A method and system for predicting power information network performance based on graph neural network

Abstract

The present invention discloses a method and system for predicting the performance of an electric power information network based on a graph neural network, and relates to the technical field of network performance estimation. Traditional network performance estimation methods are implemented through network deduction, and have the problems of being difficult to adapt to changes in network configuration, lacking flow-level statistics, leading to lags in large-scale network performance estimation, and insufficient prediction accuracy. The present invention includes: constructing a graph neural network model, extracting network topology features through the graph neural network model; calculating the mutual influence between network flows based on time series technology; extracting node features related to the target flow through the topological information of the target flow; calculating the correlation between all flows and the target flow; and fusing features to obtain network performance prediction results. This technical solution improves the accuracy of performance estimation, and has the versatility and adaptability to different types of networks. It can provide real-time performance estimation results for network operation and maintenance personnel, and is helpful for network troubleshooting and optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of network performance estimation, and in particular to a method and system for predicting power information network performance based on graph neural networks. Background Art

[0002] With the advancement of smart grid construction, power information networks carry real-time data transmission for critical services such as relay protection, safety automation, and dispatch automation. Their network performance is directly related to the safe and stable operation of the power grid. Accurate network performance estimation plays a vital role in effectively optimizing network performance. This estimation is particularly useful for researching new protocols and mechanisms, as it allows administrators to evaluate their performance before deploying them in production networks.

[0003] Traditional network performance estimation methods are mostly implemented through network simulation. However, these methods have several challenges. First, they cannot provide accurate estimation results when faced with changes in network configuration (such as topology, traffic patterns, and traffic management mechanisms), thus lacking universality. Second, most existing methods focus on packet-level performance estimation and lack attention to flow-level statistics, which makes it difficult to timely estimate the performance of large-scale networks. Furthermore, current methods for predicting power information network performance still have significant room for improvement in terms of accuracy. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is to solve the problems existing in traditional methods in terms of network configuration changes, flow level statistics and prediction accuracy.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a method for predicting the performance of a power information network based on a graph neural network, which comprises the following steps:

[0007] Construct a graph neural network model and extract the topological features of the power information network through the graph neural network model; calculate the mutual influence between network flows based on time series technology; extract node features related to the target flow through the topological information of the target flow; calculate the correlation between all flows and the target flow; and fuse the features to obtain network performance prediction results.

[0008] This technical solution deeply integrates graph neural networks with time series technology. Through dynamic-static collaborative modeling and goal-oriented feature extraction, it addresses the problem of insufficient prediction accuracy caused by the coupling of topological complexity, time-varying traffic, and service relevance in power information networks. Specifically, a graph neural network directly models the complex topology of power information networks (such as node connectivity and path dependencies), capturing spatial correlations that are difficult to describe with traditional methods (such as the cascading effects between relay protection nodes in the power grid). Time series technologies (such as Transformers and LSTMs) are introduced to model the time-varying correlations of network flows (such as traffic bursts and cyclical fluctuations), addressing the limited adaptability of traditional static traffic analysis to real-time scenarios. Key node features are filtered using target flow path information (such as the transmission path of power service flows), suppressing noise interference (such as redundant data from irrelevant nodes) and avoiding the feature dilution problem caused by the "equal treatment of all nodes" in traditional methods. The correlations between all flows and the target flow (such as bandwidth contention and routing overlap) are quantified, enabling fine-grained traffic impact modeling (such as the crowding effect of high-priority service flows on the target flow), surpassing traditional rule-based traffic prioritization. By integrating topological features (spatial dimension) with traffic interaction features (temporal dimension) through nonlinear fusion (e.g., MLP and attention mechanisms), this approach uniformly models spatiotemporal coupling effects (e.g., traffic redistribution caused by topological changes), addressing the error accumulation issues inherent in traditional step-by-step modeling. Leveraging the topological adaptability of graph neural networks and dynamic traffic modeling using timing technology, this approach supports real-time changes in network configuration (e.g., node failures, routing adjustments), outperforming traditional inference methods that rely on fixed topological assumptions. Feature screening and correlation calculations reduce redundant computational overhead, meeting the real-time requirements of power information networks (e.g., microsecond-level latency-sensitive services).

[0009] As a preferred solution of the method for predicting the performance of an electric power information network based on a graph neural network described in the present invention, the topological characteristics of the electric power information network include the connection relationship and topological characteristics between nodes.

[0010] The method of extracting the topological features of the power information network includes: a topological feature extractor uses a graph neural network to extract topological features, and uses the connection pattern and structural features between nodes to encode the network topology information. The graph neural network uses a graph convolutional network to extract, and the network topology is modeled as a graph structure. The edge features in the topological information and the node features in the state information are extracted respectively, and are expressed as , where V represents the set of all nodes in the graph, E represents the set of all edges, and each node is represented as a feature vector , the edge between node i and node j is represented as ,In the graph convolutional network, message passing is performed at each layer, expressed as,

[0011]

[0012] in, Indicates the Tier The feature vector of each node, represents the activation function, Representation node The set of neighbor nodes of It's the edge The normalization factor for normalization, Indicates the The weight matrix of the layer.

[0013] This technical solution can realize end-to-end dynamic modeling of power information network topology, solve the limitations of traditional methods in high-order relationship capture, dynamic adaptability and feature expression efficiency, and provide a high-precision, low-latency topology feature extraction basis for power network performance prediction. It is particularly suitable for scenarios with complex topology and dynamic flow coupling in smart grids.

[0014] As a preferred solution of the method for predicting the performance of a power information network based on a graph neural network described in the present invention, wherein: the mutual influence between the computing network flows includes: the encoder receives the feature vector of the flow as input, uses the multi-head attention mechanism to extract the mutual influence relationship between the flows, and models the flows as a sequence ,in Indicates the number of flows in the network, and each flow is represented by a feature vector express.

[0015] The encoder uses a multi-head attention mechanism to linearly transform the input feature vector into a query, key, and value vector. By calculating the dot product of the query and the key, the attention weight is obtained to measure the correlation between each stream and other streams. The attention weight is multiplied by the value vector to obtain a weighted stream feature representation. The multi-head attention mechanism is expressed as,

[0016]

[0017] in, Respectively represent the query, key, and value vectors generated by linear transformation, corresponding to the flow characteristics of different subspaces, Indicates the dimension of the input features.

[0018] Through the multi-head attention mechanism, the flow features are mapped to multiple subspaces (corresponding to multiple query-key-value vector groups), and each head independently captures the correlation between different dimensions of the flows (such as traffic size, time interval, protocol type, etc.), and finally fuses the multi-dimensional feature representation. Avoid the limitations of a single perspective and improve the richness of feature expression. For example, the different impact modes of "high-traffic long connections" and "low-traffic short connections" can be identified at the same time. Through linear transformation, the original flow features are decoupled into query, key, and value vectors, which correspond to "query target", "matching key value", and "information transmission carrier" respectively, clarifying the logical roles of the interaction between flows; structurally separating the use of features, so that the model can learn the dependencies between flows more accurately (for example, when a flow is used as a "query", the relevance of other flows is matched through the "key", and then the effective information is aggregated through the "value"). Through dot product operations and Normalization generates attention weights, which measure the correlation between any two flows in real time. The larger the weight value, the more significant the influence between flows. The weights between flows are adaptively adjusted. For example, in power networks, the association between key flows (such as relay protection signal flows) and other flows can be dynamically identified to suppress interference from irrelevant flows. The flows are modeled as time series, combined with the sensitivity of the attention mechanism to the order of the sequence, to capture the temporal evolution pattern of inter-flow dependencies (such as the attenuation of the influence of a certain flow on the target flow in adjacent time steps). This is suitable for scenarios where the flow in the power network fluctuates over time (such as load changes during peak hours in the morning and evening), and improves the accuracy of dynamic performance prediction. By multiplying the attention weight with the value vector, only information highly correlated with the current flow is retained, and interference from low-correlation flows is filtered out. In large-scale networks, redundant calculations are reduced and computational efficiency is improved. For example, background flows that have little impact on the target flow will be given low weights to avoid "noise" features dominating the prediction. Introducing a scaling factor Normalizing attention scores prevents vanishing softmax gradients caused by excessively large dot products in high-dimensional feature spaces, stabilizing the training process. This approach is particularly useful in scenarios with high-dimensional flow features (e.g., those containing multiple attributes such as flow rate, protocol type, and source / destination addresses). The multi-head attention mechanism can learn nonlinear and asymmetric dependencies between flows, overcoming the limitations of traditional linear models and adapting to complex traffic interaction scenarios in power grids (e.g., multiple concurrent sources competing for access to the same device).

[0019] As a preferred solution of the method for predicting the performance of an electric power information network based on a graph neural network described in the present invention, the extraction of node features related to the target flow includes purifying topological features based on a topological feature purifier, purifying node features related to the target flow, the topological feature purifier identifies the relationship between each node and the target flow through a node scoring mechanism, and uses a weight multiplication operation to extract node features related to the target flow from the output of the topological feature extractor.

[0020] This technical solution systematically solves the "information overload" problem of traditional graph neural networks in power network analysis through goal-oriented node feature purification, achieving dual optimization of feature quality and computational efficiency.

[0021] As a preferred solution of the method for predicting the performance of an electric power information network based on a graph neural network described in the present invention, the topological feature purification based on the topological feature purifier includes outputting a feature matrix representing the entire topological structure based on the topological feature extractor, wherein each row represents a feature vector of a node, and the feature vector contains information about the direct and indirect nodes of the node. The node weights related to the target flow will be set to 1, and the weights of irrelevant nodes will be set to 0 through the node scoring mechanism.

[0022] The node features related to the target flow are extracted by weighted multiplication of the score and the topological features. The eigenvalues ​​of the nodes related to the target flow are retained and the eigenvalues ​​of the irrelevant nodes are suppressed to zero by multiplying the node score with the topological feature matrix.

[0023] The node features related to the target flow are expressed as follows:

[0024]

[0025] Among them, x is the output of the topological feature extractor, representing a node feature, The path of the target flow.

[0026] The technical solution purifies topological features through a binary node scoring mechanism and matrix multiplication operations, completely eliminating the interference of irrelevant nodes (such as edge devices on non-target flow paths in the power network), so that subsequent predictions only rely on topological elements that directly affect the performance of the target flow, thereby improving feature purity. The scoring mechanism encodes the prior knowledge of "target flow path" as a feature selection rule, so that the purified feature matrix explicitly contains path structure information (such as node connection order, number of hops, etc.). Compared with unsupervised feature selection methods (such as attention mechanism automatic learning weights), this rule has a clear physical meaning, which prevents the model from learning false associations that do not conform to the business logic of the power network. The weight multiplication operation converts the original feature matrix into a sparse matrix, which greatly reduces the computational complexity of subsequent graph neural network layers. The purified non-zero nodes directly correspond to the physical path of the target flow, which can assist in quickly locating network fault nodes. When the topology of the power network changes (such as line switching causing the target flow path to change to ), only the scoring matrix needs to be regenerated ,There is no need to retrain the topology feature extractor, meeting the strict real-time requirements of the power system.

[0027] As a preferred solution of the method for predicting the performance of an electric power information network based on a graph neural network described in the present invention, the calculation of the correlation between all flows and the target flow includes calculating the correlation score between each flow and the target flow through a scoring mechanism, measuring the similarity or correlation between the two flows through the correlation score, performing a weighted multiplication operation on the correlation score and the flow feature, and extracting features that are highly correlated with the target flow.

[0028] The scoring mechanism is expressed as,

[0029]

[0030] in, Represent the characteristics of all flows and target flows in the network respectively.

[0031] This technical solution uses a cosine similarity scoring mechanism and weighted feature extraction to efficiently quantify inter-flow correlations and extract highly relevant features. While ensuring the accuracy of inter-flow correlation calculations, it also achieves an optimal balance between computational cost, generalization capability, and interpretability.

[0032] As a preferred solution of the power information network performance prediction method based on graph neural network described in the present invention, the fusion feature to obtain the network performance prediction result includes fusing the features to obtain feature representation, and predicting the target flow based on the feature representation.

[0033] The prediction is expressed as,

[0034]

[0035] in, represents the output of the topological feature refiner, represents the output of the inter-stream feature refiner, denote weights and biases respectively.

[0036] The structural information output by the topological feature purifier ( , such as node connection relationships, path features) and the dynamic information output by the inter-flow feature purifier ( Such as traffic correlation, timing dependence) through splicing operations Integration forms a joint representation; after fusion, "structure-dynamic" collaborative modeling is achieved; for example, when a switch node is overloaded (topological feature) and multiple flows related to it experience burst traffic at the same time (inter-flow feature), the model can more accurately predict performance degradation. Mapping features from different sources to a unified latent space solves the problem of feature dimension and semantic differences, avoids training instability caused by scale differences of different features, and ensures that the weights of the two types of features are balanced during the fusion process. Using two layers of ReLU nonlinear transformation, the model can learn high-order nonlinear relationships between features; for example: the first layer of ReLU captures basic interactions (such as the direct correlation between "node congestion" and "traffic surge"), and the second layer of ReLU models compound effects (such as the nonlinear amplification of delay by "node congestion + multi-flow competition"). and Two weight transformations combine low-level features into high-level semantic units, such as May indicate a "critical path state", and then Combined with the generated more abstract "network performance bottleneck" representation, the ReLU activation function alleviates the gradient disappearance problem and makes deep network training more stable.

[0037] Another object of the present invention is to provide a power information network performance prediction system based on graph neural network, which can improve the accuracy and comprehensiveness of network performance estimation by comprehensively considering the topological characteristics of the power information network and the time-varying nature of the flow, and solve the problems existing in the existing traditional methods in terms of network configuration changes, flow-level statistics and prediction accuracy.

[0038] To solve the above technical problems, the present invention provides the following technical solutions: a power information network performance prediction system based on graph neural network, comprising: a topology feature extraction module, an inter-flow feature extraction module, a target flow topology feature extraction module, a target flow inter-flow feature extraction module and a fusion module.

[0039] The topology feature extraction module is used to calculate the topology features of the power information network and can learn the representation of nodes.

[0040] The inter-flow feature extraction module is used to calculate the mutual influence between network flows, taking into account the time-varying nature of traffic.

[0041] The target flow topology feature extraction module is used to extract node features related to the target flow.

[0042] The target flow inter-flow feature extraction module is used to calculate the correlation between all flows and the target flow, and correct the inter-flow features.

[0043] The fusion module is used to fuse features to obtain network performance estimation results.

[0044] The topology feature extraction module focuses on modeling network topology. Using graph neural networks, it learns node representations (such as node connectivity and topological roles) to achieve deep encoding of static network structures. The inter-flow feature extraction module analyzes traffic dynamics, leveraging time series techniques to capture inter-flow temporal variations (such as the temporal dependencies of traffic fluctuations). Together with the topology module, this module forms a "structure-dynamic" dual-dimensional model. The topology and inter-flow feature extraction modules can run in parallel, reducing overall processing latency and enabling real-time, simultaneous analysis of multi-source data in the power grid. The target flow topology feature extraction module targets specific target flows by filtering out irrelevant node features (such as non-path nodes) and retaining only key nodes (such as source / destination nodes and routing nodes). This reduces invalid node processing, lowers computational load, and improves feature relevance. The inter-flow feature extraction module uses correlation calculations (such as cosine similarity) to suppress background flows with low correlation to the target flow and enhance the feature weights of highly correlated flows (such as collaborative flows within the same service cluster). This module compresses flow feature dimensions from the full set to a subset of highly correlated flows, reducing input noise for subsequent fusion modules. The fusion module integrates topological structural features (such as path connectivity) with dynamic features between flows (such as flow interaction patterns) to reduce prediction errors. Nonlinear transformations improve prediction accuracy in complex scenarios. When network topology changes (such as path switching due to a link failure), the target flow topology feature extraction module updates the relevant node set in real time, eliminating the need to retrain the entire system. This meets the real-time requirements of rapid recovery from power system failures.

[0045] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of the above-mentioned method for predicting the performance of an electric power information network based on a graph neural network.

[0046] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for predicting the performance of an electric power information network based on a graph neural network as described above.

[0047] Beneficial effects of the present invention: The advantage of this method is that it improves the accuracy of performance estimation and has the versatility and adaptability to different types of networks. It can provide network operation and maintenance personnel with real-time performance estimation results, which is helpful for network troubleshooting and optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 An overall flow chart of a method for predicting power information network performance based on graph neural networks provided in the first embodiment of the present invention.

[0050] Figure 2 Schematic diagram of a topological feature extractor for a method for predicting power information network performance based on a graph neural network, provided as the first embodiment of the present invention.

[0051] Figure 3 Schematic diagram of a flow feature extractor in a method for predicting power information network performance based on graph neural networks provided in the first embodiment of the present invention.

[0052] Figure 4 Schematic diagram of a topological feature purification extractor for a method for predicting power information network performance based on a graph neural network, provided as the first embodiment of the present invention.

[0053] Figure 5 Schematic diagram of a flow feature purifier for a method for predicting power information network performance based on graph neural networks, provided as the first embodiment of the present invention.

[0054] Figure 6 A system framework diagram of a graph neural network-based power information network performance prediction system provided in accordance with the second embodiment of the present invention.

[0055] Figure 7 A performance evaluation histogram of different methods under different network scales in a power information network performance prediction method based on a graph neural network provided in the third embodiment of the present invention.

[0056] Figure 8 A performance evaluation histogram under different network topologies of different methods of a power information network performance prediction method based on a graph neural network is provided as the third embodiment of the present invention.

[0057] Figure 9 A curve chart showing the influence of the feature purification module of a graph neural network-based power information network performance prediction method on the accuracy of the overall method provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0058] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0059] Example 1

[0060] Reference Figures 1 to 5 , as an embodiment of the present invention, provides a method for predicting the performance of a power information network based on a graph neural network, comprising the following steps:

[0061] S1: Build a graph neural network model and extract network topology features through the graph neural network model.

[0062] The principle of topological feature extractor is as follows Figure 2 As shown in the figure, the topological feature extractor uses GNN to extract topological features. It can encode network topology information by using the connection patterns and structural features between nodes. GNN can effectively learn the representation of nodes, including neighbor information and the position of nodes in the topological structure.

[0063] This case uses the classic implementation of GNN - Graph Convolutional Network (GCN). Of course, this part can also be assembled by other GNNs. We model the network topology as a graph structure, represented as , where V represents the set of all nodes in the graph, E represents the set of all edges, and each node is represented as a feature vector , the edge between node i and node j is represented as ,In the graph convolutional network, message passing is performed at each layer, expressed as,

[0064]

[0065] in, Indicates the Tier The feature vector of each node, represents the activation function, Representation node The set of neighbor nodes of It's the edge The normalization factor for normalization, Indicates the The weight matrix of the layer.

[0066] S2: Calculate the mutual influence between network flows based on time series technology.

[0067] The principle diagram of the inter-stream feature extractor is as follows Figure 3 As shown, the inter-stream feature extractor uses the classic encoder-decoder architecture for stream feature extraction.

[0068] Furthermore, time series technology uses Transformer to consider the time variability between flows to accurately describe the mutual influence between network flows.

[0069] Furthermore, the encoder part of the present invention adopts the Transformer encoder structure without position coding. The calculation of the mutual influence between network flows includes: the encoder receives the feature vector of the flow as input, uses the multi-head attention mechanism to extract the mutual influence relationship between the flows, and models the flows as a sequence ,in Indicates the number of flows in the network, and each flow is represented by a feature vector express.

[0070] The encoder uses a multi-head attention mechanism to linearly transform the input feature vector and convert it into a query ( ),key( ) and value ( ) vector, and calculate the dot product of the query and the key to obtain the attention weight, which is used to measure the correlation between each stream and other streams. Then, the attention weight is multiplied by the value vector to obtain the weighted stream feature representation. In this way, the encoder can capture the important correlation information between each stream and other streams.

[0071] The multi-head attention mechanism is expressed as,

[0072]

[0073] in, Represents the flow characteristics of different subspaces in high-dimensional space, Indicates the dimension of the input features.

[0074] The decoder part is a simple multi-layer perceptron that extracts inter-stream features from the high-dimensional spatial feature representation of the encoder's output.

[0075] S3: Extract node features related to the target flow through the topological information of the target flow.

[0076] Extracting node features related to the target flow includes purifying topological features based on a topological feature purifier. The principle diagram of the topological feature purifier is as follows: Figure 4 As shown, the topological feature purifier identifies the relationship between each node and the target flow through a node scoring mechanism, and uses a weighted multiplication operation to extract node features related to the target flow from the output of the topological feature extractor.

[0077] In the present invention, the output of the topological feature extractor is a feature matrix representing the entire topological structure, in which each row represents the feature vector of a node, which contains the information of the direct and indirect nodes of the node. However, for the target flow, we only focus on the nodes related to it. Therefore, we only need to set the weight of the nodes related to the target flow to 1 and the weight of the irrelevant nodes to 0 through the node scoring mechanism.

[0078] Then, the node features related to the target flow are extracted by weighted multiplication of the score and the topological features. By multiplying the node score and the topological feature matrix, the eigenvalues ​​of the nodes related to the target flow can be retained, while the eigenvalues ​​of the irrelevant nodes can be suppressed to zero.

[0079] The node features related to the target flow are purified and expressed as:

[0080]

[0081] Among them, x is the output of the topological feature extractor, representing a node feature, The path of the target flow.

[0082] Furthermore, the topological information of the target flow is used to extract node features related to the target flow to enhance the accuracy of network performance estimation.

[0083] S4: Calculate the correlation between all flows and the target flow.

[0084] The schematic diagram of the flow characteristic purifier is as follows Figure 5 As shown, we compute the correlation scores between all streams and the target stream, and we use these scores to correct the encoder output and obtain more fine-grained features for prediction.

[0085] Furthermore, the correlation degree is used to represent the degree of correlation between all flows and the target flow, and based on this, the inter-flow features are corrected to improve the accuracy of network performance estimation.

[0086] In the present invention, the output of the inter-flow feature purifier is a feature matrix representing all flows in the network, where each row represents a feature vector of a network flow. These feature vectors contain information about the target flow and other flows. However, for the target flow, we are more concerned with the features of flows that are highly correlated with it. Therefore, we need to calculate the correlation between all flows and the target flow through a scoring mechanism.

[0087] Through the scoring mechanism, we can calculate the correlation scores between each stream and the target stream. These scores can be used to measure the similarity or correlation between the two streams. Then, we perform a weighted multiplication operation on these scores with the stream features to extract features that are highly correlated with the target stream. In this way, we can obtain more accurate and targeted feature representations, which focus more on information related to the target stream.

[0088] The scoring mechanism is expressed as,

[0089]

[0090] in, Represent the characteristics of all flows and target flows in the network respectively.

[0091] S5: Fuse features to obtain network performance prediction results.

[0092] By fusing the features of S3 and S4, a more comprehensive and rich feature representation can be obtained for the prediction of the target flow.

[0093] By fusing the features of S3 and S4, we can obtain a more comprehensive and rich feature representation for the prediction of the target flow.

[0094] The feature fusion method used in this case is to concatenate these features and then send the fused features to a multilayer perceptron (MLP) for processing. In this case, we use a three-layer MLP structure, in which each neuron has a weight and a bias, and the ReLU function is used as the activation function to perform nonlinear transformation on the input.

[0095] The prediction is expressed as,

[0096]

[0097] in, represents the output of the topological feature refiner, represents the output of the inter-stream feature refiner, denote weights and biases respectively.

[0098] Example 2

[0099] Reference Figure 6 , as an embodiment of the present invention, provides a system for a power information network performance prediction method based on graph neural network. A power information network performance prediction system based on graph neural network includes a topology feature extraction module, an inter-flow feature extraction module, a target flow topology feature extraction module, a target flow inter-flow feature extraction module and a fusion module.

[0100] The topological feature extraction module is used to calculate the topological features of the power information network and can learn the representation of nodes.

[0101] The inter-flow feature extraction module is used to calculate the mutual influence between network flows, taking into account the time-varying nature of traffic.

[0102] The target flow topology feature extraction module is used to extract node features related to the target flow.

[0103] The target flow inter-flow feature extraction module is used to calculate the correlation between all flows and the target flow and correct the inter-flow features.

[0104] The fusion module is used to fuse features and obtain network performance estimation results.

[0105] If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0106] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0107] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, and then editing, interpreting, or processing in another suitable manner as necessary, and then storing it in a computer memory.

[0108] It should be understood that various components of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof may be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0109] Example 3

[0110] Reference Figure 7-Figure 9 In this embodiment, in order to verify the beneficial effects of the present invention, economic benefit calculation and simulation experiments are used for scientific demonstration. In order to demonstrate the superiority of our proposed method, two types of experimental tests are conducted: generalization experiments and ablation experiments.

[0111] First, a generalization experiment was conducted and the results are presented in Figure 7 and Figure 8 middle. Figure 7 The results show the comparison of the mean absolute percentage errors of the two models under different network topologies (NsfNet, Geant2) and different numbers of nodes (25, 30, 35). The error of the streaming network model proposed by this method is smaller than that of the routing network model, and the minimum error is only 49.20% of the error of the routing network model. This means that the streaming network model can more accurately predict the network performance under various scale network topologies, which shows that the streaming network model exhibits better stability than the routing network model in different network topologies. Therefore, it can be concluded that the streaming network model has stronger universal capabilities and stability under various network topologies. Furthermore, Figure 8The performance of our approach is demonstrated for five typical traffic scenarios: on-off traffic, constant bitrate traffic, modulated traffic, autocorrelated traffic, and mixed traffic. A bar chart compares the mean absolute percentage error (MAPE) of different models under different topologies and traffic combinations. First, in the on-off traffic scenario, the FlowSeer model demonstrates significant performance advantages over all baseline models (including the Routed Network model and the Routed Network-Elang model). Specifically, in the NsfNet topology, our proposed FlowSeer model achieves a 21.26% performance improvement, while in the Geant2 topology, the performance improvement is even more significant, reaching 28.9%. These results demonstrate that the FlowSeer model can more accurately predict network performance in on-off traffic scenarios. In the constant bitrate traffic scenario, the FlowSeer model outperforms the Routed Network-Elang model, indicating that the FlowSeer model can achieve better performance than the baseline models in constant bitrate traffic scenarios. In the modulated traffic scenario, the FlowSeer model demonstrates significant performance improvements, with performance gains of at least 28.32%. This further demonstrates the FlowSeer model's ability to more accurately predict network performance in modulated traffic scenarios. In the autocorrelated traffic scenario, the StreamView model outperformed other baseline models by up to 21.26%. In the mixed traffic scenario, the StreamView model also outperformed other baseline models. In all scenarios, the StreamView model achieved significant performance improvements, demonstrating its generalization and robustness.

[0112] Secondly, ablation experiments were conducted, and the results were as follows: Figure 9 The figure shows the impact of the feature purification module on the overall method accuracy. The vertical axis shows MAPE% (Mean Absolute Percentage Error), which measures the relative error between the predicted value and the true value. Lower values ​​indicate higher accuracy. The solid line in the figure represents the curve for the model including the feature purification module; the dashed line represents the curve for the model without the module. The MAPE% of the model including the purification module (solid line) is generally lower than that of the model without the module (dashed line) at all training epochs, demonstrating that the purification module effectively filters out irrelevant features and focuses on key information. The purification module reduces error across different topologies (NSFNET and GEANT2), demonstrating its universal applicability across various network structures. As the number of training epochs increases, the MAPE% of the model including the purification module decreases more steadily, indicating that the purification mechanism helps the model converge to a more optimal solution. The graph provides a more comprehensive understanding of the impact of the feature purification module on the overall method performance, further validating the effectiveness of our method.

[0113] Through these experimental tests, we have obtained strong experimental evidence to illustrate the superiority and reliability of the proposed method. These results have important guiding significance for the further improvement and application of our method.

[0114] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting power information network performance based on graph neural network, characterized in that: include: Build a graph neural network model and extract the topological features of the power information network through the graph neural network model; Calculate the mutual influence between network flows based on time series technology; Extract node features related to the target flow through the topological information of the target flow; Calculate the correlation between all flows and the target flow; Fusion features to obtain power information network performance prediction results; The extracting node features related to the target flow includes: purifying the topological features based on a topological feature purifier to purify the node features related to the target flow, the topological feature purifier identifying the relationship between each node and the target flow through a node scoring mechanism, and extracting the node features related to the target flow from the output of the topological feature extractor using a weighted multiplication operation; Purifying the topological features based on the topological feature purifier includes: outputting a feature matrix representing the entire topological structure based on the topological feature extractor, wherein each row represents a feature vector of a node, and the feature vector includes information about direct and indirect nodes of the node, and setting the weight of nodes related to the target flow to 1 and the weight of irrelevant nodes to 0 through a node scoring mechanism; The node features related to the target flow are extracted by weighted multiplication of the score and the topological features. The feature values ​​of the nodes related to the target flow are retained and the feature values ​​of the irrelevant nodes are suppressed to zero by multiplying the node score and the topological feature matrix. The node features related to the target flow are expressed as follows: Among them, x is the output of the topological feature extractor, representing a node feature, The path of the target flow.

2. The method for predicting power information network performance based on graph neural network according to claim 1, characterized in that: The topological characteristics of the power information network include connection relationships and topological characteristics between nodes; The extraction of network topology features includes: a topology feature extractor uses a graph neural network to extract topology features, uses the connection pattern and structural features between nodes to encode the power information network topology information, and the graph neural network uses a graph convolutional network to extract and model the power information network topology as a graph structure, which is expressed as , where V represents the set of all nodes in the graph, E represents the set of all edges, and each node is represented as a feature vector , the edge between node i and node j is represented as ,In the graph convolutional network, message passing is performed at each layer, expressed as, in, Indicates the Tier The feature vector of each node, represents the activation function, Representation node The set of neighbor nodes of It's the edge The normalization factor for normalization, Indicates the The weight matrix of the layer.

3. The method for predicting power information network performance based on graph neural network according to claim 2, characterized in that: The computation of the mutual influence between network flows includes: the encoder receives the feature vector of the flow as input, uses the multi-head attention mechanism to extract the mutual influence relationship between the flows, and models the flows as a sequence. ,in Indicates the number of flows in the network, and each flow is represented by a feature vector express; The encoder uses a multi-head attention mechanism to linearly transform the input feature vector into a query, key, and value vector. By calculating the dot product of the query and the key, the attention weight is obtained to measure the correlation between each stream and other streams. The attention weight is multiplied by the value vector to obtain a weighted stream feature representation. The multi-head attention mechanism is expressed as, in, Respectively represent the query, key, and value vectors generated by linear transformation, corresponding to the flow characteristics of different subspaces, Indicates the dimension of the input features.

4. The method for predicting power information network performance based on graph neural network according to claim 3, characterized in that: Calculating the correlation between all flows and the target flow includes calculating a correlation score between each flow and the target flow through a scoring mechanism, measuring the similarity or correlation between the two flows through the correlation score, performing a weighted multiplication operation on the correlation score and the flow features, and extracting features that are highly correlated with the target flow; The scoring mechanism is expressed as, in, Represent the characteristics of all flows and target flows in the network respectively.

5. The method for predicting power information network performance based on graph neural network according to claim 4, characterized in that: The fusing features to obtain the power information network performance prediction result includes fusing the features to obtain feature representation, and predicting the target flow based on the feature representation; The prediction is expressed as, in, represents the output of the topological feature refiner, represents the output of the inter-stream feature refiner, denote weights and biases respectively.

6. A system using the method for predicting power information network performance based on graph neural network according to any one of claims 1 to 5, characterized in that: It includes a topological feature extraction module, an inter-flow feature extraction module, a target flow topological feature extraction module, a target flow inter-flow feature extraction module and a fusion module; The topology feature extraction module is used to calculate the topology features of the power information network and can learn the representation of nodes; The inter-flow feature extraction module is used to calculate the mutual influence between network flows, taking into account the time-varying nature of traffic; The target flow topology feature extraction module is used to extract node features related to the target flow; The target flow inter-flow feature extraction module is used to calculate the correlation between all flows and the target flow, and correct the inter-flow features; The fusion module is used to fuse features to obtain power information network performance estimation results.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a method for predicting power information network performance based on graph neural network according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for predicting the performance of an electric power information network based on a graph neural network as described in any one of claims 1 to 5 are implemented.

Images