Method for testing and evaluating overall performance of unmanned aerial vehicle

Abstract

This invention discloses a method for testing and evaluating the overall performance of an unmanned aerial vehicle (UAV). The method includes: collecting multi-source operational data and unifying the time base; performing cross-platform semantic unification to generate standardized state vectors; calculating scene coverage based on a parameterized scene library, generating a supplementary testing plan with the assessment confidence prediction improvement as the target, and feeding back the newly added data; inputting the standardized state vectors and scene parameters into a digital twin model for simulation; constructing measured and simulated residuals and performing hierarchical attribution at the subsystem and coupled link layers under residual conservation constraints; performing constrained online calibration based on the residuals and outputting convergence indices; constructing a subsystem collaboration graph and performing reliability updates to obtain a reliability posterior; outputting an overall performance evaluation report and fault location clues, and triggering supplementary testing or increasing the sampling frequency when the confidence level is low. This invention improves the reliability and interpretability of the evaluation, reduces testing costs, and supports optimized design and fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) performance testing and evaluation technology, specifically to a method for testing and evaluating the overall performance of a UAV. Background Technology

[0002] Unmanned aerial vehicle (UAV) overall performance testing typically requires coverage of multiple types of operational data and external operating conditions to verify the collaborative capabilities and reliability of various subsystems under coupled conditions. Current technologies often rely on fixed test scripts or offline indicator statistics for overall system evaluation. On the one hand, the coverage of test data in the operating space lacks a unified quantitative standard, leading to an invisible risk to assessment conclusions regarding uncovered conditions, and a lack of calculable closed-loop optimization between testing investment and conclusion credibility. On the other hand, flight control logs, energy telemetry, execution feedback, navigation perception, and environmental monitoring link fields, units, coordinate systems, and sampling periods differ significantly among various UAV models. Cross-platform benchmarking often relies on manual mapping, making it difficult to establish a traceable and unified evaluation benchmark. Furthermore, conventional "actual-simulation" error comparisons usually remain at the error magnitude level, lacking a mechanism to interpret and verify the deviation between virtual and real data along subsystems and coupled links, making fault location clues difficult to verify. Finally, evaluation reports often lack quantitative expression of the reliability of conclusions, failing to directly trigger supplementary testing or sampling strategy adjustments, and thus hindering rapid iterative design and large-scale operation and maintenance diagnostics for UAVs.

[0003] Therefore, there is an urgent need for a pre-evaluation method and system for the overall performance testing of UAVs that can actively supplement tests under safety and energy constraints, hierarchically attribute virtual and real deviations to the coupled links, and output evaluation confidence and executable correction actions. Summary of the Invention

[0004] Technical Objective: To address the shortcomings of existing technologies, this invention discloses a method for testing and evaluating the overall performance of unmanned aerial vehicles (UAVs). Under cross-platform, multi-source operational data conditions, it achieves traceable semantic unification and time alignment, quantifies test coverage in a computable manner, generates supplementary test plans under constraints, decomposes virtual-real deviations to subsystems and coupled link layers while maintaining verifiable consistency and quantifying the confidence of evaluation conclusions, and drives closed-loop correction.

[0005] Technical solution: To achieve the above technical objectives, the present invention adopts the following technical solution:

[0006] A method for testing and evaluating the overall performance of a drone, specifically including the following steps:

[0007] S1. Collect multi-source operational data to characterize the UAV's operating status and external conditions, and assign a unified time reference to the multi-source operational data;

[0008] S2. Perform cross-platform semantic unified processing on multi-source operational data to generate standardized state vectors;

[0009] S3. Calculate the scene coverage index based on the parameterized scene library and standardized state vector. When the scene coverage index is less than the coverage threshold, take the predicted improvement of the evaluation confidence by the candidate supplementary test scene as the target, and generate a supplementary test plan under the condition of meeting flight safety constraints and energy budget constraints. Drive the UAV to perform supplementary test actions according to the supplementary test plan and feed back the newly added multi-source operation data.

[0010] S4. Input the standardized state vector and scene parameters into the digital twin model, and perform simulation to obtain simulation output;

[0011] S5. Construct the residual vector between the measured output and the simulation output, decompose the residual vector into subsystem residuals by subsystem layer, and allocate the link residuals by coupling link layer under the condition of satisfying the residual conservation constraint.

[0012] S6. Perform constrained online calibration on the parameter set of the digital twin model based on the subsystem residuals and link residuals to obtain the calibrated digital twin model and output the calibration convergence index.

[0013] S7. Construct a subsystem collaboration graph based on the calibrated digital twin model and standardized state vector, calculate collaboration performance index, and perform reliability update based on collaboration performance index and link residual to obtain reliability posterior.

[0014] S8. Calculate the evaluation confidence level based on the scene coverage index, residual statistics and calibration convergence index, and output the overall performance evaluation report containing the evaluation confidence level and the fault location clues corresponding to the link residuals; when the evaluation confidence level is lower than the confidence level threshold, trigger at least one of regenerating the supplementary test plan and increasing the sampling frequency.

[0015] In one embodiment, the multi-source operational data includes at least two of the following categories: flight status data, power and energy data, control and execution data, navigation and perception data, and environmental data.

[0016] In one embodiment, the cross-platform semantic unified processing includes at least field semantic mapping; the field semantic mapping is implemented through a mapping table, which includes at least source field identifier, target field identifier, unit, coordinate system, sampling period, channel delay and mapping version number, and the mapping version number is written into the metadata of the overall performance evaluation report.

[0017] In one embodiment, the cross-platform semantic unification processing further includes dimensional unification, coordinate system unification, and asynchronous sampling alignment; the asynchronous sampling alignment includes resampling alignment and delay compensation alignment based on a unified time grid to form a standardized state vector; and generates missing data markers for missing data segments, using the missing data markers as one of the deduction factors in the evaluation confidence calculation to suppress the impact of missing data segment interpolation errors on the evaluation conclusion.

[0018] In one embodiment, the predicted improvement in the confidence level is jointly determined by the predicted decrease in residual variance and the predicted decrease in posterior uncertainty of calibration parameters for the candidate supplementary testing scenario; the flight safety constraints include at least an upper limit for attitude angle and a lower limit for altitude, and the energy budget constraints include at least a lower limit for battery state of charge; the allocation of link residuals by coupled link layer includes: generating a node-link correlation matrix based on the coupled interface topology of the digital twin model, and solving for the link residuals that satisfy the residual conservation constraints under the condition of calling the coupled interface transfer matrix of the current working point, so that the link residual allocation result is consistent with the subsystem residuals and can be verified.

[0019] A system for testing and evaluating the overall performance of an unmanned aerial vehicle (UAV), comprising:

[0020] The data interface is used to access multi-source runtime data;

[0021] processor;

[0022] Memory is used to store instructions that can be executed by the processor;

[0023] When the processor executes instructions, it is used to implement the overall performance testing and evaluation method of the UAV as described above, and output an overall performance evaluation report and fault location clues.

[0024] In one embodiment, the memory stores a mapping table and a mapping version number, and the processor writes the mapping version number into the metadata of the overall performance evaluation report when performing field semantic mapping to support cross-model benchmarking and traceability.

[0025] In one embodiment, the supplementary testing plan includes at least parameters of the scenario to be supplemented, a sequence of supplementary testing actions, and a data acquisition configuration. The data acquisition configuration includes at least a sampling frequency and a channel activation list. The supplementary testing plan also includes ranking values ​​of the predicted improvement in evaluation confidence corresponding to the candidate supplementary testing scenarios, which are used to select the best candidates for supplementary testing under the condition that the constraints are met.

[0026] In one embodiment, when allocating link residuals, the processor calls the coupling interface of the digital twin model at the current operating point to transfer the matrix, and makes the sum of the inbound link residuals and outbound link residuals associated with a certain subsystem equal to the residual of that subsystem, thus satisfying the residual conservation constraint.

[0027] In one embodiment, the overall performance evaluation report further includes a confidence decomposition vector, which includes at least coverage contribution, residual contribution, and calibration convergence contribution; when the evaluation confidence is lower than the confidence threshold, the processor selects at least one of triggering regeneration of the supplementary test plan and increasing the sampling frequency based on the confidence decomposition vector.

[0028] Beneficial Effects: The method for testing and evaluating the overall performance of a drone provided by this invention has the following beneficial effects:

[0029] 1. This invention establishes a unified time benchmark for multi-source operational data and performs cross-platform semantic unification processing (including field semantic mapping, unit unification, coordinate system unification, and asynchronous sampling alignment). At the same time, the mapping version number is written into the metadata of the evaluation report, enabling data from different models and different links to be comparable and traceable in the same semantic space. Furthermore, through missing test marking and missing test deduction mechanisms, the interpolation uncertainty of missing test segments is explicitly incorporated into the confidence assessment, thereby reducing the pollution of residual calculation and reliability updates by statistical biases introduced by missing tests and asynchronous sampling. This makes the overall evaluation conclusions more consistent and verifiable when reproduced across platforms and operating conditions.

[0030] 2. This invention constructs a parameterized scenario library and calculates scenario coverage indicators. When coverage is insufficient, a supplementary test plan is generated with the predicted improvement of assessment confidence by candidate supplementary test scenarios as the target. Supplementary tests are then executed preferentially under flight safety and energy budget constraints, realizing a closed-loop test organization method of coverage-supplementary test-data feedback. Compared with the traditional method of adding tests according to fixed scripts or experience, this mechanism can concentrate limited flight resources on the scenario range with the most significant reduction in assessment uncertainty, reduce redundant testing and blind supplementary tests, shorten the test cycle required to reach the coverage threshold and confidence threshold, and improve the confidence level of the assessment report and the interpretability of its improvement path.

[0031] 3. This invention inputs standardized state vectors and scene parameters into a digital twin model to obtain simulation output. After constructing residual vectors, it decomposes them at the subsystem level and allocates them at the coupling link level to obtain subsystem residuals and link residuals. The link allocation satisfies residual conservation constraints and combines the propagation relationship of the coupling interface to constrain the distribution direction, so that the source of difference can be consistently attributed along the subsystem-link level. On this basis, constrained online calibration (parameter physical boundary, closed-loop stability and parameter change rate constraints) is used to suppress non-physical drift and output convergence index. Reliability update is performed based on the subsystem co-operation graph to obtain the reliability posterior. Finally, the output includes the overall performance evaluation report and fault location clues containing the evaluation confidence and its decomposed vector. Self-correction actions such as retesting / increasing sampling frequency are triggered when the confidence is low, thereby forming a calibrable, interpretable and operable overall pre-evaluation closed loop, which significantly improves the verifiability and error controllability of fault location, and supports the optimization design, reliability verification and preventive maintenance decision-making of UAVs. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0033] Figure 1 This is a schematic diagram of the overall performance testing and evaluation system for the unmanned aerial vehicle (UAV) of the present invention.

[0034] Figure 2 This is a schematic diagram of the overall performance testing and evaluation method for unmanned aerial vehicles (UAVs) according to the present invention.

[0035] Figure 3 This is a schematic diagram illustrating the residual construction, subsystem decomposition, and coupling link allocation of the present invention. Detailed Implementation

[0036] The present invention will now be described more clearly and completely by way of a preferred embodiment in conjunction with the accompanying drawings, but this does not limit the invention to the scope of the described embodiment.

[0037] like Figure 1 As shown, the present invention provides a system for testing and evaluating the overall performance of a drone, including a data interface, a processor, and a memory.

[0038] The data interface is used to access multi-source operational data; the multi-source operational data comes from at least two of the following: flight control log link, power and energy telemetry link, control and execution feedback link, navigation and perception link, and environmental monitoring link.

[0039] The processor is used to store instructions that can be executed by the processor;

[0040] The memory is used to store instructions that can be executed by the processor, as well as data structures and configuration files required for operation, including at least: mapping tables and their mapping version numbers, parameterized scenario libraries, digital twin model parameter sets, threshold and constraint configurations, historical evaluation reports and evaluation logs.

[0041] The processor executes instructions in memory to implement the overall performance testing and evaluation method of the UAV, and outputs an overall performance evaluation report and fault location clues. When the evaluation confidence is lower than the threshold, the processor triggers at least one of regenerating the supplementary test plan and increasing the sampling frequency based on the confidence decomposition vector, thereby realizing the pre-evaluation closed loop.

[0042] like Figure 2 As shown, the present invention also provides a method for testing and evaluating the overall performance of a drone, specifically including the following steps:

[0043] S1. Collect multi-source operational data to characterize the UAV's operating status and external conditions, and assign a unified time reference to the multi-source operational data.

[0044] In one specific embodiment, step S1, in which the system assigns a unified time reference to the multi-source operational data, specifically includes:

[0045] Configure a channel identifier and sampling frequency for each data channel; record the timestamp source when the channel carries the original timestamp; when the channel does not carry a timestamp, the system generates a timestamp at the moment of reception and records the upper limit of reception jitter.

[0046] For channels with fixed transmission or processing delays, record the channel delay parameters for subsequent alignment compensation.

[0047] Multi-source operational data should include at least two of the following categories: flight status data, power and energy data, control and execution data, navigation and perception data, and environmental data, in order to meet the basic observability of the overall aircraft status and external operating conditions.

[0048] S2. Perform cross-platform semantic unified processing on multi-source runtime data to generate standardized state vectors.

[0049] In step S2, the system unifies the differences in field meanings across different platforms to generate a standardized state vector, which includes at least:

[0050] Field semantic mapping: Construct a mapping table; the mapping table should include at least the source field identifier, target field identifier, unit, coordinate system, sampling period, channel delay, and mapping version number. At runtime, the conversion is performed according to the mapping table, and the mapping version number is written to the metadata of the overall performance evaluation report, enabling cross-model benchmarking and traceability.

[0051] Unit unification: When the source field unit is inconsistent with the target unit, the conversion is completed according to the conversion relationship in the mapping table.

[0052] Coordinate System 1: When the source field coordinate system is inconsistent with the target coordinate system, the transformation is completed according to the coordinate transformation relationship in the mapping table, so that the same physical quantity maintains a consistent meaning throughout the entire process.

[0053] Asynchronous sampling alignment: The system constructs a unified time grid; performs resampling alignment on channels with different sampling periods; and performs delay compensation alignment on channels that record channel delay parameters, so that multi-channel data are aligned on a unified time grid and form a standardized state vector.

[0054] During asynchronous sampling alignment, if a target field lacks valid observations within a certain time period, the system records this time period as a missing data segment, generates a missing data marker, and registers the interpolation strategy. The missing data marker serves as one of the deduction factors in the confidence score calculation, used to reduce the contamination of residual statistics and reliability updates by interpolation errors in the missing data segment. For ease of implementation, the system can define a missing data ratio. To assess the proportion of missing samples within a window (dimensionless), and to define a deduction factor. The confidence level is adjusted in the absence of test results using the following formula:

[0055]

[0056] in To deduct weights for dimensionless values, The revised assessment confidence level, and the missing test markers and The value is recorded in the report metadata for easy review.

[0057] S3. Calculate the scene coverage index based on the parameterized scene library and standardized state vector. When the scene coverage index is less than the coverage threshold, take the predicted improvement of the evaluation confidence of the candidate supplementary test scene as the target, and generate a supplementary test plan under the condition of meeting flight safety constraints and energy budget constraints. Drive the UAV to perform supplementary test actions according to the supplementary test plan and feed back the newly added multi-source operation data.

[0058] The scenario parameters include at least several dimensions from wind speed (m / s), ambient temperature (°C), load mass (kg), maneuverability (dimensionless), and battery state of charge (dimensionless); each dimension is given a dimensional definition and feasible range in the scenario library.

[0059] The system discretizes the scene parameter space into a grid and calculates the occupancy to obtain a coverage index. (Dimensionless). For ease of implementation, the following approach can be used:

[0060]

[0061] in The total number of grid cells. The number of covered grid cells.

[0062] when When the coverage is less than the coverage threshold, the system generates a set of candidate supplementary test scenarios. Each candidate supplementary test scenario must include at least the parameters of the scenario to be supplemented, the sequence of supplementary test actions, and the data acquisition configuration. The data acquisition configuration must include at least the sampling frequency and the list of enabled channels.

[0063] The system performs flight safety constraints and energy budget constraints checks on candidate supplementary test scenarios; flight safety constraints include at least upper limits for attitude angles and lower limits for altitude; energy budget constraints include at least lower limits for battery state of charge; only those scenarios that meet the constraints are allowed to proceed to the supplementary test plan selection stage.

[0064] To ensure that the predicted improvement in evaluation confidence from candidate supplementary testing scenarios is calculable, the system calculates the predicted improvement for each candidate supplementary testing scenario s. In one implementation, It consists of two parts:

[0065] Residual variance prediction decrease Based on the simulation sensitivity and historical residual statistics of the current calibrated digital twin model at the candidate scene parameters, the decrease in residual variance after retesting is predicted, and different observations are normalized and summarized according to their importance weights.

[0066] The decrease in predicted posterior uncertainty of parameters The additional observation information introduced by the candidate scenario is regarded as an increment to the identifiability of the parameter. The parameter covariance trace or confidence interval width is used as the uncertainty measure to predict the magnitude of the uncertainty reduction after the supplementary measurement.

[0067] The predicted increase meets the following requirements:

[0068]

[0069] in These are dimensionless weighting coefficients. The system selects from the candidate set that satisfies the constraints. The largest candidate scenario generates a retest plan, and the candidate scenarios are... The predicted improvement value for assessing confidence level is written into the supplementary testing plan and used to select the best candidates for supplementary testing under the condition that the constraints are met. After the supplementary testing is completed, the newly added multi-source running data is fed back into step S1, forming a coverage-driven closed-loop supplementary testing process.

[0070] S4. Input the standardized state vector and scene parameters into the digital twin model, and perform simulation to obtain simulation output.

[0071] A digital twin model includes at least a flight dynamics sub-model, a power and energy sub-model, a control and execution sub-model, and a navigation and perception sub-model, and exchanges state variables through a coupling interface.

[0072] S5. Construct the residual vector between the measured output and the simulation output, decompose the residual vector into subsystem residuals by subsystem layer, and allocate the link residuals by coupling link layer under the condition of satisfying the residual conservation constraint.

[0073] like Figure 3 As shown, in step S5, the system constructs the residual vector between the measured output and the simulation output:

[0074]

[0075] in For the residual vector, This is the measured output vector. The output vector is the same dimension as the simulation output vector, and the units of each component are consistent. t is the time variable.

[0076] The system decomposes the residual vector into subsystem residuals according to the observation variable-subsystem attribution relationship, and further distributes them into link residuals according to the coupling link layer. The link residual distribution satisfies the residual conservation constraint.

[0077] To ensure that the link residual allocation satisfies the residual conservation constraint and is verifiable, the system represents the subsystem and coupled links as a directed graph. The node set V corresponds to the subsystem, and the edge set E corresponds to the coupling interface links. The system automatically constructs a node-link association matrix based on the coupling interface topology. The matrix rows correspond to subsystem nodes, and the columns correspond to coupling links; for any link e, if it points from node u to node v, then... Take -1 in row u, +1 in row v, and 0 for the rest.

[0078] At the current point in operation, the system obtains the coupling interface transfer matrix from the digital twin model. This is used to characterize the propagation relationship between link variables and observation residuals, thereby suppressing backward allocation that does not conform to the propagation direction during link allocation. The system solves for the link residual vector. Make it satisfy the residual conservation constraint:

[0079]

[0080] in Let be the residual vector of the subsystem. When there is inconsistency caused by observation noise or missing measurements, the system uses least squares with regularization terms to obtain a unique stable solution, and outputs the conserved constraint residual as a verifiable index in the evaluation report.

[0081] S6. Perform constrained online calibration on the parameter set of the digital twin model based on the subsystem residuals and link residuals to obtain the calibrated digital twin model and output the calibration convergence index.

[0082] In step S6, the objective function is set to reduce residual energy, and weights are assigned to the residuals corresponding to key indicators.

[0083] The constraint set includes at least the physical boundary constraints of the parameters, the closed-loop stability constraints, and the parameter rate of change constraints.

[0084] The convergence metrics output the number of iterations, the termination residual level, and the residual decline rate, and are written into the report metadata for confidence calculation and model update credibility auditing.

[0085] S7. Construct a subsystem collaboration graph based on the calibrated digital twin model and standardized state vector, calculate collaboration performance index, and perform reliability update based on collaboration performance index and link residual to obtain reliability a posteriori.

[0086] In step S7, the system constructs a subsystem collaboration graph, where nodes represent subsystems and edges represent coupling interfaces. The system calculates collaboration performance indicators, which include at least the normalized magnitude of link residuals and a measure of cross-subsystem input-output time consistency. Reliability updates are then performed using link residual statistics to obtain the posterior reliability.

[0087] S8. Calculate the evaluation confidence level based on the scene coverage index, residual statistics and calibration convergence index, and output the overall performance evaluation report containing the evaluation confidence level and the fault location clues corresponding to the link residuals; when the evaluation confidence level is lower than the confidence level threshold, trigger at least one of regenerating the supplementary test plan and increasing the sampling frequency.

[0088] In step S8, the system calculates the evaluation confidence level and outputs a system performance evaluation report and fault location clues. For ease of implementation, the confidence level can be mapped to a range of 0–1 using a logical function:

[0089]

[0090] in It is an S-shaped logic function, where z is the function's output, which is dimensionless. For coverage metrics, This is the normalized measure of residual variance. To calibrate the convergence metric, Stab is used as the stability metric for the collaboration graph. As weight, This is the bias term. The system outputs a confidence decomposition vector, which includes at least the coverage contribution, residual contribution, and calibration convergence contribution. When the assessed confidence is lower than the confidence threshold, the system triggers at least one of the following based on the confidence decomposition vector: regenerating the supplementary test plan and increasing the sampling frequency, thus achieving a self-correction closed loop.

[0091] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for testing and evaluating the overall performance of a drone, characterized in that, Specifically, the following steps are included: S1. Collect multi-source operational data to characterize the UAV's operating status and external conditions, and assign a unified time reference to the multi-source operational data; S2. Perform cross-platform semantic unified processing on multi-source operational data to generate standardized state vectors; S3. Calculate the scenario coverage index based on the parameterized scenario library and standardized state vector. When the scenario coverage index is less than the coverage threshold, take the predicted improvement of the evaluation confidence by the candidate supplementary test scenario as the target, and generate a supplementary test plan under the condition of meeting flight safety constraints and energy budget constraints. According to the supplementary testing plan, the drone was driven to perform supplementary test actions and feed back newly added multi-source operational data; S4. Input the standardized state vector and scene parameters into the digital twin model, and perform simulation to obtain simulation output; S5. Construct the residual vector between the measured output and the simulation output, decompose the residual vector into subsystem residuals by subsystem layer, and allocate the link residuals by coupling link layer under the condition of satisfying the residual conservation constraint. S6. Perform constrained online calibration on the parameter set of the digital twin model based on the subsystem residuals and link residuals to obtain the calibrated digital twin model and output the calibration convergence index. S7. Construct a subsystem collaboration graph based on the calibrated digital twin model and standardized state vector, calculate collaboration performance index, and perform reliability update based on collaboration performance index and link residual to obtain reliability posterior. S8. Calculate the evaluation confidence level based on the scene coverage index, residual statistics and calibration convergence index, and output the overall performance evaluation report containing the evaluation confidence level and the fault location clues corresponding to the link residuals; when the evaluation confidence level is lower than the confidence level threshold, trigger at least one of regenerating the supplementary test plan and increasing the sampling frequency.

2. The method for testing and evaluating the overall performance of a drone according to claim 1, characterized in that, The multi-source operational data includes at least two of the following categories: flight status data, power and energy data, control and execution data, navigation and perception data, and environmental data.

3. The method for testing and evaluating the overall performance of a drone according to claim 1, characterized in that, The cross-platform semantic unified processing includes at least field semantic mapping; the field semantic mapping is implemented through a mapping table, which includes at least source field identifier, target field identifier, unit, coordinate system, sampling period, channel delay and mapping version number, and the mapping version number is written into the metadata of the whole machine performance evaluation report.

4. The method for testing and evaluating the overall performance of a drone according to claim 3, characterized in that, The cross-platform semantic unification processing also includes dimension unification, coordinate system unification, and asynchronous sampling alignment; the asynchronous sampling alignment includes resampling alignment and delay compensation alignment based on a unified time grid to form a standardized state vector; and generates missing data markers for missing data segments, using the missing data markers as one of the deduction factors in the evaluation confidence calculation to suppress the impact of missing data segment interpolation errors on the evaluation conclusion.

5. The method for testing and evaluating the overall performance of a drone according to claim 1, characterized in that, The predicted improvement in the assessment confidence level is jointly determined by the predicted decrease in residual variance and the predicted decrease in posterior uncertainty of calibration parameters for the candidate supplementary test scenario; the flight safety constraints include at least an upper limit for attitude angle and a lower limit for altitude, and the energy budget constraints include at least a lower limit for battery state of charge. The process of allocating link residuals according to the coupled link layer includes: generating a node-link association matrix based on the coupled interface topology of the digital twin model, and solving for the link residuals that satisfy the residual conservation constraints under the condition of calling the coupled interface transfer matrix of the current working point, so that the link residual allocation result is consistent with the subsystem residuals and can be verified.

6. A system for testing and evaluating the overall performance of an unmanned aerial vehicle (UAV), characterized in that, include: The data interface is used to access multi-source runtime data; processor; Memory is used to store instructions that can be executed by the processor; When the processor executes instructions, it is used to implement the UAV overall performance testing and evaluation method as described in any one of claims 1-5, and output an overall performance evaluation report and fault location clues.

7. The unmanned aerial vehicle (UAV) overall performance testing and evaluation system according to claim 6, characterized in that, The memory stores the mapping table and the mapping version number. When the processor performs field semantic mapping, it writes the mapping version number into the metadata of the whole machine performance evaluation report to support cross-model benchmarking and traceability.

8. The unmanned aerial vehicle (UAV) overall performance testing and evaluation system according to claim 6, characterized in that, The supplementary testing plan includes at least the parameters of the scenario to be supplemented, the sequence of supplementary testing actions, and the data acquisition configuration. The data acquisition configuration includes at least the sampling frequency and the channel activation list. The supplementary testing plan also includes the ranking value of the predicted improvement amount of the evaluation confidence corresponding to the candidate supplementary testing scenarios, which is used to select the best supplementary testing scenarios under the condition that the constraints are met.

9. The unmanned aerial vehicle (UAV) overall performance testing and evaluation system according to claim 6, characterized in that, When allocating link residuals, the processor calls the coupling interface of the digital twin model at the current working point to transfer the matrix, and makes the sum of the inbound link residuals and outbound link residuals related to a certain subsystem equal to the residual of that subsystem, thus satisfying the residual conservation constraint.

10. The unmanned aerial vehicle (UAV) overall performance testing and evaluation system according to claim 6, characterized in that, The overall performance evaluation report also includes a confidence decomposition vector, which includes at least coverage contribution, residual contribution, and calibration convergence contribution. When the evaluation confidence is lower than the confidence threshold, the processor selects at least one of triggering the regeneration of the supplementary test plan and increasing the sampling frequency based on the confidence decomposition vector.

Images