Construction of communication resilience evaluation model and dynamic optimization training method and system

Abstract

The application relates to the technical field of communication system elasticity evaluation, and provides a construction and dynamic optimization training method and system of a communication elasticity evaluation model. The method determines a characteristic parameter matrix by acquiring historical operation data and carries out characteristic transformation processing, extracts long-term communication performance indexes and short-term fluctuation indexes to calculate first elasticity evaluation parameters, obtains second elasticity evaluation parameters based on the statistical feature difference update of real-time operation data and historical data, determines optimal elasticity configuration parameters through state space decomposition and system stability analysis, realizes communication system elasticity optimization configuration, and improves system response capability and resource utilization efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication system resilience evaluation, in particular to a communication resilience evaluation model construction and dynamic optimization training method and system. BACKGROUND

[0002] With the rapid development of information technology, communication systems play an increasingly important role in modern society. The resilience of a communication system refers to the ability of the system to maintain stable service quality in the face of various external disturbances and internal failures. The communication resilience evaluation model is an important tool for evaluating the resilience of a communication system. Through the analysis and processing of system operation data, the resilience level of the system can be quantified, providing a basis for system optimization. Traditional communication resilience evaluation methods are usually based on static parameters and fixed models, and evaluate system performance through a pre-set index system.

[0003] However, the existing communication resilience evaluation technology has the following defects and deficiencies: the existing technology lacks effective combination and analysis of historical data and real-time data of the communication system. Most evaluation models only focus on static data or running data of a single time period, and cannot fully reflect the resilience performance of the communication system under different workloads and network environments, resulting in deviations between the evaluation results and the actual system state. Traditional evaluation methods often ignore the relationship between short-term fluctuations and long-term performance of the communication system, and cannot accurately capture the dynamic response characteristics of the system in the face of sudden events, making the evaluation model difficult to adapt to complex and changing network environments, reducing the accuracy and practicality of resilience evaluation. The existing resilience evaluation model also lacks a self-adaptive optimization mechanism, and cannot dynamically adjust the evaluation parameters according to the changes in the system running state, resulting in unreasonable allocation of system resources in the case of large changes in network topology structure or traffic flow, and unable to achieve optimal configuration of communication system resources, affecting the overall performance and reliability of the system. SUMMARY

[0004] The embodiments of the present application provide a communication resilience evaluation model construction and dynamic optimization training method and system, which can at least solve some of the problems existing in the prior art.

[0005] In a first aspect of the embodiments of the present application, a communication resilience evaluation model construction and dynamic optimization training method is provided, comprising:

[0006] Obtaining historical operation data of a communication system, determining a feature parameter matrix based on the historical operation data, performing feature transformation processing on the feature parameter matrix to obtain a communication system feature parameter;

[0007] Extracting long-term communication performance indicators and short-term fluctuation indicators based on the communication system feature parameter, calculating first resilience evaluation parameters of each communication endpoint based on the long-term communication performance indicators and the short-term fluctuation indicators;

[0008] obtaining real-time running data of the communication system, calculating a statistical feature difference value between the real-time running data and the historical running data, updating the first elasticity evaluation parameter based on the statistical feature difference value to obtain a second elasticity evaluation parameter;

[0009] performing state space decomposition on the second elasticity evaluation parameter to obtain a multi-dimensional feature component, calculating a system stability index of the multi-dimensional feature component, determining an optimization objective function according to the system stability index, and determining an optimal elasticity configuration parameter based on the optimization objective function;

[0010] In an optional implementation, long-term communication efficiency indicators and short-term fluctuation indicators are extracted based on the communication system feature parameters, and a first elasticity evaluation parameter of each communication endpoint is calculated based on the long-term communication efficiency indicators and the short-term fluctuation indicators, including:

[0011] performing time-frequency joint analysis based on the communication system feature parameters to obtain a multi-scale feature sequence, extracting a long-term communication efficiency indicator from a long-term stable mode of the multi-scale feature sequence, and extracting a short-term fluctuation indicator from a short-term mutation mode;

[0012] segmenting the long-term communication efficiency indicator to extract a throughput sub-indicator and a time delay sub-indicator, and performing double exponential smoothing on the throughput sub-indicator and the time delay sub-indicator to obtain a system steady state factor;

[0013] performing dynamic layering on the short-term fluctuation indicator to extract a load sub-indicator and a jitter sub-indicator, and performing distribution probability conversion on the load sub-indicator and the jitter sub-indicator to obtain a fluctuation factor;

[0014] calculating a cross-layer interaction intensity based on the system steady state factor and the fluctuation factor to obtain a correlation intensity matrix, determining a probability weighted graph using the correlation intensity matrix, and calculating a first elasticity evaluation parameter of each communication endpoint according to the probability weighted graph.

[0015] obtaining real-time running data of the communication system, calculating a statistical feature difference value between the real-time running data and the historical running data, updating the first elasticity evaluation parameter based on the statistical feature difference value to obtain a second elasticity evaluation parameter, including:

[0016] obtaining real-time running data of the communication system, performing multi-dimensional decomposition on the real-time running data, extracting a time domain data sequence and a frequency domain data sequence, and combining and reconstructing the time domain data sequence and the frequency domain data sequence to obtain real-time communication system feature parameters;

[0017] extracting a real-time communication performance index and a real-time fluctuation index from the real-time communication system characteristic parameters, performing feature fusion on the real-time communication performance index and the real-time fluctuation index to obtain a real-time state vector;

[0018] obtaining a historical state vector corresponding to historical operation data, calculating a probability distribution difference and a state transition probability of the real-time state vector and the historical state vector in multiple time scales, and obtaining a statistical feature difference value based on the probability distribution difference and the state transition probability;

[0019] performing time series decomposition and reconstruction on the statistical feature difference value, determining a state contribution degree of each dimension difference value in the statistical feature difference value, and correcting the first elasticity evaluation parameter by taking the state contribution degree as an update coefficient to obtain a second elasticity evaluation parameter.

[0020] performing time series decomposition and reconstruction on the statistical feature difference value, determining a state contribution degree of each dimension difference value in the statistical feature difference value, and correcting the first elasticity evaluation parameter by taking the state contribution degree as an update coefficient to obtain a second elasticity evaluation parameter, including:

[0021] converting the statistical feature difference value into a difference value sequence, decomposing the difference value sequence to obtain multi-scale coefficients, determining an optimal decomposition layer number according to the energy distribution of the multi-scale coefficients, and reconstructing the difference value sequence to obtain a time series component;

[0022] grouping the time series component according to fluctuation frequency, calculating the autocorrelation coefficient of each group of time series components, screening stable time series components according to the autocorrelation coefficient, and reconstructing the stable time series components to obtain a difference value reconstruction sequence;

[0023] constructing a three-dimensional state space for the difference value reconstruction sequence, calculating a phase space trajectory in the three-dimensional state space, and extracting a topological feature of the phase space trajectory to obtain a state evolution graph;

[0024] calculating a state transition probability and a residence time according to the state evolution graph, combining the state transition probability and the residence time to construct a state duration matrix, calculating the state contribution degree of each dimension difference value based on the state duration matrix, and correcting the first elasticity evaluation parameter by taking the state contribution degree as an update coefficient to obtain a second elasticity evaluation parameter.

[0025] performing state space decomposition on the second elasticity evaluation parameter to obtain multi-dimensional feature components, calculating a system stability index of the multi-dimensional feature components, determining an optimization objective function according to the system stability index, and determining an optimal elasticity configuration parameter based on the optimization objective function, including:

[0026] mapping the second elasticity evaluation parameter to a state space, constructing an orthogonal basis vector in the state space, performing dimension reduction decomposition on the second elasticity evaluation parameter based on the orthogonal basis vector to obtain a time dimension feature component and a space dimension feature component;

[0027] calculating a fluctuation period feature and a trend feature for the time dimension feature component, and calculating a distribution density feature and an aggregation feature for the space dimension feature component;

[0028] combining the fluctuation period feature and the trend feature to obtain a time stability index, and combining the distribution density feature and the aggregation feature to obtain a space stability index;

[0029] determining a target function according to the time stability index and the space stability index, obtaining a time dimension constraint condition based on a period component in the time dimension feature component, and obtaining a space dimension constraint condition based on a density component in the space dimension feature component;

[0030] determining the optimization target function with constraint conditions based on the time dimension constraint condition, the space dimension constraint condition and the target function, generating an initial solution space, iteratively calculating an optimal solution of the optimization target function in the initial solution space, and obtaining an optimal elasticity configuration parameter.

[0031] combining the fluctuation period feature and the trend feature to obtain a time stability index, and combining the distribution density feature and the aggregation feature to obtain a space stability index, comprises:

[0032] determining a period feature sequence and a trend feature sequence based on statistical moments of the fluctuation period feature and the trend feature, and calculating a time correlation coefficient of the period feature sequence and the trend feature sequence to obtain a period correlation matrix and a trend correlation matrix;

[0033] determining a density feature sequence and an aggregation feature sequence based on a local density center of the distribution density feature and an inter-cluster similarity of the aggregation feature, and calculating a space correlation coefficient of the density feature sequence and the aggregation feature sequence to obtain a density correlation matrix and an aggregation correlation matrix;

[0034] determining a first coupling strength of the period correlation matrix and the trend correlation matrix, and determining an initial time stability index according to the first coupling strength;

[0035] calculating fluctuation characteristics of the initial time stability index under multiple time scales, and correcting the initial time stability index to obtain a time stability index according to the fluctuation characteristics;

[0036] determine a second coupling strength of the density-related matrix and the aggregation-related matrix, and determine an initial spatial stability index according to the second coupling strength;

[0037] calculate distribution characteristics of the initial spatial stability index at multiple spatial scales, and correct the initial spatial stability index to obtain a spatial stability index according to the distribution characteristics.

[0038] generate a system resource scheduling strategy according to the optimal elastic configuration parameter, and perform elastic optimization configuration on the communication system, including:

[0039] calculate a system resource capacity threshold based on the optimal elastic configuration parameter, divide the system resource capacity threshold into multiple load intervals, and set a resource scheduling priority in each load interval;

[0040] extract resource utilization and service request volume in the real-time running data, and calculate variation trend characteristics of the resource utilization and the service request volume;

[0041] determine a current load interval according to the variation trend characteristics of the resource utilization, predict a next time load interval based on the variation trend characteristics of the service request volume, and calculate a transition probability of adjacent load intervals;

[0042] calculate a resource dynamic expansion and contraction capacity according to the transition probability, map the resource dynamic expansion and contraction capacity and the resource scheduling priority to a resource allocation sequence, and determine a resource recovery time sequence and a resource allocation time sequence based on the resource allocation sequence;

[0043] combine the resource recovery time sequence and the resource allocation time sequence into a resource scheduling time sequence, calculate a resource scheduling step in each scheduling period of the resource scheduling time sequence, and perform hierarchical scheduling on system resources according to the resource scheduling step, so as to perform elastic optimization configuration on the communication system.

[0044] In a second aspect of the embodiment of the application, a communication elasticity evaluation model construction and dynamic optimization training system is provided, including:

[0045] A first unit is configured to acquire historical running data of a communication system, determine a feature parameter matrix according to the historical running data, and perform feature transformation processing on the feature parameter matrix to obtain a communication system feature parameter.

[0046] A second unit is configured to extract a long-term communication efficiency index and a short-term fluctuation index based on the communication system feature parameter, and calculate a first elasticity evaluation parameter of each communication endpoint based on the long-term communication efficiency index and the short-term fluctuation index.

[0047] a third unit configured to obtain real-time operation data of the communication system, calculate a statistical feature difference value between the real-time operation data and the historical operation data, update the first elasticity evaluation parameter based on the statistical feature difference value to obtain a second elasticity evaluation parameter;

[0048] a fourth unit configured to perform state space decomposition on the second elasticity evaluation parameter to obtain a multi-dimensional feature component, calculate a system stability index of the multi-dimensional feature component, determine an optimization objective function according to the system stability index, and determine an optimal elasticity configuration parameter based on the optimization objective function;

[0049] a fifth unit configured to generate a system resource scheduling strategy according to the optimal elasticity configuration parameter, and perform elasticity optimization configuration on the communication system.

[0050] In a third aspect, an electronic device is provided, including:

[0051] a processor;

[0052] a memory for storing processor-executable instructions;

[0053] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0054] In a fourth aspect, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0055] The present application can accurately evaluate and optimize the communication system by constructing a communication elasticity evaluation model and using a dynamic optimization training method, effectively improving the operation efficiency and stability of the communication system. Through comprehensive analysis of historical operation data and real-time operation data, the present application can accurately capture the dynamic change characteristics of the communication system, realize accurate quantitative evaluation of communication elasticity, and solve the problem that traditional methods are difficult to adapt to complex communication environments. Based on state space decomposition and system stability analysis, the present application can automatically generate an optimal resource scheduling strategy, realize dynamic optimization configuration of the communication system resources, improve the anti-interference ability and service quality of the system, and reduce the operation and maintenance cost and resource waste. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 FIG. 1 is a flowchart of the construction of a communication elasticity evaluation model and a dynamic optimization training method according to an embodiment of the present application.

[0057] Figure 2 FIG. 2 is a flowchart of determining a second elasticity evaluation parameter according to an embodiment of the present application. DETAILED DESCRIPTION

[0058] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0059] The technical solutions of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and some embodiments can not be described again for the same or similar concepts or processes.

[0060] Figure 1 The flowchart of the method for constructing and dynamically optimizing the communication resilience evaluation model of the embodiments of the present application is shown in FIG. 1, and the method comprises the following steps. Figure 1

[0061] Obtaining historical running data of a communication system, determining a feature parameter matrix according to the historical running data, performing feature transformation processing on the feature parameter matrix to obtain a communication system feature parameter;

[0062] Extracting a long-term communication efficiency index and a short-term fluctuation index based on the communication system feature parameter, calculating a first resilience evaluation parameter of each communication endpoint based on the long-term communication efficiency index and the short-term fluctuation index;

[0063] Obtaining real-time running data of the communication system, calculating a statistical feature difference value of the real-time running data and the historical running data, updating the first resilience evaluation parameter based on the statistical feature difference value to obtain a second resilience evaluation parameter;

[0064] Performing state space decomposition on the second resilience evaluation parameter to obtain a multi-dimensional feature component, calculating a system stability index of the multi-dimensional feature component, determining an optimization objective function according to the system stability index, and determining an optimal resilience configuration parameter based on the optimization objective function;

[0065] Generating a system resource scheduling strategy according to the optimal resilience configuration parameter, and performing resilience optimization configuration on the communication system.

[0066] The present application calculates resilience evaluation parameters by analyzing historical data and real-time data, generates a resource scheduling strategy based on the resilience evaluation parameters, and realizes resilience optimization configuration of a communication system.

[0067] ​In this embodiment, first, the historical operation data of the communication system is obtained, which includes but is not limited to key indicators such as communication node throughput, delay time, packet loss rate, signal strength, etc. For example, the historical operation data of a certain regional 5G base station cluster contains the throughput data of each base station collected every hour in the past 30 days (average value is 850 Mbps, maximum value is 1.2 Gbps, minimum value is 350 Mbps), end-to-end delay (average value is 15 ms, peak value is 45 ms), and resource utilization (CPU average utilization is 65%, memory average utilization is 72%) and other information.

[0068] Based on the obtained historical operation data, a feature parameter matrix is determined, which contains multi-dimensional feature parameters of multiple communication endpoints. When constructing the feature parameter matrix, for each communication endpoint, its throughput, delay, packet loss rate, resource occupancy rate and other multi-dimensional feature parameters are extracted. For example, for a network containing 100 communication endpoints, 10 feature parameters are extracted for each endpoint, forming a 100x10 feature parameter matrix.

[0069] The feature parameter matrix is subjected to feature transformation processing, which can include normalization processing, principal component analysis, independent component analysis, etc. For example, the Z-score standardization method is used to normalize the feature parameters, so that the mean value of each feature parameter is 0 and the standard deviation is 1. After normalization, the principal component analysis method is applied to reduce the original 10-dimensional features to 4-dimensional principal components, which can explain 85% of the variance of the original data. These transformed feature parameters constitute the feature parameters of the communication system.

[0070] Based on the feature parameters of the communication system, long-term communication performance indicators and short-term fluctuation indicators are extracted. The long-term communication performance indicators reflect the performance of the system in a longer time period, such as the average throughput stability in 30 days (fluctuation range not more than 10%), system availability (99.95%), etc. The short-term fluctuation indicators reflect the performance fluctuations of the system in a short time, such as the throughput decrease amplitude (maximum decrease 30%) in peak hours (18:00-22:00), the delay increase rate (maximum increase 200%) under burst traffic, etc. Through time series analysis methods such as exponential weighted moving average method, the time series corresponding to the feature parameters of the communication system are decomposed into trend items and fluctuation items, where the trend items are used to calculate the long-term performance indicators, and the fluctuation items are used to calculate the short-term fluctuation indicators.

[0071] The first elasticity evaluation parameter is a comprehensive index measuring the ability of the communication endpoint to cope with load changes. The method for calculating the first elasticity evaluation parameter of each communication endpoint based on the long-term communication efficiency index and the short-term fluctuation index is to set the weight of the long-term communication efficiency index to 0.7 and the weight of the short-term fluctuation index to 0.3, and then obtain the first elasticity evaluation parameter by weighted summation. For example, if the long-term efficiency index of a certain communication endpoint is 0.85 and the short-term fluctuation index is 0.65, then the first elasticity evaluation parameter is 0.85*0.7+0.65*0.3=0.79. This calculation method places more emphasis on long-term stability while also considering short-term resilience.

[0072] The real-time running data of the communication system can be collected in real time by the system monitoring module, including current network traffic (e.g., traffic at a certain time is 750 Mbps), service response time (e.g., average response time is 25 ms), resource utilization (e.g., CPU utilization is 75%), and other indicators.

[0073] The statistical feature difference value is calculated based on the real-time running data and the historical running data, which reflects the deviation degree of the current system state from the historical normal state. The calculation method is to compare each indicator of the real-time data with the corresponding indicator of the historical data to obtain the difference ratio. For example, if the current throughput is 15% lower than the historical average, the current delay is 30% higher than the historical average, and the resource utilization is 10% higher than the historical average, the comprehensive statistical feature difference value is 0.18.

[0074] The method for updating the first elasticity evaluation parameter based on the statistical feature difference value is to dynamically adjust the first elasticity evaluation parameter according to the size of the statistical feature difference value. When the statistical feature difference value is small (e.g., less than 0.1), the first elasticity evaluation parameter remains unchanged; when the difference value is at a medium level (between 0.1 and 0.3), the first elasticity evaluation parameter is reduced by a certain proportion; and when the difference value is large (greater than 0.3), the first elasticity evaluation parameter is significantly reduced. For example, if the first elasticity evaluation parameter of a certain endpoint is 0.79 and the statistical feature difference value is 0.18, and the adjustment factor is 0.9, then the second elasticity evaluation parameter is 0.79*0.9=0.711.

[0075] The second elasticity evaluation parameter is decomposed into multiple feature components in the state space, which is a method for decomposing the second elasticity evaluation parameter into multiple components reflecting different system states. Specifically, recursive filtering technology is used to decompose the second elasticity evaluation parameter into trend components, periodic components, and random components. For example, the second elasticity evaluation parameter of a certain endpoint is 0.711, which is decomposed into a trend component of 0.68, a periodic component of 0.08, and a random component of -0.05.

[0076] The system stability index measures the ability of the system to maintain stable operation under various load conditions. The method of calculating the system stability index of the multi-dimensional characteristic component is to assign a weight of 0.5 to the trend component, a weight of 0.3 to the periodic component, and a weight of 0.2 to the random component, and to obtain the system stability index by weighted combination. For example, the system stability index of the above decomposition result is 0.68 x 0.5 + 0.08 x 0.3 + |-0.05| x 0.2 = 0.37.

[0077] The optimization objective function is determined according to the system stability index, which is a multi-objective function that comprehensively considers system stability, resource utilization efficiency and service quality. When the system stability index is higher than the threshold value 0.8, the optimization objective is biased towards resource utilization efficiency; when the system stability index is lower than the threshold value 0.5, the optimization objective is biased towards system stability; when the system stability index is between 0.5 and 0.8, the optimization objective balances stability and efficiency. In this example, the system stability index is 0.37, which is lower than the threshold value of 0.5, so the optimization objective function will focus more on improving system stability, with a stability weight of 0.7 and an efficiency weight of 0.3.

[0078] The optimal elasticity configuration parameters include resource allocation ratio, load balancing parameters, buffer size, etc. The method of determining the optimal elasticity configuration parameters based on the optimization objective function is to search the parameter space by simulated annealing algorithm under the guidance of the optimization objective function, and to find the parameter combination that optimizes the objective function value. For example, in the case of emphasizing stability, the algorithm obtains the following configuration parameters: the master-slave resource ratio is 7:3, the load balancing threshold is 65%, and the buffer size is 20% of the original traffic.

[0079] The system resource scheduling strategy is generated according to the optimal elasticity configuration parameters, including computing resource allocation strategy, bandwidth allocation strategy, task scheduling strategy, etc. For example, based on the above configuration parameters, the generated resource scheduling strategy includes: reserving 30% of the additional computing resources for critical communication nodes; triggering the load balancing mechanism when the load exceeds 65%, and directing the new traffic to the nodes with lower load; dynamically adjusting the network buffer size to maintain around 20% of the current traffic.

[0080] The elasticity optimization configuration is applied to the actual system. In specific implementation, first, the scheduling strategy is tested in a small range to verify its effectiveness; after confirming its effectiveness, it is applied to the entire system in batches, and the system performance is monitored after each batch to ensure system stability; finally, the system running status is continuously monitored, and the configuration parameters are dynamically adjusted according to real-time feedback. For example, after implementing the elasticity optimization configuration, the throughput of a regional communication network during peak hours increased by 15%, the end-to-end delay decreased by 25%, the system stability increased by 30%, and the resource utilization rate increased by 20%.

[0081] By the above, the elastic optimization configuration of the communication system can be realized, the stability, resource utilization efficiency and service quality of the system are improved, and the communication demand in different scenarios is met.

[0082] In an optional implementation, long-term communication performance indicators and short-term fluctuation indicators are extracted based on the communication system characteristic parameters, and a first elasticity evaluation parameter of each communication endpoint is calculated based on the long-term communication performance indicators and the short-term fluctuation indicators, including:

[0083] A multi-scale feature sequence is obtained by time-frequency joint analysis based on the communication system characteristic parameters, a long-term communication performance indicator is obtained by extracting a long-term stable mode from the multi-scale feature sequence, and a short-term fluctuation indicator is obtained by extracting a short-term mutation mode;

[0084] The long-term communication performance indicators are segmented to extract throughput sub-indicators and time delay sub-indicators, and double exponential smoothing is performed on the throughput sub-indicators and the time delay sub-indicators to obtain a system steady state factor;

[0085] The short-term fluctuation indicators are dynamically layered to extract load sub-indicators and jitter sub-indicators, and distribution probability conversion is performed on the load sub-indicators and the jitter sub-indicators to obtain a fluctuation factor;

[0086] The correlation strength matrix is obtained based on the system steady state factor and the fluctuation factor, the probability weighted graph is determined using the correlation strength matrix, and the first elasticity evaluation parameter of each communication endpoint is calculated according to the probability weighted graph.

[0087] In actual application, first, the characteristic parameters of the communication system are collected, which include but are not limited to network delay, packet loss rate, throughput, signal strength and other communication system performance indicators. Taking the core network of a certain telecom operator as an example, the system records network transmission rate, time delay, jitter and packet loss rate data every minute.

[0088] The collected characteristic parameters are analyzed by time-frequency joint analysis to generate a multi-scale feature sequence. The specific implementation is as follows: 48 hours of sample data are collected according to the standard of 1 minute / time of sampling frequency, forming a time sequence with a length of 2880, then the wavelet transform method is applied to decompose the signal into different frequency components to obtain a multi-scale feature sequence. In these multi-scale feature sequences, the long-term stable mode is identified as a long-term communication performance indicator, and the short-term mutation mode is extracted as a short-term fluctuation indicator. For example, in a 48-hour monitoring period, the baseline value of throughput is identified as 85 Mbps, which is part of the long-term communication performance indicator; at the same time, the phenomenon of throughput dropping to 30 Mbps in a certain time period is found, which is marked as a short-term fluctuation indicator.

[0089] The long-term communication performance indicators are segmented, and throughput sub-indicators and latency sub-indicators are extracted. The segmentation is to divide the 48-hour data into a time window of 4 hours each, and the average throughput and average latency are calculated as sub-indicators in each window. For example, in a certain 4-hour window, the average throughput is calculated to be 90 Mbps, and the average latency is 15 ms. These sub-indicators are subjected to double exponential smoothing processing. The calculation method is to first perform one exponential smoothing on the original data to obtain a first-order smoothing value, and then perform a second exponential smoothing on the first-order smoothing value to obtain a second-order smoothing value. The combination of the two obtains the system steady-state factor. For the above example, after double exponential smoothing, the system steady-state factor is determined to be 0.85, indicating that the system has high stability in long-term operation.

[0090] At the same time, the short-term fluctuation indicators are dynamically layered, and load sub-indicators and jitter sub-indicators are extracted. Dynamic layering refers to dividing short-term fluctuation indicators into different levels according to the amplitude and frequency of data fluctuations. For example, load fluctuations are divided into high, medium, and low three layers, and jitter is divided into severe, moderate, and slight three layers. In a certain short time window, it is detected that the load increases from the standard level of 60% to 95%, which is classified as high-level load fluctuation; at the same time, it is detected that the jitter increases from the average value of 3 ms to 12 ms, which is classified as moderate jitter. The distribution probability of these layered sub-indicators is converted, and the probability of each state appearing is calculated to form a fluctuation factor. For example, the probability of high load state appearing is 0.15, and the probability of moderate jitter state appearing is 0.25. After comprehensive consideration, the system calculates the fluctuation factor to be 0.30, indicating that the system has a certain degree of instability in the short term.

[0091] Based on the system steady-state factor and the fluctuation factor, the cross-layer interaction intensity is calculated, and the correlation strength matrix is generated. The calculation method is to weight and combine the system steady-state factor and the fluctuation factor, with weights of 0.7 and 0.3 respectively, and then calculate the correlation degree between different communication endpoints. For a network containing 5 communication endpoints, a 5x5 correlation strength matrix is generated, and each element in the matrix represents the correlation strength between two endpoints. For example, the correlation strength between endpoint A and endpoint B is 0.82, indicating that there is a strong interaction relationship between the two endpoints.

[0092] The correlation strength matrix is used to determine the probability weighted graph, where the nodes of the graph represent communication endpoints, and the weights of the edges represent correlation strengths. In this probability weighted graph, the centrality indicators of each node are calculated, including degree centrality, betweenness centrality, and eigenvector centrality. For example, for endpoint C, the degree centrality is 3.1, the betweenness centrality is 0.45, and the eigenvector centrality is 0.72. These centrality indicators are comprehensively considered to calculate the first elasticity evaluation parameter of endpoint C as 0.68.

[0093] By this method, the first elasticity evaluation parameters of all communication endpoints are calculated, which can help network administrators identify key nodes and weak links in the communication system, thereby optimizing the network structure and improving the overall elasticity and stability of the system. For example, if the elasticity evaluation parameter of a certain endpoint is lower than the threshold value 0.5, an early warning will be automatically sent to the administrator, suggesting to strengthen the resource configuration of the endpoint or optimize its communication strategy. Through the above, the performance and stability of the communication system under different working conditions can be effectively reflected, providing strong data support for the optimization and upgrading of the communication system.

[0094] In an optional embodiment, real-time running data of the communication system is obtained, a statistical feature difference value of the real-time running data and the historical running data is calculated, the first elasticity evaluation parameter is updated based on the statistical feature difference value to obtain a second elasticity evaluation parameter, including:

[0095] The real-time running data of the communication system is obtained, the real-time running data is decomposed in multiple dimensions, time domain data sequences and frequency domain data sequences are extracted, and the time domain data sequences and the frequency domain data sequences are combined and reconstructed to obtain real-time communication system feature parameters;

[0096] Real-time communication efficiency indicators and real-time fluctuation indicators are extracted from the real-time communication system feature parameters, the real-time communication efficiency indicators and the real-time fluctuation indicators are feature fused to obtain a real-time state vector;

[0097] A historical state vector corresponding to the historical running data is obtained, a probability distribution difference and a state transition probability of the real-time state vector and the historical state vector under multiple time scales are calculated, and a statistical feature difference value is obtained based on the probability distribution difference and the state transition probability;

[0098] The statistical feature difference value is time series decomposed and reconstructed, the state contribution degree of each dimension difference value in the statistical feature difference value is determined, the state contribution degree is taken as an update coefficient to correct the first elasticity evaluation parameter, and a second elasticity evaluation parameter is obtained.

[0099] During the operation of the communication system, real-time running data of the system is collected, including but not limited to network traffic, time delay, packet loss rate, signal strength, service quality and other indicators. The obtained real-time running data is processed by multi-dimensional decomposition, specifically, the time series data is decomposed into different frequency components by wavelet transform, and the low frequency part is extracted as the time domain data sequence and the high frequency part is extracted as the frequency domain data sequence. For example, network traffic data is collected at a sampling interval of 10 ms for 1 minute to obtain 6000 data points, and through 3-layer wavelet decomposition, low-frequency time domain features and corresponding high-frequency frequency domain features of 3 scales are obtained.

[0100] The extracted time-domain and frequency-domain data sequences are combined using a feature reconstruction algorithm to generate feature parameters for the real-time communication system. The feature reconstruction employs an adaptive weighting method, assigning weights based on the sensitivity of different frequency components to the system state; for example, the time-domain sequence has a weight of 0.6, and the frequency-domain sequence has a weight of 0.4. The reconstructed feature parameters are represented in vector form, containing comprehensive information about the system's operation.

[0101] From the characteristic parameters of the real-time communication system, real-time communication performance indicators and real-time fluctuation indicators are further extracted. Communication performance indicators include throughput, connection success rate, and service response time, characterizing the system's service capabilities. Fluctuation indicators include performance jitter, load volatility, and resource utilization changes, characterizing system stability. For example, from the reconstruction features, the average throughput is extracted to be 95 Mbps, with a fluctuation range of ±5 Mbps; the connection success rate is 99.2%, with a fluctuation range of ±0.3%.

[0102] The extracted real-time communication performance indicators and real-time fluctuation indicators are integrated into a real-time state vector through feature fusion technology. The feature fusion adopts a deep autoencoder structure to compress high-dimensional features into low-dimensional state representations. In specific implementation, the number of input layer nodes is set to the original feature dimension (e.g., 20 indicators), the number of hidden layer nodes decreases layer by layer (e.g., 20-12-8), and the output layer is an 8-dimensional state vector, with each dimension representing a core state characteristic of the system.

[0103] Simultaneously, historical state vectors corresponding to historical operational data are retrieved from the historical database. These historical state vectors are generated using the same processing flow as real-time data and represent the system's operational status over a past time period. For example, historical data from the past 30 days is extracted, generating one state vector for each day, resulting in a total of 30 historical state vectors.

[0104] The probability distribution differences between the real-time state vector and the historical state vector at multiple time scales are calculated. A kernel density estimation method is used to construct the probability distribution functions for the real-time and historical states, and the difference between the two distributions is quantified using the JS divergence measure. In practical applications, the distribution differences at the hourly, daily, and weekly levels can be calculated separately. For example, an hourly JS divergence of 0.15, a daily divergence of 0.23, and a weekly divergence of 0.31 indicates that the system state is gradually deviating from the historical pattern.

[0105] Simultaneously, the state transition probabilities are calculated, and a Markov model is used to describe the system's state transition characteristics. A state transition matrix is ​​constructed to record the probability of transitioning from one state to another. For example, the system has four typical states: normal, slightly abnormal, moderately abnormal, and severely abnormal. The current probability of the system transitioning from the normal state to the slightly abnormal state is 0.18, while in historical data this probability is 0.05, indicating a decrease in system stability.

[0106] Based on the probability distribution difference and the state transition probability, a statistical feature difference value is calculated, the difference value is calculated by using a weighted fusion method, the distribution difference weight is set to 0.6, the state transition probability weight is set to 0.4, a comprehensive difference value vector is obtained, and each dimension corresponds to the difference degree of a system feature. For example, an 8-dimensional difference value vector [0.21, 0.15, 0.32, 0.08, 0.19, 0.25, 0.12, 0.28] is calculated.

[0107] The calculated statistical feature difference value is subjected to time series decomposition and reconstruction, the difference value is decomposed into different frequency components by using an empirical mode decomposition technology, long-term trends and short-term fluctuations are identified, the contribution degree of each dimension difference value to the system state change is determined through reconstruction analysis. For example, the third dimension difference value 0.32 is decomposed, it is found that the third dimension difference value 0.32 contains 60% of a long-term trend component and 40% of a periodic fluctuation, and it is indicated that the third dimension has a greater impact on the system state.

[0108] The state contribution degree of each dimension is used as an update coefficient to correct the first elasticity evaluation parameter. It is assumed that the first elasticity evaluation parameter is [0.85, 0.92, 0.78, 0.90, 0.83, 0.87, 0.94, 0.81], the contribution degree of each dimension is [0.15, 0.12, 0.28, 0.05, 0.14, 0.18, 0.08, 0.20], and the correction formula is: the second parameter = the first parameter × (1 - the contribution degree × the adjustment factor). It is assumed that the adjustment factor is 0.8, and the second elasticity evaluation parameter is [0.83, 0.91, 0.73, 0.90, 0.81, 0.85, 0.94, 0.78].

[0109] Through the above embodiments, the dynamic update of the communication system elasticity evaluation parameter is realized, and the evaluation result can timely reflect the change of the real-time state of the system. In actual application, the method can effectively identify system abnormal fluctuations, early warn potential risks, and provide a decision basis for communication system elasticity management.

[0110] In an optional embodiment, the statistical feature difference value is subjected to time series decomposition and reconstruction, the state contribution degree of each dimension difference value in the statistical feature difference value is determined, the state contribution degree is used as an update coefficient to correct the first elasticity evaluation parameter, and a second elasticity evaluation parameter is obtained, including:

[0111] The statistical feature difference value is converted into a difference value sequence, the difference value sequence is decomposed to obtain multi-scale coefficients, the optimal decomposition layer number is determined according to the energy distribution of the multi-scale coefficients, and the difference value sequence is reconstructed to obtain a time series component;

[0112] The time series components are grouped according to their fluctuation frequency, the autocorrelation coefficient of each group of time series components is calculated, stable time series components are selected based on the autocorrelation coefficient, and the stable time series components are reconstructed to obtain the difference value reconstruction sequence.

[0113] A three-dimensional state space is constructed from the reconstructed sequence of difference values. The phase space trajectory is calculated in the three-dimensional state space, and the topological features of the phase space trajectory are extracted to obtain the state evolution diagram.

[0114] The state transition probability and dwell time are calculated based on the state evolution diagram. The state transition probability and dwell time are combined to construct a state persistence matrix. The state contribution of the difference values ​​of each dimension is calculated based on the state persistence matrix. The state contribution is used as an update coefficient to correct the first elasticity evaluation parameter to obtain the second elasticity evaluation parameter.

[0115] Figure 2 This is a schematic diagram illustrating the process of determining the second elasticity evaluation parameter according to an embodiment of the present invention. Figure 2 As shown, transforming statistical characteristic differences into a difference value sequence is the starting point for achieving time-series decomposition and reconstruction. Specifically, statistical characteristic data at different times or under different conditions are collected, the difference values ​​between the reference baseline and the current state are calculated, and the results are arranged in chronological order to form a difference value sequence. For example, for assessing the operational stability of a power system, the difference values ​​of parameters such as system frequency, voltage, and power over a continuous 24-hour period can be collected to form a difference value sequence containing 1440 sampling points.

[0116] Wavelet transform can be used to decompose the difference value sequence. By selecting a suitable wavelet basis function, such as the db4 wavelet, the sequence can be decomposed into multiple levels to obtain approximate coefficients and detail coefficients; these coefficients are collectively referred to as multi-scale coefficients. The energy distribution of the decomposition coefficients at each level is calculated, and the optimal number of decomposition levels is determined by comparing the energy proportions at different levels. Practice shows that when the rate of change in energy proportion is less than 5%, that level can be considered the optimal number of decomposition levels. Taking a power system as an example, analyzing a difference value sequence of 1440 points, the rate of change in energy proportion drops to 3.8% at the 4th level, which is below the 5% threshold; therefore, the optimal number of decomposition levels is determined to be 4.

[0117] Based on the determined optimal decomposition level, wavelet reconstruction technology is used to reconstruct the difference value sequence, obtaining time-series components including trend components, periodic components, and noise components. In the power system case, after reconstruction through 4-level wavelet decomposition, 5 time-series components are obtained, including 1 approximate component and 4 detail components.

[0118] The obtained time series components are grouped according to the fluctuation frequency characteristics. The main frequency of the time series component is calculated, and the components with similar main frequencies are grouped together. In the power system case, the five time series components are divided into three groups: low-frequency group (0-0.01 Hz), medium-frequency group (0.01-0.1 Hz), and high-frequency group (>0.1 Hz).

[0119] The autocorrelation coefficients of each group of time series components are calculated to evaluate the stability of the time series components. For each group of time series components, the autocorrelation coefficients at different lag orders are calculated. When the absolute value of the autocorrelation coefficient is greater than 0.6 and the decay is slow, the time series component is considered to be a stable component. In the example, the absolute value of the autocorrelation coefficient of the time series component in the low-frequency group and part of the medium-frequency group remains above 0.7 at the 10th lag, and is determined to be a stable component; while the autocorrelation coefficient of the high-frequency group quickly decreases to below 0.3 after 3 lags, and is determined to be a non-stable component.

[0120] The selected stable time series components are superimposed and reconstructed to obtain the difference value reconstruction sequence. In the power system example, 1 component in the low-frequency group and 2 components in the medium-frequency group are selected as stable components, and the new difference value sequence is obtained after reconstruction. This sequence retains the main characteristics of the original difference value sequence while filtering out most of the noise interference.

[0121] A three-dimensional state space is constructed for the difference value reconstruction sequence, and a delay coordinate embedding method is used to select appropriate embedding dimension and time delay parameters to map the one-dimensional time series to a three-dimensional state space. In the power system example, the embedding dimension is selected to be 3 and the time delay is selected to be 4, and the reconstruction sequence is mapped to a set of three-dimensional state vectors.

[0122] The phase space trajectory of the system is calculated in the three-dimensional state space, and the state vectors are connected to form the trajectory, which describes the evolution process of the system state. The topological features of the phase space trajectory are extracted, including the distribution area, density and shape of the trajectory, and a state evolution graph is constructed. In the power system case, the phase space trajectory presents the characteristics of rotating around certain center points and gathering in certain areas.

[0123] According to the state evolution graph, the state transition probability and residence time are calculated, and the three-dimensional state space is divided into multiple regions. The frequency of system state transition between regions is counted, and the normalized transition probability matrix is calculated. At the same time, the length of continuous residence time of system state in each region is counted to obtain the residence time distribution. In the power system example, the state space is divided into 8 regions, and the transition probability matrix shows that the system state has a 70% probability of remaining in the current state region and a 30% probability of transitioning to the adjacent region.

[0124] The state transition probability is combined with the residence time to construct a state duration matrix. The element values of the matrix comprehensively consider the transition probability and the average residence time, reflecting the stability and persistence of the system state. Based on the state duration matrix, the contribution of each dimension difference value to the system state is calculated. The contribution calculation considers the distribution proportion of the difference value in each state region and the persistence and stability of the corresponding state region. In the example, the state contribution of the frequency difference value is 0.45, the contribution of the voltage difference value is 0.35, and the contribution of the power difference value is 0.20.

[0125] The calculated state contribution is used as an update coefficient to correct the first elasticity evaluation parameter. The correction method is to multiply each dimension component of the first elasticity evaluation parameter by the corresponding state contribution and then recombine to obtain the second elasticity evaluation parameter. In power system elasticity evaluation, if the first elasticity evaluation parameter is [0.8, 0.7, 0.6], multiplied by the state contribution [0.45, 0.35, 0.20], the corrected second elasticity evaluation parameter is [0.36, 0.245, 0.12], and the comprehensive elasticity evaluation value is 0.725.

[0126] In an optional implementation, the second elasticity evaluation parameter is decomposed in the state space to obtain multi-dimensional feature components, a system stability index of the multi-dimensional feature components is calculated, an optimization objective function is determined according to the system stability index, and an optimal elasticity configuration parameter is determined based on the optimization objective function, including:

[0127] The second elasticity evaluation parameter is mapped to the state space, an orthogonal basis vector is constructed in the state space, the second elasticity evaluation parameter is decomposed in the time dimension and the space dimension based on the orthogonal basis vector, and time dimension feature components and space dimension feature components are obtained;

[0128] The time dimension feature components are calculated to obtain fluctuation period features and trend features, and the space dimension feature components are calculated to obtain distribution density features and aggregation features;

[0129] The fluctuation period features and the trend features are combined to obtain a time stability index, and the distribution density features and the aggregation features are combined to obtain a space stability index;

[0130] A target function is determined according to the time stability index and the space stability index, a time dimension constraint condition is obtained based on the period component in the time dimension feature components, and a space dimension constraint condition is obtained based on the density component in the space dimension feature components;

[0131] Based on the time dimension constraint condition, the space dimension constraint condition and the target function, the optimization target function with constraint conditions is determined, an initial solution space is generated, and the optimal solution of the optimization target function is iteratively calculated in the initial solution space to obtain optimal elastic configuration parameters.

[0132] After obtaining the second elastic evaluation parameter, state space decomposition is needed to obtain the optimal elastic configuration parameter. This method first maps the second elastic evaluation parameter to the state space. Specifically, it is assumed that the second elastic evaluation parameter is a matrix containing the load recovery rate of each node in the power system, which contains information of both time and space dimensions. Linear transformation is used to map the parameter to a high-dimensional state space to form a vector set describing the system state.

[0133] In the state space, orthogonal basis vectors are constructed to facilitate subsequent decomposition, which can be obtained by eigenvalue decomposition or singular value decomposition. For example, for the second elastic evaluation parameter containing 100 nodes and 24 hours of monitoring data, a corresponding set of orthogonal basis vectors can be constructed. Based on these orthogonal basis vectors, the second elastic evaluation parameter is decomposed by dimension reduction, thereby obtaining time dimension characteristic components and space dimension characteristic components. The time dimension characteristic components reflect the time variation characteristics in the system recovery process, while the space dimension characteristic components reflect the spatial distribution characteristics between different regions or nodes in the system.

[0134] Further processing of the time dimension characteristic components includes calculating fluctuation period characteristics and trend characteristics. Fluctuation period characteristics are obtained by analyzing the periodic changes of time series. For example, the sliding window method can be used to identify the fluctuation rules of load recovery rate, and the main fluctuation periods of 2 hours, 4 hours and 8 hours, etc. are calculated. Trend characteristics are obtained by analyzing the overall trend of time series, such as using the local weighted regression scatter smoothing method to extract the growth trend of load recovery rate with time, and obtaining the trend slope of 0.05 / hour.

[0135] The analysis of the space dimension characteristic components includes calculating the distribution density characteristics and the aggregation characteristics. The distribution density characteristics reflect the distribution of load recovery rates in different regions of the system, which can be obtained by calculating the probability density function of the space dimension characteristic components. For example, the distribution density of load recovery rate in a certain region is obtained as 0.4 nodes / recovery rate unit. The aggregation characteristics reflect the aggregation degree of nodes with similar load recovery rates in space, which can be obtained by spatial autocorrelation analysis method, such as the calculated Moran's index of 0.65, indicating that both high recovery rate regions and low recovery rate regions have strong spatial aggregation.

[0136] The time stability index can be obtained by combining the fluctuation period feature and the trend feature. Specifically, a weighted combination function can be constructed, and the fluctuation period feature is given a weight of 0.4 and the trend feature is given a weight of 0.6, so that the time stability index value is 0.78. Similarly, the spatial stability index can be obtained by combining the distribution density feature and the aggregation feature. The distribution density feature is given a weight of 0.5 and the aggregation feature is given a weight of 0.5, and the spatial stability index value is calculated to be 0.72.

[0137] Based on the time stability index and the spatial stability index, an optimized objective function is determined. The objective function can be designed as a weighted sum of the two stability indexes, for example, the time stability index weight is 0.6 and the spatial stability index weight is 0.4, and the goal is to maximize the weighted sum. In addition, the constraint conditions also need to be considered. Based on the period component in the time dimension feature component, the time dimension constraint condition is obtained, such as requiring the system recovery time to be no more than 12 hours and the main fluctuation period to be no less than 3 hours. Based on the density component in the spatial dimension feature component, the spatial dimension constraint condition is obtained, such as requiring the load recovery rate of each region to be no less than 0.4 and the difference in recovery rate between regions to be no more than 0.3.

[0138] In combination with the objective function and the constraint condition, an optimization objective function with constraint conditions is constructed. When solving the optimization problem, an initial solution space is first generated. Specifically, 100 sets of elastic configuration parameters can be randomly generated within the parameter range that meets the constraint conditions as the initial solution. Subsequently, iterative calculation is performed in the initial solution space. An optimization algorithm such as the particle swarm algorithm is used to search for the optimal solution in the solution space, and the objective function value is calculated and the parameters in the solution space are updated each time. After 500 iterations, it is found that when the elastic resource allocation ratio is 40% allocated to high-risk areas, 35% to medium-risk areas, and 25% to low-risk areas, and the recovery strategy adopts the mode of first guaranteeing important loads and then recovering general loads, the maximum stability index value 0.85 can be obtained. This set of parameters is the optimal elastic configuration parameter, which is taken as the final output result.

[0139] Through the above, the elastic configuration of the power system is optimized, the recovery ability and stability of the system in the face of sudden events are improved, and more efficient and reliable power supply guarantee is achieved.

[0140] In an optional implementation, the time stability index is obtained by combining the fluctuation period feature and the trend feature, and the spatial stability index is obtained by combining the distribution density feature and the aggregation feature, including:

[0141] Based on the statistical moments of the fluctuation period feature and the trend feature, a period feature sequence and a trend feature sequence are determined, and the time correlation coefficients of the period feature sequence and the trend feature sequence are calculated to obtain a period correlation matrix and a trend correlation matrix.

[0142] determining a density feature sequence and a clustering feature sequence based on the local density center of the distribution density feature and the inter-cluster similarity of the clustering feature, calculating a spatial correlation coefficient of the density feature sequence and the clustering feature sequence to obtain a density correlation matrix and a clustering correlation matrix;

[0143] determining a first coupling strength of the periodic correlation matrix and the trend correlation matrix, and determining an initial time stability index according to the first coupling strength;

[0144] calculating fluctuation characteristics of the initial time stability index under multiple time scales, and correcting the initial time stability index according to the fluctuation characteristics to obtain a time stability index;

[0145] determining a second coupling strength of the density correlation matrix and the clustering correlation matrix, and determining an initial spatial stability index according to the second coupling strength;

[0146] calculating distribution characteristics of the initial spatial stability index under multiple spatial scales, and correcting the initial spatial stability index according to the distribution characteristics to obtain a spatial stability index.

[0147] The specific embodiments of combining fluctuation period features and trend features to obtain a time stability index and combining distribution density features and clustering features to obtain a spatial stability index are as follows:

[0148] In this embodiment, the process of obtaining the time stability index needs to first determine a periodic feature sequence and a trend feature sequence based on the statistical moments of the fluctuation period feature and the trend feature. Specifically, the fluctuation period feature can be extracted from the input data, including statistical moment information such as maximum fluctuation amplitude, average fluctuation amplitude, and fluctuation frequency, to form a periodic feature sequence {p1, p2,..., pn}, where n is the sequence length. At the same time, the trend feature is extracted from the input data, including statistical moment information such as growth rate, trend slope, and inflection point position, to form a trend feature sequence {t1, t2,..., tn}. For example, for certain commodity sales data, the monthly fluctuation period is 3.2 months, and the fluctuation amplitude is 12.5%, which are taken as the periodic features; the annual growth rate is 8.7%, and the quarterly trend slope is 2.3%, which are taken as the trend features. n n

[0149] ​​The time correlation coefficients of the periodic characteristic sequence and the trend characteristic sequence are calculated to obtain a periodic correlation matrix C_p and a trend correlation matrix C_t. In actual applications, the sliding time window method can be used to calculate the correlation coefficients, and the window size can be set to 20% of the data length. For the above-mentioned commodity sales data, assuming that there are 60 months of data, the window size is 12 months, the adjacent window overlap rate is set to 50%, and a total of 10 time windows are generated. By calculating the correlation coefficients between the characteristic sequences in each window, a 10x10 periodic correlation matrix C_p and a trend correlation matrix C_t are formed.

[0150] The density characteristic sequence and the aggregation characteristic sequence are determined based on the similarity between the local density centers of the distribution density characteristics and the clusters of the aggregation characteristics. Specifically, the local density center points can be identified by calculating the point density distribution of different regions to form a density characteristic sequence {d1, d2,..., d m}, where m is the number of regions. At the same time, the aggregation characteristics between different regions are calculated, such as the average distance of points within a region and the separation degree between regions, to form an aggregation characteristic sequence {c1, c2,..., c m}. For example, for certain city traffic flow data, the city can be divided into 25 regions, and the traffic density value of each region is calculated as the density characteristic, and the aggregation degree of vehicles within the region is calculated as the aggregation characteristic.

[0151] The spatial correlation coefficients of the density characteristic sequence and the aggregation characteristic sequence are calculated to obtain a density correlation matrix C_d and an aggregation correlation matrix C_c. In actual applications, the spatial weight matrix method can be used to calculate the correlation coefficients, and the weight can be determined based on the distance decay function between regions. For the above-mentioned city traffic data, a 25x25 spatial weight matrix can be constructed to calculate the density correlation matrix C_d and the aggregation correlation matrix C_c, and the matrix element values are between -1 and 1.

[0152] The first coupling strength of the periodic correlation matrix C_p and the trend correlation matrix C_t is determined, and the initial time stability index is determined according to the coupling strength. The coupling strength can be calculated by weighted summation of matrix elements, and the weight can be set to the inverse of the standard deviation of the matrix elements. For the above-mentioned commodity sales data, assuming that the first coupling strength calculated is 0.75, then the initial time stability index can be set to the coupling strength multiplied by an adjustment coefficient of 1.2, obtaining 0.9.

[0153] The fluctuation characteristics of the initial time stability index at multiple time scales are calculated, and the initial time stability index is corrected according to the fluctuation characteristics to obtain a final time stability index. The time scales can be set to multiple levels such as day, week, month, quarter, etc., and the fluctuation characteristics include the fluctuation amplitude and consistency of the index at different scales. The correction method can use a penalty factor based on the fluctuation amplitude. The greater the fluctuation, the more severe the penalty. For the above data, the fluctuation amplitudes at the day, week, and month scales are 15%, 10%, and 5%, respectively. Using a penalty factor of 0.95^(average fluctuation amplitude), the final corrected time stability index is 0.87.

[0154] A second coupling strength of the density-related matrix C_d and the aggregation-related matrix C_c is determined, and an initial spatial stability index is determined according to the coupling strength. The coupling strength can be obtained by eigenvalue decomposition of the matrix, and the eigenvector corresponding to the maximum eigenvalue is taken as the index basis. For the above urban traffic data, assuming that the calculated second coupling strength is 0.68, the initial spatial stability index can be set as a function relationship of the coupling strength and the spatial complexity. When the spatial complexity is 0.35, the initial spatial stability index is 0.78.

[0155] The distribution characteristics of the initial spatial stability index at multiple spatial scales are calculated, and the initial spatial stability index is corrected according to the distribution characteristics to obtain a final spatial stability index. The spatial scales can be set to multiple levels such as point, region, city, etc., and the distribution characteristics include the distribution uniformity and aggregation degree of the index at different scales. The correction method can use an adjustment coefficient based on the entropy value. The more uniform the distribution, the higher the entropy value, and the larger the adjustment coefficient. For the above data, the entropy values at the point, region, and city scales are 0.65, 0.72, and 0.83, respectively. Using an adjustment coefficient of 1+0.1×(average entropy value), the final corrected spatial stability index is 0.81.

[0156] The time stability index is obtained by effectively combining the fluctuation period characteristics and the trend characteristics, and the spatial stability index is obtained by effectively combining the distribution density characteristics and the aggregation characteristics, which provides a quantitative basis for evaluating the temporal and spatial stability of the system.

[0157] In an optional implementation, a system resource scheduling strategy is generated according to the optimal elasticity configuration parameter, and the communication system is elastically optimized and configured, including:

[0158] A system resource capacity threshold is calculated based on the optimal elasticity configuration parameter, the system resource capacity threshold is divided into multiple load intervals, and a resource scheduling priority is set in each load interval;

[0159] The resource utilization and the service request amount in the real-time running data are extracted, and the change trend characteristics of the resource utilization and the service request amount are calculated.

[0160] determine a current load interval according to the change trend feature of the resource utilization, predict a next time load interval based on the change trend feature of the service request amount, and calculate a transition probability of adjacent load intervals;

[0161] calculate a resource dynamic expansion and contraction capacity according to the transition probability, map the resource dynamic expansion and contraction capacity and a resource scheduling priority to a resource allocation sequence, and determine a resource recovery time sequence and a resource allocation time sequence based on the resource allocation sequence;

[0162] combine the resource recovery time sequence and the resource allocation time sequence into a resource scheduling time sequence, calculate a resource scheduling step in each scheduling period of the resource scheduling time sequence, and perform hierarchical scheduling on system resources according to the resource scheduling step to perform elastic optimization configuration on the communication system.

[0163] In this embodiment, when calculating the system resource capacity threshold based on the optimal elastic configuration parameter, a resource capacity upper and lower threshold calculation method can be used. Specifically, the CPU utilization threshold in the optimal elastic configuration parameter is set to 75%, the memory utilization threshold is set to 80%, and the network bandwidth utilization threshold is set to 65%. According to these thresholds, the resource capacity is divided into three load intervals, i.e., a low load interval (0%-40%), a medium load interval (40%-70%), and a high load interval (70%-100%). In the low load interval, the resource scheduling priority is set as: network resource> memory resource> CPU resource; in the medium load interval, the resource scheduling priority is set as: memory resource> CPU resource> network resource; and in the high load interval, the resource scheduling priority is set as: CPU resource> network resource> memory resource. This priority setting can flexibly allocate resources according to different load conditions and improve the overall performance of the system.

[0164] In extracting resource utilization and business request volume in real-time running data, resource utilization data and business request volume data are collected every 5 minutes, for example, in a certain collection, CPU utilization is 65%, memory utilization is 72%, network bandwidth utilization is 58%, and business request volume is 8000 times per second. By analyzing the data of 10 consecutive sampling points, the change trend characteristics of resource utilization are calculated. Specifically, for CPU utilization, the change rate is obtained by calculating the difference between adjacent time points, such as the CPU utilization change rate of the last 10 sampling points is [+2%, +3%, +1%, -1%, +5%, +4%, +2%, -2%, +1%], and the change trend of CPU utilization is upward trend with a growth rate of about 1.67% by averaging these change rates. Similarly, the change trend of memory utilization is upward trend with a growth rate of about 1.2%, and the change trend of network bandwidth utilization is fluctuation trend with a fluctuation range of about ±3%. For business request volume, the change trend is also upward trend with a growth rate of about 500 times per minute.

[0165] According to the change trend characteristics of resource utilization to determine the current load interval, the comprehensive evaluation results of CPU utilization 65%, memory utilization 72% and network bandwidth utilization 58% determine that the current system is in the medium load interval. Based on the upward trend of business request volume, it is predicted that at the next time point (5 minutes later), the CPU utilization will reach about 66.7%, the memory utilization will reach about 73.2%, and it will still be in the medium load interval, but close to the high load interval boundary. The transition probability from the current medium load interval to the high load interval is calculated as 0.35, the transition probability to the low load interval is 0.05, and the probability of maintaining in the medium load interval is 0.6.

[0166] According to the transition probability to calculate the resource dynamic expansion and contraction capacity, it is determined that the resource amount that needs to be expanded in advance is: CPU core number increases by 2, memory capacity increases by 1GB, and network bandwidth increases by 50Mbps. These expansion capacities are mapped to resource allocation sequence according to resource scheduling priority: [memory+1GB, CPU+2core, network bandwidth+50Mbps]. According to the system load condition, the resource recycling timing is determined as: if the load interval decreases to the low load interval for 3 consecutive sampling periods (15 minutes), the resources are recycled in the order of [network bandwidth-25Mbps, CPU-1core, memory-512MB]. The resource allocation timing is: if the load interval is medium load interval in the current sampling period and the transition probability to high load interval is greater than 0.3, resources are allocated immediately in the order of [memory+512MB, CPU+1core], if the next sampling period still has the trend of migrating to high load interval, resources are allocated in the order of [memory+512MB, CPU+1core, network bandwidth+50Mbps].

[0167] After the resource recycling timing and the resource allocation timing are combined into the resource scheduling timing, the resource scheduling step is calculated in each scheduling period (5 minutes), for example, in the above scenario, the resource scheduling step of the first scheduling period is: memory + 512MB, CPU + 1 core; the resource scheduling step of the second scheduling period is: memory + 512MB, CPU + 1 core, network bandwidth + 50Mbps. According to the scheduling steps, the system resources are scheduled in stages: first, the memory resources are allocated, and the additional 512MB memory is allocated to the key business processing module; second, the CPU resources are allocated, and the increased 1 CPU core is allocated to the data processing unit; finally, the network bandwidth resources are allocated, and the increased 50Mbps bandwidth is allocated to the external interface service module.

[0168] Through the above resource scheduling strategy, the resource allocation can be dynamically adjusted according to the actual load condition and the prediction result, so that the communication system can maintain efficient and stable operation under various load conditions.

[0169] The communication elasticity evaluation model of the embodiment of the application is constructed and a dynamic optimization training system includes:

[0170] A first unit is configured to obtain historical operation data of a communication system, determine a feature parameter matrix according to the historical operation data, perform feature transformation processing on the feature parameter matrix, and obtain a communication system feature parameter.

[0171] A second unit is configured to extract a long-term communication efficiency index and a short-term fluctuation index based on the communication system feature parameter, and calculate a first elasticity evaluation parameter of each communication endpoint based on the long-term communication efficiency index and the short-term fluctuation index.

[0172] A third unit is configured to obtain real-time operation data of the communication system, calculate a statistical feature difference value of the real-time operation data and the historical operation data, update the first elasticity evaluation parameter based on the statistical feature difference value, and obtain a second elasticity evaluation parameter.

[0173] A fourth unit is configured to perform state space decomposition on the second elasticity evaluation parameter to obtain a multi-dimensional feature component, calculate a system stability index of the multi-dimensional feature component, determine an optimization objective function according to the system stability index, and determine an optimal elasticity configuration parameter based on the optimization objective function.

[0174] A fifth unit is configured to generate a system resource scheduling strategy according to the optimal elasticity configuration parameter, and perform elasticity optimization configuration on the communication system. In a third aspect, the embodiment of the application provides an electronic device, which includes:

[0175] A processor;

[0176] a memory for storing processor-executable instructions;

[0177] wherein the processor is configured to invoke the instructions stored by the memory to perform the method as described above.

[0178] In a fourth aspect, the present application provides a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the method as described above.

[0179] The present application can be a method, an apparatus, a system, and / or a computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions stored therein, which, when executed by a processor, perform various aspects of the present application.

[0180] It should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application, rather than limiting the present application; although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that the technical solutions recorded in the above-mentioned embodiments can be modified or equivalent replacements can be made to some or all of the technical features; and the modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for constructing and dynamically optimizing a communication resilience evaluation model, characterized in that, The method comprises the following steps: obtaining historical running data of a communication system, determining a characteristic parameter matrix according to the historical running data, performing characteristic transformation processing on the characteristic parameter matrix to obtain a communication system characteristic parameter; extracting a long-term communication efficiency index and a short-term fluctuation index based on the communication system characteristic parameter, and calculating a first elasticity evaluation parameter of each communication endpoint based on the long-term communication efficiency index and the short-term fluctuation index; obtaining real-time running data of the communication system, calculating a statistical characteristic difference value between the real-time running data and the historical running data, updating the first elasticity evaluation parameter based on the statistical characteristic difference value to obtain a second elasticity evaluation parameter; performing state space decomposition on the second elasticity evaluation parameter to obtain a multi-dimensional characteristic component, calculating a system stability index of the multi-dimensional characteristic component, determining an optimization objective function according to the system stability index, and determining an optimal elasticity configuration parameter based on the optimization objective function; generating a system resource scheduling strategy according to the optimal elasticity configuration parameter, and performing elasticity optimization configuration on the communication system.

2. The method of claim 1, wherein, The method comprises the following steps: performing time-frequency joint analysis based on the communication system characteristic parameter to obtain a multi-scale feature sequence, extracting a long-term stable mode from the multi-scale feature sequence to obtain a long-term communication efficiency index, and extracting a short-term mutation mode to obtain a short-term fluctuation index; segmenting the long-term communication efficiency index to extract a throughput sub-index and a time delay sub-index, and performing double exponential smoothing on the throughput sub-index and the time delay sub-index to obtain a system steady-state factor; performing dynamic layering on the short-term fluctuation index to extract a load sub-index and a jitter sub-index, and performing distribution probability conversion on the load sub-index and the jitter sub-index to obtain a fluctuation factor; calculating the cross-layer interaction strength based on the system steady-state factor and the fluctuation factor to obtain a correlation strength matrix, determining a probability weighted graph using the correlation strength matrix, and calculating a first elasticity evaluation parameter of each communication endpoint according to the probability weighted graph.

3. The method of claim 1, wherein, The method comprises the following steps: obtaining real-time running data of the communication system, performing multi-dimensional decomposition on the real-time running data to extract a time domain data sequence and a frequency domain data sequence, combining and reconstructing the time domain data sequence and the frequency domain data sequence to obtain real-time communication system characteristic parameters; extracting a real-time communication efficiency index and a real-time fluctuation index from the real-time communication system characteristic parameters, and performing feature fusion on the real-time communication efficiency index and the real-time fluctuation index to obtain a real-time state vector; obtain a historical state vector corresponding to historical operation data, calculate probability distribution differences and state transition probabilities of the real-time state vector and the historical state vector in multiple time scales, and obtain a statistical feature difference value based on the probability distribution differences and the state transition probabilities; perform time series decomposition and reconstruction on the statistical feature difference value, determine state contribution degrees of each dimension difference value in the statistical feature difference value, correct the first elasticity evaluation parameter by using the state contribution degrees as update coefficients, and obtain a second elasticity evaluation parameter.

4. The method of claim 3, wherein, perform time series decomposition and reconstruction on the statistical feature difference value, determine state contribution degrees of each dimension difference value in the statistical feature difference value, correct the first elasticity evaluation parameter by using the state contribution degrees as update coefficients, and obtain a second elasticity evaluation parameter, including: convert the statistical feature difference value into a difference value sequence, decompose the difference value sequence to obtain multi-scale coefficients, determine an optimal decomposition layer number according to energy distribution of the multi-scale coefficients, and reconstruct the difference value sequence to obtain time series components; group the time series components according to fluctuation frequencies, calculate autocorrelation coefficients of each group of time series components, filter stable time series components according to the autocorrelation coefficients, and reconstruct the stable time series components to obtain a difference value reconstructed sequence; construct a three-dimensional state space for the difference value reconstructed sequence, calculate a phase space trajectory in the three-dimensional state space, and extract topological features of the phase space trajectory to obtain a state evolution graph; calculate state transition probabilities and residence times according to the state evolution graph, construct a state duration matrix by combining the state transition probabilities and the residence times, calculate state contribution degrees of each dimension difference value based on the state duration matrix, correct the first elasticity evaluation parameter by using the state contribution degrees as update coefficients, and obtain a second elasticity evaluation parameter.

5. The method of claim 1, wherein, perform state space decomposition on the second elasticity evaluation parameter to obtain multi-dimensional feature components, calculate system stability indexes of the multi-dimensional feature components, determine an optimization objective function according to the system stability indexes, and determine optimal elasticity configuration parameters based on the optimization objective function, including: map the second elasticity evaluation parameter to a state space, construct an orthogonal basis vector in the state space, perform dimension reduction decomposition on the second elasticity evaluation parameter based on the orthogonal basis vector, and obtain time dimension feature components and space dimension feature components; calculate fluctuation period features and trend features of the time dimension feature components, and calculate distribution density features and aggregation features of the space dimension feature components; combine the fluctuation period features and the trend features to obtain a time stability index, and combine the distribution density features and the aggregation features to obtain a space stability index; determine an objective function according to the time stability index and the space stability index, obtain a time dimension constraint condition based on a period component in the time dimension feature components, and obtain a space dimension constraint condition based on a density component in the space dimension feature components; and determine an optimal elasticity configuration parameter based on the objective function, the time dimension constraint condition, and the space dimension constraint condition. Determine the optimization objective function with constraints based on the time dimension constraint condition, the space dimension constraint condition and the target function, generate an initial solution space, and iteratively calculate the optimal solution of the optimization objective function in the initial solution space to obtain optimal elasticity configuration parameters.

6. The method of claim 5, wherein, Combine the fluctuation period feature and the trend feature to obtain a time stability index, and combine the distribution density feature and the aggregation feature to obtain a space stability index, including: Determine a period feature sequence and a trend feature sequence based on the statistical moments of the fluctuation period feature and the trend feature, calculate the time correlation coefficients of the period feature sequence and the trend feature sequence to obtain a period correlation matrix and a trend correlation matrix; Determine a density feature sequence and an aggregation feature sequence based on the local density center of the distribution density feature and the inter-cluster similarity of the aggregation feature, calculate the spatial correlation coefficients of the density feature sequence and the aggregation feature sequence to obtain a density correlation matrix and an aggregation correlation matrix; Determine the first coupling strength of the period correlation matrix and the trend correlation matrix, and determine an initial time stability index according to the first coupling strength; Calculate the fluctuation characteristics of the initial time stability index at multiple time scales, and modify the initial time stability index according to the fluctuation characteristics to obtain a time stability index; Determine the second coupling strength of the density correlation matrix and the aggregation correlation matrix, and determine an initial space stability index according to the second coupling strength; Calculate the distribution characteristics of the initial space stability index at multiple spatial scales, and modify the initial space stability index according to the distribution characteristics to obtain a space stability index.

7. The method of claim 1, wherein, Generate a system resource scheduling strategy according to the optimal elasticity configuration parameters, and perform elastic optimization configuration on the communication system, including: Based on the optimal elasticity configuration parameters, calculate the system resource capacity threshold, divide the system resource capacity threshold into multiple load intervals, and set resource scheduling priorities in each load interval; Extract the resource utilization and service request volume in the real-time running data, and calculate the change trend feature of the resource utilization and service request volume; Determine the current load interval according to the change trend feature of the resource utilization, predict the load interval at the next time based on the change trend feature of the service request volume, and calculate the transition probability of adjacent load intervals; According to the transition probability, calculate the resource dynamic expansion and contraction capacity, map the resource dynamic expansion and contraction capacity and the resource scheduling priority to a resource allocation sequence, and determine the resource recovery time sequence and the resource allocation time sequence based on the resource allocation sequence; Combine the resource recovery time sequence and the resource allocation time sequence into a resource scheduling time sequence, calculate the resource scheduling step in each scheduling period of the resource scheduling time sequence, and perform hierarchical scheduling on the system resources according to the resource scheduling step to perform elastic optimization configuration on the communication system.

8. A system for constructing and dynamically optimizing a training system for a communication resilience evaluation model for implementing the method according to any one of claims 1 to 7, characterized in that, including: The first unit is configured to obtain historical running data of a communication system, determine a feature parameter matrix based on the historical running data, and perform feature transformation processing on the feature parameter matrix to obtain a communication system feature parameter. The second unit is configured to extract a long-term communication performance index and a short-term fluctuation index based on the communication system characteristic parameters, and calculate a first elasticity evaluation parameter of each communication endpoint based on the long-term communication performance index and the short-term fluctuation index. The third unit is configured to obtain real-time operation data of the communication system, calculate a statistical feature difference value between the real-time operation data and the historical operation data, update the first elasticity evaluation parameter based on the statistical feature difference value, and obtain a second elasticity evaluation parameter. The fourth unit is configured to perform state space decomposition on the second elasticity evaluation parameter to obtain a multi-dimensional feature component, calculate a system stability index of the multi-dimensional feature component, determine an optimization objective function according to the system stability index, and determine an optimal elasticity configuration parameter based on the optimization objective function. The fifth unit is configured to generate a system resource scheduling strategy according to the optimal elasticity configuration parameter, and perform elasticity optimization configuration on the communication system.

9. An electronic device, comprising: The computer program instructions are executed by the processor to implement the method in any one of claims 1 to 7. The computer program instructions are executed by the processor to implement the method in any one of claims 1 to 7. ​ ​ 10. A computer-readable storage medium having stored thereon computer program instructions, wherein, ​

Images