Electric power communication network risk assessment method based on digital twinborn and complex network theory

Abstract

The invention relates to an electric power communication network risk assessment method based on a digital twinborn and complex network theory, which comprises the following steps: firstly, effectively simulating the operation state of an electric power communication network by using a digital twinborn technology, and then comprehensively identifying and quantifying the risk in the electric power communication network, optimizing resource configuration, reducing the system fault probability through a multi-level risk assessment mechanism. And the safety and the reliability of the whole power communication network are improved. By constructing the digital twin model and applying the complex network theory, real-time monitoring, accurate evaluation and intelligent early warning of the operation risk of the electric power communication network are realized, and powerful technical support is provided for safe and stable operation of the electric power communication network. According to the method, the defects of the traditional risk assessment in real-time performance and accuracy are effectively made up, and the security and the anti-attack capability of the power communication network are remarkably enhanced. And meanwhile, dynamic risk identification and early warning of the power communication network are realized, and technical support is provided for constructing a more reliable network protection system.

Description

Technical Field

[0001] The present invention belongs to the technical field of power communication networks, and in particular to a risk assessment method for power communication networks based on digital twins and complex network theory. Background Art

[0002] In recent years, the power industry has actively promoted intelligent upgrading, and through technological innovation, management innovation and service innovation, it has accelerated the improvement of the intelligent level of the power system, promoted the modernization transformation and continuous optimization development of the industry. In this process, "strengthening the security of power communication networks and improving risk prevention and control capabilities" has become a key task, requiring the continuous improvement of the security and reliability of power communication networks through technical means to ensure the safe and stable operation of key information infrastructure. Combined with the development trend of smart grids and energy Internet, the construction of power communication networks is steadily advancing, while gradually achieving effective control and timely response to network risks.

[0003] As the key support of the power system, the security and stability of the power communication network are crucial to the reliable operation of the power grid. With the continuous improvement of the intelligence and informatization level of the power system, the security threats and operational risks faced by the power communication network are increasing. Specifically, it refers to security issues such as data leakage, service interruption, illegal intrusion, etc. that may occur in the process of power communication, as well as the resulting power grid operation risks. The risk management of power communication network is a multi-dimensional, interdisciplinary complex system engineering involving multiple fields such as network security, communication technology, and risk assessment. Through an in-depth analysis of the risk factors, vulnerabilities, and potential impacts of the existing power communication network, it is found that the current power communication network still has deficiencies in risk identification, assessment, and management.

[0004] At present, research on risk identification and management of power communication networks is still in its infancy. Based on traditional risk assessment methods, which often rely on empirical judgment and qualitative analysis, it is difficult to accurately quantify and monitor risks in real time. The risk assessment method of power communication networks based on digital twins and complex network theory provides a new technical means for risk management of power communication networks. However, how to combine digital twin technology to build an accurate network model, and how to use complex network theory for effective risk prediction and identification, are still hot topics and difficulties in current research. Summary of the invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and propose a risk assessment method for power communication networks based on digital twins and complex network theory. By constructing a digital twin model and applying complex network theory, real-time monitoring, accurate assessment and intelligent early warning of power communication network operation risks can be achieved, providing strong technical support for the safe and stable operation of the power communication network.

[0006] The present invention solves the technical problem by adopting the following technical solutions:

[0007] The risk assessment method of power communication network based on digital twin and complex network theory includes the following steps:

[0008] Step 1: Collect data on the power communication network to be modeled;

[0009] Step 2: Build a digital twin model, and use the collaboration between the demand server and the resource server to calculate the constraints and perform model training;

[0010] Step 3: Use the local model training loss function and the model long-term synchronization rate to continue training the model to ensure the accuracy and long-term synchronization of the digital twin model;

[0011] Step 4: Determine the relative importance of each service based on service requirements and list the number and characteristics of services distributed on the edge. This data is directly used as the input basis for subsequent multi-level risk identification standards to clarify the relationship between service distribution and network topology.

[0012] Step 5: After clarifying the importance and distribution characteristics of the business, establish a multi-level comprehensive risk identification standard and calculate the basic parameters of the network topology;

[0013] Step 6: Build an attack model according to steps 1 to 5 and attack the network constructed by the digital twin model, and calculate the network edge cross-layer information entropy value and the degree of network loss.

[0014] Moreover, the specific implementation method of step 1 is: the collected data includes topological structure, device status, index requirements and electromagnetic interference intensity between devices.

[0015] Moreover, the digital twin model constructed in the step 2 includes a physical layer, a perception layer, an edge layer and a central layer; wherein the physical layer includes power communication equipment, including various devices for power generation and distribution; the perception layer is deployed above the physical layer, and is used to collect equipment operating status and network environment data, and package this information into sample data, upload it to the edge layer and use it to drive the digital twin training of the local model; the edge layer is equipped with a small base station equipped with an edge server, which is used to receive status information from the perception layer and perform local model training; the edge server is divided into a demand server and a resource server: the demand server is used to receive sample data uploaded by the perception layer and carry out local model training; the resource server is used to provide computing resources to assist the demand server in completing parameter training tasks.

[0016] Moreover, the specific implementation method of calculating the constraint conditions and performing model training in step 2 is:

[0017] At the edge layer, collaboration occurs between the demand server and the resource server to calculate the edge collaboration decision variables:

[0018] X m,n (t)∈{0,1}

[0019] Where X m,n (t) = 1 indicates that the server d is required m Select Resource Server s Collaborate on model training, otherwise X m,n (t) = 0; in order to avoid the high complexity of model aggregation and reduce the efficiency of twin model synchronization, the server d m Select at most X at the same time m,max Resource servers collaborate at the edge, and are limited by their own data parallel receiving and processing capabilities. s In the same time slot, at most X n,max Demand servers provide collaborative training services:

[0020]

[0021] Use dynamic resource optimization methods to train digital twin models and combine demand servers m and resource server n s Together, we optimize the model and request server d in time slot t m and resource server n s The data transfer rate between is:

[0022]

[0023] Among them, R m,n (t), P m , G m,n (t) represents d m and n s The transmission bandwidth, transmission power and channel gain between them; N0 represents additive Gaussian white noise; δ m，n (t) represents the electromagnetic interference generated by the equipment in the communication network, then d m Select n s The maximum amount of sample data transmitted for collaborative training is:

[0024]

[0025] Among them, τ2 represents the time reserved for the demand server to transmit sample data. m Select n s The maximum amount of sample data that can be processed for collaborative training is:

[0026]

[0027] Among them, τ3 represents the time reserved for the resource server to perform collaborative training of sample data; f m,n (t) represents n in time slot t s Can be d m Provided computing resources.

[0028] Moreover, the specific implementation method of step 3 is: introducing a local model training loss function:

[0029]

[0030] Where L(·) represents the loss function of a single sample data; α i With β i They represent the input and output corresponding to the i-th sample data respectively. The loss function directly reflects the accuracy of the digital twin model of the power communication network and guides the update of the model parameters:

[0031]

[0032] Among them, λ is the update step size of the model; Denotes the loss function L m [ζ m (t),t] gradient;

[0033] At the same time, the model long-term synchronization rate is introduced to optimize the average sample data processing volume within the time period to characterize the overall synchronization situation:

[0034]

[0035] Moreover, the specific implementation method of step 4 is: in the research of power communication services, the multi-dimensional evaluation of service quality QoS is a key link to ensure the stable operation of the power system; among them, the service performance indicators include delay, bit error rate, real-time, reliability and security, and these indicators are assigned scores between 1 and 8 to quantify their impact on the overall performance of the power system.

[0036] Moreover, the specific implementation method of step 5 is: the multi-level comprehensive risk identification criteria include network topology, edge betweenness, edge business importance, edge traffic importance, physical topology layer importance, cross-layer importance and edge cross-layer information entropy;

[0037] Build network topology:

[0038] G=(V,E)

[0039] Among them, G represents the network graph, V represents the set of nodes in the network such as substations, distribution stations and relay stations, and E represents the set of links connecting these nodes in the network;

[0040] Calculate edge betweenness:

[0041]

[0042] Among them, n ij (e) represents the number of edges e that the shortest path between node i and node j passes through, n ij represents the number of all shortest paths between i and j;

[0043] Calculate edge business importance EBI:

[0044]

[0045] Among them, EBI n represents the edge business importance of the nth edge, w m represents the sum of the service importance of all services between the mth source-destination node pair, d nm is a 0-1 variable. When the nth edge is on the path between the mth source and sink node pair, d nm =1, otherwise d nm =0;

[0046] Calculate edge flow importance:

[0047]

[0048] Among them, ETI n , represents the edge flow importance of the nth edge, t m It represents the sum of the normalized traffic of all services between the mth source-destination node pair.

[0049] The edge betweenness is used to represent the importance of the edge in the physical topology layer;

[0050] Calculate cross-layer importance ECI:

[0051] ECI n =EB n ×EBTI n

[0052] Among them, ECI n Represents the cross-layer importance of the nth edge, EBTI n =(1-λ)EBI n +λETI n ,

[0053] Calculate the edge cross-layer information entropy ECE:

[0054]

[0055] Where ECI' is the normalized value of ECI:

[0056]

[0057] Moreover, the specific implementation method of step 6 is: the attack mode includes an optimization attack failure model, a greedy attack model, a random attack model and a Bernoulli failure model;

[0058] Optimized attack failure model:

[0059] The set of links that fail due to the attack is the set of h links that cause the greatest loss of service importance in the entire network. The service vulnerability of the network under this attack model is

[0060] B Lmax (x) = max{L Loss (R h )},h / L=x=0,1 / L,...,1

[0061] Greedy attack model:

[0062] The failed link set R is a set of h links obtained through greedy search. The first link attacked is the link that causes the greatest loss of importance of the entire network service; the h+1th link attacked is the link that causes the greatest loss of importance of the current network service. The current network is the network after h links fail. The network service vulnerability is:

[0063] B Lgre (x) = L Loss (R h ),h / L=x=0,1 / L,...,1

[0064] Random attack model

[0065] This model belongs to the random link failure model. The failed link set R is a set of h links randomly selected by the attacker. The service vulnerability of the network is expressed by the average value of the service importance loss caused by the failure of any h links:

[0066]

[0067] Bernoulli failure model

[0068] Each link fails naturally with probability, and the failure probability between links is independent. The vulnerability of the network is expressed as the mathematical expectation of the loss of business importance:

[0069]

[0070] Assume that the attacked edge cluster consists of the first x edges with the largest EB value in the network, x = 0, 1…, N, N is the number of nodes in the network. When x>1, let the matrix D and E x The corresponding row vectors are e1, e2, …e xThe loss of the network after the attack is:

[0071]

[0072] Where P m Represents the mth element of vector P, P = e1∨e2∨…∨e x , ∨ represents the logical “or” operation, and λ is the weight distribution coefficient of business importance and traffic importance.

[0073] The advantages and positive effects of the present invention are:

[0074] 1. Aiming at the complexity of risk identification in the power communication network, the present invention proposes a multi-level comprehensive risk identification method integrating digital twins. The method first uses digital twin technology to effectively simulate the operating status of the power communication network, and then uses a multi-level risk assessment mechanism to comprehensively identify and quantify the risks in the power communication network, optimize resource allocation, reduce the probability of system failure, and improve the safety and reliability of the entire power communication network. By constructing a digital twin model and applying complex network theory, the present invention realizes real-time monitoring, accurate assessment and intelligent early warning of the operating risks of the power communication network, providing strong technical support for the safe and stable operation of the power communication network.

[0075] 2. This patent is based on digital twin technology to build a high-precision simulation model of the power communication network, and realizes the risk identification of the power communication network through the quantitative analysis ability of complex network theory. Utilizing the characteristic of digital twin technology of "doing what normal cannot do", the failure scenarios of key nodes or links under optimization attack, greedy attack, random attack and Bernoulli failure model are simulated, and the degree of loss of the network under different attack modes is quantitatively evaluated. By calculating indicators such as edge cross-layer information entropy (ECE), high-risk nodes and links in the network are identified, providing a scientific basis for the formulation of subsequent defense strategies. The present invention effectively makes up for the shortcomings of traditional risk assessment in real-time and accuracy, and significantly enhances the security and anti-attack capability of the power communication network by identifying high-risk areas and deploying defense measures or optimizing business allocation. The present invention realizes dynamic risk identification and early warning of the power communication network, and provides technical support for building a more reliable network protection system. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a topological diagram of the electric power communication network of the present invention;

[0077] Figure 2 Constructing a schematic diagram for the digital twin model of the present invention;

[0078] Figure 3 A network vulnerability analysis curve diagram based on service importance according to an embodiment of the present invention;

[0079] Figure 4 It is a loss curve diagram of different ECE networks when being attacked according to an embodiment of the present invention. DETAILED DESCRIPTION

[0080] The present invention is further described in detail below with reference to the accompanying drawings.

[0081] The risk assessment method of power communication network based on digital twin and complex network theory includes the following steps:

[0082] Step 1: Collect data on the power communication network to be modeled. The collected data includes topology, equipment status, index requirements, and electromagnetic interference intensity between equipment.

[0083] Step 2: Build a digital twin model and use the collaboration between the demand server and the resource server to calculate the constraints and perform model training.

[0084] The present invention divides the construction of the digital twin model of the power communication network into four levels, namely the physical layer, the perception layer, the edge layer and the center layer. The physical layer mainly includes power communication equipment, including various devices for power generation and distribution; the perception layer is deployed above the physical layer, responsible for collecting equipment operating status and network environment data, and packaging this information into sample data, and uploading it to the edge layer to drive the digital twin training of the local model. Considering the wide application of edge computing, the edge layer is equipped with a small base station equipped with an edge server, which is used to receive status information from the perception layer and perform local model training. According to different computing resources, edge servers can be divided into demand servers and resource servers: the former is responsible for receiving sample data uploaded by the perception layer and carrying out local model training; the latter provides computing resources to assist the demand server in completing parameter training tasks.

[0085] At the edge layer, there will be collaboration between the demand server and the resource server, so the edge collaboration decision variable is defined

[0086] X m,n (t)∈{0,1}(1)

[0087] Where X m,n (t) = 1 indicates that the server d is required m Select Resource Server s Collaborate on model training, otherwise X m,n (t) = 0. In order to avoid the high complexity of model aggregation which will reduce the efficiency of twin model synchronization, the server d m Select at most X at the same time m,max In addition, due to the limitation of its own data parallel receiving and processing capabilities, resource servers n s In the same time slot, at most X n,maxThe demand server provides collaborative training services, namely

[0088]

[0089] The digital twin model is trained using a dynamic resource optimization method, which is characterized by combining the demand server d m and resource server n s Optimize the model together. Therefore, in time slot t, server d is required m and resource server n s The data transfer rate between

[0090]

[0091] Among them, R m,n (t), P m , G m,n (t) represents d m and n s The transmission bandwidth, transmission power and channel gain between them; N0 represents additive Gaussian white noise; δ m，n (t) represents the electromagnetic interference generated by the equipment in the communication network. Then d m Select n s The maximum amount of sample data transmitted for collaborative training is

[0092]

[0093] Where τ2 represents the time reserved for the demand server to transmit sample data. m Select n s The maximum amount of sample data that can be processed for collaborative training is

[0094]

[0095] Where τ3 represents the time reserved for the resource server to perform collaborative training on sample data; f m,n (t) represents n in time slot t s Can be d m Provided computing resources.

[0096] Step 3: Use the local model training loss function and the model long-term synchronization rate to continue training the model to ensure the accuracy and long-term synchronization of the digital twin model.

[0097] In order to better train the model, the present invention introduces a local model training loss function:

[0098]

[0099] Where: L(·) represents the loss function of a single sample data; αi With β i They represent the input and output corresponding to the i-th sample data respectively. The loss function directly reflects the accuracy of the digital twin model of the power communication network and guides the update of the model parameters, that is,

[0100]

[0101] Where λ is the update step size of the model; Denotes the loss function L m [ζ m (t),t]’s gradient.

[0102] Digital twin is a data-driven technology. The efficiency of its model training is closely related to the amount of sample data completed in each time slot. The larger the amount of sample data, the higher the degree of synchronization of the digital twin model of the power communication network with the real physical environment. In addition, the synchronization between the digital twin model and the physical entity has obvious long-term characteristics, not just limited to instantaneous or short-term performance. Based on this, the patent of this invention introduces the concept of "model long-term synchronization rate" to characterize the overall synchronization situation by optimizing the average sample data processing volume within the time period.

[0103]

[0104] Step 4: Determine the relative importance of each business according to business requirements and list the number and characteristics of business distribution on the edge. This data is directly used as the input basis for subsequent multi-level risk identification standards to clarify the relationship between business distribution and network topology.

[0105] In the study of power communication services, the multi-dimensional evaluation of service quality (QoS) is a key link to ensure the stable operation of the power system. Service performance indicators include latency, bit error rate, real-time, reliability and security, which are assigned scores between 1 and 8 to quantify their impact on the overall performance of the power system. This scoring system provides a standardized basis for the division of priorities for power communication services. Specifically, the high requirements for latency are demonstrated by the strict standard of ≤10ms, especially in key services such as 500kV and 220kV relay protection, with an importance score of 8, indicating high requirements for real-time control services. For services such as energy metering and telemetry, although the real-time requirements are lower, their reliability and security still occupy an important position, so they receive higher scores. Table 1 provides basic data support for the subsequent risk identification of the power communication network.

[0106] To simplify the calculation, the 13 types of power services in Table 1 are classified into five categories according to their importance; 500kV relay protection and 220kV relay protection are Class I services; safety and stability systems are Class II services; wide-area measurement, dispatching automation, dispatching telephones and electric energy metering telemetry are Class III services; video conferencing, substations, video monitoring, protection information management, and lightning location monitoring are Class IV services; administrative telephones and office automation are Class V services.

[0107] Table 1 shows the power business, index requirements and relative importance values

[0108]

[0109] exist Figure 1 Set the number of service types on each link and perform service weight w m calculate.

[0110] W is the distribution of service importance between all source-destination node pairs in the network. W(s, d) represents the sum of service importance of all services between source node s and destination node d, referred to as service importance sum.

[0111]

[0112] w m is the normalized service importance sum between the mth source-destination node pair, where represents the number of elements in a set; ·| represents the number of elements in a set. Table 2 shows the service distribution in the network.

[0113] Table 2 Service distribution in the network

[0114] link Type × Quantity <![CDATA[w m ]]> link Type × Quantity <![CDATA[w m ]]> <![CDATA[e 1,3 ]]> Ⅱ×5+Ⅲ×20+Ⅳ×5+Ⅴ×10 0.118 <![CDATA[e 8,9 ]]> Ⅰ×3+Ⅳ×4 0.011 <![CDATA[e 2,3 ]]> Ⅱ×3+Ⅲ×12+Ⅳ×3+Ⅴ×6 0.097 <![CDATA[e 9,10 ]]> Ⅱ×4+Ⅴ×7 0.023 <![CDATA[e 2,5 ]]> Ⅱ×2+Ⅲ×8+Ⅳ×2+Ⅴ×4 0.034 <![CDATA[e 10,13 ]]> Ⅰ×1+Ⅴ×2 0.007 <![CDATA[e 3,4 ]]> Ⅱ×2+Ⅲ×8+Ⅳ×2+Ⅴ×4 0.034 <![CDATA[e 6,11 ]]> Ⅰ×1+Ⅴ×2 0.007 <![CDATA[e 3,8 ]]> Ⅱ×2+Ⅲ×7+Ⅳ×3+Ⅴ×5 0.045 <![CDATA[e 11,12 ]]> Ⅱ×1+Ⅲ×2+Ⅳ×3+Ⅴ×3 0.051 <![CDATA[e 4,6 ]]> Ⅱ×4+Ⅲ×8+Ⅳ×6+Ⅴ×4 0.064 <![CDATA[e 11,13 ]]> Ⅰ×2+Ⅲ×8+Ⅳ×2+Ⅴ×5 0.141 <![CDATA[e 4,7 ]]> Ⅱ×2+Ⅲ×20+Ⅳ×5+Ⅴ×10 0.074 <![CDATA[e 12,14 ]]> Ⅲ×5+Ⅳ×6 0.152 <![CDATA[e 7,8 ]]> Ⅱ×3+Ⅲ×12+Ⅳ×2+Ⅴ×10 0.142 <![CDATA[e 13,14 ]]> Ⅰ×1+Ⅲ×8+Ⅳ×6+Ⅴ×4 0.108

[0115] Step 5: After clarifying the business importance and distribution characteristics, establish a multi-level comprehensive risk identification standard and calculate the basic parameters of the network topology to provide accurate parameter support for subsequent attack model construction and network loss assessment.

[0116] During the research process of the present invention, through in-depth investigation of extensive analysis results, it was observed that the statistical characteristics of power communication networks significantly deviate from the typical behavior of regular networks and random networks. This network exhibits unique topological structural characteristics, and its complexity is reflected in the non-uniformity of node connections and the diversity of network dynamics. Based on these observations, the present invention defines this type of network as a "complex network" and divides it into seven indicators for systematic definition and feature extraction.

[0117] Network topology:

[0118] G=(V,E) (14)

[0119] Among them, G represents the network graph, V represents the set of nodes in the network such as substations, distribution stations, relay stations, etc., and E represents the set of links connecting these nodes in the network.

[0120] Edge betweenness

[0121]

[0122] n ij (e) represents the number of edges e that the shortest path between node i and node j passes through, n ij Represents the number of all shortest paths between i and j.

[0123] Edge Business Importance (EBI)

[0124]

[0125] Among them, EBI n represents the edge business importance of the nth edge, w m represents the sum of the service importance of all services between the mth source-destination node pair, d nm is a 0-1 variable. When the nth edge is on the path between the mth source and sink node pair, d nm =1, otherwise d nm =0.

[0126] Edge Flow Importance (ETI)

[0127]

[0128] Among them, ETI n , represents the edge flow importance of the nth edge, t m It represents the sum of the normalized traffic of all services between the mth source-destination node pair.

[0129] Edge physical topology layer importance

[0130] The power communication network has now become a typical complex network, and the routing protocol of the power communication network is still mainly based on the Open Shortest Path First protocol (OSPF). Therefore, the betweenness of the edge indicates the number of business paths passing through the edge. The present invention defines the edge betweenness (EB) as the importance of the edge in the physical topology layer of the power communication network, which is used to indicate the importance of the edge in this layer. The calculation method of the edge betweenness is shown in formula (15).

[0131] Cross-layer Importance (ECI)

[0132] ECI n =EB n ×EBTI n (18)

[0133] Among them, ECIn Represents the cross-layer importance of the nth edge, EBTI n =(1-λ)EBI n +λETI n

[0134] Edge Cross-Layer Information Entropy (ECE)

[0135]

[0136] Among them, ECI' is the normalized value of ECI, that is,

[0137]

[0138] According to the concept of information entropy in information theory, when the probability of all random events occurring is the same, the entropy value reaches the maximum. In formula (20), ECI n ' can be understood as the distribution probability of ECI. The more balanced the distribution of ECI is, the closer its information entropy is to the maximum value log2(N). That is, the closer the ECE value of the network is to 1, the lower the risk of the network; conversely, the higher the risk of the network.

[0139] Step 6: Build an attack model and attack the network constructed by the digital twin model, calculate the network edge cross-layer information entropy value and the degree of network loss.

[0140] Optimizing attack failure model

[0141] The set of links that fail due to the attack is the set of h links that cause the greatest loss of service importance in the entire network. The service vulnerability of the network under this attack model is

[0142] B Lmax (x) = max{L Loss (R h )},h / L=x=0,1 / L,...,1 (10)

[0143] Greedy attack model

[0144] The failed link set R is a set of h links obtained through greedy search. The first link attacked is the link that causes the greatest loss of the importance of the entire network service. The h+1th link attacked is the link that causes the greatest loss of the importance of the current network service. The current network is the network after the failure of h links. The service vulnerability of the network is

[0145] B Lgre (x) = L Loss (R h ),h / L=x=0,1 / L,...,1 (11)

[0146] Random attack model

[0147] This model belongs to the random link failure model. The failed link set R is a set of h links randomly selected by the attacker. The service vulnerability of the network is expressed by the average value of the service importance loss caused by the failure of any h links, that is,

[0148]

[0149] Bernoulli failure model

[0150] Each link fails naturally with probability, and the failure probability between links is independent. The vulnerability of the network is expressed as the mathematical expectation of the loss of business importance, that is,

[0151]

[0152] Network loss (Loss(E x ))

[0153] Assume that the attacked edge cluster consists of the first x edges with the largest EB value in the network, x = 0, 1…, N, N is the number of nodes in the network. When x>1, let the matrix D and E x The corresponding row vectors are e1, e2, …e x The loss of the network after the attack is

[0154]

[0155] Where P m Represents the mth element of vector P, P = e1∨e2∨…∨e x , ∨ represents the logical “or” operation, and λ is the weight distribution coefficient of business importance and traffic importance.

[0156] like Figure 3 The comparison of network loss after different links are attacked and failed under four attack modes, namely the optimization attack model, greedy attack model, random attack model and Bernoulli failure model, is shown. The figure compares the changing trend of network loss under each attack mode as the number of attacked links increases, thus reflecting the difference in the impact of different attack modes on network vulnerability. As can be seen from the figure, the optimization attack model has the most significant impact on the network loss, while the random attack model and Bernoulli failure model have relatively low loss. This result further verifies that the destructive ability of different attack models on the network is closely related to their strategy selection.

[0157] like Figure 4The comparison of network loss of the power communication network under different ECE values ​​after being attacked by the optimization attack model, greedy attack model, random attack model and Bernoulli failure model is shown. The figure reveals the correlation between network risk and ECE value, indicating that the higher the ECE value, the lower the network loss, that is, the stronger the risk resistance of the network. It can be seen from the figure that when the ECE value is low, the network loss in all attack modes increases significantly, especially the optimization attack mode shows greater destructive power. This analysis shows that improving the ECE value plays an important role in enhancing the overall stability and anti-attack ability of the network.

[0158] Based on digital twin technology, this patent constructs a high-precision simulation model of the power communication network, and realizes risk identification of the power communication network through the quantitative analysis capability of complex network theory. By using the characteristic of digital twin technology of "doing what normal cannot do", the failure scenarios of key nodes or links under optimization attack, greedy attack, random attack and Bernoulli failure model are simulated, and the degree of loss of the network under different attack modes is quantitatively evaluated. By calculating indicators such as edge cross-layer information entropy (ECE), high-risk nodes and links in the network are identified, providing a scientific basis for the formulation of subsequent defense strategies.

[0159] This method effectively makes up for the shortcomings of traditional risk assessment in terms of real-time and accuracy. By identifying high-risk areas and deploying defense measures or optimizing business allocation, it significantly enhances the security and anti-attack capabilities of the power communication network. The present invention realizes dynamic risk identification and early warning of the power communication network, providing technical support for building a more reliable network protection system.

[0160] It should be emphasized that the embodiments described in the present invention are illustrative rather than restrictive. Therefore, the present invention includes but is not limited to the embodiments described in the specific implementation manner. Any other implementation manners derived by those skilled in the art based on the technical solution of the present invention also fall within the scope of protection of the present invention.

Claims

1. A risk assessment method for power communication networks based on digital twins and complex network theory, characterized by: The following steps are involved: Step 1: Collect data on the power communication network to be modeled; Step 2: Build a digital twin model, and use the collaboration between the demand server and the resource server to calculate the constraints and perform model training; Step 3: Use the local model training loss function and the model long-term synchronization rate to continue training the model to ensure the accuracy and long-term synchronization of the digital twin model; Step 4: Determine the relative importance of each service based on service requirements and list the number and characteristics of services distributed on the edge. This data is directly used as the input basis for subsequent multi-level risk identification standards to clarify the relationship between service distribution and network topology. Step 5: After clarifying the importance and distribution characteristics of the business, establish a multi-level comprehensive risk identification standard and calculate the basic parameters of the network topology; Step 6: Build an attack model according to steps 1 to 5 and attack the network constructed by the digital twin model, and calculate the network edge cross-layer information entropy value and the degree of network loss.

2. The power communication network risk assessment method based on digital twin and complex network theory according to claim 1 is characterized by: The specific implementation method of step 1 is: the collected data includes topological structure, device status, index requirements and electromagnetic interference intensity between devices.

3. The power communication network risk assessment method based on digital twin and complex network theory according to claim 1 is characterized by: The digital twin model constructed in the step 2 includes a physical layer, a perception layer, an edge layer and a central layer; wherein the physical layer includes power communication equipment, including various devices for power generation and distribution; the perception layer is deployed above the physical layer, and is used to collect equipment operating status and network environment data, and package this information into sample data, upload it to the edge layer and use it to drive the digital twin training of the local model; the edge layer is equipped with a small base station equipped with an edge server, which is used to receive status information from the perception layer and perform local model training; the edge server is divided into a demand server and a resource server: the demand server is used to receive sample data uploaded by the perception layer and carry out local model training; the resource server is used to provide computing resources to assist the demand server in completing parameter training tasks.

4. The power communication network risk assessment method based on digital twin and complex network theory according to claim 1 is characterized by: The specific implementation method of calculating the constraint conditions and performing model training in step 2 is: At the edge layer, collaboration occurs between the demand server and the resource server to calculate the edge collaboration decision variables: X m,n (t)∈{0,1} Where X m,n (t) = 1 indicates that server d is required m Select Resource Server s Collaborate on model training, otherwise X m,n (t) = 0; in order to avoid the high complexity of model aggregation and reduce the efficiency of twin model synchronization, the server d m Select at most X at the same time m,max Resource servers collaborate at the edge, and are limited by their own data parallel receiving and processing capabilities. s In the same time slot, at most X n,max Demand servers provide collaborative training services: Use dynamic resource optimization methods to train digital twin models and combine demand servers m and resource server n s Together, we optimize the model and request server d in time slot t m and resource server n s The data transfer rate between is: Among them, R m,n (t), P m , G m,n (t) represents d m and n s The transmission bandwidth, transmission power and channel gain between them; N0 represents additive Gaussian white noise; δ m，n (t) represents the electromagnetic interference generated by the equipment in the communication network, then d m Select n s The maximum amount of sample data transmitted for collaborative training is: Among them, τ2 represents the time reserved for the demand server to transmit sample data. m Select n s The maximum amount of sample data that can be processed for collaborative training is: Among them, τ3 represents the time reserved for the resource server to perform collaborative training of sample data; f m,n (t) represents n in time slot t s Can be d m Provided computing resources.

5. The power communication network risk assessment method based on digital twin and complex network theory according to claim 1 is characterized by: The specific implementation method of step 3 is: introducing a local model training loss function: Where L(·) represents the loss function of a single sample data; α i With β i They represent the input and output corresponding to the i-th sample data respectively. The loss function directly reflects the accuracy of the digital twin model of the power communication network and guides the update of the model parameters: Among them, λ is the update step size of the model; Denotes the loss function L m [ζ m (t),t] gradient; At the same time, the model long-term synchronization rate is introduced to optimize the average sample data processing volume within the time period to characterize the overall synchronization situation:

6. The power communication network risk assessment method based on digital twin and complex network theory according to claim 1 is characterized by: The specific implementation method of step 4 is as follows: in the research on power communication services, the multi-dimensional evaluation of service quality QoS is a key link to ensure the stable operation of the power system; among them, the service performance indicators include delay, bit error rate, real-time, reliability and security, and these indicators are assigned scores between 1 and 8 to quantify their impact on the overall performance of the power system.

7. The power communication network risk assessment method based on digital twin and complex network theory according to claim 1 is characterized by: The specific implementation method of step 5 is: the multi-level comprehensive risk identification criteria include network topology, edge betweenness, edge business importance, edge traffic importance, physical topology layer importance, cross-layer importance and edge cross-layer information entropy; Build network topology: G=(V,E) Among them, G represents the network graph, V represents the set of nodes in the network such as substations, distribution stations and relay stations, and E represents the set of links connecting these nodes in the network; Calculate edge betweenness: Among them, n ij (e) represents the number of edges e that the shortest path between node i and node j passes through, n ij represents the number of all shortest paths between i and j; Calculate edge business importance EBI: Among them, EBI n represents the edge business importance of the nth edge, w m represents the sum of the service importance of all services between the mth source-destination node pair, d nm is a 0-1 variable. When the nth edge is on the path between the mth source and sink node pair, d nm =1, otherwise d nm =0; Calculate edge flow importance: Among them, ETI n , represents the edge flow importance of the nth edge, t m It represents the sum of the normalized traffic of all services between the mth source-destination node pair. The edge betweenness is used to represent the importance of the edge in the physical topology layer; Calculate cross-layer importance ECI: ECI n =EB n ×EBTI n Among them, ECI n Represents the cross-layer importance of the nth edge, EBTI n =(1-λ)EBI n +λETI n , Calculate the edge cross-layer information entropy ECE: Where ECI' is the normalized value of ECI:

8. The power communication network risk assessment method based on digital twin and complex network theory according to claim 1 is characterized by: The specific implementation method of step 6 is: the attack mode includes an optimization attack failure model, a greedy attack model, a random attack model and a Bernoulli failure model; Optimized attack failure model: The set of links that fail due to the attack is the set of h links that cause the greatest loss of service importance in the entire network. The service vulnerability of the network under this attack model is B Lmax (x)=max{L Loss (R h )},h / L=x=0,1 / L,...,1 Greedy attack model: The failed link set R is a set of h links obtained through greedy search. The first link attacked is the link that causes the greatest loss of importance of the entire network service; the h+1th link attacked is the link that causes the greatest loss of importance of the current network service. The current network is the network after h links fail. The network service vulnerability is: B Lgre (x)=L Loss (R h ),h / L=x=0,1 / L,...,1 Random attack model This model belongs to the random link failure model. The failed link set R is a set of h links randomly selected by the attacker. The service vulnerability of the network is expressed by the average value of the service importance loss caused by the failure of any h links: Bernoulli failure model Each link fails naturally with probability, and the failure probability between links is independent. The vulnerability of the network is expressed as the mathematical expectation of the loss of business importance: Assume that the attacked edge cluster consists of the first x edges with the largest EB value in the network, x = 0, 1…, N, N is the number of nodes in the network. When x>1, let the matrix D and E x The corresponding row vectors are e1, e2, …e x The loss of the network after the attack is: Where P m Represents the mth element of vector P, P = e1∨e2∨…∨e x , ∨ represents the logical "or" operation, and λ is the weight distribution coefficient of business importance and traffic importance.

Images