Multi-dimensional mapping detection method and system for transient recording precision of fault indicator

Abstract

The invention relates to the technical field of electrical performance testing, particularly discloses a multi-dimensional mapping detection method and system for transient recording precision of a fault indicator, and aims to solve the problems that in transient recording precision detection of a traditional fault indicator, the evaluation dimension is single, deep physical reasons cannot be revealed, and a test signal is disjointed from reality. The method comprises the following steps: constructing parameterized virtual behavior twin bodies, identifying and extracting sensitive parameters, generating a dynamic precision pressure test excitation sequence, and synchronously inputting the dynamic precision pressure test excitation sequence to a to-be-tested device to perform double-ring trajectory mapping with a virtual body to obtain trajectory deviation characteristics; by analyzing the mapping relation between the track deviation characteristics and the sensitive parameters, positioning specific physical parameters causing deviation and generating a diagnosis result; finally, virtual body parameters are updated according to a diagnosis result, a calibrated virtual equipment model is generated, the model can predict equipment performance degradation, monitor an online state and simulate behaviors under specific power grid disturbance, and the value boundary of detection work is expanded.

Description

Technical Field

[0001] This invention belongs to the field of electrical performance testing technology, and relates to a multi-dimensional mapping detection method and system for the transient recording accuracy of fault indicators. Background Technology

[0002] Fault indicators are important monitoring devices widely used in power distribution networks. One of their core functions is to accurately and quickly capture and record the transient waveforms of current or voltage at the moment of a fault, i.e., transient waveform recording, when a short circuit or ground fault occurs in the power grid. The accuracy of the waveform recording data directly affects the accuracy of subsequent fault nature judgment, fault location, and accident analysis. Therefore, it is crucial to effectively test the transient waveform recording accuracy of fault indicators.

[0003] Currently, the common technical approach for testing the transient waveform recording accuracy of fault indicators is to use a standard signal generator in a laboratory environment to inject one or more preset transient test signals simulating typical faults into the fault indicator under test. Then, the waveform data recorded by the fault indicator under test is compared with its corresponding standard test signal waveform. By calculating the errors in one or more key indicators such as amplitude, phase, and response time, the accuracy of the waveform recording is evaluated to determine whether it meets the requirements of relevant standards.

[0004] However, the aforementioned existing technical solutions have significant drawbacks. First, their evaluation dimensions are relatively singular and isolated, focusing only on the errors of a few discrete indicators. This fails to comprehensively reflect the overall performance of the recorded data in terms of waveform fidelity, accuracy of response to complex frequency components, and dynamic timing stability, making misjudgments likely. Second, when out-of-tolerance accuracy is detected, this solution only presents the error result and cannot effectively reveal the underlying physical causes of the error, hindering equipment fault diagnosis and targeted improvements. Finally, the use of fixed, idealized test signals for detection is disconnected from the complexity and diversity of actual transient processes in the power grid, limiting the practical reference value of the detection results. Summary of the Invention

[0005] In view of this, in order to solve the problems mentioned in the background technology, a multi-dimensional mapping detection method and system for the transient recording accuracy of fault indicators is proposed.

[0006] The objective of this invention can be achieved through the following technical solution: The first aspect of this invention provides a multi-dimensional mapping detection method for the transient waveform recording accuracy of a fault indicator, comprising: S1, acquiring the principle model and nominal parameter set of the fault indicator under test, and constructing a parameterized virtual behavior twin based on the principle model and nominal parameter set.

[0007] S2. Identify and extract the most sensitive parameters that respond to changes in external stimuli from the parameterized virtual behavioral twin.

[0008] S3. Generate a dynamic precision stress test excitation sequence based on sensitive parameters.

[0009] S4. Synchronously input the dynamic precision pressure test excitation sequence to the fault indicator under test and the parameterized virtual behavior twin, and simultaneously collect the first response sequence of the fault indicator under test and the second response sequence of the parameterized virtual behavior twin.

[0010] S5. Based on the first response sequence and the second response sequence, perform double-loop trajectory mapping to obtain trajectory deviation characteristics.

[0011] S6. Obtain the mapping relationship between trajectory deviation features and sensitive parameters, analyze the trajectory deviation features based on the mapping relationship, locate the target sensitive parameters that cause the deviation, and generate diagnostic results containing parameter offsets.

[0012] S7. Based on the parameter offset in the diagnostic results, update the parameters of the parameterized virtual behavior twin and generate the calibrated virtual device model.

[0013] The second aspect of the present invention provides a multi-dimensional mapping detection system for the transient recording accuracy of a fault indicator, comprising: a virtual behavioral twin construction module, which acquires the principle model and nominal parameter set of the fault indicator under test, and constructs a parameterized virtual behavioral twin based on the principle model and nominal parameter set.

[0014] The sensitive parameter identification and extraction module identifies and extracts the most sensitive parameters that respond to changes in external stimuli from the parameterized virtual behavior twin.

[0015] The excitation sequence generation module generates dynamic precision stress test excitation sequences based on sensitive parameters.

[0016] The data acquisition and synchronization module synchronously inputs the dynamic precision stress test excitation sequence to the fault indicator under test and the parameterized virtual behavior twin, and synchronously acquires the first response sequence of the fault indicator under test and the second response sequence of the parameterized virtual behavior twin.

[0017] The trajectory mapping analysis module performs double-loop trajectory mapping based on the first response sequence and the second response sequence to obtain trajectory deviation characteristics.

[0018] The diagnostic result generation module obtains the mapping relationship between trajectory deviation features and sensitive parameters, analyzes the trajectory deviation features based on the mapping relationship, locates the target sensitive parameters that cause the deviation, and generates diagnostic results containing parameter offsets.

[0019] The virtual device model calibration module updates the parameters of the parameterized virtual behavior twin based on the parameter offset in the diagnostic results, and generates a calibrated virtual device model.

[0020] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention achieves a qualitative leap from appearance evaluation to physical root cause diagnosis. Traditional precision testing can only conclude whether it is qualified or not, while the present invention, by constructing a virtual behavioral twin and performing double-loop trajectory mapping, can accurately trace the abstract waveform recording precision deviation back to the specific physical parameter offset that caused the deviation, such as insufficient sensor bandwidth or sampling clock jitter, providing unprecedented depth of insight for equipment optimization and fault diagnosis.

[0021] (2) This invention transforms a one-time testing process into a dynamic, high-value modeling process. The final output of this method is not only an accuracy assessment report, but more importantly, a calibrated virtual device model that is precisely calibrated and can highly reproduce the real dynamic behavior of the device under test. This model itself is a digital twin of the device and can be used for subsequent performance degradation prediction, online operating status monitoring, and behavior simulation under specific power grid disturbances, thus expanding the value boundary of testing work.

[0022] (3) This invention constructs an intelligent adaptive closed-loop detection system. By generating dynamic stress test stimuli based on the sensitive parameters of the equipment and optimizing the model based on the diagnostic results, the entire detection process exhibits a high degree of intelligence and adaptability. It can perform targeted testing on the weak points of different equipment and ensure the accuracy of the diagnostic results through closed-loop verification. Thus, while improving the depth of detection, it also ensures the reliability and objectivity of the evaluation conclusions. The overall effect far exceeds the existing separate and open-loop detection methods. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0025] Figure 2 This is a schematic diagram of the system structure connection of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figure 1 The first aspect of the present invention provides a multi-dimensional mapping detection method for the transient recording accuracy of a fault indicator, comprising: S1, acquiring the principle model and nominal parameter set of the fault indicator under test, and constructing a parameterized virtual behavior twin based on the principle model and nominal parameter set.

[0028] Specifically, the engineering objective of constructing a parameterized virtual behavioral twin based on the principle model and nominal parameter set in this embodiment is to transform the static "design blueprints" of the physical device into a high-fidelity "white-box" mathematical computing architecture capable of running high-frequency transient signals in a computer. This model must not only present an ideal state but also reproduce the non-ideal physical characteristics of real hardware. The process begins by structurally decomposing the principle model of the fault indicator. According to the processing chain of the recorded signal, it is sequentially divided into: a sensor equivalent module (e.g., an equivalent RLC network of a Rogowski coil), an analog front-end conditioning module (e.g., a circuit network containing operational amplifiers and anti-aliasing filters), an analog-to-digital conversion module (ADC sampling and quantization network), and a digital signal processing module (digital filtering and decimation algorithms within the DSP). Next, the nominal parameter set of the device is used to construct the underlying mathematical sub-model for each module. For the analog circuit section, Kirchhoff's laws are typically used to establish state-space equations or s-domain transfer functions under Laplace transforms. For example, nominal values ​​of resistors and capacitors are substituted to establish the poles and zeros of the filter. For the digital and mixed-signal sections, discrete-time (z-domain) difference equations are established to define the quantization bits, sampling rate, and algorithm coefficients. Finally, the system cascades these mathematical sub-models according to the actual physical flow of the signal. The core engineering innovation of this step lies in the fact that the system does not simply perform an idealized series connection, but deliberately introduces additional interference variables at the cascading nodes of each module. These variables include stray inductance and parasitic capacitance models caused by PCB traces, as well as random Gaussian noise floor models simulating thermal noise and electromagnetic radiation. Through this coupling method with physical constraints and parasitic effects, the final generated parameterized virtual behavioral twin is no longer an ideal calculator on paper, but a digital twin that can highly realistically reflect the dynamic electrical characteristics of actual equipment in a complex power grid environment.

[0029] S2. Identify and extract the most sensitive parameters that respond to changes in external stimuli from the parameterized virtual behavioral twin.

[0030] In a specific embodiment of the present invention, identifying and extracting the most sensitive parameter to the change in response to external stimuli from the parameterized virtual behavioral twin includes: setting a standard broadband test signal covering the target operating frequency band in the parameterized virtual behavioral twin.

[0031] For each candidate parameter within the parameterized virtual behavior twin, a small perturbation with a preset step size is applied near its nominal value, and the corresponding output response deviation is obtained.

[0032] Calculate the partial derivative of the output response deviation with respect to the perturbation of the corresponding parameter to obtain the local sensitivity coefficient of each candidate parameter.

[0033] Sort all candidate parameters by their local sensitivity coefficients, and extract the top-ranked parameters (a preset number) or parameters whose local sensitivity coefficients exceed a preset threshold as sensitive parameters.

[0034] Specifically, the engineering objective of extracting sensitive parameters in this embodiment is to accurately screen out sensitive parameters from a complex model containing hundreds or thousands of component parameters. This effectively eliminates redundant parameters that have negligible impact on waveform recording accuracy, such as bypass capacitors in non-critical paths, thereby reducing the dimensionality of the parameter space and greatly reducing the computational complexity for subsequent dynamic excitation generation and reverse diagnosis. This process employs a sensitivity analysis method based on perturbation theory. First, the system generates a standard wideband test signal as the excitation source. This signal typically contains a continuous spectrum from DC to the Nyquist frequency to ensure that the device's response can be excited across all frequency bands. Subsequently, the system enters the parameter traversal optimization phase. For each candidate parameter within the model, such as the bias voltage of a specific operational amplifier, the sampling clock frequency, etc., at its nominal value... Based on this, a small perturbation is artificially applied to each. For example, a drift of ±1% is allowed, and the modified model is exposed to a standard broadband test signal, recording the waveform output response deviation caused by the parameter change. Next, the system calculates... and The ratio between these two values ​​is used to approximate the partial derivative of the parameter. This partial derivative value is the local sensitivity coefficient of the candidate parameter. . The larger the absolute value, the more severe the distortion in the waveform recording results will be due to even minor aging or drift of the physical parameter. Finally, the system sorts all candidate parameters according to their local sensitivity coefficients. Sort in descending order. Extract the top N parameters, or select... All parameters exceeding the preset sensitivity threshold. These standout parameters, identified as sensitive parameters, will serve as the core targets for subsequent dynamic accuracy stress testing.

[0035] In a specific embodiment of the present invention, the typical value of the preset sensitivity threshold is usually taken as 1.0 of the normalized relative sensitivity coefficient. This value is based on the "amplification and attenuation effect" mechanism of error propagation in physical systems: in the high-precision measurement link of a transient waveform recording device, if a relatively small drift of a certain underlying hardware parameter (e.g., 1% capacitor aging) leads to an equal or greater proportion of relative distortion in the final waveform recording output, i.e., a sensitivity coefficient ≥ 1.0, it indicates that the error of this parameter is directly transmitted or even amplified by the system link, representing a key bottleneck determining the device's accuracy; conversely, if the coefficient is less than 1.0, it indicates that the drift of this parameter has been effectively attenuated and suppressed by the circuit's negative feedback mechanism or digital filtering algorithm. Therefore, setting the preset sensitivity threshold to the physical critical point of 1.0 can accurately identify fatal defect sources with "error amplification effects," ensuring that reverse physical diagnosis does not overlook core hidden dangers, while maximizing the elimination of redundant parameters, thereby greatly reducing the computational complexity of high-dimensional multi-feature space trajectory mapping.

[0036] S3. Generate a dynamic precision stress test excitation sequence based on sensitive parameters.

[0037] In a specific embodiment of the present invention, a dynamic precision stress test excitation sequence is generated based on sensitive parameters, including: classifying the sensitive parameters into a first type of parameters characterizing high-frequency response, a second type of parameters characterizing low-frequency characteristics, and a third type of parameters characterizing frequency selectivity.

[0038] Based on the first type of parameters, a first excitation sequence for stimulating high-frequency response capability is generated.

[0039] Based on the second type of parameters, a second excitation sequence is generated to test low-frequency stability.

[0040] Based on the third type of parameters, a third excitation sequence is generated to evaluate the spectral fidelity.

[0041] The first, second, and third excitation subsequences are combined and scheduled in the time domain to generate a dynamic precision stress test excitation sequence.

[0042] Specifically, the engineering objective of generating the dynamic precision stress test excitation sequence in this embodiment is to create a composite test signal capable of accurately and comprehensively stimulating the potential precision defects of the fault indicator under test at different operating limits. This process first involves acquiring preset sensitive parameters in the parameterized virtual behavioral twin. These parameters are key variables in the model that are most sensitive to external excitation responses and best reflect the physical characteristics of the device. In engineering, these preset sensitive parameters are divided into three categories. The first category of parameters characterizes the high-frequency response capability of the device, such as the bandwidth of the sensor equivalent circuit or the maximum sampling rate of the AD converter. The second category of parameters characterizes the low-frequency characteristics of the device, such as the time constant of the digital high-pass filter used to filter out DC components in the software algorithm. The third category of parameters characterizes the frequency selectivity of the device, such as the cutoff frequency and roll-off characteristics of the anti-aliasing filter.

[0043] After acquiring these parameters, the system executes parallel excitation subsequence generation. For the first type of parameters, the system generates a first excitation subsequence. Its engineering implementation involves synthesizing one or more transient pulse signals with extremely steep rising edges. The rise time of these pulses is set to the nanosecond to microsecond level, and its specific value is directly related to the system response time represented by the first type of parameters, aiming to probe the device's capture limit for high-frequency transient events. For the second type of parameters, the system generates a second excitation subsequence, which is a power frequency sinusoidal fundamental wave superimposed with an exponentially decaying DC component. Its decay time constant is set to tens to hundreds of milliseconds, used to test the measurement stability of the device under long-term DC bias and the DC suppression performance of the algorithm. For the third type of parameters, the system generates a third excitation subsequence, which is engineered to synthesize a composite waveform signal containing the fundamental wave and multiple specific higher-order harmonics and interharmonics. The order and amplitude of the harmonics are generated based on a preset typical power grid disturbance model, aiming to evaluate the spectral fidelity of the device in complex electromagnetic environments.

[0044] Finally, the system combines and sorts the generated first, second, and third excitation subsequences according to a preset strategy to form the final dynamic precision stress test excitation sequence. This combination strategy is not a simple splicing, but rather a scheduling method using a time-domain windowing function to simulate the occurrence order and superposition effect of different features during a real fault process. This dynamic precision stress test excitation sequence can be expressed by the following formula: ,in, The instantaneous values ​​of the final generated dynamic precision stress test excitation sequence. The waveform function representing the first excitation subsequence. The waveform function representing the second excitation sequence. The waveform function representing the third excitation sequence. , and These are three non-overlapping time window functions whose values ​​vary between 0 and 1. They control the activation period and intensity of transient pulses, decaying DC, and harmonic groups throughout the test sequence, thereby ensuring the dynamics of the test sequence and the ability to target different performance dimensions of the equipment for stress testing.

[0045] It should be noted that regarding waveform functions , , and time window function , , Detailed explanation: 1. Regarding the waveform functions of the three excitation subsequences , , Detailed explanation of the above formula: , , These are deterministic time-domain waveform functions designed for different performance dimensions. Their specific mathematical form and physical meaning are as follows: First excitation subsequence waveform function This function is used to excite the high-frequency response limit of a device. Its typical form is a double-exponentially decaying pulse, simulating the initial steep wave front of a lightning strike or switching overvoltage. The mathematical expression is: in, This is the pulse amplitude (typical value: 1.0 pu). The rise time constant (typical value: 0.1 μs to 1 μs). This is the fall time constant (typical value: 50μs to 500μs). The value of is based on the reciprocal of the nominal bandwidth of the fault indicator under test, and is intended to generate an excitation whose spectral components cover its high-frequency measurement limit. The value of simulates the decay process of the actual transient current. The second excitation sequence waveform function. This function is used to test the measurement stability and algorithm suppression capability of the equipment under low frequency and DC components. Its typical form is a power frequency sine wave superimposed with an exponentially decaying DC component. in, It is the power frequency (50Hz or 60Hz). AC amplitude (typical value: 1.0 pu). This is the initial value for DC offset (typical value: 0.2pu to 0.5pu). This is the DC decay time constant (typical value: 100ms to 500ms). and The value is determined based on the transient DC component generated during a single-phase ground fault in a simulated low-current grounding system, and is used to evaluate the suppression performance of the equipment's digital filtering algorithm on aperiodic components. Third excitation subsequence waveform function This function is used to evaluate the fidelity of a device in complex spectral environments. Its typical form is an amplitude-modulated signal with a fundamental frequency superimposed with specific subharmonics and interharmonics. in, For a preset set of harmonic orders (e.g., the 3rd, 5th, 7th characteristic harmonics, or the 2nd, 5th, and 4th interharmonics). The values ​​represent the amplitudes of each harmonic (typical values: 0.1 pu for the 3rd harmonic and 0.05 pu for the 5th harmonic). The modulation frequency (typical: 5Hz to 20Hz). Modulation depth (typical value: 0.1). Harmonic set. With amplitude The selection criteria are based on the characteristic harmonic spectrum generated by typical nonlinear loads (such as rectifiers and electric arc furnaces) in the power grid, aiming to test the accuracy of the equipment's anti-spectral aliasing and frequency selectivity.

[0046] 2. , , Essentially, these are three "switches" that open sequentially on the timeline, with some overlap. Their function is to control when the corresponding test signals are "added" to the final overall test sequence. For the most intuitive understanding, we can define them using a simple, classic variation of a rectangular window. This window function has a value of 1 (fully open) in the core region and smoothly transitions to 0 (closed) at the edges. Its concept can be simplified as follows: : Controls the switching of high-frequency pulse signals. It is time-sensitive. Open nearby, in The area is closed. : A switch that controls the attenuation of a DC signal. It operates on time. Open nearby (with) (The closing times overlap), in The area is closed. : Controls the switching of composite harmonic signals. It is time-sensitive. Open nearby (with) (The closing times overlap), in The area is closed. (This part is incomplete and likely refers to a specific location or area.) These are four key time points.

[0047] A visual explanation of "imperfect overlap": "Imperfect overlap" means that the three switches are not all turned on simultaneously, nor are they completely separate and unrelated. Their activity times are sequential, with a deliberate overlap of some of their durations. The specific timing sequence can be arranged as follows (assuming a total test duration of 1 second): arrive Mainly Effective, allowing high-frequency pulse signals It plays a role in testing the instantaneous response capability of the equipment. arrive Mainly Effective, allowing signals with attenuated DC to... Join. Please note that in arrive This range, and When both are active, their signals will superimpose. This simulates the attenuation process that begins and continues after the initial impact in a real fault. arrive Mainly Effective, allowing complex harmonic signals Add it last. Similarly, in arrive This range, and Simultaneously effective, it simulates the phenomenon that harmonic components gradually generate and interact with attenuation components during the duration of a fault.

[0048] S4. Synchronously input the dynamic precision pressure test excitation sequence to the fault indicator under test and the parameterized virtual behavior twin, and simultaneously collect the first response sequence of the fault indicator under test and the second response sequence of the parameterized virtual behavior twin.

[0049] Specifically, the engineering purpose of synchronously inputting and acquiring the dynamic precision stress test excitation sequence in this embodiment is to overcome the heterogeneous barrier between physical hardware entities and computer digital space, and to construct a "parallel observation field" with nanosecond-level absolute alignment accuracy on the time axis. Its core significance lies in completely eliminating the "pseudo-errors" introduced by the hardware group delay, trigger jitter, or excitation distortion of the test system itself, thereby ensuring that any waveform morphology differences captured by the subsequent double-loop trajectory mapping are purely and uniquely derived from defects or offsets in the internal physical parameters of the device under test.

[0050] The process begins with high-precision physical and digital dual-track synchronous injection. In engineering implementation, since the object under test exists in two different physical dimensions, the system employs a parallel excitation delivery mechanism. For the fault indicator entity under test, the system downloads the digital dynamic precision stress test excitation sequence to a high-speed arbitrary waveform generator (AWG), which converts it into actual transient current or voltage physical analog signals via a high-fidelity wideband power amplifier and injects it directly into the device's analog input terminals via hardwiring. In parallel, for the parameterized virtual behavioral twin, this digital excitation sequence is used as a lossless digital excitation source, directly fed into the input node of the digital twin simulation engine. To ensure absolute concurrency between the virtual and real worlds at the moment of excitation, the system's main control unit broadcasts a nanosecond-precision global hardware synchronization trigger pulse. This pulse triggers the first sampling point of the physical waveform output by the AWG at the hardware level, while simultaneously and precisely initiating the "zero-time" integral calculation of the virtual simulation engine at the software level.

[0051] Next, the system enters the stage of synchronous acquisition and aggregation of heterogeneous response data. After detecting a sudden change in physical excitation, the fault indicator under test (DUT) triggers its internal transient waveform recording function at the hardware level, independently completing the sampling, quantization, and internal storage of the real physical signal. Subsequently, the system transmits this waveform recording file, which includes the device's real physical response hysteresis, component nonlinearity, and background noise, back through the industrial Ethernet communication interface, extracting it to form the first response sequence. At the same time, the parameterized virtual behavioral twin solves the transient differential equations of the excitation sequence within the simulation engine according to the set micro-step size (to ensure simulation accuracy, its step size is usually much smaller than the actual sampling period of the physical device), outputting a continuous time series representing the current model parameter characteristics, i.e., the second response sequence.

[0052] Finally, the system performs rigorous spatiotemporal locking and standardization preprocessing on the acquired first and second response sequences. Due to inherent limitations of physical devices, such as group delays caused by anti-aliasing filters and trigger jitter from AD conversion, and the fact that their sampling frequencies and quantization bit depths often differ, direct data comparison would lead to severe phase misalignment and dimensional conflicts. In engineering, this step first uses a time-domain cross-correlation algorithm to accurately calculate the microsecond-level relative time offset between the first and second response sequences near the transient transition edge. Based on this, the sequences are shifted and compensated, and their reference zero points are absolutely locked and aligned. Subsequently, high-order spline interpolation or multiphase filtering algorithms are used to resample and anti-aliasing extract the second response sequence (simulated high-frequency data), forcing its sampling rate resolution to be completely consistent with the first response sequence (actual waveform data), and uniformly performing amplitude normalization. Through this series of standardized reconstruction operations, the original data from the two heterogeneous dimensions of physical and virtual data are accurately mapped to an absolutely aligned unified digital space, paving the way for subsequent lossless extraction of multidimensional feature vectors and the construction of dual-loop dynamic trajectories.

[0053] S5. Based on the first response sequence and the second response sequence, perform double-loop trajectory mapping to obtain trajectory deviation characteristics.

[0054] In a specific embodiment of the present invention, based on the first response sequence and the second response sequence, a double-loop trajectory mapping is performed to obtain trajectory deviation features, including: for each excitation unit in the dynamic precision stress test excitation sequence, extracting a multi-dimensional feature vector from the corresponding first response sequence and the second response sequence.

[0055] In a specific embodiment of the present invention, for each excitation unit in the dynamic precision pressure test excitation sequence, a multi-dimensional feature vector is extracted from the corresponding first response sequence and second response sequence, including: performing time-frequency analysis on the corresponding response segments of the first response sequence and second response sequence to obtain a first time-frequency distribution map and a second time-frequency distribution map.

[0056] Frequency domain energy distribution features are extracted from the first and second time-frequency distribution maps.

[0057] Similarity calculation and time point statistics are performed on the response segments to obtain waveform similarity features and timing jitter statistical features, respectively.

[0058] The frequency domain energy distribution characteristics, waveform similarity characteristics, and timing jitter statistical characteristics are combined into a multi-dimensional feature vector.

[0059] Specifically, the engineering purpose of extracting multidimensional feature vectors from the response sequence in this embodiment is to convert the original one-dimensional time-domain waveform data into a structured numerical vector that can quantify its distortion from multiple physical dimensions, providing a standardized input for subsequent trajectory mapping.

[0060] The process first performs time-frequency analysis in parallel on the response segments corresponding to the same excitation unit in the first and second response sequences. In engineering, this step uses a short-time Fourier transform, employing a sliding window function with a width of several milliseconds to convert each response segment into a two-dimensional matrix, namely the first and second time-frequency distribution maps. These maps accurately display the distribution of signal energy in both time and frequency dimensions. Next, the system calculates the frequency domain energy distribution characteristics based on these two maps. The system integrates the energy in the first and second time-frequency distribution maps at a preset set of key frequencies, such as the fundamental frequency band, major harmonic bands, and transient high-frequency bands. Then, it calculates the energy ratio of the two maps in the corresponding frequency bands, forming a vector, which represents the frequency domain energy distribution characteristics. Simultaneously, the system performs waveform alignment and similarity calculations on the original response segments, generating waveform similarity features. In engineering, this step utilizes a dynamic time warping algorithm to calculate the minimum cost required to nonlinearly warp the first response segment to match the second response segment. This cost is a scalar that directly quantifies the macroscopic morphological differences between the two waveforms.

[0061] In addition, the system also performs statistical analysis on key time points in the response segments to generate timing jitter statistical features. The system automatically detects all zero-crossing points or local extrema in each response segment and records their precise timestamps. By comparing the timestamps of corresponding feature points in the first and second response sequences, a time deviation sequence is obtained, and the standard deviation of this time deviation sequence is calculated. This standard deviation is the timing jitter statistical feature, which reflects the response instability of the device under test on the time axis. Finally, the system combines the frequency domain energy distribution features, waveform similarity features, and timing jitter statistical features generated in the aforementioned steps into a single multidimensional feature vector. This combination process can be expressed by the following formula: ,in, This is the final output multidimensional feature vector. Representative at the The energy ratios calculated on each preset key frequency band are provided by the frequency domain energy distribution characteristics. This represents the waveform similarity feature calculated by the dynamic time warping algorithm. The standard deviation representing the time deviation at zero-crossing or extreme points is divided by a preset reference time value (such as the sampling interval) to convert it into a dimensionless relative value, thus obtaining the statistical characteristics of timing jitter. In this way, the original complex waveform is condensed into a standardized numerical vector that contains spectral information, morphological information, and timing information.

[0062] It should be noted that, in further detail, the waveform similarity features calculated by the dynamic time warping algorithm in the above multidimensional feature vectors will be explained. The specific engineering calculation logic is as follows. In the comparison of transient recording accuracy, due to the hardware characteristics of the anti-aliasing filter group delay and A / D conversion trigger jitter of the fault indicator under test, there is often a nonlinear small time axis distortion or local phase shift between the first response sequence recorded and the second response sequence generated by the parameterized virtual behavior twin. If the error is directly calculated using the traditional point-by-point Euclidean distance, the small time misalignment will be mistakenly amplified into a huge amplitude error. To this end, this invention introduces the DTW algorithm for elastic alignment, the specific steps of which are as follows: Step 1: Sequence Definition and Local Distance Matrix Construction. The extracted first response sequence (measured waveform segment) is... The second response sequence (virtual standard waveform segment) is ,in and These represent the number of sampling points for the two sequences, respectively. The system first calculates the sequence... Each point in with sequence Each point in The local absolute magnitude error (Euclidean distance) between them, i.e. Thus constructing a The local distance matrix.

[0063] Step 2: Dynamic Programming to Find the Minimum Cumulative Path System. Based on the local distance matrix, the dynamic programming algorithm is used to find a path from the starting point... To the finish line The optimal regularized path must satisfy monotonicity and continuity constraints (i.e., time can only advance in one direction and cannot jump). The system constructs a cumulative distance matrix. Its internal elements This indicates alignment from the starting point to the current point. The minimum cumulative cost. Its recursive calculation formula is: The physical meaning of this formula is: the cumulative error of the current point is equal to the absolute magnitude error of the current point, plus the value of the path with the smallest cumulative error among the three allowed predecessor nodes from the left, below, or lower left.

[0064] Step 3: Calculate the total DTW distance. After recursive calculation of the complete matrix, the top right element of the cumulative distance matrix is... This represents the minimum total absolute error cost of two waveforms after optimal elastic time alignment. To eliminate the influence of sequence length differences, it is divided by the total step size of the regularized path. The standardized DTW distance is obtained: The smaller the distance value, the closer the two waveforms are in shape.

[0065] Step 4: Feature Transformation (Generating Similarity Features) Since the subsequent multidimensional feature space requires a unified dimension direction (the larger the value, the more similar), the system uses a monotonically decreasing exponential mapping function to convert the DTW distance into the final waveform similarity feature. : in, The preset sensitivity adjustment coefficient (typically, it can be set to 0.5%) (This depends on the per-unit range of the waveform amplitude). Through this mapping, Strictly normalized to Interval. When the two waveforms are completely identical... , The more significant the difference in waveform morphology, the greater the cost of normalization. The value then approaches the value exponentially. This scalar That is, the core component representing the fidelity of the macroscopic shape of the waveform is pressed into the multidimensional feature vector.

[0066] Based on the multidimensional feature vectors corresponding to all excitation units, the first dynamic trajectory and the second dynamic trajectory are formed in the feature space respectively.

[0067] In a specific embodiment of the present invention, based on the multidimensional feature vectors corresponding to all excitation units, a first dynamic trajectory and a second dynamic trajectory are formed in the feature space, including: constructing a high-dimensional precision state space with waveform similarity features, timing jitter statistics features and frequency domain energy distribution features as coordinate axes.

[0068] All the multidimensional feature vectors generated by the response of the fault indicator under test are mapped to a high-dimensional precision state space in the order of excitation and connected to form the first dynamic trajectory.

[0069] All multidimensional feature vectors generated by the parameterized virtual behavior twin response are mapped to a high-dimensional precision state space in the same excitation order and connected to form a second dynamic trajectory.

[0070] Specifically, in this embodiment, the engineering purpose of forming the first dynamic trajectory and the second dynamic trajectory in the preset feature space based on the multi-dimensional feature vectors corresponding to all excitation units is to transform the discrete, multi-dimensional performance snapshots into continuous geometric paths that can intuitively represent the overall evolution of device behavior during the entire dynamic testing process, providing a computable object for subsequent morphological difference analysis.

[0071] The process begins with constructing a high-dimensional precision state space. This is achieved by defining a multi-dimensional orthogonal coordinate system, where each axis uniquely corresponds to a component of a multi-dimensional feature vector. Specifically, waveform similarity features and timing jitter statistics each have an independent coordinate axis, and the energy ratio of each frequency band in the frequency domain energy distribution features is also assigned an independent coordinate axis. To ensure a balanced contribution of features with different physical dimensions in spatial geometry, the system normalizes the historical data or expected range of each feature component before constructing the space, unifying the scale of each coordinate axis to the range of 0 to 1. Subsequently, the system begins mapping and generating the first dynamic trajectory. Following the execution order of the dynamic precision stress test excitation sequence, the system sequentially reads each multi-dimensional feature vector generated for the fault indicator under test. Each multi-dimensional feature vector is considered a point in the high-dimensional precision state space, its coordinates determined by the vector components. The system connects these points sequentially according to time, forming a spatial curve representing the evolution of the device's actual response behavior—the first dynamic trajectory. The system processes all multi-dimensional feature vectors from the parameterized virtual behavior twin response in the same manner. These vector points, representing the ideal response behavior, are mapped and connected according to the same excitation order to form a second dynamic trajectory. This trajectory serves as a baseline path in space, representing the ideal dynamic behavior that the device under test should exhibit under the current test excitation, assuming no accuracy defects.

[0072] Calculate the metric value of the region enclosed by the first dynamic trajectory and the second dynamic trajectory in the feature space to obtain the trajectory enclosed area feature.

[0073] Analyze the principal direction vectors of the first and second dynamic trajectories in the feature space, and calculate the angle between the principal direction vectors to obtain the trajectory deviation angle feature.

[0074] The trajectory morphology difference is generated by weighting and combining the trajectory enclosing area feature and the trajectory deviation angle feature.

[0075] The difference in trajectory morphology is quantified as a trajectory deviation feature.

[0076] Specifically, the engineering purpose of quantifying trajectory morphology differences into trajectory deviation features in this embodiment is to decompose the abstract, high-dimensional trajectory geometric differences into two orthogonal metrics with clear physical diagnostic significance: the magnitude of the deviation and the structure of the deviation, thereby providing more targeted input for subsequent defect root cause localization.

[0077] This process first calculates the trajectory-enclosed area feature in a predefined feature space. In engineering implementation, this is achieved by using numerical methods to approximate the volume or cross-sectional area integral of the "tubular" region enclosed by the two spatial curves of the first and second dynamic trajectories. The system connects the corresponding point pairs on the two trajectories according to the excitation sequence, forming a series of tiny polygons or high-dimensional simplexes, and then accumulates the area or volume of these tiny units. The calculation result is output as a scalar value, namely the trajectory-enclosed area feature. This feature intuitively reflects the total cumulative deviation between the actual device response and the ideal model response throughout the entire test sequence; the larger the value, the wider the deviation or the longer its duration.

[0078] In parallel, the system analyzes the principal direction vectors of the two dynamic trajectories in the feature space to obtain the trajectory deviation angle feature. In engineering, this step is achieved by performing principal component analysis on the point set constituting each trajectory. The covariance matrix of the point set of the first dynamic trajectory is calculated, and the eigenvector corresponding to the largest eigenvalue of this matrix is ​​solved. This vector is the principal direction vector of the first dynamic trajectory, representing the direction of the greatest variation in the trajectory data. The principal direction vector of the second dynamic trajectory is obtained in the same way. After obtaining the two principal direction vectors, the system calculates the angle between them; this angle value is the trajectory deviation angle feature. This feature reveals whether the macroscopic trends of the internal state evolution of the two systems are consistent when responding to dynamic excitation. A large deviation angle indicates a fundamental structural difference between the response mode of the device under test and the ideal model.

[0079] Finally, the system generates the final trajectory deviation feature based on the obtained trajectory encirclement area feature and trajectory deviation angle feature. Since these two features have different dimensions and physical meanings, they cannot be directly added. Therefore, they need to be normalized first, and then a comprehensive scalar index is generated through weighted combination. This process can be expressed by the following formula: ,in, This refers to the final generated trajectory deviation features. It is a normalized area feature obtained by dividing the trajectory-enclosed area feature by a preset baseline area value or the historical statistical maximum value. It is a normalized angle feature obtained by dividing the trajectory deviation angle feature by the maximum possible angle value (such as π). and These are preset weighting coefficients, which sum to 1. They reflect the degree of importance attached to the two different types of errors, namely deviation magnitude and deviation structure, in the final evaluation. These weighting coefficients are set based on prior engineering knowledge of equipment type and application scenario.

[0080] In a specific embodiment of the present invention, in the formula for calculating the trajectory morphology difference, the weighting coefficient... and The typical value is usually set to , The specific value is based on the engineering criterion priority of the fault indicator as a power grid safety monitoring device: the trajectory deviation angle represents the structural and systematic deviation of the device's dynamic response from the ideal model in the evolution mode. For example, insufficient core bandwidth leads to overall distortion of the response mode for all high-frequency signals. This type of error is highly concealed and usually points to fundamental defects in hardware or algorithms, with a higher risk level, and is therefore given the main weight. The area enclosed by the trajectory reflects more the cumulative magnitude or duration of the deviation and may include more random noise or non-systematic error, therefore it is given a secondary weight. ).

[0081] S6. Obtain the mapping relationship between trajectory deviation features and sensitive parameters, analyze the trajectory deviation features based on the mapping relationship, locate the target sensitive parameters that cause the deviation, and generate diagnostic results containing parameter offsets.

[0082] In a specific embodiment of the present invention, the mapping relationship between trajectory deviation features and sensitive parameters is obtained, and the trajectory deviation features are analyzed based on the mapping relationship to locate the target sensitive parameters that cause the deviation and generate a diagnostic result containing parameter offset, including: inputting the trajectory deviation features into a mapping analysis model that stores pre-simulation data.

[0083] By using a mapping analysis model, target sensitive parameters that are most relevant to the current trajectory deviation characteristics are matched.

[0084] Based on the quantization relationships recorded in the mapping analysis model, the parameter offset of the target sensitive parameters is calculated.

[0085] Specifically, the engineering purpose of using the mapping relationship between trajectory deviation characteristics and preset sensitive parameters in this embodiment to perform accuracy assessment and defect root cause location is to accurately decode a comprehensive but physically ambiguous deviation quantity into the offset of specific equipment physical parameters that cause the deviation, thereby achieving automated diagnosis from "phenomenon" to "root cause".

[0086] The process begins by inputting the trajectory deviation features calculated in the previous step into a pre-defined mapping analysis model. In engineering, this mapping analysis model is either a knowledge base pre-built through offline simulation or a trained neural network. The construction process involves: in a parameterized virtual behavioral twin, each pre-defined sensitive parameter is changed one by one and quantitatively, for example, reducing the sensor bandwidth by 5% or 10% or increasing the sampling clock jitter by 50 ppm. Then, a complete dynamic accuracy stress test excitation sequence is run, and the specific trajectory deviation features corresponding to each parameter offset are calculated, thereby establishing a database of one-to-one or many-to-one correlations between "parameter defects" and "trajectory deviation feature performance."

[0087] Upon receiving trajectory deviation features generated from actual testing, the core action of the mapping analysis model is to perform an inverse optimization or pattern matching task. The system searches the pre-built relational database for the pre-stored record(s) that best match the current input trajectory deviation features in terms of both numerical value and structure. This matching process is accomplished by minimizing the objective function, identifying the target sensitive parameter that can best explain the currently observed trajectory deviation.

[0088] Once a target sensitive parameter is identified, the system calculates its estimated offset from its nominal value using interpolation or function fitting methods, based on its quantification relationship in the database. Finally, the system combines the identified target sensitive parameters and their corresponding estimated offsets into a structured diagnostic result and outputs it. This diagnostic process can be represented by the following functional relationship: ,in, This is the final output diagnostic result, which contains target sensitivity parameters. and its estimated offset The binary tuple. It is the trajectory deviation feature of the input. This represents the inverse mapping function of the entire mapping analysis model. In engineering terms, this involves solving an optimization problem: finding the characteristics that minimize the deviation of the simulated trajectory. Deviation characteristics from measured trajectory Parameters that minimize the difference between and its offset .

[0089] It should be noted that the function This is the core algorithm of the mapping analysis model, and its function is to perform automated diagnosis by "inferring the cause from the phenomenon". Specifically, its input is the measured trajectory deviation characteristics, and the output is the diagnostic result, namely the target sensitive parameter that caused the error and its specific estimated offset. Its implementation process is based on a pre-built "defect feature knowledge base" and follows these concise steps: Knowledge base construction: Before system deployment, through numerous simulation experiments, the standard trajectory deviation characteristics corresponding to different magnitudes (e.g., +1%, -5%) of each key physical parameter in the device (such as sensor bandwidth, clock accuracy) are pre-calculated. These correspondences between "parameter offsets and feature manifestations" are stored in the knowledge base. Feature comparison and matching: After obtaining a measured trajectory deviation characteristic, the function... It automatically and quickly compares the result with all pre-stored "standard features" in the knowledge base to find the one that is closest in form and value. Output: Function The parameters and offsets corresponding to the matched "standard features" are directly output as diagnostic results. That is, through table lookup and comparison, it indicates that "the error characteristics exhibited by the current device best match the situation recorded in the knowledge base as 'a certain parameter has shifted by a certain magnitude'." Core objective: By transforming complex and hidden hardware performance degradation issues into a clear and quantifiable parameter diagnostic report, an intelligent leap from performance testing to root cause identification is achieved.

[0090] S7. Based on the parameter offset in the diagnostic results, update the parameters of the parameterized virtual behavior twin and generate the calibrated virtual device model.

[0091] In a specific embodiment of the present invention, the parameters of the parameterized virtual behavior twin are updated according to the parameter offset in the diagnostic results to generate a calibrated virtual device model, including: replacing the nominal value of the target sensitive parameter in the parameterized virtual behavior twin with the actual correction value containing the parameter offset according to the diagnostic results.

[0092] Using the actual correction value as a fixed constraint, the secondary associated parameters that are electrically coupled to the target sensitive parameters are finely adjusted through closed-loop simulation.

[0093] The updated parameter set, including the actual correction values ​​and compensatory fine-tuning, is solidified into a parameterized virtual behavioral twin, completing the model update and generating a calibrated virtual device model that can characterize the current physical state of the fault indicator under test.

[0094] Specifically, the engineering purpose of updating the model parameters to generate the calibrated virtual device model in this embodiment is to thoroughly write the "device medical record" obtained through reverse diagnosis into the virtual clone, so that it transforms from a universal model representing a "general model" into a personalized high-fidelity model that accurately depicts the current aging and drift state of "this specific device".

[0095] This process is not merely a simple numerical replacement. First, the system directly responds to the diagnostic results, incorporating the locked target sensitive parameter (e.g., diagnosing a 150ppm jitter in the AD sampling clock) into the model. The original ideal clock equation based on the nominal value (0 jitter) is replaced with a corrected equation incorporating a Gaussian random process. This step establishes the core tone of the model bias. Next, the system performs highly valuable "coupling parameter compensation." In actual physical circuits, due to impedance matching and feedback networks, a shift in a core parameter (such as a decrease in front-end amplifier gain) often leads to small changes in other secondary parameters (such as phase response within the frequency band), creating a "cascade effect." Therefore, the system uses the newly implanted corrected value as an unchangeable fixed constraint and runs the local closed-loop simulation again. In this simulation, the system employs optimization algorithms such as gradient descent to perform extremely small "compensatory fine-tuning" on secondary related parameters within the same circuit module as the sensitive parameter, ensuring that the overall response of the entire module achieves the best possible fit with the actual acquired device data, typically requiring a fitting residual within 0.1%. Finally, the system irreversibly packages and solidifies a new set of parameters containing both the "core cause (actual corrected value)" and "related manifestations (compensatory fine-tuning)." After this step, the virtual twin is calibrated. This calibrated virtual device model is no longer merely a byproduct of a testing tool, but a highly valuable digital asset. In future operation, maintenance personnel can directly apply extreme lightning strikes or short-circuit waveforms to this calibrated model to predict with remarkable accuracy how the actual fault indicator in service will react.

[0096] In a specific embodiment of the present invention, after generating the calibrated virtual device model, the method further includes: obtaining a verification stimulus independent of the dynamic accuracy stress test stimulus sequence.

[0097] The verification stimulus is simultaneously input into the fault indicator under test and the calibrated virtual device model to obtain the actual response and the simulated response.

[0098] Calculate the degree of consistency between the actual response and the simulated response.

[0099] The validity of the calibrated virtual device model is confirmed based on whether the degree of consistency exceeds a preset threshold.

[0100] Specifically, in this embodiment, the engineering purpose of performing closed-loop verification after generating the calibrated virtual device model is to independently and objectively verify the accuracy of the previous diagnostic results and the generalization ability of the calibrated model, ensuring that the model can not only fit the initial test data, but also accurately predict the actual behavior of the device in unknown transient processes.

[0101] The process begins with acquiring verification stimuli. In engineering, these stimuli come from a dataset entirely different from the initial dynamic accuracy stress test stimuli sequence. Typically, they are preset historical data of real faults, collected from actual fault events occurring in the power grid and containing complex non-ideal factors. After acquiring these stimuli, the system uses them as input, synchronously applying them to the fault indicator under test to obtain its real-time output response. Simultaneously, this response is input to a calibrated virtual device model with updated parameters to obtain its simulated output response. Next, the system quantitatively compares the acquired actual response with the simulated response to calculate their degree of consistency. This step does not repeat complex trajectory mapping but uses a more direct error evaluation metric. The system calculates the normalized root mean square error or correlation coefficient between the two response waveforms. To obtain a unified and comprehensive consistency index, the following formula can be used for calculation: ,in, This represents the final calculated level of consistency; the closer the value is to 1, the higher the consistency. For actual response in The instantaneous value of a moment. To simulate the response in The instantaneous value at time Σ. Σ represents the summation over all sampling points during the entire verification excitation period. This formula provides a dimensionless evaluation value ranging from negative infinity to 1 by calculating the proportion of error energy to actual signal energy. Finally, the system calculates the degree of consistency. Compare with a preset threshold. This preset threshold is typically set to a relatively high value in engineering practice, such as 0.95. If the calculated... If the preset threshold is exceeded, the system determines that the verification is successful and ultimately confirms that the calibrated virtual device model is valid and can accurately represent the actual dynamic performance of the fault indicator under test.

[0102] Receive real transient waveform data collected by the fault indicator under test during field operation.

[0103] The actual transient waveform data is input into the calibrated virtual device model to obtain the predicted response waveform.

[0104] By comparing the predicted response waveform with the actual transient waveform data, the accuracy and stability of the fault indicator under test in a real operating environment are evaluated.

[0105] Specifically, in this embodiment, the engineering purpose of conducting application evaluation after generating the calibrated virtual device model is to transform one-time laboratory testing capabilities into continuous, online, and precise health status monitoring of field equipment throughout its entire lifecycle, thereby enabling early warning of equipment performance drift or potential failures.

[0106] This process is triggered by actual transient events occurring on-site. The system first receives real transient waveform data collected by the fault indicator under test (FAT) during on-site operation via a communication interface. This data represents the actual response of the device to faults, disturbances, and other events in a real power grid environment without any artificial excitation. Upon receiving this data, the system uses it as an input signal and injects it into a previously validated and calibrated virtual device model. In this application scenario, the model acts as a dynamic behavior observer. Based on its internally calibrated parameter set that accurately reflects the dynamic characteristics of the specific device, it processes the input real transient waveform data to generate a predicted response waveform. This predicted response waveform can be considered an "optimal estimate" of the original waveform data in engineering terms; it filters out random noise and presents the ideal waveform that the device should produce under its current health condition. Finally, the system performs a point-by-point comparative analysis between the predicted response waveform output by the model and the input original real transient waveform data to quantitatively evaluate the accuracy stability of the FAT under test in a real operating environment. This evaluation is accomplished by calculating an accuracy stability index that varies over time, and the calculation formula is as follows: ,in, for The accuracy stability index at a given moment is a dimensionless instantaneous relative error. For real transient waveform data in The instantaneous value of a moment. To predict the response waveform in The instantaneous value of a moment. The peak amplitude of the actual transient waveform data in this transient event is used for normalization to ensure the comparability of the index. The system... The system calculates the average or peak value over the entire event period. If the value shows a continuous upward trend over time in multiple events and exceeds a preset alarm threshold, such as 0.05, the system determines that the accuracy and stability of the device are declining and generates a corresponding maintenance warning.

[0107] Reference Figure 2 The second aspect of the present invention provides a multi-dimensional mapping detection system for the transient recording accuracy of a fault indicator, comprising: a virtual behavioral twin construction module, a sensitive parameter identification and extraction module, an excitation sequence generation module, a data acquisition and synchronization module, a trajectory mapping analysis module, a diagnostic result generation module, and a virtual device model calibration module.

[0108] The virtual behavior twin construction module is connected to the sensitive parameter identification and extraction module. The sensitive parameter identification and extraction module is connected to the excitation sequence generation module. Both the virtual behavior twin construction module and the excitation sequence generation module are connected to the data acquisition and synchronization module. The data acquisition and synchronization module is connected to the trajectory mapping analysis module. Both the sensitive parameter identification and extraction module and the trajectory mapping analysis module are connected to the diagnostic result generation module. The diagnostic result generation module is connected to the virtual device model calibration module.

[0109] The virtual behavioral twin construction module obtains the principle model and nominal parameter set of the fault indicator under test, and constructs a parameterized virtual behavioral twin based on the principle model and nominal parameter set.

[0110] The sensitive parameter identification and extraction module identifies and extracts the most sensitive parameters that respond to changes in external stimuli from the parameterized virtual behavior twin.

[0111] The excitation sequence generation module generates dynamic precision stress test excitation sequences based on sensitive parameters.

[0112] The data acquisition and synchronization module synchronously inputs the dynamic precision stress test excitation sequence to the fault indicator under test and the parameterized virtual behavior twin, and synchronously acquires the first response sequence of the fault indicator under test and the second response sequence of the parameterized virtual behavior twin.

[0113] The trajectory mapping analysis module performs double-loop trajectory mapping based on the first response sequence and the second response sequence to obtain trajectory deviation characteristics.

[0114] The diagnostic result generation module obtains the mapping relationship between trajectory deviation features and sensitive parameters, analyzes the trajectory deviation features based on the mapping relationship, locates the target sensitive parameters that cause the deviation, and generates diagnostic results containing parameter offsets.

[0115] The virtual device model calibration module updates the parameters of the parameterized virtual behavior twin based on the parameter offset in the diagnostic results, and generates a calibrated virtual device model.

[0116] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A multi-dimensional mapping detection method for the transient recording accuracy of a fault indicator, characterized in that, include: S1. Obtain the principle model and nominal parameter set of the fault indicator under test, and construct a parameterized virtual behavior twin based on the principle model and nominal parameter set; S2. Identify and extract the most sensitive parameters that respond to changes in external stimuli from the parameterized virtual behavior twin; S3. Generate a dynamic precision stress test excitation sequence based on sensitive parameters; S4. Synchronously input the dynamic precision pressure test excitation sequence to the fault indicator under test and the parameterized virtual behavior twin, and simultaneously collect the first response sequence of the fault indicator under test and the second response sequence of the parameterized virtual behavior twin. S5. Based on the first response sequence and the second response sequence, perform double-loop trajectory mapping to obtain trajectory deviation characteristics; S6. Obtain the mapping relationship between trajectory deviation features and sensitive parameters, analyze the trajectory deviation features based on the mapping relationship, locate the target sensitive parameters that cause the deviation, and generate diagnostic results containing parameter offsets. S7. Based on the parameter offset in the diagnostic results, update the parameters of the parameterized virtual behavior twin and generate the calibrated virtual device model.

2. The multi-dimensional mapping detection method for the transient recording accuracy of a fault indicator according to claim 1, characterized in that, The process of identifying and extracting the most sensitive parameters to changes in response to external stimuli from the parameterized virtual behavioral twin includes: Set a standard broadband test signal covering the target operating frequency band in the parameterized virtual behavior twin; For each candidate parameter within the parameterized virtual behavior twin, a small perturbation of a preset step size is applied near its nominal value, and the corresponding output response deviation is obtained. Calculate the partial derivative of the output response deviation with respect to the perturbation of the corresponding parameter to obtain the local sensitivity coefficient of each candidate parameter; Sort all candidate parameters by their local sensitivity coefficients, and extract the top-ranked parameters (a preset number) or parameters whose local sensitivity coefficients exceed a preset threshold as sensitive parameters.

3. The multi-dimensional mapping detection method for the transient recording accuracy of a fault indicator according to claim 1, characterized in that, The generation of dynamic precision stress test excitation sequences based on sensitive parameters includes: Sensitive parameters are classified into three categories: first-class parameters characterizing high-frequency response, second-class parameters characterizing low-frequency characteristics, and third-class parameters characterizing frequency selectivity. Based on the first type of parameters, a first excitation subsequence is generated to stimulate high-frequency response capability; Based on the second type of parameters, a second excitation subsequence is generated to test low-frequency stability; Based on the third type of parameters, a third excitation subsequence is generated to evaluate the spectral fidelity; The first, second, and third excitation subsequences are combined and scheduled in the time domain to generate a dynamic precision stress test excitation sequence.

4. The multi-dimensional mapping detection method for the transient recording accuracy of a fault indicator according to claim 1, characterized in that, The process of performing double-loop trajectory mapping based on the first response sequence and the second response sequence to obtain trajectory deviation features includes: For each excitation unit in the dynamic precision stress test excitation sequence, a multi-dimensional feature vector is extracted from the corresponding first response sequence and second response sequence. Based on the multidimensional feature vectors corresponding to all excitation units, the first dynamic trajectory and the second dynamic trajectory are formed in the feature space respectively; Calculate the metric value of the region enclosed by the first dynamic trajectory and the second dynamic trajectory in the feature space to obtain the trajectory enclosed area feature; Analyze the principal direction vectors of the first dynamic trajectory and the second dynamic trajectory in the feature space, and calculate the angle between the principal direction vectors to obtain the trajectory deviation angle feature; The trajectory enclosed area feature and the trajectory deviation angle feature are weighted and combined to generate the trajectory morphology difference degree; The difference in trajectory morphology is quantified as a trajectory deviation feature.

5. The multi-dimensional mapping detection method for the transient recording accuracy of a fault indicator according to claim 4, characterized in that, For each excitation unit in the dynamic precision stress test excitation sequence, a multi-dimensional feature vector is extracted from the corresponding first response sequence and second response sequence, including: Time-frequency analysis is performed on the corresponding response segments of the first response sequence and the second response sequence to obtain the first time-frequency distribution map and the second time-frequency distribution map; Extract frequency domain energy distribution features from the first and second time-frequency distribution maps; Similarity calculation and time point statistics are performed on the response segments to obtain waveform similarity features and timing jitter statistical features, respectively; The frequency domain energy distribution characteristics, waveform similarity characteristics, and timing jitter statistical characteristics are combined into a multi-dimensional feature vector.

6. The multi-dimensional mapping detection method for the transient recording accuracy of a fault indicator according to claim 5, characterized in that, The process of forming a first dynamic trajectory and a second dynamic trajectory in the feature space based on the multidimensional feature vectors corresponding to all excitation units includes: Construct a high-dimensional precision state space with waveform similarity features, timing jitter statistics, and frequency domain energy distribution features as coordinate axes; All the multidimensional feature vectors generated by the response of the fault indicator under test are mapped to a high-dimensional precision state space in the order of excitation and connected to form the first dynamic trajectory. All multidimensional feature vectors generated by the parameterized virtual behavior twin response are mapped to a high-dimensional precision state space in the same excitation order and connected to form a second dynamic trajectory.

7. The multi-dimensional mapping detection method for the transient recording accuracy of a fault indicator according to claim 1, characterized in that, The process of acquiring the mapping relationship between trajectory deviation features and sensitive parameters, analyzing the trajectory deviation features based on the mapping relationship, locating the target sensitive parameters causing the deviation, and generating diagnostic results including parameter offsets includes: The trajectory deviation characteristics are input into a mapping analysis model that stores pre-simulation data; By using a mapping analysis model, target sensitive parameters that are most relevant to the current trajectory deviation characteristics are matched; Based on the quantization relationships recorded in the mapping analysis model, the parameter offset of the target sensitive parameters is calculated.

8. The multi-dimensional mapping detection method for the transient recording accuracy of a fault indicator according to claim 1, characterized in that, The step of updating the parameters of the parameterized virtual behavioral twin based on the parameter offset in the diagnostic results and generating a calibrated virtual device model includes: Based on the diagnostic results, the nominal values ​​of the target sensitive parameters in the parameterized virtual behavior twin are replaced with actual corrected values ​​that include parameter offsets; Using the actual correction value as a fixed constraint, the secondary associated parameters that have an electrical coupling relationship with the target sensitive parameter are finely adjusted by running closed-loop simulation. The updated parameter set, including the actual correction values ​​and compensatory fine-tuning, is solidified into a parameterized virtual behavioral twin, completing the model update and generating a calibrated virtual device model that can characterize the current physical state of the fault indicator under test.

9. The multi-dimensional mapping detection method for the transient recording accuracy of a fault indicator according to claim 1, characterized in that, After generating the calibrated virtual device model, the following steps are also included: Obtain verification stimuli independent of the dynamic accuracy stress test stimulus sequence; The verification stimulus is simultaneously input into the fault indicator under test and the calibrated virtual device model to obtain the actual response and the simulated response. Calculate the degree of consistency between the actual response and the simulated response; The validity of the calibrated virtual device model is confirmed based on whether the degree of consistency exceeds a preset threshold. Receive real transient waveform data collected by the fault indicator under test during field operation; The actual transient waveform data is input into the calibrated virtual device model to obtain the predicted response waveform; By comparing the predicted response waveform with the actual transient waveform data, the accuracy and stability of the fault indicator under test in a real operating environment are evaluated.

10. A multi-dimensional mapping detection system for the transient recording accuracy of a fault indicator, characterized in that, include: The virtual behavioral twin construction module obtains the principle model and nominal parameter set of the fault indicator under test, and constructs a parameterized virtual behavioral twin based on the principle model and nominal parameter set; The sensitive parameter identification and extraction module identifies and extracts the most sensitive parameters that respond to changes in external stimuli from the parameterized virtual behavior twin. The excitation sequence generation module generates dynamic precision stress test excitation sequences based on sensitive parameters. The data acquisition and synchronization module synchronously inputs the dynamic precision stress test excitation sequence to the fault indicator under test and the parameterized virtual behavior twin, and synchronously acquires the first response sequence of the fault indicator under test and the second response sequence of the parameterized virtual behavior twin. The trajectory mapping analysis module performs double-loop trajectory mapping based on the first response sequence and the second response sequence to obtain trajectory deviation characteristics; The diagnostic result generation module obtains the mapping relationship between trajectory deviation features and sensitive parameters, analyzes the trajectory deviation features based on the mapping relationship, locates the target sensitive parameters that cause the deviation, and generates diagnostic results containing parameter offsets. The virtual device model calibration module updates the parameters of the parameterized virtual behavior twin based on the parameter offset in the diagnostic results, and generates a calibrated virtual device model.

Images