Method and system for locating phase modulation machine inter-turn short circuit based on envelope fusion and ridge regression

By employing envelope fusion and ridge regression methods, the accuracy problem of inter-turn short circuit location in synchronous condenser rotor windings was solved, achieving stable identification and correction of fault locations and improving positioning accuracy.

CN123066640APending Publication Date: 2026-11-10STATE GRID JIANGSU ELECTRIC POWER CO LTD MAINTENANCE BRANCH +1
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202611530187.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-30
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies struggle to accurately identify the location of inter-turn short-circuit faults in the rotor windings of synchronous condensers, especially when the axial length is large. Fault reflection waves are prone to superimposed on reflection waves at the winding ends, leading to significant positioning errors.

Method used

A positioning method based on envelope fusion and ridge regression is adopted. By establishing a high-frequency distributed parameter model of the rotor winding, the differential envelope of the near-end response signal is extracted. Combined with local maximum detection and adaptive threshold, the leading edge of the reflected wave and the moment of the first main peak are identified. The initial positioning results are then corrected using the ridge regression model.

Benefits of technology

It improves the accuracy and stability of inter-turn short-circuit fault location, reduces the impact of noise and end reflection, and enhances adaptability to different fault locations and degrees.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123066640A_ABST
    Figure CN123066640A_ABST
Patent Text Reader

Abstract

The application discloses a method and system for positioning phase modulation machine inter-turn short circuit based on envelope fusion and ridge regression, which comprises the following steps: establishing a rotor winding high-frequency distributed parameter model, and obtaining a near-end response signal under a healthy working condition; collecting a near-end response signal in actual operation, differentiating the near-end response signal from the healthy working condition response, and extracting an envelope; setting a global search time range based on an injection pulse reference time and a propagation speed, determining a candidate wave packet through local maximum value detection and continuous threshold crossing search, and detecting a reflected wave front time and a first main peak time through adaptive threshold detection; fusing the front time and the main peak time to obtain a fusion positioning time, calculating an initial fault position, and outputting a final fault position through a ridge regression positioning correction model. The application fuses the front-main peak and corrects the ridge regression, offsets the front negative deviation and the main peak lag deviation, compensates systematic deviation caused by approximate propagation speed, sampling dispersion and waveform widening, and improves inter-turn short circuit positioning precision and stability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of synchronous condenser rotor winding state detection technology, specifically involving a method and system for locating short circuits between turns of a synchronous condenser based on envelope fusion and ridge regression. Background Technology

[0002] Synchronous condensers possess dynamic reactive power compensation, voltage support, and short-term overload capability, making them crucial equipment for ensuring the safe and stable operation of AC / DC hybrid power grids. As the core component of the synchronous condenser's excitation system, the rotor windings are subjected to the combined effects of electromagnetic forces, centrifugal forces, thermal stress, and mechanical vibrations over extended periods. This makes the inter-turn insulation prone to aging, wear, and localized breakdown, ultimately leading to inter-turn short-circuit faults.

[0003] In the early stages, inter-turn short circuits in rotor windings typically manifest as localized conductive paths between a few winding conductors. The changes in DC resistance, excitation current, and power frequency impedance caused by the fault are relatively small, exhibiting strong concealment and inconspicuous power frequency characteristics. If the fault location cannot be detected and determined in time, the localized short-circuit circulating current will cause increased winding temperature, uneven magnetic field distribution, and intensified mechanical vibration. In severe cases, it may develop into a rotor grounding fault, affecting the safe operation of the synchronous condenser.

[0004] Existing methods for detecting inter-turn short circuits in rotor windings mainly include the DC resistance method, AC impedance method, distributed voltage method, open transformer method, and repetitive pulse method (RSO). The DC resistance method and AC impedance method are relatively simple to operate, but they are not sensitive enough to minor inter-turn short circuits, and they can usually only determine whether there is an abnormality in the winding, but it is difficult to determine the specific fault location. The distributed voltage method has a certain location capability, but the detection process is complex and requires many manual measurement points. The open transformer method has high requirements for rotor structure and test conditions, and its applicability is limited.

[0005] Taking patent CN215986428U as an example, this patent describes a generator rotor turn-to-turn short circuit detection device based on the repetitive pulse (RSO) method. Through hardware integration of a control unit, excitation signal generation unit, excitation signal driving unit, excitation signal output control switch, acquisition signal control switch, and data acquisition unit, it eliminates the need for an external oscilloscope during on-site testing. This solves the problems of traditional RSO testing requiring an oscilloscope, complex on-site wiring, and high operator skill requirements. However, this solution has the drawback of only achieving hardware integration and simplified on-site operation. It does not address the precise identification of the arrival time of the fault reflection wave. The location result still relies on manual interpretation of the response waveform, and is susceptible to high-frequency oscillations, propagation attenuation, reflection wave superposition, and sampling discrepancies, resulting in insufficient location accuracy.

[0006] For synchronous condenser rotor windings with large axial lengths, fault reflection waves exhibit amplitude attenuation and waveform broadening during propagation. When the fault location is near the far end of the winding, the fault reflection wave may also superimpose with the reflection wave from the winding end. Using a fixed threshold or a single peak value for location can easily lead to misidentification of local noise peaks, oscillation extremes, or end reflection waves as fault reflection waves, resulting in systematic deviations in the location results.

[0007] Therefore, it is necessary to propose a method for locating short circuits between turns of a synchronous condenser rotor winding that can stably extract the arrival time of the fault reflection wave and correct the positioning deviation caused by high-frequency pulse propagation and discrete sampling. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a method and system for locating short circuits between turns of a synchronous condenser rotor winding.

[0009] The present invention adopts the following technical solution.

[0010] The first aspect of this invention proposes a method and system for locating inter-turn short circuits in a synchronous condenser based on envelope fusion and ridge regression, comprising: A high-frequency distributed parameter model of the rotor winding of the synchronous condenser is established for simulation to obtain the near-end response signal under healthy operating conditions. The near-end response signal refers to the signal collected from the pre-set near-end response acquisition channel after a high-frequency pulse injection terminal is set at one end of the rotor winding and a set high-frequency pulse is injected. During actual operation of the synchronous condenser, after injecting a set high-frequency pulse into the synchronous condenser, the proximal response signal during actual operation is obtained. The difference between the proximal response signal during actual operation and the proximal response signal under simulated healthy operating conditions is calculated as a real-time differential signal, and the envelope is extracted from the real-time differential signal. The global search time range of the fault reflection wave is set based on the injection pulse reference time and the set pulse propagation speed; within the global search time range, candidate wave packets are searched from the envelope through local maximum detection and continuous threshold search; with the candidate wave packets as the center, an adaptive threshold is set based on the maximum envelope amplitude of the set local time window; based on the adaptive threshold, the leading edge time and the first main peak time of the reflection wave are detected within the set local time window. The fused location time is obtained by weighting and summing the reflection wave leading edge time and the first main peak time according to the set weights. The initial fault location is calculated based on the fused location time. The reflection wave leading edge time, the first main peak time and the initial fault location are input into the pre-trained ridge regression localization correction model to output the final fault location.

[0011] Preferably, the simulation of establishing a high-frequency distributed parameter model of the rotor winding of the synchronous condenser specifically involves: Various fault conditions are formed by setting different fault locations and different short-circuit turns. The differential signal between the near-end response signal and the near-end response signal under the healthy condition is calculated for each fault condition. The differential reflection envelope is extracted from the differential signal corresponding to each fault condition. The reflection wave leading moment and the first main peak moment are extracted from the differential reflection envelope. The initial fault location is obtained by fusing the two. The reflection wave leading moment, the first main peak moment and the initial fault location are used as a training sample. The actual fault location of the corresponding fault condition is used as the label of the training sample. The ridge regression localization correction model is trained using all training samples.

[0012] Preferably, the calculation of the difference signal between the proximal response signal during actual operation and the simulated proximal response signal under healthy operating conditions specifically includes: The rising edge of the injection pulse of the proximal response signal during actual operation and the proximal response signal under healthy conditions are detected; the time when the rising edge of the injection pulse reaches a set threshold is used as the time reference. When the rising edge of the injection pulse cannot be directly detected, the peak time of the injection pulse is used as the time reference; the difference between the time reference of the proximal response signal during actual operation and the proximal response signal under healthy conditions is calculated as the time offset, and one set of proximal response signals is shifted according to the time offset. It should be noted that translation can be done by subtracting the time offset from the time axis of the actual running near-end response signal, or by adding the time offset to the time axis of the near-end response signal under healthy operating conditions. Based on the injected pulse amplitude under healthy operating conditions and actual operating conditions, the amplitude of the two sets of proximal response signals is normalized so that the injected pulses in the two sets of proximal response signals have the same amplitude reference. Differential operation is performed on the two sets of near-end response signals after time alignment and amplitude normalization to obtain the real-time differential signal.

[0013] Preferably, the global search time range for the fault reflection wave is set based on the injected pulse reference time and the set pulse propagation speed, specifically as follows: Based on the minimum and maximum positions where an inter-turn short circuit is allowed in the rotor winding under test, the set pulse propagation speed, and the injection pulse reference time, the global search time range of the fault reflection wave is determined. The left endpoint of the global search time range is the difference between the round-trip propagation time corresponding to the minimum position and the left protection margin, and the right endpoint is the sum of the round-trip propagation time corresponding to the maximum position and the right protection margin. The round-trip propagation time corresponding to the minimum position is the injection pulse reference time plus the set round-trip time from the high-frequency pulse injection end to the minimum position and back to the high-frequency pulse injection end. The round-trip propagation time corresponding to the maximum position is the injection pulse reference time plus the set round-trip time from the high-frequency pulse injection end to the maximum position and back to the high-frequency pulse injection end.

[0014] Preferably, the step of searching for candidate wave packets from the envelope through local maximum detection and continuous threshold search within the global search time range specifically involves: Within the global search time range, search for local maxima of the envelope, and select local maxima whose envelope amplitude is higher than the maximum amplitude of the background envelope as candidate peaks; determine whether there are more than a set number of consecutive sampling points forming an rising band before the candidate peak, and whether there are more than a set number of consecutive sampling points forming a falling band after the candidate peak. If so, select the rising band, candidate peak, and falling band as candidate wave packets. It should be noted that the background envelope is the range of fluctuations in the envelope of the near-end differential reflection signal, excluding the fault reflection packet. The background components include residual propagation modes of the winding, residual reflections at the end boundaries, measurement noise, and high-frequency oscillations, the amplitude of which is lower than the peak value of the fault reflection packet. When the amplitude of a local maximum is higher than the maximum amplitude of the background envelope, the local maximum is determined to be a candidate peak; otherwise, it is considered as background fluctuation and ignored.

[0015] When multiple candidate wave packets exist within the global search time range, the candidate wave packet with the earliest arrival time and whose envelope peak value and wave packet energy are both higher than the peak value and energy of the background oscillation are selected as the final candidate wave packet.

[0016] Preferably, the step of setting an adaptive threshold based on the maximum envelope amplitude of a set local time window, centered on the candidate wave packet, specifically involves: The adaptive threshold is the product of the maximum amplitude of the envelope within the set local time window and the ratio of the set threshold.

[0017] Preferably, the moment of the reflected wave leading edge and the moment of the first main peak are detected within a set local time window based on an adaptive threshold, specifically as follows: When the amplitude of the envelope at a set number of consecutive sampling points after the current position is greater than the adaptive threshold, the corresponding position satisfies the continuous threshold crossing condition. The position that first satisfies the continuous threshold crossing condition is determined as the candidate leading edge position. Linear interpolation is used to process the candidate leading edge position and the adjacent sampling points before the candidate leading edge position. The linear interpolation is: fitting the candidate leading edge position and the adjacent sampling points before the candidate leading edge position as a linear function of the envelope amplitude and time. The time when the envelope amplitude of the linear function is equal to the adaptive threshold is taken as the moment of the reflected wave leading edge. A main peak search time window is established starting from the moment of the reflected wave leading edge. Within this window, the moment corresponding to the local maximum value where the envelope amplitude is greater than that of two adjacent sampling points and exceeds the adaptive threshold is taken as the main peak moment, and the first main peak moment is recorded.

[0018] Preferably, the moments of the reflected wave leading edge and the first main peak are summed according to a set weight, specifically: The weights corresponding to the leading edge of the reflected wave and the first peak are set based on simulation. Initially, the weights corresponding to the leading edge of the reflected wave and the first peak are set to their initial values. For the differential signal corresponding to each fault condition, candidate wave packets are searched from the envelope of the corresponding differential signal within the global search time range through local maximum detection and continuous threshold search. If the continuous threshold condition is met only once within the local time window centered on the candidate wave packet and there is only one peak, the weight of the leading edge of the reflected wave is increased; otherwise, the weight of the first peak is increased. This process continues until all fault conditions have been traversed, resulting in the final weights corresponding to the leading edge of the reflected wave and the first peak.

[0019] It should be noted that the weights corresponding to the leading edge of the reflected wave and the first peak are always between 0 and 1, and the sum of the weights corresponding to the leading edge of the reflected wave and the first peak is always 1. If a weight is already 1, and there is a fault condition detection that still needs to increase the corresponding weight, then the weight remains unchanged.

[0020] Preferably, the calculation of the initial fault location based on the fused positioning time specifically involves: The difference between the fusion positioning time and the injection pulse reference time is calculated. The difference is multiplied by the set pulse propagation speed and divided by 2 to obtain the initial fault location.

[0021] Preferably, the input feature vector of the ridge regression localization correction model consists of the initial fault location, the leading edge propagation time, the main peak propagation time, and the leading edge-main peak time interval; wherein, the leading edge propagation time is the difference between the moment of the reflected wave leading edge and the reference moment of the injected pulse, the main peak propagation time is the difference between the moment of the first main peak and the reference moment of the injected pulse, and the leading edge-main peak time interval is the difference between the moment of the first main peak and the moment of the reflected wave leading edge.

[0022] A second aspect of the present invention proposes a short-circuit localization system for a synchronous condenser based on envelope fusion and ridge regression using the method described in the first aspect of the present invention, comprising a data acquisition module, a simulation module, an envelope extraction module, a module for extracting the leading edge time of the reflected wave and the first main peak time, and a fault localization module, characterized in that: Acquisition end setting module: A high-frequency pulse injection end is set at one end of the rotor winding. A near-end response acquisition channel is set within the set range of the high-frequency pulse injection end. After injecting the set pulse at the high-frequency pulse injection end, the signal of the near-end response acquisition channel is acquired as the near-end response signal. Simulation module: Establish a high-frequency distributed parameter model of the rotor winding of the synchronous condenser for simulation, and obtain the near-end response signal under healthy operating conditions; Envelope extraction module: When the synchronous condenser is actually running, after injecting a set pulse into the synchronous condenser, it obtains the proximal response signal during actual operation, calculates the difference signal between the proximal response signal during actual operation and the proximal response signal under simulated healthy operating conditions as a real-time differential signal, and extracts the envelope from the real-time differential signal. The module for extracting the leading edge and first peak of the reflected wave is used to set the global search time range of the fault reflected wave based on the injection pulse reference time and the set pulse propagation speed. Within the global search time range, candidate wave packets are searched from the envelope through local maximum detection and continuous threshold search. With the candidate wave packets as the center, an adaptive threshold is set based on the maximum envelope amplitude of the set local time window. Based on the adaptive threshold, the leading edge and first peak of the reflected wave are detected within the set local time window. Fault location module: It is used to obtain the fused location time by weighting and summing the reflection wave leading edge time and the first main peak time according to the set weights, and calculate the initial fault location based on the fused location time; it inputs the reflection wave leading edge time, the first main peak time and the initial fault location into the pre-trained ridge regression location correction model, and outputs the final fault location.

[0023] A third aspect of the present invention provides a terminal, including a processor and a storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method and system for inter-turn short-circuit localization of a condenser based on envelope fusion and ridge regression as described in any one of claims 1-10.

[0024] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when executed by a processor, the program implements the steps of the method and system for inter-turn short-circuit localization of a condenser based on envelope fusion and ridge regression as described in any one of claims 1-10.

[0025] The beneficial effects of the present invention are as follows: (1) By performing time alignment, amplitude normalization and differential processing on the near-end response of the healthy state and the fault state, and further extracting the Hilbert envelope, the influence of trigger offset, amplitude fluctuation and common propagation background of winding on fault characteristics can be weakened, making the additional reflected wave packet caused by the inter-turn short circuit more prominent, thus providing a basis for the stable extraction of the arrival time of the fault reflected wave. (2) By adopting a hierarchical search method from global to local, and combining adaptive threshold, continuous threshold crossing criterion and leading edge-main peak fusion positioning method, the misjudgment caused by factors such as noise spikes, local high-frequency oscillations and end reflections can be reduced; at the same time, by using the first arrival information of the reflected wave leading edge and the main energy information of the first main peak for complementarity, the positioning deviation of using the leading edge time alone is prone to being too early and using the main peak time alone is prone to being too late can be reduced, thus improving the stability of the initial positioning result. (3) To address the continuous systematic positioning error caused by factors such as approximate propagation speed, spatial discretization, time sampling, waveform broadening and threshold identification during high-frequency pulse propagation, a ridge regression model is introduced to correct the initial positioning results and related time characteristics. This can further compensate for the positioning deviation, improve the accuracy and stability of the inter-turn short circuit fault location of the synchronous condenser rotor winding, and enhance the adaptability of the method to different fault locations and different short circuit conditions. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the inter-turn short-circuit location method for a synchronous condenser based on envelope fusion and ridge regression according to the present invention; Figure 2 This is a structural diagram of the rotor core and windings of the TTS-300-2 type synchronous condenser in this invention; Figure 3 This is the wiring diagram for high-frequency detection of the synchronous condenser rotor winding RSO in this invention; Figure 4 This is a comparison diagram of the near-end differential reflection envelope and arrival time under different excitation methods in this invention; Figure 5 This is a comparison diagram of the proximal reflection envelope at different fault locations in this invention; Figure 6 This is a differential reflection waveform diagram of the near end of a fault at 900 m under different short-circuit turns in this invention; Figure 7 This is a diagram of the average absolute positioning error at different fault locations in this invention; Figure 8 This is a distribution diagram of absolute positioning error under all operating conditions. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0028] like Figure 1 As shown, Embodiment 1 of the present invention proposes a method and system for inter-turn short-circuit localization of a synchronous condenser based on envelope fusion and ridge regression, comprising: S1. Establish a high-frequency distributed parameter model of the rotor winding of the synchronous condenser for simulation, and obtain the near-end response signal under healthy operating conditions; the near-end response signal refers to the signal collected from the pre-set near-end response acquisition channel after a high-frequency pulse injection terminal is set at one end of the rotor winding and a set high-frequency pulse is injected.

[0029] It should be noted that the frequency of the injected pulse is higher than the set high-frequency threshold, and is therefore a high-frequency pulse.

[0030] Establish a high-frequency distributed parameter model for the rotor winding, specifically as follows: The rotor winding is discretized along its length into several spatial micro-elements. Each micro-element is characterized by resistance, inductance, conductance, and capacitance per unit length, and distributed parameter equations satisfying the voltage and current at position and time are established; for example... Figure 2 As shown, this embodiment takes the TTS-300-2 type synchronous condenser as the object. The total length of the winding is 2849.664m, which is discrete into 384 units and the end is open. When there is a short circuit between turns, a local equivalent short circuit branch is introduced in the corresponding unit so that the high frequency pulse generates additional reflection at the fault location.

[0031] A high-frequency distributed parameter model of the synchronous condenser rotor winding was established. The rotor winding of the TTS-300-2 synchronous condenser is distributed along the rotor slots and forms a complete excitation circuit through end connections. The rotor winding has a large axial length, and parasitic capacitance and electromagnetic coupling exist between adjacent conductors and between conductors and the rotor body. Under the action of high-frequency pulses, the rotor winding exhibits obvious traveling wave propagation, attenuation, reflection, and transmission characteristics. Therefore, it cannot be simply equivalent to a lumped parameter circuit composed of a single total resistance, total inductance, and total capacitance.

[0032] The rotor winding is discretized along its length into several spatial micro-elements, each characterized by unit-length resistance, unit-length inductance, unit-length conductance, and unit-length capacitance. Let the position... x ,time t The voltage and current at the point are respectively u (x,t) and i(x,t), then the high-frequency distributed parameter model of the winding satisfies:

[0033] In the formula, R 0 represents resistance per unit length. L 0 represents the inductance per unit length. G 0 represents the conductivity per unit length. C 0 represents capacitance per unit length.

[0034] Neglecting the effects of strong dispersion, the propagation speed of high-frequency pulses in the rotor windings Approximately expressed as:

[0035] In this implementation example, the TTS-300-2 synchronous condenser has a rated capacity of 300 MVA, a rated speed of 3000 r / min, 32 rotor slots, and a total rotor winding length of approximately 2849.664 m. The rotor winding is discretized into 384 spatial elements with a spatial step size of approximately 7.421 m. The model ends are set as open-circuit boundaries to characterize the total internal reflection phenomenon caused by high-frequency pulses at the winding ends.

[0036] When an inter-turn short circuit occurs at a specified location in the winding, a local equivalent short-circuit branch is introduced in the corresponding space cell. The inter-turn short circuit changes the equivalent impedance and loss parameters near the fault location, disrupting the continuity of the original parameter distribution of the winding and causing additional reflections of high-frequency pulses at the fault location.

[0037] S2. When the synchronous condenser is actually running, after injecting the set pulse into the synchronous condenser, the proximal response signal during actual operation is obtained, and the difference signal between the proximal response signal during actual operation and the proximal response signal under the simulated healthy operating condition is calculated as the real-time differential signal. The envelope is extracted from the real-time differential signal. It should be noted that the pulse type selected in this invention is a Gaussian pulse. The Gaussian pulse is used as the excitation, injected at one end of the winding, and the near-end response is collected near the injection end. The healthy and fault responses are first time-aligned and amplitude-normalized, and then differential analysis is performed to eliminate pseudo-differential components caused by trigger offset and amplitude fluctuations. Specifically: Determine the high-frequency excitation method and obtain the near-end response. For example... Figure 3 As shown, based on the principle of repetitive pulse detection of the rotor winding of a synchronous condenser, a high-frequency pulse injection end is set at one end of the rotor winding, and a near-end response acquisition channel is set near the injection end. The same pulse generator, sampling frequency, connection method, and sampling time are used for both the healthy state and the state to be detected.

[0038] The specific process for selecting the set pulse type is as follows: To obtain fault reflection responses with clear wavefronts and high time resolution, the near-end differential reflection envelopes under Gaussian pulses, double pulses, linear sweep pulses, and repetitive pulses were compared. The results are as follows: Figure 4 As shown. Double pulses and repetitive pulses contain multiple pulse components, which are prone to superimposing with end-reflected waves during propagation; linear sweep pulses have a long duration and are prone to envelope tailing; Gaussian pulses have the characteristics of single peak, narrow pulse and concentrated time-domain energy, and their corresponding near-end differential reflection envelope leading edge is relatively clear, making them suitable for identifying the arrival time of reflected waves.

[0039] Therefore, this implementation uses a Gaussian pulse as the unified excitation signal. The Gaussian pulse is represented as:

[0040] In the formula, A The pulse amplitude. t 0 represents the pulse center time. σ This is the pulse width parameter.

[0041] In this embodiment, the high-frequency distributed parameter model of the rotor winding for establishing the synchronous condenser is simulated, specifically as follows: Various fault conditions are formed by setting different fault locations and different short-circuit turns. The differential signal between the near-end response signal and the near-end response signal under the healthy condition is calculated for each fault condition. The differential reflection envelope is extracted from the differential signal corresponding to each fault condition. The reflection wave leading moment and the first main peak moment are extracted from the differential reflection envelope. The initial fault location is obtained by fusing the two. The reflection wave leading moment, the first main peak moment and the initial fault location are used as a training sample. The actual fault location of the corresponding fault condition is used as the label of the training sample. The ridge regression localization correction model is trained using all training samples.

[0042] The specific parameters are as follows: Gaussian pulse amplitude is set to 5 V, pulse width is set to 22 ns, and high-frequency pulse propagation speed is set to 1.8 × 10⁻⁶. 8 The fault location was set within a range of 100–2800 m from the injection end, with adjacent fault locations spaced 100 m apart. The number of short-circuit turns was set from 1 to 9 turns, used only to construct simulation conditions with different fault degrees to verify the adaptability of the location method to different fault degrees. In actual testing, the input to the location method was the high-frequency response signal in the healthy state and the state to be tested, without needing to obtain the number of short-circuit turns in advance. The near-end response in the healthy state and the near-end response in the fault state were calculated separately for each set of conditions.

[0043] Let the proximal response signal collected under healthy conditions be... u h (t), the near-end response signal acquired under fault conditions isu f (t), where t is time.

[0044] Before differential processing, the near-end responses in both healthy and faulty states are first time-aligned and amplitude-normalized. The moment when the rising edge of the injected pulse first reaches a set threshold in both acquisitions is used as the time reference, and the time offset between the two sets of responses is calculated. If the rising edge of the injected pulse crosses the set threshold multiple times, or fails to remain above the threshold after the first crossing, making it impossible to determine a unique first stable threshold crossing moment, then the rising edge is considered unstable due to noise. In this case, the peak time of the injected pulse is used as the time reference. One set of responses is then shifted according to the time offset to achieve time alignment between the two sets of responses. It should be noted that the shift can be achieved by subtracting the time offset from the time axis of the actual near-end response signals during operation, or by adding the time offset to the time axis of the near-end response signals under healthy conditions.

[0045] Subsequently, the corresponding proximal responses are normalized based on the injected pulse amplitudes in both healthy and faulty states, ensuring that the injected pulses in both sets of responses have the same amplitude reference. After time alignment and amplitude normalization, the difference operation between the healthy and faulty state responses is then performed to avoid trigger time offsets and pulse amplitude fluctuations forming pseudo-differential components.

[0046] S3. Differentiate the near-end response of the fault state and the near-end response of the healthy state after time alignment and amplitude normalization to obtain the additional reflection signal generated by the inter-turn short circuit; perform Hilbert transform on the differential reflection signal to obtain a continuous differential reflection envelope, which is used to characterize the starting position and main energy position of the fault reflection wave packet. Specifically: A near-end differential reflection signal is constructed and its Hilbert envelope is extracted. The healthy-state near-end response includes components such as injected pulses, inherent winding propagation modes, propagation attenuation, and end-boundary reflections. To highlight the local additional reflections caused by inter-turn short circuits, the fault-state response is subtracted from the healthy-state response to obtain the near-end differential reflection signal:

[0047] Differential processing can reduce the common propagation background in both states, making the additional reflected waves caused by the impedance change at the fault point more obvious.

[0048] Because the near-end differential reflection signal still exhibits high-frequency oscillations and alternating positive and negative signals, directly extracting the arrival time of the fault reflection wave from the original signal peak value is easily affected by local oscillations. Therefore, a Hilbert transform is applied to the near-end differential reflection signal to construct an analytic signal:

[0049] In the formula, H [Δu(t)] is Δ u Hilbert transform of (t).

[0050] The Hilbert envelope of the differential reflection signal is:

[0051] Hilbert envelope can transform alternating positive and negative high-frequency oscillation waveforms into continuous wave packet curves, highlighting the starting position and main energy position of the reflected wave packet.

[0052] The arrival time of the near-end reflection envelope at different fault locations gradually shifts later as the fault distance increases, indicating a clear correspondence between the arrival time of the fault reflection wave and the fault location.

[0053] For simulation samples with known fault locations, the reflected wave leading edge time, the first peak time, and the initial positioning results extracted from the differential reflection envelope are used to train the subsequent ridge regression positioning correction model. For samples to be detected, instead of directly calling the fixed envelope data corresponding to a known fault location, positioning features are extracted from the current differential reflection envelope using the same method, and these features are input into the trained ridge regression positioning correction model to obtain the final fault location.

[0054] S4. Determine the global search time interval based on the possible location range of the fault. Search for candidate wave packets within the global search time interval, and then narrow down to a local time window centered on the candidate wave packets. Determine the moment of the reflected wave leading edge using an adaptive threshold and a continuous threshold crossing criterion, and improve the time resolution using interpolation. Specifically: A hierarchical search method is used to extract the leading edge and the first main peak of the reflected wave. To avoid directly searching for the reflected wave in the entire near-end response signal, this implementation uses a hierarchical search method from global to local.

[0055] First, based on the unfolded length of the rotor winding under test, the position of the pulse injection end, and the winding section where inter-turn short circuits are permissible, determine the minimum possible location of the fault. x min and maximum position x max The aforementioned x min and x max The configuration is determined by the winding structure of the device being tested and is not arbitrarily set. In this embodiment, x min =100 m, x max=2800 m. Since the high-frequency pulse propagates from the injection end to the fault point and then reflects back to the injection end, the global search time range of the fault reflection wave is determined as follows:

[0056] In the formula, t 0 represents the reference time for the injected pulse. v For the propagation speed of a high-frequency pulse, Δ t L and Δ t R These represent the left and right protection margins for the global search time range, used to compensate for propagation speed approximation errors, numerical discretization errors, sampling errors, and time offsets caused by waveform broadening.

[0057] Secondly, a combination of local maximum detection and continuous threshold search is used to search for candidate wave packets within the global search time range. First, local maxima in the differential envelope are searched, and those with peak amplitudes higher than the maximum amplitude of the background envelope are selected as candidate peaks. Then, it is checked whether the envelope signals before and after the candidate peak form continuous rising and falling bands, and isolated spikes consisting of only a single sampling point are excluded. Continuous envelope bands that meet the above conditions are identified as candidate wave packets. If multiple candidate wave packets exist within the global search time range, the candidate wave packet with the earliest arrival time and whose envelope peak value and wave packet energy are significantly higher than the background oscillation are preferentially selected. Subsequently, the search range is narrowed down centered on this candidate wave packet, and the reflected wave front and the first main peak are further extracted within the local time window.

[0058] Set an adaptive threshold within a local time window:

[0059] In the formula, e max This represents the maximum envelope amplitude within the local search time window. η This represents the threshold ratio.

[0060] The search begins sequentially from the left end of the local search time window. When the envelope value is greater than the adaptive threshold for three consecutive sampling points... T h When the continuous threshold crossing condition is first met, the position is determined as the candidate leading edge position. Using the continuous threshold crossing condition can avoid misjudgment caused by a single noise spike or isolated oscillation point crossing the threshold.

[0061] To reduce the limitation of the sampling time interval on the timing of the reflected wave front, when the amplitude of a set number of consecutive sampling points after the current position is greater than the adaptive threshold, the corresponding position satisfies the continuous threshold crossing condition. The position that first satisfies the continuous threshold crossing condition is determined as the candidate front position. Linear interpolation is used to process the candidate front position and the adjacent sampling points before the candidate front position. The linear interpolation is: fitting the candidate front position and the adjacent sampling points before the candidate front position as a linear function of the envelope amplitude and time. The time when the envelope amplitude of the linear function is equal to the adaptive threshold is taken as the timing ta of the reflected wave front.

[0062] The moment of the reflected wave leading edge t a Located in the starting region of the reflected wave packet, it can reflect the first arrival information of the fault reflected wave, but due to the weak amplitude of the leading edge, it is not suitable for use alone. t a This can easily lead to underestimation of the positioning result. At the moment the leading edge of the reflected wave is detected... t a After that, with t a A primary peak search time window is established starting from this point. Within this window, peaks with envelope amplitudes exceeding the adaptive threshold are searched sequentially, with peak values ​​greater than the two adjacent sampling points. T h The local maxima are identified, and isolated spikes consisting of only a single sampling point are excluded. Following the chronological order, the first local maximum satisfying the above conditions is determined as the first valid main peak, and its corresponding time is denoted as [time value missing]. t p .

[0063] First peak moment t p It reflects the time when the main energy of the reflected wave packet arrives at the injection end and has good resistance to weak signal interference. However, since the main peak is located inside the wave packet, propagation attenuation and waveform broadening will cause... t p There is a lag compared to the actual first arrival time; it can be used alone. t p This can easily lead to overestimation of the location results.

[0064] Taking the fault location at 900 m and the short circuit of 5 turns as an example, such as Figure 6 As shown, the near-end differential reflection signal forms a distinct wave packet in the area where the fault reflection wave arrives, which allows the moment of the leading edge of the reflection wave and the moment of the first main peak to be extracted within the local search window.

[0065] S5, the moment of the leading edge of the reflected wave t a With the first main peak moment t pWeighted fusion is performed to obtain the fusion location time tf (the fusion coefficient α satisfies 0≤α≤1; the weight of the leading edge is increased when the wave packet is clear and the leading edge identification is stable; the weight of the main peak is increased when the reflected wave amplitude is weak or the local oscillation is complex); then, based on the time relationship of the round-trip propagation of the high-frequency pulse between the injection end and the fault point, the initial fault location x is calculated from the fusion location time. raw By fusing the leading edge and the main peak, the negative bias caused by using the leading edge moment alone and the positive bias caused by using the main peak moment alone are reduced, making the initial positioning results more stably reflect the changes in fault location. Specifically: Front-peak fusion positioning is performed. To simultaneously utilize the first-arrival information and main energy information of the reflected wave, the moment of the reflected wave's front is determined. t a and the first main peak moment t p Perform weighted fusion to obtain the fused positioning time. t f :

[0066] In the formula, α is the fusion coefficient, which satisfies 0≤α≤1.

[0067] When there is only one valid main peak exceeding the adaptive threshold within the local search time window, and the reflected wave front only exceeds the threshold once consecutively, the wave packet is determined to be clear, the front identification is stable, and the t is improved. a The weighting is adjusted; when there are two or more local peaks exceeding the adaptive threshold, or when the reflected wave front shows multiple consecutive threshold crossings, the local oscillation is deemed complex, and the weighting is increased. t p The weights are determined based on the localization error of the known fault location samples and remain constant during actual detection. The high-frequency pulse propagates from the injection end to the fault point and is then reflected back from the fault point to the injection end; therefore, the high-frequency pulse undergoes a round-trip propagation process from the injection end to the fault point and back to the injection end. The initial fault location is calculated based on the fusion localization time.

[0068] In the formula, x raw The initial fault location is obtained by fusing the frontier and the main peak. v For the propagation speed of high-frequency pulses, t 0 represents the reference time of the injected pulse, and 2 in the denominator represents the round-trip propagation process of the high-frequency pulse.

[0069] By fusing the leading edge and the main peak, the negative bias caused by using the leading edge moment alone and the positive bias caused by using the main peak moment alone can be reduced, making the initial positioning results more stably reflect the changes in the fault location.

[0070] S6. Train the ridge regression model using known fault location samples. Take the initial location result and related time features as input and the actual fault location as target. Correct the initial location result of the sample to be detected and output the final fault location to compensate for systematic deviations caused by factors such as approximate propagation speed, sampling discreteness and waveform broadening.

[0071] A ridge regression localization correction model was established. Although leading-peak fusion can improve the stability of initial localization, the propagation of high-frequency pulses in the rotor winding is still affected by factors such as approximate propagation speed, spatial discretization, time sampling, waveform broadening, and threshold identification, which can affect the initial fault location. x raw There is a certain continuous systematic deviation between the actual fault location and the location of the fault, such as Figure 7 and Figure 8 As shown.

[0072] To compensate for the systematic bias, ridge regression training data is constructed using samples from different known fault locations. The feature vector of the i-th sample... Φ i The initial fault location of the corresponding sample x raw,i Frontier dissemination time t a,i t 0,i Main peak transmission time t p,i t 0,i and the time interval between the frontier and the main peak t p,i t a,i Composition, including the frontier propagation time t a,i t 0,i The moment of the leading edge of the reflected wave in this sample t a,i Reference time of injection pulse t 0,i The difference represents the time information of the first arrival of the high-frequency pulse from the injection end to the fault point and back to the injection end; the main peak propagation time t p,i t 0,i This is the first peak moment in the sample. t p,i Reference time of injection pulse t 0,iThe difference represents the arrival time information of the main energy of the reflected wave packet at the injection end; the time interval between the leading edge and the main peak. t p,i t a,i The difference between the first peak time and the leading edge time of the reflected wave in this sample represents the wave packet width information caused by propagation attenuation and waveform broadening of the reflected wave packet. These features respectively characterize the initial positioning result, the first arrival information of the reflected wave, the arrival information of the main energy, and the degree of waveform broadening, and are based on the actual fault location. x i This is the target value.

[0073] The ridge regression model obtains its parameters by solving the following equation:

[0074] In the formula, To obtain the ridge regression model parameters, The model parameters to be optimized are... λ The regularization coefficient is . n This represents the number of training samples.

[0075] Regularization term This is used to limit the magnitude of model parameters, reduce the risk of overfitting caused by the correlation of sample features and local abnormal samples, and keep the correction relationship between the initial localization results and the actual fault location continuous and smooth.

[0076] For the sample to be detected, its initial location result and corresponding location features are input into the ridge regression model to obtain the corrected final fault location:

[0077] In the formula, This is the final output location of the inter-turn short-circuit fault.

[0078] In this invention, ridge regression is used only to correct systematic biases caused by front-peak fusion positioning, and is not used for inter-turn short-circuit fault type identification or severity classification.

[0079] The beneficial effects of the present invention are that, compared with the prior art, (1) by performing time alignment, amplitude normalization and differential processing on the near-end response of the healthy state and the fault state, and further extracting the Hilbert envelope, the influence of trigger offset, amplitude fluctuation and common propagation background of winding on fault characteristics can be weakened, making the additional reflected wave packet caused by the inter-turn short circuit more prominent, thereby providing a basis for the stable extraction of the arrival time of the fault reflected wave.

[0080] (2) By adopting a hierarchical search method from global to local and combining adaptive threshold, continuous threshold crossing criterion and front-end-main peak fusion positioning method, misjudgment caused by noise spikes, local high-frequency oscillations and end reflections can be reduced. At the same time, by using the first arrival information of the reflected wave front and the main energy information of the first main peak to complement each other, the positioning deviation of using the front moment alone is prone to being too early and using the main peak moment alone is prone to being too late can be reduced, thus improving the stability of the initial positioning result.

[0081] (3) To address the continuous systematic positioning error caused by factors such as approximate propagation speed, spatial discretization, time sampling, waveform broadening and threshold identification during high-frequency pulse propagation, a ridge regression model is introduced to correct the initial positioning results and related time characteristics. This can further compensate for the positioning deviation, improve the accuracy and stability of the inter-turn short circuit fault location of the synchronous condenser rotor winding, and enhance the adaptability of the method to different fault locations and different short circuit conditions.

[0082] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0083] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0084] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0085] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for locating short circuits between turns of a synchronous condenser based on envelope fusion and ridge regression, characterized in that, include: A high-frequency distributed parameter model of the rotor winding of the synchronous condenser was established for simulation to obtain the near-end response signal under healthy operating conditions. The near-end response signal refers to the signal collected from a pre-set near-end response acquisition channel after a high-frequency pulse injection terminal is set at one end of the rotor winding and a set high-frequency pulse is injected. During actual operation of the synchronous condenser, after injecting a set high-frequency pulse into the synchronous condenser, the proximal response signal during actual operation is obtained. The difference between the proximal response signal during actual operation and the proximal response signal under simulated healthy operating conditions is calculated as a real-time differential signal, and the envelope is extracted from the real-time differential signal. The global search time range of the fault reflection wave is set based on the injection pulse reference time and the set pulse propagation speed; within the global search time range, candidate wave packets are searched from the envelope through local maximum detection and continuous threshold search; with the candidate wave packets as the center, an adaptive threshold is set based on the maximum envelope amplitude of the set local time window; based on the adaptive threshold, the leading edge time and the first main peak time of the reflection wave are detected within the set local time window. The fused location time is obtained by weighting and summing the reflection wave leading edge time and the first main peak time according to the set weights. The initial fault location is calculated based on the fused location time. The reflection wave leading edge time, the first main peak time and the initial fault location are input into the pre-trained ridge regression localization correction model to output the final fault location.

2. The method for inter-turn short-circuit localization of a synchronous condenser based on envelope fusion and ridge regression according to claim 1, characterized in that: The simulation of the high-frequency distributed parameter model of the rotor winding of the synchronous condenser is specifically as follows: Various fault conditions are formed by setting different fault locations and different short-circuit turns; the differential signal between the near-end response signal under each fault condition and the near-end response signal under the healthy condition is calculated; the differential reflection envelope is extracted from the differential signal corresponding to each fault condition; the reflection wave leading edge time and the first main peak time are extracted from the differential reflection envelope; and the initial fault location is obtained by fusing the two; the reflection wave leading edge time, the first main peak time and the initial fault location are used as a training sample; and the actual fault location of the corresponding fault condition is used as the label of the training sample. The subsequent ridge regression localization correction model was trained using all training samples.

3. The inter-turn short-circuit location method for synchronous condensers based on envelope fusion and ridge regression according to claim 1, characterized in that: The difference signal between the proximal response signal during actual operation and the simulated proximal response signal under healthy operating conditions is specifically calculated as follows: The rising edge of the injection pulse is used to detect the proximal response signal during actual operation and the proximal response signal under healthy operating conditions. The time when the rising edge of the injected pulse reaches a set threshold is used as the time reference. When the rising edge of the injected pulse cannot be detected directly, the peak time of the injected pulse is used as the time reference. The time reference difference between the proximal response signal during actual operation and the proximal response signal under healthy operating conditions is calculated as the time offset, and one set of proximal response signals is shifted according to the time offset. Based on the injected pulse amplitude under healthy operating conditions and actual operating conditions, the amplitude of the two sets of proximal response signals is normalized so that the injected pulses in the two sets of proximal response signals have the same amplitude reference. Differential operation is performed on the two sets of near-end response signals after time alignment and amplitude normalization to obtain the real-time differential signal.

4. The method for inter-turn short-circuit localization of a synchronous condenser based on envelope fusion and ridge regression according to claim 1, characterized in that: The global search time range for the fault reflection wave is set based on the injected pulse reference time and the set pulse propagation speed, specifically as follows: Based on the minimum and maximum positions where an inter-turn short circuit is allowed in the rotor winding under test, the set pulse propagation speed, and the injection pulse reference time, the global search time range of the fault reflection wave is determined. The left endpoint of the global search time range is the difference between the round-trip propagation time corresponding to the minimum position and the left protection margin, and the right endpoint is the sum of the round-trip propagation time corresponding to the maximum position and the right protection margin. The round-trip propagation time corresponding to the minimum position is the injection pulse reference time plus the set round-trip time from the high-frequency pulse injection end to the minimum position and back to the high-frequency pulse injection end. The round-trip propagation time corresponding to the maximum position is the injection pulse reference time plus the set round-trip time from the high-frequency pulse injection end to the maximum position and back to the high-frequency pulse injection end.

5. The method for inter-turn short-circuit localization of a synchronous condenser based on envelope fusion and ridge regression according to claim 1, characterized in that: The process of searching for candidate wave packets from the envelope within the global search time range through local maximum detection and continuous threshold search specifically involves: Within the global search time range, search for local maxima of the envelope, and select local maxima whose envelope amplitude is higher than the maximum amplitude of the background envelope as candidate peaks; determine whether there are more than a set number of consecutive sampling points forming an rising band before the candidate peak, and whether there are more than a set number of consecutive sampling points forming a falling band after the candidate peak. If so, select the rising band, candidate peak, and falling band as candidate wave packets. When multiple candidate wave packets exist within the global search time range, the candidate wave packet with the earliest arrival time and whose envelope peak value and wave packet energy are both higher than the peak value and energy of the background oscillation are selected as the final candidate wave packet.

6. The method for inter-turn short-circuit localization of a synchronous condenser based on envelope fusion and ridge regression according to claim 1, characterized in that: The adaptive threshold, centered on the candidate wave packet and based on the maximum envelope amplitude of a set local time window, is set as follows: The adaptive threshold is the product of the maximum amplitude of the envelope within the set local time window and the ratio of the set threshold.

7. The method for inter-turn short-circuit localization of a synchronous condenser based on envelope fusion and ridge regression according to claim 5, characterized in that: Based on an adaptive threshold, the leading edge moment and the first main peak moment of the reflected wave are detected within a set local time window, specifically: Starting from the left end of the set local time window, the search proceeds point by point. When the envelope amplitude of a set number of consecutive sampling points starting from the k-th sampling point is greater than the adaptive threshold, and the k-th sampling point... When the envelope amplitude of a single sampling point is not greater than the adaptive threshold, the continuous threshold crossing condition is satisfied, and the k-th sampling point is determined as a candidate leading edge position; for the k-th... The envelope amplitude corresponding to the first sampling point and the kth sampling point is linearly interpolated with the sampling time. The time when the envelope amplitude of the interpolation line equals the adaptive threshold is determined as the time of the reflected wave front. A main peak search time window is established starting from the moment of the reflected wave leading edge. Within this window, the moment corresponding to the local maximum value where the envelope amplitude is greater than that of two adjacent sampling points and exceeds the adaptive threshold is taken as the main peak moment, and the first main peak moment is recorded.

8. The method for inter-turn short-circuit localization of a synchronous condenser based on envelope fusion and ridge regression according to claim 7, characterized in that: The moments of the leading edge of the reflected wave and the moment of the first main peak are summed according to a set weight, specifically as follows: The weights corresponding to the leading edge of the reflected wave and the first peak are set based on simulation. Initially, the weights corresponding to the leading edge of the reflected wave and the first peak are set to their initial values. For the differential signal corresponding to each fault condition, candidate wave packets are searched from the envelope of the corresponding differential signal within the global search time range through local maximum detection and continuous threshold search. If the continuous threshold condition is met only once within the local time window centered on the candidate wave packet and there is only one peak, the weight of the leading edge of the reflected wave is increased; otherwise, the weight of the first peak is increased. This process continues until all fault conditions have been traversed, resulting in the final weights corresponding to the leading edge of the reflected wave and the first peak.

9. The method for inter-turn short-circuit localization of a synchronous condenser based on envelope fusion and ridge regression according to claim 1, characterized in that: The calculation of the initial fault location based on the fused positioning time is specifically as follows: The difference between the fusion positioning time and the injection pulse reference time is calculated. The difference is multiplied by the set pulse propagation speed and divided by 2 to obtain the initial fault location.

10. The method for inter-turn short-circuit localization of a synchronous condenser based on envelope fusion and ridge regression according to claim 1, characterized in that: The input feature vector of the ridge regression localization correction model consists of the initial fault location, the leading edge propagation time, the main peak propagation time, and the leading edge-main peak time interval. Among them, the leading edge propagation time is the difference between the moment of the reflected wave leading edge and the reference moment of the injected pulse, the main peak propagation time is the difference between the moment of the first main peak and the reference moment of the injected pulse, and the leading edge-main peak time interval is the difference between the moment of the first main peak and the moment of the reflected wave leading edge.

11. A short-circuit localization system for synchronous condensers based on envelope fusion and ridge regression using the method described in any one of claims 1-10, comprising a data acquisition module, a simulation module, an envelope extraction module, a module for extracting the leading edge time of reflected waves and the first main peak time, and a fault localization module, characterized in that: Acquisition end setting module: A high-frequency pulse injection end is set at one end of the rotor winding. A near-end response acquisition channel is set within the set range of the high-frequency pulse injection end. After injecting the set pulse at the high-frequency pulse injection end, the signal of the near-end response acquisition channel is acquired as the near-end response signal. Simulation module: Establish a high-frequency distributed parameter model of the rotor winding of the synchronous condenser for simulation, and obtain the near-end response signal under healthy operating conditions; Envelope extraction module: When the synchronous condenser is actually running, after injecting a set pulse into the synchronous condenser, it obtains the proximal response signal during actual operation, calculates the difference signal between the proximal response signal during actual operation and the proximal response signal under simulated healthy operating conditions as a real-time differential signal, and extracts the envelope from the real-time differential signal. The module for extracting the leading edge and first peak of the reflected wave is used to set the global search time range of the fault reflected wave based on the injection pulse reference time and the set pulse propagation speed. Within the global search time range, candidate wave packets are searched from the envelope through local maximum detection and continuous threshold search. With the candidate wave packets as the center, an adaptive threshold is set based on the maximum envelope amplitude of the set local time window. Based on the adaptive threshold, the leading edge and first peak of the reflected wave are detected within the set local time window. Fault location module: It is used to obtain the fused location time by weighting and summing the reflection wave leading edge time and the first main peak time according to the set weights, and calculate the initial fault location based on the fused location time; it inputs the reflection wave leading edge time, the first main peak time and the initial fault location into the pre-trained ridge regression location correction model, and outputs the final fault location.

12. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method and system for inter-turn short-circuit localization of a condenser based on envelope fusion and ridge regression as described in any one of claims 1-10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the inter-turn short-circuit location method and system for a synchronous condenser based on envelope fusion and ridge regression as described in any one of claims 1-10.

Citation Information

Patent Citations

  • Generator rotor turn-to-turn short circuit detection device based on repetitive pulse RSO method

    CN215986428U