Knowledge Graph-Driven Design Method for Tunnel Portal Structures

Abstract

This invention relates to the field of land transportation design technology and discloses a knowledge graph-driven method for designing tunnel portal structures. The method includes: constructing a tunnel portal structure knowledge graph ontology, wherein the ontology includes tunnel portal survey parameters; inputting the ontology into a tunnel portal structure prediction model for training, resulting in a trained prediction model; and using the trained model to predict the tunnel portal design parameters based on the survey parameters. This invention trains a tunnel portal structure prediction model using a tunnel portal knowledge graph ontology, predicts tunnel portal design parameters using the trained model, and then displays the portal design scheme. This solves the problem of low design efficiency caused by poor transfer of design experience and insufficient utilization of historical cases in the design of mining tunnel portals, thereby improving the efficiency of tunnel portal design.

Description

Technical Field

[0001] This invention relates to the field of land transportation design technology, and in particular to a knowledge graph-driven method for designing tunnel portal structures. Background Technology

[0002] With the continuous development of tunnel engineering technology, tunnel portal design technology is becoming increasingly advanced. However, in the field of tunnel engineering, there is a vast and complex knowledge framework, from design specifications to engineering data, which is actually intricate. Currently, there is no research on intelligent design methods for tunnel portal structures based on knowledge graphs, and traditional machine learning does not deeply consider the internal structural information and relationships between tunnel portal exploration parameters. The poor transfer of design experience and the failure to fully utilize historical cases in the design process of mining tunnel portals lead to low efficiency in tunnel portal design, and the efficiency of tunnel portal design urgently needs to be improved. Summary of the Invention

[0003] The main objective of this invention is to provide a knowledge graph-driven tunnel portal structure design scheme, which aims to solve the technical problems of poor transfer of design experience and lack of full utilization of historical cases in the design of tunnel portals in mining methods, resulting in low design efficiency.

[0004] To achieve the above objectives, this invention provides a knowledge graph-driven method for designing tunnel portal structures, the method comprising the following steps:

[0005] Design a knowledge graph ontology of tunnel portal structure, and construct a knowledge graph of tunnel portal structure based on the knowledge graph ontology. The knowledge graph of tunnel portal structure includes tunnel portal exploration parameters.

[0006] The tunnel portal structure knowledge graph case entities are used as a sample set and input into the tunnel portal structure prediction model for training to obtain the trained tunnel portal structure prediction model. The tunnel portal structure prediction model includes a preprocessing layer, a subgraph construction layer, an entity matching layer, a graph pooling layer, and a prediction layer.

[0007] The trained tunnel portal structure prediction model is used to predict the design parameters of the tunnel portal structure based on the exploration parameters to be predicted.

[0008] Optionally, the step of inputting the case entity sample set from the tunnel portal structure knowledge graph into the tunnel portal structure prediction model for training, to obtain the trained tunnel portal structure prediction model, includes:

[0009] The preprocessing layer encodes the tunnel portal exploration parameters and outputs the target feature vector.

[0010] The subgraph construction layer constructs exploration parameter nodes based on the feature vectors and builds a fully connected graph between the exploration parameter nodes.

[0011] The attributes of the fully connected graph are interactively modeled through the entity matching layer to obtain fused features;

[0012] The fused features are learned through graph pooling layers to obtain the target embedding vector;

[0013] The prediction layer performs fusion and dimensionality reduction based on the target embedding vector to obtain the prediction result, and then performs backpropagation training based on the prediction result to obtain the tunnel portal structure prediction model.

[0014] Optionally, the step of encoding the tunnel portal exploration parameters through the preprocessing layer to output a target feature vector includes:

[0015] The discrete and continuous values ​​in the tunnel portal exploration parameters are obtained, and the discrete and continuous value attributes in the tunnel portal exploration parameters are embedded as attributes in the tunnel portal structure knowledge graph.

[0016] The discrete value attributes in the tunnel portal exploration parameters are encoded using a first encoding method to obtain a first feature vector;

[0017] The continuous value attributes in the tunnel portal exploration parameters are encoded using a second encoding method to obtain a second feature vector;

[0018] The target feature vector is obtained based on the first feature vector and the second feature vector.

[0019] Optionally, the step of interactively modeling the attributes of the fully connected graph through the entity matching layer to obtain fused features includes:

[0020] By sequentially interacting with the attributes of any node in the fully connected graph with its neighboring nodes, the original characteristics of the node are obtained.

[0021] The feature values ​​of each dimension in the original features of the node are obtained, and the feature values ​​of each dimension in the original features of the node are fully connected to obtain the interaction information between different nodes.

[0022] The interaction information between the different nodes is aggregated to obtain node matching features;

[0023] The original features of the node and the matching features of the node are fused together to obtain the fused features.

[0024] Optionally, the step of learning the fused features through a graph pooling layer to obtain the target embedding vector includes:

[0025] The fusion features and the tunnel portal structure knowledge graph ontology are fused to obtain the tunnel exploration parameter ontology.

[0026] Context information is obtained by calculating based on the tunnel exploration parameters.

[0027] The attention weight of each node is calculated based on the embedding vector of each node in the tunnel exploration parameter ontology and the context information.

[0028] The target embedding vector is obtained by calculating the embedding vectors of all nodes in the tunnel exploration parameter ontology and the corresponding node attention coefficients.

[0029] Optionally, the backpropagation training based on the prediction results, resulting in a trained tunnel portal structure prediction model, includes:

[0030] The prediction difference is obtained by calculating based on the prediction results and the actual results using a preset loss function;

[0031] The prediction accuracy of the tunnel portal structure prediction model is determined based on the predicted differences.

[0032] When the prediction accuracy reaches a preset accuracy threshold, the current tunnel portal structure prediction model is used as the trained tunnel portal structure prediction model.

[0033] If the prediction accuracy does not reach the preset accuracy threshold, backpropagation training is performed until the prediction accuracy reaches the preset accuracy threshold, thus obtaining the trained tunnel portal structure prediction model.

[0034] Optionally, the construction of the tunnel portal structure knowledge graph ontology includes:

[0035] Multiple node labels are set, and data corresponding to each node label is obtained from the tunnel entrance survey and design database based on the node labels to obtain knowledge graph nodes;

[0036] The relationships between nodes in the knowledge graph are determined based on the triple rule of the knowledge graph.

[0037] A knowledge graph ontology of tunnel portal structure is constructed based on the knowledge graph nodes and the relationships between them.

[0038] Furthermore, to achieve the above objectives, the present invention also proposes a knowledge graph-driven tunnel portal structure design device, which includes:

[0039] A construction module is used to construct a knowledge graph ontology of tunnel portal structure, wherein the knowledge graph ontology of tunnel portal structure includes tunnel portal exploration parameters;

[0040] The training module is used to input the knowledge graph ontology of the tunnel portal structure into the tunnel portal structure prediction model for training, so as to obtain the trained tunnel portal structure prediction model. The tunnel portal structure prediction model includes a preprocessing layer, a subgraph construction layer, an entity matching layer, a graph pooling layer, and a prediction layer.

[0041] The prediction module is used to predict the tunnel portal design parameters based on the survey parameters to be predicted using the trained tunnel portal structure prediction model.

[0042] Furthermore, to achieve the above objectives, the present invention also proposes a knowledge graph-driven tunnel portal structure design device, which includes: a memory, a processor, and a knowledge graph-driven tunnel portal structure design program stored in the memory and executable on the processor. The knowledge graph-driven tunnel portal structure design program is configured to implement the steps of the knowledge graph-driven tunnel portal structure design method described above.

[0043] Furthermore, to achieve the above objectives, the present invention also proposes a storage medium storing a knowledge graph-driven tunnel portal structure design program. When the knowledge graph-driven tunnel portal structure design program is executed by a processor, it implements the steps of the knowledge graph-driven tunnel portal structure design method described above.

[0044] This invention constructs a knowledge graph ontology of tunnel portal structures, which includes tunnel portal survey parameters. The knowledge graph ontology is then input into a tunnel portal structure prediction model for training, resulting in a trained prediction model. This model then predicts the tunnel portal design parameters based on the survey parameters. By training the tunnel portal structure prediction model using the knowledge graph ontology, and then using the trained model to predict the tunnel portal design parameters, this invention improves tunnel portal design efficiency. This addresses the problems of poor experience transfer and insufficient utilization of historical cases in the design of mining tunnel portals, which lead to low design efficiency. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the structure of the tunnel portal structure design device driven by the knowledge graph of the hardware operating environment involved in the embodiments of the present invention;

[0046] Figure 2 This is a flowchart illustrating the first embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0047] Figure 3 This is a knowledge graph ontology diagram of a tunnel portal structure design method driven by the knowledge graph of the present invention.

[0048] Figure 4 This is a framework diagram of a tunnel portal structure prediction model, representing an embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0049] Figure 5 This is a flowchart illustrating the second embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0050] Figure 6 This is a flowchart of Autodis encoding of continuous exploration data for tunnel portals, an embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0051] Figure 7 This is a schematic diagram of a fully connected neural network computation of knowledge graph nodes, representing an embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0052] Figure 8 This is a schematic diagram of the calculation of a GRU (Governing Recurrent Neural Network) model, which is an embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0053] Figure 9 This is a diagram showing the connection of entity nodes in the knowledge graph of the exploration parameters in an embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0054] Figure 10 This is a diagram of a convolutional layer network architecture of an embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0055] Figure 11 This is an interpretability result diagram of intelligent design of tunnel portal type, which is an embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0056] Figure 12 This is a schematic diagram of the portal type prediction confusion matrix of an embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0057] Figure 13 This is a structural block diagram of the first embodiment of the knowledge graph-driven tunnel portal structure design device of the present invention.

[0058] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0059] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0060] Reference Figure 1 , Figure 1 This is a schematic diagram of the equipment structure for a knowledge graph-driven tunnel portal structure design scheme involving the hardware operating environment of an embodiment of the present invention.

[0061] like Figure 1 As shown, the knowledge graph-driven tunnel portal structure design device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk storage device. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0062] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the design equipment for knowledge graph-driven tunnel portal structures. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0063] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a knowledge graph-driven tunnel portal structure design program.

[0064] exist Figure 1In the knowledge graph-driven tunnel portal structure design device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and memory 1005 in the knowledge graph-driven tunnel portal structure design device of the present invention can be set in the knowledge graph-driven tunnel portal structure design device, and the knowledge graph-driven tunnel portal structure design device calls the knowledge graph-driven tunnel portal structure design program stored in the memory 1005 through the processor 1001, and executes the knowledge graph-driven tunnel portal structure design method provided in the embodiment of the present invention.

[0065] This invention provides a knowledge graph-driven design method for tunnel portal structures, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0066] In this embodiment, the knowledge graph-driven tunnel portal structure design method includes the following steps:

[0067] Step S10: Construct a knowledge graph ontology of tunnel portal structure, wherein the knowledge graph ontology of tunnel portal structure includes tunnel portal survey parameters.

[0068] It should be noted that the execution subject of this embodiment is a knowledge graph-driven tunnel portal structure design device, but it can also be other devices that can achieve the same or similar functions. This embodiment does not limit this; this embodiment uses a knowledge graph-driven tunnel portal structure design device as an example for explanation.

[0069] It is understandable that the tunnel portal structure knowledge graph ontology consists of multiple entities and the relationships between them. Entities are the nodes in the tunnel portal structure knowledge graph ontology, and the relationships between entities are the connecting lines between the nodes. The tunnel portal structure knowledge graph can include tunnel entities, influencing factor entities, design parameter entities, etc. Among them, the influencing factor entities can be further divided into tunnel length, circumscribed type, circumscribed angle, etc., and the design parameter entities can be divided into tunnel portal length entities, portal type entities, and number of openings entities. This embodiment does not impose specific limitations on these. The influencing factor entities are the tunnel portal survey parameters, that is, the tunnel portal structure knowledge graph ontology includes tunnel portal survey parameters.

[0070] Furthermore, the construction of the tunnel portal structure knowledge graph ontology includes: setting multiple node labels, and obtaining data corresponding to each node label from the tunnel portal survey and design database based on the node labels to obtain knowledge graph nodes; determining the relationship between the knowledge graph nodes according to the knowledge graph triple rule; and constructing the tunnel portal structure knowledge graph ontology based on the knowledge graph nodes and the relationship between the knowledge graph nodes.

[0071] It should be noted that the tunnel entity, influencing factor entity, and design parameter entity are used as node labels, and the labels of each data in the tunnel portal survey and design database are matched to obtain the data corresponding to each node label, i.e., knowledge graph nodes.

[0072] It is understandable that mapping the relationships between entities to a knowledge graph is equivalent to mapping the relationships between different nodes. The process of establishing relationships between knowledge graph nodes can be summarized, for example, according to the rules of knowledge graph triples, namely, according to the patterns of "entity-relationship-entity" and "entity-attribute-value".

[0073] In practical implementation, based on the tunnel portal survey and design database, data under the corresponding tags in the tunnel portal survey and design database is read into the entities in the map through data mining and information retrieval from numerous tags, realizing entity extraction and data reading, and then standardizing and constraining the data layer.

[0074] like Figure 3 As shown, Figure 3 This is the ontology of the tunnel portal structure knowledge graph. The ontology is constructed from top to bottom, consisting of multiple entities and relationships between them. Entities are the nodes in the ontology, and relationships are the connecting lines between nodes. The ontology includes tunnel entities, survey parameter entities, and design parameter entities. Survey parameter entities can be categorized into tunnel length, circumscribed type, circumscribed angle, longitudinal section coordinates, key sections, and portal entities. Portal entities can be categorized into burial depth entities and geotechnical parameter entities. Geotechnical parameter entities can be categorized into lithology, construction grade, weathering degree, and bearing capacity entities. Design parameter entities can be categorized into portal structure parameters and slope parameters. Portal structure parameters include tunnel portal length entities, portal type entities, and number of openings entities. Slope parameters include slope ratio entities and side slope ratio entities.

[0075] Step S20: Input the knowledge graph ontology of the tunnel portal structure into the tunnel portal structure prediction model for training to obtain the trained tunnel portal structure prediction model, wherein the tunnel portal structure prediction model includes a preprocessing layer, a subgraph construction layer, an entity matching layer, a graph pooling layer and a prediction layer.

[0076] It should be noted that the tunnel portal structure knowledge graph ontology also includes tunnel portal design experience. By collecting tunnel portal design cases to establish a tunnel portal design case knowledge graph, and integrating the design experience from the tunnel portal knowledge graph ontology into the tunnel portal knowledge graph ontology, further training is performed to obtain a tunnel portal structure prediction model.

[0077] It is understandable that tunnel portal design cases from the tunnel portal design case knowledge graph and tunnel portal survey parameters from the tunnel portal structure ontology are used as samples, combined with tunnel portal design experience, to train the tunnel portal structure prediction model.

[0078] Understandably, the preprocessing layer is used to encode different types of input data and output vectors of the same dimension; the subgraph construction layer is used to fully learn the relationships between entities in the survey parameters; the entity matching layer is used to learn entity node features using fully connected neural networks and recurrent gate units; the graph pooling layer, also known as the graph convolutional layer, is used to strengthen the learning of internal features using an attention mechanism to obtain the embedding vector; the prediction layer uses three fully connected layers to fuse and reduce the dimensionality of the embedding vector, and finally outputs the classification result, and uses the cross-entropy loss function to calculate the difference between the classification result and the label, and performs backpropagation training.

[0079] like Figure 4 As shown, Figure 4 This is a framework diagram of a tunnel portal structure prediction model. The model includes a preprocessing layer, a subgraph construction layer, an entity matching layer, a graph convolutional layer, and a prediction layer. The input layer takes tunnel portal exploration parameters such as lithology, weathering degree, bearing capacity, tunnel length, and circumscribed angle as input to the preprocessing layer. The preprocessing layer encodes these parameters, outputting vectors of the same dimension. These vectors are then input to the subgraph construction layer to establish a fully connected graph between the exploration parameter entities. The entity matching layer uses a fully connected network (MLP) and a recurrent gated neural network (GRU) to model the interaction of attributes in the fully connected graph between the exploration parameter entities, learning entity node features and obtaining fused features. The graph convolutional layer uses a knowledge graph ontology of tunnel portal exploration parameters built based on empirical knowledge to construct a three-layer graph convolutional layer that captures the attribute relationships between the tunnel portal exploration parameters. Meanwhile, an attention mechanism is used to enhance internal feature learning and obtain the embedding vector of the ontology graph. The prediction layer uses a fully connected network MLP to fuse and reduce the dimensionality of the embedding vector to predict the tunnel portal length, portal type or number of openings.

[0080] Step S30: The tunnel portal structure prediction model, after training, is used to predict the exploration parameters to obtain the tunnel portal design parameters.

[0081] It should be noted that the tunnel portal design parameters can include the tunnel portal length, portal type, and number of openings, etc. This embodiment does not impose specific restrictions on these parameters; instead, it uses portal type as an example for detailed explanation. The trained tunnel portal structure prediction model predicts only one type of tunnel portal design parameter each time, requiring separate predictions for the tunnel portal length, portal type, and number of openings.

[0082] In the specific implementation, after obtaining the tunnel portal design parameters, the portal design parameters obtained by intelligent decision-making are presented in three dimensions through the BIM engine to form a tunnel portal structure scheme, thus realizing the three-dimensional visualization of the portal design scheme.

[0083] This embodiment constructs a knowledge graph ontology of tunnel portal structures, which includes tunnel portal survey parameters. The knowledge graph ontology is then input into a tunnel portal structure prediction model for training, resulting in a trained prediction model. This model then predicts the tunnel portal design parameters based on the survey parameters. By training the tunnel portal structure prediction model using the knowledge graph ontology, and then using the trained model to predict the tunnel portal design parameters, the embodiment demonstrates the portal design scheme. This approach addresses the problems of poor experience transfer and insufficient utilization of historical cases in the design of mining tunnel portals, leading to low design efficiency and improving overall tunnel portal design efficiency.

[0084] refer to Figure 5 , Figure 5 This is a flowchart illustrating the second embodiment of the knowledge graph-driven tunnel portal structure design method of the present invention.

[0085] Based on the first embodiment described above, step S20 in the knowledge graph-driven tunnel portal structure design method of this embodiment includes:

[0086] Step S201: Encode the tunnel entrance survey parameters through the preprocessing layer and output the target feature vector.

[0087] It should be noted that the preprocessing layer encodes the tunnel entrance survey parameters and outputs a vector of the same dimension, namely the target feature vector.

[0088] Further, the step of encoding the tunnel portal survey parameters through the preprocessing layer to output a target feature vector includes: obtaining discrete and continuous values ​​in the tunnel portal survey parameters, and embedding the discrete and continuous values ​​in the tunnel portal survey parameters as attributes in a tunnel portal structure knowledge graph; encoding the discrete values ​​in the tunnel portal survey parameters using a first encoding method to obtain a first feature vector; encoding the continuous values ​​in the tunnel portal survey parameters using a second encoding method to obtain a second feature vector; and obtaining a target feature vector based on the first feature vector and the second feature vector.

[0089] It should be noted that this embodiment does not impose specific restrictions on discrete values ​​in tunnel exploration parameters, such as lithology, weathering degree, and construction grade, or on continuous values ​​in exploration parameters, such as tunnel length, external angle, and tunnel entrance burial depth.

[0090] Understandably, the first encoding method could be one-hot encoding, which is a method of converting categorical variables into several binary columns, and the second encoding method could be Autodis encoding.

[0091] In the specific implementation, discrete values ​​are encoded using one-hot encoding. This includes: converting each discrete value into a one-hot encoded vector, with the vector dimension being the maximum value among all tunnel exploration parameters, resulting in a vector dimension of 26. For example, if the weathering degree is 1, the one-hot encoded vector would be [0,1,0,0,...,0], with a vector dimension of 26; and then multiplying the one-hot encoded value by a weight matrix. The input data is embedded to obtain the output value. This is the first feature vector. During training, gradient backpropagation is performed to update the weights, searching for an embedding method that reduces the loss, and the output value E is... i As shown in Formula 1:

[0092] E i =onehot(T) i )*W (Equation 1)

[0093] In Equation 1, E i For the output value, T i W is a one-hot encoded vector, and W is the weight matrix.

[0094] Continuous values ​​are encoded using Autodis, specifically by using a two-layer neural network to encode continuous numerical features x. j The data is discretized into 26 buckets. During the computation, the data features are normalized to a normal distribution, resulting in a vector dimension of 26 for the feature mapping results. The feature mapping results are then normalized using SoftMax to obtain the feature vectors. Define a set of shared meta-embeddings: These meta-embeddings can improve the model's expressive power for each continuous numerical feature. The input continuous numerical feature x j Perform upsampling processing to become Share meta-embeddings with x j The dot product yields the vector as These Hj embedding vectors are aggregated in a certain way to obtain the final continuous numerical feature embedding representation. During training, Meta-Embedding also belongs to a variable vector space. During the aggregation calculation, the feature vectors... with vector Perform a dot product and sum the corresponding row numbers to obtain the second encoding vector. like Figure 6 As shown, Figure 6 Autodis encoding flowchart for continuous survey data of tunnel portals.

[0095] Step S202: Construct exploration parameter nodes based on the feature vector through the subgraph construction layer, and construct a fully connected graph between the exploration parameter nodes.

[0096] It should be noted that the subgraph construction layer fully learns the interrelationships between the entities of the exploration parameters. That is, after the data preprocessing is completed, the entities of the tunnel exploration parameters in the tunnel portal structure knowledge graph are fully connected, that is, all exploration parameter entities are connected in pairs to represent the interaction relationship between the exploration parameter attribute nodes, and a fully connected graph between exploration parameter nodes is obtained.

[0097] Step S203: The attributes of the fully connected graph are interactively modeled through the entity matching layer to obtain fused features.

[0098] It should be noted that the entity matching layer performs interactive modeling of the attributes in the fully connected graph between entities of the survey parameters, and uses fully connected neural networks and recurrent gate units to learn the features of entity nodes.

[0099] Further, the step of interactively modeling the attributes of the fully connected graph through the entity matching layer to obtain fused features includes: sequentially interacting the attributes of any node in the fully connected graph with its neighboring nodes to obtain the original features of the node; obtaining the feature values ​​of each dimension in the original features of the node, and performing a full connection on the feature values ​​of each dimension in the original features of the node to obtain the interaction information between different nodes; aggregating the interaction information between different nodes to obtain node matching features; and fusing the original features of the node and the node matching features to obtain fused features.

[0100] In the specific implementation, each node in the fully connected graph of the knowledge graph is... The nodes adjacent to it in sequence Perform a dot product operation to calculate the preliminary interaction representation between node i and node j. This refers to the original characteristics of the node, as shown in Equation 2 below:

[0101] E ij =E i *E j (Equation 2)

[0102] In Equation 2, E ij E represents the initial interaction between node i and node j, i.e., the original characteristics of the nodes. i For node i, E j Let j be the neighboring node of node i.

[0103] Establish the structure of a fully connected MLP network, with the input being the preliminary interaction representation E between node i and node j. ij Each point e in the input layer k E ij The feature values ​​of one dimension are processed through two fully connected layers to obtain the further interaction result P. ij Similarly, for each point P in the output layer k P represents ij The feature value of a feature in one dimension, where the input and output dimensions are both 64, and the hidden layer dimension is 128. Each point h in the fully connected MLP network... k Each layer is connected to all points in the previous layer, thus fully fusing the features of different points. The output of each layer is passed to the next layer through a ReLU activation function. Finally, after two fully connected layers, the interaction information between different nodes is effectively captured, such as... Figure 7 As shown, Figure 7 This is a schematic diagram of the fully connected neural network calculation for knowledge graph nodes. The calculation formula is shown in Equation 3 below:

[0104] E ij =E i *E j h k =w k1 *e1+w k2 *e2+…+w k64 *e 64 +b k (Equation 3)

[0105] In Equation 3, E ij E represents the initial interaction between node i and node j, i.e., the original characteristics of the nodes. i For node i, E j Let j be the neighboring nodes of node i, e1, e2, ..., e64 The point h is the input layer. k For nodes in the hidden layer, w k1 w k2 , ..., w k64 These are the weights of the fully connected layer, b k These are the bias terms of the current fully connected layer, and both are learnable parameters.

[0106] Aggregate the attribute interaction results between node i and all its neighboring nodes to obtain node matching features. As shown in equation 4:

[0107]

[0108] In Equation 3, N i M is the number of neighboring nodes of node i. i P represents the embedding representation of node i obtained after modeling the internal interactions between node i and all its neighboring nodes, i.e., the node matching feature. ij The result of attribute interaction between node i and its neighbor node j.

[0109] The graph node matching information M obtained through the fully connected neural network MLP is processed using a recurrent gated neural network (GRU) model. t With the original feature E of the node t-1 To merge, such as Figure 8 As shown, Figure 8 This is a schematic diagram of the computation of a recurrent gated neural network (GRU) model. The GRU adaptively modifies the original features E of the nodes. t-1 Entity matching feature M t The original features and entity matching features are fully learned through fusion.

[0110] Step S204: Learn the fused features through a graph pooling layer to obtain the target embedding vector.

[0111] It should be noted that the graph pooling layer, also known as the graph convolutional layer, utilizes a knowledge graph ontology of tunnel entrance survey parameters built based on empirical knowledge to construct a three-layer graph convolutional layer that captures the attribute relationships between tunnel entrance survey parameters. Simultaneously, an attention mechanism is employed to enhance internal feature learning, resulting in the embedding vector of the ontology graph, i.e., the target embedding vector.

[0112] Furthermore, the step of learning the fused features through a graph pooling layer to obtain the target embedding vector includes: fusing the fused features with the tunnel portal structure knowledge graph ontology to obtain a tunnel exploration parameter ontology; calculating context information based on the tunnel exploration parameter ontology; calculating the attention weight of each node based on the embedding vector of each node in the tunnel exploration parameter ontology and the context information; and calculating the target embedding vector based on the embedding vectors of all nodes in the tunnel exploration parameter ontology and the corresponding node attention coefficients.

[0113] It should be noted that by integrating node entity features and knowledge graph ontology, the graph neural network learns the prior knowledge and interrelationships in the tunnel portal design process. The adjacency matrix in the fully connected tunnel graph is then replaced with an adjacency matrix in the ontology graph, thereby constructing the tunnel exploration parameter ontology graph.

[0114] like Figure 9 As shown, Figure 9 The entity node connection diagram of the knowledge graph of exploration parameters is as follows: the entities of exploration parameters can be divided into entities such as tunnel length, external type, external angle, longitudinal section coordinates, key section and tunnel entrance. The entities at the tunnel entrance can be divided into entities of burial depth and entities of geotechnical parameters. The entities of geotechnical parameters can be divided into entities such as lithology, construction grade, weathering degree and bearing capacity.

[0115] like Figure 10 As shown, Figure 10 The graph convolutional layer network architecture diagram shows a three-layer graph convolutional layer to capture the attribute interaction information of the tunnel exploration parameter ontology graph. A ReLU activation function is added after each graph convolutional layer to enhance the model's nonlinearity and improve its sparsity. To classify the tunnel exploration parameter ontology graph, all attribute nodes need to be fused into a one-dimensional vector, and attention pooling is used to enhance the features of important and minor entity nodes. Figure 10 Attention pooling in the ontology graph multiplies the feature matrix G of the exploration parameter attribute nodes with a learnable weight matrix W, and then takes the average to obtain the context information of the ontology graph. At this time, C has the global information of the ontology graph, as shown in Equation 5 below:

[0116]

[0117] In Equation 5, C represents context information, N represents the number of exploration parameter nodes in the ontology graph, and g i It is the embedding vector of each exploration parameter node in the ontology graph, g i ∈R 1×d d represents the dimension of the embedding vector. In this embodiment, N is 19 and d is 128. This embodiment does not impose specific limitations on this.

[0118] The embedding vector g of each node in the ontology graph i Transpose of context information C T Perform the inner product to obtain the attention weight 'a' for that node. i Ultimately, the attention weights of each node are combined into a vector A. This operation gives nodes in the ontology graph that are similar to the global context information a higher attention coefficient. The attention weights a of the nodes are... i As shown in equation 6:

[0119] a i =g i *C T (Equation 6)

[0120] In Equation 6, a i For each node's embedding vector, C T g is the transpose of context information. i The embedding vector for each node.

[0121] The final embedding vector v of the ontology graph is obtained by weighted summing of the embedding vectors of all nodes in the ontology graph with their corresponding attention coefficients. G That is, the target embedding vector.

[0122] Step S205: The prediction layer performs fusion and dimensionality reduction based on the target embedding vector to obtain the prediction result, and performs backpropagation training based on the prediction result to obtain the tunnel portal structure prediction model.

[0123] It should be noted that three fully connected layers are used to fuse and reduce the dimensionality of the embedded vectors, and finally output the prediction results. The cross-entropy loss function is used to calculate the difference between the prediction results and the labels, and backpropagation training is performed.

[0124] Further, the backpropagation training based on the prediction results to obtain the trained tunnel portal structure prediction model includes: calculating the prediction difference based on the prediction results and the actual results using a preset loss function; determining the prediction accuracy of the tunnel portal structure prediction model based on the prediction difference; when the prediction accuracy reaches a preset accuracy threshold, using the current tunnel portal structure prediction model as the trained tunnel portal structure prediction model; and when the prediction accuracy does not reach the preset accuracy threshold, performing backpropagation training until the prediction accuracy reaches the preset accuracy threshold to obtain the trained tunnel portal structure prediction model.

[0125] It should be noted that the preset loss function can be the cross-entropy loss function or other loss functions. This embodiment does not impose specific restrictions on this. After obtaining the embedding vector of the ontology graph, i.e., the target embedding vector, a three-layer fully connected layer is used to fuse and reduce the dimensionality of the target embedding vector. To avoid overfitting in the trained tunnel recommendation model, a dropout operation is added between the second and third fully connected layers, and the final prediction result is output. The difference between the prediction result and the true result is calculated using the cross-entropy loss function, and backpropagation training is performed. The loss function is as follows: Equation 7:

[0126]

[0127] In Equation 7, L is the cross-entropy loss function, N is the number of samples, and M is the number of categories. For example, if the number of categories M for the portal type is 5, then y ij This is a sign function; it takes the value 1 if the true class of sample i is equal to j, and 0 otherwise. P ij The model predicts the probability that sample i belongs to category j.

[0128] In the specific implementation, the trained tunnel portal structure prediction model was validated and tested to predict tunnel portal types. 400 tunnel data points were used, including 13 tunnel exploration parameters: 9 continuous parameters (tunnel length, circumscribed angle, left burial depth, middle burial depth, right burial depth, bearing capacity, left burial depth of key section, middle burial depth of key section, and right burial depth of key section) and 4 discrete parameters (circumscribed type, lithology, weathering degree, and construction grade). Furthermore, when drawing the ontology diagram of the tunnel exploration parameters, 6 additional parameters were added to describe the relationships between the various exploration parameters: exploration parameter, key section, key section burial depth, portal location, portal burial depth, and portal geotechnical parameters. To avoid interfering with the model, all 6 parameters were set to 1. Tunnel portal types are mainly divided into four categories: cut portal-0, straight-cut portal-1, wall portal-2, open portal-3, and other portal-4. The tunnel portal type is predicted by the trained tunnel portal structure prediction model.

[0129] Understandably, to explore the influence of different exploration parameters on the final model results, the interpretability algorithm GNNExplainer is used to perform interpretability analysis on the trained model, outputting the importance weight of each node in the exploration parameters and performing normalization. The influence of exploration parameters on the model output results is represented by the weights indicated on the connecting lines, such as... Figure 11 As shown, Figure 11 Intelligent design interpretability result diagrams for portal types.

[0130] It is worth noting that, through model training and learning, the final accuracy of predicting tunnel portal types is as follows: Figure 12 As shown, Figure 12 This is a schematic diagram of the confusion matrix for predicting portal types. The horizontal axis represents the predicted portal type, and the vertical axis represents the actual portal type. The prediction accuracy is 87.5%.

[0131] This embodiment encodes tunnel portal survey parameters using the preprocessing layer, outputting a target feature vector. The subgraph construction layer constructs survey parameter nodes based on the feature vector and builds a fully connected graph between these nodes. The entity matching layer interactively models the attributes of the fully connected graph to obtain fused features. The graph pooling layer learns from the fused features to obtain a target embedding vector. The prediction layer performs fusion and dimensionality reduction based on the target embedding vector to obtain a prediction result, and backpropagation training is performed based on the prediction result to train a tunnel portal structure prediction model. Through this method, the preprocessing layer encodes tunnel portal survey parameters, the subgraph construction layer constructs a fully connected graph between survey parameter nodes, the entity matching layer interactively models the attributes of the fully connected graph to obtain fused features, the graph pooling layer learns from the fused features to obtain a target embedding vector, and the prediction layer makes predictions based on the target embedding vector, thereby achieving a tunnel portal structure prediction model and improving prediction accuracy.

[0132] Reference Figure 13 , Figure 13 This is a structural block diagram of the first embodiment of the knowledge graph-driven tunnel portal structure design device of the present invention.

[0133] like Figure 13 As shown, the knowledge graph-driven tunnel portal structure design device proposed in this embodiment of the invention includes:

[0134] Construction module 10 is used to construct a knowledge graph ontology of tunnel portal structure, wherein the knowledge graph ontology of tunnel portal structure includes tunnel portal exploration parameters;

[0135] Training module 20 is used to input the knowledge graph ontology of the tunnel portal structure into the tunnel portal structure prediction model for training, so as to obtain the trained tunnel portal structure prediction model. The tunnel portal structure prediction model includes a preprocessing layer, a subgraph construction layer, an entity matching layer, a graph pooling layer and a prediction layer.

[0136] The prediction module 30 is used to predict the tunnel portal design parameters based on the survey parameters to be predicted using the trained tunnel portal structure prediction model.

[0137] This embodiment constructs a knowledge graph ontology of tunnel portal structures, which includes tunnel portal survey parameters. The knowledge graph ontology is then input into a tunnel portal structure prediction model for training, resulting in a trained prediction model. This model then predicts the tunnel portal design parameters based on the survey parameters. By training the tunnel portal structure prediction model using the knowledge graph ontology, and then using the trained model to predict the tunnel portal design parameters, the embodiment demonstrates the portal design scheme. This approach addresses the problems of poor experience transfer and insufficient utilization of historical cases in the design of mining tunnel portals, leading to low design efficiency and improving overall tunnel portal design efficiency.

[0138] In one embodiment, the training module 20 is further configured to encode the tunnel portal survey parameters through the preprocessing layer and output a target feature vector; construct survey parameter nodes based on the feature vector through the subgraph construction layer and construct a fully connected graph between the survey parameter nodes; perform interactive modeling of the attributes of the fully connected graph through the entity matching layer to obtain fused features; learn the fused features through the graph pooling layer to obtain a target embedding vector; perform fusion and dimensionality reduction based on the target embedding vector through the prediction layer to obtain a prediction result, and perform backpropagation training based on the prediction result to obtain a trained tunnel portal structure prediction model.

[0139] In one embodiment, the training module 20 is further configured to acquire discrete and continuous values ​​in the tunnel portal survey parameters, and embed the discrete and continuous values ​​in the tunnel portal survey parameters as attributes in the tunnel portal structure knowledge graph; encode the discrete values ​​in the tunnel portal survey parameters using a first encoding method to obtain a first feature vector; encode the continuous values ​​in the tunnel portal survey parameters using a second encoding method to obtain a second feature vector; and obtain a target feature vector based on the first feature vector and the second feature vector.

[0140] In one embodiment, the training module 20 is further configured to sequentially perform attribute interactions between any node in the fully connected graph and its neighboring nodes to obtain the original features of the node; obtain the feature values ​​of each dimension in the original features of the node, and perform a full connection on the feature values ​​of each dimension in the original features of the node to obtain the interaction information between different nodes; aggregate the interaction information between different nodes to obtain node matching features; and fuse the original features of the node and the node matching features to obtain fused features.

[0141] In one embodiment, the training module 20 is further configured to fuse the fusion features and the tunnel portal structure knowledge graph ontology to obtain a tunnel exploration parameter ontology; calculate context information based on the tunnel exploration parameter ontology; calculate attention weights for each node based on the embedding vector of each node in the tunnel exploration parameter ontology and the context information; and calculate the target embedding vector based on the embedding vectors of all nodes in the tunnel exploration parameter ontology and the attention coefficients of the corresponding nodes.

[0142] In one embodiment, the training module 20 is further configured to calculate the prediction difference based on the prediction result and the actual result using a preset loss function; determine the prediction accuracy of the tunnel portal structure prediction model based on the prediction difference; when the prediction accuracy reaches a preset accuracy threshold, use the current tunnel portal structure prediction model as the trained tunnel portal structure prediction model; when the prediction accuracy does not reach the preset accuracy threshold, perform backpropagation training until the prediction accuracy reaches the preset accuracy threshold to obtain the trained tunnel portal structure prediction model.

[0143] In one embodiment, the construction module 10 is further configured to set multiple node labels, and obtain data corresponding to each node label from the tunnel portal survey and design database based on the node labels to obtain knowledge graph nodes; determine the relationship between the knowledge graph nodes according to the knowledge graph triple rule; and construct a tunnel portal structure knowledge graph ontology based on the knowledge graph nodes and the relationship between the knowledge graph nodes.

[0144] Furthermore, to achieve the above objectives, the present invention also proposes a knowledge graph-driven tunnel portal structure design device, which includes: a memory, a processor, and a knowledge graph-driven tunnel portal structure design program stored in the memory and executable on the processor. The knowledge graph-driven tunnel portal structure design program is configured to implement the steps of the knowledge graph-driven tunnel portal structure design method described above.

[0145] Since the knowledge graph-driven tunnel portal structure design equipment adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be repeated here.

[0146] Furthermore, this embodiment of the invention also proposes a storage medium storing a knowledge graph-driven tunnel portal structure design program. When the knowledge graph-driven tunnel portal structure design program is executed by a processor, it implements the steps of the knowledge graph-driven tunnel portal structure design method described above.

[0147] Since this storage medium adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be repeated here.

[0148] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solutions of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any restrictions on this.

[0149] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0150] In addition, for technical details not described in detail in this embodiment, please refer to the knowledge graph-driven tunnel portal structure design method provided in any embodiment of the present invention, which will not be repeated here.

[0151] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0152] It should be understood that although the steps in the flowcharts of this application's embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0153] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0154] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0155] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A knowledge graph-driven design method for tunnel portal structures, characterized in that, The method includes: Construct a knowledge graph ontology of tunnel portal structure, wherein the knowledge graph ontology of tunnel portal structure includes tunnel portal survey parameters; The knowledge graph ontology of the tunnel portal structure is input into the tunnel portal structure prediction model for training to obtain the trained tunnel portal structure prediction model, wherein the tunnel portal structure prediction model includes a preprocessing layer, a subgraph construction layer, an entity matching layer, a graph pooling layer and a prediction layer. The tunnel portal structure prediction model, after being trained, predicts the tunnel portal design parameters based on the exploration parameters to be predicted. The step of inputting the knowledge graph ontology of the tunnel portal structure into the tunnel portal structure prediction model for training, to obtain the trained tunnel portal structure prediction model, includes: The preprocessing layer encodes the tunnel portal exploration parameters and outputs the target feature vector. The subgraph construction layer constructs exploration parameter nodes based on the feature vectors and builds a fully connected graph between the exploration parameter nodes. The attributes of the fully connected graph are interactively modeled through the entity matching layer to obtain fused features; The fused features are learned through graph pooling layers to obtain the target embedding vector; The prediction layer performs fusion and dimensionality reduction based on the target embedding vector to obtain the prediction result, and then performs backpropagation training based on the prediction result to obtain the trained tunnel portal structure prediction model.

2. The method as described in claim 1, characterized in that, The process of encoding the tunnel portal exploration parameters through the preprocessing layer and outputting the target feature vector includes: The discrete and continuous values ​​of the tunnel portal exploration parameters are obtained, and the discrete and continuous values ​​of the tunnel portal exploration parameters are embedded as attributes in the tunnel portal structure knowledge graph; The discrete values ​​in the tunnel portal exploration parameters are encoded using a first encoding method to obtain a first feature vector; The continuous values ​​in the tunnel portal exploration parameters are encoded using a second encoding method to obtain a second feature vector; The target feature vector is obtained based on the first feature vector and the second feature vector.

3. The method as described in claim 1, characterized in that, The step of interactively modeling the attributes of the fully connected graph through the entity matching layer to obtain fused features includes: By sequentially interacting with the attributes of any node in the fully connected graph with its neighboring nodes, the original characteristics of the node are obtained. The feature values ​​of each dimension in the original features of the node are obtained, and the feature values ​​of each dimension in the original features of the node are fully connected to obtain the interaction information between different nodes. The interaction information between the different nodes is aggregated to obtain node matching features; The original features of the node and the matching features of the node are fused together to obtain the fused features.

4. The method as described in claim 1, characterized in that, The step of learning the fused features through a graph pooling layer to obtain the target embedding vector includes: The fusion features and the tunnel portal structure knowledge graph ontology are fused to obtain the tunnel exploration parameter ontology. Context information is obtained by calculating based on the tunnel exploration parameters. The attention weight of each node is calculated based on the embedding vector of each node in the tunnel exploration parameter ontology and the context information. The target embedding vector is obtained by calculating the embedding vectors of all nodes in the tunnel exploration parameter ontology and the corresponding node attention coefficients.

5. The method as described in claim 1, characterized in that, The backpropagation training based on the prediction results, resulting in a tunnel portal structure prediction model, includes: The prediction difference is obtained by calculating based on the prediction results and the actual results using a preset loss function; The prediction accuracy of the tunnel portal structure prediction model is determined based on the predicted differences. When the prediction accuracy reaches a preset accuracy threshold, the current tunnel portal structure prediction model is used as the trained tunnel portal structure prediction model. If the prediction accuracy does not reach the preset accuracy threshold, backpropagation training is performed until the prediction accuracy reaches the preset accuracy threshold, thus obtaining the trained tunnel portal structure prediction model.

6. The method according to any one of claims 1 to 5, characterized in that, The knowledge graph ontology for constructing tunnel portal structures includes: Multiple node labels are set, and data corresponding to each node label is obtained from the tunnel entrance survey and design database based on the node labels to obtain knowledge graph nodes; The relationships between nodes in the knowledge graph are determined based on the triple rule of the knowledge graph. A knowledge graph ontology of tunnel portal structure is constructed based on the knowledge graph nodes and the relationships between them.

7. A knowledge graph-driven tunnel portal structure design device, characterized in that, The knowledge graph-driven tunnel portal structure design device includes: A construction module is used to construct a knowledge graph ontology of tunnel portal structure, wherein the knowledge graph ontology of tunnel portal structure includes tunnel portal exploration parameters; The training module is used to input the knowledge graph ontology of the tunnel portal structure into the tunnel portal structure prediction model for training, so as to obtain the trained tunnel portal structure prediction model. The tunnel portal structure prediction model includes a preprocessing layer, a subgraph construction layer, an entity matching layer, a graph pooling layer, and a prediction layer. The prediction module is used to predict the tunnel portal design parameters based on the survey parameters to be predicted using the trained tunnel portal structure prediction model. The training module is further configured to encode the tunnel portal survey parameters through the preprocessing layer and output the target feature vector; construct survey parameter nodes based on the feature vector through the subgraph construction layer and construct a fully connected graph between the survey parameter nodes; perform interactive modeling of the attributes of the fully connected graph through the entity matching layer to obtain fused features; learn the fused features through the graph pooling layer to obtain the target embedding vector; perform fusion and dimensionality reduction based on the target embedding vector through the prediction layer to obtain the prediction result, and perform backpropagation training based on the prediction result to obtain the trained tunnel portal structure prediction model.

8. A knowledge graph-driven tunnel portal structure design device, characterized in that, The knowledge graph-driven tunnel portal structure design device includes: a memory, a processor, and a knowledge graph-driven tunnel portal structure design program stored in the memory and executable on the processor. The knowledge graph-driven tunnel portal structure design program is configured to implement the knowledge graph-driven tunnel portal structure design method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a knowledge graph-driven tunnel portal structure design program, which, when executed by a processor, implements the knowledge graph-driven tunnel portal structure design method as described in any one of claims 1 to 6.

Images