Matrix switch test data processing system based on edge calculation

Abstract

The invention discloses a matrix switch test data processing system based on edge calculation, and relates to the technical field of edge calculation. The system comprises a feature acquisition module, an error analysis module, a configuration optimization module and a data compensation module. The method comprises the following steps: respectively obtaining test error parameters and coupling error parameters by using a pre-trained error predictor through obtaining switch characteristics such as accumulated test times of a matrix switch and task characteristics such as the number of test channels; carrying out test task distribution and edge node computing power optimization based on the error parameters and the task characteristics, and obtaining an optimal test configuration scheme through iterative optimization; and finally, calling a differential training compensation agent at the edge node to carry out distributed compensation processing on the original test data. According to the invention, accurate quantification and compensation of matrix switch test errors are realized, the problems of imperfect error analysis and low resource configuration efficiency in the prior art are solved, and the accuracy of test data and the system processing efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and more specifically to a matrix switch test data processing system based on edge computing. Background Technology

[0002] Matrix switches are core switching devices in automated testing systems. Their core function lies in their ability to flexibly construct custom connection paths between input and output signal channels according to testing requirements. This allows for the efficient completion of a series of testing tasks, including assessing the connectivity, electrical parameters, and functional effectiveness of multiple electrical signal channels, thereby improving the flexibility and reusability of the testing system. In complex testing scenarios, multiple testing tasks often need to be executed in parallel. In such cases, matrix switches need to establish multiple independent signal paths simultaneously. The performance status of their internal intersections and the mutual interference between channels directly determine the fidelity of the test signals.

[0003] Existing technologies for processing matrix switch test data often employ a centralized processing architecture, sending all test data back to a central server for unified analysis. This approach struggles to handle the large volumes of real-time data generated when multiple test tasks are executed in parallel, easily leading to network congestion and processing delays. Furthermore, during testing, performance degradation of the matrix switches due to long-term use introduces inherent testing errors, while crosstalk and load effects generated during multi-channel parallel testing can cause coupling errors. This results in unverified systematic biases in the test data, ultimately making it difficult to meet the reliability and accuracy requirements of high-precision testing applications. Summary of the Invention

[0004] This invention addresses the technical problem of the lack of analysis on the performance degradation of matrix switches and the errors introduced by multi-task channel coupling in the prior art, and provides a matrix switch test data processing system based on edge computing.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0006] This invention provides a matrix switch test data processing system based on edge computing, comprising:

[0007] The feature acquisition module is used to acquire the switch features of multiple matrix switches to be tested, and to acquire the task features of multiple test tasks to be performed, wherein each task feature includes the number of test channels, and the multiple matrix switches are connected to the test station through edge nodes.

[0008] The error analysis module is used to perform test error analysis and coupling error analysis based on multiple switch features and multiple task features, respectively, to obtain multiple test error parameters and multiple coupling error parameters;

[0009] The configuration optimization module is used to allocate test tasks and optimize edge node computing power based on the test error parameters and coupling error parameters, obtain the optimal test configuration scheme, perform test configuration, and obtain multiple raw test data, wherein auxiliary optimization strategies are configured based on multiple task characteristics;

[0010] The data compensation module is used to perform test data compensation processing within multiple edge nodes by utilizing the computing power of multiple optimal edge nodes, thereby obtaining multiple compensated test data as the data processing result.

[0011] The beneficial effects of this invention are:

[0012] Compared to existing technologies, this invention first establishes a collaborative analysis mechanism for switch features and task features, enabling precise quantification of test errors and coupling errors, providing a reliable basis for subsequent optimization. Secondly, it innovatively combines error parameters with edge computing power allocation, comprehensively considering error compensation needs during the test configuration phase, achieving collaborative optimization of test tasks and computing resources. Thirdly, through a distributed data compensation architecture, compensation computation tasks are pushed down to edge nodes, effectively reducing data transmission latency and improving system response speed. Finally, it employs an agent-based compensation mechanism, dynamically invoking corresponding compensation models based on different error characteristics, improving the accuracy and reliability of test data, and overall solving the problems of incomplete error analysis and low resource utilization efficiency in existing technologies. Attached Figure Description

[0013] Figure 1 A schematic diagram of the structure of a matrix switch test data processing system based on edge computing provided by the present invention;

[0014] Figure 2 This is a schematic diagram of the principle of a matrix switch test data processing system based on edge computing provided by the present invention.

[0015] In the attached diagram, the components represented by each number are as follows:

[0016] Feature acquisition module 11, error analysis module 12, configuration optimization module 13, data compensation module 14. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0019] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0020] Example 1, as Figure 1 , Figure 2 As shown, this embodiment of the invention provides a matrix switch test data processing system based on edge computing, comprising:

[0021] The feature acquisition module 11 is used to acquire the switch features of multiple matrix switches to be tested, and to acquire the task features of multiple test tasks to be performed, wherein each task feature includes the number of test channels, and the multiple matrix switches are connected to the test station through edge nodes.

[0022] Specifically, the switching characteristics of multiple matrix switches to be tested are obtained, as well as the task characteristics of multiple test tasks to be performed, including:

[0023] Obtain the switching characteristics of multiple matrix switches to be tested, whereby the switching characteristics include the cumulative number of tests on the matrix switches;

[0024] Obtain the task characteristics of multiple test tasks to be performed, where each task characteristic includes the number of test channels for the test task.

[0025] First, the switching characteristics of the multiple matrix switches to be tested are obtained. Switching characteristics are a set of technical parameters that comprehensively reflect the current performance status of the matrix switches, including the cumulative number of tests. The cumulative number of tests is a core parameter characterizing the usage level of the matrix switch; its increase is directly related to the physical wear of the internal mechanical contacts and the gradual degradation of electrical performance. As the cumulative number of tests increases, the contact resistance of the matrix switch gradually increases, and the signal transmission quality decreases accordingly. This performance change directly leads to systematic errors in the testing process. Therefore, accurately obtaining this characteristic parameter of the cumulative number of tests provides an important data foundation for subsequent quantitative analysis of test errors.

[0026] Secondly, the task characteristics of the multiple test tasks to be performed are obtained. Task characteristics are a set of basic attributes that fully describe the technical requirements of the test tasks, with each task characteristic including the number of test channels. The number of test channels not only determines the complexity of the test task but also directly affects the signal integrity during the test process. When multiple test channels operate simultaneously, electromagnetic coupling occurs between them, forming mutually interfering coupled electrical signals. This inter-channel interference increases significantly with the number of test channels, introducing additional coupling errors and severely affecting the accuracy of the test data. Therefore, obtaining the number of test channels provides the necessary input for subsequent analysis of coupling error characteristics.

[0027] By synchronously acquiring switch and task characteristics, a complete test environment parameter system was established. The performance degradation reflected by the cumulative number of tests and the inter-channel interference determined by the number of test channels together constitute the main sources of test errors. This dual-feature acquisition mechanism lays a solid data foundation for subsequent error analysis and compensation processing, enabling the system to adopt corresponding processing strategies for different error sources.

[0028] Meanwhile, in the system architecture, multiple matrix switches establish network connections with the remote test center through edge nodes deployed at the test site. This connection method enables the test center to centrally manage the distributed matrix switches, while making full use of the local computing power of the edge nodes, laying the foundation for subsequent distributed data processing.

[0029] Error analysis module 12 is used to perform test error analysis and coupling error analysis based on multiple switch features and multiple task features, respectively, to obtain multiple test error parameters and multiple coupling error parameters;

[0030] Based on the multiple switch features and multiple task features, test error analysis and coupling error analysis are performed respectively to obtain multiple test error parameters and multiple coupling error parameters, including:

[0031] Obtain the error predictor configured in the test bench during training, wherein the error predictor includes a switching error prediction branch and a coupled error prediction branch;

[0032] The multiple switching features and multiple task features are respectively input into the switching error prediction branch and the coupling error prediction branch in the error predictor, and multiple test error parameters and multiple coupling error parameters are output.

[0033] First, a pre-trained error predictor is acquired and configured in the test bench. This error predictor adopts a dual-branch architecture design, which includes a switching error prediction branch specifically for handling switching features, and a coupled error prediction branch specifically for handling task features, enabling accurate modeling for different types of error sources.

[0034] Among them, the switch error prediction branch maps the nonlinear relationship between switch characteristics and test error, and is used to accurately predict the magnitude of the test error introduced by the matrix switch based on the cumulative number of tests; the coupling error prediction branch maps the functional relationship between task characteristics and channel interference, and is used to accurately assess the degree of coupling error generated during multi-channel parallel testing based on the number of test channels.

[0035] Specifically, the configuration steps of the error predictor include:

[0036] Based on historical test and verification data of matrix switches, sample switch feature sets and sample task feature sets are collected.

[0037] The test obtains the test data error introduced under different sample switching characteristics, and obtains the sample test error parameter set; and obtains the coupling data error introduced under different sample task characteristics, and obtains the sample coupling error parameter set.

[0038] Based on machine learning, a switching error prediction branch and a coupling error prediction branch are constructed.

[0039] Using the sample switch feature set and sample test error parameter set, the switch error prediction branch is trained under supervision. Using the sample task feature set and sample coupling error parameter set, the coupling error prediction branch is trained under supervision. Convergence is tested separately to obtain the error predictor, which is then configured in the test center.

[0040] First, based on the historical test verification data of the matrix switches accumulated during historical testing, sample switch feature sets and sample task feature sets are collected. The historical test verification data of the matrix switches consists of real test records and verification results recorded during long-term testing activities, including the difference between standard signal source test data and actual output test data, signal crosstalk measurements between different test channels, and the statistical variance of multiple repeated tests, obtained through high-precision measurement instrument calibration and continuous monitoring by a data acquisition system. The collected sample switch feature sets and sample task feature sets contain rigorously selected and labeled historical data samples, which are used to construct the basic dataset required for training the error predictor.

[0041] Secondly, test data error values ​​introduced under different sample switch characteristic conditions are obtained through standard testing procedures, forming a sample test error parameter set. This test error parameter is quantified as a percentage of error amplitude, specifically reflecting the impact of performance degradation phenomena such as contact wear and increased contact resistance caused by the cumulative number of tests on the accuracy of the test data. Simultaneously, coupling data error values ​​introduced under different sample task characteristic conditions are measured, forming a sample coupling error parameter set. This coupling error parameter is also quantified as a percentage of error amplitude, specifically reflecting the impact of interference phenomena such as electromagnetic coupling and signal crosstalk between channels on the integrity of the test data when the number of test channels increases. Finally, the standardized and quantified sample test error parameter set and sample coupling error parameter set establish numerical correspondences between switch performance status and test error, and between task characteristics and coupling error, respectively, providing accurate supervisory signals for the subsequent training of the error predictor.

[0042] Furthermore, an initial model architecture for a switching error prediction branch and a coupling error prediction branch is constructed based on machine learning algorithms. The switching error prediction branch is specifically designed to learn the complex mapping relationship from switching features to test errors; the coupling error prediction branch is specifically designed to learn the intrinsic correlation from task features to coupling errors.

[0043] Furthermore, the switch error prediction branch is trained under supervision using a sample switch feature set and a sample test error parameter set, enabling it to predict test error parameters based on the input switch features. Simultaneously, the coupling error prediction branch is trained under supervision using a sample task feature set and a sample coupling error parameter set, enabling it to predict coupling error parameters based on the input task features.

[0044] Once the switching error prediction branch and the coupled error prediction branch have both passed testing and verification and reached the convergence criteria, a usable error predictor is obtained. This error predictor is then deployed and configured within the test center to accurately predict and quantify various errors during real-time testing. The convergence criteria are set based on the trend of the loss function during training. For example, when the mean squared error of the training and validation sets no longer decreases significantly for 15 consecutive training epochs and reaches a preset error threshold range, such as within 3% of the relative error range, convergence is considered achieved.

[0045] For example, since there are highly nonlinear and complex relationships between switch features and test error parameters, and between task features and coupling error parameters, and deep neural network models have significant advantages in multi-level feature abstraction and complex pattern recognition, deep neural network models can be selected to construct this error predictor.

[0046] Specifically, the error predictor employs a dual-branch parallel architecture, consisting of a switching error prediction branch and a coupled error prediction branch. Each branch includes an input layer, a feature abstraction layer, and a prediction output layer. The input layer of the switching error prediction branch receives a standardized switching feature vector, which contains key performance parameters such as the cumulative number of tests. The input layer of the coupled error prediction branch receives a standardized task feature vector, which contains core task parameters such as the number of test channels. The feature abstraction layers of both branches utilize a three-layer fully connected neural network structure, with 64, 32, and 16 nodes per layer, respectively. Each layer employs the ReLU activation function to introduce non-linear transformation capabilities, and Dropout layers are embedded between network layers with a dropout rate of 0.3 to effectively suppress overfitting and improve generalization performance. The output layers of both branches use linear activation functions to map the final abstract features to continuous error magnitude percentages.

[0047] During training, key hyperparameters included a learning rate of 0.001, 200 training epochs, and a batch size of 32. The learning rate was set to balance training stability and convergence speed; the number of training epochs ensured the model fully learned error patterns from the data; and the batch size balanced training efficiency with memory consumption. Specifically, a supervised learning approach was used. The collected sample switch feature set was used as the input sample set for the switch error prediction branch, and the corresponding sample test error parameter set was used as the supervision label. Simultaneously, the sample task feature set was used as the input sample set for the coupling error prediction branch, and the corresponding sample coupling error parameter set was used as the supervision label. The input sample sets and corresponding label sample sets for each branch were divided into training, validation, and test sets in a 7:2:1 ratio.

[0048] Furthermore, using the sample feature vectors from the training set as input and the corresponding error parameter labels as supervision signals, two prediction branches are trained independently using the backpropagation algorithm and the Adam optimizer. The mean squared error loss function is used to measure the deviation between the percentage of predicted error and the actual error parameter, and the training process is monitored using a validation set. Training is terminated when the validation set loss function values ​​of both branches no longer decrease for 15 consecutive rounds and the average relative error between the percentage of predicted error and the actual value reaches a predetermined threshold, such as within 3%, resulting in a converged error predictor.

[0049] The resulting error predictor effectively captures the complex nonlinear relationship between switching features and test errors, and between task features and coupling errors, achieving accurate error prediction and providing reliable data support for subsequent test configuration optimization.

[0050] Specifically, during the error predictor phase, multiple collected switch features are input into the switch error prediction branch. This branch, based on the cumulative number of tests in the switch features and through an internally established mapping relationship, outputs quantified test error parameters. These parameters are represented as a percentage of error amplitude, specifically reflecting the degree of deviation in test results caused by performance degradation due to long-term use of the matrix switches. Simultaneously, multiple collected task features are input into the coupling error prediction branch. This branch, based on the number of test channels in the task features and through an internally established correlation model, outputs quantified coupling error parameters. These parameters are also represented as a percentage of error amplitude, specifically reflecting the degree of impact on test results caused by signal interference during multi-channel parallel testing.

[0051] This dual-branch parallel processing mechanism enables the acquisition of test error parameters for each matrix switch and coupling error parameters for each test task, providing precise numerical basis for subsequent test configuration optimization and data processing, and ensuring that error compensation is based on scientific quantification.

[0052] The configuration optimization module 13 is used to allocate test tasks and optimize edge node computing power according to the test error parameters and coupling error parameters, obtain the optimal test configuration scheme, perform test configuration, and obtain multiple raw test data, wherein auxiliary optimization strategies are configured according to multiple task characteristics;

[0053] Based on the test error parameters and coupling error parameters, test tasks are allocated and edge node computing power is optimized to obtain the optimal test configuration scheme. Test configuration is then performed, and multiple raw test data sets are obtained, including:

[0054] Obtain a computing power resource pool of multiple edge nodes, randomly configure the computing power of multiple first edge nodes, wherein the computing power of each edge node includes a computing power configuration coefficient, and the sum of the computing power of multiple edge nodes satisfies the total computing power constraint.

[0055] Multiple test tasks are randomly combined with multiple matrix switches, and the computing power of the multiple first edge nodes is combined to obtain a first test configuration scheme.

[0056] Based on the multiple test error parameters, multiple coupling error parameters, and multiple task characteristics, the first test fitness of the first test configuration scheme is calculated and obtained.

[0057] Continue iterative optimization. When stuck in a local optimum, configure the number of exits based on multiple task characteristics, generate multiple exit test configuration schemes, and optimize them.

[0058] The optimization continues until convergence, obtaining the optimal test configuration scheme with the greatest test fitness. Test configuration is then performed, and multiple raw test data are obtained.

[0059] First, a computing power resource pool consisting of multiple edge nodes is acquired, and the computing power of multiple first edge nodes is randomly allocated. Specifically, an initial computing power configuration that meets the total computing power constraint is randomly generated. The computing power of each edge node is represented by a computing power allocation coefficient, which is a dimensionless value used to quantify the relative computing power allocated to that edge node. This computing power allocation coefficient is positively correlated with the speed and accuracy of the node in actually processing test data. The sum of the computing power allocation coefficients of all nodes must meet the total computing power limit set by the system to ensure that the resource allocation meets the system's carrying capacity.

[0060] Secondly, multiple test tasks are randomly matched and combined with multiple matrix switches, and combined with the initial computing power configuration, i.e., the computing power of multiple first edge nodes, to form a first test configuration scheme. This first test configuration scheme fully defines the mapping relationship between test tasks and matrix switches, as well as the computing power allocation of each edge node.

[0061] Furthermore, the test fitness of the first test configuration scheme is calculated based on test error parameters, coupling error parameters, and task characteristics. Test fitness is a comprehensive evaluation index, whose calculation process comprehensively considers the impact of errors and the efficiency of computing power allocation. It is used to characterize the overall performance level of the current test configuration scheme in achieving the optimal balance between test accuracy and processing efficiency under specific resource constraints.

[0062] Specifically, based on the multiple test error parameters, multiple coupling error parameters, and multiple task characteristics, the first test fitness of the first test configuration scheme is calculated, including:

[0063] Based on multiple test error parameters and multiple coupling error parameters, multiple fusion error parameters of multiple matrix switches within the first test configuration scheme are calculated.

[0064] The ratios of computing power of multiple first edge nodes and multiple fusion error parameters are calculated respectively, and the first compensation fitness is obtained by processing them.

[0065] The first processing fitness is calculated based on the computing power of multiple first edge nodes and the characteristics of multiple tasks.

[0066] The first test fitness is calculated based on the first compensation fitness and the first processing fitness.

[0067] First, the test error parameters and coupling error parameters are fused. Based on the test error parameters and coupling error parameters corresponding to each matrix switch in the first test configuration scheme, a weighted summation algorithm is used to calculate the fused error parameter for each matrix switch. This fused error parameter comprehensively characterizes the overall error level that may be generated when a specific test task is executed on a specified matrix switch. Specifically, the fused error parameter = test error parameter × α + coupling error parameter × β. The weighting coefficients are set according to the criticality level of the test task. For example, for high-precision test tasks, since the performance accuracy of the equipment itself plays a decisive role in the test results, the weighting coefficient α of the test error parameter is set to 0.7, and the weighting coefficient β of the coupling error parameter is set to 0.3; for regular test tasks, since inter-channel interference becomes the main source of error during multi-channel parallel testing, a weighting coefficient α of 0.4 and a weighting coefficient β of 0.6 are adopted.

[0068] Secondly, compensation fitness is calculated. Specifically, the computing power of the first edge node configured for each edge node is compared and analyzed with the fusion error parameters generated by the corresponding test task. The ratio of the computing power configuration coefficient of each node to the fusion error parameter is calculated. All calculated ratios are normalized and arithmetically averaged to obtain the first compensation fitness. This first compensation fitness reflects the ability of the current computing power configuration to compensate for test errors. The larger the value, the higher the matching degree between computing power allocation and error compensation requirements, reflecting the optimized configuration effect of edge computing resources in dealing with device performance degradation and channel interference.

[0069] Simultaneously, processing fitness is calculated. Based on the computing power parameters of multiple first edge nodes and the number of test channels in multiple task characteristics, the single-channel allocated computing power for each edge node is calculated. The single-channel allocated computing power is obtained by dividing the computing power configuration coefficient of each edge node by the total number of channels of the test tasks it undertakes. Based on the calculated single-channel allocated computing power values, the variance of the multiple single-channel allocated computing power values ​​is further calculated, and the reciprocal of this variance is used as the first processing fitness. Through this calculation method, the closer the single-channel allocated computing power of each edge node is, the smaller the variance value, and the larger its reciprocal, i.e., the first processing fitness, is, indicating that the distribution of computing power resources among different test channels is more balanced, and the system load balance is better.

[0070] Finally, test fitness is synthesized. The first compensation fitness and the first processing fitness are linearly weighted according to preset importance weights to obtain the first test fitness. This first test fitness index simultaneously considers the system's ability to compensate for test errors and the degree of balanced allocation of computing resources, comprehensively and objectively reflecting the overall performance level of the first test configuration scheme in terms of both accuracy assurance and resource utilization, providing an accurate quantitative evaluation basis for subsequent optimization iterations. Specifically, the first test fitness = first compensation fitness × weight coefficient γ + first processing fitness × weight coefficient δ. The weight coefficients are set according to the data accuracy requirements of the test task. For example, for test tasks with high accuracy requirements, γ is set to 0.7 and δ is set to 0.3; for test tasks with high throughput requirements, a ratio of γ of 0.4 and δ of 0.6 is used.

[0071] Further, the process enters the iterative optimization phase, continuously generating new test configuration schemes and calculating the corresponding test fitness. When the change in test fitness value during a preset number of optimizations is less than a fitness change threshold, it is determined that the system has entered a local optimum. A local optimum refers to a situation where the optimization process stagnates near a suboptimal solution, and a better configuration scheme cannot be found through a local search strategy. At this point, it is necessary to dynamically determine the number of exits based on the current task characteristics. Specifically, the more test channels there are and the higher the task complexity, the larger the number of exits should be set to enhance the global search capability. Multiple exit test configuration schemes different from historical schemes will be randomly generated to continue the optimization process.

[0072] Specifically, iterative optimization continues. When stuck in a local optimum, the number of exits is configured based on multiple task characteristics, generating multiple exit test configuration schemes for optimization, including:

[0073] During the optimization process, it is determined whether the change in fitness value during the preset number of optimizations is less than the fitness change threshold. If not, the optimization continues; if so, the system gets stuck in a local optimum.

[0074] When trapped in a local optimum, the number of configuration exits is calculated based on the aforementioned multiple task characteristics;

[0075] Randomly generate test configuration schemes with different numbers of bounces than those in previous optimization processes, and use these as multiple bounce test configuration schemes for further optimization.

[0076] During the optimization process, the changes in test fitness are continuously monitored within a preset number of iterations. The relative change in test fitness values ​​during the preset number of optimization iterations is calculated and compared with a preset fitness change threshold. If the change is greater than or equal to the fitness change threshold, it indicates that the optimization process is still progressing effectively, and the regular optimization steps continue. If the change remains less than the fitness change threshold, the current optimization process is determined to have fallen into a local optimum. Specifically, the preset number of iterations is set according to the complexity of the optimization problem; for example, 20 iterations are set for a medium-complexity scenario with more than ten test tasks. The fitness change threshold is set according to the test accuracy requirements; for example, a relative change threshold of 0.5% is set for a test scenario requiring an accuracy of 95% or higher.

[0077] When a local optimum is identified, the required number of bounces is calculated based on the task characteristics of all current test tasks. The number of bounces refers to the number of test configuration schemes regenerated during the optimization process to escape the local optimum. The specific calculation process uses the arithmetic mean of multiple task characteristics as a basis, rounding off this mean to obtain the final number of bounces. Here, the task characteristic is the number of test channels; a larger number of test channels results in a correspondingly larger number of bounces, ensuring that a sufficient number of new configuration schemes can be generated when facing complex test tasks with a large number of channels. This effectively expands the search space, thereby improving the global search capability of the optimization process and the accuracy of the final solution.

[0078] Furthermore, based on the calculated number of bounces, a corresponding number of new test configuration schemes are randomly generated as bounce test configuration schemes. These bounce test configuration schemes are significantly different from all schemes that appeared in the historical optimization process. By introducing new task allocation combinations and computing power configuration methods, the limitations of local optima are effectively broken. The newly generated bounce test configuration schemes will be incorporated into the subsequent optimization process to continue to advance the global optimization search until the optimal test configuration scheme that meets the requirements is obtained.

[0079] The optimization process continues until the convergence condition is met, meaning the fitness value no longer significantly improves within a preset number of iterations. The preset number of iterations is set based on system computing resources and real-time requirements; for example, it's set to 100 iterations in scenarios with sufficient computing resources but no real-time requirements, and to 30 iterations in scenarios with limited resources and a need for rapid response. At this point, the configuration scheme with the highest fitness in historical iterations is selected as the optimal test configuration. Based on this optimal test configuration, test tasks are assigned, matrix switches are configured, and edge node computing power is scheduled. Actual test operations are executed, and multiple sets of raw test data are collected.

[0080] In summary, this optimization process achieves collaborative optimization of test resources and computing resources through intelligent optimization algorithms. It considers the impact of error characteristics on test quality and makes full use of the compensation capabilities of edge computing resources to ensure the acquisition of high-quality raw test data.

[0081] The data compensation module 14 is used to perform test data compensation processing within multiple edge nodes by utilizing the computing power of multiple optimal edge nodes, and obtain multiple compensated test data as the data processing result.

[0082] After obtaining the raw test data, further distributed intelligent compensation processing is required because the raw test data contains test errors introduced by the performance degradation of the matrix switch and coupling errors generated by multi-channel parallel testing. This is to eliminate system errors and improve the accuracy and reliability of the test data, thereby obtaining optimized data processing results.

[0083] Specifically, within multiple edge nodes, the computing power of multiple optimal edge nodes is configured to perform test data compensation processing, obtaining multiple compensated test data sets as the data processing results, including:

[0084] Obtain a preset number, and calculate and determine the number of multiple intelligent agents based on the computing power of multiple optimal edge nodes and the preset number;

[0085] Based on the number of agents, within multiple edge nodes, the corresponding number of compensation agents are called. Multiple original test data, multiple test error parameters, and multiple coupling error parameters are input respectively, and multiple compensation test data are output as the data processing result.

[0086] Each edge node is configured with a preset number of compensation agents, and each compensation agent is trained using different sets of original sample test data, sample test error parameters, sample coupling error parameters, and sample compensation test data.

[0087] First, a preset number is obtained, which refers to the baseline number of agents configured in advance by the system. Combining the optimal computing power allocated to each edge node, the number of agents that need to be activated for each edge node is calculated. Number of agents = Preset number × (Node computing power configuration coefficient / Total system computing power configuration coefficient). Nodes with higher computing power will be allocated more agents proportionally, while nodes with lower computing power will be allocated fewer agents, thus achieving a precise match between computing resources and processing needs.

[0088] Secondly, based on the determined number of agents, a corresponding number of compensation agents are invoked within each edge node. These compensation agents are specially trained intelligent models used to correct errors and optimize accuracy on the test data. Each invoked compensation agent receives three sets of input data: the original test data, test error parameters, and coupling error parameters. After processing by the compensation agent's internal data processing algorithm, the error-compensated test data is calculated and output, constituting the final data processing result.

[0089] It should be noted that each edge node pre-deploys a predetermined number of compensation agents. The predetermined number is set based on the edge node's hardware storage capacity and maximum concurrent processing requirements; for example, 10 agents are set for a node with 1TB of storage, and 20 agents are set for a node with 4TB of storage. Each compensation agent is independently trained using different training datasets, including diverse sets of original test data, sets of test error parameters, sets of coupling error parameters, and corresponding sets of compensation test data. This differentiated training method ensures that the agent group can cover multiple error patterns and compensation scenarios, thereby improving the accuracy and robustness of the compensation processing.

[0090] The final data processing result is the calibrated matrix switch test data, which specifically includes high-precision measurement values ​​that have eliminated systematic errors introduced by switch performance degradation and channel coupling. These values ​​can be directly used for final test result determination and equipment performance evaluation, thus improving the quality and reliability of the test task.

[0091] In summary, the embodiments of this application have at least the following technical effects:

[0092] Compared to existing technologies, this application firstly achieves systematic quantification of test errors and coupling errors by constructing a collaborative analysis mechanism of switch features and task features, laying a data foundation for accurate error compensation; secondly, it innovatively combines error prediction with resource allocation, completing the joint optimization of error source tracing and resource matching during the test configuration phase, improving the system's resource utilization efficiency; thirdly, through a distributed intelligent agent compensation architecture, it pushes computational tasks down to edge nodes, effectively reducing data transmission latency and enhancing system real-time performance; finally, it adopts differentiated intelligent agent training strategies to ensure that the compensation model can cover various complex working conditions, improving the accuracy of test data and the system's adaptability in different scenarios.

[0093] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0094] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0095] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A matrix switch test data processing system based on edge computing, characterized in that, The system includes: The feature acquisition module is used to acquire the switch features of multiple matrix switches to be tested, and to acquire the task features of multiple test tasks to be performed, wherein each task feature includes the number of test channels, and the multiple matrix switches are connected to the test station through edge nodes. The error analysis module is used to perform test error analysis and coupling error analysis based on multiple switch features and multiple task features, respectively, to obtain multiple test error parameters and multiple coupling error parameters; The configuration optimization module is used to allocate test tasks and optimize edge node computing power based on the test error parameters and coupling error parameters, obtain the optimal test configuration scheme, perform test configuration, and obtain multiple raw test data, wherein auxiliary optimization strategies are configured based on multiple task characteristics; The data compensation module is used to perform test data compensation processing within multiple edge nodes by utilizing the computing power of multiple optimal edge nodes, thereby obtaining multiple compensated test data as the data processing result.

2. The matrix switch test data processing system based on edge computing according to claim 1, characterized in that, The execution steps of the feature acquisition module include: Obtain the switching characteristics of multiple matrix switches to be tested, whereby the switching characteristics include the cumulative number of tests on the matrix switches; Obtain the task characteristics of multiple test tasks to be performed, where each task characteristic includes the number of test channels for the test task.

3. The matrix switch test data processing system based on edge computing according to claim 1, characterized in that, The execution steps of the error analysis module include: Obtain the error predictor configured in the test bench during training, wherein the error predictor includes a switching error prediction branch and a coupled error prediction branch; The multiple switching features and multiple task features are respectively input into the switching error prediction branch and the coupling error prediction branch in the error predictor, and multiple test error parameters and multiple coupling error parameters are output.

4. The matrix switch test data processing system based on edge computing according to claim 3, characterized in that, The execution steps of the error analysis module also include: Based on historical test and verification data of matrix switches, sample switch feature sets and sample task feature sets are collected. The test obtains the test data error introduced under different sample switching characteristics, and obtains the sample test error parameter set; and obtains the coupling data error introduced under different sample task characteristics, and obtains the sample coupling error parameter set. Based on machine learning, a switching error prediction branch and a coupling error prediction branch are constructed. Using the sample switch feature set and sample test error parameter set, the switch error prediction branch is trained under supervision. Using the sample task feature set and sample coupling error parameter set, the coupling error prediction branch is trained under supervision. Convergence is tested separately to obtain the error predictor, which is then configured in the test center.

5. The matrix switch test data processing system based on edge computing according to claim 1, characterized in that, The execution steps of the configuration optimization module include: Obtain a computing power resource pool of multiple edge nodes, randomly configure the computing power of multiple first edge nodes, wherein the computing power of each edge node includes a computing power configuration coefficient, and the sum of the computing power of multiple edge nodes satisfies the total computing power constraint. Multiple test tasks are randomly combined with multiple matrix switches, and the computing power of the multiple first edge nodes is combined to obtain a first test configuration scheme. Based on the multiple test error parameters, multiple coupling error parameters, and multiple task characteristics, the first test fitness of the first test configuration scheme is calculated and obtained. Continue iterative optimization. When stuck in a local optimum, configure the number of exits based on multiple task characteristics, generate multiple exit test configuration schemes, and optimize them. The optimization continues until convergence, obtaining the optimal test configuration scheme with the greatest test fitness. Test configuration is then performed, and multiple raw test data are obtained.

6. The matrix switch test data processing system based on edge computing according to claim 1, characterized in that, The execution steps of the configuration optimization module also include: Based on multiple test error parameters and multiple coupling error parameters, multiple fusion error parameters of multiple matrix switches within the first test configuration scheme are calculated. The ratios of computing power of multiple first edge nodes and multiple fusion error parameters are calculated respectively, and the first compensation fitness is obtained by processing them. The first processing fitness is calculated based on the computing power of multiple first edge nodes and the characteristics of multiple tasks. The first test fitness is calculated based on the first compensation fitness and the first processing fitness.

7. The matrix switch test data processing system based on edge computing according to claim 1, characterized in that, The execution steps of the configuration optimization module also include: During the optimization process, it is determined whether the change in fitness value during the preset number of optimizations is less than the fitness change threshold. If not, the optimization continues; if so, the system gets stuck in a local optimum. When trapped in a local optimum, the number of configuration exits is calculated based on the aforementioned multiple task characteristics; Randomly generate test configuration schemes with different numbers of bounces than those in previous optimization processes, and use these as multiple bounce test configuration schemes for further optimization.

8. The matrix switch test data processing system based on edge computing according to claim 1, characterized in that, The execution steps of the data compensation module include: Obtain a preset number, and calculate and determine the number of multiple intelligent agents based on the computing power of multiple optimal edge nodes and the preset number; Based on the number of agents, within multiple edge nodes, the corresponding number of compensation agents are called. Multiple original test data, multiple test error parameters, and multiple coupling error parameters are input respectively, and multiple compensation test data are output as the data processing result. Each edge node is configured with a preset number of compensation agents, and each compensation agent is trained using different sets of original sample test data, sample test error parameters, sample coupling error parameters, and sample compensation test data.

Images