Detection device and early warning method for voltage abnormity of bushing end screen of converter transformer

Abstract

The invention relates to the technical field of power equipment on-line monitoring and fault diagnosis, and discloses a converter transformer bushing end screen voltage abnormity detection device and early warning method.The method comprises the steps that voltage data are collected through a wireless intelligent sensor, soft synchronization and self-adaptive windowing are achieved through a commutation notch feature locking mechanism, and a preprocessing signal is output; topological correlation clustering is executed to construct a virtual health reference, a dynamic residual sequence is extracted through time domain difference, and system side and equipment side features are decoupled; constructing a variational mode decomposition model, and calculating a characteristic entropy value of the intrinsic mode function; and comparing the entropy value with a threshold value, triggering an active impedance disturbance verification mechanism, controlling a sensor to access a load and comparing an actual voltage drop rate with a theoretical voltage drop rate, and determining an insulation degradation or poor contact fault. High-precision synchronization can be achieved without a hardware clock, and the problems that weak fault features are difficult to extract and virtual connection and insulation faults are difficult to distinguish in a high-impedance loop are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of online monitoring and fault diagnosis technology for power equipment, specifically to a detection device and early warning method for abnormal voltage at the end screen of a converter transformer bushing. Background Technology

[0002] Converter transformers are core equipment in high-voltage direct current transmission systems, and their operational reliability directly affects the safety and stability of the power grid. Bushings, as a key component of converter transformers, have their end-screen grounding leads serving as important signal sampling points for monitoring the insulation dielectric loss factor and capacitance, as well as crucial grounding points to ensure stable voltage distribution within the bushing capacitor core.

[0003] With the development of ubiquitous power Internet of Things technology, low-power smart sensors based on wireless transmission are gradually replacing traditional wired monitoring devices and becoming the mainstream development direction for online monitoring of bushing end screens due to their advantages such as convenient installation and good electrical isolation.

[0004] However, existing wireless monitoring technologies still face many technical bottlenecks in practical applications, making it difficult to meet the high-precision diagnostic requirements of converter transformers under complex operating conditions. The inherent random delays and packet loss in wireless data transmission result in discrete arrival times of data packets. Furthermore, low-power sensors, limited by cost and battery capacity, are usually difficult to configure with high-precision GPS or BeiDou timing modules, making it impossible to achieve strict physical time synchronization between multiple devices.

[0005] This lack of a time reference renders phase analysis algorithms based on lateral comparisons ineffective, making it impossible to accurately assess state differences between devices.

[0006] In addition, the converter transformer operates in a complex electromagnetic environment with AC and DC hybrid circuits. The periodic commutation operation of the converter valve will generate high-frequency commutation overshoot and rich harmonic components on the valve-side voltage waveform. This common-mode interference energy on the system side is often much greater than the weak discharge or voltage drift signal generated by early insulation defects inside the bushing.

[0007] Existing monitoring methods mostly rely on fixed threshold judgment or simple amplitude trend analysis, lacking effective signal decoupling means. This makes it easy for weak features reflecting specific equipment faults to be drowned out by strong background noise, resulting in missed faults or insufficient detection sensitivity.

[0008] More importantly, the bushing end-screen measurement circuit is a typical high-impedance capacitive circuit, which is sensitive to changes in series resistance. In actual operation, whether it is the equivalent impedance drift caused by the deterioration of the insulating medium or the increase in contact resistance caused by poor contact of the end-screen grounding lead, both will manifest as abnormal voltage fluctuations at the measurement port in passive monitoring mode.

[0009] Existing passive voltage monitoring methods can only sense changes in voltage signals, but cannot detect the physical mechanisms that cause these changes. They are unable to distinguish whether the fault source is irreversible damage to the internal insulation layer or a loose connection in the external leads. This ambiguity in the nature of the fault can easily lead to misdiagnosis and affect the formulation of operation and maintenance strategies. Summary of the Invention

[0010] To address the shortcomings of existing technologies, this invention provides a detection device and early warning method for abnormal voltage at the end screen of a converter transformer bushing. It solves the problems in existing technologies, such as the lack of a high-precision synchronous reference for data acquisition leading to failure of lateral analysis, the easy masking and difficulty in extracting weak fault features under strong electromagnetic interference, and the inability of passive monitoring mode to effectively distinguish between insulation degradation and poor contact faults.

[0011] The first aspect of this invention provides a detection device for abnormal voltage at the end screen of a converter transformer bushing, comprising: Multiple wireless smart sensors are installed at the bushing end screen of each converter transformer to collect the ground voltage data of the bushing end screen and to respond to control commands to perform load connection operations. The main control unit is communicatively connected to the wireless intelligent sensor; The main control unit is configured to: receive the ground voltage data and perform soft synchronization processing to obtain a preprocessed signal; construct a virtual health reference signal based on the preprocessed signal and extract a dynamic residual sequence; perform variational mode decomposition and characteristic entropy calculation on the dynamic residual sequence; send an active disturbance command to the wireless smart sensor when a suspected fault is detected; calculate the actual voltage drop rate based on the feedback effective voltage value; and compare the actual voltage drop rate with the theoretical voltage drop rate to output a fault diagnosis result.

[0012] Furthermore, the wireless intelligent sensor integrates a controllable load module, which includes: The load resistor has a known resistance value; An internal switch is connected in series with the load resistor and then connected in parallel to the signal acquisition circuit. The wireless smart sensor is configured to: upon receiving the active disturbance command, control the internal switch to connect the load resistor to the circuit, and measure the effective voltage value before and after connection.

[0013] The second aspect of this invention provides an intelligent detection and early warning method for abnormal voltage at the end screen of a converter transformer bushing. This method is based on the detection device described in the first aspect and includes the following steps: First, the wireless intelligent sensors are used to collect the ground voltage data of the bushing end screen of each converter transformer. The main control unit uses the commutation gap feature locking mechanism to identify the characteristic synchronization points in the ground voltage data, and performs a soft synchronization mechanism and adaptive windowing operation based on the characteristic synchronization points, and outputs the preprocessed signals of each wireless intelligent sensor.

[0014] Secondly, topological correlation clustering is performed on the preprocessed signal to construct a virtual health reference signal that reflects the common-mode voltage characteristics of the system side. The preprocessed signal and the virtual health reference signal are then subjected to time-domain difference to extract a dynamic residual sequence containing device-side specific features.

[0015] Next, a variational mode decomposition model is constructed to decompose the dynamic residual sequence to obtain intrinsic mode functions, and the characteristic entropy value of the intrinsic mode functions is calculated.

[0016] Finally, the feature entropy value is compared with a preset safety threshold. When the determination result is a suspected poor contact fault, the active impedance disturbance verification mechanism is triggered. The active impedance disturbance verification mechanism controls the wireless smart sensor to perform load access operation by sending an active disturbance command, and outputs the fault diagnosis result based on the comparison result of the actual voltage drop rate and the theoretical voltage drop rate.

[0017] Furthermore, in the above acquisition and preprocessing steps, the specific execution process of the commutation gap feature locking mechanism is as follows: perform a first-order difference operation on the discrete time series of the ground voltage data of the end screen of each converter transformer bushing to obtain the voltage change slope; monitor the voltage change slope, and lock the moment when the voltage change slope first exceeds the preset negative threshold as the feature synchronization point, which corresponds to the starting moment of the leading edge of the commutation gap generated when the converter valve commutates.

[0018] The preset negative threshold is set in advance based on the statistical characteristics of the voltage drop slope at the commutation gap of the converter transformer under normal operating conditions.

[0019] Furthermore, the specific execution process of the soft synchronization mechanism and adaptive windowing operation is as follows: select the wireless smart sensor with the highest signal-to-noise ratio as the reference benchmark, and calculate the time offset of the feature synchronization point locked by the other wireless smart sensors relative to the feature synchronization point of the reference benchmark. Based on the time offset, a time shift correction operation is performed on each of the discrete time series to achieve physical alignment of the signals. An adaptive time window centered on the characteristic synchronization point is constructed, and the data weights of the aligned signals within the adaptive time window are reset to zero or removed, retaining only the stationary segment data to obtain the preprocessed signal.

[0020] The width of the adaptive time window is dynamically set based on the operating trigger angle and commutation overlap angle of the converter transformer.

[0021] Furthermore, in the above-mentioned virtual benchmark reconstruction and residual generation steps, the process of constructing the virtual health benchmark signal and extracting the dynamic residual sequence is as follows: calculate the Pearson correlation coefficient between each pair of the preprocessed signals, and calculate the normalized weight coefficient of each preprocessed signal based on the Pearson correlation coefficient. The preprocessed signals from multiple sources are weighted and synthesized using the normalized weighting coefficients. The virtual health baseline signal is then calculated based on the virtual health baseline signal construction formula. The preprocessed signals from each wireless smart sensor are then subjected to point-by-point time-domain difference analysis with the constructed virtual health baseline signal. The dynamic residual sequence is obtained by subtracting the virtual health baseline signal from the preprocessed signals using the dynamic residual sequence extraction formula.

[0022] Furthermore, in the above decomposition and feature extraction steps, the process of constructing and solving the variational mode decomposition model and calculating the feature entropy value is as follows: by solving the constrained variational problem, a set of intrinsic mode functions and their center frequencies are found such that the sum of the estimated bandwidths of each mode is minimized, thereby establishing the variational mode decomposition model; Based on the variational mode decomposition model, and according to the variational mode decomposition objective function formula, the alternating direction multiplier method is used for iterative solution to obtain multiple intrinsic mode functions corresponding to the dynamic residual sequence; from the multiple intrinsic mode functions obtained by decomposition, the intrinsic mode functions in the high frequency band are selected, and phase space reconstruction is performed to generate reconstruction vectors. The relative probabilities of different permutation patterns appearing in the reconstruction vectors are statistically analyzed, and the feature entropy value is calculated according to the permutation entropy feature calculation formula.

[0023] Furthermore, in the above-mentioned anomaly detection and verification steps, the detection logic that triggers the active impedance perturbation verification mechanism is: comparing the feature entropy value with the preset security threshold; If the feature entropy value exceeds the preset safety threshold, and the energy of the corresponding intrinsic mode function is mainly concentrated in the high-frequency mode, it is determined to be a suspected poor contact or floating discharge, triggering the active impedance perturbation verification mechanism.

[0024] The preset safety threshold is pre-set based on the statistical distribution of the characteristic entropy values ​​of the converter transformer under the health history data.

[0025] Furthermore, the execution process of the active impedance perturbation verification mechanism is as follows: the active perturbation command is sent to the corresponding wireless smart sensor; the wireless smart sensor responds to the active perturbation command, controls the internal switch to operate, and connects the load resistor with a known resistance value in parallel to the signal acquisition circuit to perform the load connection operation; The effective value of the open-circuit voltage before the load connection operation and the effective value of the loaded voltage after the load connection operation are measured respectively; the ratio of the difference between the effective value of the open-circuit voltage and the effective value of the loaded voltage to the effective value of the open-circuit voltage is calculated to obtain the actual voltage drop rate.

[0026] Furthermore, the process of the active impedance disturbance verification mechanism outputting fault diagnosis results is as follows: Based on the physical structure of the bushing end screen of the converter transformer, a Thevenin equivalent circuit model is established, and the Thevenin equivalent output impedance of the bushing end screen port is determined. Based on the theoretical voltage sag rate calculation formula, the theoretical voltage sag rate is calculated using the Thevenin equivalent output impedance and the resistance value of the load resistor; the actual voltage sag rate is then compared with the theoretical voltage sag rate. If the difference between the actual voltage drop rate and the theoretical voltage drop rate is within a preset error range, the fault type is determined to be insulation degradation, and this is output as the fault diagnosis result. If the actual voltage drop rate is greater than the theoretical voltage drop rate and exceeds the preset error range, or if the monitored voltage signal disappears, the fault type is determined to be poor contact of the end screen lead, and this is output as the fault diagnosis result.

[0027] The preset error range is pre-set based on the measurement accuracy and circuit parameter tolerance range of the wireless smart sensor.

[0028] This invention provides a detection device and early warning method for abnormal voltage at the end screen of a converter transformer bushing. It has the following beneficial effects: 1. This invention utilizes the unique commutation gap characteristics of the converter transformer as a physical synchronization reference and achieves soft synchronization of multi-source data by locking the slope change point in the voltage waveform. This method overcomes the time dispersion problem of wireless sensor networks in the absence of a high-precision hardware synchronization clock. It can correct the impact of data transmission delay on phase analysis without adding extra synchronization hardware costs, and ensures the timing consistency of signal comparison among multiple devices.

[0029] 2. This invention constructs a virtual health benchmark signal through topological correlation clustering and extracts dynamic residual sequences using time-domain differential technology, effectively decoupling the common-mode voltage fluctuations on the grid side from the specific fault characteristics on the equipment side. In the absence of an external standard voltage transformer reference, it filters out fundamental and background harmonic interference, improving the signal-to-noise ratio of weak partial discharge or poor contact signals at the end screen under strong electromagnetic environments.

[0030] 3. This invention establishes a closed-loop diagnostic logic from passive monitoring to active verification by introducing an active impedance disturbance verification mechanism. By controlling the sensor to connect to a known load when an abnormal characteristic entropy value is detected and comparing the actual and theoretical voltage drop rates, it can accurately distinguish between poor contact of the end-screen lead and insulation medium degradation faults in high-impedance circuits from a physical perspective. This effectively solves the technical problem of difficulty in identifying the nature of faults based solely on passive monitoring data and reduces the risk of false alarms. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of a device for detecting abnormal voltage at the end screen of a converter transformer bushing according to the present invention. Figure 2 This is a flowchart of an early warning method for abnormal voltage at the end screen of a converter transformer bushing, according to the present invention. Figure 3 This is a flowchart of the multi-source data acquisition and soft synchronization preprocessing based on commutation characteristics of the present invention. Figure 4 This is a flowchart of the virtual benchmark reconstruction and dynamic residual generation process of the present invention. Detailed Implementation

[0032] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see the appendix Figure 1 This invention provides a detection device for abnormal voltage at the end screen of a converter transformer bushing. This device is applied in converter station scenarios and specifically includes: Several wireless intelligent sensors are installed at the bushing grounding leads of multiple converter transformers operating under the same conditions (i.e., same pole, same valve group, same phase) within the converter station.

[0034] The signal acquisition module connects to the bushing end-screen test interface to convert the analog voltage signal from the end-screen to ground into a digital signal. Internally, it includes a high-impedance input interface circuit and a high-precision analog-to-digital converter (ADC) to ensure the accuracy of measurements on high-impedance capacitive circuits.

[0035] The controllable load module includes an internal switch (such as a high-voltage relay, solid-state relay, or optocoupler switch) and a load resistor.

[0036] The load resistor has a pre-calibrated, known resistance value with an extremely low temperature drift coefficient. The internal switch is connected in series with the load resistor and then connected in parallel to the input terminal of the signal acquisition circuit. In normal monitoring mode, the internal switch is open; upon receiving a disturbance command, the internal switch closes, connecting the load resistor to the measurement circuit.

[0037] The microprocessor module, as the control core of the sensor (e.g., using STM32 series, DSP or FPGA chips), is responsible for controlling ADC acquisition, controlling internal switching actions and encapsulating data packets.

[0038] The wireless communication module is used to send voltage data and receive control commands from the main control unit via wireless networks (such as LoRa, NB-IoT, ZigBee or industrial Wi-Fi).

[0039] The main control unit establishes a two-way wireless communication connection with each of the wireless smart sensors.

[0040] The main control unit can specifically be an edge computing gateway, an industrial control computer (IPC), or a high-performance server deployed at the substation site. Its hardware architecture includes: Processors, such as high-performance multi-core CPUs (central processing units) or GPUs (graphics processing units), are used to execute complex signal processing algorithms (soft synchronization, VMD decomposition, entropy calculation, etc.).

[0041] The memory, including random access memory (RAM) and non-volatile storage media (such as hard disks and SSDs), is used to temporarily store the high-frequency data streams acquired, store historical health data, and pre-store the ledger topology information of the converter transformer (for identifying equipment under the same operating conditions).

[0042] The communication interface includes a network interface for connecting to a wireless base station / gateway, enabling data interaction with front-end sensors.

[0043] The main control unit runs dedicated monitoring and diagnostic software. Specifically, the main control unit is responsible for receiving multi-source voltage data, performing soft synchronization based on commutation gaps, constructing a virtual reference and extracting residuals, calculating holographic entropy features, issuing active disturbance commands when a suspected fault is identified, and finally calculating the voltage drop rate based on feedback to confirm the fault.

[0044] Please see the appendix Figure 2 The present invention also provides a method for early warning of abnormal voltage at the end screen of a converter transformer bushing, which mainly includes the following steps: Step S100: Multi-source data acquisition and soft synchronization preprocessing based on commutation characteristics. Intelligent sensors acquire ground voltage data from the bushing end screen of each converter transformer and transmit it to the main control unit via a wireless network. To address the time asynchrony issue caused by wireless transmission, the main control unit implements a commutation gap feature locking mechanism. This mechanism utilizes the highly synchronized physical time commutation gap generated on the valve-side voltage waveform during converter valve commutation as a reference.

[0045] The main control unit performs a first-order difference operation on the discrete-time series collected by each sensor to obtain the voltage change slope. The system locks the moment when the slope first exceeds a preset negative threshold as the characteristic synchronization point of that cycle.

[0046] Subsequently, the system executes a soft synchronization mechanism. Using the sensor with the highest signal-to-noise ratio as a reference, the time offset of the characteristic synchronization points of the other sensors relative to the reference is calculated, and a timing shift correction operation is performed to achieve physical alignment of the signals.

[0047] To eliminate the interference of high-frequency nonlinear abrupt changes in the commutation gap itself on the analysis of weak fault characteristics, the system performs an adaptive windowing operation. An adaptive time window centered on the characteristic synchronization point is constructed, and the data weights of the aligned signal within the window are reset to zero or removed, retaining only the data in the stable segment, thus obtaining the preprocessed signal.

[0048] Step S200: Virtual Reference Reconstruction and Dynamic Residual Generation. The homogeneity of multiple devices is utilized to eliminate system-level common-mode interference. The system performs topological correlation clustering, calculates the correlation coefficients between each preprocessed signal, and calculates the normalized weight coefficients accordingly. The virtual health reference signal under this operating condition is reconstructed using a weighted average method, based on the following virtual health reference signal construction formula: ; In the formula: For a moment The virtual health reference signal voltage value; for the summation operator; This represents the total number of sensors under the same operating conditions. This is the sensor's serial number index, with a value range of [value range missing]. arrive ; For the first Normalized weighting coefficients for each sensor signal; For the first Each sensor at time The preprocessed signal value.

[0049] Subsequently, the system performs time-domain difference analysis between the preprocessed signals from each monitoring point and the virtual health baseline signal to obtain a dynamic residual sequence, which is calculated based on the following dynamic residual sequence extraction formula: ; In the formula: For the first The dynamic residual sequence values ​​of each sensor; For the first Each sensor at time The preprocessed signal value; For a moment The virtual health reference signal voltage value.

[0050] This dynamic residual sequence retains specific high-frequency components and minute voltage drifts that reflect changes in the insulation medium of the end screen or poor lead contact, while eliminating fundamental fluctuations and background harmonics on the grid side.

[0051] Step S300: Variational Mode Decomposition and Holographic Entropy Feature Extraction of the Residual Signal. Given the non-stationarity of the dynamic residual sequence, the system employs a variational mode decomposition model to decompose the dynamic residual sequence into... Layer-specific intrinsic mode functions. This model finds a set of mode functions and their center frequencies by solving a constrained variational problem, such that the sum of the estimated bandwidths of all modes is minimized. Its core objective is based on the following variational mode decomposition (VMD) objective function formula: ; In the formula: This is the minimize operator; The set of intrinsic mode functions obtained from decomposition; This is the set of center frequencies corresponding to each modal function; The total number of layers of the intrinsic mode functions obtained from the decomposition; This is the index of the intrinsic mode function components, with values ​​ranging from 1 to... ; This refers to the L2 norm operator. Regarding time The partial derivative operator; The Dirac distribution function; The imaginary unit; Pi is a constant. This is the convolution operator; For the first Layer intrinsic mode function with time A changing function; is the base of the natural logarithm; For the first The center frequency of each modal function.

[0052] The system selects the high-frequency intrinsic mode functions obtained from decomposition, constructs a reconstruction vector through phase space reconstruction, and calculates the probability of the occurrence of permutation patterns in the vector to quantify the complexity and abrupt change of the signal. The calculation is based on the following permutation entropy characteristic calculation formula: ; In the formula: The calculated feature entropy value; For the embedding dimension; for The factorial of represents the total number of all possible permutations; This is the index of the arrangement pattern, with a value range of 1. arrive ; For the first Arrangement patterns The relative probability of occurrence in the reconstructed sequence; This is the operator for the natural logarithm.

[0053] Step S400: Anomaly Detection and Active Impedance Disturbance Verification. The system performs an initial state judgment based on the characteristic entropy value. When the characteristic entropy value... When the threshold is exceeded and accompanied by high-frequency pulse characteristics, it is determined to be a suspected poor contact or floating discharge, and the active impedance disturbance verification mechanism is triggered.

[0054] The main control unit issues an active disturbance command, triggering the corresponding wireless sensor to perform a load access operation, connecting a load resistor of known resistance in parallel in the signal acquisition loop. The system measures the voltage change before and after load access and calculates the actual voltage drop rate. Simultaneously, it calculates the theoretical voltage drop rate based on the equivalent circuit model of the final screen, using the following theoretical voltage drop rate calculation formula: ; In the formula: This represents the theoretical voltage drop rate. Thevenin equivalent output impedance of the bushing end screen port; This refers to the resistance value of the load resistor connected during active verification.

[0055] The system compares the actual voltage drop rate with the theoretical voltage drop rate. If the difference between the two is consistent within the error range, the fault is determined to be insulation degradation, i.e., the change in capacitor dielectric parameters causes Thevenin equivalent output impedance drift. If the actual voltage drop rate is much greater than the theoretical voltage drop rate or the signal disappears completely, the fault is determined to be poor contact of the end screen lead, i.e., there is unstable nonlinear contact resistance in the circuit.

[0056] Please see the appendix Figure 3In step S100, multi-source data acquisition and soft synchronization preprocessing based on commutation characteristics, the process specifically includes the following sub-steps: Step S110: Perform multi-source isomorphic data acquisition and filtering. The main control unit receives bushing end-screen to ground voltage data from multiple wireless intelligent sensors within the substation. To ensure the physical validity of subsequent differential signals, the system filters sensor data groups under the same operating conditions based on pre-stored converter transformer ledger topology information. Here, "same operating conditions" specifically refers to operating states of the same pole, same valve group, and same phase. For example, selecting phases A, B, and C of the YY converter transformer at the lower end of pole I, or the corresponding phase data of another converter transformer operating in parallel.

[0057] This screening method is based on the electrical coupling characteristics of converter transformers. Specifically, when the converter valves of converter transformers in the same valve group are commutating, their valve-side bushing voltages are simultaneously affected by commutation overvoltage and voltage gap. This physical synchronicity provides an objective benchmark for addressing the time discrepancies in wireless transmission. The collected data is a discrete-time series after analog-to-digital conversion.

[0058] To address the unavoidable packet jitter and latency during wireless transmission, this step will not perform hard alignment based on timestamps, but will instead retain the original sampling sequence for subsequent feature processing.

[0059] Step S120: Execute the commutation gap feature locking mechanism. For each discrete time series obtained in step S110, the system locks the physical synchronization point by analyzing the waveform morphology. Due to the periodic on and off of the converter valve, the valve-side voltage waveform will produce a significant voltage drop in a specific phase interval, i.e., a commutation gap. The occurrence time of this gap on the time axis is determined by the grid frequency and firing angle, and it is strictly synchronized for equipment in the same valve group.

[0060] The system performs a first-order difference operation on the discrete-time series collected by each sensor to calculate the voltage change rate between adjacent sampling points. When the voltage change rate (slope) is detected to increase sharply in a negative direction and exceed the preset negative threshold for the first time, this moment corresponds to the starting point of the leading edge of the commutation gap. The system marks this moment as the characteristic synchronization point within this power frequency cycle.

[0061] The preset negative threshold is set in advance based on the statistical characteristics of the voltage drop slope at the commutation gap of the converter transformer under normal operating conditions.

[0062] For example, the system can calculate the mean of the first-order difference sequence of the voltage waveform during the device's historical normal operating cycles. and standard deviation And set the threshold to to ; Alternatively, considering that the rate of decrease of the commutation gap is much faster than the natural change of the power frequency sine wave, this threshold can be set to more than 10 times the maximum rate of change of the power frequency voltage (for example, if the maximum slope of the normalized sampled value is...). Then the negative threshold is set to This ensures that the steep drop-off edge at the moment of commutation can be accurately captured, while eliminating subtle fluctuations caused by background harmonics in the power grid.

[0063] This processing method can effectively avoid the influence of voltage amplitude fluctuations and relies solely on the structural abrupt changes in the waveform, thereby achieving microsecond-level feature localization without an external high-precision clock source.

[0064] Step S130: Perform time-shift correction based on feature points. After obtaining the characteristic synchronization points of each signal, the system evaluates the signal quality of each sensor and selects the sensor with the highest signal-to-noise ratio or the best waveform integrity as the reference benchmark.

[0065] The system calculates the time offset of the characteristic synchronization point of each of the other sensors relative to the reference characteristic synchronization point. This time offset directly reflects the data transmission delay difference and sampling start phase difference between different sensors.

[0066] Based on the calculated time offset, the system performs a reverse time shift operation on each original discrete time sequence, so that the commutation gap start time of all signals coincides on the time axis, thereby achieving soft synchronization at the physical level.

[0067] Step S140: Perform adaptive windowing and signal reassembly. Although the signal has been aligned, the commutation gap itself contains a large number of high-frequency components and nonlinear transient noise. These components can easily mask the fault characteristics caused by weak discharges inside the end screen in the spectrum, or generate false high-frequency modes in mode decomposition. Therefore, the system constructs an adaptive time window centered on the characteristic synchronization point. The width of this adaptive time window is dynamically set according to the current converter transformer operating trigger angle and commutation overlap angle, or set to a fixed width covering the commutation process.

[0068] For example, considering that the commutation overlap angle of the converter valve under rated operating conditions is usually between 15° and 25° electrical angle, in order to ensure complete shielding of the commutation gap and its accompanying high-frequency oscillation, the width of the adaptive time window can be set to a fixed value greater than the maximum overlap angle, such as setting it to 30° electrical angle (the corresponding time width at 50Hz power frequency is about 1.67ms); or set to an absolute time width including safety margins before and after (such as 2ms to 3ms) to ensure that non-stationary signals within the window are completely zeroed, and only the stationary sine wave band outside the window is retained for subsequent analysis.

[0069] The system performs an adaptive windowing operation, resetting the data weights of each aligned signal within the window range to zero, or directly discarding the data for that time period, retaining only the stable voltage data after commutation. The data after the above processing is defined as a preprocessed signal. This signal is synchronous in time and its waveform has been freed from large system-level disturbances, providing a clean data foundation for subsequent extraction of weak device-level residuals.

[0070] The general digital signal processing hardware implementations involved in the aforementioned data acquisition and differential operations are well-known to those skilled in the art and will not be elaborated upon here.

[0071] Please see the appendix Figure 4 In step S200, virtual benchmark reconstruction and dynamic residual generation, the process specifically includes the following sub-steps: Step S210: Perform topological correlation clustering analysis. In order to extract the reference component representing the common-mode voltage waveform on the system side from the multi-source signals, the system first performs correlation analysis on each preprocessed signal output in step S100.

[0072] Since all converter transformers are connected to the same AC power grid bus, under normal conditions, the induced voltage waveforms of their terminal screens should have high similarity, mainly containing the fundamental frequency and system-side background harmonics. The system calculates the Pearson correlation coefficients between each pair of preprocessed signals and constructs a correlation coefficient matrix.

[0073] Based on this matrix, the system evaluates the average correlation between each signal and all other signals. A signal with a higher correlation indicates that it is closer to the common characteristics of the group under the current operating conditions, that is, closer to the actual grid voltage waveform; conversely, if a signal has a low correlation with other signals, it indicates that the signal may contain specific fault components or interference.

[0074] The system assigns normalized weight coefficients based on the correlation value. Signals with high correlation are given greater weights, while signals that deviate from the group characteristics are given smaller weights, thereby reducing the contamination of the reference signal by a single faulty device.

[0075] Step S220: Construct a virtual health baseline signal. The system uses determined normalized weighting coefficients to perform weighted synthesis of multiple preprocessed signals. This synthesis process is performed point-by-point in the time domain, essentially utilizing the topological redundancy of multiple devices to establish an idealized reference model. This reference model is defined as a virtual health baseline signal, and its physical meaning lies in reconstructing a clean grid voltage waveform stripped of individual device differences.

[0076] Compared to the traditional method of relying on an external voltage transformer (PT) to obtain the reference voltage, this self-referencing reference construction method avoids introducing additional measurement errors and hardware costs. The virtual health reference signal is calculated based on the following virtual health reference signal construction formula: ; In the formula: For a moment The virtual health reference signal voltage value; for the summation operator; This represents the total number of sensors under the same operating conditions. This is the sensor's serial number index, with a value range of [value range missing]. arrive ; For the first Normalized weighting coefficients for each sensor signal; For the first Each sensor at time The preprocessed signal value.

[0077] Step S230: Extract the dynamic residual sequence. After acquiring the virtual health baseline signal representing the common mode component of the system, the system performs a time-domain differential operation, subtracting the measured preprocessed signal of each sensor from the virtual health baseline signal.

[0078] The purpose of this operation is to achieve signal decoupling, that is, to filter out the grid fundamental voltage, system-side voltage regulation fluctuations and common-mode harmonic components generated by the converter valve contained in the original measurement signal, and retain only the differentiated components unique to the end screen of this device.

[0079] This differential component is defined as the dynamic residual sequence. Under ideal insulation conditions, this residual sequence should be approximately zero or contain only white noise with a noise floor; however, when there is insulation degradation or poor contact in the equipment, the residual sequence will contain weak high-frequency pulses, voltage drifts, or step signals caused by local defects. The dynamic residual sequence is calculated based on the following dynamic residual sequence extraction formula: ; In the formula: For the first The dynamic residual sequence values ​​of each sensor; For the first Each sensor at time The preprocessed signal value; For a moment The virtual health reference signal voltage value.

[0080] Through the above steps, the system successfully transforms the problem of extracting weak fault features under strong backgrounds into a residual analysis problem under low-noise backgrounds, providing a high signal-to-noise ratio data foundation for subsequent feature quantization. The specific algorithm implementation for calculating the Pearson correlation coefficient and the weighted average is well-known to those skilled in the art and will not be elaborated upon here.

[0081] In step S300, the residual signal variational mode decomposition and holographic entropy feature extraction specifically include the following sub-steps: Step S310: Construct and solve the variational mode decomposition model. Given that the dynamic residual sequence obtained in step S200 often exhibits significant non-stationarity and nonlinearity, containing intermittent transient components caused by insulation defects, traditional Fourier transform is insufficient to effectively analyze its time-frequency local characteristics. Therefore, the system employs a variational mode decomposition model to adaptively process the dynamic residual sequence.

[0082] This model decomposes the complex residual signal into a preset number of elements. The system has intrinsic mode functions, each of which is an amplitude-modulated (AM) and frequency-modulated (FM) signal with a finite bandwidth around its center frequency. The system constructs a constrained variational problem to find a set of mode functions and their center frequencies such that the sum of the estimated bandwidths of all modes is minimized, while the sum of the bandwidths equals the original input signal. Its core objective is based on the following variational mode decomposition (VMD) objective function formula: ; In the formula: This is the minimize operator; The set of intrinsic mode functions obtained from decomposition; This is the set of center frequencies corresponding to each modal function; The total number of layers of the intrinsic mode functions obtained from the decomposition; This is the index of the intrinsic mode function components, with values ​​ranging from 1 to... ; This refers to the L2 norm operator. Regarding time The partial derivative operator; The Dirac distribution function; The imaginary unit; Pi is a constant. This is the convolution operator; For the first Layer intrinsic mode function with time A changing function; is the base of the natural logarithm; For the first The center frequency of each modal function.

[0083] To solve the aforementioned constrained variational problem, the system introduces a quadratic penalty factor and Lagrange multipliers, transforming it into an unconstrained variational problem. It then uses the Alternating Direction Multiplier Method (ADMM) for iterative updates until the convergence condition is met, ultimately outputting a distribution from low to high frequencies. Layer intrinsic mode function components.

[0084] Step S320: Screening high-frequency characteristic modes. After obtaining all intrinsic mode functions, the system screens them based on the distribution characteristics of the center frequency. Since faults such as poor contact of the end-screen lead or floating discharge mainly excite high-frequency pulse signals, while aging of insulating oil or slow drift of capacitance parameters mainly manifests in low-frequency components.

[0085] Therefore, the system automatically selects the intrinsic mode functions corresponding to the high-frequency bands obtained from the decomposition as the feature extraction objects, or directly selects the index number. Modal components exceeding preset values ​​are used to focus on abrupt fault characteristics.

[0086] Step S330: Calculate the holographic spectrum permutation entropy features. To quantify the disorder and abrupt change characteristics in the selected high-frequency intrinsic mode functions, the system introduces a permutation entropy algorithm. For a given modal signal sequence, the system first performs phase space reconstruction, generating a series of reconstruction vectors.

[0087] Subsequently, the system maps the element values ​​in the reconstructed vector to ordinal patterns and counts the relative frequency of each specific pattern among all possible permutations.

[0088] The statistical results reflect the complexity of the signal's local structure: when the signal contains regular power frequency components, the permutation entropy is low; when the signal contains random abrupt pulses caused by discharge, the permutation entropy increases significantly. The system calculates the characteristic entropy value as the final fault criterion, based on the following permutation entropy characteristic calculation formula: ; In the formula: The calculated feature entropy value; For the embedding dimension; for The factorial of represents the total number of all possible permutations; This is the index of the arrangement pattern, with a value range of 1. arrive ; For the first Arrangement patterns The relative probability of occurrence in the reconstructed sequence; This is the operator for the natural logarithm.

[0089] Through the above steps, the system transforms the complex raw voltage waveform into a single quantitative indicator, namely the characteristic entropy value. This indicator is not only sensitive to minor faults but also effectively distinguishes between steady-state drift and transient impacts, providing robust feature input for subsequent intelligent diagnosis.

[0090] The specific iterative process of the ADMM algorithm and the selection of specific parameters for phase space reconstruction are well-known techniques to those skilled in the art and will not be elaborated upon here.

[0091] In step S400, the anomaly detection and active impedance perturbation verification process specifically includes the following sub-steps: Step S410: Initial Status Assessment and Graded Early Warning. The system receives the feature entropy value output in step S300 and compares it with a preset safety threshold. The safety threshold is set based on the statistical distribution of the converter transformer under historical health data. For example, the average value of the feature entropy value during the equipment's historical normal operation cycle plus three times the standard deviation is selected as the safety threshold.

[0092] When the feature entropy value is below the safety threshold, the system determines that the device is in normal operating condition. When the feature entropy value exceeds the safety threshold, the system further analyzes the energy distribution characteristics of the corresponding intrinsic mode function.

[0093] If the energy is mainly concentrated in the low-frequency mode, it indicates that the voltage waveform has undergone overall distortion or drift. The system is initially judged to be due to abnormal insulation dielectric loss factor or slow change in capacitance parameters.

[0094] If the characteristic entropy value increases significantly and the energy is mainly concentrated in the high-frequency mode, the waveform exhibits intermittent high-frequency pulse characteristics. The system determines that there is a suspected poor contact of the end screen lead or an internal floating discharge defect. At this time, it is difficult to distinguish between severe internal discharge and loose lead connection based solely on passive monitoring data. Therefore, the system triggers the active impedance perturbation verification mechanism.

[0095] Step S420: Execute the active disturbance command and connect the load. The main control unit sends the active disturbance command to the corresponding wireless smart sensor.

[0096] The sensor integrates a controllable load module, which includes an electronic switch or relay and a high-precision load resistor. Upon receiving a command, the sensor controls the electronic switch to connect the load resistor of known resistance value in parallel to the signal acquisition circuit, maintaining this connection for several power frequency cycles.

[0097] The principle for selecting the load resistor value is that its resistance should be significantly smaller than the capacitive reactance of the bushing end screen to ground at power frequency, and is usually selected in the range of 1. Up to 100 This is to ensure that a noticeable voltage drop occurs after connection, while limiting the loop current to the safe current-carrying range of the electronic switch.

[0098] During this period, the sensor continuously acquires the voltage signal to ground from the terminal screen and records the effective value of the open-circuit voltage before the load is connected and the effective value of the applied voltage after the load is connected. The system calculates the actual voltage drop rate based on these two measurements, which reflects the actual degree of voltage attenuation at the port after the external load is connected.

[0099] Step S430: Calculate the theoretical voltage drop rate. To verify the integrity of the current electrical circuit connection, the system establishes a Thevenin equivalent circuit model based on the physical structure of the converter transformer bushing end screen. In this model, the main insulation capacitance of the bushing... and the end screen to ground insulation capacitor This forms a passive voltage divider network, and its impedance seen from the end-screen measurement port is defined as the Thevenin equivalent output impedance.

[0100] This impedance value is typically in the megaohm range and is relatively stable when no breakdown occurs. The system calculates the expected voltage drop ratio based on equipment ledger parameters when a known load resistance is connected, using the following theoretical voltage drop rate calculation formula: ; In the formula: This represents the theoretical voltage drop rate. Thevenin equivalent output impedance of the bushing end screen port; This refers to the resistance value of the load resistor connected during active verification.

[0101] Step S440: Execute closed-loop diagnostic logic. The system compares and analyzes the actual voltage drop rate calculated from the measured values ​​with the theoretical voltage drop rate. If the difference between the two is consistent within the preset error range, it indicates that the response of the actual circuit conforms to the prediction of the linear capacitive voltage divider model, and the electrical connection between the end-screen lead and the signal conditioning circuit is good.

[0102] At this point, the entropy anomaly detected in the preceding steps should be attributed to a change in the parameters of the insulating medium itself, namely the capacitance. Internal partial breakdown may have caused the equivalent output impedance to be reduced. A drift occurred, and the system diagnosed the fault type as insulation degradation.

[0103] Conversely, if the actual voltage drop rate is much greater than the theoretical voltage drop rate, or if the voltage signal monitored at the moment the load is connected disappears completely (the drop rate is close to 100%), this indicates that an unstable nonlinear contact resistor is connected in series in the circuit.

[0104] The preset error range is pre-set based on the sensor's measurement accuracy and the tolerance range of the circuit parameters, and is used to accommodate calculation deviations caused by non-fault factors.

[0105] For example, taking into account the nominal error of the load resistance (such as...) The capacitance drift of the end-screen capacitor due to temperature and the quantization error of the sensor ADC analog-to-digital conversion can be addressed by setting the error range to the theoretical voltage drop rate. to .

[0106] In open-circuit or high-impedance measurement modes, the contact resistance may be masked by the high input impedance; once a relatively low-resistance load resistor is connected, the loop current increases, causing a sharp increase in the voltage drop across the contact resistance, which in turn causes the voltage at the measurement port to drop significantly or open the circuit directly.

[0107] Based on this, the system eliminates insulation medium faults, confirms the fault type as poor contact of the end screen lead, and issues a high-level alarm. Through this proactive physical verification, the present invention effectively solves the technical problem of distinguishing between loose connections and actual faults in the monitoring of weak high-impedance signals.

[0108] In summary, the detection device and early warning method for abnormal voltage at the end screen of a converter transformer bushing provided in this embodiment constructs a complete monitoring system from physical feature locking to active closed-loop verification.

[0109] This method breaks through the dependence of traditional monitoring technology on high-precision hardware synchronization clocks. It utilizes the unique commutation gap characteristics of the converter transformer as a physical anchor point to achieve microsecond-level multi-source data soft synchronization in a wireless transmission environment.

[0110] By constructing a virtual health benchmark through topological correlation clustering, this invention effectively decouples the common-mode voltage fluctuations on the system side from the personalized fault characteristics on the equipment side, making it possible to extract weak dynamic residual sequences under strong electromagnetic interference environments.

[0111] Furthermore, by combining variational mode decomposition and holographic spectrum permutation entropy algorithm, this invention transforms non-stationary residual signals into fault-sensitive quantization entropy values, thereby improving the ability to capture early insulation defects and intermittent discharge signals.

[0112] Most importantly, by introducing an active impedance perturbation verification mechanism, this invention changes the limitation of traditional monitoring that can only passively receive data. By utilizing the controllable load access function of sensor hardware, a closed-loop diagnostic logic of anomaly detection, hardware verification, and fault diagnosis is established.

[0113] This mechanism effectively solves the common industry problem of difficulty in distinguishing between loose connections in the end screen lead and internal insulation deterioration in high impedance measurement circuits from a physical perspective, reduces false alarm rate, and provides reliable technical support for the safe and stable operation of converter transformers.

Claims

1. A device for detecting abnormal voltage at the end screen of a converter transformer bushing, characterized in that, The application relates to a detection device for voltage abnormality of a bushing tap of a converter transformer, and comprises the following steps: S1, collecting ground voltage data of bushing taps of each converter transformer by using wireless intelligent sensors, identifying a characteristic synchronization point in the ground voltage data by using a commutation gap characteristic locking mechanism, and performing a soft synchronization mechanism and an adaptive windowing operation based on the characteristic synchronization point to output pretreatment signals of the wireless intelligent sensors; S2, performing topology correlation clustering on the pretreatment signals, constructing a virtual health reference signal reflecting a system side common mode voltage characteristic, and performing time domain difference between the pretreatment signals and the virtual health reference signal to extract a dynamic residual sequence containing a device side individualized characteristic; S3, constructing a variational mode decomposition model, decomposing the dynamic residual sequence to obtain an intrinsic mode function, and calculating a characteristic entropy value of the intrinsic mode function; 2. The device for detecting abnormal voltage of the bushing tap of the converter transformer according to claim 1, characterized in that, S4, comparing the characteristic entropy value with a preset safety threshold value, triggering an active impedance disturbance verification mechanism when a suspected contact failure fault is determined, sending an active disturbance instruction to control the wireless intelligent sensors to perform a load access operation, and outputting a fault diagnosis result according to a comparison result of an actual voltage drop rate and a theoretical voltage drop rate. In the S1 step, the specific execution process of the commutation gap characteristic locking mechanism comprises the following steps: performing a first-order difference operation on discrete time sequences of the collected ground voltage data of the bushing taps of each converter transformer to obtain a voltage change slope; monitoring the voltage change slope, locking a time point when the voltage change slope first exceeds a preset negative threshold value as the characteristic synchronization point, and the characteristic synchronization point corresponds to a front starting time of a commutation gap generated during commutation of a converter valve; 3. A method for early warning of abnormal voltage of a converter transformer bushing end shield, characterized in that, ​ ​ ​ ​ ​ 4. The method of Claim 3, wherein the method further comprises: ​ ​ ​ The preset negative threshold is preset according to a voltage falling edge slope statistical characteristic of a commutation gap of the converter transformer under a normal operation condition.

5. The method of Claim 4, wherein the method further comprises: The specific execution process of the soft synchronization mechanism and the adaptive windowing operation in the S1 step includes: Selecting a wireless intelligent sensor with the highest signal-to-noise ratio as a reference benchmark, and calculating a time offset of a feature synchronization point of each of the remaining wireless intelligent sensors relative to the feature synchronization point of the reference benchmark; According to the time offset, performing a time sequence translation correction operation on each of the discrete time sequences to realize physical alignment of the signals; An adaptive time window is constructed with the feature synchronization point as the center, data weights of the aligned signals in the adaptive time window are set to zero or removed, only smooth section data is retained, and the preprocessed signals are obtained; The width of the adaptive time window is dynamically set according to an operation trigger angle and a commutation overlap angle of the converter transformer.

6. The pre-warning method for abnormal voltage of the bushing tap of the converter transformer according to claim 3, characterized in that, The S2 step specifically includes: Calculating a Pearson correlation coefficient between each of the preprocessed signals, and calculating a normalized weight coefficient of each of the preprocessed signals according to the Pearson correlation coefficient; The virtual healthy reference signal is calculated according to a formula constructed by using the normalized weight coefficient to weight and synthesize the preprocessed signals; A point-by-point time domain difference is performed between the preprocessed signals of each of the wireless intelligent sensors and the constructed virtual healthy reference signal; According to a dynamic residual sequence extraction formula, the preprocessed signals are subtracted from the virtual healthy reference signal to obtain the dynamic residual sequence.

7. The pre-warning method for abnormal voltage of the bushing tap of the converter transformer according to claim 3, characterized in that, The S3 step specifically includes: A set of intrinsic mode functions and a center frequency are found by solving a constrained variation problem, so that the sum of the estimated bandwidths of each mode is minimized, thereby establishing the variational mode decomposition model; Based on the variational mode decomposition model, the intrinsic mode functions corresponding to the dynamic residual sequence are obtained by using an alternating direction multiplier method to iteratively solve a variational mode decomposition objective function formula; From the multiple intrinsic mode functions obtained by decomposition, the intrinsic mode functions in a high-frequency band are selected, a reconstructed vector is generated by phase space reconstruction, the relative probability of different permutation patterns in the reconstructed vector is counted, and the feature entropy value is calculated according to a permutation entropy feature calculation formula.

8. The pre-warning method for abnormal voltage of the bushing tap of the converter transformer according to claim 3, characterized in that, In the S4 step, the discrimination logic of the active impedance disturbance verification mechanism includes: The feature entropy value is compared with the preset safety threshold; If the feature entropy value exceeds the preset safety threshold and the energy of the corresponding intrinsic mode function is mainly concentrated in a high-frequency mode, it is determined that there is suspected poor contact or floating discharge, and the active impedance disturbance verification mechanism is triggered. The preset safety threshold is preset according to a feature entropy value statistical distribution of the converter transformer under healthy historical data.

9. The method of Claim 3, wherein the method further comprises: The execution process of the active impedance disturbance verification mechanism in the S4 step includes: The active disturbance instruction is sent to the corresponding wireless intelligent sensor. The wireless intelligent sensor controls internal switches to connect a load resistor with a known resistance in parallel to the signal acquisition circuit in response to the active disturbance instruction to perform the load connection operation; The open circuit voltage effective value before performing the load connection operation and the loaded voltage effective value after performing the load connection operation are measured respectively; The ratio of the difference between the open circuit voltage effective value and the loaded voltage effective value to the open circuit voltage effective value is calculated to obtain the actual voltage drop rate.

10. The method of Claim 9, wherein the method further comprises: In the S4 step, the process of outputting the fault diagnosis result by the active impedance disturbance verification mechanism includes: A Thevenin equivalent circuit model is established based on the physical structure of the bushing tap of the converter transformer, and the Thevenin equivalent output impedance of the bushing tap is determined; The theoretical voltage drop rate is calculated by using the Thevenin equivalent output impedance and the resistance of the load resistor according to the theoretical voltage drop rate calculation formula; The actual voltage drop rate is compared with the theoretical voltage drop rate; If the difference between the actual voltage drop rate and the theoretical voltage drop rate is consistent within a preset error range, it is determined that the fault type is insulation deterioration, and the fault diagnosis result is output; If the actual voltage drop rate is greater than the theoretical voltage drop rate and exceeds the preset error range, or the monitored voltage signal disappears, it is determined that the fault type is poor contact of the bushing tap lead, and the fault diagnosis result is output; The preset error range is set in advance according to the measurement accuracy of the wireless intelligent sensor and the tolerance range of the circuit parameters.

Images