Weapon equipment data consistency verification system based on virtual-real combination

Abstract

The application discloses a weapon equipment data consistency verification system based on virtual-real combination and relates to the technical field of weapon equipment test identification.The system comprises a space-time alignment module, a virtual-real registration module, a consistency measurement module and a hierarchical judgment module;the virtual-real data of different sources, different time and different dimensions are subjected to space-time alignment, semantic mapping and time delay registration;consistency components are calculated from five dimensions of amplitude, trend, distribution, dynamic and frequency domain;an integrated consistency index and a lower confidence limit are obtained through combination and weighting;and four-level judgment of consistency, acceptable deviation, significant inconsistency and failure is made accordingly.The application solves the problems in the prior art that virtual-real data is difficult to directly compare, inconsistent sources cannot be distinguished, there is lack of quantitative confidence and grading, online verification and anti-tampering are difficult and the like, and significantly improves the reliability and traceability of virtual-real data consistency verification.

Description

Technical Field

[0001] This invention belongs to the field of weapon and equipment testing and evaluation technology, specifically involving a weapon and equipment data consistency verification system based on virtual and real integration. Background Technology

[0002] In the research, development, testing and evaluation of weapons and equipment, semi-physical simulation and digital twin technologies are widely used to combine virtual simulation models with physical equipment and measured data in order to reduce testing costs and expand testing boundaries.

[0003] The effectiveness of combining virtual and real data relies on the consistency between virtual simulation data and physical test data. Therefore, data consistency verification is necessary to assess the fidelity of the digital twin model, verify the status of the physical equipment, and support mutual verification decisions that use virtual data to supplement real data and real data to verify virtual data. Existing data consistency verification technologies face challenges due to issues such as heterogeneous data sources, asynchronous sampling, inconsistent dimensions and coordinates, and inconsistent field semantics, making direct point-by-point comparison difficult. Furthermore, existing methods often employ single thresholds or point-by-point error criteria, failing to distinguish between different sources of inconsistency, such as insufficient model fidelity, actual equipment anomalies, data quality issues (latency, missing data points, noise, drift), and human tampering, leading to high false positive rates. The lack of quantitative confidence assessment and hierarchical judgment of consistency makes it difficult to support reliable mutual verification decisions. Additionally, existing systems face difficulties in online consistency verification, inconsistency localization, and tamper-proofing of verification processes and conclusions under high-frequency, large-volume, and multi-channel data conditions.

[0004] Given the above shortcomings, there is an urgent need for a virtual-real integrated weapon and equipment data consistency verification system. Summary of the Invention

[0005] This invention addresses the aforementioned problems of existing technologies by providing a weapon and equipment data consistency verification system based on a combination of virtual and real data. It aims to resolve core technical deficiencies such as the difficulty in directly comparing virtual and real data, the inability to distinguish the sources of inconsistencies, the lack of quantitative confidence and classification, and the challenges of online verification and tamper prevention. The technical solution is as follows:

[0006] The weapon and equipment data consistency verification system based on virtual and real integration includes a spatiotemporal alignment module, a virtual and real registration module, a consistency measurement module, and a hierarchical judgment module.

[0007] The spatiotemporal alignment module is used to access virtual simulation data output as a digital twin model and physical test data as physical equipment measurement or telemetry data, and to perform field semantic mapping, time reference unification, resampling, and dimension and coordinate normalization on them to generate virtual and physical data pairs under a unified spatiotemporal reference.

[0008] The virtual-real registration module is used to extract comparable state variable feature sequences from virtual-real data pairs, estimate the time delay between virtual and real data based on cross-correlation, and align them accordingly to obtain registered virtual-real feature sequence pairs.

[0009] The consistency measurement module is used to calculate the consistency components of amplitude, trend, distribution, dynamics and frequency domain from five dimensions, based on the registered virtual and real feature sequence pairs.

[0010] The graded judgment module is used to combine and weight the five consistency components to obtain a comprehensive consistency index, calculate its confidence lower bound, and output four-level judgment results of consistency, acceptable deviation, significant inconsistency and failure based on the graded threshold.

[0011] Furthermore, the system also includes a source tracing and localization module and an evidence storage and feedback module; the source tracing and localization module is used to attribute inconsistencies to insufficient model fidelity, actual equipment anomalies, data quality problems, or human tampering, and to locate the time and channel of such inconsistencies; the evidence storage and feedback module is used to construct Merkle trees and chain-based evidence storage for virtual and real data blocks and verification conclusions, detect and locate tampering, and update the classification threshold and measurement weights online based on the measured false alarm rate to form a closed loop.

[0012] Furthermore, the spatiotemporal alignment module generates a unified timescale with a unified sampling interval, linearly interpolates and resamples the virtual simulation data and physical test data to the unified timescale, and performs z-score normalization on the resampling results, thereby eliminating the differences between virtual and real data in terms of time reference, sampling rate and dimensions, and obtaining virtual and real data pairs under a unified spatiotemporal reference.

[0013] Furthermore, the virtual-real registration module calculates the normalized cross-correlation between the normalized virtual feature sequence and the physical feature sequence, takes the delay that makes the normalized cross-correlation reach its maximum value as the estimated delay between the virtual and physical data, and uses it to shift and align, thereby compensating for the transmission and processing delay between the virtual and physical data, and obtaining the registered virtual-real feature sequence pair.

[0014] Furthermore, the consistency measurement module measures the consistency of the registered virtual and real feature sequences from five dimensions: amplitude, trend, distribution, dynamics, and frequency domain. The normalized root mean square error (RMSE) characterizes the degree of amplitude deviation, thus determining the amplitude consistency component. The Pearson correlation coefficient characterizes the correlation of the changing trend, thus determining the trend consistency component. The Kolmogorov-Smirnov statistic of the empirical cumulative distribution function characterizes the probability distribution difference, thus determining the distribution consistency component. The dynamic time warping distance characterizes the morphological difference of the time series waveform, thus determining the dynamic consistency component. The spectral coherence function characterizes the consistency of the frequency domain structure, thus determining the frequency domain consistency component. All five consistency components are normalized to the [0, 1] interval; a larger value indicates higher consistency between the virtual and real data in that dimension.

[0015] Furthermore, the grading judgment module adaptively determines the weight of each consistency component based on its dispersion across multiple evaluation windows using the entropy weight method. It then weights and sums the five consistency components to obtain a comprehensive consistency index. Based on the sample standard deviation of the comprehensive consistency index and a given confidence level, it calculates the half-width of the confidence interval, thereby obtaining the lower confidence bound of the comprehensive consistency index. The comprehensive consistency index and the lower confidence bound are as follows: ; In the formula, As a comprehensive consistency index, For the first Entropy weight of a consistent component For the first A consistent component The confidence interval is half-width. To establish a lower bound for confidence.

[0016] Based on the grading threshold , , , The confidence lower bound domain [0, 1] is divided into four levels of consistency judgment results: failure, significant inconsistency, acceptable deviation, and consistency.

[0017] Furthermore, when the source tracing and positioning module determines a significant inconsistency or failure, it calculates the discrimination scores for four types of inconsistencies—insufficient model fidelity, actual equipment anomalies, data quality problems, and human tampering—based on the system deviation and significance test of the virtual and real residuals, the mutation detection and mutation positioning of the residual cumulative sum, and the data missing rate. The module takes the largest discrimination score as the attribution result and uses the time when the residual cumulative sum exceeds the control limit and the corresponding channel as the spatiotemporal positioning result of the inconsistency, thereby distinguishing the source of inconsistency and locating the time and channel of its occurrence.

[0018] Furthermore, the evidence storage and feedback module calculates cryptographic hashes for each virtual and real data block participating in the verification, constructs a Merkle tree root by concatenating hashes layer by layer, and concatenates the hashes of each data block with the corresponding verification conclusions in chronological order to form a chain of evidence storage; by recalculating the chain of evidence storage values ​​and comparing them with the original values, tampering is detected, and the tampered data block is located via the Merkle path; at the same time, the hierarchical threshold and the measurement weights of each consistency component are updated online based on the difference between the measured false alarm rate and the target false alarm rate, forming a closed loop of access alignment, feature registration, multi-dimensional measurement, comprehensive evaluation, source tracing and location, evidence storage feedback and parameter update.

[0019] Compared with the prior art, the beneficial effects of this invention are: This invention unifies the time base, resamples, normalizes, and performs cross-correlation-based time delay registration on heterogeneous, asynchronous, and dimensionless virtual and real data. It calculates and weights consistency components from five dimensions: amplitude, trend, distribution, dynamics, and frequency domain. Through a comprehensive consistency index and its confidence lower bound, and a four-level classification, it traces inconsistencies and locates them spatiotemporally based on system bias, residual accumulation, mutation detection, and missing rate. This solves the problem of distinguishing between insufficient model fidelity, real equipment anomalies, data quality issues, and human tampering, and overcomes the difficulty of directly comparing virtual and real data. It achieves tamper-proof and traceability of the verification process and conclusions. This invention significantly improves the credibility, interpretability, and traceability of weapon and equipment data consistency verification under virtual-real combined conditions. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the weapon and equipment data consistency verification system based on the combination of virtual and real data according to the present invention. Figure 2 This is a schematic diagram illustrating the principle of virtual-real data consistency verification in this invention. Figure 3 This is a schematic diagram of the comparison between virtual and real hyperbolas and the multidimensional consistency classification of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some embodiments of this invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the protection scope of this invention.

[0022] like Figure 1 As shown, the weapon equipment data consistency verification system based on virtual-real integration of the present invention includes a spatiotemporal alignment module, a virtual-real registration module, a consistency measurement module, a hierarchical judgment module, a source tracing and positioning module, and an evidence storage and feedback module connected in sequence.

[0023] The evidence storage and feedback module feeds back the updated classification threshold and measurement weights to the classification judgment module. The classification judgment module combines and assigns weights to each consistency component according to the updated measurement weights and performs a four-level consistency judgment according to the updated classification thresholds, forming a closed-loop parameter update.

[0024] like Figure 2 As shown, the system receives two data streams: virtual simulation data and physical test data. After spatiotemporal alignment and virtual-real registration, consistency components are calculated from five dimensions: amplitude, trend, distribution, dynamics, and frequency domain. The comprehensive consistency index is obtained by combining and weighting the components, and the confidence lower bound is calculated. Based on this, a four-level classification judgment is made, and at the same time, inconsistency tracing and location and tamper-proof evidence storage are performed.

[0025] The following example illustrates the consistency verification of virtual and real data for the missile's pitch angular velocity state quantity during a semi-physical test of a certain type of missile. In this embodiment, the sampling rate of the physical test data is 1000Hz, the sampling rate of the virtual digital twin output is 500Hz, the verification window length is 2s, the unified sampling interval Δt=1ms, and the unified sequence length N=2000.

[0026] The spatiotemporal alignment module performs field semantic mapping between virtual simulation data and physical test data, and generates a unified timescale with a unified sampling interval Δt. The unified timescale is: .

[0027] The virtual simulation data and the physical test data are then linearly interpolated and resampled to the unified time scale. The resampled value at point is: .

[0028] The resampled sequence was then normalized using z-scores, with the normalized values ​​being: In the formula, For the k-th unified time scale, The starting time is Δt, and the uniform sampling interval is Δt. , For adjacent original sampling times, , For the corresponding original sampled values; Normalized value The original value, These are the mean and standard deviation of the sequence, respectively.

[0029] The virtual-real registration module performs normalization on the virtual feature sequence. With physical feature sequence Calculate the normalized cross-correlation. The normalized cross-correlation is: .

[0030] The time delay that maximizes the normalized cross-correlation is taken as the estimated time delay, and the estimated time delay is: In the formula, For the normalized cross-correlation with a time delay of τ, , These are the normalized virtual and physical feature sequences, respectively. To estimate the time delay.

[0031] In this embodiment, after the virtual data is linearly interpolated and resampled to 1000Hz and aligned with the physical data to a unified time scale through the aforementioned spatiotemporal alignment, and then normalized by z-score, the virtual real-time delay is estimated based on the normalized cross-correlation. =8ms, based on which the virtual and real feature sequence pairs are translated and aligned to obtain the registered virtual and real feature sequence pairs.

[0032] The consistency measurement module calculates the consistency components from five dimensions: amplitude, trend, distribution, dynamics, and frequency domain, based on the registered virtual and real feature sequence pairs.

[0033] The magnitude dimension is characterized by the normalized root mean square error (RMSE), which is: The amplitude uniformity component is: In the formula, For normalized root mean square error, These are the maximum and minimum values ​​of the physical sequence, respectively, and N is the sequence length; For amplitude uniformity components, This represents the value of the virtual feature sequence at the k-th sampling point; This indicates that the sequence of physical features has been shifted after the estimated time delay. The value at the kth sampling point after that.

[0034] In this embodiment ≈0.068, therefore the amplitude uniformity component =1 0.068 ≈ 0.93.

[0035] The trend dimension uses the Pearson correlation coefficient to characterize the degree of correlation of changing trends. The Pearson correlation coefficient is: ; The trend consistency component is: In the formula, The Pearson correlation coefficient is used. This is a trend consistency component.

[0036] In this embodiment, the Pearson correlation coefficient =0.96, therefore the trend consistency component =(0.96+1) / 2=0.98.

[0037] The distribution dimension is characterized by the Kolmogorov-Smirnov statistic of the empirical cumulative distribution function, which represents the difference in probability distributions. The Kolmogorov-Smirnov statistic is: The distributional consistency component is: In the formula, For Kolmogorov-Smirnov statistics, , Let be the empirical cumulative distribution functions of the virtual and physical sequences, respectively. This is a component with consistent distribution.

[0038] In this embodiment =0.09, therefore the distribution is consistent. =1 0.09 = 0.91.

[0039] The dynamic dimension is characterized by the dynamic time warp distance, which represents the morphological differences of time series waveforms. The dynamic time warp distance is: The dynamic consistency component is: In the formula, DTW is the dynamic time warping distance, and W is the warped path. The scaling factor. This is a dynamically consistent component.

[0040] In this embodiment, the dynamic time warp distance (DTW) is 180, and the scaling factor is taken. =0.6, therefore the dynamic consistency component is: .

[0041] The frequency domain dimension is characterized by the spectral coherence function, which represents the degree of consistency of the frequency domain structure. The spectral coherence function is: ; The frequency domain uniformity component is: In the formula, Let f be the spectral coherence function at frequency f. The virtual-real cross-power spectrum, , These represent the virtual and physical self-power spectra, respectively, where F is the number of frequency points. For frequency domain consistent components.

[0042] Average spectral coherence in this embodiment =0.90. The obtained consistency components and their entropy weights are shown in Table 1: Table 1. Consistency components and their entropy weights in this embodiment.

[0043] The hierarchical determination module uses the entropy weight method to determine the weight of each consistency component, and the entropy value of each consistency component is: The entropy weights of each consistency component are: In the formula, For the first The entropy value of each consistent component. For the first The weight of each component in the k-th evaluation window, i.e. K is the number of evaluation windows; For the first The entropy weight of a consistent component.

[0044] The overall consistency index is obtained by weighted summation of the five consistency components. The overall consistency index is: In the formula, As a comprehensive consistency index, For the first A consistent component.

[0045] In this embodiment, the overall consistency index calculated using the entropy weight obtained from Table 1 is:

[0046] The confidence interval half-width is calculated based on the sample standard deviation of the consensus index and a given confidence level. The confidence interval half-width is: The lower confidence bound is: In the formula, The confidence interval is half-width. For standard normal quantiles, Here, denoted as the sample standard deviation of the overall consistency index, and M represents the number of evaluation windows. To establish the lower bound for confidence.

[0047] In this embodiment, the evaluation window number M=20 and the sample standard deviation are set to 20. =0.05, 95% confidence level ( =1.96), then the half-width of the confidence interval and the lower bound of the confidence interval are: ; .

[0048] Take the grading threshold =0.6、 =0.8、 =0.9, due to the lower confidence bound =0.89 falls within the interval [0.8, 0.9), and the system determines that the consistency between virtual and real data is at an acceptable deviation level, since it has not reached the consistency threshold. This triggers the source tracing and location module.

[0049] The source tracing and localization module performs attribution and localization based on the systematic bias and significance of virtual and real residuals, mutation detection and mutation localization of residual cumulative sum, and missing rate. The systematic bias of virtual and real residuals is: The significance statistic for systematic bias is: In the formula, This refers to the systematic bias of the virtual and real residuals. This is the significance statistic for the deviation. This represents the standard deviation of the residuals.

[0050] In this embodiment, the systematic bias and its significance statistic are: ; The significance statistic of the bias is less than the critical value of 1.96, indicating that there is no significant systematic bias, thus excluding inconsistencies caused by systematic bias in model fidelity.

[0051] The cumulative sum of residuals is: .

[0052] The mutation location of the residual cumulative sum is: In the formula, Let k be the residual at point k. The mean of the residuals, For reference purposes, when Alarm triggered when control time limit is exceeded; This is the mutation location.

[0053] In this embodiment, a reference value is used. =0.25, control limit hc=2.5, the residual suddenly increases at t≈1.5s, causing the cumulative sum to exceed the control limit and triggering an alarm, thus locating the point of change. =1500 (t≈1.5s).

[0054] The data missing rate is: In the formula, The number of missing points; the missing rate in this embodiment. =30 / 2000=0.015.

[0055] The attribution result is the one with the highest score among the four categories of inconsistency discrimination. The attribution result is as follows: In the formula, As an attribution result, The score is the discrimination score for the j-th type of inconsistency.

[0056] The attribution rules are as follows: when the trend correlation is high and the bias is significant, it is attributed to insufficient model fidelity; when the residual accumulation and abrupt change at a certain moment meet the physical constraints, it is attributed to real equipment anomalies; when the estimated latency or missing rate is significant, it is attributed to data quality issues; when the residual changes abruptly and does not match the evidence hash, it is attributed to human tampering.

[0057] In this embodiment, the missing rate is only 1.5% and the estimation delay is... =8ms is relatively small, ruling out data quality issues; the data has been verified by the evidence hash to have not been tampered with, ruling out human intervention; the residual mutation conforms to the physical constraints of projectile dynamics, therefore it is attributed to a genuine equipment anomaly, located at t≈1.5s, such as Figure 3 As shown, this is the measured mutation that occurred at that moment.

[0058] The evidence storage and feedback module divides the virtual and real data involved in the verification into blocks and calculates their hashes. The b-th data block... The hash is: The Merkle tree parent node is calculated by piecing together the left and right child nodes layer by layer. The Merkle tree parent node is: .

[0059] The hashes of each data block and their corresponding verification conclusions are concatenated in chronological order to form a chained evidence value. The chained evidence value at time t is: . For the b-th data block The hash of H, where H is the cryptographic hash function. The parent node of the Merkle tree. , For its left and right child nodes, calculate layer by layer down to the Merkle root; Let t be the chained evidence value. The previous value of the evidence. For the hash of the current data block, This is the current verification conclusion.

[0060] In this embodiment, a SHA-256 hash is calculated for each data block, and the Merkle tree parent node is calculated layer by layer until the root of the Merkle tree. The hash of each block and the corresponding verification conclusion are included in the chain-stored evidence value, thereby binding the verification conclusion with the original data. When the chain-stored evidence value is recalculated for the evidence data and compared with the original evidence value, if a data block is tampered with, the recalculated value is inconsistent with the original value, thereby detecting the tampering and locating the tampered data block through the Merkle path.

[0061] The evidence storage and feedback module updates the classification threshold online based on the measured false alarm rate. The updated classification threshold is as follows: In the formula, The updated grading threshold, To update the step size, , These are the measured false alarm rate and the target false alarm rate, respectively.

[0062] In this embodiment, the measured false alarm rate during the statistical period is set. =0.04, Target False Alarm Rate =0.02, Update step size If the value is 0.1, then the consistency threshold is updated online as follows: That is, when the measured false alarm rate is higher than the target value, the threshold is appropriately relaxed to suppress false alarms; at the same time, the entropy weight of each consistency component is re-estimated based on the sliding window, so that the measurement weight is updated adaptively with the data characteristics, thereby forming a closed-loop verification of access alignment, feature registration, multi-dimensional measurement, comprehensive evaluation, source tracing and location, evidence storage feedback and parameter update.

[0063] This invention system achieves reliable, interpretable, and traceable verification of weapon and equipment data consistency under virtual-real combination conditions through spatiotemporal alignment and registration of virtual and real data, multidimensional consistency measurement of amplitude / trend / distribution / dynamic / frequency domain, four-level classification judgment of comprehensive consistency index and confidence lower bound, source tracing and spatiotemporal positioning of inconsistencies, and the coordinated cooperation of Merkle tree and chain hash storage.

[0064] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A weapon and equipment data consistency verification system based on virtual-real integration, characterized in that, The system is used to verify the consistency between virtual simulation data and physical test data in the semi-physical and digital twin tests of weapon equipment. The system includes a spatiotemporal alignment module, a virtual-physical registration module, a consistency measurement module, and a hierarchical judgment module. The spatiotemporal alignment module is used to access virtual simulation data and physical test data. The virtual simulation data is the output data of the digital twin model, and the physical test data is the measurement or telemetry data of physical equipment. The module performs field semantic mapping, time reference unification, resampling, and dimension and coordinate normalization on the virtual simulation data and physical test data to generate virtual and real data pairs under a unified spatiotemporal reference. The virtual-real registration module is used to extract comparable state variable feature sequences from the virtual-real data pairs, estimate the time delay between virtual and real data based on cross-correlation, and align them accordingly to obtain registered virtual-real feature sequence pairs. The consistency measurement module is used to calculate the amplitude consistency component, trend consistency component, distribution consistency component, dynamic consistency component and frequency domain consistency component from five dimensions: amplitude, trend, distribution, dynamic and frequency domain, based on the registered virtual and real feature sequence pairs. The hierarchical judgment module is used to combine and weight the five consistency components to obtain a comprehensive consistency index, calculate the confidence lower bound of the comprehensive consistency index, and output a four-level consistency judgment result of consistency, acceptable deviation, significant inconsistency and failure based on the hierarchical threshold.

2. The system according to claim 1, characterized in that, The system also includes a source tracing and location module and an evidence storage and feedback module; The source tracing and location module is used to attribute inconsistencies to insufficient model fidelity, real equipment anomalies, data quality problems, or human tampering when a significant inconsistency or failure is determined, based on the system deviation and significance of virtual and real residuals, residual accumulation and mutation detection and missing rate, and to locate the time and channel where the inconsistency occurred. The evidence storage and feedback module is used to calculate hashes for the virtual and real data blocks participating in the verification and the verification conclusion, and construct Merkle trees and chain-based evidence storage. It detects tampering and locates the tampered data blocks by recalculating the hash chain, and updates the hierarchical thresholds and combined weighted metrics online according to the measured false alarm rate, forming a closed loop of access alignment, feature registration, multi-dimensional measurement, comprehensive evaluation, source tracing and location, evidence storage feedback and parameter update.

3. The system according to claim 2, characterized in that, The virtual-real registration module performs linear interpolation resampling on the virtual-real data pair using a unified time scale, performs z-score normalization on the resampled virtual feature sequence and the real feature sequence, estimates the time delay between the virtual feature sequence and the real feature sequence based on the normalized cross-correlation, and takes the time delay that makes the normalized cross-correlation reach its maximum value as the estimated time delay, and then shifts and aligns them to obtain the registered virtual-real feature sequence pair.

4. The system according to claim 3, characterized in that, The consistency measurement module calculates the amplitude consistency component based on the normalized root mean square error of the registered virtual and real feature sequence pairs, the trend consistency component based on the Pearson correlation coefficient between the virtual and real feature sequences, and the distribution consistency component based on the Kolmogorov-Smirnov statistic of the empirical cumulative distribution function of the virtual and real feature sequences.

5. The system according to claim 4, characterized in that, The consistency measurement module calculates the dynamic consistency component based on the dynamic time warping distance between the virtual feature sequence and the physical feature sequence, and calculates the frequency domain consistency component based on the mean of the spectral coherence function of the virtual feature sequence and the physical feature sequence at each frequency point.

6. The system according to claim 5, characterized in that, The classification judgment module uses the entropy weight method to determine the weights of the amplitude consistency component, trend consistency component, distribution consistency component, dynamic consistency component and frequency domain consistency component, and calculates the comprehensive consistency index by weighted summation of the five consistency components. Based on the sample standard deviation of the comprehensive consistency index and the given confidence level, the confidence interval half-width and confidence lower bound are calculated. The overall consistency index and confidence lower bound are as follows: ; In the formula, As a comprehensive consistency index, For the first Entropy weight of a consistent component For the first A consistent component The confidence interval is half-width. To establish a lower bound; Based on the grading threshold , , , The confidence lower bound domain [0, 1] is divided into four levels of consistency judgment results: failure, significant inconsistency, acceptable deviation, and consistency.

7. The system according to claim 6, characterized in that, The source tracing and localization module calculates the discrimination scores for four types of inconsistencies based on the system deviation and significance of virtual and real residuals, mutation detection and mutation localization of residual cumulative sum, and data missing rate. The module takes the largest discrimination score as the attribution result and uses the time when the residual cumulative sum exceeds the control limit and the corresponding channel as the spatiotemporal localization result of the inconsistency.

8. The system according to claim 7, characterized in that, The evidence storage and feedback module calculates cryptographic hashes for each virtual and real data block participating in the verification, concatenates the hashes layer by layer to obtain the Merkle root, and concatenates the hashes of each data block with the corresponding verification conclusions in chronological order to form a chain-like evidence storage value.

9. The system according to claim 8, characterized in that, The evidence storage and feedback module recalculates the chained evidence storage value and compares it with the original chained evidence storage value. If the two are inconsistent, it determines that the data has been tampered with and locates the tampered data block through the Merkel path. The module also updates the classification threshold online based on the difference between the measured false alarm rate and the target false alarm rate and the update step size. The classification threshold is relaxed when the measured false alarm rate is higher than the target false alarm rate, and tightened otherwise.

10. The system according to claim 9, characterized in that, The hierarchical determination module is also used to make a virtual-to-real mutual verification decision based on the confidence lower bound: when the confidence lower bound is not less than the consistency threshold. And the confidence level is not lower than the set value, that is, it satisfies: When the data is consistent with the actual data, it is allowed to use virtual simulation data to replace or supplement physical test data; When the confidence lower bound is in any other interval, it is determined to be an acceptable deviation, a significant inconsistency, or a failure, respectively, and the source tracing and localization module is triggered. The confidence level is determined by the confidence interval half-width and the overall consistency index.

Images