Electric power communication network path evaluation method and system based on feature fusion

Abstract

The invention discloses an electric power communication network path evaluation method and system based on feature fusion. Comprising the following steps: acquiring communication data of physical nodes in an electric power communication network, preprocessing the communication data, performing feature extraction on the preprocessed communication data, and acquiring time sequence features of the physical nodes; constructing a power communication knowledge graph based on the power communication network topological graph and priori knowledge in the power field; carrying out graph volume accumulation operation on the electric power communication knowledge graph in a hyperbolic space, and extracting structural features of physical nodes; carrying out adaptive fusion on the time sequence features and the structure features, obtaining a feature fusion vector, inputting the feature fusion vector into a risk assessment model for processing, and obtaining a risk prediction result; and combining the risk prediction result with the relationship between the physical nodes in the path to obtain a path evaluation result. The accuracy and comprehensiveness of electric power communication network path evaluation can be improved, and the problem that the existing electric power communication network path evaluation effect needs to be improved is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power communication network data analysis, and in particular to a power communication network path evaluation method and system based on feature fusion. Background Art

[0002] The statements in this section merely mention background art related to the present invention and do not necessarily constitute prior art.

[0003] With the continuous growth of energy demand and the complexity of power systems, the importance of power communication networks has become increasingly prominent. Modern power systems not only need to transmit electric energy reliably, but also need to realize precise monitoring, protection and control functions, all of which cannot be separated from the support of efficient communication networks. The power communication network is a large-scale and well-developed private network in modern dedicated communication networks. With the rapid development of communication technology and the acceleration of the fiber optic process, the data transmission of the power communication network has become more frequent, its structure has become more complex, and the number of services it carries has also increased day by day. As an indispensable part of the power system, the power communication network provides a communication channel for the dispatching, production and management functions of the power system. Once the communication network is paralyzed, the series of chain reactions generated may cause the entire power system to stagnate.

[0004] In order to ensure the stable operation of the power system, it is urgent to carry out risk prediction and transmission efficiency assessment of power grid communication paths for different businesses. Reliability assessment of power communication network transmission paths is one of the key means to ensure the stability of power communications. Through path assessment, potential problems in the network can be discovered, the network structure can be optimized, communication efficiency can be improved, and the risk of failure can be reduced, thereby ensuring the safe and stable operation of the power system. Especially in the face of emergencies such as natural disasters, equipment failures, or cyber attacks, reliable path assessment methods can quickly provide path alternatives to avoid communication line or equipment failures affecting the transmission of daily business data of the power grid, thereby ensuring the stable operation of the power grid.

[0005] Traditional path assessment methods mostly use statistical analysis strategies to conduct quantitative assessments of risk nodes, paths or systems by building mathematical models and designing simulation experiments. In recent years, with the rapid development of artificial intelligence technology, related technologies such as machine learning and deep learning have gradually been applied to risk assessment. For example, risk quantitative assessments of paths in complex physical topologies are conducted based on graph neural networks and reinforcement learning.

[0006] However, the inventors found that for power communication networks, due to the dynamic nature of daily business volume, the diversity of transmission quality requirements for different businesses, and the differences in network topology and communication environment, although the above method can also achieve quantitative risk assessment, it ignores the timing information of communication and the heterogeneity of relationships in the topology diagram, that is, the different characteristics or attributes presented by the relationships. The comprehensiveness and accuracy of the assessment needs to be improved.

[0007] In addition, the above methods often only focus on one aspect of time or space, and fail to effectively combine time series characteristics and spatial structure characteristics, which affects the performance of risk assessment. Summary of the invention

[0008] In order to address the deficiencies of the prior art, the present invention provides a method, system, electronic device, computer-readable storage medium and computer program product for evaluating the path of an electric power communication network based on feature fusion, which extracts and adaptively fuses the timing characteristics of each time window of physical nodes in different business communication networks and the structural characteristics of physical nodes, outputs the risk score of the physical node according to the fused characteristics, and performs risk assessment on the paths in the network based on the risk score, thereby improving the accuracy and comprehensiveness of the risk assessment.

[0009] In a first aspect, the present invention provides a method for evaluating a path in a power communication network based on feature fusion;

[0010] A method for evaluating a power communication network path based on feature fusion, comprising:

[0011] Acquire and preprocess the communication data of the physical nodes in the power communication network, extract features from the preprocessed communication data, and acquire the timing features of the physical nodes;

[0012] Obtain the power communication network topology map and prior knowledge in the power field, and construct a power communication knowledge graph based on the power communication network topology map and prior knowledge in the power field; perform graph convolution aggregation operations in hyperbolic space on the power communication knowledge graph to extract the structural features of physical nodes;

[0013] The timing features and structural features of the physical nodes are adaptively fused to obtain a feature fusion vector and input it into a preset risk assessment model for processing to obtain a risk prediction result; based on the risk prediction result, a path assessment result is obtained.

[0014] In some embodiments, the feature extraction of the preprocessed communication data to obtain the timing characteristics of the physical nodes is specifically as follows: the preprocessed communication data is input into a Bi-LSTM-based autoencoder for processing, the front and back timing information is captured through the forward propagation and backward propagation LSTM, and the timing characteristics of the physical nodes are extracted.

[0015] In some embodiments, before performing a graph convolution aggregation operation in a hyperbolic space on the power communication knowledge graph, it also includes: initializing the physical nodes in the power communication knowledge graph using position coding and feature coding.

[0016] In some implementations, performing a graph convolution aggregation operation in a hyperbolic space on the power communication knowledge graph to extract structural features of physical nodes specifically includes:

[0017] Project the initialized power communication knowledge graph into the hyperbolic space of non-Euclidean geometry to obtain the first-order embedding of physical nodes;

[0018] The attention aggregation operation is iteratively performed on the first-order embedded neighborhood information through hyperbolic graph convolution and combined to obtain the structural feature vector.

[0019] In some implementations, adaptively fusing the temporal features and structural features of the physical node to obtain a feature fusion vector specifically includes:

[0020] Obtain fusion gating signals based on the timing characteristics and structural characteristics of the physical nodes;

[0021] A feature fusion vector is obtained according to the fusion gating signal, the timing characteristics and the structural characteristics of the physical node.

[0022] In some embodiments, the feature fusion vector is input into a preset risk assessment model for processing, and the risk prediction result is obtained by inputting the feature fusion vector into a preset multi-layer perceptron (MLP) for feature transformation, and using an activation function to determine the probability of risk occurring in the physical node.

[0023] In a second aspect, the present invention provides a power communication network path evaluation system based on feature fusion;

[0024] A power communication network path evaluation system based on feature fusion, comprising:

[0025] The timing feature extraction module is configured to: obtain communication data of physical nodes in the power communication network and perform preprocessing, perform feature extraction on the preprocessed communication data, and obtain timing features of the physical nodes;

[0026] The structural feature extraction module is configured to: obtain the power communication network topology map and the prior knowledge of the power field, and construct the power communication knowledge graph based on the power communication network topology map and the prior knowledge of the power field; perform the graph convolution aggregation operation in the hyperbolic space on the power communication knowledge graph to extract the structural features of the physical nodes;

[0027] The path assessment module is configured to: adaptively fuse the timing features and structural features of the physical node, obtain the feature fusion vector and input it into a preset risk assessment model for processing to obtain the risk prediction result; and obtain the path assessment result based on the risk prediction result.

[0028] In a third aspect, the present invention provides an electronic device;

[0029] An electronic device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-mentioned method for evaluating a power communication network path based on feature fusion.

[0030] In a fourth aspect, the present invention provides a computer-readable storage medium;

[0031] A computer-readable storage medium stores a computer program / instruction thereon, which, when executed by a processor, implements the steps of the above-mentioned method for evaluating a power communication network path based on feature fusion.

[0032] In a fifth aspect, the present invention provides a computer program product;

[0033] A computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned method for evaluating a power communication network path based on feature fusion.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. The technical solution provided by the present invention is based on the topological map and time series data of the power communication network for different businesses, and proposes a power communication network path evaluation method based on Bi-LSTM and hyperbolic GCN, which solves the risk assessment problem of different business paths in the field of power communication and improves the performance of path transmission utility detection.

[0036] 2. The technical solution provided by the present invention introduces the timing characteristics in the communication process into the path risk assessment method, and uses the Bi-LSTM-based autoencoder to extract the timing characteristics of each physical node in different time windows, fully mining the timing information in the power communication network, and realizing multi-step node risk prediction in the time dimension.

[0037] 3. The technical solution provided by the present invention utilizes the GCN aggregation operation in hyperbolic space to obtain the structural characteristics of physical nodes. Compared with the Euclidean space method, it can greatly reduce the distortion of node embedding in heterogeneous graphs, thereby enriching the information expression capability of the embedded representation and improving the risk prediction performance of the model.

[0038] 4. The technical solution provided by the present invention greatly improves the performance of risk assessment by designing a temporal feature extraction module and a structural feature extraction module for physical nodes and efficiently combining the two types of feature information. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0040] Figure 1 A schematic diagram of a flow chart of a method for evaluating a power communication network path based on feature fusion provided in an embodiment of the present invention;

[0041] Figure 2 A schematic diagram of a model framework provided by an embodiment of the present invention;

[0042] Figure 3 A schematic diagram of the network architecture of a Bi-LSTM encoder provided in an embodiment of the present invention;

[0043] Figure 4 A schematic diagram of the network architecture of a single LSTM unit provided in an embodiment of the present invention;

[0044] Figure 5 A schematic diagram of the structure of a power communication network path evaluation system based on feature fusion provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0046] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0047] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0048] Explanation of terms

[0049] Poincare sphere model: This implementation uses a Poincare sphere model with constant curvature. The curvature c is set to 1. The Poincare sphere can be defined as:

[0050]

[0051] Logarithmic and Exponential Maps: Tangent Spaces represent The n-dimensional Euclidean space of the point z represents the first-order approximation of the point z. Since it is not clear how to define feature transformations and aggregation operations in hyperbolic space, the tangent space is necessary to perform operations in the hyperbolic space. Usually, the tangent space and the hyperbolic space need to be converted continuously. The exponential mapping can convert the tangent space Mapping to hyperbolic space In contrast, the logarithmic map transforms the hyperbolic space Map to The definitions of the two at point z are as follows:

[0052]

[0053] in, Represents the Möbius addition method:

[0054]

[0055] Autoencoder: An autoencoder is an unsupervised learning neural network consisting of an encoder (E) and a decoder (D). The encoder converts the input data into a low-dimensional encoded representation that usually captures the important features of the data. The decoder maps this encoding back to the original data space, trying to reconstruct the input data. The training goal of the autoencoder is to minimize the difference between the input data and the reconstructed data so that the encoding can retain key information.

[0056] In anomaly detection scenarios, the reconstruction error is usually used as an anomaly score. The time window with a score higher than the threshold will be considered an anomaly. The reconstruction error of the autoencoder is defined as:

[0057]

[0058] Among them, AE(W)=D(Z), Z=E(W), W represents the time window, and Z represents the latent variable.

[0059] Embodiment 1

[0060] The existing power communication network path assessment method ignores the timing information of communication data and the heterogeneity of relationships in the topology diagram, resulting in the accuracy and comprehensiveness of risk assessment to be improved; therefore, the present invention provides a power communication network path assessment method based on feature fusion based on the topology diagram and time series data of the power communication network for different businesses.

[0061] Next, combine Figure 1-Figure 4 , a method for evaluating a power communication network path based on feature fusion disclosed in this embodiment is described in detail. The method for evaluating a power communication network path based on feature fusion comprises the following steps:

[0062] S1. Obtain communication data of physical nodes in the power communication network and perform preprocessing.

[0063] In this embodiment, the communication data is a time series feature, including network packet loss rate, communication status, communication delay, power flow entropy, service flow characteristics, channel type, channel bandwidth, flow and node connection number characteristics, which are expressed as a time series feature matrix According to the window size in the training label, it is divided into a time window sequence W = (W k ,W k+1 ,W k+2 ,…,W T ), each node in the topology graph corresponds to a set of time window sequences.

[0064] Here, the topology graph refers to the network topology graph that includes the communication network between physical nodes. Different types of links can be regarded as different relationships. On the basis of the topology graph, the power prior knowledge is introduced to form the power communication knowledge graph. For example, if two physical nodes belong to the same category, they are connected to the same prior knowledge entity.

[0065] Due to environmental factors, the time series features of physical nodes in the power communication network may be missed. To address this problem, in this embodiment, the missing data is supplemented by the mean of adjacent data. For example, when the feature at time t is missed, the mean of the data at time t-1 and time t+1 is used to supplement it.

[0066] At the same time, the erroneous items and duplicate items in the time window sequence are cleaned and the data is normalized, and the time window sequence after data normalization is used as the input of the Bi-LSTM-based autoencoder.

[0067] S2. Perform feature extraction on the preprocessed time window sequence to obtain the timing features of the physical node.

[0068] In order to better capture the preceding and following time sequence information in the time window sequence, in this embodiment, a Bi-LSTM-based autoencoder is used to encode the preprocessed time window sequence.

[0069] Combination Figure 3-Figure 4 Specifically, Bi-LSTM can be understood as a LSTM network in the forward and reverse directions. The traditional LSTM formula is as follows:

[0070] f t =Sigmoid(W f ·[h t-1 ,X t ]+b f ) (6)

[0071] i t =Sigmoid(W i ·[h t-1 ,X t ]+b i ) (7)

[0072] o t =Sigmoid(W o ·[h t-1 ,X t ]+b o ) (8)

[0073] c t =f t ⊙c t-1 +i t ☉Tanh(W c ·[h t-1 , X t ]+b c ) (9)

[0074] h t =o t ☉Tanh(c t ) (10)

[0075] Among them, Sigmoid is the activation function, X t is the input vector at time t, W f ,W i ,W o is the weight matrix, b f ,b i ,b o is the offset vector, f t ,i t ,o t represents the gate vector, c t is the cell state vector, h t is the state output vector.

[0076] In this embodiment, the Bi-LSTM used captures the previous and next time sequence information better through the forward propagation and backward propagation LSTM, and combines the hidden states in the two directions to obtain the output of the Bi-LSTM layer in a time window. Finally, the hidden state H is activated by the Tanh function. t Encoded as a hidden variable Z, the hidden variable Z is the extracted temporal feature of the physical node. The specific formula is as follows:

[0077]

[0078] Z=Tanh(H t ·W z +b z ) (14)

[0079] Similar to the encoder E, this embodiment also uses a decoder D based on a Bi-LSTM network. The training process of the autoencoder AE is completed by minimizing the reconstruction error in formula (5).

[0080] S3. Obtain the power communication network topology map and prior knowledge in the power field, and construct a power communication knowledge graph based on the power communication network topology map and prior knowledge in the power field.

[0081] Specifically, first, the data of the power communication network topology and prior knowledge are collected and organized; then, entity recognition and relationship extraction are performed, and the data is normalized, including entity alignment and disambiguation; finally, triples are constructed to connect entities and relationships into a power communication knowledge graph.

[0082] S4. Perform graph convolution aggregation operations in hyperbolic space on the power communication knowledge graph to extract the structural features of physical nodes.

[0083] In order to fully mine the complex structural information of nodes in the knowledge graph, before executing S4, it also includes: initializing the physical nodes in the power communication knowledge graph using position coding and feature coding.

[0084] Specifically, for position coding, in this embodiment, Laplacian position coding (LapPE) is used to encode the position information, because this coding method is sensitive to Euclidean distance, which can make adjacent nodes have similar characteristics, while the opposite is true for distant nodes. The positional encoding is represented as

[0085] For feature coding, in this embodiment, MLP is used to transform the original nodes to mine the potential features of the nodes, which is expressed as

[0086] Final Node The initial structural embedding of can be expressed as:

[0087]

[0088] Next, the initial structure embedding is projected into a hyperbolic space of non-Euclidean geometry and used as the first-order embedding for the hyperbolic GCN aggregation:

[0089]

[0090] Then, in this embodiment, a relation-aware hyperbolic GCN is used to perform representation learning on the nodes in the power communication knowledge graph.

[0091] For the convenience of description, h, r, and t represent the head entity, relationship, and tail entity in the knowledge graph. The first-order neighbor triples of entity h represent the node structure features. By continuously aggregating the information of neighbor nodes, the node structure features can be extracted. h The first-order attention aggregation operation of neighborhood information is defined as:

[0092]

[0093] Among them, exp represents exponential mapping, which is responsible for mapping the embedding into the hyperbolic space; log represents logarithmic mapping, which is responsible for mapping the embedding back into the tangent space, that is, the Euclidean space; μ (0) Represents a trainable parameter.

[0094] In addition, in this embodiment, combined with the Softmax function, the attention score π(h, r, t) is defined as:

[0095]

[0096] Among them, LeakyReLU represents the activation function, W r Represents the trainable parameter matrix. Similarly, executing the above process L times, we can get node v h The L-order embedding representation of is expressed as:

[0097]

[0098] After L-order aggregation, the L-order embedding representations are combined to obtain the final structural features, which are expressed as:

[0099]

[0100] S5. Adaptively fuse the temporal features and structural features of the physical node to obtain a feature fusion vector.

[0101] As an implementation method, for the timing feature and structural features This embodiment proposes an attention fusion method, which uses a gating signal to adaptively adjust the fusion ratio. The specific formula is as follows:

[0102]

[0103] in, represents the fusion gate signal, W1, W2 represent the trainable parameter matrix, Represents the final fused embedding representation, that is, the feature fusion vector.

[0104] S6. The feature fusion vector is input into a preset risk assessment model for processing to obtain a risk prediction result; based on the risk prediction result, a path assessment result is obtained.

[0105] In this embodiment, the risk assessment model is MLP, the feature fusion vector is input into the trained MLP for feature transformation, and the activation function is set to Sigmoid, and the probability of risk occurring in the physical node is obtained by calculation, which is expressed as:

[0106]

[0107] in, Indicates the probability of a physical node experiencing risk. By calculating this probability, the risk prediction label of the communication node can be obtained.

[0108] For the path set L of business i i , use the predicted value of the node risk label to Carry out risk quantitative assessment, and regard the paths where the proportion of risk nodes in the path exceeds the preset threshold, that is, the paths where the proportion of risk nodes in the path is higher than the total nodes as high-risk paths, and finally obtain L i The paths in the set are sorted from high to low in terms of the percentage of risky nodes.

[0109] As an implementation method, before the risk assessment model is applied, it is optimized using Binary Cross-Entropy Loss. The Binary Cross-Entropy Loss function is expressed as follows:

[0110]

[0111] Among them, M represents the number of services, N represents the number of nodes, and K represents the number of time windows. represents the true value of the node risk label, Represents the predicted value of the node risk label.

[0112] Next, in order to verify the advancement of the power communication network path evaluation method based on feature fusion described in this embodiment, this embodiment conducted a set of experiments.

[0113] The adopted data set includes 6092 physical nodes involved in different businesses in the field of power communication. Each node includes network packet loss rate, communication status, communication delay, power flow entropy and business flow characteristics, and the lines between nodes include channel type, channel bandwidth, flow, and node connection number characteristics. In this embodiment, the communication data of each node from September to December 2023 is collected to form a time series, which is collected every 1 hour. Then, it is divided into several time windows using the sliding window mechanism, and the risk label of each node in each time window is provided.

[0114] In addition, this embodiment also defines 1054 knowledge entities, covering the prior knowledge in the power field, such as power system structure, three-phase electricity, the basis of transformers and substations, transmission line characteristics, and the basic principles of power system stability and protection mechanisms. Based on the prior knowledge in the field of power communication and the relationship between physical nodes, a power communication knowledge graph is constructed. Finally, the data set is divided into a training set, a validation set, and a test set in a ratio of 6:2:2.

[0115] The experimental parameter settings used in this embodiment are shown in Table 1. Wherein L represents the depth of GCN aggregation, Lr represents the model learning rate, Dropout represents the random drop ratio, and Epoch represents the number of training cycles.

[0116] Table 1 Experimental parameter settings

[0117]

[0118]

[0119] This example uses three indicators, Area Under Curve (AUC), Accuracy (ACC) and F1-Score, to evaluate the risk identification performance of different models. The model in this example is compared with six methods, including GCN and GAT. The experimental results are shown in Table 2:

[0120] Table 2 Performance comparison of different models

[0121]

[0122] As can be seen from Table 2, the model described in this embodiment outperforms other methods in all indicators. At the same time, it can be noted that the performance of the methods based on graph neural networks and time series prediction are distributed in two intervals, and the method based on time series prediction is significantly better than the method based on graph neural networks, which shows that in this data set, time series provides more valuable information for the risk classification of nodes. Even so, the structural feature extraction module proposed in this embodiment also promotes performance improvement. Compared with the method based on graph neural networks, the AUC of this implementation model has increased by at least 4.9%, the ACC has increased by at least 5.1%, and the F1-Score has increased by at least 4.3%; compared with the method based on time series prediction, the AUC of this implementation model has increased by at least 2.4%, the ACC has increased by at least 2.2%, and the F1-Score has increased by at least 2.0%. This shows that the combination of knowledge graph and time series information in this implementation model can break through the limitations of their respective original fields and further improve model performance.

[0123] In order to verify the effectiveness of each module of the model, this embodiment also designs the following five variants for ablation experiments. No temporal feature extraction module means that the variant is only trained by structural features; no structural feature extraction module means that the variant is only trained by temporal features; no attention fusion module means that the variant uses an additive method to process temporal features and structural features; no Bi-LSTM means that the variant removes the Bi-LSTM module and uses an LSTM-based autoencoder instead; no hyperbolic space means that the variant abandons the use of hyperbolic space and performs GCN aggregation in Euclidean space instead. The comparison results of the ablation experiment are shown in Table 3:

[0124] Table 3 Ablation experiment

[0125]

[0126] As can be seen from Table 3, time series feature extraction and structural feature extraction, as the two main modules of this implementation model, both play an important role in improving performance. Specifically, after removing the time series feature extraction module, the model can only use the structural features in the static knowledge graph, and cannot grasp the changing trend of risk nodes at the time level, resulting in a decline in various indicators; after removing the structural feature extraction module, the model cannot obtain the logical association between each physical node through the knowledge graph, which will also reduce the accuracy of risk identification. In addition, the attention fusion, Bi-LSTM, and hyperbolic space modules proposed in this embodiment can also improve performance to a certain extent. The above results fully illustrate the rationality and effectiveness of the design of each module of this model.

[0127] Embodiment 2

[0128] Combination Figure 5, this embodiment discloses a power communication network path evaluation system based on feature fusion, including:

[0129] The timing feature extraction module is configured to: obtain communication data of physical nodes in the power communication network and perform preprocessing, perform feature extraction on the preprocessed communication data, and obtain timing features of the physical nodes;

[0130] The structural feature extraction module is configured to: obtain the power communication network topology map and the prior knowledge of the power field, and construct the power communication knowledge graph based on the power communication network topology map and the prior knowledge of the power field; perform the graph convolution aggregation operation in the hyperbolic space on the power communication knowledge graph to extract the structural features of the physical nodes;

[0131] The path assessment module is configured to: adaptively fuse the timing features and structural features of the physical node, obtain the feature fusion vector and input it into a preset risk assessment model for processing to obtain the risk prediction result; and obtain the path assessment result based on the risk prediction result.

[0132] In some embodiments, the feature extraction of the preprocessed communication data to obtain the timing characteristics of the physical nodes is specifically as follows: the preprocessed communication data is input into a Bi-LSTM-based autoencoder for processing, the front and back timing information is captured through the forward propagation and backward propagation LSTM, and the timing characteristics of the physical nodes are extracted.

[0133] In some embodiments, before performing a graph convolution aggregation operation in a hyperbolic space on the power communication knowledge graph, it also includes: initializing the physical nodes in the power communication knowledge graph using position coding and feature coding.

[0134] In some implementations, performing a graph convolution aggregation operation in a hyperbolic space on the power communication knowledge graph to extract structural features of physical nodes specifically includes:

[0135] Project the initialized power communication knowledge graph into the hyperbolic space of non-Euclidean geometry to obtain the first-order embedding of physical nodes;

[0136] The attention aggregation operation is iteratively performed on the first-order embedded neighborhood information through hyperbolic graph convolution and combined to obtain the structural feature vector.

[0137] In some implementations, adaptively fusing the temporal features and structural features of the physical node to obtain a feature fusion vector specifically includes:

[0138] Obtain fusion gating signals based on the timing characteristics and structural characteristics of the physical nodes;

[0139] A feature fusion vector is obtained according to the fusion gating signal, the timing characteristics and the structural characteristics of the physical node.

[0140] In some embodiments, the feature fusion vector is input into a preset risk assessment model for processing, and the risk prediction result is obtained by inputting the feature fusion vector into a preset multi-layer perceptron for feature transformation, and using an activation function to determine the probability of risk occurring in the physical node.

[0141] It should be noted that the above-mentioned time series feature extraction module, structural feature extraction module and path evaluation module correspond to the steps in Embodiment 1, and the examples and application scenarios implemented by the above-mentioned modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Embodiment 1. It should be noted that the above-mentioned modules as part of the system can be executed in a computer system such as a set of computer executable instructions.

[0142] Embodiment 3

[0143] Based on the first embodiment, the third embodiment of the present invention provides an electronic device, including a memory and a processor, and a computer instruction stored in the memory and executed on the processor. When the computer instruction is executed by the processor, the following steps are completed:

[0144] Acquire and preprocess the communication data of the physical nodes in the power communication network, extract features from the preprocessed communication data, and acquire the timing features of the physical nodes;

[0145] Obtain the power communication network topology map and prior knowledge in the power field, and construct a power communication knowledge graph based on the power communication network topology map and prior knowledge in the power field; perform graph convolution aggregation operations in hyperbolic space on the power communication knowledge graph to extract the structural features of physical nodes;

[0146] The timing features and structural features of the physical nodes are adaptively fused to obtain a feature fusion vector and input it into a preset risk assessment model for processing to obtain a risk prediction result; based on the risk prediction result, a path assessment result is obtained.

[0147] In some embodiments, the feature extraction of the preprocessed communication data to obtain the timing characteristics of the physical nodes is specifically as follows: the preprocessed communication data is input into a Bi-LSTM-based autoencoder for processing, the front and back timing information is captured through the forward propagation and backward propagation LSTM, and the timing characteristics of the physical nodes are extracted.

[0148] In some embodiments, before performing a graph convolution aggregation operation in a hyperbolic space on the power communication knowledge graph, it also includes: initializing the physical nodes in the power communication knowledge graph using position coding and feature coding.

[0149] In some implementations, performing a graph convolution aggregation operation in a hyperbolic space on the power communication knowledge graph to extract structural features of physical nodes specifically includes:

[0150] Project the initialized power communication knowledge graph into the hyperbolic space of non-Euclidean geometry to obtain the first-order embedding of physical nodes;

[0151] The attention aggregation operation is iteratively performed on the first-order embedded neighborhood information through hyperbolic graph convolution and combined to obtain the structural feature vector.

[0152] In some implementations, adaptively fusing the temporal features and structural features of the physical node to obtain a feature fusion vector specifically includes:

[0153] Obtain fusion gating signals based on the timing characteristics and structural characteristics of the physical nodes;

[0154] A feature fusion vector is obtained according to the fusion gating signal, the timing characteristics and the structural characteristics of the physical node.

[0155] In some embodiments, the feature fusion vector is input into a preset risk assessment model for processing, and the risk prediction result is obtained by inputting the feature fusion vector into a preset multi-layer perceptron for feature transformation, and using an activation function to determine the probability of risk occurring in the physical node.

[0156] Embodiment 4

[0157] Based on the first embodiment, the fourth embodiment of the present invention provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the following steps are completed:

[0158] Acquire and preprocess the communication data of the physical nodes in the power communication network, extract features from the preprocessed communication data, and acquire the timing features of the physical nodes;

[0159] Obtain the power communication network topology map and prior knowledge in the power field, and construct a power communication knowledge graph based on the power communication network topology map and prior knowledge in the power field; perform graph convolution aggregation operations in hyperbolic space on the power communication knowledge graph to extract the structural features of physical nodes;

[0160] The timing features and structural features of the physical nodes are adaptively fused to obtain a feature fusion vector and input it into a preset risk assessment model for processing to obtain a risk prediction result; based on the risk prediction result, a path assessment result is obtained.

[0161] In some embodiments, the feature extraction of the preprocessed communication data to obtain the timing characteristics of the physical nodes is specifically as follows: the preprocessed communication data is input into a Bi-LSTM-based autoencoder for processing, the front and back timing information is captured through the forward propagation and backward propagation LSTM, and the timing characteristics of the physical nodes are extracted.

[0162] In some embodiments, before performing a graph convolution aggregation operation in a hyperbolic space on the power communication knowledge graph, it also includes: initializing the physical nodes in the power communication knowledge graph using position coding and feature coding.

[0163] In some implementations, performing a graph convolution aggregation operation in a hyperbolic space on the power communication knowledge graph to extract structural features of physical nodes specifically includes:

[0164] Project the initialized power communication knowledge graph into the hyperbolic space of non-Euclidean geometry to obtain the first-order embedding of physical nodes;

[0165] The attention aggregation operation is iteratively performed on the first-order embedded neighborhood information through hyperbolic graph convolution and combined to obtain the structural feature vector.

[0166] In some implementations, adaptively fusing the temporal features and structural features of the physical node to obtain a feature fusion vector specifically includes:

[0167] Obtain fusion gating signals based on the timing characteristics and structural characteristics of the physical nodes;

[0168] A feature fusion vector is obtained according to the fusion gating signal, the timing characteristics and the structural characteristics of the physical node.

[0169] In some embodiments, the feature fusion vector is input into a preset risk assessment model for processing, and the risk prediction result is obtained by inputting the feature fusion vector into a preset multi-layer perceptron for feature transformation, and using an activation function to determine the probability of risk occurring in the physical node.

[0170] Embodiment 5

[0171] Based on the first embodiment, the fifth embodiment of the present invention provides a computer program product, including a computer program / instruction, which implements the following steps when executed by a processor:

[0172] Acquire and preprocess the communication data of the physical nodes in the power communication network, extract features from the preprocessed communication data, and acquire the timing features of the physical nodes;

[0173] Obtain the power communication network topology map and prior knowledge in the power field, and construct a power communication knowledge graph based on the power communication network topology map and prior knowledge in the power field; perform graph convolution aggregation operations in hyperbolic space on the power communication knowledge graph to extract the structural features of physical nodes;

[0174] The timing features and structural features of the physical nodes are adaptively fused to obtain a feature fusion vector and input it into a preset risk assessment model for processing to obtain a risk prediction result; based on the risk prediction result, a path assessment result is obtained.

[0175] In some embodiments, the feature extraction of the preprocessed communication data to obtain the timing characteristics of the physical nodes is specifically as follows: the preprocessed communication data is input into a Bi-LSTM-based autoencoder for processing, the front and back timing information is captured through the forward propagation and backward propagation LSTM, and the timing characteristics of the physical nodes are extracted.

[0176] In some embodiments, before performing a graph convolution aggregation operation in a hyperbolic space on the power communication knowledge graph, it also includes: initializing the physical nodes in the power communication knowledge graph using position coding and feature coding.

[0177] In some implementations, performing a graph convolution aggregation operation in a hyperbolic space on the power communication knowledge graph to extract structural features of physical nodes specifically includes:

[0178] Project the initialized power communication knowledge graph into the hyperbolic space of non-Euclidean geometry to obtain the first-order embedding of physical nodes;

[0179] The attention aggregation operation is iteratively performed on the first-order embedded neighborhood information through hyperbolic graph convolution and combined to obtain the structural feature vector.

[0180] In some implementations, adaptively fusing the temporal features and structural features of the physical node to obtain a feature fusion vector specifically includes:

[0181] Obtain fusion gating signals based on the timing characteristics and structural characteristics of the physical nodes;

[0182] A feature fusion vector is obtained according to the fusion gating signal, the timing characteristics and the structural characteristics of the physical node.

[0183] In some embodiments, the feature fusion vector is input into a preset risk assessment model for processing, and the risk prediction result is obtained by inputting the feature fusion vector into a preset multi-layer perceptron for feature transformation, and using an activation function to determine the probability of risk occurring in the physical node.

[0184] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0185] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0186] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0187] The description of each embodiment in the above embodiments has different emphases. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0188] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A power communication network path evaluation method based on feature fusion, characterized in that: include: Acquire and preprocess the communication data of the physical nodes in the power communication network, extract features from the preprocessed communication data, and acquire the timing features of the physical nodes; Obtain the power communication network topology map and prior knowledge in the power field, and build a power communication knowledge graph based on the power communication network topology map and prior knowledge in the power field; Perform graph convolution aggregation operations in hyperbolic space on the power communication knowledge graph to extract the structural features of physical nodes; The timing features and structural features of the physical nodes are adaptively fused to obtain a feature fusion vector and input it into a preset risk assessment model for processing to obtain a risk prediction result; based on the risk prediction result, a path assessment result is obtained.

2. The method for evaluating a power communication network path based on feature fusion according to claim 1, characterized in that: The feature extraction of the preprocessed communication data to obtain the timing characteristics of the physical nodes is specifically as follows: the preprocessed communication data is input into the Bi-LSTM-based autoencoder for processing, the front and back timing information is captured through the forward propagation and backward propagation LSTM, and the timing characteristics of the physical nodes are extracted.

3. The method for evaluating a power communication network path based on feature fusion according to claim 1, characterized in that: Before performing the graph convolution aggregation operation in the hyperbolic space on the power communication knowledge graph, it also includes: initializing the physical nodes in the power communication knowledge graph using position encoding and feature encoding.

4. The method for evaluating a power communication network path based on feature fusion according to claim 1, characterized in that: The graph convolution aggregation operation is performed on the power communication knowledge graph in a hyperbolic space to extract the structural features of the physical nodes, specifically including: Project the initialized power communication knowledge graph into the hyperbolic space of non-Euclidean geometry to obtain the first-order embedding of physical nodes; The attention aggregation operation is iteratively performed on the first-order embedded neighborhood information through hyperbolic graph convolution and combined to obtain the structural feature vector.

5. The method for evaluating a power communication network path based on feature fusion according to claim 1, characterized in that: Adaptively fusing the temporal features and structural features of the physical node to obtain a feature fusion vector specifically includes: Obtain fusion gating signals based on the timing characteristics and structural characteristics of the physical nodes; A feature fusion vector is obtained according to the fusion gating signal, the timing characteristics and the structural characteristics of the physical node.

6. The method for evaluating a power communication network path based on feature fusion according to claim 1, characterized in that: The feature fusion vector is input into the preset risk assessment model for processing to obtain the risk prediction result. Specifically, the feature fusion vector is input into the preset multi-layer perceptron for feature transformation, and the activation function is used to determine the probability of risk occurring in the physical node.

7. The power communication network path evaluation system based on feature fusion is characterized by: include: The timing feature extraction module is configured to: obtain communication data of physical nodes in the power communication network and perform preprocessing, perform feature extraction on the preprocessed communication data, and obtain timing features of the physical nodes; The structural feature extraction module is configured to: obtain the power communication network topology map and the prior knowledge of the power field, and construct the power communication knowledge graph based on the power communication network topology map and the prior knowledge of the power field; perform the graph convolution aggregation operation in the hyperbolic space on the power communication knowledge graph to extract the structural features of the physical nodes; The path assessment module is configured to: adaptively fuse the timing features and structural features of the physical node, obtain the feature fusion vector and input it into a preset risk assessment model for processing to obtain the risk prediction result; and obtain the path assessment result based on the risk prediction result.

8. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the power communication network path evaluation method based on feature fusion as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instruction is executed by a processor, the steps of the method for evaluating a power communication network path based on feature fusion described in any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the method for evaluating a power communication network path based on feature fusion described in any one of claims 1 to 6 are implemented.

Images