Test parameter determination method and related device

Abstract

Embodiments of the present application disclose a test parameter determination method and related equipment; the method comprises: determining a target test parameter type corresponding to a test requirement of a user; initializing an agent model, a security evaluation model and a multi-fidelity strategy model by using global initialization parameters; performing a first operation iteration to update the agent model and the security evaluation model until a stop condition is met, and determining a current candidate test parameter as an optimal test parameter; wherein the first operation comprises: generating a candidate test parameter in a search range corresponding to the target test parameter type, inputting the candidate test parameter filtered by a security constraint of the security evaluation model into the agent model for performance prediction, performing actual measurement on the candidate test parameter by the multi-fidelity strategy model according to a performance prediction result, and updating the agent model and the security evaluation model according to an actual measurement result.

Description

Technical Field

[0001] This application relates to the field of electronic measuring instrument testing, specifically to a method for determining test parameters and related equipment. Background Technology

[0002] In related technologies, the testing process for electronic measuring instruments heavily relies on manual operation. Testers manually set the test parameters of the electronic measuring instruments (such as the center frequency of the spectrum analyzer and the output power of the signal source), record data, and analyze it. Specifically, the selection of optimal test parameters depends on manual experience or fixed rules, making it difficult to quickly find the optimal test parameters, resulting in low testing efficiency. Summary of the Invention

[0003] To address the technical problems existing in the above-mentioned technologies, this application provides a method for determining test parameters and related equipment.

[0004] In a first aspect, embodiments of this application provide a method for determining test parameters, the method comprising:

[0005] Determine the target test parameter types corresponding to the user's testing requirements; Initialize the proxy model, security assessment model, and multi-fidelity strategy model using global initialization parameters; The first operation iteratively updates the proxy model and the security assessment model until the stopping condition is met, at which point the current candidate test parameters are determined to be the optimal test parameters. The first operation includes: generating candidate test parameters within the search range corresponding to the target test parameter type; inputting the candidate test parameters selected by the security constraints of the security assessment model into the proxy model for performance prediction; conducting actual tests on the candidate test parameters using a multi-fidelity strategy model based on the performance prediction results; and updating the proxy model and the security assessment model based on the actual test results.

[0006] In some embodiments, the proxy model, security assessment model, and multi-fidelity strategy model are initialized using global initialization parameters, including: Retrieve global initialization parameters from the meta-knowledge base; The proxy model, security assessment model, and multi-fidelity strategy model are initialized using the initial proxy model parameters, initial security assessment model parameters, and initial multi-fidelity strategy model parameters in the global initialization parameters, respectively.

[0007] In some embodiments, global initialization parameters are determined in the following ways: Obtain the meta-training set; wherein the meta-training set contains at least one historical test task data; Using historical test task datasets, a model-independent meta-learning algorithm is employed to train the agent model, security assessment model, and multi-fidelity strategy model to obtain global initialization parameters.

[0008] In some embodiments, using a historical test task dataset, a model-independent meta-learning algorithm is employed to train the agent model, security assessment model, and multi-fidelity strategy model to obtain global initialization parameters, including: Iterate through the historical test task data in the meta-training set, and perform the second operation on the currently iterated historical test task data until the current meta-parameters converge, and obtain the global initialization parameters. The second operation includes: Initialize the proxy model, security assessment model, and multi-fidelity strategy model using the current meta-parameters; Using a portion of the historical test task data currently being traversed, the agent model, security assessment model, and multi-fidelity strategy model are trained using the gradient descent algorithm to obtain the task adaptation parameters of the agent model, security assessment model, and multi-fidelity strategy model. Calculate the loss value of the task adaptation parameters using the remaining data in the current traversal of historical test task data; Update the current meta-parameters by backpropagating based on the loss value.

[0009] In some embodiments, candidate test parameters are generated within the search range corresponding to the target test parameter type, including: A first candidate test parameter set is generated within the search range corresponding to the target test parameter type using a grid search strategy; The second candidate test parameter set, selected through the security constraints of the security assessment model, is input into the proxy model for performance prediction, and the predicted mean and uncertainty of the performance index corresponding to each second candidate test parameter are obtained. Each second candidate test parameter is scored using a data acquisition function, and candidate test parameters are selected from the second candidate test parameters based on the scoring results.

[0010] In some embodiments, the safety assessment model is a binary Gaussian process classifier; safety constraint screening includes: Input the candidate test parameters into the security assessment model to obtain the risk probability of the candidate test parameters; When the risk probability exceeds the risk probability threshold, the candidate test parameter is determined to have failed the safety constraint screening.

[0011] In some embodiments, the proxy model is a hybrid proxy model, which includes a physical model and a data-driven compensation model; wherein, the data-driven compensation model is constructed based on a sparse Gaussian process and is used to learn the residual between the predicted values ​​of the basic performance indicators output by the physical model and the actual performance indicators. The process of performance prediction using a proxy model is as follows: The candidate test parameters are input into the physical model and the data-driven compensation model, respectively. Using a physical model, the predicted values ​​of basic performance indicators are output based on the physical relationship between candidate test parameters and performance indicators; and using a data-driven compensation model, the residuals between the predicted values ​​of basic performance indicators and the actual performance indicators are output. The predicted values ​​of the basic performance indicators are superimposed with the residuals to obtain the performance prediction results; the performance prediction results include the predicted mean and uncertainty of the performance indicators corresponding to the candidate test parameters.

[0012] In some embodiments, the multi-fidelity policy model is built based on reinforcement learning and is used to determine the target evaluation accuracy level from multiple evaluation accuracy levels according to the current state, and to test the candidate test parameters using the target evaluation accuracy level. The current state of reinforcement learning includes the uncertainty in the performance prediction results output by the agent model, the consumed test budget, and the variance of the historical evaluation results of the multi-fidelity policy model. Multiple assessment accuracy levels are available, including low-fidelity accuracy and high-fidelity accuracy, with different accuracy levels corresponding to different testing costs; The reward function for reinforcement learning is defined as the negative of the weighted sum of test cost and uncertainty.

[0013] In some embodiments, determining the target test parameter type corresponding to the user's test requirements includes: Obtain users' testing requirements; Perform intent analysis on the test requirements to determine the target test tasks; Based on the target test task, match the corresponding target test template from the template library; Extract the target test parameter types and the search range corresponding to each target test parameter type from the target test template.

[0014] In some embodiments, matching a corresponding target test template from the template library according to the target test task includes: Perform a first-level matching based on the test subject in the target test task to determine the instrument control template corresponding to the target test task; Based on the secondary matching of the collaborating instruments in the target test task, determine the corresponding process control template for the target test task; Perform a three-level matching based on the test objectives in the target test task to determine the data collection template corresponding to the target test task; The instrument control template, process control template, and data acquisition template are combined to form the target test template.

[0015] In some embodiments, intent parsing is performed on the test requirements to determine the target test task, including: Semantic parsing is performed on test requirements to extract the test subject, test objective, test constraints, and output requirements. Generate an initial structured task description based on the test subject, test objective, test constraints, and output requirements; The initial structured task description is validated for completeness. If the initial structured task description is incomplete, guidance information is generated to obtain supplementary information from the user. The initial structured task description is then updated based on the information to obtain the target test task.

[0016] In some embodiments, the stopping condition is: The uncertainty in the performance prediction results output by the proxy model is lower than the preset uncertainty threshold; Or / and, the number of iterations for the first operation reaches the preset maximum number of iterations; Or / and, the performance prediction results corresponding to the candidate test parameters output in consecutive iterations no longer change.

[0017] In some embodiments, updating the proxy model and security assessment model based on actual test results includes: The candidate test parameters of the current iteration and the corresponding test results are used as training samples to fine-tune the proxy model and the security assessment model using gradient descent, respectively, in order to update the parameters of the proxy model and the security assessment model.

[0018] In some embodiments, after determining that the current candidate test parameter is the optimal test parameter, the method provided by the first aspect further includes: The optimal test parameters are used to assign values ​​to the corresponding parameters in the target test template, and instrument control commands are generated based on the assignment results. According to the instrument control instructions, control the corresponding instrument to perform the test on the test subject corresponding to the test requirements, and collect the test data returned by the test subject.

[0019] In some embodiments, after collecting the test data returned by the test subject, the method provided in the first aspect further includes: A test report is generated based on the test data.

[0020] Secondly, embodiments of this application provide a test parameter determination device, the device comprising: The target test parameter type determination module is used to determine the target test parameter type corresponding to the user's test requirements; The optimal test parameter determination module is used to initialize the proxy model, security assessment model, and multi-fidelity strategy model using global initialization parameters; it performs a first operation to iteratively update the proxy model and security assessment model until a stopping condition is met, and then determines the current candidate test parameters as the optimal test parameters; wherein, the first operation includes: generating candidate test parameters within the search range corresponding to the target test parameter type; inputting the candidate test parameters selected by the security constraints of the security assessment model into the proxy model for performance prediction; performing actual tests on the candidate test parameters through the multi-fidelity strategy model based on the performance prediction results; and updating the proxy model and security assessment model based on the actual test results.

[0021] Thirdly, embodiments of this application provide an electronic device, including: a processor and a memory for storing a computer program capable of running on the processor, wherein, when the processor runs the computer program, it performs the method described in any embodiment of the first aspect.

[0022] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the methods described in any embodiment of the first aspect.

[0023] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, causes the processor to perform the method described in any embodiment of the first aspect.

[0024] The test parameter determination method provided in this application includes: determining the target test parameter type corresponding to the user's test requirements; Initialize the proxy model, security assessment model, and multi-fidelity strategy model using global initialization parameters; The first operation iteratively updates the proxy model and the security assessment model until the stopping condition is met, at which point the current candidate test parameters are determined to be the optimal test parameters. The first operation includes: generating candidate test parameters within the search range corresponding to the target test parameter type; inputting the candidate test parameters selected by the security constraints of the security assessment model into the proxy model for performance prediction; conducting actual tests on the candidate test parameters using a multi-fidelity strategy model based on the performance prediction results; and updating the proxy model and the security assessment model based on the actual test results.

[0025] The test parameter determination method provided in this application initializes the proxy model, security assessment model, and multi-fidelity strategy model through global initialization parameters. In new test tasks, the initial search direction can be determined without relying on human experience, thus solving the problem of dependence on human experience. Through the closed-loop iterative process of generating candidate test parameters within the search range in the first operation and sequentially filtering them through security constraints, predicting the performance of the proxy model, and testing the multi-fidelity strategy model, the model-driven automatic optimization replaces the manual traversal or fixed rule parameter selection method, reducing the number of invalid tests. By repeatedly executing the first operation to iteratively update the proxy model and security assessment model until the stopping condition is met, the method automatically converges to the optimal test parameters, avoiding the time-consuming process of repeatedly adjusting parameters manually, thereby improving test efficiency.

[0026] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are merely embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort, and this application can be applied to other similar scenarios based on the provided drawings.

[0028] Figure 1 A flowchart illustrating a method for determining test parameters provided in an embodiment of this application; Figure 2 A flowchart illustrating the second operation provided in an embodiment of this application; Figure 3 A flowchart illustrating the method for generating candidate test parameters provided in an embodiment of this application; Figure 4 A flowchart illustrating the method for determining the type of target test parameters provided in an embodiment of this application; Figure 5 A flowchart illustrating the method for determining a target test template provided in an embodiment of this application; Figure 6 This is a schematic diagram of a test parameter determination device provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0029] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. The described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0030] It should be noted that the terms "system," "device," "unit," and / or "module" used in this application are methods of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they can be replaced by other expressions.

[0031] Hereinafter, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first," "second," or "third" may explicitly or implicitly include one or more of that feature.

[0032] In related technologies, the testing process for electronic measuring instruments heavily relies on manual operation. Testers manually set the test parameters (such as the center frequency of the spectrum analyzer and the output power of the signal source), record data, and analyze it. With the development of technologies such as 5G / 6G, the Internet of Things, and high-speed interconnection, the complexity of the electronic measuring instruments under test is increasing exponentially, the number of test parameters is growing, and the testing process is becoming increasingly cumbersome. Therefore, the following problems are becoming increasingly prominent: First, it is highly dependent on human experience: the configuration of test parameters, result analysis, and troubleshooting rely heavily on the experience and judgment of senior engineers, which is difficult for novices to handle, resulting in high personnel costs and low testing efficiency.

[0033] Second, the level of automation is shallow: Although existing automated testing methods have achieved the encapsulation of the instrument under test and the execution of scripts, they only execute tests according to preset scripts. They cannot dynamically adjust the test parameter search strategy according to the test results, nor can they autonomously find the optimal solution for test parameters.

[0034] Third, insufficient generalization: Different batches of the same type of testing instrument have different performance. When the testing instrument is changed, the test parameters need to be recalibrated and the optimal test parameters need to be selected again to obtain better test performance.

[0035] Although existing technologies attempt to introduce Natural Language Processing (NLP) techniques to generate test scripts, their essence is still to transform one static description (user input) into another static description (script), without addressing the issue of automatically finding the optimal solution for test parameters.

[0036] To address at least one of the aforementioned technical problems, embodiments of this application provide a method for determining test parameters and related equipment.

[0037] Figure 1 This is a flowchart illustrating a method for determining test parameters provided in an embodiment of this application.

[0038] like Figure 1 As shown, the first method for determining test parameters provided in this application embodiment can be executed by an electronic device. Electronic devices include, but are not limited to, devices with data processing capabilities such as servers, personal computers, laptops, tablets, and test instrument control terminals. The method for determining test parameters includes the following steps: Step 101: Determine the target test parameter type corresponding to the user's test requirements; In some embodiments, test requirements refer to a description of the test tasks that the user expects to perform on the instrument under test. Test requirements may specifically include information such as the test subject, test objective, and test constraints. The test subject refers to the instrument under test, such as a signal generator, spectrum analyzer, vector network analyzer, or other electronic measuring instruments. The test objective refers to the performance indicators that the user expects to measure, such as output power stability, spurious radiation, and frequency response. Test constraints refer to the limiting conditions that need to be met during the test, such as frequency band range (e.g., 2.4 GHz band) and accuracy indicators (e.g., power fluctuation ≤ ±0.5 dB).

[0039] In some embodiments, the target test parameter type refers to the type (name) of the adjustable parameters of the instrument that need to be set when performing a test task, such as the center frequency, resolution bandwidth (RBW), scan start frequency, scan end frequency, number of scan points, and averaging count in spectrum analyzer testing, or the output power and modulation method in signal source testing. The search range corresponding to the target test parameter type refers to the range of values ​​that each target test parameter type can take during the optimization process, for example, the search range for RBW is 1kHz to 1MHz.

[0040] In some embodiments, users can input test requirements in the form of natural language text or voice. The electronic device interprets the test requirements to determine the type of target test parameters.

[0041] In some implementations, the test requirements include target test parameter types explicitly specified by the user. For example, if the user inputs "optimize the resolution bandwidth and scan points of the spectrum analyzer in the 2.4GHz band to quickly complete spurious emissions testing," the electronic device directly extracts "resolution bandwidth" and "scan points" as target test parameter types through intent parsing, and determines the search range for each target test parameter type based on the frequency band information (2.4GHz) and the instrument's physical range.

[0042] In some implementations, the target test parameter type is not explicitly specified in the test requirements. For example, the user may only input "perform an output stability test on the signal source". The electronic device parses the test requirements, extracting the test subject, test objective, and test constraints, and then matches the corresponding target test template from the template library based on this information. The template library is a pre-built database storing multiple test templates. Each test template corresponds to a test scenario and includes the type definitions of the test parameters required to execute that test scenario, as well as the corresponding test operation procedures. For example, the "Signal Source Output Stability Test Template" includes type definitions for test parameters such as center frequency, output power, and modulation method. The electronic device extracts the type of test parameter from the matched target test template as the target test parameter type and determines the search range corresponding to each target test parameter type based on the instrument's physical range.

[0043] Step 102: Initialize the proxy model, security assessment model, and multi-fidelity strategy model using global initialization parameters; In some embodiments, the agent model, security assessment model, and multi-fidelity strategy model constitute a meta-learning-enhanced hybrid modeling combined multi-fidelity optimization (Meta-HCMFO) framework. The Meta-HCMFO framework continuously perceives and analyzes the current test state by performing closed-loop operations of hypothesis generation, verification, result analysis, and strategy adjustment, dynamically generating test parameter adjustment strategies, and continuously correcting the optimization direction using a feedback mechanism. Its core objective is to efficiently converge to the optimal test parameters in the testing process with minimal expected time and economic cost under multiple performance constraints and absolute security constraints.

[0044] In some embodiments, global initialization parameters refer to a set of parameters obtained through offline meta-training for initializing the agent model, security assessment model, and multi-fidelity strategy model. These global initialization parameters have generalization capabilities across testing tasks, enabling the agent model, security assessment model, and multi-fidelity strategy model to learn from scratch in new testing tasks. The global initialization parameters are pre-stored in a meta-knowledge base. The electronic device retrieves the global initialization parameters from the meta-knowledge base and initializes the agent model, security assessment model, and multi-fidelity strategy model using the sub-parameters corresponding to these parameters.

[0045] The surrogate model is used to predict the performance of candidate test parameters.

[0046] Specifically, the surrogate model takes candidate test parameters as input and outputs the predicted mean and uncertainty of the performance metric corresponding to each candidate test parameter. The predicted mean quantifies the expected performance level of the candidate test parameter, while the uncertainty quantifies the confidence level of the prediction. The surrogate model acquires reasonable initial predictive power during initialization, enabling it to provide performance predictions with some reference value even when there is no or very little experimental data.

[0047] The safety assessment model is used to screen candidate test parameters for safety constraints.

[0048] Specifically, the safety assessment model takes candidate test parameters as input and outputs the probability of those parameters causing equipment damage or malfunction. Upon initialization, the model acquires risk discrimination capabilities based on historical experience, enabling it to identify potential risks in candidate test parameters early in a new task and eliminate high-risk candidates before actual testing.

[0049] The multi-fidelity strategy model is used to determine the evaluation accuracy level to be used when measuring candidate test parameters based on the current state, and to measure candidate test parameters according to the corresponding evaluation accuracy level.

[0050] Specifically, the input to the multi-fidelity strategy model is the current state, which includes the uncertainty in the performance prediction results output by the surrogate model, the consumed test budget, and the variance of the historical evaluation results of the multi-fidelity strategy model. The output is the target evaluation accuracy level, which is selected from multiple evaluation accuracy levels, including low-fidelity accuracy and high-fidelity accuracy, with different accuracy levels corresponding to different test costs. The multi-fidelity strategy model acquires near-optimal resource allocation decision-making capabilities during initialization, enabling it to make reasonable choices among different accuracy levels of testing methods in the early stages of a new task, thereby minimizing the number of high-cost tests while ensuring prediction accuracy.

[0051] Step 103: Perform the first operation to iteratively update the proxy model and security assessment model until the stopping condition is met, and then determine the current candidate test parameters as the optimal test parameters. The first operation includes: generating candidate test parameters within the search range corresponding to the target test parameter type; inputting the candidate test parameters selected by the security constraints of the security assessment model into the proxy model for performance prediction; conducting actual tests on the candidate test parameters through a multi-fidelity strategy model based on the performance prediction results; and updating the proxy model and the security assessment model based on the actual test results.

[0052] Specifically, the electronic device generates candidate test parameters within the search range corresponding to the target test parameter type. A candidate test parameter refers to a set of parameter values ​​obtained by assigning specific values ​​to each target test parameter type within the search range. For example, if the target test parameter types are resolution bandwidth and scan point count, and the search ranges are 1kHz~1MHz and 101~10001 points respectively, then a candidate test parameter could be "resolution bandwidth = 10kHz, scan point count = 601 points".

[0053] In some embodiments, the electronic device can generate candidate test parameters within a search range using a grid search strategy. The grid search strategy discretizes the search space into grid points, sequentially traversing these points to generate a set of candidate test parameters, ensuring complete coverage of the search range. For example, the search range of resolution bandwidth (1kHz to 1MHz) can be discretized into several discrete values ​​using logarithmic coordinates, and the search range of scan points (101 to 10001) can be discretized into several discrete values ​​using an arithmetic sequence, and then combined to form a set of candidate test parameters.

[0054] The electronic device inputs candidate test parameters into the safety assessment model. The safety assessment model evaluates whether the candidate test parameters will cause equipment damage or malfunction. The input to the safety assessment model is the candidate test parameter, and the output is the probability of that candidate test parameter causing equipment damage or malfunction. When the probability of risk exceeds a preset risk probability threshold, the electronic device determines that the candidate test parameter has not passed the safety constraint screening and removes it from the subsequent process, without performing the actual test. When the probability of risk does not exceed the risk probability threshold, the electronic device determines that the candidate test parameter has passed the safety constraint screening and allows it to enter the subsequent proxy model prediction stage.

[0055] In some embodiments, when a candidate test parameter fails the security constraint screening, the electronic device also feeds the candidate test parameter back to the security assessment model as a high-risk sample to correct subsequent security constraint screening, so that the security assessment model can more accurately identify similar high-risk candidate test parameters in subsequent iterations.

[0056] Electronic devices input candidate test parameters, selected through security constraints, into a surrogate model for performance prediction. The surrogate model takes the candidate test parameters as input and outputs the predicted mean and uncertainty of the corresponding performance metrics. The predicted mean quantifies the expected performance level of the candidate test parameter, such as the predicted signal-to-noise ratio, detection probability, and test latency. The uncertainty quantifies the confidence level of the prediction; higher uncertainty indicates a less comprehensive understanding of the surrogate model in that region, making further exploration more worthwhile. By predicting performance, the surrogate model can evaluate the quality of candidate test parameters without performing real tests, avoiding a large amount of invalid experimental data.

[0057] The electronic device inputs the performance prediction results output by the surrogate model into the multi-fidelity strategy model. The multi-fidelity strategy model determines the evaluation accuracy level to be used when conducting actual measurements on candidate test parameters based on the current state. The input to the multi-fidelity strategy model is the current state, which includes the uncertainty in the performance prediction results output by the surrogate model, the consumed test budget, and the variance of the historical evaluation results of the multi-fidelity strategy model. The output of the multi-fidelity strategy model is the target evaluation accuracy level. The target evaluation accuracy level is selected from multiple evaluation accuracy levels, including low-fidelity accuracy and high-fidelity accuracy, with different accuracy levels corresponding to different test costs. Low-fidelity accuracy uses low-cost configurations for rapid coarse screening, such as setting a wider resolution bandwidth, fewer sampling points, or using faster auxiliary instruments; high-fidelity accuracy uses high-cost configurations for precise measurement, such as setting a narrower resolution bandwidth, a complete frequency sweep range, or using full-specification precision instruments. The multi-fidelity strategy model selects the evaluation accuracy level with the highest cost-effectiveness under the current state, minimizing the number of high-cost actual measurements while ensuring prediction accuracy.

[0058] The electronic device performs actual measurements on candidate test parameters according to the target evaluation accuracy level output by the multi-fidelity strategy model. Actual measurement refers to driving a real electronic measuring instrument to perform tests according to the candidate test parameters to obtain the actual performance data corresponding to those parameters. For example, when the target evaluation accuracy level is low-fidelity, the electronic device drives the instrument to perform rapid testing with a wider resolution bandwidth and fewer sampling points to obtain the actual measurement results; when the target evaluation accuracy level is high-fidelity, the electronic device drives the instrument to perform precise testing with a narrower resolution bandwidth and a complete frequency sweep range to obtain the actual measurement results.

[0059] Electronic devices update their proxy model and security assessment model based on actual test results. The update process involves using the candidate test parameters of the current iteration and the corresponding test results as training samples to fine-tune the parameters of the proxy model and security assessment model. This makes the proxy model's performance predictions more accurate, and the security assessment model's risk identification more precise. After the update, the prediction uncertainty of the proxy model in the explored area is reduced, and the security boundary of the security assessment model is clearer, thus enabling more accurate recommendations of high-performance and safe candidate test parameters in subsequent iterations.

[0060] The electronic device repeats the first operation described above until a stopping condition is met. The stopping condition includes, but is not limited to: the uncertainty in the performance prediction result output by the surrogate model is below a preset uncertainty threshold, indicating that the optimization has converged to the optimal region; the number of iterations of the first operation reaches a preset upper limit for the number of iterations; and the performance prediction results corresponding to the candidate test parameters output in consecutive iterations no longer change, indicating that it is impossible to obtain better test parameters through further iterations. When the stopping condition is met, the electronic device determines the current candidate test parameter as the optimal test parameter.

[0061] In some embodiments, during the iteration process, the multi-fidelity strategy model tends to select a low-fidelity accuracy level for rapid exploration in the early stages of the iteration. As the uncertainty of the surrogate model gradually decreases and the number of iterations increases, the multi-fidelity strategy model adaptively increases the selection ratio of the high-fidelity accuracy level according to the change of the current state, so as to improve the accuracy of the final output optimal test parameters.

[0062] In summary, the test parameter determination method provided in this application initializes the proxy model, security assessment model, and multi-fidelity strategy model with global initialization parameters. This allows for the determination of the initial search direction in new test tasks without relying on human experience, thus solving the problem of dependence on human experience. The first operation generates candidate test parameters within the search range and sequentially filters them through security constraints, predicts proxy model performance, and conducts practical tests using the multi-fidelity strategy model. This closed-loop iterative process replaces manual traversal or fixed-rule parameter selection with model-driven automatic optimization, reducing the number of invalid tests. By repeatedly executing the first operation to iteratively update the proxy model and security assessment model until the stopping condition is met, the method automatically converges to the optimal test parameters, avoiding the time-consuming process of repeatedly adjusting parameters manually, thereby improving testing efficiency.

[0063] In some embodiments, based on the description of step 102, step 102 specifically includes the following steps: Retrieve global initialization parameters from the meta-knowledge base; The proxy model, security assessment model, and multi-fidelity strategy model are initialized using the initial proxy model parameters, initial security assessment model parameters, and initial multi-fidelity strategy model parameters in the global initialization parameters, respectively.

[0064] In some embodiments, the meta-knowledge base is a knowledge storage system built on a graph-structured database, used to store task descriptors, parameter optimization trajectory sequences, final optimal solutions, and globally initialized parameters obtained through offline meta-training for historical optimization tasks. The task descriptor characterizes the basic attributes and contextual features of each test task, including the test objective, device type, signal characteristics, constraints, and environmental state. The parameter optimization trajectory sequence records the candidate test parameters proposed in each iteration of the test parameter optimization process and the corresponding measured results, reflecting the search path and convergence process. The final optimal solution refers to the optimal test parameters and their corresponding performance indicators obtained under the conditions of satisfying safety and performance constraints.

[0065] Global initialization parameters are a set of parameters obtained through offline meta-training, used to initialize the agent model, security assessment model, and multi-fidelity strategy model. They include three subsets: initial agent model parameters, initial security assessment model parameters, and initial multi-fidelity strategy model parameters. The global initialization parameters stored in the meta-knowledge base remain unchanged until updated during offline meta-training. Regardless of the specific content of the current test task, the electronic device loads the same set of global initialization parameters from the meta-knowledge base.

[0066] The initialization of the three models is performed independently. The electronic device assigns the initial proxy model parameters to the proxy model, the initial security assessment model parameters to the security assessment model, and the initial multi-fidelity strategy model parameters to the multi-fidelity strategy model, respectively. After each model obtains the initial parameters corresponding to its own function, it possesses general prior knowledge across tasks.

[0067] In some embodiments, global initialization parameters are determined in the following way: Obtain the meta-training set; wherein the meta-training set contains at least one historical test task data; Using historical test task datasets, a model-independent meta-learning algorithm is employed to train the agent model, security assessment model, and multi-fidelity strategy model to obtain global initialization parameters.

[0068] In some embodiments, the global initialization parameters are pre-trained in an offline meta-training phase, which is executed independently of the current test task.

[0069] In some embodiments, the Model-Agnostic Meta-Learning (MAML) algorithm has a two-layer optimization structure, consisting of an inner loop and an outer loop. By continuously comparing and optimizing across multiple tasks, the MAML algorithm extracts initialization parameters with generalization capabilities, enabling the model to quickly fine-tune and achieve better performance when facing new tasks with only a small amount of evaluation data.

[0070] Specifically, using historical test task datasets, a model-independent meta-learning algorithm was employed to train the agent model, security assessment model, and multi-fidelity strategy model, obtaining globally initialized parameters, including: Traverse the historical test task data in the meta-training set, and perform the second operation on the currently traversed historical test task data until the current meta-parameters converge, thus obtaining the global initialization parameters.

[0071] Specifically, Figure 2 This is a flowchart illustrating the second operation provided in an embodiment of this application.

[0072] like Figure 2 As shown, the second operation includes the following steps: Step 201: Initialize the proxy model, security assessment model, and multi-fidelity strategy model using the current meta-parameters; In some embodiments, the current meta-parameters refer to the meta-parameters that are iteratively updated during offline meta-training at the current moment. The initial values ​​of the meta-parameters can be set randomly or based on experience. The electronic device assigns the current meta-parameters to the proxy model, the security assessment model, and the multi-fidelity strategy model, respectively, so that each model obtains its initial model parameters in the current state.

[0073] Step 202: Using a portion of the historical test task data currently being traversed, train the proxy model, security assessment model, and multi-fidelity strategy model using the gradient descent algorithm to obtain the task adaptation parameters of the proxy model, security assessment model, and multi-fidelity strategy model. In some embodiments, the partial data refers to the initial small amount of measurement data obtained from the historical test task data being traversed. The electronic device uses only this partial data to fine-tune the three models initialized in step 201 using the gradient descent algorithm. Since this step belongs to the inner loop stage of the MAML algorithm, its purpose is to allow the model to quickly adapt to the characteristics of the current historical task and obtain the task adaptation parameters under that historical task.

[0074] Step 203: Calculate the loss value of the task adaptation parameters using the remaining data in the current traversal of historical test task data; In some embodiments, the remaining data refers to the measurement data in the historical test task data being traversed, excluding the data used in step 202. The electronic device applies the task adaptation parameters obtained in step 202 to the remaining data, calculates the model's prediction error on the remaining data, and obtains a loss value. This loss value reflects the actual performance of the model after adapting to the current task.

[0075] Step 204: Update the current meta-parameters by backpropagating based on the loss value.

[0076] In some embodiments, the electronic device updates the current meta-parameters based on the loss value calculated in step 203 via backpropagation of the second-order gradient. Step 204 belongs to the outer loop phase of the MAML algorithm. The updated meta-parameters will be used as new current meta-parameters in the training of the next historical task.

[0077] The electronic device iterates through each historical test task data in the meta-training set, and sequentially executes steps 201 to 204 above for each historical task, continuously repeating the process of task adaptation (step 203) and cross-task update (step 204) until the current meta-parameter converges. After the current meta-parameter converges, the electronic device determines the converged current meta-parameter as the global initialization parameter.

[0078] Figure 3 This is a flowchart illustrating the method for generating candidate test parameters provided in an embodiment of this application.

[0079] In some embodiments, such as Figure 3 As shown, step 103, generating candidate test parameters within the search range corresponding to the target test parameter type, includes: Step 301: Generate a first candidate test parameter set within the search range corresponding to the target test parameter type using a grid search strategy; In some embodiments, the grid search strategy is a method for generating candidate test parameter vectors by discretizing samples within a defined parameter space according to preset rules. For each target test parameter type, the electronic device discretizes its corresponding search range into multiple discrete value points according to a preset step size or preset interval. Then, the discrete value points of each target test parameter type are combined to form a candidate test parameter set covering the entire search space, i.e., the first candidate test parameter set. The first candidate test parameter set contains multiple candidate test parameter vectors, and each candidate test parameter vector corresponds to a specific combination of test parameter values.

[0080] Step 302: Input the second candidate test parameter set selected by the security constraints of the security assessment model into the proxy model for performance prediction, and obtain the predicted mean and uncertainty of the performance index corresponding to each second candidate test parameter; In some embodiments, the electronic device inputs each candidate test parameter vector from the first candidate test parameter set into the security assessment model for security constraint screening, eliminates candidate test parameter vectors that do not meet the security constraints, and uses the candidate test parameter vectors that pass the screening as the second candidate test parameters.

[0081] The electronic device inputs the second candidate test parameters into the surrogate model. The surrogate model performs performance prediction for each second candidate test parameter and outputs the predicted mean and uncertainty of the corresponding performance index. The predicted mean quantifies the expected performance level of the second candidate test parameter, and the uncertainty quantifies the confidence level of the prediction.

[0082] Step 303: Score each second candidate test parameter using the acquisition function, and select candidate test parameters from the second candidate test parameters based on the scoring results.

[0083] The electronic device inputs the predicted mean and uncertainty of each second candidate test parameter output by the surrogate model into a data acquisition function. The data acquisition function, considering factors such as the predicted mean and uncertainty, quantifies the value of each second candidate test parameter. The score reflects the potential benefit and exploratory significance of the second candidate test parameter. Based on the score results, the electronic device selects the second candidate test parameters with higher scores from the set of second candidate test parameters as the candidate test parameters for the current iteration.

[0084] In some embodiments, the safety assessment model is a binary Gaussian process classifier; safety constraint screening includes: Input the candidate test parameters into the security assessment model to obtain the risk probability of the candidate test parameters; When the risk probability exceeds the risk probability threshold, the candidate test parameter is determined to have failed the safety constraint screening.

[0085] In some embodiments, the binary Gaussian Process Classifier (GPC) is a probabilistic classification model based on Gaussian processes, used in real time during the optimization process to determine whether candidate test parameters may pose a risk to the device or instrument under test. The input to the safety assessment model is a vector of candidate test parameters. The output is the probability of the candidate test parameters causing equipment damage or malfunction. .

[0086] The training data for the safety assessment model mainly comes from two sources: first, samples marked as "dangerous" or "near-miss" in historical tasks, such as situations where a sudden increase in power causes the current or voltage to exceed the safety threshold; second, extreme operating condition samples constructed based on system boundary conditions, which are used to enhance the model's ability to distinguish in edge regions.

[0087] After the electronic device inputs the candidate test parameters into the security assessment model, the security assessment model outputs the risk probability corresponding to the candidate test parameters.

[0088] Electronic devices will compare the risk probability output by the security assessment model with a preset risk probability threshold. (like Compare them. When If the electronic device determines that the candidate test parameter has failed the safety constraint screening, it will not perform a real measurement. Simultaneously, the electronic device will feed this candidate test parameter back to the safety assessment model as a high-risk sample for subsequent model correction.

[0089] when At that time, the electronic device determines that the candidate test parameter has passed the safety constraint screening and allows it to enter the subsequent process.

[0090] In some embodiments, the initial parameters of the security assessment model are provided by the meta-learning phase, enabling it to have reasonable risk discrimination capabilities from the early stages of a new task.

[0091] In some embodiments, the proxy model is a hybrid proxy model, which includes a physical model and a data-driven compensation model; wherein, the data-driven compensation model is constructed based on a sparse Gaussian process and is used to learn the residual between the predicted values ​​of the basic performance indicators output by the physical model and the actual performance indicators. The process of performance prediction using a proxy model is as follows: The candidate test parameters are input into the physical model and the data-driven compensation model, respectively. Using a physical model, the predicted values ​​of basic performance indicators are output based on the physical relationship between candidate test parameters and performance indicators; and using a data-driven compensation model, the residuals between the predicted values ​​of basic performance indicators and the actual performance indicators are output. The predicted values ​​of the basic performance indicators are superimposed with the residuals to obtain the performance prediction results; the performance prediction results include the predicted mean and uncertainty of the performance indicators corresponding to the candidate test parameters.

[0092] The electronic device takes the candidate test parameters as inputs to the physical model and the data-driven compensation model, respectively.

[0093] In some embodiments, the physical model is constructed based on a model-based systems engineering (MBSE) approach, establishing a system-level physical simulation model for each instrument and its hardware links in the automated test system. The physical model is used to characterize the fundamental physical relationships between candidate test parameters and performance indicators.

[0094] Electronic devices utilize physical models to output predicted values ​​of basic performance indicators based on the physical relationship between candidate test parameters and performance indicators, denoted as... ;in, These are the candidate test parameters. These are the parameters of the physical model.

[0095] In some embodiments, the data-driven compensation model is built on a sparse Gaussian process (Sparse GP) to learn the nonlinear residuals between the predicted values ​​of the underlying performance metrics output by the physical model and the actual performance metrics.

[0096] Electronic devices utilize data-driven compensation models to output the residual between the predicted values ​​of basic performance indicators from the physical model and the actual performance indicators, denoted as . ;in, These are the trainable parameters (hyperparameters of a Gaussian process) of the data-driven model, including the kernel function and parameter noise variance, used to control the smoothness, nonlinear fitting ability, and sensitivity to observation noise of the data-driven model. Sparse GP significantly reduces computational complexity by setting a set of inducing points to perform a low-rank approximation of the entire dataset, enabling the model to be updated efficiently during online optimization. The kernel function used is the Matérn 5 / 2 kernel function to adapt to high-frequency variations in the measured signal while ensuring smoothness.

[0097] Electronic devices superimpose the predicted basic performance indicators from the physical model with the residuals from the data-driven compensation model to obtain the performance prediction result: that is, the prediction form of the hybrid proxy model is as follows. The performance prediction results include the predicted mean and uncertainty estimate of the performance metrics corresponding to the candidate test parameters. Performance metrics may include, for example, detection probability, signal-to-noise ratio, false alarm rate, and test latency.

[0098] In some embodiments, the multi-fidelity policy model is built based on reinforcement learning and is used to determine the target evaluation accuracy level from multiple evaluation accuracy levels according to the current state, and to test the candidate test parameters using the target evaluation accuracy level. The current state of reinforcement learning includes the uncertainty in the performance prediction results output by the agent model, the consumed test budget, and the variance of the historical evaluation results of the multi-fidelity policy model. Multiple assessment accuracy levels are available, including low-fidelity accuracy and high-fidelity accuracy, with different accuracy levels corresponding to different testing costs; The reward function for reinforcement learning is defined as the negative of the weighted sum of test cost and uncertainty.

[0099] In some embodiments, reinforcement learning (RL) is a machine learning method that learns the optimal decision-making policy by having an agent interact with the environment, select actions based on the state, and obtain rewards. In this embodiment, a multi-fidelity policy model acts as the agent, used to determine a target evaluation accuracy level from multiple evaluation accuracy levels based on the current state, and to use the target evaluation accuracy level to test candidate test parameters.

[0100] The current state of reinforcement learning includes three dimensions: First, there is uncertainty in the performance prediction results output by the surrogate model. This uncertainty reflects the cognitive confidence of the surrogate model in the region where the current candidate test parameters are located. The higher the uncertainty, the more insufficient the exploration of that region is, and the more necessary it is to conduct actual tests to obtain more information.

[0101] Second, the consumed test budget; the test budget refers to the upper limit of the allowed time and economic costs for executing tests. The consumed test budget records the test costs used in the current optimization process and is used to constrain the allocation of subsequent test resources.

[0102] Third, the variance of historical evaluation results in multi-fidelity strategy models. The variance of historical evaluation results reflects the consistency of past decision-making outcomes and is used to guide the stability of subsequent decisions.

[0103] The multi-fidelity strategy model's action space encompasses multiple evaluation accuracy levels, including low-fidelity and high-fidelity levels. Different accuracy levels correspond to different testing costs. Low-fidelity accuracy levels utilize fast, low-cost configurations (e.g., wider resolution bandwidth, fewer sampling points, or faster but less accurate auxiliary instruments) for large-scale coarse screening. High-fidelity accuracy levels, on the other hand, call upon full-specification instruments for precise measurements, offering high accuracy but at a high cost. The multi-fidelity strategy model intelligently selects between evaluation methods at different accuracy levels, prioritizing low-fidelity evaluation to narrow down the optimal solution region. Only when a region shows high potential in the low-fidelity results is a high-fidelity evaluation initiated for final confirmation.

[0104] The reward function for reinforcement learning is defined as the negative of the weighted sum of test cost and uncertainty, specifically expressed as: The reward function, denoted by λ, represents a coefficient that balances cost and accuracy. This function incentivizes both cost-effective evaluation methods to reduce resource consumption and uncertainty-reducing methods to improve model prediction accuracy. λ balances the weights of cost and uncertainty: a larger λ indicates a greater tendency to choose evaluation methods that reduce uncertainty, while a smaller λ indicates a greater tendency to choose cost-effective methods. The multi-fidelity policy network learns to select the most cost-effective evaluation action under different conditions, minimizing the number of expensive physical tests while maintaining accuracy.

[0105] Figure 4 This is a flowchart illustrating a method for determining the type of target test parameters provided in an embodiment of this application.

[0106] In some embodiments, such as Figure 4 Step 101 includes the following steps: Step 401: Obtain the user's testing requirements; In some embodiments, users can input test requirements in the form of natural language text or voice. Test requirements include information about the test tasks the user expects to perform, such as the test object, frequency band range, accuracy metrics, and stability requirements. The electronic device receives the user's input test requirements as input for subsequent intent parsing.

[0107] Step 402: Perform intent parsing on the test requirements to determine the target test tasks; In some embodiments, the electronic device performs intent parsing and semantic disambiguation on the user's input test requirements. Intent parsing refers to identifying key information related to the test task from the user's natural language description, including elements such as the test subject, test objective, constraints, and output requirements. The electronic device converts the parsing results into a machine-readable structured task description to obtain the target test task.

[0108] Step 403: Match the corresponding target test template from the template library according to the target test task; In some embodiments, the template library is a predefined test process "script library" that stores standardized test process templates covering multiple instruments and scenarios. Each test template corresponds to a test scenario and includes core elements such as the type definition of test parameters required to execute the test scenario, execution logic, and exception handling. The electronic device matches the corresponding target test template from the template library based on three dimensions: the test subject, collaborating instruments, and test target in the target test task.

[0109] The template library is shown in Table 1 below: Table 1. Template Library Illustration

[0110] Step 404: Extract the target test parameter types and the search range corresponding to each target test parameter type from the target test template.

[0111] In some embodiments, the target test template contains type definitions of test parameters required to perform the corresponding test task. For example, the "signal source output control template" contains type definitions of test parameters such as center frequency, output power, and modulation method; the "spectrum analyzer scanning and acquisition template" contains type definitions of test parameters such as scan start frequency, end frequency, and resolution bandwidth. The electronic device extracts the types of test parameters from the target test template as target test parameter types, and determines the search range corresponding to each target test parameter type in combination with the instrument's physical range.

[0112] Figure 5 This is a flowchart illustrating the method for determining a target test template provided in an embodiment of this application.

[0113] In some embodiments, such as Figure 5 Step 403 includes the following steps: Step 501: Perform first-level matching based on the test subject in the target test task to determine the instrument control template corresponding to the target test task; In some embodiments, the test subject refers to the object under test, such as electronic measuring instruments like signal generators, spectrum analyzers, and vector network analyzers. As shown in Table 1, the template library stores instrument control templates corresponding to different instrument types. The electronic device matches the corresponding instrument type template category from the template library based on the test subject in the target test task. For example, when the test subject is a "signal generator," the first-level matching locks the "signal generator control template"; when the test subject is a "spectrum analyzer," the first-level matching locks the "spectrum analyzer control template."

[0114] Step 502: Perform secondary matching based on the collaborating instruments in the target test task to determine the process control template corresponding to the target test task; In some embodiments, collaborating instruments refer to other instruments that work in conjunction with the test subject during the execution of a test task. When a test task involves the collaborative work of multiple instruments, a corresponding flow control template is required to define the collaboration logic and synchronization mechanism between the instruments. The electronic device matches the corresponding flow control template from the template library based on the collaborating instruments in the target test task. For example, when a test task requires collaborative testing of a "signal source + spectrum analyzer," a secondary matching lock is placed on the "multi-instrument collaborative test template." The flow control template includes the parameter variable definitions, synchronization trigger conditions, and collaborative execution logic for each instrument.

[0115] Step 503: Perform three-level matching based on the test objectives in the target test task to determine the data collection template corresponding to the target test task; In some embodiments, the test target refers to the performance indicators that the user expects to measure, such as output power stability testing, spurious radiation testing, and frequency response testing. The template library stores data acquisition templates corresponding to different test scenarios. Each data acquisition template includes the definition of acquisition parameter variables, acquisition execution logic, and data output format for that test scenario. The electronic device matches the corresponding data acquisition template from the template library based on the test target in the target test task. For example, when the test target is "output stability testing," the three-level matching locks the "continuous scan acquisition template"; when the test target is "spurious radiation testing," the three-level matching locks the "spurious scan acquisition template."

[0116] Step 504: Combine the instrument control template, process control template, and data acquisition template to form the target test template.

[0117] In some embodiments, the electronic device combines the three templates matched in steps 501 to 503 to form a complete target test template. This target test template includes all parameter variable definitions, execution logic, and data format specifications required for this test task, and is used for subsequent test parameter optimization and instrument control command generation.

[0118] In some embodiments, step 402 includes the following steps: Semantic parsing is performed on test requirements to extract the test subject, test objective, test constraints, and output requirements. Generate an initial structured task description based on the test subject, test objective, test constraints, and output requirements; The initial structured task description is validated for completeness. If the initial structured task description is incomplete, guidance information is generated to obtain supplementary information from the user. The initial structured task description is then updated based on the information to obtain the target test task.

[0119] In some embodiments, semantic parsing refers to identifying key information related to the testing task from the user's natural language description. In this application embodiment, semantic parsing adopts a BERT+Deepseek hybrid architecture, parsing the user's input testing requirements through a pre-trained testing domain corpus. The testing domain corpus contains a terminology database for typical testing scenarios such as "spectrum analyzer calibration" and "signal source output stability testing".

[0120] The test subject refers to the object under test, such as a "signal source," "spectrum analyzer," or "vector network analyzer." The test objective refers to the performance indicators that the user expects to measure, such as "spurious emissions ≤ -80dBm" or "output power fluctuation ≤ ±0.5dB." The test constraints refer to the limiting conditions that need to be met during the test, such as the test scenario (e.g., "2.4GHz band" or "normal temperature environment") and the instrument type (e.g., "needs to call a signal source + spectrum analyzer for joint testing"). The output requirements refer to the user's requirements for the test report, such as the report format (e.g., "PDF / Excel") and the level of detail (e.g., "conclusion only" or "includes raw data").

[0121] In some embodiments, the structured task description may be in JSON format and include fields such as "test_subject", "test_goal", "constraints", and "output_requirements".

[0122] In some embodiments, integrity verification refers to checking whether key information is missing from the structured task description, such as whether necessary fields like "test frequency band" or "metric threshold" are missing. When the initial structured task description is incomplete, the electronic device generates guidance information to obtain supplementary information from the user. The guidance information prompts the user to supplement the missing key information; for example, when the test frequency band is missing, the device generates guidance information such as "What is the frequency band range you need to test?". The electronic device obtains the supplementary information input by the user based on the guidance information and updates the initial structured task description accordingly, completing the missing fields to obtain the complete target test task. When the initial structured task description is complete, the electronic device directly determines the initial structured task description as the target test task.

[0123] In some embodiments, after step 103, the test parameter determination method provided in this application further includes the following steps: The optimal test parameters are used to assign values ​​to the corresponding parameters in the target test template, and instrument control commands are generated based on the assignment results. According to the instrument control instructions, control the corresponding instrument to perform the test on the test subject corresponding to the test requirements, and collect the test data returned by the test subject.

[0124] In some embodiments, the electronic device fills the corresponding parameter variables in the target test template with the optimal test parameters. For example, if the target test template is a combination of a "signal source output control template + spectrum analyzer scanning and acquisition template", then the optimal parameter values ​​such as "center frequency = 2.4 GHz", "output power = 3 dBm", "scan range = 2.39-2.41 GHz", and "RBW = 10 kHz" are filled into the variables of the corresponding templates. After filling, the electronic device generates instrument control commands based on the assigned target test template.

[0125] Electronic devices send instrument control commands to the corresponding instruments through the standardized interface of the signal analyzer / signal generator (SA / SG) and collect the test data returned by the instruments.

[0126] The SA / SG standardized interface is used to shield the underlying differences in communication protocols between different instruments. Specifically, the SA / SG standardized interface includes functions such as a protocol adaptation layer, command encapsulation, and data parsing. The protocol adaptation layer encapsulates the low-level operations of mainstream communication protocols such as VISA, GPIB, and LAN, including TCP / IP connections and SCPI command parsing, providing a unified calling interface for upper layers. Command encapsulation encapsulates the instrument's proprietary commands (such as "FREQ:CENT" for a spectrum analyzer) into standardized operation functions (such as set_center_frequency(frequency)). When calling these functions, upper layers only need to pass in the physical quantity (such as 2400MHz) without needing to concern themselves with the details of the underlying protocol.

[0127] Taking the collaborative testing of signal source and spectrum analyzer as an example, the electronic device sends the command "set frequency 2.4GHz, power 3dBm, start output" to the signal source through the SA / SG standardized interface, and at the same time sends control commands such as "set scan range 2.39-2.41GHz, RBW=10kHz, start scan" to the spectrum analyzer, driving the device to execute the test according to the established process.

[0128] In some embodiments, the SA / SG standardized interface employs a loosely coupled "main program + plugin" design to support different types of instruments. The main program defines the common interfaces for instrument drivers (e.g., initialize(), set_parameters(), acquire_data()), with each instrument corresponding to an independent driver plugin. The electronic device supports dynamically loading new driver plugins at runtime via a plugin manager (e.g., when adding a vector network analyzer, only the .dll or .so format plugin file needs to be placed in the specified directory), without restarting the main program. When a plugin is loaded, the electronic device automatically verifies the interface specifications (e.g., whether the initialize() method is implemented); plugins that do not conform to the specifications will be rejected and an error message will be displayed.

[0129] During testing, the electronic equipment acquires test data returned by the instrument in real time. This test data includes raw measurement data such as spectrum curves, power values, and noise levels. The data parsing function in the SA / SG standardized interface converts the raw data returned by the instrument (e.g., binary scan data from a spectrum analyzer) into structured data (e.g., a JSON array containing frequency and power values). The electronic equipment automatically timestamps the converted test data, performs preliminary preprocessing, and saves the preprocessed test data to a local database according to a template-defined format (e.g., CSV or JSON), forming a structured experimental dataset.

[0130] In some embodiments, after collecting the test data returned by the test subject, the test parameter determination method provided in this application embodiment further includes the following steps: A test report is generated based on the test data.

[0131] The test report is a standardized, visualized document automatically generated based on test data. The report's content is based on test conditions, optimal test parameters, raw data curves, and AI analysis results. Test conditions include information on the test subject, collaborating instruments, and test environment involved in the test. Optimal test parameters are the optimal parameter configuration determined by the electronic device through iterative optimization using the aforementioned steps. Raw data curves are visualized curves corresponding to the test data collected during test execution, such as spectrum curves. AI analysis results are conclusive information automatically derived by the electronic device based on test data and test constraints.

[0132] The electronic device generates a standardized, visual report that includes a summary, test procedure, data charts, conclusions, and improvement suggestions. The summary briefly outlines the test objective, results, and key conclusions. The test procedure records the complete test execution flow, including test parameter settings, test steps, and test environment information. Data charts present test data and analysis results in a visual format, such as spectrum graphs and data tables. The conclusions determine whether the test meets the test constraints. Improvement suggestions provide a reference for future test optimization directions.

[0133] In some embodiments, if the test report shows that the test results do not meet the standards, the test parameter determination method provided in this application embodiment further includes the following steps: When the test results fail to meet the standards, the electronic device automatically triggers the debugging process. The device analyzes the template execution logs and parameter mappings to pinpoint the root cause of the problem and mark the anomaly type. For example, the electronic device detects anomalies such as "the scan interval was incorrectly filled with 15s" and marks the corresponding problem type.

[0134] In some embodiments, the electronic device provides a web interface for users to view parameter execution records and instrument logs. Users can manually identify the source of problems and correct templates or rules through the web interface. The electronic device then sends the corrected template, along with the parameter configuration, execution results, and evaluation results for this task, back to the meta-knowledge base. The meta-knowledge base is updated based on the sent data.

[0135] Through the aforementioned debugging and optimization mechanisms, electronic devices continuously accumulate real-world testing experience. In subsequent tasks, these devices can search for optimal test parameter combinations more quickly and accurately. As testing tasks are continuously executed, the electronic devices achieve a continuous self-optimization capability, becoming "more accurate with testing and smarter with use."

[0136] To implement the test parameter determination method provided in the embodiments of this application, the embodiments of this application also provide a test parameter determination device. For example... Figure 6 As shown, Figure 6 A schematic diagram of a test parameter determination device provided in an embodiment of this application. The test parameter determination device includes: The target test parameter type determination module 601 is used to determine the target test parameter type corresponding to the user's test requirements; The optimal test parameter determination module 602 is used to initialize the proxy model, security assessment model, and multi-fidelity strategy model using global initialization parameters; perform a first operation to iteratively update the proxy model and security assessment model until a stopping condition is met, and determine the current candidate test parameters as the optimal test parameters; wherein, the first operation includes: generating candidate test parameters within the search range corresponding to the target test parameter type, inputting the candidate test parameters selected by the security constraints of the security assessment model into the proxy model for performance prediction, performing actual tests on the candidate test parameters through the multi-fidelity strategy model based on the performance prediction results, and updating the proxy model and security assessment model based on the actual test results.

[0137] In some embodiments, the optimal test parameter determination module 602 is specifically used for: Retrieve global initialization parameters from the meta-knowledge base; The proxy model, security assessment model, and multi-fidelity strategy model are initialized using the initial proxy model parameters, initial security assessment model parameters, and initial multi-fidelity strategy model parameters in the global initialization parameters, respectively.

[0138] In some embodiments, the optimal test parameter determination module 602 is further specifically used for: Obtain the meta-training set; wherein the meta-training set contains at least one historical test task data; Using historical test task datasets, a model-independent meta-learning algorithm is employed to train the agent model, security assessment model, and multi-fidelity strategy model to obtain global initialization parameters.

[0139] In some embodiments, the optimal test parameter determination module 602 is further specifically used for: Iterate through the historical test task data in the meta-training set, and perform the second operation on the currently iterated historical test task data until the current meta-parameters converge, and obtain the global initialization parameters. The second operation includes: Initialize the proxy model, security assessment model, and multi-fidelity strategy model using the current meta-parameters; Using a portion of the historical test task data currently being traversed, the agent model, security assessment model, and multi-fidelity strategy model are trained using the gradient descent algorithm to obtain the task adaptation parameters of the agent model, security assessment model, and multi-fidelity strategy model. Calculate the loss value of the task adaptation parameters using the remaining data in the current traversal of historical test task data; Update the current meta-parameters by backpropagating based on the loss value.

[0140] In some embodiments, the optimal test parameter determination module 602 is further specifically used for: A first candidate test parameter set is generated within the search range corresponding to the target test parameter type using a grid search strategy; The second candidate test parameter set, selected through the security constraints of the security assessment model, is input into the proxy model for performance prediction, and the predicted mean and uncertainty of the performance index corresponding to each second candidate test parameter are obtained. Each second candidate test parameter is scored using a data acquisition function, and candidate test parameters are selected from the second candidate test parameters based on the scoring results.

[0141] In some embodiments, the security assessment model is a binary Gaussian process classifier; the optimal test parameter determination module 602 is further specifically used for: Input the candidate test parameters into the security assessment model to obtain the risk probability of the candidate test parameters; When the risk probability exceeds the risk probability threshold, the candidate test parameter is determined to have failed the safety constraint screening.

[0142] In some embodiments, the proxy model is a hybrid proxy model, which includes a physical model and a data-driven compensation model; wherein, the data-driven compensation model is constructed based on a sparse Gaussian process and is used to learn the residual between the predicted values ​​of the basic performance indicators output by the physical model and the actual performance indicators; the optimal test parameter determination module 602 is further specifically used for: The candidate test parameters are input into the physical model and the data-driven compensation model, respectively. Using a physical model, the predicted values ​​of basic performance indicators are output based on the physical relationship between candidate test parameters and performance indicators; and using a data-driven compensation model, the residuals between the predicted values ​​of basic performance indicators and the actual performance indicators are output. The predicted values ​​of the basic performance indicators are superimposed with the residuals to obtain the performance prediction results; the performance prediction results include the predicted mean and uncertainty of the performance indicators corresponding to the candidate test parameters.

[0143] In some embodiments, the multi-fidelity policy model is built based on reinforcement learning and is used to determine the target evaluation accuracy level from multiple evaluation accuracy levels according to the current state, and to test the candidate test parameters using the target evaluation accuracy level. The current state of reinforcement learning includes the uncertainty in the performance prediction results output by the agent model, the consumed test budget, and the variance of the historical evaluation results of the multi-fidelity policy model. Multiple assessment accuracy levels are available, including low-fidelity accuracy and high-fidelity accuracy, with different accuracy levels corresponding to different testing costs; The reward function for reinforcement learning is defined as the negative of the weighted sum of test cost and uncertainty.

[0144] In some embodiments, the target test parameter type determination module 601 is specifically used for: Obtain users' testing requirements; Perform intent analysis on the test requirements to determine the target test tasks; Based on the target test task, match the corresponding target test template from the template library; Extract the target test parameter types and the search range corresponding to each target test parameter type from the target test template.

[0145] In some embodiments, the target test parameter type determination module 601 is further specifically used for: Perform a first-level matching based on the test subject in the target test task to determine the instrument control template corresponding to the target test task; Based on the secondary matching of the collaborating instruments in the target test task, determine the corresponding process control template for the target test task; Perform a three-level matching based on the test objectives in the target test task to determine the data collection template corresponding to the target test task; The instrument control template, process control template, and data acquisition template are combined to form the target test template.

[0146] In some embodiments, the target test parameter type determination module 601 is further specifically used for: Semantic parsing is performed on test requirements to extract the test subject, test objective, test constraints, and output requirements. Generate an initial structured task description based on the test subject, test objective, test constraints, and output requirements; The initial structured task description is validated for completeness. If the initial structured task description is incomplete, guidance information is generated to obtain supplementary information from the user. The initial structured task description is then updated based on the information to obtain the target test task.

[0147] In some embodiments, the stopping condition is: The uncertainty in the performance prediction results output by the proxy model is lower than the preset uncertainty threshold; Or / and, the number of iterations for the first operation reaches the preset maximum number of iterations; Or / and, the performance prediction results corresponding to the candidate test parameters output in consecutive iterations no longer change.

[0148] In some embodiments, the optimal test parameter determination module 602 is further specifically used for: The candidate test parameters of the current iteration and the corresponding test results are used as training samples to fine-tune the proxy model and the security assessment model using gradient descent, respectively, in order to update the parameters of the proxy model and the security assessment model.

[0149] In some embodiments, the test parameter determination apparatus provided in this application further includes a test module, which is used for: The optimal test parameters are used to assign values ​​to the corresponding parameters in the target test template, and instrument control commands are generated based on the assignment results. According to the instrument control instructions, control the corresponding instrument to perform the test on the test subject corresponding to the test requirements, and collect the test data returned by the test subject.

[0150] In some embodiments, the test parameter determination apparatus provided in this application further includes a test report generation module, which is used for: A test report is generated based on the test data.

[0151] The test parameter determination device provided in the embodiments of this application above describes the test parameter determination method provided in the embodiments of this application. To implement the functions of the test parameter determination method provided in the embodiments of this application above, the test parameter determination device may include hardware structures and software modules, and may implement the above functions in the form of hardware structures, software modules, or a combination of hardware structures and software modules. One of the above functions may be executed in the form of hardware structures, software modules, or a combination of hardware structures and software modules.

[0152] Figure 7 This is a block diagram illustrating an electronic device 700 for implementing the above-described test parameter determination method, according to an exemplary embodiment. For example, the electronic device 700 may be a computer, a personal digital assistant, etc.

[0153] Reference Figure 7 The electronic device 700 may include a communication interface 701, capable of interacting with other devices; a processor 702, connected to the communication interface 701 to interact with other devices, used to execute the methods provided by one or more of the above-described technical solutions when running a computer program; and a memory 703, on which the computer program is stored. Specifically, the specific processing operations of the processor 702 can refer to the test parameter determination method described in the above embodiments of this disclosure.

[0154] Of course, in practical applications, the various components in electronic device 700 are coupled together through bus system 704. It can be understood that bus system 704 is used to realize the connection and communication between these components. In addition to a data bus, bus system 704 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, in... Figure 7 The general designated all buses as Bus System 704.

[0155] The memory 703 in this embodiment is used to store various types of data to support the operation of the electronic device 700. Examples of such data include any computer program used to operate on the electronic device 700.

[0156] The methods disclosed in the embodiments of this application can be applied to processor 702, or implemented by processor 702. Processor 702 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 702 or by instructions in software form. The processor 702 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 702 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. A general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in memory 703. Processor 702 reads information from memory 703 and, in conjunction with its hardware, completes the steps of the aforementioned method.

[0157] In an exemplary embodiment, the electronic device 700 may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned method.

[0158] Embodiments of this disclosure also provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the test parameter determination method described in the above embodiments of this disclosure.

[0159] Embodiments of this disclosure also provide a computer program product, including a computer program that, when executed by a processor, implements the test parameter determination method described in the above embodiments of this disclosure.

[0160] Embodiments of this disclosure also propose a chip including one or more interface circuits and one or more processors; the interface circuits are used to receive signals from the memory of an electronic device and send signals to the processors, the signals including computer instructions stored in the memory, which, when executed by the processor, cause the electronic device to perform the test parameter determination method described in the above embodiments of this disclosure.

[0161] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0162] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0163] Any description of operation or method in the flowchart or otherwise described herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or operation, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0164] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processing module, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (control method), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic device, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0165] It should be understood that various parts of the embodiments of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0166] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by instructing related hardware through an operation sequence, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0167] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc.

[0168] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for determining test parameters, characterized in that, The method includes: Determine the target test parameter types corresponding to the user's testing requirements; Initialize the proxy model, security assessment model, and multi-fidelity strategy model using global initialization parameters; The first operation iteratively updates the proxy model and the security assessment model until a stopping condition is met, at which point the current candidate test parameter is determined to be the optimal test parameter. The first operation includes: generating candidate test parameters within the search range corresponding to the target test parameter type; inputting the candidate test parameters selected through the security constraints of the security assessment model into the proxy model for performance prediction; performing actual tests on the candidate test parameters using the multi-fidelity strategy model based on the performance prediction results; and updating the proxy model and the security assessment model based on the actual test results.

2. The method according to claim 1, characterized in that, The initialization of the proxy model, security assessment model, and multi-fidelity strategy model using global initialization parameters includes: The global initialization parameters are obtained from the meta-knowledge base; The proxy model, the security assessment model, and the multi-fidelity strategy model are initialized using the initial proxy model parameters, the initial security assessment model parameters, and the initial multi-fidelity strategy model parameters in the global initialization parameters, respectively.

3. The method according to claim 1 or 2, characterized in that, The global initialization parameters are determined in the following way: Obtain a meta-training set; wherein the meta-training set contains at least one historical test task data; Using historical test task datasets, the agent model, the security assessment model, and the multi-fidelity strategy model are trained using a model-independent meta-learning algorithm to obtain the global initialization parameters.

4. The method according to claim 3, characterized in that, The method utilizes historical test task datasets and employs a model-independent meta-learning algorithm to train the proxy model, the security assessment model, and the multi-fidelity strategy model to obtain the global initialization parameters, including: Traverse the historical test task data in the meta-training set, perform the second operation on the currently traversed historical test task data, until the current meta-parameters converge, and obtain the global initialization parameters; The second operation includes: The proxy model, the security assessment model, and the multi-fidelity strategy model are initialized using the current meta-parameters. Using a portion of the historical test task data currently being traversed, the proxy model, the security assessment model, and the multi-fidelity strategy model are trained using the gradient descent algorithm to obtain the task adaptation parameters of the proxy model, the security assessment model, and the multi-fidelity strategy model. Using the remaining data in the currently traversed historical test task data, calculate the loss value of the task adaptation parameter; The current meta-parameters are updated by backpropagation based on the loss value.

5. The method according to claim 1, characterized in that, The step of generating candidate test parameters within the search range corresponding to the target test parameter type includes: A first candidate test parameter set is generated within the search range corresponding to the target test parameter type using a grid search strategy; The second candidate test parameter set, filtered by the security constraints of the security assessment model, is input into the proxy model for performance prediction, and the predicted mean and uncertainty of the performance index corresponding to each second candidate test parameter are obtained. Each of the second candidate test parameters is scored using a data acquisition function, and candidate test parameters are selected from the second candidate test parameters based on the scoring results.

6. The method according to claim 1, characterized in that, The safety assessment model is a binary Gaussian process classifier; the safety constraint screening includes: The candidate test parameters are input into the security assessment model to obtain the risk probability of the candidate test parameters. When the risk probability exceeds the risk probability threshold, the candidate test parameter is determined to have failed the safety constraint screening.

7. The method according to claim 1, characterized in that, The proxy model is a hybrid proxy model, which includes a physical model and a data-driven compensation model; wherein, the data-driven compensation model is constructed based on a sparse Gaussian process and is used to learn the residual between the predicted values ​​of the basic performance indicators output by the physical model and the actual performance indicators. The process of performance prediction by the proxy model is as follows: The candidate test parameters are input to the physical model and the data-driven compensation model, respectively. Using the physical model, the predicted value of the basic performance index is output based on the physical relationship between the candidate test parameters and the performance index; and using the data-driven compensation model, the residual between the predicted value of the basic performance index and the actual performance index is output. The predicted values ​​of the basic performance indicators are superimposed with the residuals to obtain the performance prediction results; wherein, the performance prediction results include the predicted mean and uncertainty of the performance indicators corresponding to the candidate test parameters.

8. The method according to claim 1, characterized in that, The multi-fidelity strategy model is built based on reinforcement learning and is used to determine the target evaluation accuracy level from multiple evaluation accuracy levels according to the current state, and to use the target evaluation accuracy level to test the candidate test parameters. The current state of the reinforcement learning includes the uncertainty in the performance prediction results output by the agent model, the consumed test budget, and the variance of the historical evaluation results of the multi-fidelity policy model. The various evaluation accuracy levels include low-fidelity accuracy level and high-fidelity accuracy level, with different accuracy levels corresponding to different testing costs; The reward function for reinforcement learning is defined as the negative of the weighted sum of the test cost and the uncertainty.

9. The method according to claim 1, characterized in that, The determination of the target test parameter type corresponding to the user's test requirements includes: Obtain users' testing requirements; The intent of the test requirements is parsed to determine the target test tasks; Based on the target test task, match the corresponding target test template from the template library; Extract the target test parameter types and the search range corresponding to each target test parameter type from the target test template.

10. The method according to claim 9, characterized in that, The step of matching the corresponding target test template from the template library according to the target test task includes: Perform a first-level matching based on the test subject in the target test task to determine the instrument control template corresponding to the target test task; Based on the collaborative instruments in the target test task, perform secondary matching to determine the process control template corresponding to the target test task; Perform a three-level matching based on the test objectives in the target test task to determine the data collection template corresponding to the target test task; The instrument control template, the process control template, and the data acquisition template are combined to form the target test template.

11. The method according to claim 9, characterized in that, The process of parsing the intent of the test requirements to determine the target test task includes: Semantic parsing is performed on the test requirements to extract the test subject, test objective, test constraints and output requirements. Based on the test subject, the test objective, the test constraints, and the output requirements, an initial structured task description is generated; The initial structured task description is checked for completeness. If the initial structured task description is incomplete, guidance information is generated to obtain supplementary information from the user, and the initial structured task description is updated according to the information to obtain the target test task.

12. The method according to claim 1, characterized in that, The stopping condition is: The uncertainty in the performance prediction results output by the proxy model is lower than a preset uncertainty threshold. Or / and, the number of iterations of the first operation reaches the preset maximum number of iterations; Or / and, the performance prediction results corresponding to the candidate test parameters output in consecutive iterations no longer change.

13. The method according to claim 1, characterized in that, The step of updating the proxy model and the security assessment model based on the actual test results includes: The candidate test parameters of the current iteration and the corresponding test results are used as training samples to fine-tune the proxy model and the security assessment model using gradient descent, respectively, in order to update the parameters of the proxy model and the security assessment model.

14. The method according to claim 9, characterized in that, After determining that the current candidate test parameter is the optimal test parameter, the method further includes: The optimal test parameters are used to assign values ​​to the corresponding parameters in the target test template, and instrument control commands are generated based on the assignment results. According to the instrument control command, the corresponding instrument is controlled to perform the test on the test subject corresponding to the test requirement, and the test data returned by the test subject is collected.

15. The method according to claim 14, characterized in that, After collecting the test data returned by the test subject, the method further includes: A test report is generated based on the test data.

16. A device for determining test parameters, characterized in that, The device includes: The target test parameter type determination module is used to determine the target test parameter type corresponding to the user's test requirements; The optimal test parameter determination module is used to initialize a proxy model, a security assessment model, and a multi-fidelity strategy model using global initialization parameters; perform a first operation to iteratively update the proxy model and the security assessment model until a stopping condition is met, and determine the current candidate test parameters as the optimal test parameters; wherein, the first operation includes: generating candidate test parameters within the search range corresponding to the target test parameter type; inputting the candidate test parameters filtered by the security constraints of the security assessment model into the proxy model for performance prediction; performing actual tests on the candidate test parameters through the multi-fidelity strategy model based on the performance prediction results; and updating the proxy model and the security assessment model based on the actual test results.

17. An electronic device, characterized in that, include: A processor and a memory for storing a computer program capable of running on the processor, wherein the processor, when running the computer program, performs the method of any one of claims 1 to 15.

18. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 15.

19. A computer program product, characterized in that, It includes a computer program, which, when executed by a processor, causes the processor to perform the method of any one of claims 1 to 15.

Images