Fault Identification Method for Generator Excitation Rectifier Circuit Based on Information Fingerprint

Abstract

A method for identifying faults in the excitation rectifier circuit of a generator based on information fingerprints belongs to the technical field of power system fault diagnosis. This method uses multi-signal acquisition to calculate the fingerprint information of faults and perform fault detection. The set rough fingerprints are used to judge the damaged rectifier cabinet, and the refined fingerprints are used to identify the damaged thyristors in the rectifier cabinet. To improve the identification accuracy, the rough error rate GBER and the refined error rate ABER are defined as fingerprint similarity measurement indicators, and an energy correction coefficient is introduced for correction. By calculating the error rate, the system can accurately match the fault fingerprints with the fingerprints in the level-I and level-II fault identification libraries to determine the fault type. In addition, the present invention also improves the self-expansion ability of the system, which can continuously update the fault identification library during operation and collect more data by establishing a supplementary fault identification library to enhance the fault identification function. This method improves the speed and accuracy of fault detection in the excitation system, thereby enhancing the stability and security of the power system.

Description

Technical Field

[0001] The present invention relates to the technical field of power system fault diagnosis, and particularly to a method for identifying faults in a generator excitation rectification circuit based on information fingerprints. Background Art

[0002] With the transformation of the global energy structure, the large-scale construction of new energy power generation systems and the operation of large-capacity thermal power units have put forward higher requirements for the reliability, safety, and stability of modern power grids. As a key component to ensure the stable operation of the power grid, the synchronous generator excitation system can dynamically adjust the excitation current of the generator and maintain the stability of the output voltage when the power load changes rapidly. However, the timely diagnosis of excitation system faults is crucial for ensuring the safe operation of generator sets. Existing fault diagnosis methods rely on the comprehensive analysis of electrical signals at different positions of the excitation system. These methods have high requirements for the accuracy and diversity of signal information sources and are difficult to implement in practice. Since there are direct electrical connections between the excitation voltage signal, excitation current signal, trigger angle signal and the excitation transformer, excitation rectifier cabinet, and generator rotor winding, and rich state information is contained in their signal waveforms, it is of great significance to diagnose excitation system faults based on the excitation voltage signal.

[0003] The information fingerprint technology has been effectively applied in the field of audio and video recognition. By analyzing the unique features of signals and constructing hash functions, signal classification and rapid search can be achieved. At present, some studies have attempted to apply the information fingerprint technology to the field of fault identification in electrical systems, such as the identification of fault locations in DC power grids. However, these methods mainly rely on the time-frequency information of the original signals, have poor robustness to signal time translation, are highly dependent on the starting time of signal acquisition, and are difficult to make consistent fingerprint mappings for signals without obvious trigger marks. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a method for identifying faults in a generator excitation rectification circuit based on information fingerprints. This method uses the autocorrelation function of the signal over a period of time to replace the original signal to extract time-frequency features and perform fingerprint mapping, improving the robustness of fingerprint coding to signal time translation, and making the identification scheme of matching fingerprint coding with the fault fingerprint library applicable to periodic stable analog signals, thereby effectively realizing the rapid and accurate identification of different fault types in the three-phase bridge rectifier circuit of the excitation system.

[0005] The present invention adopts the following technical solutions:

[0006] A method for identifying faults in a generator excitation rectification circuit based on information fingerprints, wherein the generator excitation rectification circuit includes a group of parallel three-phase bridge rectifier cabinets, and each three-phase bridge rectifier cabinet includes a group of thyristors connected in a bridge; the specific steps of the fault identification method are as follows:

[0007] Step 1: Set the fault identification of the generator excitation rectifier circuit to include the fault identification of the first-level rectifier cabinet and the fault identification of the second-level thyristor. The fault identification of the first-level rectifier cabinet is used to determine the damaged rectifier cabinet, and the fault identification of the second-level thyristor is used to further determine the damaged thyristor in the rectifier cabinet;

[0008] Step 2: Simulate each first-level rectifier cabinet fault, and map the time-domain information of the directly collected quantities within 1 second after each first-level rectifier cabinet fault to form corresponding rough fingerprints of the first-level rectifier cabinet faults. 1 < t 1 < 5, and establish a first-level fault identification library. Among them, the directly collected quantities include the firing angle and the excitation voltage; t 1 < 5, and establish a first-level fault identification library. Among them, the directly collected quantities include the firing angle and the excitation voltage;

[0009] Step 3: Simulate each second-level thyristor fault, calculate the time-domain information of the non-collected quantities according to the time-domain information of the directly collected quantities within 2 seconds after each second-level thyristor fault. 10 < t 2 < 30, map the calculated time-domain information of the non-collected quantities to form corresponding fine fingerprints of the second-level rectifier cabinet faults respectively, and establish a second-level fault identification library. Among them, the non-collected quantities include the excitation power, the AC-side excitation current, and the average energy; t 2 < 30, map the calculated time-domain information of the non-collected quantities to form corresponding fine fingerprints of the second-level rectifier cabinet faults respectively, and establish a second-level fault identification library. Among them, the non-collected quantities include the excitation power, the AC-side excitation current, and the average energy;

[0010] Step 4: Transplant the first-level fault identification library and the second-level fault identification library to the on-site measurement and control platform of the generator excitation system. After the generator excitation rectifier circuit fails, collect t the time-domain information of the directly collected quantities within 1 second, generate a fingerprint code through the mapping algorithm and compare it with the rough fingerprints of the first-level fault identification library. By defining the rough error rate GBER and calculating the rough error rate GBER to determine the rough fingerprint with the highest similarity, and send the first-level rectifier cabinet fault corresponding to the rough fingerprint with the highest similarity as the identification result for the short-time relay protection of the generator excitation system;

[0011] Step 5: After the generator excitation rectifier circuit fails, collect t the time-frequency information of the directly collected quantities within 2 seconds, calculate and generate the time-domain information of the non-collected quantities, generate a fingerprint code through the mapping algorithm and compare it with the fine fingerprints of the second-level fault identification library. By defining the fine error rate ABER and calculating the fine error rate ABER to determine the fine fingerprint with the highest similarity, and send the second-level thyristor fault corresponding to the fine fingerprint with the highest similarity as the fault identification result for the operation and maintenance of the excitation rectifier circuit and the replacement of thyristor devices after the generator stops;

[0012] Step 6: After the I-level fault recognition library and the II-level fault recognition library have run for a set period in the generator excitation system measurement and control platform, according to the determination logic of the recognition library, dynamically update the corresponding rough fingerprints and fine fingerprints in the fault recognition library; for the generator excitation rectifier circuit faults that cannot be recognized by the I-level fault recognition library and the II-level fault recognition library, establish corresponding III-level supplementary fault recognition libraries and IV-level supplementary fault recognition libraries respectively.

[0013] Further: In the said Step 2, the formation steps of the rough fingerprint are as follows:

[0014] Step 2.1) Based on the time-domain information of the directly collected quantity, calculate its autocorrelation function and extract the first k cycles, and the expression is as follows:

[0015] ;

[0016] In the formula, is the time-domain signal of the directly collected quantity, is the length of the time-domain signal sequence, T is the signal acquisition step size;

[0017] Step 2.2) Perform wavelet transform reconstruction on the extracted autocorrelation function, divide the wavelet transform reconstruction result into signal frames, and calculate the signal frame energy and the zero-crossing number , and the expressions are:

[0018] ;

[0019] ;

[0020] In the formula, is the energy of the th layer and the th signal frame, is the zero-crossing number of the th layer and the th signal frame, represents the reconstructed signal of the th layer; represents the number of sampling points per frame; represents the sign function;

[0021] Step 2.3) Map and to obtain the corresponding fingerprint information as follows:

[0022] ;

[0023] ;

[0024] In the formula, represents the signal frame energy of the th layer and the th frame, is the corresponding fingerprint bit value, is the th layer and the th frame's zero-crossing number, is the corresponding fingerprint bit value.

[0025] Furthermore: In step 3, the steps for forming the refined fingerprint are as follows:

[0026] Step 3.1) Based on the time-domain information of the directly collected quantity, calculate the non-collected quantity excitation power, AC-side excitation current, and average energy, which are defined as follows:

[0027] Step 3.1.1) Excitation power The formula is as follows:

[0028] ;

[0029] In the formula, is the DC-side excitation voltage, is the excitation current;

[0030] Step 3.1.2) Assume that M the excitation currents in the rectifier cabinets are equal, then the average excitation current

[0031] ;

[0032] ;

[0033] In the formula, is the firing angle, is the rectifier energy balance coefficient, is the current sharing coefficient, is the thyristor switching loss, is the thyristor heating loss, is the loss of the rectifier system cooling system, is the control system loss; if the rectifier cabinet does not have the intelligent current sharing function, then the control system loss is ;

[0034] Step 3.1.3) Average energy The formula is as follows:

[0035] ;

[0036] ;

[0037] In the formula, N is the length of the time-domain signal sequence, T is the signal acquisition step size, ceiling represents the ceiling function;

[0038] Step 3.2) Based on the calculated time-domain information of the non-acquired quantity, calculate its autocorrelation function and extract the first k cycles, and the expression is as follows:

[0039] ;

[0040] In the formula, is the time-domain signal of the non-acquired quantity, is the length of the time-domain signal sequence, T is the signal acquisition step size;

[0041] Step 3.3) Perform wavelet transform reconstruction on the extracted autocorrelation function, divide the wavelet transform reconstruction result into signal frames, and calculate the signal frame energy and the zero-crossing number , and the expression is:

[0042] ;

[0043] ;

[0044] In the formula, is the energy of the th signal frame in the th layer, is the zero-crossing number of the signal frame in the th layer, represents the reconstructed signal of the th layer; represents the number of sampling points per frame;

[0045] Step 3.4) Map and to obtain the corresponding fingerprint information as follows:

[0046] ;

[0047] ;

[0048] In the formula, represents the signal frame energy of the th frame in the th layer, is the corresponding fingerprint bit value; is the The number of zero crossings of the frame, is the corresponding fingerprint bit value.

[0049] Furthermore: The process of step 4 is as follows:

[0050] Define the rough error rate GBER to measure the similarity between rough fingerprints, and introduce an energy correction coefficient , and the expression is:

[0051] ;

[0052] ;

[0053] In the formula, is the number of different bit values in the layer of rough fingerprints, is the total number of fingerprints in each layer, is the total number of layers of rough fingerprints, is the weight of each layer of fingerprints; , are respectively the average energy of the time-domain waveform of the directly acquired quantity measured during actual operation, and the average energy of the time-domain waveform of the directly acquired quantity stored in the first-level fault recognition library. The expression is:

[0054] ;

[0055] ;

[0056] In the formula, , are respectively the time-domain signals of the directly acquired quantity measured during actual operation and the time-domain signals of the directly acquired quantity stored in the first-level fault recognition library, and are respectively the signal sequence lengths of the two;

[0057] Compare the rough fingerprints generated by the actual operation fault with the rough fingerprints of each fault in the first-level fault recognition library in sequence to obtain the minimum rough error rate GBER min , and the minimum rough error rate GBER min The corresponding rough fingerprint is the rough fingerprint with the highest similarity to the actual rectifier cabinet fault in the first-level fault recognition library.

[0058] Furthermore: The process of step 5 is as follows:

[0059] Define the fine error rate ABER to measure the similarity between fine fingerprints, and introduce an energy correction coefficient , and the expression is:

[0060] ;

[0061] ;

[0062] Wherein, is the number of different bit values in the layer of refined fingerprints, is the total number of fingerprints in each layer, is the total number of layers of refined fingerprints, is the weight of each layer of fingerprints; and are respectively the average energy of the non - acquisition time - domain waveform measured during actual operation and the average energy of the non - acquisition time - domain waveform stored in the level - II fault recognition library. The expression is:

[0063] ;

[0064] ;

[0065] Wherein, and are respectively the time - domain signals of the non - acquisition quantity measured during actual operation and the time - domain signals of the non - acquisition quantity stored in the level - II fault recognition library, and are respectively the signal sequence lengths of the two;

[0066] Compare the refined fingerprints generated by the actual operation fault with the refined fingerprints of each fault in the level - II fault recognition library in sequence to obtain the minimum refined error rate ABER min The minimum refined error rate ABER min The corresponding refined fingerprint is the refined fingerprint with the highest similarity to the actual thyristor fault in the level - II fault recognition library.

[0067] Furthermore: In step 6, according to the numerical values of the coarse / fine error rates matched by the actual fault, update and supplement the fault recognition library. The specific process is as follows:

[0068] Step 6.1) When the matched coarse / fine error rate is less than the set threshold , the dynamic update steps of the fingerprint coding in the level - I / level - II fault recognition library are as follows:

[0069] Step 6.1.1) Compare the coarse / refined fingerprints in the level - I / level - II fault recognition library with the coarse / refined fingerprints generated by the actual operation fault, and identify the positions where the bit values are different in the corresponding fingerprints;

[0070] Step 6.1.2) For the different positions, according to the set probability Replace the rough / fine elements in the level-I / level-II fault recognition library with the corresponding rough / fine fingerprint elements in the actual fault fingerprint;

[0071] Step 6.2) When the matched rough / fine error rate is greater than or equal to the threshold , the steps for establishing the level-III / level-IV supplementary fault recognition library are as follows:

[0072] Step 6.2.1) Re-collect the time-domain information of the directly collected quantities during the fault process twice. According to the above-mentioned step 4 / step 5), calculate the rough / fine error rates of the two collected signals respectively, and obtain the corresponding fault recognition results;

[0073] Step 6.2.2) If both of the two groups of rough / fine error rates calculated in step 6.2.1) are less than the set threshold , and the obtained fault recognition results are consistent, then send the result as the level-I rectifier cabinet / level-II thyristor fault recognition result this time; otherwise, add all the rough / fine fingerprints of this fault and the time-domain waveforms of the corresponding directly collected quantities to the level-III / level-IV supplementary fault recognition library for subsequent recognition and research, and output a special fault warning.

[0074] A method for identifying faults in a generator excitation rectification circuit based on information fingerprints provided by the present invention has the following beneficial effects:

[0075] (1) The fingerprint data generated by the method of the present invention is small in quantity. For fault situations under different parameters, it can achieve a large amount of storage and fast matching, and has a strong ability to distinguish different asymmetric fault states under different trigger angles;

[0076] (2) The method of the present invention extracts the time-frequency information of the autocorrelation function of the original signal based on wavelet transform reconstruction, realizing that the excitation electric fingerprint has strong robustness to noise influence and signal time translation, and reducing the error rate between the signal fingerprint and the original fingerprint;

[0077] (3) The multi-fault library parallel and update scheme in the method of the present invention enables the simulation fault fingerprint library to be updated in real time according to the actual operation situation, and output warnings for doubtful data in a timely manner, making the recognition method have strong reliability. Description of the Drawings

[0078] Figure 1 is the schematic diagram of the excitation system of the present invention;

[0079] Figure 2 is the schematic diagram of the information fingerprint structure of the present invention;

[0080] Figure 3 is the trigger angle of the present invention , Sampling signal waveforms of the normal working part of the excitation system, 3a is the waveform of the excitation voltage signal, 3b is the waveform of the excitation current signal, and 3c is the waveform of the excitation active power signal;

[0081] Figure 4 For the firing angle of the present invention , Only the sampling signal waveforms before and after the short circuit of thyristor VT1, 4a is the waveform of the excitation voltage signal, 4b is the waveform of the excitation current signal, and 4c is the waveform of the excitation active power signal;

[0082] Figure 5 For the firing angle of the present invention , Only the time-frequency diagram of the excitation voltage before and after the short circuit of thyristor VT1, 5a is the waveform of the excitation voltage signal, 5b is the waveform of the excitation current signal, and 5c is the waveform of the excitation active power signal;

[0083] Figure 6 In the present invention, after the sampling signal is translated in time, the relationship between the bit error rate between the fingerprints generated by the original signal and the autocorrelation function of the original signal and the original fingerprint;

[0084] Figure 7 In the present invention, after different degrees of white noise are introduced into the original signal, the bit error rate of the fingerprint between it and the original signal;

[0085] Figure 8 In the present invention, under the fault type of only the short circuit of thyristor VT1, different firing angles The bit error rate between the voltage fingerprint at and the voltage fingerprint at the firing angle

[0086] Figure 9 In the present invention, the bit error rate between fingerprints under 12 selected different fault types. Detailed implementation manners

[0087] In order to describe the present invention more specifically, the technical solutions of the present invention will be described in detail below with reference to the drawings and specific implementation manners.

[0088] Embodiment: A method for identifying faults in a generator excitation rectification circuit based on information fingerprints. The generator excitation rectification circuit is composed of at least 3 three-phase bridge rectifier cabinets connected in parallel, and each rectifier cabinet is bridge-connected by 6 thyristor devices; the specific process of the fault identification method is as follows:

[0089] Step 1, Set that the fault identification of the generator excitation rectification circuit includes the first-level rectifier cabinet fault identification and the second-level thyristor fault identification. Among them, the first-level rectifier cabinet fault identification is used to determine the damaged rectifier cabinet, and the second-level thyristor fault identification is used to further determine the damaged thyristor in the rectifier cabinet.

[0090] Step 2: Simulate the faults of each level-I rectifier cabinet, map the time-domain information of the directly collected quantities within several seconds after each fault respectively to form a rough fingerprint, and thus establish a level-I fault recognition library, where the directly collected quantities include the firing angle and the excitation voltage.

[0091] Step 3: Simulate the faults of each level-II thyristor, calculate the time-domain information of the non-collected quantities from the time-domain information of the directly collected quantities within dozens of seconds after each fault, map them respectively to form a refined fingerprint, and thus establish a level-II fault recognition library, where the non-collected quantities include the excitation power, the AC-side excitation current, and the average energy.

[0092] Step 4: Transplant the level-I and level-II fault recognition libraries to the on-site measurement and control platform of the generator excitation system. After a fault occurs in the generator excitation rectifier circuit, collect the time-domain information of the directly collected quantities within several seconds, generate a fingerprint code through the mapping algorithm and compare it with the rough fingerprint of the level-I fault recognition library, and determine the rough fingerprint with the highest similarity by defining and calculating the rough error rate GBER Send the level-I rectifier cabinet fault corresponding to the rough fingerprint as the recognition result for the short-time relay protection of the generator excitation system.

[0093] Step 5: After a fault occurs in the generator excitation rectifier circuit, collect the time-frequency information of the directly collected quantities within dozens of seconds, calculate and generate the time-domain information of the non-collected quantities, generate a fingerprint code through the mapping algorithm and compare it with the refined fingerprint of the level-II fault recognition library, and determine the refined fingerprint with the highest similarity by defining and calculating the refined error rate ABER Send the level-II thyristor fault corresponding to the refined fingerprint as the fault recognition result for the operation and maintenance of the excitation rectifier circuit and the replacement of thyristor devices after the generator stops.

[0094] Step 6: After the level-I and level-II fault recognition libraries run for a certain period in the generator excitation system measurement and control platform, dynamically update the rough fingerprint and the refined fingerprint of the fault recognition library according to the determination logic of the recognition library; at the same time, establish a level-III and a level-IV supplementary fault recognition library respectively for the faults of the generator excitation rectifier circuit that cannot be recognized by the level-I and level-II fault recognition libraries, for outputting special fault warnings.

[0095] The schematic diagram of the excitation system principle is as Figure 1 , the generator terminal voltage is input into the three-phase bridge rectifier circuit through the excitation transformer and the synchronous transformer, and after being rectified by the thyristors VT1-VT6, a stable DC current is obtained, and finally through the generator rotor winding, the generator set is started and operated. The automatic voltage regulator (AVR) provides corresponding 6-pulse signals according to the firing angle to control the enabling of the thyristors VT1-VT6. Figure 3 For When the thyristor is working normally, the excitation voltage, excitation current, and excitation active power curves. When any thyristor has a short - circuit or open - circuit fault, the excitation voltage signal will change and, after reaching stability, present a curve with different characteristics with a period of 0.02 s. The fault of a certain thyristor will destroy the symmetry of the three - phase AC source, and the proportion of high - frequency components will increase (see Figure 4 , Figure 5 ), so the characteristics of different frequency bands of the signal can be obtained through wavelet decomposition and reconstruction, and the fingerprint of the voltage signal can be constructed based on this.

[0096] In step 2, the formation steps of the rough fingerprint are as follows:

[0097] Step 2.1) Based on the time - domain information of the directly collected quantity, calculate its autocorrelation function and extract the first k cycles, and the expression is as follows:

[0098] ;

[0099] In the formula, is the time - domain signal of the directly collected quantity, is the length of the time - domain signal sequence, T is the signal acquisition step size;

[0100] Step 2.2) Perform wavelet transform and reconstruction on the intercepted autocorrelation function, divide the wavelet transform and reconstruction result into signal frames, and calculate the signal frame energy and the zero - crossing number , and the expressions are:

[0101] ;

[0102] ;

[0103] In the formula, is the energy of the th layer and the th signal frame, is the zero - crossing number of the th layer and the th signal frame, represents the reconstructed signal of the th layer; represents the sign function.

[0104] Step 2.3) Map and to obtain the corresponding fingerprint information as follows:

[0105] ;

[0106] ;

[0107] In the formula, represents the signal frame energy of the th layer and the th frame, is the corresponding fingerprint bit value; is the zero-crossing number of the th layer and the th frame, is the corresponding fingerprint bit value.

[0108] In step 3, the steps for forming the refined fingerprint are as follows:

[0109] Step 3.1) Based on the time-domain information of the directly collected quantity, calculate the excitation power of the non-collected quantity, the excitation current on the AC side, and the average energy, and their definitions are as follows:

[0110] Step 3.1.1) Excitation power The formula for

[0111] is as follows:

[0112] In the formula, is the excitation voltage on the DC side, is the excitation current;

[0113] Step 3.1.2) Assume that M the excitation currents in the rectifier cabinets are equal, then the average excitation current

[0114] on the AC side has the following formula:

[0115] ;

[0116] In the formula, is the firing angle, is the rectifier energy balance coefficient, is the current sharing coefficient, is the thyristor switching loss, is the thyristor heating loss, is the loss of the rectifier system cooling system, is the loss of the control system; if the rectifier cabinet does not have the intelligent current sharing function, the loss of the control system is ;

[0117] Step 3.1.3) Average energy The formula for

[0118] is as follows:

[0119] ;

[0120] In the formula, N is the length of the time-domain signal sequence, T is the signal acquisition step size, ceiling represents the ceiling function.

[0121] Step 3.2) Based on the calculated time-domain information of the non-acquired quantity, calculate its autocorrelation function and extract the first k cycles, and the expression is as follows:

[0122] ;

[0123] In the formula, is the time-domain signal of the non-acquired quantity, is the length of the time-domain signal sequence, T is the signal acquisition step size.

[0124] Step 3.3) Perform wavelet transform reconstruction on the intercepted autocorrelation function, divide the wavelet transform reconstruction result into signal frames, and calculate the signal frame energy and the zero-crossing number , and the expressions are:

[0125] ;

[0126] ;

[0127] In the formula, is the energy of the th layer and the th signal frame, is the zero-crossing number of the th layer and the th signal frame, represents the reconstructed signal of the th layer; represents the number of sampling points per frame; represents the sign function.

[0128] Step 3.4) Map and to obtain the corresponding fingerprint information as follows:

[0129] ;

[0130] ;

[0131] In the formula, represents the signal frame energy of the th layer and the th frame, is the corresponding fingerprint bit value; is the The number of zero-crossings in the frame, is the corresponding fingerprint bit value.

[0132] In step 4, the gross error rate GBER is defined to measure the similarity between rough fingerprints, and an energy correction coefficient is introduced. The expression is:

[0133] ;

[0134] ;

[0135] In the formula, is the number of different bit values in the rough fingerprint of the layer, is the total number of fingerprints per layer, is the total number of rough fingerprint layers, , are the average energies of the time-domain waveforms of the directly collected quantities measured during actual operation and stored in the first-level fault identification library respectively. The expression is:

[0136] ;

[0137] ;

[0138] In the formula, and are the time-domain signals of the directly collected quantities measured during actual operation and stored in the first-level fault identification library respectively, and are the signal sequence lengths of the two respectively.

[0139] The rough fingerprints generated by the actual operation fault are compared with the rough fingerprints of each fault in the first-level fault identification library in turn, and the minimum gross error rate GBER min can be obtained. The corresponding one is the rough fingerprint with the highest similarity to the actual rectifier cabinet fault in the first-level fault identification library.

[0140] In step 5, the fine error rate ABER is defined to measure the similarity between fine fingerprints, and an energy correction coefficient is introduced. The expression is:

[0141] ;

[0142] ;

[0143] In the formula, is the The number of different bit values in the refined fingerprint of each layer, is the total number of fingerprints in each layer, is the total number of refined fingerprint layers, is the weight of each fingerprint layer; 、 are respectively the average energy of the non-acquired quantity time-domain waveforms measured during actual operation and stored in the Class-II fault recognition library. The expression is:

[0144] ;

[0145] ;

[0146] In the formula, and are respectively the time-domain signals of the non-acquired quantity measured during actual operation and stored in the Class-II fault recognition library. and are respectively the signal sequence lengths of the two.

[0147] Compare the refined fingerprints generated by the actual operation fault with the refined fingerprints of each fault in the Class-II fault recognition library in sequence, and the minimum refined bit error rate ABER min can be obtained. The corresponding one is the refined fingerprint with the highest similarity to the actual thyristor fault in the Class-II fault recognition library.

[0148] In step 2-5: Simulate the sampling signals of the circuit under different error states at different trigger angles . Store the fault state, direct acquisition quantity time-domain signal, rough fingerprint, and signal average energy together to establish a Class-I fault recognition library; store the fault state, calculated non-acquired quantity time-domain signal, refined fingerprint, and signal average energy together to establish a Class-II fault recognition library. After a fault occurs, collect the fault sampling signal waveforms with a time length of 1 s and 10 s, respectively make rough and refined fingerprint information, and match them with the fingerprints in the Class-I and Class-II fault recognition libraries and calculate the rough bit error rate GBER and the refined bit error rate ABER . Use the fault type corresponding to the minimum rough bit error rate as the Class-I rectifier cabinet fault recognition result for the short-term relay protection of the generator excitation system; use the fault type corresponding to the minimum refined bit error rate as the Class-II thyristor fault recognition result for the operation and maintenance of the excitation rectification circuit and the replacement of thyristor devices after the generator stops.

[0149] In step 6, update and supplement the fault recognition library according to the magnitude of the rough (refined) bit error rate value matched by the actual fault. The specific process is as follows:

[0150] 6.1) When the matched rough (refined) bit error rate is less than the set threshold When there is a Class-I (Class-II) fault, the dynamic update steps of the fingerprint coding in the fault recognition library are as follows:

[0151] Step 6.1.1) Compare the original fingerprint in the Class-I (Class-II) fault recognition library with the rough (fine) fingerprint generated by the actual operating fault, and identify the positions where the bit values in the corresponding fingerprints are different;

[0152] Step 6.1.2) For each different position, according to the set probability Replace the elements of the original fingerprint in the Class-I (Class-II) fault recognition library with the corresponding elements in the actual fault fingerprint.

[0153] Step 6.2) When the matching rough (fine) error rate is greater than or equal to the threshold the establishment steps of the Class-III (Class-IV) supplementary fault recognition library are as follows:

[0154] Step 6.2.1) Re-collect the time-domain information of the directly collected quantities during the fault process twice. According to Step 4 (Step 5), calculate the rough (fine) error rates of the two collected signals respectively, and obtain the corresponding fault recognition results;

[0155] Step 6.2.2) If the two rough (fine) error rates calculated in Step 6.2.1) are both less than the set threshold and the obtained fault recognition results are consistent, then send the result as the fault recognition result of this Class-I rectifier cabinet (Class-II thyristor) fault; otherwise, add all the rough (fine) fingerprints of this fault and the time-domain waveforms of the corresponding directly collected quantities to the Class-III (Class-IV) supplementary fault recognition library for subsequent recognition and research, and output a special fault warning.

[0156] The process of updating and supplementing the fault recognition library is shown in Figure 4 .

[0157] To achieve accurate recognition of various fault types, the fingerprint mapping method should have good robustness and anti-collision ability. That is, for the same type of fault, when affected by noise or time shift, the error rate between its fingerprints should be small; while for fingerprints of different fault types, there should be a large error rate. For this reason, in this embodiment, a simulation model of the generator excitation system is built in MATLAB / Simulink software to verify the proposed method through simulation. Figure 1 shows the schematic diagram of the excitation system, and Table 1 shows the simulation parameters of the generator excitation system. Figure 2 shows the information fingerprint structure, where Layers 1-5 are generated by the signal frame energy of a6 - a 10 and a 10 and Layers 6-8 are generated by the zero-crossing number of the signal frame of d7 - d9.

[0158] Table 1: Simulation Parameters of Generator Excitation System

[0159]

[0160] To verify the strong robustness result of the method of generating fingerprints by the autocorrelation function to the time translation of the original signal, and considering that the rough fingerprints and the refined fingerprints have almost the same mapping formation and the fault matching process, therefore, in this embodiment, the formation and fault matching of the refined fingerprints are taken as examples for verification and description.

[0161] Figure 6 It shows the relationship between the bit error rate between the fingerprints generated by the original signal and the autocorrelation function of the original signal and the original fingerprint after the sampled signal (taking the active power of excitation as an example) is translated in time. From Figure 6 it can be seen that after changing to the autocorrelation function, the influence of the time translation is greatly reduced and controlled below 5%. Therefore, the fingerprint mapping method in this embodiment uses the time-frequency domain information of the signal autocorrelation function instead of the original signal, which greatly improves the influence of the time translation of the periodic voltage signal on the fingerprint.

[0162] Figure 7 It shows the bit error rate between the fingerprints of the original signal after introducing different degrees of white noise and the original signal. From Figure 7 it can be seen that when the signal-to-noise ratio is greater than 10 dB, the bit error rate is always less than 5%, while for the voltage fingerprints under different fault types, the bit error rate between the fingerprints is 20% - 80%. Therefore, the fingerprint mapping method in this embodiment can maintain good stability when the signal-to-noise ratio is above 10 dB and has strong robustness to noise influence and signal time translation.

[0163] Figure 8 It shows the bit error rate between the fingerprint information at different trigger angles and the fingerprint information at the trigger angle under the fault type of only VT1 thyristor short circuit. From it can be seen that when the bit error rate is less than 5%, Figure 8 it is controlled within . Figure 9 It shows the bit error rate between the fingerprints under 12 different fault types of the selected excitation system, all of which are above 20%. The corresponding relationship between the fault number and the fault type is shown in Table 2.

[0164] Table 2: Corresponding Numbers of Fault States of the Rectifier Circuit of the Excitation System

[0165]

[0166] ​In this embodiment, a 6-bit ternary array is used to represent the circuit fault state, where 1 represents an open circuit, -1 represents a short circuit, and 0 represents normal operation. For example, in the case where both VT1 and VT2 are open circuits, the fault state is represented as (1, 1, 0, 0, 0, 0).

[0167] Therefore, this embodiment effectively distinguishes the sampled signal waveforms at different trigger angles under the same fault condition, and the fingerprint mapping method of this embodiment has good anti-collision performance and can effectively distinguish different fault types.

[0168] Comprehensively Figures 6 - 9 , the method of this embodiment can effectively identify the sampled signal waveforms at different trigger angles and fault states, and has strong robustness against interference noise with a signal-to-noise ratio of more than 10 dB relative to the original signal and the influence of time translation. In this embodiment, due to the symmetry between the phases of the input three-phase alternating current, three-phase symmetric faults (such as only VT1 short circuit and only VT3 short circuit) have consistent steady-state voltage waveforms, and these two types of faults are classified as the same fault type.

[0169] The above description of the embodiments is to enable those of ordinary skill in the art to understand and apply the present invention. Obviously, those skilled in the art can easily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without creative labor. Therefore, the present invention is not limited to the above embodiments, and the improvements and modifications made by those skilled in the art based on the disclosure of the present invention should be within the protection scope of the present invention.

Claims

1. A method for identifying faults in a generator excitation rectifier circuit based on information fingerprint, characterized in that: The generator excitation rectifier circuit includes a group of parallel three-phase bridge rectifier cabinets, and each three-phase bridge rectifier cabinet includes a group of bridge-connected thyristors; the specific steps of the fault identification method are as follows: Step 1, setting the generator excitation rectifier circuit fault identification to include level I rectifier cabinet fault identification and level II thyristor fault identification; Step 2: simulate the faults of each level I rectifier cabinet, form a corresponding rough fingerprint of the level I rectifier cabinet fault according to the information collected after the fault of each level I rectifier cabinet, and establish a level I fault identification library; Step 3: Simulate the faults of each level II thyristor, form a corresponding level II rectifier fault fingerprint according to the information collected after the fault of each level II thyristor, and establish a level II fault identification library; Step 4: Generate corresponding fingerprint codes according to the information collected by the actual generator excitation rectifier fault system, compare them with the coarse fingerprint and fine fingerprint in the corresponding fault identification library, and define the coarse bit error rate GBER and the fine bit error rate ABER respectively. Determine the coarse fingerprint and fine fingerprint with the highest similarity through the calculated coarse bit error rate GBER and fine bit error rate ABER, and send the faults corresponding to the coarse fingerprint and fine fingerprint with the highest similarity as the identification results; Step 5: Dynamically update the corresponding coarse fingerprint and fine fingerprint in the fault identification library; For the generator excitation rectifier circuit faults that cannot be identified by the Level I fault identification library and the Level II fault identification library, corresponding Level III supplementary fault identification library and Level IV supplementary fault identification library are established respectively. In step 2, the time domain information of the directly collected quantities within t1 seconds after each level I rectifier cabinet fails is mapped to form a corresponding rough fingerprint of the level I rectifier cabinet fault, and a level I fault identification library is established; the directly collected quantities include the trigger angle and the excitation voltage; wherein the steps of forming the rough fingerprint are as follows: Step 2.1) Based on the time domain information of the directly collected quantity, the autocorrelation function is calculated and the first k periods are extracted; Step 2.2) The extracted autocorrelation function is reconstructed by wavelet transform, and the wavelet transform reconstruction result is divided into M G signal frames, calculate the signal frame energy E G and zero crossing number ZC G ; Step 2.3) For E G and ZC G Mapping is performed to obtain the corresponding fingerprint information. In step 3, the time domain information of the non-collected quantity is calculated according to the time domain information of the directly collected quantity within t2 seconds after each level II thyristor fails, and the calculated non-collected quantity time domain information is mapped to form the corresponding level II rectifier cabinet fault fine-control fingerprint, and a level II fault identification library is established; the non-collected quantity includes excitation power, AC side excitation current and average energy; wherein, the steps of forming the fine-control fingerprint are as follows: Step 3.1) Based on the time domain information of the directly collected quantity, calculate the non-collected quantity excitation power, AC side excitation current and average energy; Step 3.2) Based on the calculated time domain information of the non-collected quantity, calculate its autocorrelation function and extract the first k periods; Step 3.3) Perform wavelet transform reconstruction on the extracted autocorrelation function and divide the wavelet transform reconstruction result into M A signal frames, calculate the signal frame energy E A and zero crossing number ZC A ; Step 3.4) For E A and ZC A Mapping is performed to obtain the corresponding fingerprint information.

2. The method for identifying faults in a generator excitation rectifier circuit based on information fingerprint according to claim 1 is characterized in that: The process of step 4 is as follows: 4.1) Define the coarse bit error rate GBER to measure the similarity between coarse fingerprints, and introduce the energy correction coefficient k EG ; The rough fingerprint generated by the actual operation fault is compared with the rough fingerprint of each fault in the level I fault identification library in turn to obtain the minimum rough bit error rate GBER min , minimum coarse bit error rate GBER min The corresponding rough fingerprint is the rough fingerprint with the highest similarity to the actual rectifier cabinet fault in the level I fault identification library; 4.2) Define the bit error rate ABER to measure the similarity between detailed fingerprints, and introduce the energy correction coefficient k EA ; The refined fingerprint generated by the actual operation fault is compared with the refined fingerprint of each fault in the level II fault identification library in turn to obtain the minimum fine bit error rate ABER min , minimum bit error rate ABER min The corresponding refined fingerprint is the refined fingerprint with the highest similarity to the actual thyristor fault in the Level II fault identification library.

3. The method for identifying faults in a generator excitation rectifier circuit based on information fingerprint according to claim 1 is characterized in that: In step 5, the fault identification library is updated and supplemented according to the coarse / fine bit error rate values ​​of the actual fault matching. The specific process is as follows: Step 5.1) When the matched coarse / fine bit error rate is less than the set threshold β G / β A When the fingerprint code in the level I / level II fault identification library is dynamically updated, the steps are as follows: Step 5.1.1) Compare the coarse / fine fingerprints in the level I / level II fault identification library with the coarse / fine fingerprints generated by the actual operation fault, and identify the positions with different bit values ​​in the corresponding fingerprints; Step 5.1.2) For different positions, replace the coarse / fine elements in the level I / level II fault identification library with the elements of the corresponding coarse / fine fingerprint in the actual fault fingerprint according to the set probability p; Step 5.2) When the matched coarse / fine bit error rate is greater than or equal to the threshold β G / β A ,The steps for establishing the Level III / Level IV supplementary fault identification library are as follows: Step 5.2.1) Re-collect the time domain information of the directly collected quantity during the fault process twice, calculate the coarse / fine bit error rate of the two collected signals respectively according to the step 4, and obtain the corresponding fault identification result; Step 5.2.2) If the two groups of coarse / fine bit error rates calculated in step 5.2.1) are both less than the set threshold β G / β A , and the fault identification results obtained are consistent, the result will be sent as the I-level rectifier cabinet / II-level thyristor fault identification result; otherwise, all the coarse / fine fingerprints of this fault and the time domain waveforms of the corresponding directly collected quantities are added to the III-level / IV-level supplementary fault identification library for subsequent identification and research, and a special fault warning is output.

Images