Graph Structure Reconstruction Method, Device, Storage Medium and Program Product

Abstract

The present application discloses a graph structure reconstruction method, device, storage medium, and program product. The method includes: determining hyperbolic embedding vectors corresponding to each graph node in an initial graph structure to be processed, and parsing target prior knowledge that semantically matches each hyperbolic embedding vector; for each hyperbolic embedding vector, extracting a layer-level representation corresponding to the hyperbolic embedding vector, and fusing the corresponding target prior knowledge to obtain a graph node fusion feature; inputting each graph node fusion feature into a reinforcement learning model to determine a corresponding target graph reconstruction action, and performing deletion and modification operations on the edge connections between graph nodes in the initial graph structure according to the target graph reconstruction action to obtain a corresponding target reconstructed graph structure. Thus, it is possible to utilize the topological representation characteristics of the graph and the adaptive ability of reinforcement learning to reconstruct a graph structure that meets the actual task requirements.

Description

Technical Field

[0001] This application relates to the technical field of data analysis, and particularly to a method, device, storage medium and program product for graph structure reconstruction. Background Art

[0002] Graph structures are widely used in various data processing scenarios, such as social network analysis, transportation networks, material molecular science, and communication networks. In these applications, graph structures are usually used to represent the relationships or interaction behaviors between entities, where nodes represent entities and edges represent the relationships or connections between nodes, making graph structures an indispensable part of data analysis and machine learning.

[0003] For example, in the social network scenario, nodes represent users and edges represent the friendship relationships between users, and a huge social network graph can be constructed. By analyzing the social network graph, the interests and hobbies of users can be accurately predicted, and personalized content recommendations can be provided for them, thus significantly improving the user experience and user stickiness of the platform. In the transportation field, nodes represent transportation hubs, etc., and edges represent road connections, and a traffic network topology graph can be constructed. Using this graph structure, the traffic flow distribution can be effectively optimized, travel routes can be reasonably planned, the transportation efficiency can be improved, and the traffic congestion can be alleviated.

[0004] However, the complexity and scale of graph structure data are increasing day by day, and traditional graph reconstruction methods are difficult to cope with these challenges. Most traditional methods are based on geometric properties in Euclidean space and use graph adjacency matrices or distance matrices to construct the relationships between nodes and edges. This method can achieve good results when dealing with small-scale and relatively regular graph data, but when facing large-scale, sparse and high-dimensional complex graph structures, its performance is limited. Especially in graph data with a hierarchical structure, the embedding method in Euclidean space cannot fully express the distances and relationships between graph nodes, which leads to problems such as loss of key information and low reconstruction accuracy in the graph structure reconstruction process.

[0005] In response to the above problems, the industry has not yet proposed a better solution. Summary of the Invention

[0006] This application provides a method, device, storage medium and program product for graph structure reconstruction, so as to at least solve the problem that the embedding method in the traditional Euclidean space has great limitations in the processing of complex graph data.

[0007] In a first aspect, an embodiment of the present application provides a method for graph structure reconstruction, including: determining hyperbolic embedding vectors corresponding to each graph node in an initial graph structure to be processed, and parsing target prior knowledge that semantically matches each of the hyperbolic embedding vectors; for each of the hyperbolic embedding vectors, extracting a layer-level representation corresponding to the hyperbolic embedding vector, and fusing the corresponding target prior knowledge to obtain a graph node fusion feature; inputting each of the graph node fusion features into a reinforcement learning model to determine a corresponding target graph reconstruction action, and performing a deletion or modification operation on the edge connections between graph nodes in the initial graph structure according to the target graph reconstruction action to obtain a corresponding target reconstructed graph structure.

[0008] In a second aspect, an embodiment of the present application provides an electronic device, which includes: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the steps of the graph structure reconstruction method according to any embodiment of the present application.

[0009] In a third aspect, an embodiment of the present application provides a storage medium, on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the graph structure reconstruction method according to any embodiment of the present application are implemented.

[0010] In a fourth aspect, an embodiment of the present application provides a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the graph structure reconstruction method according to any embodiment of the present application are implemented.

[0011] The beneficial effects of the embodiments of the present application are as follows:

[0012] By using hyperbolic embedding vectors, it is possible to more accurately represent graph nodes in a non-Euclidean space and better capture the complex non-linear relationships between nodes in graph data with a hierarchical structure. In addition, by fusing target prior knowledge, the construction of graph node features is not only based on the characteristics of the nodes themselves, but also can combine external domain semantic information. Furthermore, by introducing a reinforcement learning model to automatically delete or modify the connections between graph nodes, based on the training and feedback mechanism of reinforcement learning, it is possible to select graph reconstruction actions corresponding to the optimal graph reconstruction strategy and improve the graph reconstruction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0014] Figure 1 Shows a flowchart of an example of the graph structure reconstruction method according to an embodiment of the present application;

[0015] Figure 2 Shows an operation flowchart of an example of determining the hyperbolic embedding vectors of each graph node according to an embodiment of the present application;

[0016] Figure 3 Shows a schematic structural connection diagram of an example of a reinforcement learning model according to an embodiment of the present application;

[0017] Figure 4 Shows a schematic diagram of the effect of an example of the chemical variable relationship graph structure according to an embodiment of the present application;

[0018] Figure 5 Shows a flowchart of an example of the chemical variable relationship graph reconstruction method based on graph reinforcement learning provided by an embodiment of the present application;

[0019] Figure 6 Shows a schematic structural diagram of an example of the graph attribute-knowledge tree according to an embodiment of the present application;

[0020] Figure 7 Shows a schematic diagram of the effect of an example of the corpus of graph structure features and prior knowledge according to an embodiment of the present application;

[0021] Figure 8 Shows a visualization diagram of an example of the prior knowledge weight vector;

[0022] Figure 9 Shows a schematic architecture diagram of an example of graph reinforcement learning according to an embodiment of the present application;

[0023] Figure 10 Shows a schematic framework diagram of an example of the chemical variable relationship graph reconstruction according to an embodiment of the present application;

[0024] Figure 11 Shows a schematic diagram of the experimental simulation effect comparison of an example of the identification result of system variable 1 according to an embodiment of the present application;

[0025] Figure 12 Shows a schematic diagram of the experimental simulation effect comparison of an example of the identification result of system variable 2 according to an embodiment of the present application;

[0026] Figure 13 Is a schematic structural diagram of an embodiment of the electronic device of the present application. Detailed implementation manners

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are some, but not all, of the embodiments of this application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of this application without creative efforts shall fall within the protection scope of this application.

[0028] It should be noted that Graph Neural Networks (GNNs) have made breakthrough progress in the field of graph data analysis. Graph neural networks achieve feature learning of graph nodes and edges by iteratively aggregating the feature information of nodes and their neighbor nodes. However, existing graph neural networks usually learn based on a fixed graph structure and lack flexibility. When the structure of the graph changes dynamically, these models are difficult to adapt to the new structure. The Reinforcement Learning (RL) mechanism is introduced to enhance the learning ability for dynamic graphs. Reinforcement learning can continuously optimize and adjust the graph structure through interaction with the environment. Nevertheless, existing reinforcement learning methods usually only focus on the optimization of the graph topology and ignore the semantic information represented by prior knowledge in the graph. Therefore, there are still certain limitations in dealing with complex graph structures.

[0029] Figure 1 The flowchart of an example of the graph structure reconstruction method according to the embodiments of this application is shown.

[0030] As Figure 1 shown, in step S110, the hyperbolic embedding vectors corresponding to each graph node in the initial graph structure to be processed are determined, and the target prior knowledge that semantically matches each hyperbolic embedding vector is parsed.

[0031] In some embodiments, for the graph structure to be processed, each graph node will be mapped to a low-dimensional hyperbolic space, which is completed by a hyperbolic geometric embedding algorithm. A vector of each graph node in the hyperbolic space is mapped, aiming to capture the complex non-linear relationships between nodes. Compared with the traditional Euclidean space, the hyperbolic space can better express the hierarchy and topology between nodes. It should be understood that the types of hyperbolic spaces can be diverse, such as the Poincaré ball model, negative curvature surface, or Lorentz model, etc., which are not limited here for the time being.

[0032] In addition, according to the specific business application scenarios of the graph structure (such as social networks, transportation networks, etc.), target prior knowledge is introduced, such as external knowledge like user behavior and traffic flow, and it is semantically matched with the embedding vectors of each graph node. Through semantic matching, it can help enhance the expressive ability of the graph node embedding vectors, further enrich the semantic information of the nodes, enabling the representation of the nodes to not only consider the structure of the graph itself but also incorporate external information, thereby improving the accuracy and practicality of the graph node representation.

[0033] In step S120, for each hyperbolic embedding vector, the layer-level representation corresponding to the hyperbolic embedding vector is extracted, and the corresponding target prior knowledge is fused to obtain the graph node fusion feature.

[0034] In some embodiments, for the hyperbolic embedding vector of each graph node, the hierarchical structure information of its position in the graph is extracted. For example, through a Graph Convolutional Network (GCN), a Graph Attention Network (GAT), or other graph-based deep learning methods, the local and global hierarchical representations of the graph node in the graph structure are extracted. Thus, the hierarchical representation of the graph node is extracted, making each node not only limited to its own characteristics but also considering its position in the overall structure of the graph, effectively improving the richness of the node features.

[0035] Furthermore, the target prior knowledge is fused with the hierarchical features of the graph nodes, and it can be completed through simple weighted fusion, concatenation, or more complex attention mechanisms, so that the features of the graph nodes can not only reflect the graph structure but also express richer semantic information.

[0036] In step S130, each graph node fusion feature is input into a reinforcement learning model to determine the corresponding target graph reconstruction action, and the edge connections between the graph nodes in the initial graph structure are deleted or modified according to the target graph reconstruction action to obtain the corresponding target reconstructed graph structure.

[0037] In some embodiments, the reinforcement learning model decides the graph reconstruction operation by evaluating the fusion features of each graph node to optimize the connectivity of the graph structure, making the relationships between the nodes in the graph clearer and enhancing the practical application effect of the graph structure expression. It should be understood that the architecture of the reinforcement learning model can be diverse, and algorithms based on policy gradients, such as PPO (Proximal Policy Optimization) or DQN (Deep Q-Network), can be used to learn how to delete or modify the edge connections between graph nodes.

[0038] It should be noted that the reinforcement learning model makes graph reconstruction action decisions based on the current graph structure state, generates target graph reconstruction actions, and the reconstruction actions include operations such as deleting unimportant edges, adding missing edges, or adjusting the weights of edges. Through continuous iterative training, the model can learn the optimal graph reconstruction strategy, and then modify the edge connections in the initial graph structure according to the actions generated by the reinforcement learning model, so as to obtain the target reconstructed graph structure, which can better express the relationships between nodes and improve the accuracy of graph analysis.

[0039] Through the embodiments of the present application, the reinforcement learning model automatically selects the optimal graph reconstruction strategy according to the characteristics of the graph structure, making the graph structure reconstruction process more intelligent and dynamic. Compared with the traditional manually designed graph reconstruction method, the model can adapt to the characteristics of different graph data and provide a more accurate reconstruction effect. In addition, optimizing the edge connections through action decisions can reduce redundant edges and irrelevant information, improve the accuracy and interpretability of the graph structure, which is particularly important for large-scale, sparse, and high-dimensional graph data, and can ensure that the graph structure used in graph analysis tasks (such as node classification, graph matching, etc.) is more efficient and accurate.

[0040] In the hybrid graph reconstruction method based on hyperbolic space and semantic graph reinforcement learning provided by the embodiments of the present application, the graph data is embedded into the hyper-surface space, and the geometric characteristics of the hyper-surface space are used to enhance the representation ability of complex graph structures. At the same time, the semantic information of the extracted graph is combined to construct a semantic graph, and the geometric graph and the semantic graph are jointly optimized through the graph reinforcement learning framework, comprehensively considering the geometric distance and semantic association between graph nodes, and significantly improving the accuracy and efficiency of the graph reconstruction task.

[0041] Figure 2 The operation flowchart showing an example of determining the hyperbolic embedding vectors of each graph node according to the embodiments of the present application is shown.

[0042] As Figure 2 shown, in step S210, the initial graph embedding space corresponding to the initial graph structure in the Euclidean space is determined.

[0043] In some embodiments, it is assumed that the initial graph structure to be processed is , where is the node set, is the edge set. Each graph node Both contain a node feature vector, whose features can be the basic attributes of the node, historical behaviors, or other semantic information. Furthermore, existing graph embedding algorithms (such as DeepWalk, Node2Vec, etc.) can be used to map the graph structure into a Euclidean space, and each graph node is embedded into a low-dimensional vector space in the Euclidean space, such as 128 or 256 dimensions, so as to obtain an initial graph embedding space representation, making the relationships between nodes mainly measured based on the Euclidean distance.

[0044] In step S220, based on the Poincaré ball model, the node features of each graph node in the initial graph embedding space are mapped to corresponding points in the Poincaré space, and the corresponding first hyperbolic embedding vectors are obtained through a learnable linear transformation layer.

[0045] In some embodiments, the Poincaré ball model is a non-Euclidean space model, which can effectively represent the non-linear relationships in high-dimensional, sparse, and complex graph structures. In this space, the embedding space of node features is transformed into a hyperbolic geometric space, which can capture the complex dependencies between graph nodes.

[0046] Specifically, by using the mapping function of the Poincaré ball model, the Euclidean embedding vectors of each graph node are mapped to corresponding points in the Poincaré space, ensuring that the structural relationships between nodes are better expressed in the hyperbolic space. Especially for graph data with a hierarchical structure, the Poincaré space can effectively retain this hierarchical information. Furthermore, through a learnable linear transformation layer, the mapping of each graph node is adjusted to generate more reasonable hyperbolic embedding vectors.

[0047] As a further preferred embodiment, in order to implement the node aggregation and update operation of the graph structure.

[0048] In step S230, each of the first hyperbolic embedding vectors is remapped back to the Euclidean space, and graph node aggregation is performed through a graph attention algorithm to obtain the updated graph nodes.

[0049] It should be noted that since directly performing the graph node aggregation operation in the Poincaré hyperbolic space will introduce additional computational overhead, and most graph operations (such as convolution, etc.) are defined in the Euclidean space, before performing information aggregation, the hyperbolic embedding vectors of each graph node will be remapped back to the Euclidean space, which ensures that the subsequent aggregation operations can utilize the computational advantages of the Euclidean space and reduce the computational overhead.

[0050] In Euclidean space, node feature aggregation is performed through the graph attention algorithm. By using the graph attention mechanism to assign dynamically learnable attention weights to each edge, the updated feature of each node is the weighted sum of the features of its neighbor nodes, enabling each node to adaptively select meaningful information from its neighbor nodes for aggregation, thereby improving the quality and efficiency of node feature update.

[0051] In step S240, each updated graph node is remapped back to the Poincaré hyperbolic space and undergoes a non-linear transformation through an activation function to obtain the corresponding second hyperbolic embedding vector.

[0052] In some embodiments, using the mapping function of hyperbolic geometry, the new feature vector of each graph node is converted into a new vector in the Poincaré space, ensuring that the geometric properties of the graph structure are fully preserved. To further enhance the expressive power of node features, a non-linear activation function (such as ReLU, tanh, etc.) is used to transform the feature vector remapped to the Poincaré space, introducing more non-linear factors into the features of the nodes, making the representation of node features more abundant, and thus enabling the capture of more complex relationships and patterns.

[0053] Through the embodiments of the present application, by first mapping node features to Euclidean space for information aggregation, then remapping them back to the Poincaré hyperbolic space and performing non-linear transformation, the computational advantages of Euclidean space and the superior representation ability of the Poincaré hyperbolic space for the complexity of the graph structure can be effectively combined. Thereby, both the efficiency of graph node aggregation is ensured, and the expressive power of node features is improved, enabling more accurate capture of the non-linear and hierarchical relationships between nodes in the graph, and thus enhancing the overall graph embedding and task performance.

[0054] Regarding the specific operation details of parsing the target prior knowledge that semantically matches the hyperbolic embedding vector, in some examples of the embodiments of the present application, for each hyperbolic embedding vector, the semantic similarity between the hyperbolic embedding vector and each prior knowledge weight word vector in the graph corpus is calculated to screen the matching target prior knowledge.

[0055] Here, the prior knowledge weight word vector is a word vector containing learned weights converted from the graph attribute-knowledge tree, and the graph attribute-knowledge tree is constructed based on the fusion of the task attribute tree and the graph structure training samples.

[0056] Regarding the construction details of the graph attribute - knowledge tree, which can be high - order attributes and hierarchical information in the extraction task, representing them as nodes or branches in the tree to obtain the task attribute tree. Then, in the context of the graph - structured training samples, high - order attributes related to the task are extracted. These attributes may include the hierarchical structure between nodes, the functional relationships of nodes, the similarity between nodes, or other meaningful task - related information, and may also involve the upstream - downstream relationships of nodes, the clustering information of nodes, etc. Then, the extracted high - order attributes are integrated with the hierarchical information of the nodes and represented as a tree structure in the graph. The root node in this tree structure represents the central node of the high - order attributes in the graph, and each original graph node is regarded as a leaf node of the tree. The branch part of the tree represents the membership relationship of the nodes in the layer - level structure.

[0057] The graph attribute - knowledge tree constructs a tree structure containing multi - level information by fusing the structural information of the graph and the high - order attributes of the task (or downstream task). The secondary root nodes are divided into a structure layer and a node layer according to the position and level of the nodes in the graph. The structure layer represents the order of the nodes relative to the central node, that is, the level or distance of the nodes in the graph structure in the graph. The node layer contains the original nodes representing the actual variables in the graph, which are directly related to the original data in the graph. In this way, the original graph nodes of the graph - structured training samples are divided into different categories according to their high - order features in the layer - level, and new nodes and connections are automatically generated to further supplement the hierarchical information in the original graph. Thus, the graph attribute - knowledge tree not only retains the structural information of the original graph but also embeds the task attributes in a hierarchical manner, enabling the effective integration of the contextual semantic information of the task into the graph and improving the expressive power of the graph structure.

[0058] In addition, through the graph attribute - knowledge tree, the task attributes in the graph are mapped into word vectors with learning weights. These weighted word vectors reflect the hierarchical relationships between nodes and the importance of each node in the task in the graph. Each prior - knowledge word vector is optimized through learning to ensure that they are closely related to the structure of the graph and the task attributes. By calculating the semantic similarity between the hyperbolic embedding vectors and the prior - knowledge weighted word vectors, it is determined which prior knowledge matches the features of the current graph node, for example, by calculating the cosine similarity or Manhattan distance, to filter out the prior knowledge with higher similarity, which can effectively enhance the representation ability of the graph nodes, not only providing additional semantic support for the graph nodes but also helping the graph model better understand the roles and meanings of each graph node in the graph structure.

[0059] Through the embodiments of the present application, calculating the similarity between the hyperbolic embedding vectors and the prior knowledge weighted word vectors can accurately match the high-order attributes of nodes semantically. Through semantic matching, each graph node can not only utilize its own structural information but also combine external prior knowledge, further enhancing the intelligence and task relevance of graph data. Thus, by effectively screening and integrating prior knowledge, the node features can be dynamically adjusted according to the requirements of the task, enabling the model to automatically introduce the most relevant context information when processing graph data, thereby improving the accuracy and efficiency of task execution.

[0060] The graph attribute-knowledge tree constructed by combining the graph structure and the task attribute tree can process and represent the high-order information of graph nodes more precisely. By introducing prior knowledge that semantically matches the node features, it not only provides rich context support for graph nodes but also improves the accuracy of node feature representation and the relevance of tasks, enhancing the intelligence and flexibility of graph structure analysis and contributing to improving the depth analysis and reasoning accuracy of complex graph data.

[0061] Figure 3 The structural connection diagram of an example of the reinforcement learning model according to the embodiments of the present application is shown.

[0062] As Figure 3 shown, the reinforcement learning model adopts a graph reinforcement learning agent. The reinforcement learning model 300 includes a GAT layer 310 and a fully connected layer 320.

[0063] Specifically, based on the GAT layer 310, feature extraction is performed on the fused features of each graph node, and a heuristic graph reconstruction action group is output through the fully connected layer 320.

[0064] It should be understood that the number of GAT layers 310 can be one or more. The features of each graph node and its adjacent nodes are weighted and aggregated through the self-attention mechanism. The fused feature vector of the graph node is passed as input to the GAT layer for further processing. The GAT layer uses the attention mechanism for feature aggregation according to the relationship between each node and its neighbors, and the local structure and neighbor information of the node in the graph will be effectively fused, enhancing the context expression ability of the node features.

[0065] In the fully connected layer 320, the node features are further processed through a multi-layer perceptron or linear transformation, mapping the features of the graph nodes to a higher-dimensional space, and generating a heuristic graph reconstruction action group through the learned heuristic policy. These actions include deletion and modification operations of the edges between nodes to guide the optimization of the graph structure under the action of different action strategies.

[0066] Successively execute the deletion and modification operations of the edges connected to each heuristic graph reconstruction action in the heuristic graph reconstruction action group to obtain the corresponding candidate reconstructed graph structures respectively.

[0067] Exemplarily, based on the graph reconstruction action, when the connection between nodes is considered no longer meaningful, the edge is deleted. Additionally, when the model discovers that the relationship between certain nodes is underestimated or missed, new edges are added to connect them. After each execution of a graph reconstruction action, a new candidate graph structure is generated, which represents a possible way of graph reconstruction and aims to optimize the connectivity and expressive power of the graph through deletion and modification operations.

[0068] Evaluate each candidate reconstructed graph structure according to the performance metrics of the downstream tasks corresponding to it to determine the corresponding target reconstructed graph structure.

[0069] In some embodiments, the performance metrics of the downstream tasks, such as system identification accuracy or fault diagnosis accuracy, can be used as rewards for graph reconstruction, thereby evaluating the optimization effect of the candidate reconstructed graph structures to select the best reconstructed graph structure. By comparing the task application performances of the candidate graph structures, the optimization process of the reconstructed graph has a clear goal, ensuring that the graph structure is not only optimized in form but also can provide the best support for specific tasks. Thus, through the task-oriented graph structure optimization method, the accuracy and efficiency of graph analysis tasks are improved, supporting applications in complex tasks that require adaptive adjustment of the graph structure.

[0070] Regarding the description of the training details of the reinforcement learning model, an enhanced design based on mutual information is adopted for the loss function of the GAT layer 310. Specifically, the loss function of the GAT layer 310 is determined by maximizing the sum of the mutual information between the graph feature representations of each candidate reconstructed graph structure and the graph feature representation of the global reconstructed graph structure. Here, the global reconstructed graph structure is constructed by fusing the graph feature representations of all candidate reconstructed graph structures. For example, the feature representations of all candidate graph structures are combined (such as through weighted averaging, concatenation, etc.) to obtain a comprehensive global graph feature representation. The global graph structure represents the overall information of all candidate graph structures and provides a global perspective for graph reconstruction.

[0071] It should be noted that mutual information measures the degree of information sharing between the feature representations of two graph structures. Maximizing mutual information helps to improve the similarity between the candidate graph and the global graph, thereby ensuring that each candidate structure can be adjusted towards the goal of global optimization during the graph reconstruction process. Thus, each candidate graph not only improves the quality of the local structure during the optimization process but also keeps consistent with the global structure, avoiding overfitting to local data and improving the accuracy of the reconstruction action.

[0072] It should be noted that in the chemical process industry, chemical processes often involve multiple variables, including controlled variables, manipulated variables, and other measurable variables involved in the process. These variables each represent different physical states, such as temperature, pressure, flow rate, etc., and there are complex coupling relationships between them. It is impossible to accurately describe the relationships between these chemical variables relying solely on mechanism relationships or manual experience, which causes difficulties for relevant industry personnel to understand and optimize the corresponding chemical processes, and also poses challenges to tasks such as system identification and fault diagnosis of chemical processes using graph learning. However, unfortunately, the construction of chemical variable graphs (for example, chemical knowledge graphs) in the chemical field often relies heavily on expert experience at present, lacks generalization, and is prone to subjective omissions.

[0073] In some business application scenarios combining the embodiments of the present application, the initial graph structure is a chemical variable relationship graph structure, the graph nodes of which are defined by controlled variables, manipulated variables or measurable variables, and the edge connections in the chemical variable relationship graph structure are defined by the mechanism relationships between variables.

[0074] Figure 4 The effect schematic diagram of an example of the chemical variable relationship graph structure according to the embodiment of the present application is shown.

[0075] As Figure 4 shown, the chemical process is characterized by a graph from the graph perspective and decomposed into nodes and edges. Specifically, by analyzing the mechanism model of the chemical process, the controlled variables involved (such as the inlet condensate temperature and the condenser vacuum), the manipulated variables (such as the water pump frequency and the fan frequency), and the measurable variables (such as the condenser inlet and outlet pressures) are correspondingly characterized as nodes in the chemical variable relationship graph, and the connections between these nodes can be initialized according to their original mechanism relationships.

[0076] Through the chemical variable relationship graph reconstruction method based on hyperbolic space and semantic graph reinforcement learning provided by the embodiments of the present application, the graph data is embedded into the hyperbolic space, and the geometric characteristics of the hyperbolic space are used to enhance the representation ability of complex graph structures. At the same time, the semantic information of the graph is extracted by combining the word2vec algorithm, the chemical variable relationship graph is automatically jointly optimized through the graph reinforcement learning framework, and it is evaluated according to the performance of downstream tasks.

[0077] Specifically, through the method of reconstructing the chemical process variable relationship graph based on graph reinforcement learning, the problem of exploring variable relationships in the chemical process field is effectively solved to find the complex coupling relationships between chemical variables, provide knowledge reference, and improve the performance of downstream tasks. The exemplary operations are as follows: chemical process characterization based on graphs; graph embedding of node representations in hyperbolic space; constructing a graph attribute-knowledge tree to enrich the representation of the original graph; using the word2vec algorithm for semantic embedding to retain important feature information in the graph and integrate prior knowledge; integrating into a semantic graph reinforcement learning framework for graph reconstruction. Thus, by utilizing the topological representation characteristics of the graph and the adaptive ability of reinforcement learning, a graph structure that conforms to the coupling relationship of the actual chemical process can be reconstructed, providing a reference for industry personnel to understand the corresponding process and improving the performance of graph tasks.

[0078] Figure 5 FIG. 4 shows a flowchart of an example of the method for reconstructing a chemical process variable relationship graph based on graph reinforcement learning provided by an embodiment of the present application.

[0079] In an embodiment of the present application, a method for reconstructing a chemical process variable relationship graph based on hyperbolic space and semantic graph reinforcement learning is provided. By embedding the topological features of the graph into a hyper-surface space and combining with the semantic information of the graph, a graph reinforcement learning framework is used to jointly optimize the topological graph and the semantic graph. This method can provide higher reconstruction accuracy and downstream task efficiency in the graph reconstruction task of large-scale complex graphs.

[0080] As Figure 5 shown in the method for reconstructing a chemical process variable relationship graph based on graph reinforcement learning, it aims to solve the problem of exploring variable relationships in the chemical process field to find the complex coupling relationships between chemical variables, provide knowledge reference, and improve the performance of downstream tasks.

[0081] It should be noted that the graph reinforcement learning framework constructed based on the embodiment of the present application aims to provide solutions for graph construction tasks in different application scenarios. To prove the effectiveness of the present application, a system identification experiment was conducted based on real cooling tower process data and used as an embodiment. It should be noted that the description of this part of the embodiment in combination with the application scenario of the cooling tower process is only used as an example and should not be regarded as a limitation on the implementation scope of the embodiment of the present application. And downstream system identification tasks or other chemical processes, as well as downstream graph task applications such as fault diagnosis and fault tracing, all fall within the implementation scope of the embodiment of the present application.

[0082] Aiming at the problem of complex chemical variable coupling relationships in the chemical process industry, a method for reconstructing a chemical process variable relationship graph based on hyperbolic space and semantic graph reinforcement learning is proposed. The specific process framework can be described as follows:

[0083] In step S1, the chemical process is graphically characterized from a graph perspective and decomposed into nodes and edges. The chemical process is abstracted in the form of a graph, where nodes represent various variables in the chemical process, such as temperature, pressure, etc., and edges represent the potential relationships between these variables. The characteristics of each node are constructed by collecting relevant physical properties, historical data, and on-site observations to form a feature vector. The initial node connection relationship can be based on expert experience or randomly initialized to form a basic chemical variable relationship graph.

[0084] Specifically, step S1 specifically includes the following steps:

[0085] S11: Analyze the chemical process from a graph perspective and characterize it as , list the relevant variables involved, including controlled variables, manipulated variables, and other measurable variables, and use them as nodes in the characterized chemical variable relationship graph .

[0086] S12: Use the physical properties, historical data, on-site observations, etc. of each variable as node characteristics and characterize them as feature vectors .

[0087] S13: According to artificial experience or randomly initialize the connections between nodes , as the adjacency matrix of the chemical variable relationship graph , and form the original chemical variable relationship graph .

[0088] In step S2, the constructed original graph data is embedded into the hyperbolic space to generate hyperbolic embedding vectors for each node.

[0089] Specifically, a hyperbolic space graph embedding algorithm is adopted. This algorithm uses a hyperbolic geometry model to embed the original Euclidean graph into the hyperbolic space and utilizes the characteristics of hyperbolic geometry to enhance the expression ability for complex graph structures. Through specific mathematical transformations, such as the Poincaré ball model, the nodes are transformed from the Euclidean space to the hyperbolic space to generate hyperbolic embedding vectors. Further, through linear transformations, the hyperbolic embedding vectors are adjusted to ensure that they have a suitable geometric structure and better reflect the hierarchical relationships between nodes. And it includes the following steps:

[0090] S21: Construct an initial graph embedding space in the Euclidean space and initialize the node vectors in the Euclidean space , indicating that this vector is used to characterize a certain node;

[0091] S22: Select the Poincaré ball as the hyperbolic space model and use the exponential mapping formula of Poincaré embedding to map each node to the corresponding point in the Poincaré space 。Initial vector is converted into a hyperbolic vector :

[0092] Equation (1)

[0093] where is Möbius addition, is a parameter that can be set manually, represents the origin in hyperbolic space, characterizes the initial vector of the i-th node v, characterizes the representation vector of the i-th node v in hyperbolic space H.

[0094] S23: Through a learnable linear transformation layer, ensure that the mapped hyperbolic space embedding has an appropriate hyperspherical geometric structure.

[0095] Equation (2)

[0096] where, is Möbius multiplication, represents the transformation matrix, represents the bias, represents the transformed representation vector of the i-th node v in hyperbolic space H.

[0097] As a further preferred embodiment, since the mathematical steps of aggregating adjacent node features in a graph network are inherently complex and operations such as convolution are defined in Euclidean space, directly performing these operations in the Poincaré hyperbolic space will incur additional computational overhead. Therefore, before information aggregation, the node feature vectors are remapped back to the Euclidean space.

[0098] Equation (3)

[0099] where is Möbius addition, is a parameter that can be set manually, represents the origin in hyperbolic space, represents the mapping reference point in hyperbolic space, characterizes the representation vector of the i-th node v after being remapped back to the Euclidean space τ.

[0100] After the aggregation is completed according to the steps of the GAT algorithm, the new features of each node are remapped back to the Poincaré hyperbolic space and subjected to a non-linear transformation through an activation function.

[0101] Equation (4)

[0102] where, represents a transformation matrix, represents a non-linear activation function, represents the attention coefficient between node i and node j in the GAT, represents node i, represents the neighbor node j of node i, represents the set of all neighbor nodes of node i, characterizes the representation vector of the i-th node in the Euclidean space τ at the l-th GAT layer, characterizes the representation vector of the i-th node in the hyperbolic space H at the (l + 1)-th GAT layer.

[0103] In step S3, a graph attribute-knowledge tree is constructed, combining the tree structure with the graph structure to retain the high-order information of the graph.

[0104] Here, the graph attribute-knowledge tree is used to represent the high-order information and prior knowledge in the task to supplement the content of the graph representation. To enrich the graph representation, a graph attribute-knowledge tree is constructed to integrate the high-order information and prior knowledge of the graph. The tree structure classifies the nodes in the graph hierarchically, forming a new hierarchical structure that contains the high-order features and prior knowledge of the nodes. In this way, not only the structural information of the original graph is retained, but also additional semantic information is introduced, enhancing the expressive power of the graph. Specifically, it includes the following steps:

[0105] S31: Extract the high-order attributes and hierarchical information in the task and represent them as nodes or branches in the tree; use the layer-level information of the nodes as the new root nodes, and the original graph nodes as the new leaf nodes. The new branches represent the membership relationship of the nodes at the layer level.

[0106] S32: Construct a graph attribute-knowledge tree, combining the tree structure with the graph structure to retain the high-order information of the graph; the secondary root nodes are divided into two levels: the structure layer and the node layer. The structure layer includes the node order relative to the central node, and the underlying node layer contains the original nodes representing variables in the graph. In this way, the original nodes are classified into different categories according to the high-order features at the layer level, and new nodes and connections are generated during this process as a supplement to the layer-level information of the original graph.

[0107] S33: Since the tree structure has excellent scalability, new attribute nodes containing prior knowledge such as "important" and "unimportant" are additionally generated in the structure layer or the node layer as a new knowledge layer to flexibly integrate the prior knowledge into the newly generated tree to enrich the content of the tree.

[0108] Specifically, a method for constructing a graph attribute-knowledge tree from the perspective of high-order graph features such as the original graph structure and node attributes is adopted. Figure 6A structural schematic diagram showing an example of a graph property-knowledge tree according to an embodiment of the present application is shown.

[0109] As Figure 6 shown, a knowledge tree is constructed according to the properties and structure of the graph, combining the nodes and edges in the graph with the tree structure to retain the high-order information in the graph. The original nodes are divided into different categories according to the high-order features of the layer level, and new nodes and connections are generated in this process as a supplement to the layer level information of the original graph. In addition, due to the excellent scalability of the tree structure, the method additionally generates new attribute nodes containing prior knowledge such as "important" and "unimportant" as new knowledge layers at the graph or node layer to flexibly integrate prior knowledge into the newly generated tree to enrich the content of the tree.

[0110] In step S4, a corpus containing graph structure features and prior knowledge is constructed and semantic embedding is performed using the word2vec algorithm to integrate prior knowledge.

[0111] Here, the semantic graph construction adopts the word2vec algorithm to extract the semantic information of the local neighborhood nodes in the graph, convert the graph property-knowledge tree into a text form, and construct a corpus. By training this corpus, vectors that can reflect the semantic information of the nodes are generated, and these vectors can capture the semantic associations between the nodes. Calculate the similarity between the node vectors and the prior knowledge vectors to generate weight vectors for use in the subsequent graph reinforcement learning process. Specifically, it includes the following steps:

[0112] S41: The constructed graph property-knowledge tree is converted into text knowledge; by constructing a graph corpus, which includes words reflecting graph structure features and prior knowledge, including the nodes of the original graph, connection information, hierarchical information in the tree, the actual physical meanings of nodes, edges and node attributes, and the learning weights representing prior knowledge.

[0113] S42: Based on the converted text knowledge, digitize it into vectors, and use the word2vec algorithm to train the corpus so that the distribution of the word vectors in the semantic space reflects prior knowledge.

[0114] S43: Calculate the word vectors of each node and the word vectors of the prior knowledge to extract the learning focus. Convert it into a weight vector, integrate it into the layer level features, and participate in the training of the subsequent graph reinforcement learning.

[0115] Equation (5)

[0116] In the formula, represents the word vector of a single node, represents the word vector of the prior knowledge, represents the cosine similarity.

[0117] Here, the text information in the corpus is vectorized, and these terms are mapped to the semantic space in the form of word vectors to form a vocabulary list as shown Figure 7 for distinction. Figure 7 FIG. 8 shows an effect schematic diagram of an example of a corpus of graph structure features and prior knowledge according to an embodiment of the present application. During the process of semantic embedding, a corpus including graph structure features and prior knowledge and a corresponding vocabulary list are constructed. Subsequently, the agent needs to understand the meanings of words such as "node", "level", and "importance", which requires these terms to be distributed in the semantic space according to their original semantics.

[0118] In this method, the training of the graph corpus is combined with the word2vec algorithm for semantic embedding, and the Skipgram model is used. This model is trained by predicting the surrounding context words of a given center word. During the training process, a parameterizable backpropagation (BP) neural network is used to learn word vectors.

[0119] Assume a constructed vocabulary list in which each word has a unique index. For each center word, the goal is to maximize the conditional probability of the context words. Specifically, for a given center word and its context words, the training objective of the corpus is to maximize. The specific method is to perform a dot product on the vector of the center word and the vector of the context word, and then pass the result to the SoftMax function to obtain the conditional probability. The definition of this function is as follows:

[0120] Equation (6)

[0121] where is the center word, is the center word vector, and is the context word vector, and t represents the index interval from the center word.

[0122] Equation (7)

[0123] where is the word vector of the context word, t represents the upper bound of the index, is the average word vector of the context words.

[0124] However, the computational cost of the SoftMax function increases with the increase in the vocabulary size. Therefore, in actual training, negative sampling is introduced to accelerate the training process. The goal is to maximize the probability of the correct context words while minimizing the likelihood of randomly sampled incorrect context words. Specifically, for each center word and its correct context word, their inner product is maximized, and the inner product between the center word and the incorrect context word is minimized. The objective function of negative sampling is defined as follows:

[0125] Equation (8)

[0126] where is the word vector of a certain center word ; represents the word vector of the context word of the center word ; represents the word vector of the incorrect context word of the center word ; represents the library of incorrect context words of the center word ; represents the non-linear activation function.

[0127] For the entire graph corpus, the loss function can be defined as:

[0128] Equation (9)

[0129] where is the set of all center words in the corpus, represents an adjustable coefficient between (0, 1).

[0130] After completing the corpus learning, the arrangement of the original word vectors in the new embedding space not only reflects the original structure of the graph but also embodies the guiding role of prior knowledge.

[0131] The training process of word embedding is independent of the subsequent graph reinforcement learning training process. The feature vectors learned in this step reflect the high-order information and prior knowledge of the graph and do not participate in the subsequent information aggregation in the graph network, thus avoiding the over-smoothing phenomenon that may occur due to subsequent graph reinforcement learning. Figure 8 shows a visualization diagram of an example of the prior knowledge weight vector. As Figure 8 shown, the similarity reflecting the importance of each node will form a prior knowledge weight matrix, which is added to the subsequent transformation from node representation to graph representation.

[0132] In step S5, the original graph structure is iteratively optimized according to the reward function of graph reinforcement learning, and the connection relationship between nodes is gradually adjusted to obtain a reconstructed graph for further applications.

[0133] Here, the topological and semantic properties of the graph structure are comprehensively considered, so that the reconstructed graph can satisfy semantic associations while retaining topological characteristics. The reinforcement learning policy network adopts the DDQN (Double Deep Q-Network) algorithm to guide graph reconstruction by maximizing the reward function of downstream tasks.

[0134] Specifically, it includes the following steps:

[0135] S51: Use the improved GAT and DDQN algorithms as the basic algorithms for graph reinforcement learning; the input of graph reinforcement learning is the original features of each node and the adjacency matrix . These features are then converted into hyperbolic vectors through the hyperbolic embedding in step S2, and then feature extraction is performed through the GAT layer.

[0136] S52: Expand the node-level representation of the graph network into a layer-level representation through a flattening layer ; add prior knowledge weights to the layer-level representation to fuse semantic knowledge and make the node representation richer;

[0137] Equation (10)

[0138] where is a transformation matrix, represents the flattening operation, represents the concatenation operation of vectors, represents the node-level representation vector of the i-th node, represents the prior knowledge weight vector of the i-th node.

[0139] S53: Convert the extracted layer-level representation into an action probability vector through the FC layer, which represents the probability that the agent executes the actions in the designed action rule library. The graph reconstruction actions executed by the agent are reflected as the deletion and modification of the edges between nodes on the variable graph. After an action cycle of the agent, the reconstructed graph is obtained and evaluated according to the performance of the downstream task, so as to train the agent.

[0140] S54: After the agent is trained, it can be generalized to other chemical processes to autonomously reconstruct the chemical variable relationship graph and find out the complex coupling relationships between variables.

[0141] Through the graph reinforcement learning framework, using the results of hyperbolic embedding and semantic embedding, the original graph structure is optimized. The improved GAT and DDQN algorithms are used to intelligently adjust the connection relationships between nodes according to the preset reward function. Through iterative optimization, a graph structure that more conforms to the actual chemical process is gradually reconstructed, improving the understanding and prediction ability of the chemical process.

[0142] Figure 9 Shows a schematic diagram of the architecture of an example of graph reinforcement learning according to an embodiment of the present application.

[0143] As Figure 9 shown, the main body of the graph reinforcement learning agent consists of a GAT. Based on the improved graph attention network and the double-layer deep Q-network algorithm, a graph reinforcement learning framework is constructed to perform the graph reconstruction task. The last layer of the graph reinforcement learning is a fully connected layer (FC layer), and the extracted graph features are converted into an action probability vector through this layer. The action vector corresponds to a set of heuristic graph reconstruction actions, which are reflected in the modification of the edges of the graph. After completing the action, a new graph is generated, which is a step of graph reconstruction. After generating the new graph, it is evaluated using the reward function R, the loss function L is calculated, and the policy network in the graph reinforcement learning is updated. The loss of the GAT layer is designed to maximize the sum of the mutual information between all local and global graph features in the global graph:

[0144] Equation (11)

[0145] where represents the graph feature representation of the i-th single graph, represents the graph feature representation of the j-th step of the entire reconstruction process, represents the mutual information calculation.

[0146] Figure 10 Shows a schematic diagram of the framework of an example of the reconstruction of the chemical variable relationship graph according to an embodiment of the present application.

[0147] As Figure 10 shown, the graph reconstruction framework includes two main parts: one is the GAT-FC network as the graph reinforcement learning agent, which outputs actions to reconstruct the variable graph; the other is the graph-recurrent neural network for downstream tasks, and its task performance is used as the reward of the reinforcement learning. When the downstream tasks of the graph reconstruction are different, the reward R of the graph reinforcement learning will also be adjusted accordingly. The results of the downstream tasks, such as the system identification accuracy or the fault diagnosis accuracy, are used as the reward R of the graph reconstruction.

[0148] In the process of the experiment combining the embodiments of the present application, the evaluation index of whether the graph reconstruction is reasonable is the accuracy of using this graph for system identification. The higher the identification accuracy, the more the reconstructed graph can reflect the complex coupling relationship between the process variables. Therefore, before training the graph reinforcement learning agent, the network used for the downstream task is pre-trained first to eliminate the influence of the untrained downstream task network on the graph reinforcement learning process, so that the performance difference in the graph reinforcement learning training process is completely determined by the quality of the graph reconstruction.

[0149] Figure 11A schematic diagram comparing the experimental simulation effects showing an example of the identification result of system variable 1 according to an embodiment of the present application. Figure 12 A schematic diagram comparing the experimental simulation effects showing an example of the identification result of system variable 2 according to an embodiment of the present application.

[0150] Specifically, Figure 11 The identification results of the downstream system identification task using the initialized chemical variable relationship diagram and the chemical variable relationship diagram after graph reconstruction are shown. It can be seen from the figure that the performance of system identification using the reconstructed graph is significantly better than that using the initial graph. The reconstructed graph is further applied to the system identification of another target control variable (vacuum degree), and the identification results are compared with those using the initial graph. The comparison results are as Figure 12 shown. The experimental results show that the result of graph reconstruction has a certain universality in this process. It is not only the best solution completely driven by the data of a specific downstream task, but also reflects to a certain extent the relationship between chemical variables in the actual chemical process.

[0151] In the embodiments of the present application, by integrating graph representation learning, graph reinforcement learning, hyperbolic space embedding, and semantic embedding technologies, a hybrid graph reconstruction method based on hyperbolic space and semantic graph reinforcement learning is implemented, providing a general graph reinforcement learning framework to solve the graph reconstruction problem of complex problems, especially suitable for the graph structure optimization and reconstruction of processing large-scale, sparse, and high-dimensional data. Thus, the topological characterization characteristics of the graph and the adaptive ability of reinforcement learning can be utilized to reconstruct a graph structure that conforms to the coupling relationship of the actual chemical process, providing a reference for industry personnel to understand the corresponding process and improving the graph task performance.

[0152] It should be noted that this study used publicly available human subject data for retrospective analysis. Since the data used is open access and the attached license confirms that ethical approval is not required, this study does not require ethical approval.

[0153] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of actions combined. However, those skilled in the art should know that the present application is not limited by the described action sequence, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present application. In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0154] In some embodiments, the embodiments of the present application provide a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute any one of the above-mentioned graph structure reconstruction methods of the present application.

[0155] In some embodiments, the embodiments of the present application further provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer is enabled to execute any one of the above-mentioned graph structure reconstruction methods.

[0156] In some embodiments, the embodiments of the present application further provide an electronic device, which includes: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the graph structure reconstruction method.

[0157] Figure 13 is a schematic hardware structure diagram of an electronic device for executing the graph structure reconstruction method provided by another embodiment of the present application, as Figure 13 shown, the device includes:

[0158] One or more processors 1310 and a memory 1320, Figure 13 Taking one processor 1310 as an example.

[0159] The device for executing the graph structure reconstruction method may further include: an input device 1330 and an output device 1340.

[0160] The processor 1310, the memory 1320, the input device 1330, and the output device 1340 may be connected by a bus or other means, Figure 13 Taking the connection by a bus as an example.

[0161] The memory 1320, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the graph structure reconstruction method in the embodiments of the present application. The processor 1310 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 1320, that is, implements the graph structure reconstruction method in the above method embodiments.

[0162] The memory 1320 may include a program storage area and a data storage area. The program storage area may store an operating system and application programs required for at least one function. The data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 1320 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 1320 may optionally include a memory remotely disposed relative to the processor 1310, and these remote memories may be connected to the electronic device through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0163] The input device 1330 may receive input digital or character information, and generate signals related to user settings and function control of the electronic device. The output device 1340 may include a display device such as a display screen.

[0164] The one or more modules are stored in the memory 1320 and, when executed by the one or more processors 1310, execute the graph structure reconstruction method in any of the above method embodiments.

[0165] The above product may execute the method provided in the embodiments of the present application, and has function modules and beneficial effects corresponding to the execution of the method. Technical details not described in detail in this embodiment may be referred to the method provided in the embodiments of the present application.

[0166] The electronic device in the embodiments of the present application exists in various forms, including but not limited to:

[0167] (1) Mobile communication devices: These devices are characterized by having mobile communication functions and mainly aim to provide voice and data communication. Such terminals include: smart phones, multimedia phones, functional phones, and low-end phones, etc.

[0168] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have the characteristic of mobile Internet access. Such terminals include: PDAs, MIDs, and UMPC devices, etc.

[0169] (3) Portable entertainment devices: These devices can display and play multimedia content. Such devices include: audio and video players, handheld game consoles, e-books, and smart toys and portable vehicle navigation devices.

[0170] (4) Other on-board electronic devices with data interaction functions, such as in-vehicle device installed on a vehicle.

[0171] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0172] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. A graph structure reconstruction method, comprising: Determine the hyperbolic embedding vector corresponding to each graph node in the initial graph structure to be processed, and parse the target prior knowledge that semantically matches each of the hyperbolic embedding vectors; For each of the hyperbolic embedding vectors, extract the graph level representation corresponding to the hyperbolic embedding vector, and fuse the corresponding target prior knowledge to obtain the graph node fusion feature; Inputting the fusion features of each of the graph nodes into a reinforcement learning model to determine a corresponding target graph reconstruction action, and performing deletion and modification operations on the edge connections between the graph nodes in the initial graph structure according to the target graph reconstruction action to obtain a corresponding target reconstructed graph structure; The step of analyzing the semantically matching target prior knowledge of each of the hyperbolic embedding vectors includes: For each of the hyperbolic embedding vectors, the semantic similarity between the hyperbolic embedding vector and each prior knowledge weight word vector in the graph corpus is calculated to screen the matching target prior knowledge; the prior knowledge weight word vector is a word vector containing learning weights converted from a graph attribute-knowledge tree, and the graph attribute-knowledge tree is constructed based on the fusion of the task attribute tree and the graph structure training sample; Wherein, the initial graph structure is a chemical variable relationship graph structure; The graph nodes in the chemical variable relationship graph structure are defined by controlled variables, manipulated variables or measurable variables, and the edge connections in the chemical variable relationship graph structure are defined by the mechanism relationship between variables.

2. The method according to claim 1, wherein: The step of determining the hyperbolic embedding vector corresponding to each graph node in the initial graph structure to be processed includes: Determine the initial graph embedding space corresponding to the initial graph structure in the Euclidean space; Based on the Poincare sphere model, the node features of each graph node in the initial graph embedding space are mapped to corresponding points in the Poincare space, and a corresponding first hyperbolic embedding vector is obtained through a learnable linear transformation layer.

3. The method according to claim 2, after mapping the node features of each graph node in the initial graph embedding space to the corresponding point in the Poincare space based on the Poincare sphere model and obtaining the corresponding first hyperbolic embedding vector through a learnable linear transformation layer, the method further comprises: Re-mapping each of the first hyperbolic embedding vectors back to the Euclidean space, and performing graph node aggregation through a graph attention algorithm to obtain updated graph nodes; Each updated graph node is remapped back to the Poincare hyperbolic space and nonlinearly transformed through an activation function to obtain the corresponding second hyperbolic embedding vector.

4. The method according to claim 1, wherein: The reinforcement learning model includes a GAT layer and a fully connected layer. The step of inputting the fusion features of each of the graph nodes into a reinforcement learning model to determine a corresponding target graph reconstruction action, and performing deletion and modification operations on the edge connections between the graph nodes in the initial graph structure according to the target graph reconstruction action to obtain a corresponding target reconstructed graph structure includes: Based on the GAT layer, feature extraction is performed on the fusion features of each graph node, and a heuristic graph reconstruction action group is output through the fully connected layer; Sequentially executing the deletion and modification operations of the edge connections corresponding to each of the heuristic graph reconstruction actions in the heuristic graph reconstruction action group to obtain corresponding candidate reconstructed graph structures respectively; Each of the candidate reconstructed graph structures is evaluated according to the downstream task performance index corresponding to each of the candidate reconstructed graph structures to determine a corresponding target reconstructed graph structure.

5. The method according to claim 4, wherein: Determine the loss function of the GAT layer by maximizing the sum of mutual information between the graph feature representations of each candidate reconstructed graph structure and the graph feature representation of the global reconstructed graph structure; The global reconstructed graph structure is constructed by fusing the graph feature representations of all candidate reconstructed graph structures.

6. A storage medium having a computer program stored thereon, wherein: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 5 are implemented.

7. An electronic device, comprising: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method described in any one of claims 1 to 5.

8. A computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 5.

Images