Network performance evaluation method and system based on multi-dimensional entropy weight fusion

Abstract

The invention discloses a network performance evaluation method and system based on multi-dimensional entropy weight fusion, and belongs to the field of network performance evaluation. The method comprises the following steps: acquiring key performance index values of a network system at each moment in a time window, correlation among key performance indexes and correlation between the key performance indexes and other performance indexes; carrying out standard price processing on the key performance index value at each moment; according to the standard values of the key performance indexes and the correlation among the key performance indexes, calculating original weight values of the key performance indexes; processing the original weight value in a time dimension and a space dimension to obtain a final weight value of each key performance index at each moment; and obtaining a network system efficiency evaluation result in the time window based on the standard values of the key performance indexes and the final weight value of each key performance index at each moment. According to the method, the service capability and the operation effect in various network environments can be comprehensively and quantitatively evaluated.

Description

Technical Field

[0001] The present invention belongs to the field of dynamic network performance evaluation, and particularly relates to a network performance evaluation method and system based on multi-dimensional entropy weight fusion. Background Art

[0002] Dynamic network performance evaluation is a key technical field for studying how to quantify the operating state of a network system in real time and comprehensively. Its core goal is to provide a scientific basis for decision-making such as network resource scheduling and service quality management through multi-dimensional index fusion analysis. This field needs to solve the limitations of traditional evaluation methods in dealing with emerging complex network forms. In the current field of network performance evaluation, traditional methods have core challenges such as fragmented evaluation dimensions, insufficient dynamic adaptability, and weak anti-interference ability, which seriously restrict the evaluation accuracy and real-time performance in complex scenarios.

[0003] First, the single-dimensional evaluation mechanism uses standardized network performance indicators and linearly scores a single dimension (such as network throughput or end-to-end delay) through a fixed threshold. Although it is simple to implement, it is difficult to depict the coupling relationship of hierarchical indicators in emerging complex network forms such as multi-modal networks. For example, in a scientific computing scenario, traditional methods may only evaluate the network state by calculating the bandwidth utilization rate between computing nodes (system-level indicator), but ignore the correlation between the task completion rate at the application layer (such as job success rate) and the resource occupancy rate of hardware devices at the resource layer. This fragmented evaluation causes the system to fail to identify abnormal states such as "high bandwidth but low task completion rate". For example, when a distributed computing task encounters a storage I / O bottleneck or network congestion causes control signaling delay, although the physical link bandwidth utilization rate is high, the actual service completion rate drops significantly, leading to resource waste and performance bottlenecks.

[0004] Second, the static weight allocation method uses a predefined expert scoring mechanism to calculate fixed weight values by constructing a judgment matrix to determine the relative importance of each indicator. Taking cloud computing resource scheduling as an example, traditional methods usually preset fixed weight ratios for CPU, memory, and disk I / O. Although the problem of multi-index fusion is solved, in a sudden AI training task, when the GPU utilization rate surges instantaneously and causes video memory contention, this preset static weight model cannot be dynamically adjusted to reflect the real impact of GPU video memory contention on system performance, resulting in a deviation between the evaluation result and the actual situation. Summary of the Invention

[0005] The present invention provides a network performance evaluation method and system based on multi-dimensional entropy weight fusion, which can comprehensively and quantitatively evaluate the service capabilities and operation effects in various network environments, and is applicable to various application scenarios such as cloud computing platforms, distributed computing systems, and data center networks. By constructing a cross-level dynamic evaluation system, this method can comprehensively reflect key performance indicators such as the service quality, resource utilization efficiency, and task execution performance of the network system, providing a reliable quantitative basis for network performance optimization and service quality assurance.

[0006] To achieve the above object, the technical solution of the present invention includes the following contents.

[0007] A network performance evaluation method based on multi-dimensional entropy weight fusion, the method comprising:

[0008] Obtaining the key performance indicator values of the network system at each moment within a time window, the correlation between the key performance indicators, and the correlation between the key performance indicators and other performance indicators;

[0009] Performing standard price processing on the key performance indicator values at each moment to obtain the standard values of the key performance indicators at each moment;

[0010] Calculating the original weight values of the key performance indicators according to the standard values of the key performance indicators and the correlation between the key performance indicators;

[0011] On the basis of the correlation between the key performance indicators and the correlation between the key performance indicators and other performance indicators, introducing a time decay factor to process the original weight values in the time dimension and the space dimension to obtain the final weight values of the key performance indicators at each moment;

[0012] Based on the standard values of the key performance indicators at each moment and the final weight values of the key performance indicators at each moment, obtaining the network system performance evaluation result within the time window.

[0013] The key performance indicators include: application-level key performance indicators, system-level key performance indicators, and resource-level key performance indicators. The application-level key performance indicators include: the difference τ between the average task delay and the deadline avg , the task delay variance and the time-transmission time ratio R comp / trans . The system-level key performance indicators include: the number of parallel tasks N par , the fairness factor F fair and the system task throughput T throughput . The resource-level key performance indicators include: the memory bandwidth occupancy rate B mem and the storage IO efficiency I IO .

[0014] Further, the other performance metrics include: temperature.

[0015] Further, based on the standard values of the key performance metrics and the correlations between the key performance metrics, calculate the original weight values of the key performance metrics, including:

[0016] Based on the standard values of the key performance metrics, calculate the information entropy e of the key performance metrics j , where j represents the ID of the key performance metric;

[0017] According to the correlations between the key performance metrics, calculate the correlation correction factor ρ of the key performance metrics j ;

[0018] Based on the information entropy e j and the correlation correction factor ρ j , obtain the original weight values w of the key performance metrics j .

[0019] Further, based on the correlations between the key performance metrics and on the basis of the correlations between the key performance metrics and other performance metrics, introduce a time decay factor to process the original weight values in the time dimension and the space dimension, and obtain the final weight values of the key performance metrics at each moment, including:

[0020] Determine whether the correlation between a key performance metric and a performance metric exceeds a set threshold, where the performance metrics include: key performance metrics and their characteristic metrics;

[0021] In the case where the correlation between the key performance metric and a performance metric exceeds the set threshold, update the original weight value based on the correlation between the key performance metric and the performance metric to obtain the latest weight value of the key performance metric

[0022] By introducing the time decay factor γ to update the weight value , obtain the final weight value of the key performance metric at time t

[0023] Further, based on the standard values of the key performance metrics at each moment and the final weight values of the key performance metrics at each moment, obtain the evaluation result of the network system efficiency within the time window, including:

[0024] According to the standard values of the key performance metrics at each moment, calculate the mean value of the key performance metric values within the time window

[0025] Based on the mean value Based on the standard values of the key performance indicators at each moment, calculate the penalty value of the j-th key performance indicator at each moment; where T is the length of the i-time window;

[0026] Based on the penalty values of the key performance indicators at each moment, the standard values of the key performance indicators at each moment, and the final weight values of the key performance indicators at each moment, obtain the evaluation result of the network system effectiveness within the time window.

[0027] A network effectiveness evaluation system based on multi-dimensional entropy weight fusion, the system includes:

[0028] A data acquisition module for obtaining the key performance indicator values of the network system at each moment within the time window, the correlation between the key performance indicators, and the correlation between the key performance indicators and other performance indicators;

[0029] A standardization module for performing standard price processing on the key performance indicator values at each moment to obtain the standard values of the key performance indicators at each moment;

[0030] An entropy weight fusion module for calculating the original weight values of the key performance indicators according to the standard values of the key performance indicators and the correlation between the key performance indicators; on the basis of the correlation between the key performance indicators and the correlation between the key performance indicators and other performance indicators, introduce a time decay factor to process the original weight values in the time dimension and space dimension to obtain the final weight values of the key performance indicators at each moment;

[0031] An effectiveness evaluation module for obtaining the evaluation result of the network system effectiveness within the time window based on the standard values of the key performance indicators at each moment and the final weight values of the key performance indicators at each moment.

[0032] An electronic device, the electronic device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the network effectiveness evaluation method based on multi-dimensional entropy weight fusion described in any one of the above is implemented.

[0033] A computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the network effectiveness evaluation method based on multi-dimensional entropy weight fusion described in any one of the above is implemented.

[0034] A computer program product, characterized in that when the computer program product runs on a computer device, the computer device is enabled to execute the network effectiveness evaluation method based on multi-dimensional entropy weight fusion described in any one of the above.

[0035] Compared with the prior art, the present invention has at least the following beneficial effects.

[0036] · Cross - hierarchical multi - modal evaluation system: Aiming at the fragmentation problem of traditional single - dimensional evaluation mechanisms, an innovative "three - level and nine - dimension" evaluation matrix is proposed to achieve cross - hierarchical coupling analysis of the resource layer, system layer, and application layer. The resource layer focuses on the utilization rate of hardware resources, the system layer quantifies network throughput and scheduling fairness, and the application layer tracks task success rate and latency fluctuations. Through the collaborative modeling of nine - dimension indicators (each layer includes time efficiency, stability, and resource efficiency),

[0037] the problem of fragmented evaluation dimensions in traditional methods is solved.

[0038] · Dynamic correlation entropy - weight model: Breaking through the limitations of static weight allocation, a dynamic weight calculation formula based on double correction of information entropy and correlation degree is proposed. By quantifying the information concentration of index data with entropy values (suppressing low - information interference terms) and calculating the cross - hierarchical index correlation intensity with the Pearson correlation coefficient (strengthening the weights of pivotal indicators), the adaptive adjustment of weights with load fluctuations is realized, and the problem of insufficient dynamic adaptability of evaluation methods is solved.

[0039] · Outlier suppression mechanism: An outlier suppression factor is introduced in the efficiency fusion calculation, and the contribution weight of abnormal indicators is dynamically attenuated through an exponential function, solving the problem of weak anti - interference ability of evaluation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Flowchart of the network efficiency evaluation method based on multi - dimensional entropy - weight fusion.

[0041] Figure 2 Comparison chart of relative errors of three schemes of the present invention and the prior art under noise interference scenarios in 600 samplings.

[0042] Figure 3 Comparison chart of volatility of the present invention and the prior art under noise interference scenarios. DETAILED DESCRIPTION OF THE INVENTION

[0043] The method of the present invention will be further described below through specific embodiments, so that those skilled in the art can have a more thorough understanding of the features and advantages of the present invention.

[0044] In a complex network environment, traditional performance evaluation methods face core challenges such as fragmented evaluation dimensions, insufficient dynamic adaptability, and weak anti-interference ability. To address this, the present invention proposes a multi-level dynamic architecture that accurately reflects key performance indicators such as the service quality, resource utilization efficiency, and task execution performance of a network system through real-time data collection, dynamic entropy weight calculation, and cross-level performance fusion, providing a reliable quantitative basis for network performance optimization and service quality assurance. The core model construction of the present invention follows the technical route of "indicator decoupling - dynamic weight assignment - cross-layer fusion", and realizes the accurate characterization of network performance through the synergistic mechanism of establishing a three-level nine-dimensional evaluation system and improving the entropy weight calculation model. The technical depth of this model is reflected in the semantic decoupling design of multi-modal indicators, the joint modeling of spatio-temporal correlation features, and the innovative implementation of the dynamic weight mechanism. The multi-level dynamic architecture adopted by the network performance evaluation method of the present invention, as Figure 1 shown, includes the following core processes.

[0045] Step 1: Obtain the key performance indicator values of the network system at each moment within a time window, the correlation between key performance indicators, and the correlation between key performance indicators and other performance indicators.

[0046] The present invention collects three-level indicators such as the application layer, system layer, and resource layer through distributed probes, thereby constructing the design concept of a three-level nine-dimensional evaluation system. The core lies in constructing an indicator system with semantic independence and functional completeness through hierarchical decoupling and orthogonal screening. Specifically as follows:

[0047]

[0048] At the architecture design level, the evaluation object is divided into three logical levels: application level, system level, and resource level. Each level corresponds to three key indicators that have passed strict orthogonality tests, forming a 3×3 indicator matrix M.

[0049] The application-level indicators focus on task execution efficiency, and the average task delay and deadline difference τ avg are selected as the core metrics. This metric can effectively characterize the degree to which the system meets the real-time requirements by calculating the absolute difference between the actual completion time and the predetermined deadline. The task delay variance is used as its supplementary indicator, and the second-order moment statistical method is used to quantify the delay fluctuation amplitude, which can detect the system load balancing state. The innovatively introduced calculation time - transmission time ratio R comp / trans highlights the efficiency game relationship between the calculation and communication links while eliminating the dimension difference through the non-linear mapping of log(1 + calculation time / network transmission time).

[0050] The system-level indicators focus on resource scheduling efficiency, and the number of parallel tasks N parDirectly reflects the system's concurrent processing ability, the scheduling fairness factor F fair Calculated using the max-min fairness algorithm to ensure the balance of resource allocation, the system task throughput T throughput Reflects the number of tasks processed by the system per unit time.

[0051] Resource-level metrics start from the hardware dimension, the CPU utilization rate u CPU An improved calculation method is adopted: First, the original time slice data collected from the operating system kernel is classified, the running time of the system state and the user state is statistically analyzed respectively, and a dynamic weighting strategy is adopted to assign different weight coefficients to the time slice according to the current task type (for example, a higher weight is given to the user state time for compute-intensive tasks). Finally, the ratio of the sum of the weighted time slices to the total sampling period time after normalization is used as the final utilization rate index u CPU . This CPU utilization rate calculation method can solve the measurement deviation problem caused by frequent system calls in traditional CPU utilization rate calculations. The memory bandwidth occupancy rate B mem The command ratio of the memory controller is collected in real time through the performance counter. Compared with the traditional memory usage monitoring, it can better reveal the performance bottleneck caused by bandwidth competition. The storage IO efficiency I IO Represents the ratio of the actual IOPS of the storage device to the theoretical maximum value, and can reflect whether the overall performance of data-intensive tasks (such as big data analysis) is dragged down due to low storage efficiency.

[0052] In addition, for the convenience of subsequent data processing, the present invention also needs to obtain the correlations between key performance indicators and the correlations between key performance indicators and other performance indicators.

[0053] Step 2: Perform standard price processing on the key performance indicator values at each moment to obtain the standard values of the key performance indicators at each moment.

[0054] In the data standardization processing stage, an adaptive sliding window mechanism is adopted to dynamically adjust the normalization interval according to the index fluctuation characteristics, enhancing the sensitivity to sudden states while retaining the long-term trend characteristics. The probe acquisition layer injects spatio-temporal context information into the original data through metadata tagging technology, laying a semantic foundation for subsequent multi-dimensional correlation analysis. The sample at time t of the jth index is processed using the following formula:

[0055]

[0056] Step 3: Calculate the original weight values of the key performance indicators based on the standard values of the key performance indicators and the correlations between the key performance indicators.

[0057] The present invention combines traditional information entropy with spatio-temporal features using a dynamic entropy weight calculation model. First, the standardized data matrix is normalized to obtain P = x″ j (t), where x″ j (t) = x′ j (t) / ∑x′ j (t). Based on this, the information entropy (when x″ j (t) = 0, this item is agreed to be 0) of each index is calculated. The entropy value e j ∈[0,1] reflects the dispersion degree of the index data. At the same time, a correlation degree correction factor is introduced to quantify the overall correlation between the key performance index j and other key performance indexes. ρ j approaching 1 indicates that there is serious redundancy in this index. Finally, the original weight w j is jointly determined by the information entropy e j and the correlation degree ρ j . The specific calculation formula is as follows:

[0058]

[0059] Step 4: Based on the correlation between key performance indexes and the correlation between key performance indexes and other performance indexes, a time decay factor is introduced to process the original weight value in the time dimension and space dimension, and the final weight value of each key performance index at each moment is obtained.

[0060] The correlation entropy weight calculation module of the present invention constructs a spatio-temporal two-dimensional dynamic weight optimization system by integrating a time decay mechanism and spatial correlation correction.

[0061] In the space dimension, the system constructs an n×n Pearson correlation coefficient matrix R = [r jk to identify the implicit dependence relationship between indexes, where r jk represents the correlation coefficient between index j and k, and the value range is [-1,1]. When strong correlation is detected (such as the correlation coefficient r jk > 0.8 between CPU utilization rate and temperature), the system will automatically trigger a weight penalty mechanism and reduce the weight of the corresponding key performance index through a linear decay function .

[0062] In the time dimension, the system adopts an exponential fading memory model (decay factor γ = 0.9, where γ∈(0,1) controls the decay rate of historical data) to assign higher weights to recent data, ensuring that the weight allocation can timely reflect the transient characteristics of the system. Specifically, the weight of the latest data at the current moment t = 0 is γ 0 = 1, and the weight of the data at the previous moment is γ1 , and so on. After that, combined with the above, the weight values are obtained to obtain the final weight values at each time point Among them, the sum of the weights within the time window T satisfies the normalization condition (1 - γ T ) / (1 - γ).

[0063] Step 5: Based on the standard values of the key performance indicators at each moment and the final weight values of each key performance indicator at each moment, obtain the evaluation result of the network system effectiveness within the time window.

[0064] In the effectiveness fusion stage, a dual mechanism of outlier suppression and dynamic aggregation is adopted to ensure the evaluation robustness. This mechanism combines the statistical characteristics of the index data with the dynamic weights through mathematical modeling, and constructs an evaluation system with adaptive filtering characteristics.

[0065] The outlier suppression function applies non-linear attenuation based on the degree of deviation of the index from the mean value, and its core expression is the exponential attenuation term where x′ j (t) represents the normalized value of the j-th index at time t (the value after min-max normalization processing), is the mean value of this index within the sliding time window T β is the outlier suppression intensity coefficient (β > 0, typical value 1.5 - 3.0), and this coefficient controls the penalty strength for outliers. The larger the β value, the more severe the weight attenuation imposed on the index deviating from the mean value. When the instantaneous value x′ j (t) of a certain index deviates from the mean value by more than 2 standard deviations (i.e., where σ is the standard deviation), this exponential term will reduce its weight to less than 15% of the original value, thus effectively suppressing the influence of instantaneous interference and hardware false alarms on the comprehensive evaluation value.

[0066] The dynamic aggregation algorithm realizes the evaluation effect with band-pass filtering characteristics by fusing the time-varying weight w j (t) and the outlier suppression factor. Among them, w j (t) represents the dynamic weight of the j-th index at time t (satisfying ), and its value is updated in real time by the correlation entropy weight calculation module, which can reflect the change of the importance of the index in the spatio-temporal dimension. The final fusion evaluation value E(t) is calculated by the weighted sum formula:

[0067]

[0068] Among them, through the exponential term non-linear punishment is imposed on the index deviating from the mean value . In this way, the present invention calculates the comprehensive effectiveness value and locates the bottleneck based on the weight matrix and the outlier suppression function

[0069] The performance analysis of the network effectiveness evaluation method based on multi-dimensional entropy weight fusion of the present invention is carried out through an experiment below.

[0070] a) Experiment introduction

[0071] To verify the effectiveness of the network effectiveness evaluation method (MEF) based on multi-dimensional entropy weight fusion, a systematic experiment with 600 samplings is designed in this study. The experiment adopts a three-level nine-dimensional index generation mechanism, in which the resource layer indexes (such as CPU utilization rate, memory bandwidth occupancy rate) are collected in real time through hardware performance counters, the system layer indexes (such as network throughput, scheduling fairness) are generated by simulation, and the application layer indexes (such as task delay, completion rate) are dynamically synthesized through a discrete event simulator. To simulate the complexity of the real network environment, multi-modal noise is introduced: in the time dimension, Gaussian white noise (signal-to-noise ratio 10 - 15 dB) is superimposed on 30% of the sampling points; in the space dimension, 20% of the indexes are randomly selected and injected with pulse interference (amplitude fluctuation ±30%). The comparison schemes are the traditional static weight allocation method (STATIC evaluation scheme) and the entropy weight method based on a sliding window (SINGLE one-dimensional evaluation scheme). The STATIC evaluation scheme adopts a fixed weight ratio (application layer: system layer: resource layer = 4:3:3), and the SINGLE one-dimensional evaluation scheme uses a fixed time window (5 seconds) to calculate the information entropy weight.

[0072] b) Experiment analysis

[0073] Figure 2 Shows the relative error changes of the three methods within 600 sampling periods. First, a reference truth value is established as the reference standard, and then the output results of each evaluation method are compared with the reference truth value, and the formula is used to calculate the relative error of each sampling point.

[0074] The experimental data shows that the MEF method shows significant advantages under the dynamic anti-noise mechanism: the overall fluctuation range of its relative error is narrow, and only short peaks appear under the influence of pulse noise in some periods. In contrast, the error value of the Single method is always close to 1.0, indicating that its one-dimensional evaluation completely fails; the Static method is stable in the low-noise stage, but the error quickly climbs above 0.6 under noise interference. The superiority of the MEF method stems from its dynamic weight adjustment mechanism: in the stage of pulse noise interference, through cross-level index correlation analysis, the abnormal fluctuation of the application layer delay index is identified, and then its weight is dynamically attenuated to avoid the conduction of noise to the comprehensive evaluation value. The Static method cannot suppress the interference of abnormal indexes due to the use of fixed weight allocation, resulting in a continuously high error; the Single method completely deviates from the real load under noise interference due to relying on a single index.

[0075] Figure 3 The fluctuation coefficients of three methods under different noise intensities were compared. Experimental data show that the volatility of the MEF method is always lower than 0.05, and it only slightly rises from 0.03 to 0.045 as the noise intensity increases, with significantly better stability than other schemes. The volatility of the Static method is 0.08 at low noise, but rises to 0.10 at high noise; the volatility of the Single method sharply climbs from 0.1 to 0.35, indicating that its evaluation is completely dominated by noise. The stability of MEF benefits from a dual mechanism: when the noise intensity is 0.3, inefficient indicators with information entropy > 0.8 are dynamically screened out by entropy weight, and the weights are concentrated on high-information indicators, greatly reducing the impact of redundant noise; at the same time, through an exponential outlier suppression mechanism, a non-linear weight decay is imposed on abnormal data deviating from the mean, restricting the fluctuation range of the comprehensive evaluation value. In extreme noise scenarios, the volatility of MEF is only 12.9% of that of the Single method, and its fluctuation growth slope is significantly lower than that of Static and Single, fully demonstrating that through cross-level indicator coupling and a dynamic anti-noise mechanism, it can effectively meet the effectiveness evaluation requirements in complex noise environments.

[0076] Finally, it should be noted that the above implementation cases are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail using examples, those skilled in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A network performance evaluation method based on multi-dimensional entropy weight fusion, characterized in that, The method includes: Obtaining the key performance indicator values of the network system at each moment within the time window, the correlations between the key performance indicators, and the correlations between the key performance indicators and other performance indicators; Performing standard price processing on the key performance indicator values at each moment to obtain the standard values of the key performance indicators at each moment; Calculating the original weight values of the key performance indicators according to the standard values of the key performance indicators and the correlations between the key performance indicators; Based on the correlations between the key performance indicators and the correlations between the key performance indicators and other performance indicators, introducing a time decay factor to process the original weight values in terms of time dimension and space dimension, and obtaining the final weight values of the key performance indicators at each moment; Based on the standard values of the key performance indicators at each moment and the final weight values of the key performance indicators at each moment, obtaining the evaluation result of the network system efficiency within the time window.

2. The method according to claim 1, wherein The key performance indicators include: application-level key performance indicators, system-level key performance indicators, and resource-level key performance indicators. The application-level key performance indicators include: the difference τ between the average task delay and the deadline avg , the task delay variance and the time-transmission time ratio R comp / trans . The system-level key performance indicators include: the number of parallel tasks N par , the fairness factor F fair and the system task throughput T throughput . The resource-level key performance indicators include: the memory bandwidth occupancy rate B mem and the storage I / O efficiency I IO .

3. The method according to claim 1, wherein The other performance indicators include: temperature.

4. The method according to claim 1, wherein Calculating the original weight values of the key performance indicators according to the standard values of the key performance indicators and the correlations between the key performance indicators, including: Calculate the information entropy e of the key performance indicator based on the standard value of the key performance indicator j , where j represents the ID of the key performance indicator; Calculate the correlation degree correction factor ρ of the key performance indicators according to the correlation between the key performance indicators j ; Based on the information entropy e j and the correlation degree correction factor ρ j , the original weight value w of each key performance indicator is obtained j .

5. The method according to claim 1, characterized in that Based on the correlations between the key performance indicators and the correlations between the key performance indicators and other performance indicators, introducing a time decay factor to process the original weight values in terms of time dimension and space dimension, and obtaining the final weight values of the key performance indicators at each moment, including: Judging whether the correlation between a key performance indicator and a performance indicator exceeds a set threshold, where the performance indicators include: key performance indicators and their characteristic indicators; When the correlation between the key performance indicator and a performance indicator exceeds the set threshold, update the original weight value based on the correlation between the key performance indicator and the performance indicator to obtain the latest weight value of the key performance indicator By introducing a time decay factor γ for the weight value to update and obtain the final weight value of the key performance indicator at time t 6. The method according to claim 1, wherein Based on the standard values of the key performance indicators at each moment and the final weight values of the key performance indicators at each moment, obtaining the evaluation result of the network system efficiency within the time window, including: Calculate the mean value of the key performance indicator values within the time window based on the standard values of the key performance indicators at each moment According to the mean value and the standard value of the key performance indicator at each moment, calculate the penalty value of the j-th key performance indicator at each moment; where T is the length of the i time window; Based on the penalty values of the key performance indicators at each moment, the standard values of the key performance indicators at each moment, and the final weight values of the key performance indicators at each moment, obtaining the evaluation result of the network system efficiency within the time window.

7. A network performance evaluation system based on multi-dimensional entropy weight fusion, characterized in that, The system includes: A data acquisition module, configured to obtain the key performance indicator values of the network system at each moment within the time window, the correlations between the key performance indicators, and the correlations between the key performance indicators and other performance indicators; A standardization module, configured to perform standard price processing on the key performance indicator values at each moment to obtain the standard values of the key performance indicators at each moment; An entropy weight fusion module, configured to calculate the original weight values of the key performance indicators according to the standard values of the key performance indicators and the correlations between the key performance indicators; based on the correlations between the key performance indicators and the correlations between the key performance indicators and other performance indicators, introducing a time decay factor to process the original weight values in terms of time dimension and space dimension, and obtaining the final weight values of the key performance indicators at each moment; An efficiency evaluation module, configured to obtain the evaluation result of the network system efficiency within the time window based on the standard values of the key performance indicators at each moment and the final weight values of the key performance indicators at each moment.

8. An electronic device, characterized in that, The electronic device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the network performance evaluation method based on multi-dimensional entropy weight fusion as described in any one of claims 1-6 is implemented.

9. A computer-readable storage medium, characterized in that, Computer program instructions are stored on the computer-readable storage medium, and when the computer program instructions are executed by a processor, the network performance evaluation method based on multi-dimensional entropy weight fusion as described in any one of claims 1-6 is implemented.

10. A computer program product, characterized in that, When the computer program product runs on a computer device, the computer device is caused to execute the network performance evaluation method based on multi-dimensional entropy weight fusion as described in any one of claims 1-6.

Images