Intelligent measuring switch with fault self-recognition and risk self-warning functions

Abstract

This invention provides an intelligent measurement switch with fault self-identification and risk self-early warning functions. By dynamically increasing the sampling rate under natural operation event triggering, it accurately acquires and preprocesses transient response waveforms. It extracts intrinsic response features using multi-level wavelet packet decomposition and energy distribution analysis, and then generates typical fingerprints and feature boundaries to identify the device status through fingerprint clustering and discreteness evaluation. Combined with a preset lookup table, it achieves dynamic adaptive adjustment of the fault judgment threshold, thereby enabling online self-calibration of the diagnostic model. This invention solves the problem of increased diagnostic errors caused by environmental changes and aging of equipment electrical performance, and achieves high precision, real-time performance, and environmental adaptability in equipment fault judgment under different operating conditions, significantly improving the reliability and intelligence level of diagnostic results.

Description

Technical Field

[0001] This invention relates to the field of intelligent power equipment measurement and adaptive fault diagnosis technology, and in particular to an intelligent measurement switch with fault self-identification and risk self-early warning functions. Background Technology

[0002] Currently, in the field of intelligent measurement switch fault diagnosis, identification schemes based on electrical characteristic thresholds are commonly used. These schemes mainly compare key electrical parameters such as overcurrent rise rate, zero-point offset, and harmonic distortion rate with preset static thresholds to determine abnormal states. Most of these methods complete multi-condition calibration and solidify the thresholds before the equipment leaves the factory. Although some schemes are optimized with environmental compensation or sliding statistical correction, they still rely heavily on external parameter inputs and manual experience settings. They lack the ability to adaptively adjust to the long-term operating status of the equipment and cannot meet the needs of long-term, stable, and autonomous monitoring in the field.

[0003] In practical engineering applications, existing technologies exhibit significant limitations:

[0004] First, the diagnostic model relies excessively on external environmental sensing data such as temperature, humidity, and load rate, as well as cloud-based modeling and calculations. This requires continuous uploading of large amounts of data, resulting in high computational costs, high response latency, and an inability to achieve lightweight, localized, and autonomous operation on embedded terminals. Second, the fault identification process uses static thresholds throughout, which cannot be dynamically adjusted to adapt to changes in the intrinsic electrical characteristics caused by factors such as equipment aging and temperature rise drift. This easily leads to misjudgments and missed judgments, making it difficult to guarantee diagnostic accuracy and reliability. Third, threshold updates rely on manual verification or remote updates, lacking an online self-calibration mechanism driven by the equipment's own operating behavior. This results in lagging strategy updates, insufficient system robustness, and an inability to adapt to complex and ever-changing field conditions.

[0005] The aforementioned problems collectively result in traditional fault diagnosis methods exhibiting poor adaptability and insufficient stability during long-term operation, making it difficult to meet the engineering requirements of intelligent measurement switches for high reliability, high real-time performance, and high adaptability. Summary of the Invention

[0006] In order to solve the above-mentioned technical problems, the present invention provides an intelligent measurement switch with fault self-identification and risk self-warning functions.

[0007] The technical solution of this invention is implemented as follows: an intelligent measuring switch with fault self-identification and risk self-warning functions, wherein the intelligent measuring switch achieves diagnosis and warning through the following methods:

[0008] S1: Acquire the high sampling rate electrical transient response waveform of the intelligent measurement switch under natural operation event triggering to form the original transient signal sequence to be processed;

[0009] S2: Perform fixed-order wavelet packet decomposition on the original transient signal sequence to generate a set of decomposition coefficients containing multi-band energy distribution information;

[0010] S3: Based on the decomposition coefficient set, extract the energy distribution entropy, main frequency offset and first half-wave rise time consistency index within the preset frequency band to construct an intrinsic response fingerprint characterizing the characteristics of a single event.

[0011] S4: Input multiple consecutive intrinsic response fingerprints into the local fingerprint clusterer for aggregation operation to generate typical fingerprint cluster centers and discrete boundaries that characterize the current device operating state;

[0012] S5: Based on the numerical value of the discreteness boundary, query the fixed fingerprint-threshold calibration lookup table to generate dynamic fault judgment tolerance parameters that adapt to the current electrical characteristic drift level;

[0013] S6: Update the register configuration of the fault identification module using the dynamic fault determination tolerance parameter to replace the original static feature threshold and complete the online self-calibration of the diagnostic model;

[0014] S7: Based on the updated dynamic fault judgment tolerance parameters, compare and judge the real-time collected electrical characteristic data to output a preliminary diagnostic result with environmental adaptability;

[0015] S8: Monitor the stability of the preliminary diagnostic results and feed them back to the local fingerprint clusterer to drive the cumulative update of the intrinsic response fingerprints and the dynamic correction of the dispersion boundary in the next round.

[0016] The present invention also provides an intelligent measurement switch with fault self-identification and risk self-warning functions. The intelligent measurement switch with fault self-identification and risk self-warning functions is used to perform fault self-identification and risk self-warning of the intelligent measurement switch.

[0017] The present invention provides an intelligent measuring switch with fault self-identification and risk self-warning functions, which has the following beneficial effects:

[0018] (1) The electrical equipment condition monitoring implementation path proposed in this invention avoids the perception and modeling process of external variables. It does not introduce environmental parameters such as temperature and humidity, load rate, and equipment service life, nor does it use neural network fitting or generative models for trend prediction. Instead, it focuses on non-fault electrical events triggered by the physical behavior of the intelligent measurement switch itself, such as normal opening and closing, periodic load switching, and communication-triggered internal logic jumps. These naturally occurring operations generate transient response waveforms with high repeatability and individual specificity in the current / voltage sampling sequence. By using a lightweight module on the edge side to start high sampling rate capture only at the moment the event occurs, and combining fixed-order wavelet packet decomposition to extract three types of low-dimensional features, namely energy distribution entropy, main frequency offset, and first half-wave rise time consistency, a "genuine response fingerprint" is formed, realizing unsupervised and self-triggered characterization of the equipment's intrinsic electrical characteristics. This mechanism does not require continuous sampling and storage of full data, significantly reducing power consumption and storage pressure, while avoiding statistical dependence on historical data distribution, fundamentally overcoming the risk of misjudgment caused by environmental drift or changes in operating conditions in traditional methods.

[0019] (2) This invention establishes the dynamic adjustment of fault criteria based on local fingerprint clustering results: the intrinsic response fingerprints generated by multiple consecutive events are sent to the edge clusterer to generate typical fingerprint cluster centers and their dispersion boundaries under the current operating conditions in real time; the dispersion directly reflects the consistency level of the electrical response of the equipment. When the equipment is in a stable state, the fingerprints are highly concentrated, and the system appropriately relaxes the tolerance based on this to reduce false alarms; when the equipment ages more or the contact deteriorates, the fingerprint dispersion increases, and the system automatically tightens the judgment tolerance of key features (such as overcurrent rise rate and zero-point offset) to improve sensitivity. This threshold adaptive mechanism driven by the evolution of the equipment's own behavior has a mapping relationship based on the fingerprint-threshold calibration experiment completed before leaving the factory under multiple temperature zones, multiple loads, and multiple life stages, which is solidified into a read-only lookup table. The edge only needs to perform a table lookup operation to complete the threshold update. The whole process takes less than 10 milliseconds and does not require any model inference, parameter tuning, or cloud collaboration. Compared to traditional solutions that rely on complex algorithm iterations or remote feedback, this approach achieves true local closed-loop adaptive adjustment, featuring ultra-low latency, zero communication overhead, and strong robustness. It is particularly suitable for field terminal deployment scenarios with no stable network coverage and limited power supply capabilities. Attached Figure Description

[0020] Figure 1 A flowchart of an intelligent measurement switch with fault self-identification and risk self-warning functions according to the present invention;

[0021] Figure 2 This is a sub-flowchart of an intelligent measurement switch with fault self-identification and risk self-early warning functions according to the present invention;

[0022] Figure 3This is another sub-flowchart of an intelligent measurement switch with fault self-identification and risk self-early warning functions according to the present invention. Detailed Implementation

[0023] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0024] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0025] like Figure 1 As shown, this invention provides an intelligent measurement switch with fault self-identification and risk self-warning functions. The intelligent measurement switch achieves diagnostic warning through the following methods:

[0026] S1: Acquire the high sampling rate electrical transient response waveform of the intelligent measurement switch under natural operation event triggering to form the original transient signal sequence to be processed;

[0027] S2: Perform fixed-order wavelet packet decomposition on the original transient signal sequence to generate a set of decomposition coefficients containing multi-band energy distribution information;

[0028] S3: Based on the decomposition coefficient set, extract the energy distribution entropy, main frequency offset and first half-wave rise time consistency index within the preset frequency band to construct an intrinsic response fingerprint characterizing the characteristics of a single event.

[0029] S4: Input multiple consecutive intrinsic response fingerprints into the local fingerprint clusterer for aggregation operation to generate typical fingerprint cluster centers and discrete boundaries that characterize the current device operating state;

[0030] S5: Based on the numerical value of the discreteness boundary, query the fixed fingerprint-threshold calibration lookup table to generate dynamic fault judgment tolerance parameters that adapt to the current electrical characteristic drift level;

[0031] S6: Update the register configuration of the fault identification module using the dynamic fault determination tolerance parameter to replace the original static feature threshold and complete the online self-calibration of the diagnostic model;

[0032] S7: Based on the updated dynamic fault judgment tolerance parameters, compare and judge the real-time collected electrical characteristic data to output a preliminary diagnostic result with environmental adaptability;

[0033] S8: Monitor the stability of the preliminary diagnostic results and feed them back to the local fingerprint clusterer to drive the cumulative update of the intrinsic response fingerprints and the dynamic correction of the dispersion boundary in the next round.

[0034] Step S1: Acquire the high-sampling-rate electrical transient response waveform of the intelligent measurement switch under natural operation event triggering, in order to form the original transient signal sequence to be processed. Specifically, this includes:

[0035] S1.1: Real-time correlation monitoring and processing of the logic state register inside the intelligent measurement switch with the instantaneous sampled values ​​of current and voltage to generate a natural operation event trigger signal containing the opening and closing action identifier or load change characteristics;

[0036] Periodically read the logic status register configured inside the intelligent measuring switch to collect data bytes that identify the current status of each physical contact and control logic unit;

[0037] The instantaneous output of the logic state register is synchronously associated with the instantaneous sampled value of the current / voltage sampling unit, and the register state is kept consistent with the electrical sampling time reference through an internal clock pulse alignment mechanism.

[0038] In the synchronized and associated dataset, the status flag bit change detection algorithm is used to capture the register displacement pattern corresponding to the contact opening and closing action or control logic jump, and to achieve double cross-verification in conjunction with the abrupt change characteristics of electrical sampling values.

[0039] Based on the verified state change events, load mutation detection processing is performed. The rate of change of instantaneous amplitude is calculated in the current sampling sequence, and the mutation rate metric is constructed using the following formula:

[0040]

[0041] in, Represents the current sampling amplitude function. At the current sampling time, The sampling interval;

[0042] The mutation rate metric is compared with a preset rate threshold to filter out rate peaks that match the characteristics of natural operational events.

[0043] For events that meet the peak conditions, perform event identifier encapsulation processing, and output the opening / closing action identifier or load change identifier in a standardized event signal format as a natural operation event trigger signal;

[0044] By real-time correlation monitoring and processing of status registers and instantaneous sampled values, the register status data and electrical sampled data from the previous step are transformed into event trigger signals containing specific action properties and electrical fluctuation patterns, thereby enabling the construction of precise trigger conditions for high sampling rate capture windows.

[0045] For example, in a smart metering switch with a rated voltage of 12kV and a rated current of 630A, the logic state register is read every 50 microseconds, and the current / voltage sampling unit operates at a conventional 10 kHz sampling rate. During synchronous correlation processing, the current sampling abrupt change rate metric corresponding to the register state change is calculated. A / s, sampling interval for s, applying the above formula, yields the instantaneous rate of change as A, where the preset rate threshold is A. The event meets the peak rate condition. The encapsulated event signal contains a "1" for the opening and closing action flag and a "0" for the load change flag. The output format is a 16-bit structured data word, which is sent to the high sampling rate acquisition module via the link. In actual verification, this mechanism can trigger the high sampling rate acquisition configuration within 1 millisecond of the physical action of the switch, effectively capturing transient response waveforms and significantly improving the accuracy and real-time performance of operation event detection.

[0046] S1.2: Based on the natural operation event trigger signal, the sampling frequency control parameters of the embedded data acquisition module are dynamically switched to instantly increase the conventional monitoring sampling rate to a high sampling rate configuration state at the megahertz level.

[0047] Based on the natural operation event trigger signal output by step S1.1, the sampling frequency control register of the embedded data acquisition module is detected in real time to confirm that the module is currently in a low sampling rate configuration state under normal monitoring.

[0048] Perform parameter loading preprocessing on the sampling frequency control register, combine the control parameters required for high sampling rate mode into a configuration word that can be directly written to the register, and include key fields such as target sampling frequency, sampling channel synchronization flag and data buffer depth in the configuration word;

[0049] By using the trigger signal to drive the register write interface, the transient switching of the sampling frequency control parameters is completed, thereby increasing the sampling frequency of the data acquisition module from the power frequency level to the megahertz level, so as to enhance the ability to capture details of transient response signals.

[0050] The sampling frequency after switching is calibrated by actual measurement. By comparing the sampling clock count value with the expected value, a calibration report confirming the accuracy of the sampling frequency is generated, and compensatory adjustments are performed when a deviation is detected.

[0051] Based on the calibrated high sampling rate configuration, the high-speed buffer management strategy of the data acquisition module is activated to ensure that data is not lost due to buffer overflow while the sampling rate is increased.

[0052] Through the above processing method, the natural operation event trigger signal of the previous step is transformed into an execution command to control the sampling rate, thereby improving the transient sampling rate and providing hardware conditions for high-precision transient capture.

[0053] For example, in a smart measurement switch, the sampling frequency in the normal monitoring mode is set to 1024 points per second, and the current value in the sampling frequency control register is 1024. After a natural operation event trigger signal arrives, the system generates a configuration word with a target sampling frequency value of 1000000, including a sampling channel synchronization flag of "1" and a buffer depth of 8192 points. Upon receiving this configuration word, the register write interface switches the sampling frequency from 1024 to 1000000, and verifies the sampling period using an internal clock count within the following 10 milliseconds. The actual counting result is... With a sampling rate of 2 points per second and a deviation of only 2 points per second, the accuracy requirement is met. The high-speed buffer management strategy automatically allocates dual buffers for alternating sampling and data forwarding, ensuring that no data frames are lost during the 5-millisecond transient process triggered by the capture switch contact action. The final high-sampling-rate transient signal completely covers the transient oscillation period, providing clean and high-resolution raw data for the windowed capture of S1.3.

[0054] S1.3: Utilize the high sampling rate configuration state to perform windowed interception processing on the weak electrical transient process triggered by the instantaneous action of the switch contacts, so as to form an original transient signal sequence covering the complete transient oscillation cycle;

[0055] S1.4: Perform DC component removal and power frequency fundamental wave suppression filtering on the original transient signal sequence to eliminate background noise interference and generate clean high-frequency transient response waveform data to be processed;

[0056] S1.5: Perform amplitude normalization and time-domain alignment standardization processing on the high-frequency transient response waveform data to be processed, so as to output a standard original transient signal sequence with device-specific characteristics that can be directly used for feature extraction.

[0057] Step S2: Perform fixed-order wavelet packet decomposition on the original transient signal sequence to generate a set of decomposition coefficients containing multi-band energy distribution information. Specifically, this includes:

[0058] S2.1: The original transient signal sequence is discretized and digitally sampled and reconstructed to generate a discrete transient voltage and current dual-channel digital signal sequence with a unified time base, which serves as the standard input data source for the wavelet packet decomposition algorithm.

[0059] A unified digital sampling control parameter is set for the original transient signal sequence obtained by the S1.5 step standardization process to clarify the sampling interval and sampling accuracy indicators;

[0060] For the voltage channel, the embedded sampling control unit is invoked to set the sampling frequency and time reference, so that each sampling point is strictly synchronized with the preset clock source;

[0061] For the current channel, the same time base binding operation is performed, and the sampling trigger signal is aligned with the sampling point number of the voltage channel to achieve a synchronization structure for the two channels;

[0062] Interpolation reconstruction is performed on the dual-channel sampling results, and a bandpass-limited linear interpolation algorithm is used to correct the waveform distortion caused by the sampling gap during the natural operational event.

[0063] Quantization mapping is performed on the interpolated and reconstructed dual-channel data to convert the analog amplitude into a digital code with a preset bit width, while retaining the amplitude precision to a specified number of decimal places;

[0064] The dual-channel digital sampling point numbers are uniformly mapped using a time base association table to generate a discrete transient voltage and current dual-channel digital signal sequence with a unified time base, which serves as a standardized input data source for the wavelet packet decomposition algorithm.

[0065] Through the above discretized digital sampling and reconstruction process, the standard original transient signal from the previous step is transformed into a dual-channel digital signal with consistent structure, time synchronization, and controllable accuracy, thereby achieving the input normalization required for wavelet domain decomposition.

[0066] For example, in a single device natural opening and closing event, the voltage channel is set to a sampling frequency of [frequency value missing]. Hertz, current channel sampling frequency set to Hertz Hertz, and sampling point alignment is achieved through phase-locked loop (PLL) of the system clock. The original sampling sequences of the two channels respectively contain For each sampling point, third-order linear interpolation is used for reconstruction to fill the sampling gaps into a continuous time series. The analog amplitude is quantized and mapped using a 12-bit ADC, and the voltage channel corresponds to the full-scale range. Encoding step size, current channel full scale corresponding to the same step size, quantization accuracy controlled within Minimum encoding unit. The time base association table will assign the voltage channel number... To the The sampling points correspond to the same numbered positions as the current channels, so that each sampling point forms... and The synchronized pair. The completed discrete transient voltage and current dual-channel digital signal sequence showed high stability in wavelet packet decomposition tests, and the calculation error of the decomposition coefficients was significantly reduced, verifying the effective supporting role of the synchronized discretization processing for subsequent frequency domain feature extraction;

[0067] S2.2: Perform a multi-level wavelet packet tree decomposition operation with a preset number of decomposition layers based on the discrete transient voltage and current dual-channel digital signal sequence to generate a multi-level wavelet packet node coefficient matrix set covering the entire frequency band;

[0068] Based on the discrete transient voltage and current dual-channel digital signal sequence with a unified time base output from the preceding step S2.1, the initial input matrix required by the wavelet packet decomposition algorithm is determined, and the preset wavelet basis type and decomposition level parameters are loaded.

[0069] The dual-channel digital signal sequence is input into the wavelet packet decomposition processing module. The full-band splitting operation is performed on the signal of each channel in the first decomposition layer to generate the corresponding parent node coefficient matrix. The index mapping is retained during the splitting process for subsequent frequency band positioning.

[0070] Using the coefficient matrix of each node output from the first decomposition layer as input, a sub-band splitting operation is performed on the specified node in the second decomposition layer to ensure that the splitting range meets the preset frequency coverage target, and the cross-band leakage effect is eliminated through boundary control parameters.

[0071] The node coefficient matrix of the second decomposition layer is iteratively split to the preset final decomposition layer number. After each layer split, coefficient precision truncation is performed to reduce the impact of floating-point calculation errors on energy distribution. During the iteration process, a recursive tree index structure is used to clarify the coordinate position of each node in the entire frequency band.

[0072] At the end of the decomposition process, the wavelet packet node coefficient matrix sets output by each decomposition level are globally indexed and sorted according to a tree structure, and the corresponding frequency band ranges are marked to form a multi-level wavelet packet node coefficient matrix set covering the entire frequency band.

[0073] By performing multi-level wavelet packet tree decomposition, the discrete transient voltage and current dual-channel digital signal sequence is transformed into a complete set of multi-level node coefficient matrices, enabling multi-scale and multi-frequency band analysis of the original transient response, and providing a structured input basis for subsequent high-frequency noise removal and frequency band energy calculation.

[0074] For example, in a natural opening and closing event acquisition of a smart metering switch, the input signal length of the discrete transient voltage channel and current channel is set to 1024 points, the sampling frequency is 1MHz, Daubechies-8 is selected as the wavelet basis, and the preset decomposition level is 3. In the first decomposition level, the voltage channel signal is split into two node coefficient matrices, corresponding to frequency bands of 0~250kHz and 250~500kHz, respectively; the current channel is split in the same way. In the second decomposition level, the 0~250kHz node is further split into two nodes of 0~125kHz and 125~250kHz, and the high-frequency band is also split into nodes of 250~375kHz and 375~500kHz, forming four sub-frequency band nodes. In the third decomposition level, the 0~125kHz frequency band is further split into nodes of 0~62.5kHz and 62.5~125kHz, recursively completing the splitting of the full-frequency band sub-bands. In each splitting step, the coefficient matrix is ​​preserved to 6 decimal places with floating-point precision, and the frequency coordinates of each node are recorded using a recursive tree index. Taking the coefficient matrix of the voltage channel from 0 to 62.5 kHz as an example, the energy calculation formula is used:

[0075]

[0076] in For node energy; This refers to the nth wavelet coefficient of this node; The total number of node coefficients is given. The node energy value is obtained by the sum of squares of the node coefficients. This value is used for subsequent energy normalization mapping. Finally, a complete set of multi-level wavelet packet node coefficient matrices is output, realizing the full-band structured analysis of transient signals. The verification results show that this decomposition method can significantly improve the stability and distinguishability of frequency band energy features.

[0077] S2.3: Perform threshold denoising filtering on the high-frequency noise interference components in the multi-level wavelet packet node coefficient matrix set to generate a clean wavelet packet node coefficient matrix set after filtering out background electromagnetic interference.

[0078] S2.4: Perform frequency band energy rearrangement and normalization mapping calculation based on the pure wavelet packet node coefficient matrix set to generate a standardized frequency band energy distribution vector characterizing the relative energy intensity of each frequency band;

[0079] S2.5: Perform structured encapsulation processing on the standardized frequency band energy distribution vector to generate a set of decomposition coefficients containing complete multi-frequency band energy distribution information, which serves as the direct feature input for constructing the intrinsic response fingerprint.

[0080] like Figure 2As shown, step S3 involves extracting energy distribution entropy, dominant frequency offset, and first half-wave rise time consistency index within a preset frequency band based on the decomposition coefficient set, in order to construct an intrinsic response fingerprint characterizing the properties of a single event. Specifically, this includes:

[0081] S3.1: Normalize the energy distribution information of each frequency band contained in the decomposition coefficient set to generate a standardized frequency band energy probability distribution sequence and eliminate the interference of absolute amplitude difference caused by load fluctuation.

[0082] S3.2: Perform Shannon entropy calculation based on the standardized frequency band energy probability distribution sequence to generate an energy distribution entropy value that characterizes the signal spectral complexity and disorder, and quantify the spectral structure stability of the current transient response;

[0083] S3.3: Perform spectral centroid shift detection using the peak frequency position within the preset main frequency band in the decomposition coefficient set to generate main frequency shift data reflecting the slight changes in the inductance and capacitance parameters of the device, thereby capturing the subtle drift trend of electrical characteristics;

[0084] S3.4: Perform zero-crossing and peak time difference measurement on the first half-wave period in the original transient signal sequence to generate the original metric value of the first half-wave rise time, and process the original metric value of the first half-wave rise time of multiple consecutive events through the sliding variance algorithm to generate the first half-wave rise time consistency index characterizing the repeatability and consistency of the waveform.

[0085] S3.5: The energy distribution entropy value, the main frequency offset data, and the first half-wave rise time consistency index are vector-concatenated and encapsulated to generate an intrinsic response fingerprint vector containing multi-dimensional feature information, which serves as a unique digital identifier for the characteristics of a single event.

[0086] like Figure 3 As shown, step S4 involves inputting multiple consecutive intrinsic response fingerprints into a local fingerprint clusterer for aggregation operations to generate typical fingerprint cluster centers and dispersion boundaries characterizing the current device operating state. Specifically, this includes:

[0087] S4.1: Obtain multiple consecutive intrinsic response fingerprint vectors generated by the previous steps, and perform sliding time window filtering on the intrinsic response fingerprint vectors to remove outlier abnormal fingerprint data caused by communication interference or extreme operating conditions, and generate a valid fingerprint dataset with temporal consistency.

[0088] S4.2: Initialize the iterative calculation parameters of the local fingerprint clusterer based on the effective fingerprint dataset, and use the weighted centroid update algorithm to perform aggregation and convergence operations on the multidimensional feature coordinates in the effective fingerprint dataset to generate typical fingerprint cluster center coordinates that characterize the average state of electrical characteristics under the current aging stage and temperature rise level of the device.

[0089] Clustering initialization is performed on valid fingerprint datasets with temporal consistency to ensure the matching of parameters with input data during iterative computation;

[0090] During initialization, a normalization scaling factor is set based on the numerical range of each dimension feature in the valid fingerprint dataset to eliminate the impact of dimensional differences between features on the clustering convergence speed.

[0091] In the normalized feature space, the initial centroid position of the clusterer is set. The initial centroid position is determined by calculating the arithmetic mean of all valid fingerprint data points.

[0092] The initial centroid vector is assigned a weight coefficient, which is determined based on the stability quantification index of the most recent diagnosis, so that fingerprint data with stronger recent representativeness accounts for a higher proportion in the aggregation calculation.

[0093] Iterative operations are performed based on the weighted centroid update algorithm. Each iteration includes calculating the distance from all valid fingerprint data points to the current centroid position, increasing the weight of data points with smaller distances and recalculating the centroid coordinates, and decreasing the weight of data points with larger distances and adjusting the centroid coordinates.

[0094] The following formula is used to calculate the update of the centroid position:

[0095]

[0096] in, The updated centroid coordinate vector, Let be the weight coefficient of the i-th fingerprint data point. The multidimensional feature coordinates of the i-th fingerprint data point;

[0097] The iterative process continues until the change in the centroid position is less than the preset convergence threshold, and then outputs the typical fingerprint cluster center coordinates that characterize the average state of electrical characteristics under the current aging stage and temperature rise level of the device.

[0098] By using an initialization and weighted centroid iterative update method based on an effective fingerprint dataset, the time-series consistency data filtered in the previous step is transformed into typical fingerprint cluster centers that reflect the mean of the intrinsic characteristics of the equipment under stable operating conditions, thereby achieving the accurate establishment of a clustering reference benchmark.

[0099] For example, in an embedded measurement switch of a medium- and low-voltage power distribution system, the effective fingerprint dataset contains 120 recently collected intrinsic response fingerprint vectors, each vector being 3-dimensional feature data: energy distribution entropy, dominant frequency offset, and first half-wave rise time consistency index, with numerical ranges of [1.5, 2.3], [...]. [0.05, 0.08], [0.92, 0.98]. Normalization scale factors were set to 0.5, 0.01, and 0.05, respectively. The initial centroid positions were calculated using the arithmetic mean. The weighting coefficients were set to 1.2 for recent data points and 0.8 for earlier data points, based on the recent diagnostic stability index. During the iterative update process, the change in centroid position was calculated in each round, using a convergence threshold. As a termination condition, the updated coordinates of the typical fingerprint cluster center are finally obtained. This output serves as the benchmark for Euclidean distance calculation in subsequent steps, resulting in significantly improved clustering stability and fingerprint drift quantization accuracy.

[0100] S4.3: Using the center coordinates of the typical fingerprint cluster as a reference, calculate the Euclidean distance distribution from each intrinsic response fingerprint vector in the effective fingerprint dataset to the center coordinates of the typical fingerprint cluster, so as to quantify the degree of deviation of the single event characteristics from the average state and generate the original distance deviation sequence.

[0101] The coordinates of the center of a typical fingerprint cluster output by step S4.2 are analyzed to clarify the range of multidimensional feature coordinate values ​​for each dimension as a calculation reference benchmark.

[0102] The multidimensional feature coordinates of each intrinsic response fingerprint vector in the effective fingerprint dataset are compared with the coordinates of the center of a typical fingerprint cluster to obtain the deviation vector of each dimension of the feature.

[0103] Applying the Euclidean distance formula to the deviation vector converts the multidimensional deviation coordinates into a single scalar distance value, quantifying the overall deviation of each intrinsic response fingerprint vector relative to the cluster center. The formula is as follows:

[0104]

[0105] in This represents the i-th dimension coordinate value of the fingerprint vector. This represents the corresponding dimensional coordinate value of the cluster center;

[0106] The above distance calculation is applied sequentially to all fingerprint vectors in the valid fingerprint dataset to form the original distance deviation sequence;

[0107] The original distance deviation sequence is serialized and stored to ensure that it is arranged in the order of event occurrence for subsequent statistical analysis and confidence interval estimation.

[0108] By using Euclidean distance as a metric, the clustering results from the previous step are transformed into quantitative numerical data that characterize the degree of deviation of the characteristics of a single event, thus achieving a precise characterization of the fluctuations in the electrical characteristics of the equipment.

[0109] For example, in the operational status monitoring of a medium- and low-voltage distribution switch, the center coordinates of a typical fingerprint cluster are represented by a multidimensional feature vector {0.12, 0.05, 0.18}. A fingerprint vector in the valid fingerprint dataset has coordinates {0.15, 0.07, 0.20}, and the difference vector is {0.03, 0.02, 0.02}. Substituting the difference vector into the Euclidean distance formula: The distance value obtained is 0.0416. This distance value is recorded in the original distance deviation sequence and reflects the slight offset of the event relative to the average state in the subsequent discrete boundary calculation. When the corresponding fluctuation characteristics appear multiple times in the sequence, it indicates that the equipment is aging faster. Tightening the tolerance during dynamic threshold mapping can significantly improve the stability of fault diagnosis.

[0110] S4.4: Perform statistical variance analysis and confidence interval estimation on the original distance deviation sequence, and extract the upper boundary value of the distance distribution based on the preset confidence probability threshold to generate the dispersion boundary value characterizing the fluctuation range and stability of the current equipment electrical characteristics.

[0111] S4.5: The coordinates of the center of the typical fingerprint cluster and the discrete boundary value are structurally encapsulated to generate an aggregated calculation result package containing complete current device operating status features, which serves as the direct input basis for querying the fixed fingerprint threshold calibration lookup table to drive the mapping generation of dynamic tolerance parameters.

[0112] The data type unification process is performed on the typical fingerprint cluster center coordinates and discrete boundary values ​​obtained from the preceding sub-steps, converting their numerical structures into a numerical matrix format that conforms to the multi-dimensional feature encapsulation specification of the embedded processor.

[0113] Metadata injection is performed on the transformed numerical matrix, embedding the device's unique identifier, timestamp information, and clustering calculation round index at the matrix header to ensure that the data source and operation stage can be accurately located during subsequent table lookup and mapping processes;

[0114] A segmented storage encoding method is used to map the center coordinates and discrete boundary values ​​of typical fingerprint clusters to predefined feature fields, which are then arranged according to fixed field length and offset rules to form a structured and persistent aggregated operation result carrier.

[0115] The aggregated operation result carrier is processed to generate a check code. The CRC-16 algorithm is used to perform cyclic redundancy check calculation on the entire carrier to generate an integrity check field and append it to the end of the carrier to ensure that no data is corrupted or misaligned during the table lookup process.

[0116] The aggregation result carrier of the additional integrity verification field is encapsulated into an input packet format for the lookup module, and the data interaction protocol specified by the fixed fingerprint threshold mapping table is followed. The result packet is configured as a data structure that can be directly called by the lookup operation.

[0117] Through structured encapsulation, the aggregation result of the previous step is transformed into a result package with a complete description of the device's operating status characteristics, thus achieving the expected technical effect of providing a direct input basis for the generation of dynamic tolerance parameter mapping.

[0118] For example, in the embedded instruction processor of an intelligent measurement switch, the coordinates of the typical fingerprint cluster centers obtained from clustering calculation are set to [0.82, 0.76, 0.91], and the discreteness boundary value is set to 0.14. When constructing the numerical matrix, a 4×1 matrix is ​​generated in the form of three-dimensional cluster centers plus one-dimensional discreteness, and the device ID code "SWCH-202306", event timestamp 1685521200, and clustering round index 12 are embedded in the matrix header. In the segmented storage encoding, the three-dimensional cluster centers are written into feature fields F1-F3, each field being 16 bits long with offsets of 0, 16, and 32 bits respectively; the discreteness boundary is written into field F4, with a length of 16 bits and an offset of 48 bits. A checksum is generated using CRC-16 verification. = This is then appended to the end of the field. The encapsulated input packet structure is 64 bytes in size, conforming to the standard format of the fixed fingerprint threshold mapping table lookup interface. After the lookup interface reads this result packet, the dynamic tolerance parameter mapping module can accurately extract the cluster center and dispersion boundary and perform mapping operations to obtain a tolerance configuration value adapted to the current state. In actual verification, this can significantly improve the accuracy of fault identification when the state offset is small.

[0119] Step S5: Based on the numerical value of the dispersion boundary, query the fixed fingerprint-threshold calibration lookup table to generate dynamic fault judgment tolerance parameters adapted to the current electrical characteristic drift level. Specifically, this includes:

[0120] S5.1: Obtain the current typical fingerprint cluster dispersion boundary value output by the local fingerprint clusterer, and perform normalization and quantization processing on the dispersion boundary value to generate a standardized drift level index value, which serves as the input basis for subsequent table lookup operations.

[0121] The discrete boundary value of the current typical fingerprint cluster is obtained from the local fingerprint clusterer. This value is then loaded into the normalization processing module in the embedded computing unit as the input object to ensure that the discrete boundary under different devices or different cluster sizes has a unified and comparable benchmark.

[0122] In the normalization processing module, linear interval mapping is performed based on the preset global maximum and minimum values ​​of dispersion to convert the original dispersion boundary values ​​into dimensionless scaling coefficients, ensuring the accuracy of subsequent drift degree classification index.

[0123] Perform quantization and grading operations on the normalized scaling coefficients, and map the scaling coefficients to integer level values ​​according to the fixed drift level classification table to ensure the uniqueness of the address code when looking up the table.

[0124] The generated integer drift level value is passed to the index encapsulation unit, and combined with the current device's unique identifier, it is encapsulated in two fields to form a standardized drift level index value data packet.

[0125] In the output interface logic, the above data packet is formatted according to the protocol to ensure that the drift level index value can be directly used as the input basis for subsequent table lookup operations, so as to achieve correct matching with the fixed fingerprint-threshold calibration lookup table.

[0126] By using linear interval mapping and quantization grading, the discrete boundary values ​​from the previous step are transformed into standardized drift level index values, thus achieving a unified input benchmark for dynamic tolerance parameter mapping.

[0127] For example, in the clustering results of the operating status of a switchgear, the discrete boundary value output by the local fingerprint clusterer is 0.37, the global maximum value is set to 1.00, and the global minimum value is set to 0.05. The embedded computing unit inputs 0.37 into the normalization processing module and uses the following formula to calculate the scaling factor:

[0128]

[0129] The numerator is the difference between the current boundary value and the global minimum, and the denominator is the difference between the global maximum and the global minimum. The calculation result is 0.337, indicating that the current electrical characteristic drift of the device accounts for 33.7% of the total range. The quantization and grading calculation unit compares the proportional coefficient with the set drift level classification table. The classification table specifies that a proportional coefficient of 0.30~0.40 corresponds to a level value of 2, and finally generates a drift level value of 2. The index encapsulation unit concatenates the level value 2 with the device's unique identifier 20240615 to form a data packet {ID:20240615,Level:2}, and outputs it to the lookup table retrieval unit after protocol formatting. The processing effect of this embodiment is to convert the original physical unit value into a unified, directly searchable integer level index value, ensuring the synchronization and comparability of the tolerance parameter update steps between different devices;

[0130] S5.2: Based on the drift level index value, perform address mapping and data retrieval operations in the fixed fingerprint threshold calibration lookup table to extract the original tolerance correction coefficient set that uniquely corresponds to the drift level index value;

[0131] S5.3: Perform a linear weighted operation on the basic static feature thresholds of the fault identification module using the original set of tolerance correction coefficients to generate intermediate dynamic fault judgment tolerance parameters that adapt to the current degree of electrical characteristic drift;

[0132] The original set of tolerance correction coefficients obtained by retrieving the fixed fingerprint threshold calibration lookup table is used as input data. The basic static feature threshold vector stored in the fault identification module register is used as the execution object, and the corresponding parameter mapping relationship is established in the local computing unit.

[0133] Based on the proportional adjustment mode of the correction coefficients of each feature dimension depending on the basic static feature threshold, the product operation is first performed on the basic static feature threshold and the corresponding correction coefficient of each dimension to form the correction contribution vector of each dimension.

[0134] The intermediate-state dynamic fault determination tolerance parameter vector after linear weighting is calculated using the following formula:

[0135]

[0136] in, This is the vector of tolerance parameters for determining intermediate-state dynamic faults. Based on the static feature threshold vector, This is the original set vector of tolerance correction coefficients;

[0137] After the calculations are completed in each dimension, multidimensional correlation detection is performed on the resulting tolerance parameter vector to eliminate cross-dimensional interference caused by inconsistent adjustment magnitudes of different feature dimensions.

[0138] Subsequently, a uniform scaling calculation is performed on the tolerance vector after correlation detection to ensure that the overall tolerance adjustment conforms to the preset drift degree influence weight, using the following formula for the scaling factor:

[0139]

[0140] in, This is the proportionality coefficient. This is the index value for the current drift level. and These are the maximum and minimum values ​​for the drift severity level, respectively;

[0141] The scaling factor is applied to each component of the tolerance vector to complete the generation of intermediate dynamic tolerance weighted by drift level;

[0142] Through the above chain-weighted and proportional adjustment process, the result of the previous step is transformed into intermediate tolerance data that can be directly entered into the amplitude limiting and legality verification stage, thereby realizing targeted dynamic optimization of tolerance parameters.

[0143] For example, in a medium-low voltage intelligent measuring switch that has been in operation for five years, the basic static characteristic threshold vector is set to [120, 5.0, 0.02], corresponding to the overcurrent rise rate, zero-point offset, and harmonic distortion rate, respectively. The original tolerance correction coefficient set obtained by retrieving the fixed calibration lookup table is [0.15, -0.10, 0.05], the drift level index value is 2.0, and the maximum and minimum levels are 5.0 and 0.0, respectively. According to the formula... First, the corrected contribution vector is calculated as [18, -0.5, 0.001], which is then added to the base threshold to obtain the initial intermediate tolerance vector [138, 4.5, 0.021]. Then, according to the scaling factor formula... The proportionality coefficient was found to be 0.4, which, when applied to the tolerance vector, yielded an adjusted tolerance of [55.2, 1.8, 0.0084]. This result demonstrated significantly improved adaptability to electrical characteristic drift caused by equipment aging in subsequent limiting and verification processes, ultimately effectively reducing the probability of false positives and false negatives.

[0144] S5.4: Perform boundary limiting and legality verification on the intermediate dynamic fault judgment tolerance parameter to eliminate abnormal values ​​that exceed the physical allowable range and generate the final valid dynamic fault judgment tolerance parameter.

[0145] S5.5: Encapsulate the final effective dynamic fault determination tolerance parameter into a standard register configuration instruction format to generate a dynamic fault determination tolerance parameter configuration package that can be directly written into the control register of the fault identification module.

[0146] The final effective dynamic fault judgment tolerance parameter after amplitude limiting and legality verification is used as the input object for encapsulation processing. Before the register instruction conversion is executed, it is split into fields and the tolerance values ​​corresponding to different features are mapped into independent low-level data units.

[0147] For each tolerance data unit after splitting, the pre-compiled encoding rule set of the embedded register interface protocol is called to combine the values ​​into an opcode field that conforms to the instruction format through bit operations and mask matching.

[0148] The combined opcode fields are sequentially bound to register addresses, and each field is mapped one-to-one to the physical address space of the fault identification module control register, forming a command data structure with precise write location.

[0149] Parity check code embedding is performed on the command data structure to improve the controllability and anti-interference capability of register writing by adding check code segments;

[0150] The command data structure, opcode field and check code segment are assembled into a data packet of standard register configuration instruction format using structured encapsulation functions. Length and type identifier fields are added to the header of the data packet to ensure that the write process can be quickly identified and executed during register interface parsing.

[0151] This encapsulation process transforms the result of the previous step into a dynamic fault judgment tolerance parameter configuration package that can be directly loaded into the control register of the fault identification module, enabling uninterrupted writing of dynamic threshold replacement.

[0152] For example, in the embedded computing unit of a smart metering switch, the final effective dynamic fault determination tolerance parameter includes three characteristic thresholds: the overcurrent rise rate threshold is... The zero-point offset threshold is The harmonic distortion rate threshold is The system splits this vector into three parts according to feature order. Data unit, each unit occupies The byte address space encodes numerical values ​​into a fixed decimal format through bitwise operations and combines them into a register opcode field. For example, the overcurrent rise rate threshold is encoded as... Binary form. During the address binding phase, the three fields are mapped to register addresses respectively. , and And append the result generated by the XOR operation to the end of the data. The parity check segment is a 1-bit code. The encapsulation function combines all fields, addresses, checksums, header length identifiers, and type identifiers into a single segment with a total length of [value missing]. Byte-level register configuration instruction format configuration package. Tested and executed continuously in field testing. The configuration package remains stable under conditions of load fluctuations and environmental electromagnetic interference. The updated fault identification module significantly improves the sensitivity and anti-false alarm capability under the corresponding drift level.

[0153] Step S6: Update the register configuration of the fault identification module using the dynamic fault determination tolerance parameter to replace the original static feature threshold and complete the online self-calibration of the diagnostic model. Specifically, this includes:

[0154] S6.1: Obtain the dynamic fault judgment tolerance parameter output by the fingerprint-threshold calibration lookup table, and perform data type conversion and bit width alignment processing on the dynamic fault judgment tolerance parameter to generate a standardized tolerance data word that conforms to the embedded register interface standard.

[0155] Obtain the dynamic fault judgment tolerance parameter configuration package generated by the previous step S5.5 as the initial input condition, and identify the original numerical field of the tolerance parameter and its data encoding format carried in the configuration package;

[0156] Data type mapping is performed on the original numerical field to convert it from the storage type (e.g., fixed-point integer, floating-point or half-precision floating-point) output from the lookup table to the standard numerical type supported by the fault identification module register. The fixed type mapping rules ensure that the numerical precision is not lost and the sign bit is correctly preserved.

[0157] Perform bit width alignment operation on the tolerance parameter after type conversion, and fill the high bits or truncate the low bits according to the effective bit width specified by the register interface protocol, so that the tolerance data will remain unsigned overflow or high bit misalignment risk during subsequent bus transmission and register writing.

[0158] By employing data mask logic and check bit generation algorithm, a standardized tolerance data word conforming to the register interface standard is constructed. This data word contains a valid value segment, a reserved flag segment, and an error detection segment, ensuring that fast consistency verification can be performed during register update.

[0159] Through the above chained processing, the dynamic tolerance parameters of the previous steps are transformed from the lookup table output state into standardized tolerance data words that conform to the embedded register writing specification, realizing dual adaptation of data type and bit width, and providing reliable input for subsequent register unlocking and dynamic threshold overlay.

[0160] For example, in an intelligent measurement switch system with a rated voltage of 380V and a rated current of 63A, the final effective dynamic fault judgment tolerance parameter output by the local fingerprint clusterer is 0.003125 amperes per millisecond. This value is encoded in the lookup table as a 16-bit fixed-point integer, and the scaling factor is... This parameter first converts a 16-bit fixed-point integer to a 32-bit floating-point number through type mapping. The conversion formula is as follows: ,in This is the converted tolerance value. The value is a fixed-point original code. At this point, C=3, which, after conversion, yields V=0.0029296875 and matches it to the floating-point format. Then, bit-width alignment is performed, converting the floating-point number to the IEEE 754 single-precision format conforming to the register protocol, occupying the lower end of the 32-bit bus width, with unused bits at the higher end padded with zeros. After masking logic, reserved flag bits and a cyclic redundancy check (CRC) segment are added to form a standardized tolerance data word. Its binary structure includes a 24-bit value segment, a 4-bit reserved segment, and a 4-bit CRC check segment. This data word was successfully read and echoed through the register interface in simulation testing, showing that the tolerance value deviated from the original value by no more than [percentage missing] in the accuracy comparison test. With an ampere per millisecond, the reliability of tolerance updates and data consistency are significantly improved, ensuring the physical correctness and operational stability of subsequent dynamic threshold coverage processes;

[0161] S6.2: Based on the standardized tolerance data word, perform an atomic write protection unlocking operation on the static feature threshold storage area inside the fault identification module to generate a temporary write enable signal that allows threshold updates;

[0162] S6.3: Using the temporary write enable signal, the standardized tolerance data word is mapped to the address space of the key comparison register of the fault identification module to complete the in-situ overlay of the original static feature threshold and generate the updated dynamic threshold configuration vector.

[0163] A mapping table is established for the standardized tolerance data words with bit width alignment completed in the embedded register interface, and the address of the key comparison register is used as the target addressing parameter;

[0164] Based on the temporary write enable signal, the standardized tolerance data word is pushed to the critical comparison register address space inside the fault identification module through the bus interface protocol adaptation layer.

[0165] During register writing, bit-by-bit verification is performed on the mapped data word to ensure that no bit flips or illegal state encodings occur during the writing process, and the register mirror cache is updated after each byte is written.

[0166] Synchronously refresh each key threshold field in the register mirror cache to keep it consistent with the newly written standardized tolerance data words, thereby achieving in-situ coverage of static feature thresholds;

[0167] The refreshed register image cache is structured and encapsulated to form an updated dynamic threshold configuration vector with the current dynamic tolerance parameters.

[0168] Through the above mapping and mirror refresh processing, the result of the previous step is transformed into a dynamic threshold configuration vector that can be used for real-time comparison operations of the fault identification module, thereby realizing online self-calibration of the threshold.

[0169] For example, in the embedded control system of a three-phase intelligent measuring switch, the standardized tolerance data word length of the dynamic fault determination tolerance parameter is... Position, of which high The corresponding overcurrent rise rate threshold is low. The bit corresponds to the zero-point offset threshold. The starting address of the critical comparison register is... The temporary write enable signal is located in the control register. Bit. SPI mode is used when performing bus mapping. Protocol, clock frequency set to MHz, dynamically tolerant data words are written in high-order priority. During the writing process, a CRC-8 check is performed on each byte, with the CRC polynomial being... To ensure data transmission integrity, after the register mirror cache is refreshed, the generated dynamic threshold configuration vector contains two threshold fields: the overcurrent rise rate threshold is updated to... Unit amplitude / ms, zero offset threshold updated to Unit amplitude. This configuration vector significantly improves the adaptability to electrical characteristic drift caused by equipment aging in real-time fault determination, resulting in a substantial reduction in false alarm rate and a significant improvement in diagnostic stability;

[0170] S6.4: Perform cyclic redundancy check calculation on the updated dynamic threshold configuration vector to generate a check pass flag and trigger the reset and restart process of the diagnostic model state machine;

[0171] S6.5: Based on the reset and restart process, load the updated dynamic threshold configuration vector into the real-time comparison engine to activate the new diagnostic model with environmental adaptability and complete the online self-calibration closed loop;

[0172] Based on the updated dynamic threshold configuration vector that has passed cyclic redundancy check and the reset and restart process output signal of the diagnostic model state machine, the dynamic threshold configuration vector is initialized with memory address mapping according to the interface definition of the real-time comparison engine to ensure that the data maintains bit width consistency and timing integrity during the loading process.

[0173] The dynamic threshold configuration vector that has been initialized by mapping is written sequentially into the feature threshold register group inside the real-time comparison engine. At the same time, the synchronous refresh mechanism of the register group is triggered to complete the parallel update of all threshold boundaries within a single clock cycle, avoiding cross-cycle inconsistencies during the refresh process.

[0174] The synchronous refresh flag bit updated by the register group drives the parameter loading unit of the real-time comparison engine, and caches the updated threshold configuration vector into the high-speed computing cache area for immediate use by the multi-dimensional feature determination module of the comparison logic unit.

[0175] After loading the dynamic threshold data in the high-speed computing cache, the rule matching and recompilation process of the real-time comparison engine is started. The logical expressions built on static thresholds in the dynamic fault judgment rule base are replaced item by item to form a self-updating rule set that is adapted to the current device operating status.

[0176] The completion signal of rule matching recompilation triggers the real-time comparison engine to enter full-speed operation mode, enabling it to perform judgment calculations based on the latest dynamic thresholds when receiving real-time electrical feature data, thereby realizing the activation of a new diagnostic model with environmental adaptability and online self-calibration closed loop.

[0177] Through the above loading and activation process, the dynamic threshold configuration vector of the previous step is transformed into a judgment rule that can be effective in the real-time comparison engine, so as to achieve the goal of continuous high-precision operation of the fault identification module under non-static environmental conditions.

[0178] For example, in an intelligent measurement switchgear with a high aging stage and operating under high summer temperatures, the dynamic threshold configuration vector includes an overcurrent rise rate tolerance setting of 1.8, a zero-point offset tolerance setting of 0.0025, and a harmonic distortion rate tolerance setting of 0.035. The real-time comparison engine interface definition requires that each tolerance setting be stored in a 16-bit fixed-point format, with the overcurrent rise rate tolerance mapped as follows: Zero-point offset tolerance mapping is Harmonic distortion rate tolerance is mapped as After interface conversion, the values ​​are sequentially written to the register group, triggering a synchronous refresh. Upon refresh, the computational buffer immediately loads the aforementioned tolerance values. The rule-matching recompilation process replaces the original static threshold judgment statement with a judgment expression based on these values. For example, the overcurrent fault judgment condition is updated to... ,in This is a real-time sampled value of the overcurrent rise rate, with value 1843 corresponding to the tolerance setting after interface conversion. Verification shows that the new diagnostic model maintains significantly improved comparison accuracy and a significantly reduced false alarm rate under this extreme condition. The output accurately identifies instantaneous overcurrent events exceeding the tolerance and responds quickly when a real fault is triggered, achieving the expected technical effect of environmentally adaptable online self-calibration.

[0179] Step S7: Based on the updated dynamic fault judgment tolerance parameters, the real-time collected electrical characteristic data is compared and judged to output a preliminary diagnostic result with environmental adaptability. Specifically, this includes:

[0180] S7.1: Obtain the original voltage and current timing signals output by the real-time sampling circuit, and perform synchronization alignment and noise filtering processing on the original voltage and current timing signals to generate a standardized real-time electrical feature data sequence, providing a clean input reference for subsequent feature extraction;

[0181] The input conditions include the original voltage timing signal and the original current timing signal output by the real-time sampling circuit, as well as the updated dynamic fault judgment tolerance parameters that have been loaded into the fault identification module. The execution object is the synchronously acquired dual-channel analog signal and its trigger timing identifier.

[0182] The original voltage and current timing signals are subjected to sampling clock drift detection respectively. A timestamp sequence is constructed using the period reference of the reference clock source. The sampling offset is dynamically corrected by comparing the timestamp difference between the two channels, thereby achieving synchronization and alignment of the dual-channel signals.

[0183] Preprocessing gain balancing is performed based on the aligned dual-channel signals to ensure that the voltage channel and the current channel do not accumulate errors due to differences in the hardware front end across the amplitude ratio range.

[0184] The dual-channel signal is filtered within a preset effective frequency band using a bandpass filter to remove invalid components below the power frequency fundamental wave and above the transient response frequency band, thus avoiding interference from environmental broadband noise in subsequent feature calculations.

[0185] An adaptive noise suppression operation is performed on the filtered signal, and the suppression weight is adjusted according to the dynamic tolerance parameter to enhance the high-frequency noise suppression when the drift is large, thereby maintaining the stability of the eigenvalues.

[0186] A sliding window mechanism is used to perform phase consistency detection on dual-channel signal segments, eliminating local segments with phase abrupt changes caused by sampling instantaneous anomalies, and restoring missing points through interpolation algorithms to generate a complete and continuous real-time electrical feature data sequence.

[0187] Through the above-mentioned synchronization alignment and noise filtering processing methods, the results of the previous step are transformed into a standardized real-time electrical feature data sequence that simultaneously possesses temporal consistency and spectral purity, thereby achieving the expected technical effect of providing high-quality input data for subsequent multi-dimensional feature calculations.

[0188] For example, in a set of intelligent measurement switches for medium and low voltage power distribution systems, the real-time sampling circuit outputs the original voltage and current signals at a sampling frequency of 256kHz. The dual-channel sampling clock offset detection result is 15 nanoseconds, which is reduced to 0 nanoseconds after timestamp correction, achieving complete alignment. The gain balance coefficient is set to 1.02 to eliminate differences in front-end hardware. The passband range of the bandpass filter is set to 2kHz to 50kHz to ensure complete preservation of transient characteristic frequency bands. The adaptive noise suppression weight is automatically increased to 0.85 when the dispersion boundary is large to enhance noise filtering efficiency. The sliding window length is set to 5 milliseconds to eliminate phase abrupt changes with amplitudes greater than 5 milliseconds in phase consistency detection. The missing sampling points were recovered using cubic spline interpolation. After processing, a dual-channel standardized real-time electrical characteristic data sequence was obtained, with its waveform phase difference maintained at [value missing]. The effective signal-to-noise ratio was increased from 35dB before processing to 52dB after processing, which significantly improved the accuracy of subsequent feature calculations in determining the overcurrent rise rate, zero-point offset and harmonic distortion rate.

[0189] S7.2: Based on the standardized real-time electrical feature data sequence, multi-dimensional feature processing is performed using a predefined time-frequency domain feature extraction algorithm to generate a set of real-time fault discrimination feature vectors containing overcurrent rise rate, zero-point offset and harmonic distortion rate.

[0190] S7.3: Read the dynamic fault judgment tolerance parameter stored in the fault identification module register, and parse the dynamic fault judgment tolerance parameter into an adaptive threshold boundary interval for each dimension in the real-time fault discrimination feature vector set, so as to construct a dynamic fault judgment rule base under the current working condition.

[0191] Read the dynamic fault judgment tolerance parameters written in the online self-calibration step from the register area of ​​the fault identification module, and parse the tolerance parameters item by item according to the predefined dimension labels to form a numerical comparison table that corresponds one-to-one with the dimensions of the real-time fault discrimination feature vector set.

[0192] The parsed tolerance parameters are classified into amplitude, time and spectrum subsets according to their characteristic properties, and the construction mode of the threshold boundary interval is determined according to the sensitivity level of each subset.

[0193] For amplitude-type features, an interval symmetric expansion calculation is performed, using the tolerance value as the half-width of the central interval. Upper and lower limit points are generated around the center point of the original static threshold, and the boundaries are determined by the following formula:

[0194]

[0195] in The center point of the static threshold. This is a dynamic tolerance value;

[0196] For time-related features, an asymmetric interval correction is performed, weighting the tolerance values ​​according to their positive and negative drift trends, and calculating them using the following form:

[0197] and

[0198] in This serves as the static threshold time base. For spectral features, a relative proportional relaxation is applied, converting the tolerance value into a boundary multiplier coefficient. A frequency range extension value is then generated through multiplication mapping, using the following formula:

[0199]

[0200] in This represents the static threshold frequency boundary.

[0201] The above calculation results are combined into rule base record units according to their original dimensional positions, and consistency checks are performed to ensure that the upper and lower limits of the intervals do not overlap and that the numerical meanings are physically feasible.

[0202] By generating and verifying boundaries by category, the tolerance parameters of the previous step are transformed into an adaptive threshold boundary rule base for real-time feature comparison, thereby enabling environmentally adaptive fault determination for different feature dimensions.

[0203] For example, in a machine with a rated current of In An's intelligent measurement switch, the overcurrent rise rate tolerance value is obtained after parsing the register tolerance parameter. Amperes per millisecond, zero-point offset tolerance Second, harmonic distortion rate tolerance The center point of the static threshold for the rate of rise of amplitude-type overcurrent is... Amperes per millisecond, boundary calculations are Get the upper and lower limits and Amperes / milliseconds. Static baseline for time-based zero-point offset. Seconds, after asymmetric correction, yield a positive upper limit. = Seconds, negative lower limit = Seconds. Static boundary of spectral harmonic distortion rate. The upper limit was obtained after the multiplier was relaxed. = This rule base significantly improves drift adaptability in real-time fault identification and comparison, and effectively reduces false alarm rates under short-term load shocks;

[0204] S7.4: Perform a logical comparison operation between each feature value in the real-time fault discrimination feature vector set and the corresponding adaptive threshold boundary interval in the dynamic fault judgment rule base to generate a binary status flag bit sequence representing whether each feature exceeds the limit;

[0205] The feature values ​​of each dimension in the real-time fault discrimination feature vector set are used as the input objects for comparison operation, and the adaptive threshold boundary interval in the dynamic fault judgment rule base is used as the comparison reference standard.

[0206] Perform interval boundary analysis on each feature value and extract the corresponding minimum and maximum allowed values ​​as boundary limit arrays;

[0207] Based on the limit array, the interval inclusion detection operation is performed on the input feature values. The limit-crossing state is determined by judging whether the feature values ​​strictly fall between the minimum and maximum values.

[0208] For double-boundary detection that includes both upper and lower limits, a closed-interval judgment logic is used. That is, when the feature value is less than the minimum allowable value or greater than the maximum allowable value, the judgment result is marked as exceeding the limit. For features that only define a single boundary, a one-sided inequality judgment logic is used. For example, when only an upper limit is set, when the feature value is greater than the upper limit, it is judged as exceeding the limit.

[0209] The detection results of each dimension are converted into binary status flags, where the qualified status is mapped as follows: The out-of-limit state is mapped as And generate a sequence of state flag bits in order of feature dimensions;

[0210] Through the above comparison operation, the dynamic fault judgment rule base generated in the previous step is transformed into a binary status flag sequence that represents whether each feature exceeds the limit and can be used for fusion judgment, so as to realize accurate fault judgment input based on real-time dynamic threshold.

[0211] For example, the real-time fault discrimination feature vector set for a certain medium- and low-voltage distribution switch includes the overcurrent rise rate. Zero offset Harmonic distortion rate The three feature dimensions and their corresponding adaptive threshold boundary intervals in the dynamic fault determination rule base are as follows: , ]、[ , ]、[ , ] Perform closed-interval inclusion detection on the rate of rise of the overcurrent, because lie in[ , Within, the flag is Perform the same detection method on the zero-point offset. lie in[ , Within, the flag is When detecting harmonic distortion rate lie in[ , Within, the flag is The final output binary state flag sequence is [ , , This indicates that none of the features triggered the limit under the current operating conditions, effectively supporting subsequent fusion decision processing, significantly improving the stability of the diagnostic results, and greatly reducing the false alarm rate;

[0212] S7.5: The binary status flag sequence is fused and decided according to the preset fault logic combination strategy to generate a preliminary diagnostic result with environmental adaptability. The result clearly indicates whether the equipment is currently in normal operation or a specific type of fault-triggered state.

[0213] Obtain the binary state flag sequence generated by the preceding sub-step S7.4 as the input dataset for the fusion decision processing, so as to ensure that each flag has a synchronized timing reference and a unified encoding method before logical combination;

[0214] Based on a fixed fault logic combination strategy library, the system calls predefined logic functions for different fault types, maps the input binary status flag sequence to each fault type decision unit, and performs parameter matching and condition verification operations in each decision unit to generate the corresponding single fault judgment result.

[0215] For a single fault determination result set, perform multi-fault mutual exclusion analysis and processing, use Boolean matrix operations to eliminate the contradictory state of multiple faults being triggered simultaneously due to feature overlap, and retain independent fault identifier bits that meet mutual exclusion constraints;

[0216] The fault identifier bit set after mutual exclusion analysis is input into the priority decision module. Priority sorting and highest priority fault type filtering operations are performed according to the preset fault severity level table to determine the most risky fault mode under the current working condition.

[0217] By using the fault mode determination result and the normal operation status flag bit to perform status fusion processing, the fault mode information and status flag bit are structured and encapsulated into a preliminary diagnosis result data packet, realizing the logical chain transformation from binary flag bits to environmentally adaptable preliminary diagnosis results;

[0218] Through the above-mentioned fusion decision processing method, the feature over-limit binarization result of the previous step is transformed into clear status judgment data, realizing accurate identification of the current operating status of the equipment or specific fault types, with dynamic threshold adaptation capability and low false judgment rate.

[0219] For example, in an embedded intelligent measurement switch, the dynamic fault judgment rule base defines the overcurrent rise rate exceeding the limit flag as b1, the zero-point offset exceeding the limit flag as b2, and the harmonic distortion rate exceeding the limit flag as b3. The logic combination strategy specifies that the short-circuit fault judgment condition is (b1 AND b2) and NOT (b3), the ground fault judgment condition is (b2 AND b3) and NOT (b1), and the overload fault judgment condition is (b1 AND NOT (b2) AND NOT (b3)). When the real-time comparison result output flag sequence is [1,1,0], the short-circuit fault identification module is triggered, and the single fault result set is short circuit = 1, ground = 0, and overload = 0. In the mutual exclusion analysis, only the short-circuit fault satisfies the mutual exclusion condition. The priority decision module queries the severity level table, and the short-circuit fault level is higher than other types, and is finally confirmed as a short-circuit fault mode. In the state fusion processing, the short-circuit fault identifier is combined with the normal state flag to output a preliminary diagnostic result data packet. This data packet was used as a diagnostic stability monitoring benchmark in the subsequent closed-loop feedback. The verification results showed that it could still maintain the consistency and effectiveness of fault type determination in scenarios with high temperature rise and equipment aging, which significantly improved the system's environmental adaptability and misjudgment suppression capability.

[0220] Step S8: Monitor the stability of the preliminary diagnostic results and feed them back to the local fingerprint clusterer to drive the cumulative update of the intrinsic response fingerprints and the dynamic correction of the dispersion boundary in the next round. Specifically, this includes:

[0221] S8.1: Obtain the sequence of preliminary diagnostic results within a continuous time window, and perform sliding variance calculation on the sequence of preliminary diagnostic results to generate a quantitative index of diagnostic stability that characterizes the degree of fluctuation in diagnostic output;

[0222] S8.2: Based on the comparison logic between the diagnostic stability quantification index and the preset steady-state threshold range, a state transition trigger signal is generated to indicate whether the current operating state of the device has undergone a significant drift.

[0223] S8.3: Activate the incremental update mode of the local fingerprint clusterer using the state transition trigger signal, and perform weight reallocation operation on the intrinsic response fingerprint set to be aggregated in the cache to generate a weighted intrinsic response fingerprint dataset containing the latest time series information.

[0224] S8.4: Perform cluster center iterative convergence calculation on the weighted intrinsic response fingerprint dataset to update the typical fingerprint cluster center coordinates and discrete boundary values ​​that characterize the current device operating state;

[0225] S8.5: Replace the original discrete boundary register data with the updated discrete boundary value to complete the initial condition configuration for the next round of intrinsic response fingerprint cumulative update cycle and drive the remapping of dynamic fault judgment tolerance parameters.

[0226] The present invention also provides an intelligent measurement switch with fault self-identification and risk self-warning functions. The intelligent measurement switch with fault self-identification and risk self-warning functions is used to perform fault self-identification and risk self-warning of the intelligent measurement switch.

[0227] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0228] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent measuring switch with fault self-identification and risk self-warning functions, characterized in that, The intelligent measurement switch achieves diagnostic early warning through the following methods: S1: Acquire the high sampling rate electrical transient response waveform of the intelligent measurement switch under the triggering of natural operation events, and form the original transient signal sequence to be processed; S2: Perform fixed-order wavelet packet decomposition on the original transient signal sequence to generate a set of decomposition coefficients; S3: Based on the set of decomposition coefficients, extract the energy distribution entropy, the main frequency offset, and the consistency index of the first half-wave rise time within the preset frequency band to construct the intrinsic response fingerprint; S4: Input multiple consecutive intrinsic response fingerprints into the local fingerprint clusterer for aggregation operation to generate typical fingerprint cluster centers and dispersion boundaries; S5: Based on the numerical value of the discreteness boundary, query the fixed fingerprint-threshold calibration lookup table to generate dynamic fault judgment tolerance parameters; S6: Update the register configuration of the fault identification module using the dynamic fault determination tolerance parameter, replace the original static feature threshold, and complete the online self-calibration of the diagnostic model; S7: Based on the updated dynamic fault judgment tolerance parameters, compare and judge the real-time collected electrical characteristic data, and output preliminary diagnostic results; S8: Monitor the stability of the preliminary diagnostic results and feed them back to the local fingerprint clusterer to drive the cumulative update of the intrinsic response fingerprints and the dynamic correction of the dispersion boundary in the next round.

2. The intelligent measuring switch with fault self-identification and risk self-warning functions according to claim 1, characterized in that, The wavelet packet decomposition includes synchronous multi-level decomposition of the current channel and the voltage channel, generating a full-band structured feature coefficient matrix, and obtaining a standardized band energy vector through band energy normalization calculation.

3. The intelligent measuring switch with fault self-identification and risk self-warning functions according to claim 1, characterized in that, Step S3 specifically includes: The energy distribution information of each frequency band contained in the decomposition coefficient set is normalized to generate a standardized frequency band energy probability distribution sequence. Based on the standardized frequency band energy probability distribution sequence, Shannon entropy calculation is performed to generate energy distribution entropy values; Spectral centroid shift detection is performed using the peak frequency position within a preset main frequency band in the set of decomposition coefficients to generate main frequency shift data; For the first half-wave period in the original transient signal sequence, the time difference between the zero-crossing point and the peak point is measured to generate the original metric value of the first half-wave rise time. The original metric value of the first half-wave rise time of multiple consecutive events is processed by the sliding variance algorithm to generate the first half-wave rise time consistency index. The energy distribution entropy value, the main frequency offset data, and the first half-wave rise time consistency index are vector-concatenated and encapsulated to generate an eigenresponse fingerprint vector.

4. The intelligent measuring switch with fault self-identification and risk self-warning functions according to claim 3, characterized in that, The energy distribution entropy value characterizes the signal spectrum complexity and disorder; the main frequency offset data reflects the slight changes in the inductance and capacitance parameters of the equipment; and the first half-wave rise time consistency index characterizes the waveform repeatability and consistency.

5. The intelligent measuring switch with fault self-identification and risk self-warning functions according to claim 1, characterized in that, Step S4 specifically includes: Obtain multiple consecutive intrinsic response fingerprint vectors, perform sliding time window filtering on the intrinsic response fingerprint vectors, and generate a valid fingerprint dataset; The iterative calculation parameters of the local fingerprint clusterer are initialized based on the effective fingerprint dataset, and the aggregation and convergence operation is performed on the multidimensional feature coordinates in the effective fingerprint dataset to generate the center coordinates of typical fingerprint clusters. Using the center coordinates of the typical fingerprint cluster as a reference, the Euclidean distance distribution from each intrinsic response fingerprint vector in the effective fingerprint dataset to the center coordinates of the typical fingerprint cluster is calculated, generating the original distance deviation sequence. Statistical variance analysis and confidence interval estimation are performed on the original distance deviation sequence. Based on the preset confidence probability threshold, the upper boundary value of the distance distribution is extracted to generate the dispersion boundary value.

6. The intelligent measuring switch with fault self-identification and risk self-warning functions according to claim 5, characterized in that, The center coordinates of the typical fingerprint cluster represent the average state of electrical characteristics under the current aging stage and temperature rise level of the device; the discreteness boundary value represents the fluctuation range and stability of the current device's electrical characteristics.

7. The intelligent measuring switch with fault self-identification and risk self-warning functions according to claim 5, characterized in that, Step S4 also includes: The coordinates of the center of the typical fingerprint cluster and the discrete boundary values ​​are structured and encapsulated to generate an aggregation result package.

8. The intelligent measuring switch with fault self-identification and risk self-warning functions according to claim 1, characterized in that, Step S5 specifically includes: Obtain the current typical fingerprint cluster dispersion boundary value output by the local fingerprint clusterer, and perform normalization quantization processing on the current typical fingerprint cluster dispersion boundary value to generate a standardized drift level index value. Based on the drift level index value, address mapping and data retrieval operations are performed in the fixed fingerprint threshold calibration lookup table to extract the original tolerance correction coefficient set that uniquely corresponds to the drift level index value; The original set of tolerance correction coefficients is used to perform a linear weighted operation on the basic static feature thresholds of the fault identification module to generate intermediate dynamic fault judgment tolerance parameters. Boundary limiting and validity checks are performed on the intermediate dynamic fault determination tolerance parameters to generate the final valid dynamic fault determination tolerance parameters.

9. The intelligent measuring switch with fault self-identification and risk self-warning functions according to claim 8, characterized in that, Step S5 also includes: The final effective dynamic fault determination tolerance parameters are encapsulated into a standard register configuration instruction format to generate a dynamic fault determination tolerance parameter configuration package.

10. The intelligent measuring switch with fault self-identification and risk self-warning functions according to claim 1, characterized in that, In step S7, based on the updated dynamic fault judgment tolerance parameters, the electrical characteristic data collected and preprocessed in real time are compared and judged. By comparing the overcurrent rise rate, zero-point offset and harmonic distortion rate with the adaptive threshold boundary interval, a binary status flag is generated, and then the preliminary diagnosis result is output through fusion decision.

Images