Communication network fault prediction and self-healing system based on digital twinning

Abstract

The application is suitable for the technical field of communication network, and provides a communication network fault prediction and self-healing system based on digital twinning, which executes an operation mechanism including the following steps: in step S1, actual telemetry state data of a node in a physical communication network within a current time window and ideal benchmark state data synchronously output by a digital twinning benchmark model are acquired, and a twinning state deviation scalar quantifying a local state degradation degree of the node is calculated in combination with a feature distribution of a historical normal operation state of the node; in step S2, the physical communication network is modeled as a graph topology structure composed of nodes and edges, and a diffusion dynamics mechanism based on a spatial topology is utilized to take the twinning state deviation scalar obtained in the step S1 as an external excitation source to predict a fault evolution probability of each node in the whole network at a future time step.

Description

Technical Field

[0001] This invention belongs to the field of communication network technology, and in particular relates to a communication network fault prediction and self-healing system based on digital twins. Background Technology

[0002] As the core infrastructure of the information society, the reliability of communication networks directly affects the stable operation of the national economy and social life. With the rapid deployment of 5G / 6G mobile communications, large-scale IoT terminal access, and edge computing nodes, modern communication networks exhibit significant characteristics such as a surge in the number of nodes, dynamic changes in topology, and uneven distribution of service traffic. The exponential expansion of network scale leads to the multi-source and concealed nature of fault causes, posing a severe challenge to traditional fault management mechanisms.

[0003] Existing network management systems generally adopt an alarm mode based on single-point threshold triggering, which only initiates the response process when the node performance parameters exceed a preset threshold. This mechanism ignores the inherent topological interconnectivity of communication networks and cannot capture the diffusion effect of node state degradation in the spatial dimension and the cumulative process in the temporal dimension.

[0004] In the fault self-healing stage, current technical solutions mainly rely on static routing table switching or pre-configured redundant link activation strategies. These methods simplify resource scheduling to a priority ranking problem, failing to establish a quantitative correlation model between resource allocation utility and network topology importance. When the network encounters distributed faults, fixed rules can easily lead to an over-concentration of self-healing resources in local areas, resulting in a contradictory phenomenon of idle resources on critical nodes and overloaded resources on non-critical nodes.

[0005] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0006] The purpose of this invention is to provide a communication network fault prediction and self-healing system based on digital twins, aiming to solve the above-mentioned problems.

[0007] This invention is implemented as follows: a communication network fault prediction and self-healing system based on digital twins, wherein the system executes an operating mechanism including the following steps: Step S1: Obtain the actual telemetry status data of the node in the physical communication network within the current time window, as well as the ideal reference status data synchronously output by the digital twin reference model, and calculate the twin state deviation scalar that quantifies the degree of local state degradation of the node by combining the characteristic distribution of the node's historical normal operation status. Step S2: Model the physical communication network as a graph topology structure composed of nodes and edges. Using the diffusion dynamics mechanism based on spatial topology, the twin state deviation scalar obtained in step S1 is used as an external excitation source to predict the failure evolution probability of each node in the network in the future time step. Step S3: Construct the network self-healing process as a static optimal quadratic resource allocation problem with constraints on resource conversion utility and call cost. Based on the fault evolution probability output in step S2, analyze and solve the optimal self-healing resource dynamic reconstruction strategy that minimizes the combined cost of instantaneous fault risk and reconstruction cost. Step S4: Before executing the optimal self-healing resource dynamic reconstruction strategy in the physical communication network, inject it into the digital twin benchmark model. Utilize the algebraic connectivity characteristics of the network topology and the total resource consumption to perform closed-loop verification and comprehensive evaluation of the overall network structure stability and connectivity robustness after strategy execution. If the comprehensive evaluation index meets the preset security threshold, then execute the strategy in the physical communication network; otherwise, return to step S3 and re-solve the strategy by increasing the resource call cost parameter.

[0008] A further technical solution involves calculating the twin state deviation scalar using multidimensional phase space mapping in step S1. The specific calculation method is as follows:

[0009] First, subtract the actual telemetry state column vector of the physical network node at the current moment from the ideal state column vector synchronously output by the digital twin baseline model to obtain the deviation column vector. Then, multiply the transpose of this deviation column vector by the inverse matrix of the covariance matrix of the node's historical normal operating state characteristics, and then multiply by the aforementioned deviation column vector to obtain the instantaneous deviation quadratic value at that moment. Next, integrate and accumulate the instantaneous deviation quadratic values ​​of all moments within an integration time window length from the current moment. Finally, divide the accumulated integral value by the integration time window length to obtain the quantized scalar of the twin state deviation at the current moment.

[0010] A further technical solution involves calculating the fault evolution probability based on graph Laplace dynamics. The specific calculation method is as follows:

[0011] First, calculate the product of the spatial propagation rate coefficient of the fault in the network topology and the prediction time step, and then multiply this product by the product of the initial fault probability column vector of all nodes in the network at the current time to obtain the first term; then calculate the deviation excitation coupling coefficient multiplied by the prediction time step, and then multiply it by the global deviation excitation column vector formed by the aggregation of the twin state deviations of all nodes to obtain the second term; add the first term and the second term to obtain the fault evolution probability column vector of all nodes in the network at the future time.

[0012] A further technical solution involves using a static optimal quadratic control algorithm in step S3 to solve for the optimal self-healing resource dynamic reconfiguration strategy. The specific calculation method is as follows:

[0013] First, calculate the transpose of the self-healing resource utility mapping matrix, multiply it by the positive definite diagonal matrix of the network node's fault risk penalty, and then multiply it by the self-healing resource utility mapping matrix to obtain the first matrix term. Then, add the first matrix term to the positive definite diagonal matrix of the self-healing resource call cost, and invert the sum matrix to obtain the inverse matrix. Next, calculate the transpose of the self-healing resource utility mapping matrix, multiply it by the positive definite diagonal matrix of the fault risk penalty, and then multiply it by the future fault evolution probability column vector predicted in step S2 to obtain the column vector term. Finally, multiply the above inverse matrix by the above column vector term to obtain the optimal self-healing resource allocation strategy column vector at the current time.

[0014] A further technical solution involves calculating the network structure stability evaluation index in step S4 using graph theory. The specific calculation method is as follows:

[0015] First, calculate the sum matrix obtained by adding the original Laplace matrix to the increment of the Laplace matrix generated by the optimal self-healing resource allocation strategy. Then, calculate the second smallest eigenvalue of this sum matrix to obtain the corrected algebraic connectivity. Next, calculate the largest eigenvalue of the sum matrix. Then, divide the corrected algebraic connectivity by the largest eigenvalue of the sum matrix to obtain the topology connectivity ratio. Next, calculate the L1 norm of the column vector of the optimal self-healing resource allocation strategy to obtain the absolute amount of total system resources consumed in this self-healing process. Divide this absolute amount of total resources by the total limit of the maximum self-healing resource capacity pool of the communication network, square the ratio, and then subtract the squared value from one to obtain the resource consumption attenuation penalty term. Finally, multiply the topology connectivity ratio by the resource consumption attenuation penalty term to obtain the comprehensive evaluation index of network structure stability after applying the self-healing strategy.

[0016] In a further technical solution, in step S4, if the comprehensive evaluation index does not meet the preset safety threshold, the diagonal elements of the positive definite diagonal matrix of the self-healing resource call cost in step S3 are incremented by a preset step size, and steps S3 and S4 are re-executed for iterative optimization; if the comprehensive evaluation index still does not meet the safety threshold after the number of iterations reaches the preset maximum number of iterations, a system alarm is triggered and the current self-healing strategy is maintained and not executed, waiting for manual intervention.

[0017] A further technical solution is that the Laplace matrix increment is constructed in the following manner:

[0018] First, for each node in the communication network, calculate the amount of self-healing resources allocated to that node multiplied by the resource-connection gain coefficient to obtain the resource gain weight of that node; then multiply the resource gain weight by the local Laplacian matrix centered on that node to obtain the contribution matrix of that node to the Laplacian matrix increment; finally, sum the contribution matrices of all nodes to obtain the Laplacian matrix increment generated by the optimal self-healing resource allocation strategy.

[0019] A further technical solution involves a rolling time-domain optimization mechanism in step S3, where the optimal self-healing resource allocation strategy is solved: at each sampling time, a cumulative cost function is constructed using the fault evolution probability sequence over multiple future time steps predicted in step S2 as input. The cumulative cost function is calculated as follows:

[0020] For each future prediction step, calculate the transpose of the failure probability column vector for that step, multiply it by the positive definite diagonal matrix of the failure risk penalty, and then multiply it by the failure probability column vector for that step to obtain the failure risk cost term. Then, calculate the transpose of the self-healing resource allocation strategy column vector corresponding to that step, multiply it by the positive definite diagonal matrix of the self-healing resource call cost, and then multiply it by the self-healing resource allocation strategy column vector to obtain the resource call cost term. Add the above two terms to obtain the single-step cost for that step. Then, sum the single-step costs of all prediction steps to obtain the cumulative cost function. Solve for the control sequence that minimizes the cumulative cost function, and output only the first element of the control sequence as the optimal self-healing resource allocation strategy at the current time. When the next sampling time arrives, repeat the above process and re-predict and optimize with the latest system state.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0022] In step S1, this invention employs multi-dimensional phase space mapping. By accumulating a weighted quadratic deviation between the actual telemetry state and the ideal state output by the digital twin baseline model, and using the inverse matrix of the characteristic covariance matrix of the historical normal operating state for decoupling and normalization, the influence of different physical dimensions is eliminated. This mechanism can sensitively capture minute anomalies of nodes in multiple dimensions while suppressing false alarms caused by normal fluctuations. It provides accurate, stable, and physically meaningful local degradation measurements for fault prediction, overcoming the insensitivity of traditional single-point threshold alarm modes to early, minor faults.

[0023] In step S2, this invention abstracts the communication network into a graph topology and uses the graph Laplace dynamics equations, taking local twin state deviations as external excitation sources to drive the diffusion of fault probabilities along the topological connections in the network. This model not only characterizes the spatial propagation effect and temporal accumulation process of faults, but also ensures the numerical stability of iterative predictions by introducing stability constraints. It can reliably predict the fault evolution probability of each node in complex dynamic networks over the long term, breaking through the bottleneck of existing systems lacking the ability to model the spatiotemporal evolution of the entire network.

[0024] In step S3 of this invention, the self-healing resource allocation is modeled as an optimal quadratic form control problem with a resource conversion utility matrix and a call cost penalty matrix, and the optimal strategy that minimizes the combined cost of instantaneous failure risk and reconstruction cost is analytically solved. This scheme reflects the topological importance of nodes through a positive definite diagonal matrix of failure risk penalty and suppresses network oscillations caused by over-adjustment through a resource call cost matrix. This ensures that resource allocation is based not only on failure probability but also on node criticality and call economy, avoiding the problems of local resource overload or idleness caused by traditional static routing switching or fixed priority strategies. Attached Figure Description

[0025] Figure 1 Here is the logic block diagram for the twin state deviation calculation module; Figure 2 Here is the logic block diagram of the fault evolution probability prediction module; Figure 3 Logic block diagram for solving the dynamic reconfiguration strategy; Figure 4 This is a logic block diagram for the strategy closed-loop verification and evaluation module. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0027] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0028] like Figures 1-4 As shown, an embodiment of the present invention provides a communication network fault prediction and self-healing system based on digital twins, wherein the system executes an operating mechanism including the following steps:

[0029] Step S1: Obtain the actual telemetry status data of the node in the physical communication network within the current time window, as well as the ideal reference status data synchronously output by the digital twin reference model, and calculate the twin state deviation scalar that quantifies the degree of local state degradation of the node by combining the characteristic distribution of the node's historical normal operation status.

[0030] For example, telemetry data such as CPU utilization, memory usage, interface traffic, and packet loss rate can be periodically collected from network devices. This data can be organized into time series or vector formats. The digital twin baseline model then outputs ideal state data within the same time window based on a preset normal operating mode. This is used to calculate the twin state deviation scalar.

[0031] Step S2: Model the physical communication network as a graph topology structure composed of nodes and edges. Using the diffusion dynamics mechanism based on spatial topology, the twin state deviation scalar obtained in step S1 is used as an external excitation source to predict the failure evolution probability of each node in the network in the future time step.

[0032] For example, devices such as routers, switches, and base stations in a communication network can be abstracted as nodes in a graph, and the physical or logical connections between them can be abstracted as edges. The weight of an edge can reflect the bandwidth, latency, or reliability of the connection. One implementation is to treat a node as a fault "hotspot" when its twin state deviation scalar is high, and assume that the fault risk will propagate to its neighboring nodes at a fixed rate. This propagation can be simply achieved by iteratively updating the fault probability of each node, i.e., the fault probability of each node is the sum of its own current probability and the probability contributions from its neighboring nodes.

[0033] Step S3: Construct the network self-healing process as a static optimal quadratic resource allocation problem with constraints on resource conversion utility and call cost. Based on the fault evolution probability output in step S2, analyze and solve the optimal self-healing resource dynamic reconstruction strategy that minimizes the combined cost of instantaneous fault risk and reconstruction cost.

[0034] For example, when a high probability of failure evolution is predicted in the network, the system needs to decide how to allocate limited self-healing resources (such as backup bandwidth, computing resources, backup links, etc.) to reduce the risk.

[0035] Step S4: Before executing the optimal self-healing resource dynamic reconstruction strategy in the physical communication network, inject it into the digital twin benchmark model. Utilize the algebraic connectivity characteristics of the network topology and the total resource consumption to perform closed-loop verification and comprehensive evaluation of the overall network structure stability and connectivity robustness after strategy execution. If the comprehensive evaluation index meets the preset security threshold, then execute the strategy in the physical communication network; otherwise, return to step S3 and re-solve the strategy by increasing the resource call cost parameter.

[0036] For example, after obtaining a self-healing strategy, instead of immediately deploying it to the physical network, the strategy's execution effect is first simulated in a digital twin model. One implementation is to simply check whether there are isolated nodes or excessive loads on critical links in the network after the strategy is implemented. Simultaneously, the total resources consumed by the strategy are calculated and compared with the system's total resource capacity. If resource consumption is excessive or isolated nodes appear, the strategy is considered insecure.

[0037] In this embodiment, the communication network refers to a system consisting of a series of interconnected communication devices (such as routers, switches, base stations, etc.) and transmission links, used to realize the transmission and exchange of information such as data, voice, and video. This network can be a wireless network, a wired network, or a hybrid of both.

[0038] Physical communication networks specifically refer to the actual deployed and operating communication infrastructure, including hardware equipment, optical fibers, cables, and the real network environment that carries service traffic.

[0039] Telemetry status data refers to various operational parameters and performance indicators collected in real time from nodes in a physical communication network, such as traffic, latency, packet loss rate, CPU utilization, memory utilization, and temperature. This data reflects the current operational status of the nodes.

[0040] A digital twin baseline model is a virtual model used in a digital twin system to characterize the behavior patterns and performance baseline of a physical communication network under ideal or normal operating conditions. This model is trained on historical data or built using expert experience, and can output ideal state data under specific input conditions.

[0041] The twin state deviation scalar is a numerical indicator that quantifies the difference between the actual operating state of a physical network node and the ideal state output by the digital twin baseline model. This scalar reflects the degree of degradation of the node's local state; a larger value indicates a more severe deviation from the normal baseline.

[0042] Graph topology refers to the structure that abstracts nodes (such as devices) and edges (such as links) in a communication network into vertices and edges in mathematical graph theory. This structure can intuitively represent the connections and interactions between elements in the network.

[0043] The diffusion dynamics mechanism of spatial topology refers to the dynamic process of how a certain "matter" (such as fault risk) propagates from one node to its neighboring nodes and spreads throughout the network over time on a graph topology. This mechanism takes into account the influence of network connectivity strength and topology on propagation.

[0044] Fault evolution probability refers to the likelihood that a node in a communication network will fail within a certain time step in the future. This probability is dynamic and is affected by factors such as network topology, node state deviations, and external stimuli.

[0045] Network self-healing refers to the process by which a communication network, when detecting a potential or actual fault, automatically adjusts its network configuration, resource allocation, or routing strategies to restore network functionality, reduce the impact of the fault, and maintain service continuity.

[0046] The static optimal quadratic resource allocation problem models resource allocation during network self-healing as a mathematical optimization problem. The objective function is typically quadratic and is subject to constraints such as resource conversion utility, resource allocation cost, and other conditions. Solving this problem yields the optimal resource allocation scheme under given conditions.

[0047] The optimal self-healing resource dynamic reconfiguration strategy refers to a specific scheme obtained by solving a static optimal quadratic resource allocation problem, used to guide the resource adjustment and configuration optimization of physical communication networks. This strategy is dynamic and updates in real time according to changes in the network state.

[0048] Algebraic connectivity is an important metric in graph theory used to measure network connectivity and robustness, typically represented by the second smallest eigenvalue of the graph Laplacian matrix. A larger value indicates a stronger ability of the network to maintain connectivity when faced with node or link failures.

[0049] Closed-loop verification and comprehensive evaluation refers to simulating the self-healing strategy in a digital twin model before its actual implementation, and evaluating various aspects such as network performance, structural stability, and resource consumption after the strategy is implemented. This process forms a feedback loop to ensure the effectiveness and security of the strategy.

[0050] The safety threshold is a standard used in closed-loop verification and comprehensive evaluation to determine whether a self-healing strategy meets preset safety and performance requirements. If the evaluation index is lower than this threshold, the strategy is considered to have risks and needs to be adjusted.

[0051] In a preferred embodiment of the present invention, the twin state deviation scalar is calculated using multidimensional phase space mapping, and the specific calculation formula is as follows: in, The quantization scalar represents the twin state deviation of node i at the current time t, which characterizes the weighted cumulative deviation energy of the node state from the baseline model. Its value is a dimensionless normalized deviation measure, which is weighted by the inverse of the variance of each telemetry physical quantity to eliminate the influence of different physical dimensions. The integral time window length represents the state sampling; t represents the current system time. This represents the integral variable within the integration time window; This represents the column vector of the actual telemetry state of physical network node i at time τ; This represents the ideal state column vector synchronously output by the digital twin baseline model at time τ; This is the matrix transpose symbol; The inverse matrix represents the feature covariance matrix of the historical normal operation state of node i, used to eliminate the variance differences in each dimension of the actual telemetry state column vector and achieve decoupling; the covariance matrix If it is a positive definite matrix; if it is a positive semi-definite matrix, then use... After performing regularization, the inverse is calculated, where δ is a minimal positive number. This represents an identity matrix with the same dimension as the node telemetry state vector (i.e., if the node telemetry state column vector...). If the dimension is m, then (where m is the identity matrix) is a dimensionless parameter.

[0052] In this embodiment, the multidimensional phase space mapping calculation of the twin state deviation scalar refers to mapping multiple physical quantity measurements (consisting of a multidimensional state vector) of a node at a certain moment into a single scalar value through mathematical transformation. This scalar value can comprehensively reflect the degree of deviation of the multidimensional state. In addition to the weighted quadratic form based on the inverse of the covariance matrix used in this application, other methods can also be used. For example, principal component analysis can be used to reduce the dimensionality of the multidimensional state vector, and then the Euclidean distance between the dimensionality-reduced vector and the ideal state vector can be calculated as the deviation scalar; or, a regression model based on a neural network can be constructed, taking the multidimensional state vector as input and directly outputting a scalar value representing the degree of deviation.

[0053] Twin state deviation quantization scalar It is a dimensionless normalized deviation measure whose magnitude directly reflects the degree to which the node state deviates from the ideal baseline model. This scalar integrates the deviation information of multidimensional states through weighted accumulation, providing a unified and physically meaningful input for subsequent fault prediction.

[0054] Actual telemetry state column vector It includes multiple real-time monitoring data points of physical network nodes at time τ, such as CPU utilization, memory usage, network interface bandwidth utilization, packet loss rate, latency, and error packet count. This data can be collected in real time using standard network management protocols, streaming data analysis techniques, or custom probes deployed on network devices.

[0055] Ideal state column vector A digital twin baseline model simulates or predicts the multidimensional state values ​​of a node under healthy and normal operating conditions based on factors such as current network configuration, traffic load, service demand, and environmental conditions. This ideal state column vector can be generated based on physical models or simulation models, or it can be learned and predicted through statistical analysis of historical normal operating data and machine learning modeling (such as Gaussian process regression, long short-term memory networks, etc.).

[0056] The inverse matrix of the covariance matrix of historical normal operation status characteristics It plays a crucial role in weighting and decoupling in the calculation. It utilizes the statistical correlation and fluctuation characteristics among various telemetry physical quantities during normal node operation. By multiplying the deviation vector with this inverse matrix, the influence of different physical quantities can be eliminated, and weighting is performed according to the inverse of the variance of each physical quantity. This ensures that physical quantities with smaller fluctuations under normal conditions contribute more to the total deviation when they exhibit the same absolute deviation. Simultaneously, the off-diagonal elements of the covariance matrix reflect the correlation between different physical quantities. Its inverse matrix decouples these correlations, avoiding redundancy or mutual interference between multidimensional data, thus more accurately reflecting the true deviation of the node's state. This covariance matrix... It can be obtained by continuously collecting multidimensional telemetry data from nodes during the long-term stable operation of the network and performing statistical calculations.

[0057] When the covariance matrix When the data is positive semidefinite (e.g., some telemetry dimensions are almost invariant during normal operation, or there is a strict linear correlation between different dimensions), direct inversion can lead to numerical instability or irreversibility. In this case, using... The matrix is ​​inverted after regularization, where δ is a very small positive number. This regularization method, such as Tikhonov regularization, ensures that the matrix becomes positive definite and invertible by adding a small positive value to the diagonal, thus guaranteeing computational robustness and numerical stability.

[0058] In a preferred embodiment of the present invention, in step S2, the failure evolution probability is calculated based on graph Laplace dynamics, and the specific calculation formula is as follows: in, This represents the column vector of predicted fault evolution probabilities for all nodes in the entire network at the predicted future time t+Δt, and is a dimensionless parameter. The coefficient representing the spatial propagation rate of a fault in the network topology is the reciprocal of the time dimension; Δt represents the prediction time step and has the time dimension. The Laplace matrix represents the topology of a communication network, reflecting the strength of physical or logical connections between network nodes, and is a dimensionless parameter. This represents the initial fault probability column vector of all nodes in the network at time t, and is a dimensionless parameter. This represents the deviation excitation coupling coefficient, with the dimension being the reciprocal of the time dimension; This represents the deviation of all node twin states calculated by step S1. A globally aggregated bias excitation column vector; to ensure the numerical stability of the iterative prediction, the prediction time step Δt must satisfy the stability constraint condition. ,in For the Laplace matrix The largest eigenvalue. For the fault evolution probability column vector obtained in this iteration. Each component in Where i represents the node number in the communication network, if If so, then the component will be forcibly assigned a value of 0; if If the value of the component is 1, then the component will be forcibly assigned the value 1; if the original value of the component is within the closed interval [0,1], then its calculation result will remain unchanged.

[0059] In this embodiment, graph Laplace dynamics is a mathematical framework used to describe the diffusion process of information, matter, or state on a network. It utilizes the Laplace matrix of a graph to capture the influence of network topology on dynamic processes, effectively simulating the propagation behavior of faults in communication networks. This dynamic model can be used to analyze various diffusion phenomena in networks, such as information flow, heat conduction, or fault propagation. Its core lies in quantifying the mutual influence through the connection strength between nodes. The fault evolution probability is a quantitative indicator used to represent the likelihood of each node in the communication network failing at a future time. This probability value can provide network administrators with a forward-looking fault risk assessment, thereby supporting preventative maintenance and resource allocation decisions. This probability can be a dimensionless numerical value reflecting the degree of deviation from the node's health status, or it can be the result of a comprehensive evaluation based on historical fault data and real-time monitoring data.

[0060] In the above formula This represents the predicted failure probability of all nodes in the entire network after a time step Δt. This predicted value is based on the failure probability at the current moment. and the global deviation activation column vector aggregated from twin state deviations The result of combined action. Among them, item... The intrinsic diffusion and decay process of fault probability in network topology is described, where As a Laplace matrix, it precisely characterizes the connection relationships and strength between nodes, and This modulates the rate of this diffusion. Another factor... This introduces an external stimulus, namely the direct impact of the degree of local node degradation detected by the digital twin model on the failure evolution. This is the coupling coefficient of the bias excitation. To ensure the numerical stability of the prediction process and avoid unreasonable divergence or oscillations in the model during iterative calculations, the prediction time step Δt is limited to a specific range, i.e., it needs to satisfy the stability constraint condition. This condition is expressed by the Laplace matrix. Maximum eigenvalue This determination ensures the convergence and physical rationality of the model at discrete time steps.

[0061] The solution in this application uses the node local twin state deviation obtained in step S1 above. Using the external excitation source and the communication network graph topology established in step S2 above, a graph Laplace dynamics model is employed to dynamically predict the failure evolution probability of each node in the entire network. Specifically, this scheme uses the deviation excitation coupling coefficient to transmit local degradation information. This is incorporated into the dynamic equations to drive the propagation of fault probabilities within the network. Simultaneously, the Laplace matrix... By capturing the strength of physical or logical connections between network nodes, this method ensures that the propagation path and speed of faults accurately reflect the impact of the network topology. This approach not only considers the degradation state of individual nodes but, more importantly, quantifies and predicts how this degradation affects other nodes through network connections, ultimately leading to changes in the overall network failure risk. By introducing stability constraints, the reliability and validity of the prediction results are ensured, avoiding potential numerical instability issues that may arise during long-term predictions. Therefore, this scheme can transform discrete, localized state deviation information into continuous, global fault evolution trends, providing crucial predictive basis for subsequent self-healing decisions.

[0062] In a preferred embodiment of the present invention, in step S3, the optimal self-healing resource dynamic reconstruction strategy is solved using a static optimal quadratic control algorithm, and the specific calculation formula is as follows: in, This represents the column vector of the optimal self-healing resource allocation strategy solved at time t, which represents the amount of self-healing resources allocated to each node and has the absolute dimension of resources. This represents the self-healing resource utility mapping matrix, whose dimension is the reciprocal of the absolute dimension of the resource, representing the magnitude of how a unit of self-healing resource is converted into the probability of offsetting failure. The transpose of the self-healing resource utility mapping matrix; The fault risk penalty positive definite diagonal matrix represents the fault risk penalty of a network node, reflecting the topological importance of the node in the communication network, and is a dimensionless parameter. This represents a positive definite diagonal matrix representing the cost of self-healing resource allocation, used to penalize network oscillations caused by over-adjustment. Its dimension is the negative square of the absolute dimension of the resource (i.e., ...). ), to ensure with To ensure that the dimensions of the matrix items are consistent. It has absolute dimensions of resources and guarantees that the matrix inversion operation holds true. -1 is the symbol for matrix inversion. The column vector of fault evolution probability at the future time t+Δt is the output of step S2.

[0063] In this embodiment, the static optimal quadratic control algorithm is a classic optimization method. Its core lies in determining the optimal control input by minimizing a quadratic objective function. This algorithm can provide analytical solutions or efficient numerical solutions and is suitable for systems requiring fast response and precise control.

[0064] Optimal self-healing resource allocation strategy column vector It is the amount of self-healing resources allocated by the system to each communication network node at the current moment. These resources can be bandwidth, computing power, backup link capacity or power supply, etc., and their role is to directly guide the physical communication network to perform self-healing operations.

[0065] Self-healing resource utility mapping matrix A quantitative relationship was established between the amount of self-healing resources allocated and the effect of reducing the probability of failure. The elements of this relationship can be determined by statistical analysis of historical data, setting based on expert experience, or by modeling the importance of network topology and node functions.

[0066] Positive definite diagonal matrix of network node failure risk penalty This is used to weight the failure risks of different nodes, so that the system can prioritize nodes that have a greater impact on the overall stability of the network or critical services when allocating resources. Its diagonal elements can be set according to topological indicators such as degree centrality and betweenness centrality of nodes, or according to business indicators such as the amount of services carried by the nodes and service level agreements.

[0067] Self-healing resource retrieval cost positive definite diagonal matrix This is used to quantify and penalize the cost of using self-healing resources, avoiding excessive or unnecessary resource calls. Its diagonal elements can be set according to factors such as the actual economic cost of the resource, energy consumption, deployment difficulty, or impact on existing business.

[0068] The proposed solution precisely models the resource allocation problem during the self-healing process of a communication network as a static optimal quadratic control problem and solves it analytically. Specifically, the system first uses the future fault evolution probability column vector predicted in step S2. Combined with a pre-defined self-healing resource utility mapping matrix Network node failure risk penalty positive definite diagonal matrix And the positive definite diagonal matrix of self-healing resource call costs Construct a quadratic optimization problem. The goal of this optimization problem is to find a set of optimal self-healing resource allocation strategy column vectors. This approach minimizes the risk of transient failures while also considering the cost of resource allocation. By performing matrix operations on the above formulas, the optimal self-healing resource allocation strategy column vector can be directly and analytically solved. This method closely integrates fault prediction results with resource allocation decisions, making the generation of self-healing strategies no longer a simple empirical judgment, but a scientific decision based on quantitative models and optimization algorithms. By introducing fault risk penalties and resource allocation costs, the system can achieve a balance between reducing fault risk and controlling resource consumption, thereby generating a self-healing strategy that is both effective and economical.

[0069] In a preferred embodiment of the present invention, in step S4, the network structure stability evaluation index is calculated using graph theory, and the specific calculation formula is as follows: in, λ represents the comprehensive evaluation index of network structure stability after applying the self-healing strategy, and is a dimensionless parameter; λ2 represents the second smallest eigenvalue of the corresponding matrix, i.e., the algebraic connectivity, and is a dimensionless parameter. The Laplace matrix representing the original communication network topology is a dimensionless parameter. This indicates the optimal self-healing resource allocation strategy. The incremental Laplacian matrix generated by the mapping is used to correct the off-diagonal equivalent edge weights between nodes to improve network connectivity; The largest eigenvalue of the sum matrix (i.e., the sum matrix obtained by adding the original Laplacian matrix to the Laplacian matrix increment) is a dimensionless parameter. The L1 norm of the column vector representing the optimal self-healing resource allocation strategy is used to characterize the absolute amount of total system resources consumed in this self-healing process. It represents the total limit of the maximum self-healing resource capacity pool that the communication network can provide, and has the absolute dimension of resources; It is a resource consumption quadratic decay penalty term used to suppress high-cost self-healing behavior that approaches the total limit.

[0070] In this embodiment, the comprehensive evaluation index of network structure stability This index is used to quantify the overall structural stability and connectivity robustness of communication networks after applying self-healing strategies. It is a dimensionless parameter whose value directly reflects the network's resistance and recovery ability in the face of faults or disturbances. The calculation of this index comprehensively considers the network's topological connectivity enhancement effect and the cost of self-healing resource consumption, aiming to provide a comprehensive and quantifiable evaluation result to decide whether to implement the current self-healing strategy. Here, λ² represents the second smallest eigenvalue of the corresponding matrix, i.e., algebraic connectivity, an important concept in graph theory used to measure the strength of network connectivity and robustness. A larger value indicates a stronger ability of the network to maintain connectivity when nodes or edges are removed, i.e., better network connectivity robustness. In practical applications, it can be obtained by calculating the eigenvalues ​​of the network's Laplacian matrix, or estimated through iterative algorithms or approximation methods.

[0071] This indicates the optimal self-healing resource allocation strategy. This refers to the modification or enhancement of the original network topology. It improves network connectivity by adjusting the equivalent edge weights between nodes, thus reflecting the impact of self-healing resources on the strength of physical or logical connections in the network. The incremental matrix can be constructed in various ways. For example, besides using a resource-weighted local Laplacian matrix, it can directly modify the off-diagonal elements related to the affected nodes in the original Laplacian matrix based on resource allocation, or enhance connectivity by introducing virtual edges. The L1 norm of the column vector representing the optimal self-healing resource allocation strategy is the sum of the absolute values ​​of all elements in the vector. Here, it is used to characterize the absolute amount of total system resources consumed in this self-healing operation. It provides an intuitive metric reflecting the total amount of resources invested to achieve the current self-healing goal. This represents the total limit of the maximum self-healing resource capacity pool that the communication network can provide. It is a preset upper limit with absolute resource dimensions, used to constrain the resource consumption of the self-healing strategy and prevent excessive self-healing from exhausting system resources or causing other negative impacts.

[0072] Resource consumption quadratic decay penalty term It is a dimensionless factor used to suppress the self-healing strategy from approaching the total resource limit. High-cost behavior. When resource consumption... near When this happens, the value of the penalty item will decrease significantly, thereby lowering the overall evaluation index. This causes the system to tend to choose strategies with lower resource consumption. This quadratic decay mechanism can effectively balance the relationship between self-healing effect and resource consumption, avoiding resource waste.

[0073] In the aforementioned communication network fault prediction and self-healing system based on digital twins, in order to ensure the optimal self-healing resource dynamic reconfiguration strategy is found... To ensure the security and effectiveness of implementation in real-world physical networks, this application introduces a closed-loop verification and comprehensive evaluation mechanism. The core of this mechanism lies in calculating a comprehensive evaluation index of network structural stability using graph theory before policy execution. Specifically, the system first obtains the Laplace matrix of the original communication network topology. This matrix reflects the strength of physical or logical connections between network nodes. Next, based on the optimal self-healing resource allocation strategy to be verified... Generate a Laplace matrix increment This incremental matrix characterizes the potential corrective or enhancing effects of the self-healing strategy on the network topology, such as improving the connection strength between certain nodes through resource allocation. Subsequently, the original Laplace matrix is... With Laplace matrix increment Adding these together yields a modified Laplacian matrix that reflects the network topology changes after policy implementation. The second smallest eigenvalue of this modified matrix, namely the algebraic connectivity, is then calculated. This allows for the quantification of network connectivity robustness after policy execution. A higher algebraic connectivity indicates a stronger ability of the network to maintain connectivity in the face of failures. Simultaneously, to avoid excessive resource consumption, the system also calculates the L1 norm of the column vector of the optimal self-healing resource allocation policy. This represents the absolute amount of total system resources consumed in this self-healing operation. This resource consumption is relative to the total limit of the network's maximum self-healing resource capacity pool. The values ​​are compared and weighted by a quadratic decay penalty term for resource consumption. This penalty term effectively suppresses high-cost self-healing behavior that approaches the total limit, prompting the system to ensure network stability while also considering the economy of resource use. Finally, the algebraic connectivity is compared with the largest eigenvalue of the sum matrix (i.e., the sum matrix obtained by adding the original Laplace matrix to the Laplace matrix increment). The ratio of the two values ​​is multiplied by a quadratic attenuation penalty term for resource consumption to obtain a comprehensive evaluation index for network structure stability. This metric comprehensively reflects the improvement in network connectivity robustness and the rationality of resource consumption after strategy implementation. In this way, the system can perform comprehensive closed-loop verification of the predicted self-healing strategy, ensuring that it improves network stability without causing new risks due to excessive resource consumption. If this comprehensive evaluation metric... If the preset security threshold is met, the strategy is considered safe and effective, and can be executed in the physical communication network; otherwise, the system will return to step S3 and adjust the resource call cost parameters. The strategy is re-evaluated until a strategy that can effectively cope with faults while meeting safety and resource constraints is found. This mechanism effectively compensates for the network structure stability and resource economy issues that may be ignored by simply relying on fault probability prediction and resource optimization, making the entire fault prediction and self-healing system more robust and practical.

[0074] In a preferred embodiment of the present invention, in step S4, if the comprehensive evaluation index does not meet the preset safety threshold, the positive definite diagonal matrix of self-healing resource call cost in step S3 is increased by a preset step size. The diagonal elements are then processed, and steps S3 and S4 are re-executed for iterative optimization. If the comprehensive evaluation index still does not meet the safety threshold after the number of iterations reaches the preset maximum number of iterations, a system alarm is triggered and the current self-healing strategy is maintained and not executed, awaiting manual intervention.

[0075] In this embodiment, when the comprehensive evaluation index does not meet the preset safety threshold, the system will increment the positive definite diagonal matrix of the self-healing resource call cost in step S3 by a preset step size. The diagonal elements. An increasing self-healing resource call cost positive definite diagonal matrix. The diagonal elements of the equation imply that when solving for the optimal self-healing strategy, the system considers resource usage more costly, thus prompting the optimization algorithm to choose strategies that consume fewer resources or have less impact on the network. This preset step size can be a fixed increment, for example, increasing the step size by a factor of 100 in each iteration. The diagonal elements are increased by a constant, or it could be a percentage increment, for example, by increasing the diagonal elements in each iteration. The diagonal elements are increased by a certain percentage of their current value. In some cases, the step size can also be dynamically adjusted based on how much the comprehensive evaluation index deviates from the safety threshold. For example, if the evaluation index is much lower than the threshold, a larger step size is used to accelerate convergence.

[0076] After increasing the cost of self-healing resource invocation, the system will re-execute steps S3 and S4 for iterative optimization. Re-execution of steps S3 (strategy solving) and S4 (strategy verification) forms a feedback loop, enabling the system to adjust parameters based on the previous verification results (i.e.,...). The system then performs a matrix-based algorithm and recalculates and validates the strategy. This process continues until a strategy that meets the safety threshold is found or the preset maximum number of iterations is reached. The maximum number of iterations is set to prevent the system from getting stuck in an infinite loop and to ensure that a decision is made within a finite amount of time. This maximum number of iterations can be a fixed integer value, such as 5 or 10, or it can be dynamically adjusted according to the size of the network, real-time requirements, or computational resource limitations.

[0077] If the comprehensive evaluation index still fails to meet the safety threshold after the preset maximum number of iterations, the system will trigger an alarm and maintain the current self-healing strategy, awaiting manual intervention. The purpose of triggering the system alarm is to notify operators that the current automated self-healing strategy cannot meet safety requirements and requires manual intervention. Alarms can be implemented in various ways, such as audible and visual alarms, sending SMS or email notifications, or displaying warning information on the network monitoring interface. Maintaining the current self-healing strategy means that the system will prevent any self-healing operations that might cause network instability from being sent to the physical network, thereby avoiding the introduction of new risks. Awaiting manual intervention entrusts decision-making power to experienced operators, who can judge and handle the situation based on the alarm information and the actual network conditions, such as conducting deeper fault diagnosis, manually adjusting network configurations, or deploying contingency plans.

[0078] Furthermore, as a preferred embodiment of the present invention, in order to improve convergence efficiency and avoid strategy oscillations that may be caused by unidirectional adjustment of resource cost parameters during the above iterative optimization process, the system first determines the total consumption of the current self-healing resource allocation strategy before executing each iteration. The deviation direction relative to the preset optimal resource consumption range. If the total consumption exceeds the upper limit of the preset range, the positive definite diagonal matrix of the self-healing resource call cost is increased by a preset step size. The diagonal elements are used to suppress redundant resource allocation; if the total consumption is lower than the lower limit of the preset range, the amount is reduced by a preset step size. The diagonal elements are used to free up necessary resources to ensure enhanced network connectivity. Simultaneously, a comprehensive evaluation metric is recorded after each iteration. If two consecutive iterations If the improvement in the value is less than a preset convergence threshold (e.g., the relative change is less than 1%), then the current strategy is determined to have approached the feasible optimal solution, the iteration is terminated early, and the current strategy is output without waiting for the maximum number of iterations. This preferred embodiment is only a specific illustration of the iterative adjustment mechanism and does not constitute the sole limitation on the iterative steps described in the claims.

[0079] As a preferred embodiment of the present invention, the Laplace matrix increment Construct it in the following manner: in, Indicates the total number of nodes in the communication network; The resource-connectivity gain coefficient, with the dimension being the reciprocal of the absolute dimension of the resource, is used to determine the amount of self-healing resources allocated to node k. The dimensionless gain mapped to the topological connectivity strength; Let represent the local Laplacian matrix centered at node k. If the network topology is an unweighted graph, its off-diagonal elements are -1 between node k and each of its neighbors. If the network topology is a weighted graph, the corresponding element is the negative of the edge weight, and the diagonal elements are the sum of the weights of all neighboring edges of that node. The diagonal elements represent the degree of node k, and the remaining elements are 0, representing dimensionless parameters. The increment of the Laplacian matrix is... The self-healing resource is obtained by weighting and summing the basis matrices contributed by all nodes according to their respective resource allocations, and is used to characterize the comprehensive enhancement effect of the self-healing resource on the overall network topology edge weight strength.

[0080] In this embodiment, the total number of nodes N in the communication network refers to the set of all independent communication entities or logical nodes that constitute the physical communication network. For example, in a cellular network, N can represent the number of all base stations; in a data center network, N can represent the number of all servers or switches.

[0081] Resource-connection gain coefficient This is a scaling factor used to quantify the conversion of self-healing resources into network connectivity gains. The coefficient's dimension is the reciprocal of the absolute dimension of the resource, ensuring the correct conversion between resource quantity and dimensionless topology connectivity gains. Its value can be preset based on the specific characteristics of the network, resource type, and desired topology enhancement effect, or calibrated using historical data. For example, the average connectivity increase per unit of resource input can be determined through experiments or simulations, thereby setting this coefficient.

[0082] Self-healing resources , representing the specific amount of resources allocated to node k in the communication network at the current time t, according to the optimal self-healing resource allocation strategy. These resources can be bandwidth, computing power, backup links, power capacity, etc., and their dimensions correspond to the physical resource types they represent.

[0083] Local Laplacian matrix This is a matrix reflecting the local network topology centered at node k. The off-diagonal elements of this matrix are -1 between node k and each of its neighbors, representing the connectivity between node k and its neighbors; the diagonal elements are the degree of node k, i.e., the number of directly connected neighbors, representing the connectivity activity of node k; the remaining elements are 0. This construction method makes... Capturing the connectivity characteristics of node k and its direct neighbors is the fundamental unit for constructing global topology increments. Laplace matrix increment. The construction method involves using the local Laplacian matrix contributed by all nodes. Based on the amount of self-healing resources they each obtained Perform a weighted summation and multiply by the resource-connectivity gain coefficient. This weighted summation mechanism allows nodes allocated more self-healing resources to have a greater weight in the global Laplace matrix increment for their local topology enhancement effect, thus characterizing the comprehensive enhancement effect of self-healing resources on the overall network topology edge weight strength.

[0084] In a preferred embodiment of the present invention, in step S3, the optimal self-healing resource allocation strategy The solution employs a rolling time-domain optimization mechanism: at each sampling time t, the future N output is predicted using step S2. p Fault evolution probability sequence within a time step For the input, construct the following cumulative cost function: And solve for the cumulative cost function. Minimize control sequence And only the first element in the control sequence The optimal self-healing resource allocation strategy at the current moment Output: When the next sampling time t+Δt arrives, repeat the above process and re-predict and optimize with the latest system state;

[0085] Wherein, the prediction step size of the rolling time-domain optimization The value is a positive integer greater than or equal to 2 to ensure the multi-step forward-looking nature of fault evolution prediction and the long-term optimality of the self-healing strategy.

[0086] In this embodiment, the rolling time-domain optimization mechanism is an advanced control strategy. Its core idea is to solve an optimization problem at each sampling time based on the current system state and predictions for a finite future time window to obtain a control sequence for a future period, but only the first control action of this sequence is applied to the system. Subsequently, at the next sampling time, after the system state is updated, the above optimization process is repeated. This mechanism can effectively handle system dynamics, uncertainties, and constraints, achieving real-time, adaptive control of complex systems. For example, this mechanism can be implemented using model predictive control or predictive control.

[0087] Future N pThe fault evolution probability sequence within a given time step represents a set of predictions made by the system at the current moment regarding the likelihood of faults occurring at various discrete time points in the network. It serves as input to optimization algorithms, enabling the system to proactively consider the dynamic evolution trend of faults when formulating self-healing strategies, rather than focusing solely on instantaneous states. For example, this sequence can be generated by iteratively applying a graph Laplace dynamics model, or by combining historical data and machine learning models to modify and extend the predictions of the graph Laplace dynamics model.

[0088] Cumulative cost function This is a mathematical expression that quantifies the merits of a self-healing strategy. It comprehensively considers the cumulative cost of network failure risk and the cumulative cost of self-healing resource invocation within the prediction time window. The purpose of introducing a penalty term is to guide the system to reach or approach a desired stable state at the end of the prediction time window during the optimization process, avoiding unexpected behavior or instability at the end of the prediction period. For example, the penalty can be a weighted penalty on the failure probability at the end of the prediction time window, or a penalty for the system state deviating from the expected value. Solving for the cumulative cost function... Minimizing the control sequence refers to finding a series of control sequences within a future prediction time window using mathematical optimization methods that minimize the cumulative cost function. The self-healing resource allocation action aims to minimize the risk of failure and the cost of resources in order to achieve the best long-term self-healing effect. For example, since the cumulative cost function is usually quadratic and may have linear constraints, it can be solved using a quadratic programming algorithm or efficiently computed using a convex optimization toolbox. Only the first element in the control sequence is considered. The optimal self-healing resource allocation strategy at the current moment Output is one of the core operations of rolling time-domain control. Although the optimization process calculates the control sequence for the entire future time window, the system only executes the first control action in that sequence. This is because, as time progresses, new system state information and more accurate predictions become available. By re-optimizing at each sampling time, the system can promptly correct previous decisions, thereby improving its adaptability to uncertainties and disturbances. For example, this mechanism ensures that the system always makes decisions based on the latest information, avoiding over-reliance on long-term prediction uncertainties. At the next sampling time... Upon arrival, the above process is repeated, and prediction and optimization are performed again with the latest system state, describing the iterative characteristics of the rolling time-domain optimization mechanism. When a time step... Afterward, the system acquires the latest physical network state data and uses it as a new starting point to re-predict fault evolution and optimize self-healing strategies. This continuous "perception-prediction-decision-execution-re-perception" cycle enables the system to continuously adapt to changes in the network environment, achieving dynamic and real-time self-healing. For example, this can be achieved through a periodically triggered scheduler, which... A new optimization cycle begins when the interval ends.

[0089] In a preferred embodiment of the present invention, before the optimal self-healing resource dynamic reconfiguration strategy is executed in step S4, the following security assurance sub-steps are also performed: Step S41: Apply the optimal self-healing resource allocation strategy The resource allocation of each node is compared with the maximum acceptable resource capacity of the corresponding node. If the allocation of any node exceeds its capacity limit, the allocation is truncated to the capacity limit of that node, and the truncation difference is recorded. Step S42: The total amount of the truncated difference in resources is reinjected into the system's backup resource pool and marked as a reserved emergency reserve, which will not participate in this round of self-healing allocation; Step S43: After executing the truncated strategy, the strategy execution effect is verified a second time using a digital twin benchmark model. The strategy is allowed to be executed in the physical network only if the comprehensive evaluation index of the second verification still meets the security threshold. If the comprehensive evaluation index of the second verification does not meet the security threshold, the truncated difference is substituted into step S3 as an additional resource constraint, and the process returns to step S3. The strategy is solved and verified again by increasing the resource call cost parameter. The system alarm is triggered and the current strategy is not executed only if the preset maximum number of iterations is reached and the requirement is still not met.

[0090] In this embodiment, the solution of this application, in a digital twin-based communication network fault prediction and self-healing system, firstly obtains the actual telemetry state data of physical network nodes and the ideal baseline state data of the digital twin baseline model in step S1, and calculates a twin state deviation scalar that quantifies the degree of local state degradation of nodes by combining historical data. Subsequently, in step S2, using a diffusion dynamics mechanism based on spatial topology, these deviation scalars are used as external excitation sources to predict the fault evolution probability of each node in the entire network in future time steps. Based on this prediction result, step S3 constructs the network self-healing process as a static optimal quadratic resource allocation problem with constraints on resource conversion utility and call cost, and analytically solves the optimal self-healing resource dynamic reconfiguration strategy that minimizes the combined cost of instantaneous fault risk and reconfiguration cost. However, before directly applying this optimal strategy to the physical network, this application introduces additional security safeguards to further enhance the strategy's practicality and security. Specifically, in step S41, the system applies the optimal self-healing resource allocation strategy output in step S3. The resource allocation for each node is compared one by one with its maximum acceptable resource capacity. This aims to identify and prevent any node from becoming overloaded or unstable due to being allocated resources beyond its capacity. Once a node's allocation exceeds its capacity limit, the system immediately truncates the allocation to that node's capacity limit and accurately records the truncated difference. Subsequently, in step S42, all truncated resource differences are intelligently reinjected into the system's backup resource pool and explicitly marked as reserved emergency reserves. These resources will not be used in this round of self-healing allocation, thus avoiding resource waste and providing additional protection for potential future emergencies. Finally, in step S43, the system uses a digital twin benchmark model to perform a secondary verification of this strategy after node capacity truncation and resource reclamation. This verification is crucial, ensuring that even after considering physical constraints and adjusting the strategy, it still meets the preset network structure stability requirements. The system considers safety thresholds for comprehensive evaluation indicators such as connectivity and robustness. Only when the secondary verification confirms the strategy's safety and effectiveness will the system allow its actual execution in the physical communication network. If the comprehensive evaluation indicators of the secondary verification do not meet the safety thresholds, the truncation difference is substituted as an additional resource constraint into step S3, and the system returns to step S3, resolving and verifying the strategy by increasing the resource call cost parameter. Only when the preset maximum number of iterations is reached and the strategy still not satisfied will a system alarm be triggered, and the current strategy will remain inactive.

[0091] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A communication network fault prediction and self-healing system based on digital twins, characterized in that, The system operates by following these steps: Step S1: Obtain the actual telemetry status data of the node in the physical communication network within the current time window, as well as the ideal reference status data synchronously output by the digital twin reference model, and calculate the twin state deviation scalar that quantifies the degree of local state degradation of the node by combining the characteristic distribution of the node's historical normal operation status. Step S2: Model the physical communication network as a graph topology structure composed of nodes and edges. Using the diffusion dynamics mechanism based on spatial topology, the twin state deviation scalar obtained in step S1 is used as an external excitation source to predict the failure evolution probability of each node in the network in the future time step. Step S3: Construct the network self-healing process as a static optimal quadratic resource allocation problem with constraints on resource conversion utility and call cost. Based on the fault evolution probability output in step S2, analyze and solve the optimal self-healing resource dynamic reconstruction strategy that minimizes the combined cost of instantaneous fault risk and reconstruction cost. Step S4: Before executing the optimal self-healing resource dynamic reconstruction strategy in the physical communication network, inject it into the digital twin benchmark model. Utilize the algebraic connectivity characteristics of the network topology and the total resource consumption to perform closed-loop verification and comprehensive evaluation of the overall network structure stability and connectivity robustness after the strategy is executed. If the comprehensive evaluation index meets the preset security threshold, the strategy is executed in the physical communication network; otherwise, return to step S3 and solve the strategy again by increasing the resource call cost parameter.

2. The system according to claim 1, in step S1, the twin state deviation scalar is calculated using multidimensional phase space mapping, specifically as follows: First, subtract the actual telemetry state column vector of the physical network node at the current moment from the ideal state column vector synchronously output by the digital twin baseline model to obtain the deviation column vector. Then, multiply the transpose of this deviation column vector by the inverse matrix of the covariance matrix of the node's historical normal operating state characteristics, and then multiply by the aforementioned deviation column vector to obtain the instantaneous deviation quadratic value at that moment. Next, integrate and accumulate the instantaneous deviation quadratic values ​​of all moments within an integration time window length from the current moment. Finally, divide the accumulated integral value by the integration time window length to obtain the quantized scalar of the twin state deviation at the current moment.

3. The system according to claim 2, characterized in that, The failure evolution probability is calculated based on graph Laplace dynamics, specifically as follows: First, calculate the product of the spatial propagation rate coefficient of the fault in the network topology and the prediction time step, and then multiply this product by the product of the initial fault probability column vector of all nodes in the network at the current time to obtain the first term; then calculate the deviation excitation coupling coefficient multiplied by the prediction time step, and then multiply it by the global deviation excitation column vector formed by the aggregation of the twin state deviations of all nodes to obtain the second term; add the first term and the second term to obtain the fault evolution probability column vector of all nodes in the network at the future time.

4. The system according to claim 3, characterized in that, In step S3, the optimal self-healing resource dynamic reconfiguration strategy is solved using a static optimal quadratic control algorithm. The specific calculation method is as follows: First, calculate the transpose of the self-healing resource utility mapping matrix, multiply it by the positive definite diagonal matrix of the network node's failure risk penalty, and then multiply it by the self-healing resource utility mapping matrix to obtain the first matrix term; Next, add the first matrix term to the positive definite diagonal matrix of the self-healing resource mobilization cost, and invert the sum matrix to obtain the inverse matrix; then calculate the transpose of the self-healing resource utility mapping matrix, multiply it by the positive definite diagonal matrix of the fault risk penalty, and then multiply it by the future fault evolution probability column vector predicted in step S2 to obtain the column vector term; finally, multiply the above inverse matrix by the above column vector term to obtain the optimal self-healing resource allocation strategy column vector at the current moment.

5. The system according to claim 4, characterized in that, In step S4, the stability evaluation index of the network structure is calculated using graph theory. The specific calculation method is as follows: First, calculate the sum matrix obtained by adding the original Laplacian matrix to the increment of the Laplacian matrix generated by the optimal self-healing resource allocation strategy mapping. Then, calculate the second smallest eigenvalue of the sum matrix to obtain the corrected algebraic connectivity. Finally, calculate the largest eigenvalue of the sum matrix. Then, the modified algebraic connectivity is divided by the largest eigenvalue of the sum matrix to obtain the topology connectivity ratio. Next, the L1 norm of the column vector of the optimal self-healing resource allocation strategy is calculated to obtain the absolute amount of total system resources consumed in this self-healing process. This absolute amount of total resources is divided by the total limit of the maximum self-healing resource capacity pool of the communication network. The ratio is squared, and then the squared value is subtracted from one to obtain the resource consumption attenuation penalty term. Finally, the topology connectivity ratio is multiplied by the resource consumption attenuation penalty term to obtain the comprehensive evaluation index of network structure stability after applying the self-healing strategy.

6. The system according to claim 5, characterized in that, In step S4, if the comprehensive evaluation index does not meet the preset safety threshold, the diagonal elements of the positive definite diagonal matrix of the self-healing resource call cost in step S3 are incremented by a preset step size, and steps S3 and S4 are re-executed for iterative optimization; if the comprehensive evaluation index still does not meet the safety threshold after the number of iterations reaches the preset maximum number of iterations, a system alarm is triggered and the current self-healing strategy is maintained and not executed, waiting for manual intervention.

7. The system according to claim 5, characterized in that, The Laplace matrix increment is constructed as follows: First, for each node in the communication network, calculate the amount of self-healing resources allocated to that node multiplied by the resource-connection gain coefficient to obtain the resource gain weight of that node. Then multiply the resource gain weight by the local Laplacian matrix centered on the node to obtain the node's contribution matrix to the Laplacian matrix increment; then sum the contribution matrices of all nodes to obtain the Laplacian matrix increment generated by the optimal self-healing resource allocation strategy.

8. The system according to claim 4, characterized in that, In step S3, the optimal self-healing resource allocation strategy is solved using a rolling time-domain optimization mechanism: at each sampling time, a cumulative cost function is constructed using the fault evolution probability sequence over multiple future time steps predicted in step S2 as input. The cumulative cost function is calculated as follows: For each future prediction step, calculate the transpose of the failure probability column vector for that step, multiply it by the positive definite diagonal matrix of the failure risk penalty, and then multiply it by the failure probability column vector for that step to obtain the failure risk cost term. Then, calculate the transpose of the self-healing resource allocation strategy column vector corresponding to that step, multiply it by the positive definite diagonal matrix of the self-healing resource call cost, and then multiply it by the self-healing resource allocation strategy column vector to obtain the resource call cost term. Add the above two terms to obtain the single-step cost for that step. Then, sum the single-step costs of all prediction steps to obtain the cumulative cost function. Solve for the control sequence that minimizes the cumulative cost function, and output only the first element of the control sequence as the optimal self-healing resource allocation strategy at the current time. When the next sampling time arrives, repeat the above process and re-predict and optimize with the latest system state.

Images