesian network, defining a node system including environmental conditions, ship status, traffic complexity, and target node modules, mapping the directed acyclic graph of the Bayesian network to a structural prior of a graph neural network, and constructing a spatiotemporal adjacency matrix. The spatiotemporal graph sequence is input into a GNN spatial encoder, using the Bayesian adjacency matrix as a structural prior, and employing multi-layer graph convolution and residual propagation mechanisms. Spatial dependency features are extracted to obtain a spatial feature embedding matrix. The spatial feature embedding matrix is input into the Transformer temporal encoder, and temporal dependency features are captured through temporal embedding, positional encoding, and multi-head self-attention mechanism to obtain temporal feature representation. During the training phase, the BN-SoftHead module is introduced, and the Bayesian network inference results are used as teacher signals to construct a hybrid loss function including prediction loss, imitation loss, and regularization loss to jointly optimize the model. After the model converges, the prediction head outputs the feasibility prediction results of ship navigation at future times, and the prediction results are either allowed or not allowed to navigate.
[0094] This invention constructs a clear causal structure through Bayesian networks, providing a foundation for model interpretability. Simultaneously, it utilizes GNNs and Transformers to capture spatial dependence and temporal dynamic features, respectively, achieving an organic integration of causal inference and data-driven modeling. The introduction of the BN-SoftHead module and a hybrid loss function ensures that the model follows causal logic while learning data features, improving prediction robustness. Addressing the unique characteristics of the Three Gorges Reservoir area waterway, data preprocessing and node design are optimized to adapt to prediction needs under different hydrological conditions, significantly improving the accuracy of navigation feasibility prediction and providing reliable decision support for waterway safety management, vessel scheduling, and resource optimization.
[0095] To better implement the waterway traffic feasibility prediction method in this embodiment of the invention, based on the waterway traffic feasibility prediction method, correspondingly, such as Figure 9 As shown, this embodiment of the invention also provides a waterway traffic feasibility prediction device, the waterway traffic feasibility prediction device 900 comprising: The data acquisition module 901 is used to acquire multi-source monitoring data of the target channel during the current target time period, and to construct a spatiotemporal map sequence of the target channel based on the multi-source monitoring data and a Bayesian network. The feature extraction module 902 is used to extract spatial dependency features from the spatiotemporal graph sequence using a preset graph neural network to determine the spatial feature embedding matrix of the target channel; The prediction module 903 is used to use the spatial feature embedding matrix as input to the preset waterway traffic feasibility prediction model to predict the traffic feasibility of the target waterway and obtain the prediction result.
[0096] The waterway feasibility prediction device 900 provided in the above embodiments can realize the technical solutions described in the above waterway feasibility prediction method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above waterway feasibility prediction method embodiments, and will not be repeated here.
[0097] like Figure 10 As shown, the present invention also provides an electronic device 1000. The electronic device 1000 includes a processor 1001, a memory 1002, and a display 1003. Figure 10 Only some components of the electronic device 1000 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.
[0098] In some embodiments, processor 1001 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in memory 1002 or process data, such as the waterway navigability feasibility prediction method of the present invention.
[0099] In some embodiments, processor 1001 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 1001 may be local or remote. In some embodiments, processor 1001 may be implemented on a cloud platform. In some embodiments, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, internal cloud, multi-cloud, or any combination thereof.
[0100] In some embodiments, memory 1002 may be an internal storage unit of electronic device 1000, such as a hard disk or memory of electronic device 1000. In other embodiments, memory 1002 may also be an external storage device of electronic device 1000, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 1000.
[0101] Furthermore, the memory 1002 may include both internal storage units of the electronic device 1000 and external storage devices. The memory 1002 is used to store application software and various types of data installed on the electronic device 1000.
[0102] In some embodiments, display 1003 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 1003 is used to display information from electronic device 1000 and to display a visual user interface. Components 1001-1003 of electronic device 1000 communicate with each other via a system bus.
[0103] In some embodiments, when the processor 1001 executes the waterway navigability prediction program in the memory 1002, the following steps may be implemented: Acquire multi-source monitoring data of the target channel for the current target time period, and construct a spatiotemporal map sequence of the target channel based on the multi-source monitoring data and a Bayesian network; A pre-defined graph neural network is used to extract spatial dependency features from the spatiotemporal graph sequence to determine the spatial feature embedding matrix of the target channel; The spatial feature embedding matrix is used as the input to a pre-defined waterway traffic feasibility prediction model to predict the traffic feasibility of the target waterway and obtain the prediction results.
[0104] It should be understood that when the processor 1001 executes the channel passage feasibility prediction program in the memory 1002, in addition to the functions mentioned above, it can also perform other functions, as can be found in the description of the corresponding method embodiments above.
[0105] Furthermore, the embodiments of the present invention do not specifically limit the type of the electronic device 1000 mentioned. The electronic device 1000 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, the electronic device 1000 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).
[0106] Accordingly, this application also provides a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the steps or functions of the waterway passage feasibility prediction method provided in the above-described method embodiments.
[0107] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.
[0108] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for predicting the feasibility of waterway passage, characterized in that, include: Acquire multi-source monitoring data of the target channel for the current target time period, and construct a spatiotemporal map sequence of the target channel based on the multi-source monitoring data and a Bayesian network; A preset graph neural network is used to extract spatial dependency features from the spatiotemporal graph sequence to determine the spatial feature embedding matrix of the target waterway; The spatial feature embedding matrix is used as input to a preset waterway traffic feasibility prediction model to predict the traffic feasibility of the target waterway and obtain the prediction result.
2. The method for predicting the feasibility of waterway passage according to claim 1, characterized in that, The acquisition of multi-source monitoring data of the target channel for the current target time period includes: Acquire AIS navigation data, hydrological monitoring data, navigation management and scheduling data, and meteorological monitoring data for the target waterway during the current target time period; The AIS navigation data, hydrological monitoring data, navigation management and scheduling data, and meteorological monitoring data are synchronized in time, outlier removal is performed, missing value interpolation is performed, and feature standardization is performed to obtain a multi-source monitoring data matrix of continuous time series of the target waterway.
3. The method for predicting the feasibility of waterway passage according to claim 2, characterized in that, The construction of the spatiotemporal map sequence of the target channel based on the multi-source monitoring data and a Bayesian network includes: A static adjacency matrix of multi-source monitoring data at the same time is constructed based on the predefined node system of the Bayesian network and the multi-source monitoring data matrix. A time-extended adjacency matrix is constructed based on the temporal relationship of the static adjacency matrix of multi-source monitoring data at different times and the state propagation relationship of nodes at adjacent times; The temporal extended adjacency matrix is block-diagonalized and spliced to obtain the spatiotemporal graph sequence of the target channel.
4. The method for predicting the feasibility of waterway passage according to claim 1, characterized in that, The step of extracting spatial dependency features from the spatiotemporal graph sequence using a preset graph neural network to determine the spatial feature embedding matrix of the target channel includes: The temporal extended adjacency matrix in the spatiotemporal graph sequence is subjected to self-connection addition and symmetric normalization to obtain a normalized adjacency matrix; Feature propagation is performed using a multi-layer stacked graph convolutional network to obtain the spatial feature embedding matrix of the target channel.
5. The method for predicting the feasibility of waterway passage according to claim 4, characterized in that, The step of using the spatial feature embedding matrix as input to a preset waterway traffic feasibility prediction model to predict the traffic feasibility of the target waterway and obtaining the prediction result includes: The spatial feature embedding matrix is reorganized into a node center time series, and a sinusoidal position code is added to obtain a time embedding sequence; The dependency strength at different times in the temporal embedding sequence is calculated using a multi-head self-attention mechanism; The dependency strength is nonlinearly mapped through a feedforward network, and after layer normalization and residual connection, the temporal features are obtained. The future navigability prediction result of the target waterway is obtained by predicting the time-series features using a prediction head.
6. The method for predicting the feasibility of waterway passage according to claim 5, characterized in that, The preset waterway traffic feasibility prediction model, during training, includes: Using the Bayesian network inference results as the teacher signal, a hybrid loss function including prediction loss, imitation loss and regularization loss is constructed to jointly optimize the waterway passage feasibility prediction model.
7. The method for predicting the feasibility of waterway passage according to claim 6, characterized in that, The prediction loss is: in, The number of samples in one training iteration; This represents the actual risk status of the i-th sample; To predict the probability that the i-th sample is at risk for the model; The imitation loss is: in, For Bayesian networks to the first i The inference probability of each sample category; For the BGS-T model, the first i The inference probability of each sample category; Indexed by category; The regularization loss is: in, The L2 regularization coefficient; These are the learnable parameters of the model; The square of the L2 norm; For structural constraint coefficients; The learned adjacency matrix; The prior adjacency matrix; It is the square of the Frobenius norm.
8. A device for predicting the feasibility of waterway passage, characterized in that, include: The data acquisition module is used to acquire multi-source monitoring data of the target channel during the current target time period, and construct a spatiotemporal map sequence of the target channel based on the multi-source monitoring data and a Bayesian network. The feature extraction module is used to extract spatial dependency features from the spatiotemporal graph sequence using a preset graph neural network to determine the spatial feature embedding matrix of the target channel; The prediction module is used to use the spatial feature embedding matrix as input to a preset waterway traffic feasibility prediction model to predict the traffic feasibility of the target waterway and obtain the prediction result.
9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the waterway navigability feasibility prediction method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the waterway navigability feasibility prediction method according to any one of claims 1 to 7.