An environmental element information embedded heterogeneous network key node mining method

Abstract

The application discloses a kind of environmental element information embedding heterogeneous network key node mining method, according to cross-layer rotation embedding, multi-modal environmental element embedding and reinforcement learning optimization strategy, the application not only can accurately model the cross-layer relationship of multilayer power network, improve the accuracy of edge prediction task, simultaneously can be combined with reinforcement learning to identify high-risk key nodes, provide scientific basis for disaster response, risk assessment and safety scheduling of smart grid, improve the security, resilience and intelligent level of power grid system.

Description

Technical Field

[0001] This invention relates to the field of power network technology, and in particular to a method for mining key nodes in heterogeneous networks by embedding environmental element information. Background Technology

[0002] Current research on power energy networks largely focuses on the topological analysis of single-layer networks, neglecting the complex interactive characteristics inherent in multi-layer network systems. In multi-layer power networks (such as generation, transmission, and distribution networks), the connections between layers are influenced not only by the internal topology but also by dynamic disturbances from external environmental factors (such as typhoons, fires, and earthquakes). However, existing edge prediction methods primarily rely on static modeling, considering only the fixed structural characteristics of single-layer networks. This makes it difficult to effectively model cross-layer relationships in multi-layer networks, and the prediction accuracy is low under natural disaster conditions, failing to meet the security requirements of complex power systems. Summary of the Invention

[0003] To address the limitations and shortcomings of existing technologies, this invention provides a method for mining key nodes in heterogeneous networks by embedding environmental element information, comprising:

[0004] Obtain information on environmental factors and power grid status;

[0005] According to the dynamic embedding mechanism, the environmental element information and the power network status information are embedded as node attributes into the multi-layer network model, and the multi-layer network model adaptively adjusts the embedding representation according to the dynamic neural network and the environmental element information.

[0006] By mapping cross-layer edges to rotation operations in complex space, multiple patterns of cross-layer relationships are captured, enhancing the ability to model node associations between different network levels. These multiple patterns include symmetry, antisymmetry, inverse, and combinatoriality.

[0007] The ability to model cross-layer relationships is evaluated using the edge prediction task. A key node mining strategy is constructed based on the reinforcement learning framework to identify key nodes whose importance to the stability of the power network under natural disasters is greater than or equal to a preset value.

[0008] Optionally, the step of constructing a key node mining strategy based on a reinforcement learning framework to identify key nodes whose importance value to power network stability under natural disasters is greater than or equal to a preset value includes:

[0009] Construct a node importance evaluation mechanism based on deep reinforcement learning;

[0010] The importance of nodes to the stability of the power network under natural disasters is assessed based on the aforementioned node importance assessment mechanism.

[0011] Nodes whose importance value is greater than or equal to a preset value are identified as key nodes.

[0012] Optional, also includes:

[0013] Graph convolutional neural networks are used to extract features from the topology of single-layer networks and learn the node connection patterns within each layer.

[0014] Based on the environmental state information and the node connection pattern, the multi-layer network model's ability to dynamically perceive the structure within each layer is enhanced, providing feature representations for subsequent key node mining.

[0015] Optionally, the environmental element information includes wind speed, temperature, humidity, precipitation, air pressure, earthquake magnitude, tidal amplitude, and thermal radiation intensity, and the power network status information includes voltage and power.

[0016] Optionally, the step of adaptively adjusting the embedding representation includes:

[0017] An environment-aware dynamic embedding mechanism is introduced to adaptively adjust the node embedding representation according to changes in the external environment. The expression for this adaptive adjustment is as follows:

[0018] z′ i =z i +β·g(R i )

[0019] Among them, z i For the initial embedding of node i; g(R) i ) is R i A nonlinear mapping function, R, is used to adjust the embedding vector; i β is the natural disaster risk score for node i; β is the adjustment coefficient.

[0020] Optionally, the step of mapping cross-layer edges to rotation operations in complex space includes:

[0021] We use rotation operations over the complex field to model the embedding of entities and relations. The expression for the rotation operation is: Among them, h u Here, r is the embedding of node u, and r is the embedding vector of the relation, used to model cross-layer relationships between nodes across layers. It represents the product of each variable over the complex field.

[0022] The present invention has the following beneficial effects:

[0023] This invention provides a method for mining key nodes in heterogeneous networks by embedding environmental element information. Based on cross-layer rotation embedding, multimodal environmental element embedding, and reinforcement learning optimization strategies, this invention can not only accurately model the cross-layer relationships of multi-layer power networks and improve the accuracy of edge prediction tasks, but also identify high-risk key nodes by combining reinforcement learning. This provides a scientific basis for disaster response, risk assessment, and safe scheduling of smart grids, thereby improving the safety, resilience, and intelligence level of the power grid system. Attached Figure Description

[0024] Figure 1 This is an architecture diagram of a heterogeneous network key node mining method for embedding environmental element information, as provided in Embodiment 1 of the present invention.

[0025] Figure 2 This is a schematic diagram of the unlayered network framework structure and its abstract layered structure provided in Embodiment 1 of the present invention.

[0026] Figure 3 This is an attribute environment feature architecture diagram provided in Embodiment 1 of the present invention.

[0027] Figure 4 This is a schematic diagram of the intra-layer relationship modeling provided in Embodiment 1 of the present invention.

[0028] Figure 5 This is a schematic diagram of cross-layer relationship modeling provided in Embodiment 1 of the present invention.

[0029] Figure 6 This is a schematic diagram of the key node mining method provided in Embodiment 1 of the present invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the technical solution of the present invention, the following describes in detail the method for mining key nodes in heterogeneous networks by embedding environmental element information provided by the present invention, with reference to the accompanying drawings.

[0031] Example 1

[0032] This embodiment provides a method for key node mining in multimodal, multi-layer power networks based on cross-layer rotation embedding. The method includes: constructing a multi-layer power network model and analyzing the interaction characteristics between generation, transmission, and distribution networks; mapping cross-layer edges to rotation operations in complex space to capture multiple patterns of cross-layer relationships and enhance node association modeling capabilities; and introducing a dynamic environmental element embedding mechanism, embedding real-time environmental conditions such as wind speed, humidity, and temperature as node attributes into the network, and achieving adaptive model adjustment through a dynamic neural network. To verify the model's effectiveness, the cross-layer relationship modeling capability is first evaluated using an edge prediction task. Then, reinforcement learning methods are combined to optimize the key node mining strategy and identify key nodes affecting power network stability under severe weather or sudden disaster conditions. This method provides efficient decision support for disaster response and fault repair in smart grids, improving the security and risk management capabilities of power networks.

[0033] This embodiment proposes a multimodal key node mining method based on cross-layer rotation embedding. This method, building upon multi-layer power network modeling, incorporates edge prediction as an effectiveness verification mechanism and further introduces reinforcement learning to optimize the key node mining strategy. Specifically, in the relationship modeling of multi-layer networks, cross-layer edges are mapped to rotation operations in complex space, thereby capturing various patterns of cross-layer relationships, such as symmetry, antisymmetry, inverseness, and combinatoriality, enhancing the ability to model node associations between different network levels.

[0034] To adapt to complex and dynamic environments, this embodiment further introduces a dynamic embedding mechanism for environmental elements. Real-time environmental conditions such as wind speed, humidity, and temperature are embedded as node attributes into the network model. Combined with a dynamic neural network, this enables the model to adaptively adjust its embedding representation and inference strategy based on changes in the external environment. For example, during a typhoon, the model can identify the potential impact of wind speed on power transmission lines and dynamically optimize its prediction strategy. In the event of emergencies such as fires or earthquakes, the model can focus on the connectivity of key nodes in the affected area, thereby enhancing the model's environmental adaptability.

[0035] To further optimize the resilience analysis and disaster response capabilities of smart grids, this embodiment employs a reinforcement learning framework and optimizes the key node discovery strategy based on the effectiveness verification results of the edge prediction task. By constructing a node importance assessment mechanism based on deep reinforcement learning, the model can identify nodes crucial to grid stability under natural disasters and provide targeted control suggestions. For example, the model can identify key nodes that have the most significant impact on power supply during typhoons or earthquakes, thereby providing efficient decision support for proactive grid defense and post-disaster recovery.

[0036] This embodiment proposes a method for mining key nodes in power networks under complex and variable environments by combining cross-layer rotation embedding, multimodal environmental element embedding, and reinforcement learning optimization strategies. This method can not only accurately model the cross-layer relationships of multi-layer power networks and improve the accuracy of edge prediction tasks, but also identify high-risk key nodes using reinforcement learning. This provides a scientific basis for disaster response, risk assessment, and safe scheduling in smart grids, thereby improving the safety, resilience, and intelligence level of the power grid system.

[0037] This embodiment, based on dynamic embedding technology, incorporates environmental factors such as typhoons and fires, along with power network status information such as voltage and power, into a multimodal graph convolutional neural network embedding framework. Building upon accurate and robust prediction of cross-layer and intra-layer power network connections, it further combines reinforcement learning to optimize the key node mining strategy. This method comprises three main parts:

[0038] The attribute environment feature processing module dynamically embeds environmental elements (such as wind speed, humidity, and temperature) and power network status information (such as voltage and power) as node attributes into the multi-layer network model. This ensures that the model can adapt to complex dynamic environments and improves the robustness of the prediction task by optimizing the embedding representation through an adaptive mechanism.

[0039] Intra-layer relation modeling module: Graph Convolution Neural Network (GCNN) is used to extract features of the topology of a single-layer network, learn the node connection patterns within each layer, and combine environmental state information to enhance the model's ability to dynamically perceive the intra-layer structure, providing high-quality feature representations for subsequent key node mining.

[0040] The cross-layer relationship modeling and critical node mining module models cross-layer edge relationships in multi-layer power networks, mapping cross-layer interactions to rotation operations in complex space to capture various patterns of cross-layer relationships. It then incorporates reinforcement learning to optimize the critical node mining strategy. Based on the effectiveness verification of the edge prediction task, a critical node selection strategy is constructed using a reinforcement learning framework to identify the critical nodes with the greatest impact on power grid stability under natural disasters, and to generate optimized control schemes to improve the power grid's risk management capabilities and disaster response efficiency.

[0041] Figure 1 This is an architecture diagram of a heterogeneous network key node mining method based on environmental element information embedding provided in Embodiment 1 of the present invention. This embodiment identifies key nodes affecting power grid stability under natural disasters by fusing cross-layer rotation embedding, multimodal environmental feature embedding, and reinforcement learning optimization strategies, thereby improving the power grid's disaster resistance and fault recovery efficiency.

[0042] In terms of data: a multi-layered heterogeneous energy network in a real-world scenario was constructed, involving the collection and preprocessing of its node characteristic data.

[0043] Figure 2 This is a schematic diagram of the unlayered network framework structure and its abstract layered structure provided in Embodiment 1 of the present invention.

[0044] I. Attribute Environment Feature Processing Module Flowchart

[0045] Figure 3 This is an attribute-environment feature architecture diagram provided in Embodiment 1 of the present invention. In the modeling process of multi-layer power networks, the data collection, preprocessing, and fusion of physical and environmental attributes are crucial steps in model construction, and their quality directly affects the model's generalization ability and computational efficiency. To enhance the effectiveness of multimodal feature representation and improve the performance of edge prediction and key node mining by combining dynamic embedding technology, this embodiment proposes a systematic data processing flow. This flow covers the standardized transformation of physical attributes, the structured expression of spatial information, and the dynamic modeling of environmental attributes to ensure that the model can achieve high-precision inference in complex power network environments. This embodiment proposes the following standardized data processing flow:

[0046] 1. Physical property processing

[0047] 1.1 Discrete Attribute One-Hot Encoding

[0048] Discrete attributes in power networks (such as equipment type and voltage level) have a finite but discontinuous range of values. Traditional numerical assignment methods may introduce meaningless numerical relationships. Therefore, one-hot encoding is used for their numerical transformation. Specifically, assuming there are n categories of equipment types, for equipment belonging to category...

[0049] For device i, its one-hot encoding is represented as:

[0050] OneHot(i) = [0,…,1,…,0]

[0051] In this encoding method, only the i-th dimension has a value of 1, while the other dimensions are all 0. This encoding method avoids unreasonable ordinal relationships between categorical variables and enhances the model's ability to distinguish between different categories, enabling it to effectively learn the differences in the roles of power equipment in a multi-layer network structure.

[0052] 1.2 Normalization of Continuous Attributes

[0053] For continuous physical attributes such as transformer capacity and population density in the service area, the Min-Max Normalization method is used to eliminate the dimensional differences between different attributes, thereby improving the convergence of the model during optimization. The normalization transformation formula is as follows:

[0054]

[0055] Numerical features are mapped to the interval [0,1] to ensure that different attributes participate in model calculation at the same scale, while avoiding the impact of unbalanced attribute numerical ranges on the stability of model weight learning.

[0056] 1.3 Convert latitude and longitude to distance

[0057] The spatial distribution characteristics of energy networks are crucial for network edge prediction and key node identification. To avoid the impact of the non-Euclidean nature of the raw latitude and longitude data on model calculations, a great-circle distance calculation method is adopted. Using the central substation or designated anchor point of the power network as a reference, the geographical location information of each node is converted into a spherical distance between it and the anchor point. The calculation formula is as follows: The great-circle distance formula is used for calculation:

[0058]

[0059] Where r is the Earth's radius (approximately 6371 km); φ and λ are latitude and longitude, respectively (in radians). Subsequently, to ensure that the distance characteristics maintain a consistent numerical scale with other physical properties, the calculated inter-node distances are normalized:

[0060]

[0061] 2. Environmental Attribute Processing

[0062] To adapt to the impact of various disasters (such as typhoons, earthquakes, fires, and floods) on power grids, a general disaster early warning model is designed. Based on multi-dimensional environmental attributes, the disaster risk score R of nodes is calculated. i It is used to dynamically adjust the embedded representation of the model.

[0063] In complex power network systems, natural disasters (such as fires and floods) have a profound impact on the dynamic changes in network topology. Therefore, this embodiment proposes an environment-aware dynamic embedding modeling method, which calculates the disaster risk score (R) of nodes based on multi-dimensional environmental attributes. i The model dynamically adjusts the embedding representation of nodes based on this, thereby enhancing its adaptability to external disturbances.

[0064] 2.1 Dynamic Modeling of Environmental Attributes

[0065] Assume the environmental attributes include n influencing factors: wind speed (v), humidity (h), temperature (T), seismic intensity (e), etc. The environmental attribute vector of node i is represented as:

[0066] E i =[v i ,h i ,T i ,e i ,p f,i ,…]

[0067] Where p f,i This indicates the probability or intensity of node i being affected by a specific disaster (such as a typhoon or fire). This information can be obtained through historical statistics or real-time monitoring data.

[0068] 2.2 Disaster Risk Scoring Function

[0069] Based on both environmental and node characteristics, a disaster risk score for each node is defined.

[0070]

[0071] Where: E i,j f is the feature value of the environment attribute j of node i; j (E i,j α is the normalization or characteristic function of attribute j; j Here, represents the weighting parameter, indicating the contribution of different attributes to the risk score. Common feature function example: Wind speed: humidity: earthquake intensity:

[0072] 2.3 Dynamic Embedding Adjustment

[0073] Combining disaster risk score R i Furthermore, an environment-aware dynamic embedding mechanism is introduced, enabling the node embedding representation to adaptively adjust according to changes in the external environment. The adjustment formula is as follows:

[0074] z′ i =z i +β·g(R i )

[0075] Where z i Initial embedding of nodes; g(R) i Risk score R i A nonlinear mapping function is used to adjust the embedding vector; β is the adjustment coefficient. This mechanism ensures that the model can update the node representation in real time under natural disasters, improving its prediction accuracy under disaster conditions.

[0076] II. Intra-layer Relationship Modeling

[0077] Figure 4 This is a schematic diagram of intra-layer relationship modeling provided in Embodiment 1 of the present invention. Graph Neural Networks (GNNs) can learn node representations by aggregating information from neighbors, thereby capturing potential connection patterns, which is beneficial for link prediction tasks. Multi-layer networks provide valuable information for enhancing link prediction. Therefore, our method involves using GNNs in multi-layer networks, which simultaneously learn cross-network embeddings, integrating intra-layer and inter-layer structural features.

[0078] Various GNNs have been developed for node representations. Generally, a typical GNN layer looks like this:

[0079]

[0080]

[0081] in The output vector (representation) of node u at the k-th layer; The representation of node u in the previous level (k-1); The weight matrix adjusts the contribution of the node's own representation and the neighboring node representations to the update, respectively. Neighbors of node u; a uv θ: Attention weights of the edges between u and v; θ: Activation function used to introduce a non-linear transformation (e.g., ReLU), indicating that the node representation in the k-th layer is calculated by combining its own representation and the representations of its neighbors.

[0082] III. Cross-layer Relationship Modeling

[0083] Figure 5 This is a schematic diagram of cross-layer relationship modeling provided in Embodiment 1 of the present invention. Aggregating messages from other layers helps to supplement node information, thereby enhancing the understanding of connection patterns. To enhance node representation learning in cross-layer networks, we compute cross-layer anchor embeddings based on a rotation-based modeling method, while supporting the aggregation of multiple anchors. By utilizing the rotation embedding mechanism, we can effectively capture cross-layer relationships between nodes and combine a weighting mechanism to perform weighted fusion of information from multiple anchors. While maintaining the original intra-layer neighbor structure information, we further optimize the integration of cross-layer information, making the embedding representation more accurate and semantic.

[0084] RotatE is a knowledge graph embedding method that uses rotation operations over complex domains to model the embedding of entities and relations. Its formula is: Where h uLet be the embedding of node u, and r be the embedding vector of the relation, modeling cross-layer relationships between nodes across layers. This represents a product of elements over the complex field. The node embeddings are typically represented as complex numbers, and rotation relationships are modeled using complex numbers with a modulus of 1 (ensuring that rotations do not change the modulus).

[0085] There are multiple anchor points for other layers. Assume that each node u corresponds to a set of anchor points. Each anchor point They all have their embedded And the corresponding relationship r uu′ The cross-GNN that integrates inter-layer information is as follows:

[0086]

[0087] This involves: multi-anchor point aggregation, rotation-embedded updates, and preserving original structural information (within layers). For all anchor points... Perform a weighted summation, with weights w. uu′ Used to measure the importance of node u and its cross-layer anchor point u′. u This represents the original feature vector of node u, which is usually a fixed input feature. It is a linear transformation matrix used to combine node feature information. Through... To aggregate; Re(·) is used to take the real part of the complex number, ensuring the embedding result is in the real domain. The original formula is based on the real domain, while the rotation operation is based on the complex domain. By taking the real part Re(·), the result in the complex domain can be mapped back to the real domain to maintain compatibility with the original node representation. Using the rotation formula To calculate the embedding of cross-layer anchor points, where r uu′ It is the rotation vector relating u and u′. The first and second terms... Used to preserve the structure and neighbor information within the node layer.

[0088] The objective function includes intra-layer loss and inter-layer loss, where intra-layer loss primarily preserves intra-layer structural features, and inter-layer loss preserves inter-layer structural features. The total loss is then jointly optimized to better unify the multiple networks. If two nodes frequently appear together in a fixed-length random walk, it indicates a higher probability that a link exists between them. Therefore, the random walk captures higher-order proximity and provides insights into intra-network connection patterns and potential links. Thus, to learn the embedding z... i , Applying a random walk-based objective function in an unsupervised setting:

[0089]

[0090] Where j is the node that co-occurs with i in the random walk sequence window, σ is the sigmoid function, and P n (v) represents the negative sampling probability distribution, and Q defines the number of negative samples. Neighboring nodes encourage similar embeddings, while discrete nodes are distinct in the embedding space. The intra-layer loss is the sum of the intra-layer losses based on random walks: Where n is the number of network layers, the intra-layer structural features of the network can be preserved by minimizing the intra-layer loss.

[0091] When designing the cross-layer loss function, a rotation embedding mechanism is introduced into cross-layer modeling to more accurately capture the cross-layer relationships between nodes. Specifically, based on the rotation embedding of nodes and the complex rotation operation of relation vectors, the updated loss function improves the accuracy of cross-layer alignment by calculating the cosine similarity of cross-layer node embeddings. In the positive sample loss, the semantic consistency between cross-layer anchors is ensured by combining the rotated embedding alignment method; while in the negative sample part, the problem of sample imbalance is effectively alleviated by screening and optimizing hard negative samples. The formula is as follows:

[0092]

[0093] For each node and its cross-layer anchor point set have negative sample S - The choice comes from The set of neighbors ensures that the sampling process is consistent with the distribution of anchor points, thereby improving the representativeness of hard-to-bear samples. Represents a node The embedding is performed through the relation vector r ij After rotation, the real part is used to calculate the result. The similarity. For each pair of negative samples Similarly, cosine similarity based on rotation embedding is calculated. Hard negative samples with similarity exceeding a certain threshold ∈ are filtered using max(cos-∈,0). This cross-layer loss function design fully combines the characteristics of cross-layer networks and the advantages of rotation embedding, providing more robust theoretical support and practical guidance for node alignment tasks in multi-layer networks.

[0094] The overall objective function is defined as a linear combination of intra-layer and inter-layer losses; therefore, we can preserve intra-layer and inter-layer structural information through joint optimization.

[0095] L=α·L intra +(1-α)·L inter

[0096] Here, α is the weight parameter that balances the two components of the objective function, and the Adam optimizer is used to update the parameters of the GNN layers. The node embedding update of each layer depends on the embedding of its previous layer, the embedding of its neighboring nodes, and cross-layer information. Specifically, cross-layer embedding models the relationship between cross-layer nodes through rotation operations, making the cross-layer nodes more accurate in the embedding space. To perform edge prediction, the edge embedding of each pair of nodes (x, y) needs to be calculated first. The edge embedding is constructed by the embedding vectors of nodes x and y. Logistic Regression is used to predict the probability of edge existence, which is then used to learn the relationship between edge embedding features and edge existence. By learning the relationship between the input edge embedding and the label (existence of the edge), we predict P(x~y), which is the probability that there is an edge between a given node pair (x,y).

[0097] IV. Key Node Mining Based on Graph Convolutional Neural Networks

[0098] Figure 6 This is a schematic diagram of the key node mining method provided in Embodiment 1 of the present invention. It employs a combination of graph convolutional neural networks (GMMs) and reinforcement learning, using an end-to-end optimization algorithm to transform the key node selection problem into a decision-making process for the agent. First, the network is encoded into a low-dimensional space using a GMM algorithm. The size of the largest connected group in the network is defined as the reinforcement learning reward. Then, at the decoding end, a reinforcement learning DQN architecture is used, cross-multiplying the node state vector with a predefined action vector, and using a multilayer perceptron to generate a Q-value function. The Q-value represents the importance of possible edges between node pairs for network robustness. Considering the agent's interaction with the environment, actions are taken and feedback rewards are obtained. The node vector representation is updated using a graph neural network as the next state. Finally, a large number of small-scale multilayer networks are generated based on the GMM multilayer network generation model to train the reinforcement learning model and ultimately achieve accurate key node finding.

[0099] This study investigates the impact of virtual natural disasters on network nodes, examining early warning signals of network collapse and quantifying the changes in network resilience caused by these nodes. First, a percolation model is used to define the network node ΔN that needs to collapse to cause a network crash. Then, the importance of node i, predicted by a reinforcement learning agent, and v are used to further analyze this. i v represents the degree of damage to network robustness caused by removing the node. i The calculations are derived from the reinforcement learning agent trained in the previous stage. S0 represents the critical value for network collapse, and S0 represents the set of nodes actually removed from the multi-layer network. Ω represents the estimated degree of network damage after removing a set of nodes S. s =∑ n∈s pn Finally, the network's warning value Ω is expressed by the following formula, 0≤Ω m ≤1 is used to quantify the level of damage a network can tolerate before it collapses.

[0100]

[0101] In network topology optimization, three-dimensional tensors are used to store constraint relationships based on the type of constraint and the type of interaction with the environment. For example, geographical reachability between different nodes is represented by binary vectors, and the construction cost of network lines is represented by normalized values. Subsequently, the node representation vectors are multiplied bitwise to obtain the vector representation of the relationships between nodes. The change in connected groups after removing nodes and breaking corresponding edges is used as the reward function for reinforcement learning. A reinforcement learning DQN architecture consistent with the aforementioned key nodes is adopted, using a large number of small-scale multilayer networks instantiated from the GMM multilayer network generation model as the training dataset. Using the good model and value evaluation function of the representation reinforcement learning model, the importance score Q of the edges between any pair of nodes is calculated. By adjusting the weights of the hyperparameters, the model is combined with the actual situation of the network system to calculate the weighted resilience contribution value of node pairs. Based on this value, node pairs are ranked from high to low. The node with the largest weighted resilience contribution value is the key node discovered in the network.

[0102] This embodiment proposes a method for edge prediction and key node mining in multimodal, multi-layer power networks based on cross-layer rotation embedding. This method addresses the shortcomings of traditional network edge prediction and node mining methods that only focus on single-layer network topology and neglect the interaction characteristics of multi-layer networks and the influence of external environmental factors. It constructs a multi-layer power network model and deeply analyzes the interaction characteristics between generation, transmission, and distribution networks. Cross-layer edges are mapped to rotation operations in complex space, effectively capturing multiple patterns of cross-layer relationships and significantly enhancing the node association modeling capability. Simultaneously, a dynamic environmental factor embedding mechanism is introduced, embedding real-time environmental states such as wind speed, humidity, and temperature as node attributes into the network. A dynamic neural network enables adaptive adjustment of the model, allowing it to comprehensively consider attribute factors and achieve key node mining under severe weather or sudden disaster conditions.

[0103] At the technical implementation level, this method consists of an attribute-environment feature processing module, an intra-layer relationship modeling module, and a cross-layer relationship modeling and key node mining module. The attribute-environment feature processing module uses one-hot encoding and normalization to preprocess the physical attributes (such as voltage level and transformer capacity) and environmental attributes (such as wind speed, humidity, and temperature) of the power grid. It then constructs a disaster early warning model based on multi-dimensional environmental factors, calculates the disaster risk score of nodes, and dynamically adjusts the node embedding representation. The intra-layer relationship modeling module employs a graph convolutional neural network, learning node representations by aggregating neighbor information, thereby capturing intra-layer topological features and potential connection patterns, improving the accuracy of local edge prediction. The cross-layer relationship modeling and key node mining module calculates cross-layer anchor point embeddings based on a rotation embedding method, models cross-layer edge relationships using rotation transformation, and optimizes multi-anchor point information fusion through a weighting mechanism to improve cross-layer information integration capabilities. Furthermore, an intra-layer-inter-layer joint optimization objective function is constructed, and the structural consistency and predictability of node representations are optimized through random walks, cosine similarity, and a rotation embedding strategy. Finally, a reinforcement learning framework is introduced to optimize the key node mining strategy based on the effectiveness verification of edge prediction, identify key nodes that affect the stability of the power grid under natural disasters, and thus improve the disaster resistance and fault recovery efficiency of the power grid.

[0104] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for mining key nodes in heterogeneous networks by embedding environmental element information, characterized in that, include: Obtain information on environmental factors and power grid status; According to the dynamic embedding mechanism, the environmental element information and the power network status information are embedded as node attributes into the multi-layer network model, and the multi-layer network model adaptively adjusts the embedding representation according to the dynamic neural network and the environmental element information. By mapping cross-layer edges to rotation operations in complex space, multiple patterns of cross-layer relationships are captured, enhancing the ability to model node associations between different network levels. These multiple patterns include symmetry, antisymmetry, inverse, and combinatoriality. The ability to model cross-layer relationships is evaluated by using the edge prediction task. A key node mining strategy is constructed based on the reinforcement learning framework to identify key nodes whose importance value to the stability of the power network under natural disasters is greater than or equal to a preset value. The step of adaptively adjusting the embedded representation includes: An environment-aware dynamic embedding mechanism is introduced to adaptively adjust the node embedding representation according to changes in the external environment. The expression for this adaptive adjustment is as follows: ， in, For nodes i The initial embedding; for A nonlinear mapping function is used to adjust the embedding vector; For nodes i Natural disaster risk score; This is for adjusting the coefficient.

2. The method for mining key nodes in heterogeneous networks by embedding environmental element information according to claim 1, characterized in that, The step of constructing a key node mining strategy based on a reinforcement learning framework to identify key nodes whose importance value to power network stability is greater than or equal to a preset value under natural disasters includes: Construct a node importance evaluation mechanism based on deep reinforcement learning; The importance of nodes to the stability of the power network under natural disasters is assessed based on the aforementioned node importance assessment mechanism. Nodes whose importance value is greater than or equal to a preset value are identified as key nodes.

3. The method for mining key nodes in heterogeneous networks by embedding environmental element information according to claim 2, characterized in that, Also includes: Graph convolutional neural networks are used to extract features from the topology of single-layer networks and learn the node connection patterns within each layer. Based on the environmental element information and the node connection pattern, the multi-layer network model's ability to dynamically perceive the structure within each layer is enhanced, providing feature representations for subsequent key node mining.

4. The method for mining key nodes in heterogeneous networks by embedding environmental element information according to claim 3, characterized in that, The environmental information includes wind speed, temperature, humidity, precipitation, air pressure, earthquake magnitude, tidal amplitude, and thermal radiation intensity. The power network status information includes voltage and power.

5. The method for mining key nodes in heterogeneous networks by embedding environmental element information according to claim 1, characterized in that, The step of mapping cross-layer edges to rotation operations in complex space includes: We use rotation operations over the complex field to model the embedding of entities and relations. The expression for the rotation operation is: ,in, It is a node Embedded, It is the embedding vector of the relationship, used to model cross-level relationships between nodes across different levels. It represents the product of each variable over the complex field.

Images