Real-time fault detecting and positioning method for high-precision relay protection system

Abstract

The invention belongs to the field of power system relay protection, and particularly relates to a real-time fault detection positioning method based on double-end synchronous high-frequency sampling and multi-dimensional feature fusion, which extracts transient traveling wave features through dynamic window function time-frequency analysis, intelligently identifies fault types in combination with a relevance vector machine, and improves fault detection accuracy. A three-dimensional space parameter correction factor is used for dynamically compensating an environment error, and finally millisecond-level accurate positioning of a complex fault is realized through a hybrid positioning equation and a three-level verification system, so that the safety protection capability of a power grid is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power system relay protection, more particularly, the present application relates to a real-time fault detection and positioning method of a high-precision relay protection system. BACKGROUND

[0002] The new power system construction promotes the evolution of the power grid towards high proportion of new energy access and multi-level AC / DC hybrid connection, and the amplitude, phase and frequency spectrum characteristics of fault current show strong uncertainty. According to the "2023 China Power Grid Safety Report", the proportion of complex faults (high resistance grounding, conversion fault, etc.) in the past five years has risen to 42%, and the contradiction between the speed and selectivity of traditional relay protection is intensifying, and it is urgent to break through the millisecond-level accurate positioning technology bottleneck to meet the stringent requirements of ultra-high voltage projects and urban cable networks for fault isolation accuracy ≤0.1% line length. The existing technology has the following deficiencies:

[0003] 1. Power frequency quantity protection is seriously affected by system disturbance

[0004] The impedance method based on power frequency electrical quantity fails in the new energy feeding scene: when the photovoltaic inverter provides short-circuit current, the phase shift of fault component causes the error amplification of impedance measurement, and the actual measurement of 220kV line positioning deviation is more than 3km (accounting for 10% of the total length); the misoperation rate is as high as 35% during system oscillation, which cannot meet the stability demand of modern power grid.

[0005] 2. There is a physical detection blind area in the traveling wave method

[0006] When the single-ended traveling wave positioning device is in fault within 5km from the measurement end, the initial traveling wave and the reflected wave superimpose to form waveform distortion, and the wave head recognition failure rate is >45%; the reflection path ambiguity in multi-branch line leads to false fault point false alarm, and the actual measurement of a certain 500kV substation shows that 30% of the fault positioning results need to be manually reviewed.

[0007] 3. Single dimension of feature extraction leads to misjudgment

[0008] The mainstream protection device relies on single-dimensional features such as current rate of change or harmonic content, and the confusion rate of lightning disturbance and real fault is as high as 28%; without fusion of time and frequency domain transient characteristics, the recognition delay of conversion fault (such as phase-to-phase short circuit to ground) is more than 20ms, causing protection overstep tripping accident.

[0009] 4. Lack of environmental parameters causes wave speed drift

[0010] The traditional traveling wave speed is fixed at 98% of the speed of light, ignoring the coupling effects of temperature, humidity and altitude: the wave speed decreases by 0.15% for every 10℃ increase in temperature, and decreases by 0.3% for every 1000m increase in altitude. The positioning deviation of a certain highland line (altitude 3800m) in summer and winter is 2.1km, which is equivalent to 1.4% of the total length of the line.

[0011] Therefore, a real-time fault detection and positioning method of a high-precision relay protection system is proposed to solve the above problems. SUMMARY

[0012] In order to overcome the above-mentioned defects of the prior art, embodiments of the present application provide a real-time fault detection and positioning method of a high-precision relay protection system to solve the problems raised in the above background.

[0013] To achieve the above object, the present application provides the following technical scheme: a real-time fault detection and positioning method of a high-precision relay protection system, comprising the following consecutive steps: collecting voltage and current original signals at both ends of a power transmission line in real time by a double-end synchronous measurement device at a sampling frequency of not less than 200 kHz, performing time-frequency joint analysis on the original signals by using an improved S transform algorithm to extract specific frequency band energy features of the fault transient traveling wave, and constructing a multi-dimensional fault feature recognition model based on a correlation vector to determine the fault type and severity level in real time according to the extracted feature vector, and when determining that a fault occurs, combining the double-end traveling wave arrival time difference and the distributed parameters of the line to calculate the fault distance by using a mixed positioning equation, wherein the equation takes one-half of the total length of the line as a reference value, and adds a dynamic correction result of one-half of the product of the traveling wave propagation speed and the time difference, wherein the correction process uses a three-dimensional space parameter correction factor based on the line topology structure, and finally outputs the fault positioning result to a protection execution unit through a hardware acceleration channel, and the delay of the whole process from signal collection to result output is strictly controlled within 3 milliseconds.

[0014] Preferably, the improved S transform uses a dynamically adjustable Gaussian window function, the window width factor of which is determined by joint optimization of the line characteristic frequency and the noise level, and specifically includes a basic window width coefficient, a frequency sensitive adjustment coefficient and a transient response gain coefficient, all of which are generated by training a deep learning model of a large amount of historical fault data, and the window function adjusts the window width in real time during signal processing to match the feature extraction requirements under different fault conditions.

[0015] Preferably, the input layer of the fault feature recognition model includes five-dimensional feature parameters: traveling wave high-frequency band energy proportion, transient signal wavelet entropy value, double-end waveform cross-correlation, zero-sequence current mutation rate and negative-sequence voltage distortion rate, and the output layer generates a nine-dimensional fault probability vector covering the whole category probability distribution of single-phase ground fault, two-phase inter-phase short-circuit fault, two-phase ground fault, three-phase short-circuit fault and conversion fault, and when the probability value of any fault type exceeds the confidence threshold of 0.97, the positioning mechanism is automatically triggered.

[0016] Preferably, the calculation of the three-dimensional space parameter correction factor fuses line electrical parameters, spatial structure characteristics and environmental variables, and is specifically expressed as a static structure compensation term superimposed on a dynamic impedance correction term, and then subtracting a distributed capacitance effect compensation term with an exponential decay relationship with fault distance, wherein the static structure compensation term takes into account the tower coordinate elevation data, the dynamic impedance correction term is associated with real-time data of line temperature and humidity sensors, and all compensation coefficients are dynamically generated through a line digital twin model.

[0017] Preferably, the double-end synchronous measurement adopts a quadruple time synchronization guarantee mechanism: the main channel is a GPS and Beidou dual-mode satellite time system, the standby channel is an IEEE 1588 precise time protocol network, the third channel is a local high-stability crystal oscillator time-keeping system, and the fourth channel is a bidirectional timestamp calibration channel based on optical fiber medium, ensuring that the time synchronization error is not more than 50 nanoseconds and the data interaction period is compressed to 500 nanoseconds.

[0018] Preferably, the traveling wave arrival time detection adopts an enhanced Teager-Kaiser energy operator, which detects energy mutation points in the signal sampling sequence through third-order difference operation combined with sliding window energy integral algorithm, and determines effective traveling wave arrival when the operator output values of five consecutive sampling points exceed four times the adaptive threshold of background noise, while phase consistency verification is adopted to eliminate reflected wave interference.

[0019] Preferably, a multi-stage fault verification process is added: in the first stage, the traditional impedance method calculation result is compared with the traveling wave positioning result, and when the deviation is not more than 1.5% of the full length of the line, the result is confirmed; in the second stage, adjacent monitoring node data collaborative verification is started, and the positioning consistency is verified through the waveform propagation time chain; in the third stage, the last three historical fault data are called for pattern matching verification.

[0020] Preferably, the collaborative verification adopts a space-time joint Bayesian inference algorithm, fuses at least three adjacent substation voltage and current transient data and topological connection relationships, constructs a fault location probability cloud map, determines the final positioning coordinates through the maximum probability density area, and the whole verification process is completed within 1.5 milliseconds.

[0021] Technical effects and advantages of the present application:

[0022] Compared with the prior art, the technical effects and advantages of the present application are concentrated in that through multi-source feature cooperative processing and dynamic environment adaptation mechanism, the three breakthroughs of precision, speed and reliability in the field of relay protection are realized: first, the double-end synchronous high-frequency sampling architecture is adopted, combined with the dynamic window function time-frequency analysis technology of improved S transform, the multi-dimensional fault features such as high-frequency energy, wavelet entropy value and waveform mutual correlation of transient traveling wave are accurately extracted in strong noise background; then the associated vector machine intelligent model is constructed, the sudden change characteristics of electrical quantity and the distortion characteristics of sequence component are fused, and the millisecond-level accurate identification of complex fault types is realized; on this basis, the three-dimensional space parameter correction factor is innovatively introduced, the tower elevation data, real-time temperature and humidity parameters and line distributed capacitance effect are comprehensively considered, and the dynamic compensation of traveling wave propagation speed deviation is realized; finally, through the three-level verification mechanism of impedance method checking, adjacent node time chain cooperation and historical data matching, the closed-loop reliability guarantee system is formed. The technical scheme forms the whole process optimization from signal acquisition, feature extraction, intelligent decision to verification execution, which significantly improves the positioning accuracy and anti-interference ability under complex working conditions while maintaining the inherent speed of relay protection. BRIEF DESCRIPTION OF DRAWINGS

[0023] Fig. 1 The system framework diagram of the present application.

[0024] Fig. 2 The working flow chart of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0026] Embodiment one:

[0027] As shown in the accompanying Figs. 1-2As shown, (1) a real-time fault detection and positioning method of a high-precision relay protection system, comprising the following consecutive steps: collecting the voltage and current original signals at both ends of the transmission line in real time by a double-end synchronous measurement device at a sampling frequency not less than 200 kHz, using an improved S transform algorithm to perform time-frequency joint analysis on the original signals to extract the specific frequency band energy features of the fault transient traveling wave, and based on the multi-dimensional fault feature recognition model constructed by the correlation vector, the fault type and severity level are determined in real time according to the extracted feature vector, when the fault occurs, the fault distance is calculated by a mixed positioning equation combining the double-end traveling wave arrival time difference and the distributed parameters of the line, the equation takes one half of the length of the line as the reference value, and superimposes the dynamic correction result of one half of the product of the traveling wave propagation speed and the time difference, wherein the correction process uses a three-dimensional space parameter correction factor based on the line topology structure, and finally the fault positioning result is output to the protection execution unit through a hardware acceleration channel, the delay of the whole process from signal collection to result output is strictly controlled within 3 ms, wherein the double-end synchronous measurement device installs a high-precision merging unit at both ends of the transmission line to synchronously collect three-phase voltage and current signals at a sampling frequency of 200 kHz; the original signal is input into an improved S transform processor to extract the transient traveling wave energy features in the 2-100 kHz frequency band through time-frequency joint analysis; the feature vector is input into a pre-trained correlation vector machine model in real time, which outputs the probability distribution containing the fault type and severity level; when the single-phase ground fault probability exceeds the threshold of 0.97, the positioning engine is immediately activated: the double-end traveling wave arrival time difference Δt is collected, combined with the line wave impedance parameters and the full length data, and substituted into the mixed positioning equation to calculate the fault distance (the calculation reference value is 1 / 2 of the length of the line, superimposed with 1 / 2 of the product of the traveling wave speed and Δt and multiplied by a three-dimensional correction factor); the calculation result is output to the protection execution unit through a hardware acceleration channel within 3 ms, triggering the circuit breaker tripping action.

[0028] (2) The improved S transform uses a dynamically adjustable Gaussian window function, the window width factor of which is determined by the joint optimization of the line characteristic frequency and the noise level, specifically including a basic window width coefficient, a frequency sensitive adjustment coefficient and a transient response gain coefficient, all of which are generated by a deep learning model trained on a large amount of historical fault data, and the window width of the window function is adjusted in real time during signal processing to match the feature extraction requirements under different fault conditions, wherein in the improved S transform processing, the parameters of the dynamic Gaussian window function are configured in real time by an embedded system: the basic window width coefficient α = 0.35 is determined by the line length, the frequency sensitive adjustment coefficient β = 1.2e -3|f|The transient response gain coefficient γ=0.8 is dynamically adjusted with signal frequency and is self-adaptively corrected by fault current rate of change; the window function parameter configuration module connects the historical fault database, and the coefficient mapping table is updated online once every 30 minutes; during the processing, when the window function convolution operation is performed on the A-phase current signal, the window width is automatically adjusted in the interval of 0.1-1.2 ms, ensuring that the signal-to-noise ratio of the high-frequency feature extraction under lightning interference is greater than 40 dB.

[0029] (3) The input layer of the fault feature recognition model contains five-dimensional feature parameters: high-frequency band energy proportion of traveling wave, wavelet entropy value of transient signal, double-end waveform cross-correlation, zero-sequence current rate of change, and negative-sequence voltage distortion rate. The output layer generates a nine-dimensional fault probability vector, covering the full-class probability distribution of single-phase ground fault, two-phase inter-phase short-circuit fault, two-phase ground fault, three-phase short-circuit fault, and conversion fault. When the probability value of any fault type exceeds the confidence threshold of 0.97, the positioning mechanism is automatically triggered. The input layer of the fault feature recognition model receives five-dimensional feature vectors: energy proportion above 10 kHz (FFT calculation), wavelet entropy value (db4 wavelet basis 5-layer decomposition), double-end waveform cross-correlation (time window 1 ms sliding calculation), zero-sequence current rate of change (dI0 / dt>5 kA / ms), and negative-sequence voltage distortion rate (THD>15%). The output layer generates a nine-dimensional probability vector [P ae ,P βe ,Pce,P αβ ,P β c,Pca,Pa β e,Pa β c,Ptrans] through the Softmax function. When P ae >0.97 is detected, an initiation instruction is immediately sent to the positioning engine.

[0030] (4) The calculation of the three-dimensional space parameter correction factor fuses line electrical parameters, spatial structure characteristics, and environmental variables. Specifically, it is expressed as a static structure compensation term superimposed with a dynamic impedance correction term, and then subtracted by a distributed capacitance effect compensation term with an exponential decay relationship with fault distance. The static structure compensation term takes into account the tower coordinate elevation data, and the dynamic impedance correction term is related to real-time data of line temperature and humidity sensors. All compensation coefficients are dynamically generated through the line digital twin model. The generation process of the three-dimensional space parameter correction factor is as follows: first, obtain the tower coordinate elevation data from the SCADA system to generate the static structure compensation term Ks=1+0.02×(h / 100)(h is the elevation difference); then read the temperature and humidity sensor data, and generate the dynamic correction term Kt=0.003×(T-25)-0.001×(RH-60) through the impedance-environment mapping table; finally, calculate the distributed capacitance compensation term Kc=0.15 e-0.8D / L(D is the fault distance); the correction factor Kcorr=Ks+Kt-Kc, the calculation result is transmitted to the positioning engine through the OPC UA protocol.

[0031] (5) The double-ended synchronous measurement adopts a quadruple time synchronization guarantee mechanism: the main channel is a GPS and Beidou dual-mode satellite time system, the standby channel is an IEEE 1588 precise time protocol network, the third channel is a local high-stability crystal oscillator time-keeping system, and the fourth channel is a bidirectional time stamp calibration channel based on an optical fiber medium, which ensures that the time synchronization error is not more than 50 nanoseconds and the data interaction period is compressed to 500 nanoseconds. The implementation of the quadruple time synchronization mechanism: the main clock adopts a Beidou III / GPS III dual-mode receiver, outputs a 1PPS pulse signal (precision ± 15 ns); the standby channel runs the IEEE 1588v2 protocol through the substation switch (jitter < 30 ns); the time-keeping system adopts an OCXO constant-temperature crystal oscillator (time-keeping error 1 μs / 24h); the optical fiber calibration channel transmits B code time stamps through a special core bidirectionally (calibration period 200 ms); the four-channel input time synchronization processor adopts a weighted average algorithm to output the final time stamp, ensuring that the sampling time deviation at both ends is ≤ 50 ns.

[0032] (6) The traveling wave arrival time detection adopts an enhanced Teager-Kaiser energy operator. The operator detects energy mutation points in the signal sampling sequence through third-order difference operation combined with sliding window energy integral algorithm, and determines that the effective traveling wave arrives when the operator output values of five consecutive sampling points exceed four times the adaptive threshold of background noise, while phase consistency verification is used to eliminate reflected wave interference. The traveling wave arrival time detection adopts an enhanced Teager-Kaiser operator: the current signal i a (n) is subjected to third-order difference operation Δ 3 i a (n) = i a (n+1)-3i a (n)+3i a (n-1)-i a (n-2); after 10 kHz high-pass filtering, the energy integral value E = Σ[Δ 3 i a (k)]2 is calculated for every 5 consecutive sampling points; when the E value is greater than 4 times the background noise energy (the background noise updates the standard deviation every 100 ms) and continuously satisfies 5 sampling points, it is marked as the traveling wave arrival time; at the same time, the phase relationship of the three-phase signals is compared, and if the C-phase lags behind the A-phase by 120°, the reflected wave interference is excluded.

[0033] (7) Add multi-stage fault verification process: the first stage compares the results of traditional impedance method calculation with the results of traveling wave positioning, and confirms the results when the deviation is not more than 1.5% of the total length of the line; the second stage starts the data collaborative verification of adjacent monitoring nodes, and checks the consistency of positioning through the wave propagation time chain; the third stage calls the historical fault data in the last three times for pattern matching verification, wherein, the three-stage fault verification process is implemented: the first stage starts within 0.5 ms after positioning is completed, calls the R-X impedance method to calculate the fault distance Limp, and verifies when |Limp-Lm| / Ltotal≤1.5%; the second stage obtains the fault recording data of the adjacent three substations through IEC 61850-90-1 protocol, and checks the consistency of the traveling wave arrival time chain (time deviation tolerance ±2μs); the third stage calls the historical fault record under similar environmental parameters within 72 hours before the fault, and performs waveform correlation coefficient matching (threshold value>0.85).

[0034] (8) The collaborative verification adopts a space-time joint Bayesian inference algorithm, fuses at least three adjacent substations of voltage and current transient data and topological connection relationship, constructs a fault location probability cloud map, determines the final positioning coordinates through the maximum probability density area, and the whole verification process is completed within 1.5 ms, wherein, the space-time joint Bayesian inference execution process is: a 5km×5km grid map (grid precision 10m) is constructed, the voltage mutation variables ΔU1, ΔU2, ΔU3 of the adjacent substations S1 / S2 / S3 and the topological impedance matrix are imported; the probability of each grid point is calculated through the likelihood function P(L|ΔU)=Πexp[-0.5×(ΔUmeas-ΔUtheory) 2 / σ 2 ]; the traveling wave positioning result is superimposed as the prior probability to generate a fault probability cloud map; the final positioning coordinates are taken from the geometric center of the maximum probability density area, and the whole calculation is completed within 1.5 ms and output through the GOOSE message.

[0035] Embodiment two: multi-source data joint modeling scene

[0036] Step 1: device installation and initialization configuration

[0037] Install high-precision merging units (model: MU-8000) at the substations at both ends of the transmission line, and connect the central processing host (CPU: Intel Xeon Gold 6348) in the main control room through optical fiber. The following initialization operations are performed when starting:

[0038] 1. Satellite clock synchronization: GPS / Beidou dual-mode receiver calibrates the clocks at both ends, with an error controlled within ±50 nanoseconds.

[0039] 2. Sampling rate setting: the three-phase voltage and current signal acquisition is fixed at 200,000 times per second (200kHz).

[0040] 3. Parameter preloading: download line parameters (total length 150 km, wave speed 294 km / ms) from database.

[0041] Example: Initialization completed at 14:05:25.123.456.789, A station clock shows 14:05:25.123.456.789, B station shows 14:05:25.123.456.839.

[0042] Step 2: Real-time acquisition of fault signals

[0043] The merging unit continuously acquires three-phase electrical quantities:

[0044] Voltage signals: A / B / C phases are respectively recorded as Ua, Ub, and Uc (units: kilovolts);

[0045] Current signals: A / B / C phases are respectively recorded as Ia, Ib, and Ic (units: kiloamperes);

[0046] When Ia mutation is detected (e.g., 1.2 kA→15.8 kA / 0.1 ms), it is immediately marked as "suspected fault" and the timestamp is stored.

[0047] Fault scenario: At 14:05:25.123.456.789, A station measures Ia = 15.8 kA (normal value 1.2 kA).

[0048] The central host performs improved S-transform on the fault phase current (Ia):

[0049] 1. Dynamic window length adjustment:

[0050] If the signal contains >50 kHz components, enable short window (0.1 ms window length);

[0051] If the signal contains <10 kHz components, enable long window (1.2 ms window length);

[0052] 2. Frequency band energy calculation:

[0053] Divide the 10-100 kHz frequency band, calculate the energy value of this frequency band ÷ total signal energy value × 100%

[0054] 3. Anti-interference processing:

[0055] When a pulse with duration <5 μs is detected, enable a 5-point sliding average filter

[0056] Processing result: Within 1 ms after the fault, the energy proportion of the 85 kHz frequency band is 72% (normal <5%)

[0057] Five features are extracted from the time-frequency analysis results:

[0058] 1. High-frequency energy ratio: energy above 10 kHz accounts for 82% (calculated value);

[0059] 2. Wavelet entropy value:

[0060] Decompose 5 layers with db4 wavelet basis,

[0061] Calculate entropy value = -Σ (coefficient square × log coefficient square) → measured 3.7.

[0062] 3. Waveform similarity:

[0063] Compare the Ia waveforms of A station and B station, calculate the correlation coefficient in a 1ms window (range 0-1, in this case 0.15).

[0064] 4. Zero sequence mutation rate:

[0065] Zero sequence current I0 = (Ia + Ib + Ic) / 3,

[0066] Mutation rate = |I0(t) - I0(t-0.1ms)| ÷ 0.1ms → measured 12.5kA / ms.

[0067] 5. Negative sequence distortion rate:

[0068] Negative sequence voltage U2 = (Ua + α 2 Ub + αUc) / 3 (α = 120° rotation factor),

[0069] Distortion rate = U2 harmonic content ÷ U2 fundamental content × 100% → measured 38%.

[0070] Step 5: Intelligent identification of fault type

[0071] Relevance vector machine model execution process:

[0072] 1. Input normalization: compress [82%, 3.7, 0.15, 12.5, 38%] to [0.82, 0.74, 0.15, 0.83, 0.76].

[0073] 2. Similarity calculation: calculate the Euclidean distance between the input vector and 500 historical samples (the smaller the distance, the more similar).

[0074] 3. Probability weighting:

[0075] Single-phase grounding probability = average similarity with 50 single-phase grounding samples × 0.95 weight

[0076] Similarly, calculate the probabilities of the other 8 types of faults.

[0077] 4. Decision output: Generate probability vector [single-phase ground: 0.992, two-phase short circuit: 0.003,...], trigger positioning when single-phase ground probability > 0.97.

[0078] Step 6: Traveling wave arrival time detection

[0079] Enhanced Teager energy detection process:

[0080] 1. Third-order difference calculation:

[0081] First-order difference D1 = Ia(n) - Ia(n-1);

[0082] Second-order difference D2 = D1(n) - D1(n-1);

[0083] Third-order difference D3 = D2(n) - D2(n-1);

[0084] 2. Energy integration: Calculate E = Σ[D3(k)]2 in a 0.05ms window (10 sample points).

[0085] 3. Threshold determination: When E > 4 × background noise value (updated every 100ms) and lasts for 5 sample points → mark the arrival time.

[0086] Result: Arrival time at A station 14:05:25.124.036.789, B station 14:05:25.124.095.089, time difference Δt = 58.3μs.

[0087] Step 7: Generation of three-dimensional correction factor

[0088] Real-time acquisition of parameters from database:

[0089] 1. Structural compensation term:

[0090] Elevation difference of towers on both sides of fault point h = 35m → compensation value = 1 + 0.002 × h = 1.07.

[0091] 2. Environmental correction term:

[0092] Temperature T = 32℃ → temperature correction = 0.0003 × (T-25),

[0093] Humidity RH = 75% → humidity correction = -0.0001 × (RH-60),

[0094] Total correction value = 0.0003 × 7 - 0.0001 × 15 = 0.0006.

[0095] 3. Capacitive attenuation term:

[0096] Estimated distance D≈84km→ Attenuation value = 0.12 x e-0.8 x D / L = 0.12 x 0.58 = 0.07.

[0097] Final correction factor K = 1.07 + 0.0006 - 0.07 = 1.0006.

[0098] Step 8: Fault distance calculation

[0099] Mixed positioning equation execution:

[0100] Reference distance = Total line length ÷ 2 = 150 ÷ 2 = 75 km,

[0101] Traveling wave component = (wave speed x time difference x correction factor) ÷ 2

[0102] = (294 km / ms x 0.0583 ms x 1.0006) ÷ 2

[0103] = 8.58 km

[0104] Final fault distance = 75 km + 8.58 km = 83.58 km (from A end).

[0105] Step 9: Three-level reliability verification

[0106] First level - impedance method verification:

[0107] Calculate fault loop impedance Z = U fault phase ÷ I fault phase,

[0108] Convert distance L_imp = (Z x total line length) ÷ unit impedance → 84.7 km,

[0109] Deviation verification: |83.58 - 84.7| ÷ 150 = 0.75% < 1.5% threshold → pass.

[0110] Second level - adjacent node cooperation:

[0111] Get adjacent station S1 data: Traveling wave arrival time 14:05:25.124.052.089.

[0112] Theoretical propagation time: S1 to fault point should be more than A end (distance difference ÷ wave speed) = (83.58 - 68) ÷ 294 = 15.2 μs.

[0113] Measured difference: (124.052.089 - 124.036.789) = 15.3 μs → deviation 0.1 μs → pass. Third level - historical data matching: Retrieve the B phase grounding fault waveform at 83 km last week, calculate the correlation coefficient of the current waveform:

[0114] Take 1 ms before the fault point + 5 ms after the fault point for a total of 1200 points,

[0115] Correlation coefficient = ∑ [current value x history value] ÷ sqrt(∑ current value2 x ∑ history value2) → get 0.91. Verification: 0.91 > 0.85 threshold → pass.

[0116] Step 10: Protection instruction execution

[0117] The central host generates a GOOSE instruction message (format as follows):

[0118] {Fault type: "A phase ground fault",

[0119] Positioning result: "83.58 km away from A end",

[0120] Confidence: 99.1%,

[0121] Action instruction: "trip A phase circuit breaker",

[0122] Time scale: "14:05:25.126.456.789"}

[0123] The message is transmitted to the circuit breaker through the optical fiber network, and the actuator completes the tripping operation within 8ms, and the total time consumption from fault occurrence to removal is ≤15ms.

[0124] Finally, a few points should be noted: first, in the description of the present application, it should be pointed out that unless otherwise specified and limited, the terms "installation", "connection", "connection" should be understood broadly, which can be mechanical connection or electrical connection, or the communication between two elements, or direct connection, "up", "down", "left", "right" and the like are only used to indicate the relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may change;

[0125] Secondly: the present application discloses the structure involved in the embodiment of the present application, other structures can refer to the usual design, in the case of no conflict, the same embodiment and different embodiments of the present application can be combined with each other;

[0126] Finally: the above only for the preferred embodiment of the present application, and does not limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the protection scope of the present application.

Claims

1. A real-time fault detection and location method for a high-precision relay protection system, characterized in that The method comprises the following consecutive steps: collecting the original voltage and current signals at both ends of the transmission line in real time at a sampling frequency of not less than 200 kHz through a dual-end synchronous measuring device, performing a time-frequency joint analysis on the original signals using an improved S-transform algorithm to extract the specific frequency band energy characteristics of the fault transient traveling wave, and using a multi-dimensional fault feature recognition model constructed based on an association vector machine to distinguish the fault type and severity level in real time according to the extracted feature vectors. When a fault is determined to have occurred, the hybrid positioning equation is used to calculate the fault distance in combination with the dual-end traveling wave arrival time difference and the line distributed parameters. The equation uses half of the total length of the line as a reference value and superimposes a dynamic correction result of half the product of the traveling wave propagation speed and the time difference. The correction process uses a three-dimensional spatial parameter correction factor based on the line topology structure. Finally, the fault positioning result is output to the protection execution unit through a hardware acceleration channel. The delay of the entire process from signal acquisition to result output is strictly controlled within 3 milliseconds.

2. A real-time fault detection and location method for a high-precision relay protection system according to claim 1, characterized in that The improved S transform adopts a dynamically adjustable Gaussian window function, whose window width factor is determined by the joint optimization of the line characteristic frequency and the noise level. Specifically, it includes the basic window width coefficient, the frequency sensitivity adjustment coefficient, and the transient response gain coefficient. All coefficients are generated through deep learning model training based on massive historical fault data. The window function adjusts the window width in real time during signal processing to match the feature extraction requirements under different fault conditions.

3. A real-time fault detection and location method for a high-precision relay protection system according to claim 1, characterized in that The input layer of the fault feature recognition model contains characteristic parameters of five dimensions: the proportion of energy in the high-frequency band of the traveling wave, the wavelet entropy value of the transient signal, the cross-correlation of the two-terminal waveform, the zero-sequence current mutation rate, and the negative-sequence voltage distortion rate. The output layer generates a nine-dimensional fault probability vector, covering the full category probability distribution of single-phase grounding faults, two-phase interphase short-circuit faults, two-phase grounding faults, three-phase short-circuit faults, and conversion faults. When the probability value of any fault type exceeds the confidence threshold of 0.97, the positioning mechanism is automatically triggered.

4. A real-time fault detection and location method for a high-precision relay protection system according to claim 1, characterized in that The calculation of the three-dimensional spatial parameter correction factor integrates the line electrical parameters, spatial structural characteristics and environmental variables, and is specifically expressed as a static structure compensation term superimposed on a dynamic impedance correction term, and then subtracting a distributed capacitance effect compensation term that has an exponential decay relationship with the fault distance. The static structure compensation term incorporates the tower coordinate elevation data, and the dynamic impedance correction term is associated with the real-time data of the line temperature and humidity sensors. All compensation coefficients are dynamically generated through the line digital twin model.

5. The real-time fault detection and location method for a high-precision relay protection system according to claim 1 is characterized in that The dual-end synchronous measurement adopts a four-fold time synchronization guarantee mechanism: the main channel is the GPS and Beidou dual-mode satellite timing system, the backup channel is the IEEE 1588 precision time protocol network, the third channel is a local high-stability crystal oscillator timing system, and the fourth channel is a bidirectional timestamp calibration channel based on optical fiber media, ensuring that the time synchronization error does not exceed 50 nanoseconds and the data exchange cycle is compressed to 500 nanoseconds.

6. A real-time fault detection and location method for a high-precision relay protection system according to claim 1, characterized in that The traveling wave arrival time detection adopts an enhanced Teager-Kaiser energy operator, which detects energy mutation points in the signal sampling sequence through third-order difference operation combined with a sliding window energy integration algorithm. When the operator output value of five consecutive sampling points exceeds four times the background noise adaptive threshold, it is determined that the traveling wave has arrived effectively. At the same time, phase consistency check is used to eliminate reflected wave interference.

7. The real-time fault detection and location method for a high-precision relay protection system according to claim 1 is characterized in that A multi-level fault verification process has been added: in the first stage, the calculation results of the traditional impedance method are compared with the traveling wave positioning results, and the result is confirmed when the deviation does not exceed 1.5% of the total length of the line; in the second stage, the collaborative verification of adjacent monitoring node data is initiated, and the positioning consistency is verified through the waveform propagation time chain; in the third stage, the data of the three most recent historical faults are called for pattern matching verification.

8. A real-time fault detection and location method for a high-precision relay protection system according to claim 7, characterized in that The collaborative verification adopts a spatiotemporal joint Bayesian inference algorithm, integrating the voltage and current transient data and topological connection relationships of at least three adjacent substations, constructing a probability cloud map of the fault location, and determining the final positioning coordinates through the maximum probability density area. The entire verification process is completed within 1.5 milliseconds.

Images