Station territory protection method based on comprehensive multiple determination of fault discriminator

Abstract

The invention provides a comprehensive multi-judgment station domain protection method based on fault discriminators, and the method comprises the steps: determining a cooperative protection domain which comprises a plurality of protection domains, and each protection domain is provided with at least one fault discriminator; determining a first fault discriminator and a second fault discriminator under the condition that the fault discriminator discriminates that a fault occurs in the protection domain of the fault discriminator; according to the fault discrimination result of the first fault discriminator and the fault discrimination result of the second fault discriminator, cooperatively discriminating whether the fault of the first fault discriminator is misjudged; under the condition that the fault occurring in the first fault discriminator is not misjudged, cooperatively discriminating the fault type and position of the fault occurring in the first fault discriminator according to the fault information discriminated by the first fault discriminator and the fault information discriminated by the second fault discriminator; and performing fault isolation on the fault type. Rapid and accurate fault positioning is carried out through cooperative discrimination, the backup protection action can be accelerated, and the overall protection performance of the transformer substation is improved.

Description

Technical Field

[0001] The present invention relates to the field of protection technology of intelligent substations, and in particular to a comprehensive multiple-determination substation protection method based on a fault discriminator. Background Art

[0002] With the emergence of new energy grid connection and increasingly complex grid structure, traditional power system relay protection faces limited interactive information, increasingly difficult configuration and setting of protection devices, and may find it difficult to adapt to rapidly changing network conditions.

[0003] As the amount of network data in the power system increases, the response time required becomes faster and faster. Traditional relay protection methods are difficult to achieve protection purposes quickly and conveniently. The network topology of each line is difficult to control the overall situation, resulting in a large amount of normal and fault data, which is difficult to distinguish and process.

[0004] At present, some studies have applied artificial intelligence algorithms to fault identification of components such as transformers and busbars, but most of them still require the auxiliary setting of additional starting criteria. It is impossible for artificial intelligence algorithms to automatically identify the entire process of fault occurrence. Moreover, most studies have not constructed protection schemes from the perspective of the "four properties" of protection, and it is difficult to adapt well to the complexity and dynamics of the power system. Summary of the invention

[0005] The purpose of this application is to propose a station protection method based on a comprehensive multiple-determination fault discriminator, which can independently complete fault identification and positioning without starting the judgment criteria, thereby improving the overall performance of the system.

[0006] In a first aspect, an embodiment of the present application provides a site domain protection method for collaborative identification, including:

[0007] Determine a collaborative protection domain, the collaborative protection domain comprising a plurality of protection domains, each of the plurality of protection domains being configured with at least one fault discriminator;

[0008] In the case where the fault discriminator discriminates that a fault occurs in its protection domain, determining the fault discriminator as a first fault discriminator, and determining the fault discriminator corresponding to at least another protection domain in the collaborative protection domain as a second fault discriminator;

[0009] According to the fault determination result of the first fault determiner on its protection domain and the fault determination result of the second fault determiner on its protection domain, collaboratively determine whether the fault occurring in the protection domain corresponding to the first fault determiner is a misjudgment;

[0010] The preliminary judgment result of the fault judgment module of the collaborative protection domain is recorded as 0 and 1 according to no fault and fault respectively, that is, the normal operation state is recorded as 0, and the internal and external faults are both recorded as 1;

[0011] During operation, if the judgment result of one of the fault discriminators in the collaborative protection domain changes from 0 to 1, while the result of another fault discriminator in the collaborative protection zone is still 0, then after waiting for the first fault discriminator to pass a preset time, the two fault discriminators continuously output the preliminary judgment results of a quarter of a cycle again. If the results of a quarter of a cycle continuously output by the two fault discriminators are inconsistent, it is considered that the first fault discriminator misjudged;

[0012] After the first fault discriminator initially discriminates that a fault occurs in its protection domain and a preset discrimination time has elapsed, based on the fault discrimination result of the first fault discriminator on its protection domain and the fault discrimination result of the second fault discriminator on its collaborative protection domain, it is collaboratively discriminated whether the fault occurring in the protection domain corresponding to the first fault discriminator is misjudged;

[0013] If the judgment results of the two fault discriminators outputting a quarter of a cycle are consistent, it is considered that a fault has indeed occurred in the collaborative protection domain. At this time, the fault discriminator performs collaborative judgment, and the fault discriminator no longer outputs 0 and 1. The fault discriminator now classifies the twenty types of faults outside and inside the area in sequence, and the output results correspond to 1 to 20 one by one. The corresponding results of each fault output are then collaboratively judged by the two fault discriminators, and the fault type and location can be determined at this time. According to the fault type and location of the fault in the protection domain corresponding to the first fault discriminator, the collaborative protection domain is fault isolated.

[0014] On the second aspect, the present application proposes a fault discriminator trained with an artificial intelligence model. The fault discriminator is first trained using a data training set, and then configured using a transfer learning method. According to the different deep learning models configured according to the bus, line and transformer, the fault discriminator can adapt to various changing power grid operation scenarios; according to the configuration of the fault discriminator, the fault discriminator can eventually identify the fault types of single-phase ground short circuit, two-phase short circuit, two-phase ground short circuit, and three-phase ground short circuit.

[0015] In the third aspect, the present application proposes a method for setting a preset judgment time length for optimizing a fault discriminator, characterized in that the preset judgment time length of the fault discriminator is set after three eighths of a cycle to prevent misjudgment caused by frequent disturbances of the first quarter of a cycle and bad data points, while the quarter to three eighths of a cycle provide a margin for the judgment time. In addition, when the fault occurring in the protection domain corresponding to the first fault discriminator is not misjudged and after the preset judgment time length, the fault occurring in the protection domain corresponding to the first fault discriminator is again judged based on the fault information judged by the first fault discriminator and the fault information of any fault discriminator in the collaborative protection domain except the first fault discriminator, and the preset judgment time length is reduced, and the adjusted preset judgment time length is greater than or equal to one eighth of a cycle, so as to accelerate the exit from the judgment state.

[0016] The station domain protection scheme proposed in the present application can locate faults more quickly and accurately through collaborative protection, and when the main protection refuses to operate, it can accelerate the backup protection action, thereby improving the overall protection performance of the substation; in addition, it is also proposed to configure the artificial intelligence model through transfer learning. This method can realize fast, small sample size deep learning model training, and realize a station domain protection method for fault identification and positioning. This method is an artificial intelligence model that does not require the startup of judgment criteria, can independently complete the whole process of fault identification and meet the relay protection scheme required by the "four properties" of protection. It is specifically used for on-site protection of transformers and busbars, as well as station domain protection of smart substations, which improves the reliability of the system; in addition, the present application proposes a method for setting the preset judgment time of the optimized fault discriminator, which improves the accuracy of the data and accelerates the speed of fault identification.

[0017] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory and cannot limit the disclosure of the embodiments of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0019] Figure 1 It is a flowchart of a multiple determination station domain protection method provided by an application embodiment;

[0020] Figure 2 This is a schematic diagram of the station protection zoning of a 220kV substation;

[0021] Figure 3 It is a schematic diagram of the research case model of the busbar;

[0022] Figure 4 It is a schematic diagram of the changes in the indicator parameters of each model during the training process;

[0023] Figure 5 It is a schematic diagram of the discrimination performance of each model for faults under CT saturation;

[0024] Figure 6 It is a schematic diagram of the changes in the indicator parameters of each model during the training process;

[0025] Figure 7 It is a schematic diagram of the accuracy of the transfer learning model in distinguishing the faults within the bus area and the faults outside the bus area;

[0026] Figure 8 It is a schematic diagram of the changes in the indicator parameters of each model during the training process;

[0027] Fig. 9 It is a schematic diagram of the discrimination results of the models in the collaborative protection areas A1, A2 and A3;

[0028] Fig.10 It is a schematic diagram of the discrimination accuracy of the model in the collaborative protection areas A1, A2 and A3;

[0029] Fig.11 It is a schematic diagram of the discrimination results of the models in the collaborative protection areas A1 and B1;

[0030] Fig.12 It is a schematic diagram of the discrimination accuracy of the model in the collaborative protection areas A1 and B1;

[0031] Fig.13 It is a schematic diagram of the online discrimination results of the model in the collaborative protection area B1;

[0032] Fig.14 It is a schematic diagram of the online discrimination accuracy of the model in the collaborative protection area B1. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0034] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0035] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0036] See also Figure 1 , Figure 1 : is a flowchart of a multiple determination site domain protection method provided by an embodiment of the present application, the determination method comprises the following steps:

[0037] Step S110: determining a coordinated protection domain, the coordinated protection domain includes multiple protection domains, and the protection domains are mainly divided into bus protection domains, transformer protection domains and line protection domains according to protection objects.

[0038] For example, see Figure 2 The station protection zoning method of a 220kV substation is shown, and the specific information of each protection domain is shown in Table 1. The station protection area is divided according to the controllable circuit breakers to ensure that each circuit breaker is included in at least one protection domain, and each protection domain contains two single-component protection domains of electrical components or lines. According to the type of protection object, the coordinated protection domain of the station protection is divided into the busbar-transformer coordinated protection domain and the busbar-line coordinated protection zone.

[0039]

[0040]

[0041] Table 1

[0042] Step S120: configure at least one fault discriminator for each protection domain, and each of the fault discriminators is trained using an artificial intelligence model. Using the training strategy adopted in this application and optimizing it using a transfer learning algorithm enables the SECapsNet model to achieve online discrimination of transformer faults and magnetizing inrush currents without setting additional startup criteria.

[0043] For example, for Figure 3 The bus shown in the figure is analyzed, with the bus flow direction as the positive direction. First, the three-phase current signals, current fault components, bus voltage signals and voltage fault components of all branches connected to the bus are selected as the basic electrical characteristic quantities. The current of any line connected to the bus is selected as i sI , the sum of the currents of the remaining lines is i sII, the currents flowing through lines L1, L2, L3 and L4 are i1, i2, i3 and i4 respectively, and i sI =i1, then i sII =i2+i3+i4. When the busbar is operating normally or an out-of-zone fault occurs, i sI and i sII are equal in magnitude and 180° in phase; when a busbar fault occurs within the area, i sI and i sII The phase difference is 0°.

[0044] For the busbar, when a fault occurs outside the zone, the CT at the branch end where the fault point is located may face serious saturation problems. For example, when a fault occurs in line L2, the current flowing from the busbar to the remaining branches will all flow to the fault point. At this time, the current i2 flowing through line L2 will increase rapidly, and the CT of line L2 will face saturation problems. Therefore, i sI and i sII The current polarity ratio characteristic quantity composed of the fault characteristics is used to retain the fault characteristics, and the A phase current is used for analysis. sIa 、i sIIa I sI 、i sII The A phase current.

[0045] (1) The A-phase polarity ratio p of the current on both sides of the busbar a The calculation formula is as follows:

[0046]

[0047] (2) The difference dp between the currents on both sides of the busbar and the polarity ratio of phase A a , is the current moment p a Subtract p corresponding to the fault-free time a .

[0048] (3) The A-phase polarity ratio and p of the current on both sides of the busbar sa , is the sum of the current moment and the historical moment p, and the formula is as follows:

[0049]

[0050] Using the bus three-phase voltage and three-phase voltage fault components, the three-phase current signals and three-phase current fault components of all branches, i sI with i sII The three-phase current polarity ratio p, the three-phase current polarity ratio difference dp and the three-phase current polarity ratio and p s , combined to form the data feature model of the bus, the time window lengths are 1 / 4 cycle and 1 / 2 cycle respectively.

[0051] Model 1: bus three-phase voltage signal, three-phase current signals of all branches connected to the bus, and the summed three-phase current signal of all branches i x A combination of, where each phase i x is the sum of all branch currents of the corresponding phase. The model includes u a -u b -u c -i 1a -i 1b -i 1c -i 2a -i 2b -i 2c -i 3a -i 3b -i 3c -i 4a -i 4b -i 4c -i xa -i xb -i xc .

[0052] Model 2: The fault component of the busbar three-phase voltage, the combination of the three-phase current fault components of all branches connected to the busbar, and the fault component di of the sum of the three-phase currents of all branches x Combination of a -du b -du c -di 1a -di 1b -di 1c

[0053] -di 2a -di 2b -di 2c -di 3a -di 3b -di 3c -di 4a -di 4b -di 4c -di xa -di xb -di xc .

[0054] Model 3: Based on Model 2, the current i is increased sI with i sII The three-phase current polarity ratio p combination includes:

[0055] p a -p b -p c .

[0056] Model 4: Based on Model 2, the current i is increased sI with isII The three-phase current polarity ratio difference dp combination includes dp a -dp b -dp c .

[0057] Model 5: Based on Model 2, add current i sI with i sII The three-phase current polarity ratio and p s Combination, including p sa -p sb -p sc .

[0058] Exemplarily, firstly, the fault discriminator is trained by deep learning using a data training set to establish a fault discriminator that can realize bus fault discrimination.

[0059] See Table 2 for the SECapsNet model parameters for bus fault identification, and Table 3 for the training-related parameters of the SECapsNet model. This example extracts features of electrical quantities based on the capsule neural network model, and stacks the electrical quantities into slices for training. SECapsNet is used to classify bus faults. The bus status can be divided into three categories: normal operation, out-of-zone faults, and in-zone faults. Out-of-zone faults and in-zone faults each contain 10 fault types, so there are 21 bus states in total.

[0060] In addition, due to the large number of lines connected to the bus, the number of signals fused by the bus data feature model also increases, so a 5×5 convolution kernel is used to replace the original convolution layer C2 and the 3×3 convolution kernel of the main capsule layer. At the same time, the output of the digital capsule layer becomes 21×16, corresponding to 20 fault types and 1 non-fault type of the bus.

[0061]

[0062] Table 2

[0063]

[0064] Table 3

[0065] This application is for Figure 3The busbar simulation case shown in the figure is studied. The specific system parameters are: power supply G1 = 330∠25°, power supply G2 = 330∠0°, power supply G3 = 330∠10°, power supply G4 = 330∠5°, the length of line L1 is 10km, the length of line L2 is 45km, the length of line L3 is 50km, the length of line L4 is 15km, and the line parameters are R1 = 0.027Ω / km, X1 = 0.282Ω / km, R0 = 0.195Ω / km, X0 = 0.693Ω / km. The power angle before the fault is the difference between the phase angles of power supply G1 and power supply G2, which is adjusted by changing the phase angle of power supply G1. S1~S4 are protection units, and the fault distance is the distance from the fault point to the protection unit of the fault line.

[0066] Simulation samples with comprehensive fault parameters were collected from protection units S1~S4. Each fault sample contains one cycle of data before and after the fault occurs. The specific simulation parameters of the training set and test set samples are shown in Tables 4 and 5. The intra-zone faults set on the bus and the extra-zone faults set on lines L1~L4 include single-phase ground short circuit, two-phase short circuit, two-phase ground short circuit and three-phase ground short circuit, a total of 10 fault types. At the same time, the samples also include samples of CT saturation caused by intra-zone and extra-zone faults of the bus. The CT saturation samples include 10 intra-zone and extra-zone fault types, and the number of samples is two-fifths of the number of non-saturated samples. The samples are preprocessed according to the data feature quantity model, and the samples are randomly selected as the training, verification and test subsets of the training set at a ratio of 7:2:1.

[0067]

[0068] Table 4

[0069]

[0070] Table 5

[0071] After data preprocessing using data feature models 1 to 5, the accuracy and loss changes of the trained SECapsNet models BM1 to BM5 during the training process are shown in the following figure: Figure 4 As shown in the figure, the accuracy of models BM1 to BM5 increases with the number of iterations, while the loss value gradually decreases. The accuracy of models BM1 to BM5 on the test subset of the training set is above 99%.

[0072] Internal and external faults are accompanied by CT saturation. Based on the bus test set, it is set that after an external fault occurs, the CT at the branch end where the fault point is located reaches a saturated state. For internal bus faults, the CT at the S1 side is set to be saturated. Figure 5The test results of models BM2 to BM5 are shown. When using saturated training samples, the duration of the fuzzy zone of models BM2 to BM5 does not exceed the length of the time window used, but only models BM4 and BM5 can maintain 100% accuracy in the stable area, while models BM2 and BM3 are at risk of misjudgment.

[0073] When saturated training samples are not used, the accuracy of models BM2 to BM5 shows a trend of first decreasing, then increasing, then decreasing again, and then increasing again after the fault. The area corresponding to the time window length after the fault occurs is selected as the stable area, and the average accuracy of the stable area of ​​each model is calculated. Among them, the average accuracy of the stable area of ​​model BM5 is the highest, but the highest is only 95%, and there is still a large risk of false operation. Therefore, a small amount of CT saturated samples still need to be added in the training to assist in training.

[0074] Exemplarily, a model-based transfer learning method is used to achieve fast, small-sample SECapsNet model optimization to improve the robustness and versatility of the model, so that it can achieve accurate fault identification in new scenarios.

[0075] The SECapsNet models obtained through transfer learning using training sets TB1 to TB3 are denoted as models TBM1 to TBM3 respectively. The changes in accuracy and loss of each model during training are shown in the following table. Figure 6 shown.

[0076] After 20 cycles of iteration, the training accuracy of each model increased rapidly, and the loss value continued to decrease. During the training process, the time required for each model to complete a training round was recorded. The specific data are as follows: model TBM1 took 27s, model TBM2 took 76s, model TBM3 took 133s, and pre-trained model BM5 took 181s. As the number of training set samples doubled, the training time also basically showed a trend of doubling, which once again confirmed the significant advantages of the transfer learning method in saving time and data costs. In the testing phase, the time required for each model to test a single sample was recorded. The test time of models TBM1, TBM2, TBM3 and BM5 were 105us, 107us, 107us and 104us respectively, and the difference in test time between the models was very small.

[0077] Next, the performance of models TBM1~TBM3 was tested. Ten types of out-of-zone faults were set on lines L1~L4, and ten types of in-zone faults were set on the busbar. Test samples of CT saturation and single-phase high-resistance grounding faults caused by in-zone and out-of-zone faults were simulated. The test results are shown in Figure 2. Figure 7 shown.

[0078] exist Figure 7Among them, model TBM3, which uses the most training samples, performs best under various test conditions. The duration of the fuzzy zone is the shortest among the three, and the accuracy is maintained at 100% in the stable zone. Although model TBM1, which uses the least sample training, is similar to model TBM3 in the duration of the fuzzy zone, its accuracy fluctuates in the stable zone, resulting in poor overall performance. Model TBM2 is not much different from model TBM3 in the duration of the fuzzy zone, and can maintain 100% accuracy in the stable zone. Considering that the time and number of samples required for training model TBM2 and model TBM3 differ by twice, model TBM2 is the best choice that can take into account both training cost and model performance.

[0079] The SECapsNet models obtained through transfer learning using training sets TB4 and TB5 are denoted as models TBM4 and TBM5 respectively. The changes in accuracy and loss of each model during training are shown in the following table. Figure 8 As shown in the figure. After 20 cycles of iteration, the training accuracy of each model increased rapidly, and the loss value continued to decrease. The time required for models TBM4 and TBM5 to complete a training round was 76s and 78s respectively, and there was no significant difference in training time between the two. During the test, the time required for the two models to test a single sample was 103us and 105us respectively, which was basically consistent.

[0080] Next, models TBM4 and TBM5 were tested. Ten in-zone faults were set on bus B2, and ten out-of-zone faults were set on line L5 and power supply G6. Test samples of CT saturation and single-phase high-resistance grounding faults caused by faults were simulated. The performance of models TBM4 and TBM5 on the test set was statistically analyzed according to three performance zones. The test results are shown in Table 6.

[0081]

[0082] Table 6

[0083] Depend on Figure 8 It can be seen that the performance of model TBM4 is significantly better than that of model TBM5. The effect of using zero filling to complete the missing data is much better than using the same-side current data to fill in the missing data. The accuracy of model TBM4 in the stable area remains 100%, and the maximum duration of the fuzzy area does not exceed 1 / 4 cycle. After transfer learning, it shows good performance in the new target domain.

[0084] By using training sets of different sizes for transfer learning, it is verified that using a training set with a data size of 1 / 4 of the original data set can better balance the sample size, training time and model performance. At the same time, for the problem of missing data, the zero-filling method is more suitable for optimizing the bus transfer learning model. In the simulation case test, both models TBM2 and TBM5 can complete the recovery of accuracy within 1 / 4 cycle after the fault, showing superior performance in the new scenario.

[0085] Therefore, the safety margin is set to 1 / 8 cycle, and the set durations corresponding to models BM5, TBM2 and TBM5 are all 3 / 8 cycle. s=10 is selected, and the bus fault discrimination model based on SECapsNet is integrated. A larger-scale data set is used to train the bus fault discrimination model based on SECapsNet. Then, the possible operating scenarios of the bus are analyzed, and the transfer learning method is used to optimize the SECapsNet model to obtain a very good fault discriminator model.

[0086] Step S130: The fault discriminator performs preliminary discrimination and performs collaborative discrimination verification.

[0087] The step of collaboratively determining whether a fault occurring in the protection domain corresponding to the first fault discriminator is a misjudgment according to the fault discrimination result of the first fault discriminator on its protection domain and the fault discrimination result of the second fault discriminator on its protection domain comprises:

[0088] After the first fault discriminator initially determines that a fault occurs in its protection domain and after a preset determination time has passed, based on the fault determination result of the first fault discriminator on its protection domain and the fault determination result of the second fault discriminator on its protection domain, it is collaboratively determined whether the fault occurring in the protection domain corresponding to the first fault discriminator is a misjudgment.

[0089] The preliminary judgment result of the fault judgment module of the collaborative protection domain is recorded as 0 and 1 according to no fault and fault respectively, that is, the normal operation state is recorded as 0, and the internal and external faults are both recorded as 1;

[0090] During operation, if the judgment result of one of the fault discriminators in the collaborative protection domain changes from 0 to 1, while the result of the other fault discriminator in the collaborative protection zone is still 0, then after waiting for the first fault discriminator for a preset period of time, the two fault discriminators will re-continuously output the preliminary judgment results of a quarter of a cycle. If the results of a quarter of a cycle continuously output by the two fault discriminators are inconsistent, it is considered that the first fault discriminator has made a wrong judgment, and the collaborative judgment verification state is jumped out; if the results of a quarter of a cycle continuously output by the two fault discriminators are consistent, the next collaborative judgment state is entered to confirm the fault location and fault type.

[0091] Step S141: After the collaborative identification confirms that a fault is misdiagnosed, the fault identifier exits the collaborative identification waiting state.

[0092] The first fault discriminator initially discriminates that a fault occurs in its protection domain, and after a preset discrimination time has passed, based on a fault discrimination result of the first fault discriminator on its protection domain and a fault discrimination result of the second fault discriminator on its collaborative protection domain, collaboratively discriminates whether a fault occurring in the protection domain corresponding to the first fault discriminator is misjudged, including:

[0093] When the first fault discriminator initially discriminates that a fault occurs in its protection domain, if the fault discriminator's current fault discriminator's result of discriminating the coordinated protection domain indicates that no fault occurs in the coordinated protection domain of the second fault discriminator, then after a preset discrimination time has elapsed, the first fault discriminator's result of discriminating the fault of its protection domain is obtained again, and the second fault discriminator's result of discriminating the fault of its coordinated protection domain is obtained again;

[0094] If the fault determination result of the first fault determiner obtained again indicates that a fault occurs in its protection domain and the fault determination result of the second fault determiner obtained again indicates that no fault occurs in its collaborative protection domain, it is determined that a fault misjudgment occurs in the protection domain corresponding to the first fault determiner, and the fault determiner exits the collaborative determination waiting state.

[0095] Step S142: After collaborative identification confirms that the fault has occurred correctly, collaborative identification is performed again to confirm the fault location and fault type, and a trip signal is sent to achieve fault isolation.

[0096] The first fault discriminator initially discriminates that a fault occurs in its protection domain, and after a preset discrimination time has passed, based on a fault discrimination result of the first fault discriminator on its protection domain and a fault discrimination result of the second fault discriminator on its collaborative protection domain, collaboratively discriminates whether a fault occurring in the protection domain corresponding to the first fault discriminator is misjudged, including:

[0097] When the first fault discriminator initially discriminates that a fault occurs in its protection domain, if the fault discriminator's current fault discriminator's result of discriminating the coordinated protection domain indicates that no fault occurs in the coordinated protection domain of the second fault discriminator, then after a preset discrimination time has elapsed, the first fault discriminator's result of discriminating the fault of its protection domain is obtained again, and the second fault discriminator's result of discriminating the fault of its coordinated protection domain is obtained again;

[0098] If the collaborative discrimination is performed again, the fault discrimination result of the first fault discriminator obtained indicates that a fault occurs in its protection domain and the fault discrimination result of the second fault discriminator obtained again indicates that a fault occurs in its collaborative protection domain, then it is determined that the fault occurring in the protection domain corresponding to the first fault discriminator is not misjudged.

[0099] After confirming the occurrence of a fault, the fault discriminator performs collaborative discrimination again. The fault discriminator no longer outputs 0 and 1. The fault discriminator now classifies the twenty types of faults inside and outside the zone in turn, and the output results correspond to 1 to 20 one by one. The corresponding result of each fault output is then collaboratively discriminated by the multiple fault discriminators. At this time, the fault type and location can be determined, and then a trip signal is sent to isolate the fault.

[0100] Exemplarily, in response to the problem of long action time of existing backup protection, a specific implementation method of a station domain protection scheme with comprehensive multiple judgments is proposed, and a method for setting a preset judgment time of an optimized fault discriminator is used to improve data accuracy and accelerate the speed of fault identification.

[0101] Models such as Figure 2 As shown in the figure, fault simulation is performed on the high-voltage side, medium-voltage side, and low-voltage side of transformer T1, high-voltage side bus Bus1, medium-voltage side bus Bus2, low-voltage side bus Bus3, high-voltage side line L1, medium-voltage side line L2, and low-voltage side line L3, respectively. The specific simulation parameter settings are shown in Table 7. The station protection method with comprehensive multiple judgments proposed in this chapter is used to test the samples in the test set, and the fault judgment process of line L1, bus Bus1, and high-voltage side of transformer T1 is taken as an example to elaborate the fault collaborative judgment process between the models.

[0102]

[0103] Table 7

[0104] A phase A ground fault is set on the high voltage side of transformer T1, and the judgment results of the models in the coordinated protection zones A1, A2 and A3 are as follows: Fig. 9As shown. Among them, the output result 11 of the transformer T1 model corresponds to the A-phase grounding fault in the zone, and the output result 1 of the busbar Bus1, Bus2 and Bus3 models corresponds to the A-phase grounding fault outside the zone. At the moment of the fault, the output of the transformer T1 model changes from 0 to 11, and then detects the output of the busbar Bus1, Bus2 and Bus3 models in the busbar-transformer collaborative protection zone A1, A2 and A3 where it is located. The busbar Bus1, Bus2 and Bus3 models respectively determine the fault type after 1, 3 and 2 sampling points after the fault, and then enter the set time waiting. In the continuous segment interval [94,103], the output results of the transformer T1 model are all 11, and the output results of the busbar Bus1, Bus2 and Bus3 models are all 1. The judgment results of the four models match each other, and it is judged that the fault occurs on the transformer T1 and is an A-phase grounding fault.

[0105] The test was conducted using 10 fault samples set on the high voltage side of transformer T1. The discrimination accuracy of transformer T1 model, bus Bus1 model, Bus2 model and Bus3 model is as follows: Fig.10 As shown in the figure, within 7.5ms after the fault, the accuracy of the four models all reached 100%, meeting the judgment requirements and entering the delayed waiting period. During the delayed waiting period, each model can still maintain 100% accuracy. Although there are three collaborative protection zones associated with the transformer, and at most four models can participate in the collaborative judgment, in fact, three models can realize the redundant setting of information and functions. Therefore, the first three models that enter the set time can be selected as the basis for judgment to reduce the waiting time and the amount of data synchronization.

[0106] A phase A ground fault is set on bus Bus1. The judgment results of bus Bus1 model, line L1 model and transformer T1 model are as follows: Fig.11 As shown. Among them, the output result 11 of the bus Bus1 model corresponds to the A-phase grounding fault in the zone, and the output result 1 of the line L1 model and the transformer T1 model corresponds to the A-phase grounding fault outside the zone. After one sampling time after the fault occurs, the output of the bus Bus1 model changes from 0 to 11, and the type of fault is determined. The judgment results of the other models in the related bus-transformer coordinated protection zone A1 and the bus-line coordinated protection zone Bus1 are detected. The line L1 model and the transformer T1 model are both judged as the type of fault, and the three models enter the set time [61,90] to wait. Subsequently, in the continuous segment interval [91,100], the output results of the bus Bus1 model are all 11, and the output results of the line L1 model and the transformer T1 model are all 1. The judgment results of the three models match each other, and it is judged that the fault occurs on the bus Bus1 and is an A-phase grounding fault.

[0107] Next, we use the 10 fault samples set at the bus Bus1 in the test set to test. The discrimination accuracy of the bus Bus1 model, line L1 model and transformer T1 model is as follows: Fig.12 As shown in the figure, within 7.5ms after the fault, the accuracy of the three models all reached 100%, meeting the discrimination requirements.

[0108] A phase A ground fault is set on line L1. The discrimination results of line L1 model and bus Bus1 model are as follows: Fig.13 As shown. The output result 11 of the line L1 model corresponds to the A-phase grounding fault in the zone, and the output result 1 of the bus Bus1 model corresponds to the A-phase grounding fault outside the zone. After one sampling moment after the fault occurs, the output result of the line L1 model is 17, which is judged as a faulty type, and enters the set time length [61,90] to wait, and then detects the judgment result of the bus Bus1 model in the same collaborative protection zone as the line L1 model. The bus Bus1 model is judged as a faulty type after two sampling moments, and then enters the set time length [62,91]. In the continuous segment [92,101], the judgment results of the line L1 model are all 11, and the judgment results of the bus Bus1 model are all 1. The judgment results of the two models match each other, and it is judged that the fault occurs on the line L1 and is an A-phase grounding fault.

[0109] Using the 10 fault samples set on line L1 in the test set, the discrimination accuracy of the line L1 and bus Bus1 models in the collaborative protection zone B1 is as follows: Fig.14 After a fault occurs in the L1 line, the bus1 model and the L1 line model both enter the set waiting time. After the set 3 / 8 cycle time, the identification accuracy of the two models reaches 100%, which can achieve accurate fault identification.

[0110] The model accuracy and duration of the fuzzy zone when other components or lines fail are shown in Table 4h. The duration of the fuzzy zone of each model is less than the set duration of 7.5ms, and the accuracy of all models in the stable zone reaches 100%. During the delay waiting period, each model can still maintain 100% accuracy and can act quickly after the delay ends. Therefore, the station protection method with comprehensive multiple judgments proposed in this paper can realize the coordinated cooperation of multi-component fault discrimination models, realize fast and accurate fault discrimination, and accelerate the backup protection action when the main protection refuses to operate.

[0111]

[0112] Table 8

[0113] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A site protection method based on comprehensive multiple determination of fault discriminator, characterized in that: include: Determine a collaborative protection domain, the collaborative protection domain comprising a plurality of protection domains, each of the plurality of protection domains being configured with at least one fault discriminator; In the case where the fault discriminator discriminates that a fault occurs in its protection domain, determining the fault discriminator as a first fault discriminator, and determining the fault discriminator corresponding to at least another protection domain in the collaborative protection domain as a second fault discriminator; According to the fault determination result of the first fault determiner on its protection domain and the fault determination result of the second fault determiner on its cooperative protection domain, collaboratively determine whether the fault occurring in the protection domain corresponding to the first fault determiner is a misjudgment; In the case where the fault occurring in the protection domain corresponding to the first fault discriminator is not misjudged, the fault type and location of the fault occurring in the protection domain corresponding to the first fault discriminator are collaboratively determined based on the fault information determined by the first fault discriminator and the fault information determined by the second fault discriminator; According to the fault type and location of the fault occurring in the protection domain corresponding to the first fault determiner, the fault type is isolated.

2. The station domain protection method according to claim 1, characterized in that: The protection domain is divided into a line protection domain, a transformer protection domain, a busbar protection domain and a circuit breaker protection domain according to the protection object.

3. The station domain protection method according to claim 1, characterized in that: The fault discriminator is trained using an artificial intelligence model, and the artificial intelligence model is configured according to the line, transformer, busbar and circuit breaker; the fault discriminator can determine the fault types of single-phase ground short circuit, two-phase short circuit, two-phase ground short circuit and three-phase ground short circuit.

4. The station domain protection method according to claim 1, characterized in that: The method further comprises: In case of a fault misjudgment occurring in the protection domain corresponding to the first fault discriminator, fault isolation is not performed on the fault type and location of the fault occurring in the protection domain corresponding to the first fault discriminator.

5. The station domain protection method according to claim 1, characterized in that: The step of collaboratively determining whether a fault occurring in the protection domain corresponding to the first fault discriminator is misjudged according to the fault discrimination result of the first fault discriminator on its protection domain and the fault discrimination result of the second fault discriminator on its collaborative protection domain comprises: After the first fault discriminator initially determines that a fault occurs in its protection domain and after a preset determination time has passed, based on the fault determination result of the first fault discriminator on its protection domain and the fault determination result of the second fault discriminator on its collaborative protection domain, it is collaboratively determined whether the fault occurring in the protection domain corresponding to the first fault discriminator is a misjudgment.

6. The station domain protection method according to claim 5, characterized in that: The first fault discriminator initially discriminates that a fault occurs in its protection domain, and after a preset discrimination time has passed, based on a fault discrimination result of the first fault discriminator on its protection domain and a fault discrimination result of the second fault discriminator on its collaborative protection domain, collaboratively discriminates whether a fault occurring in the protection domain corresponding to the first fault discriminator is misjudged, including: When the first fault discriminator initially discriminates that a fault occurs in its protection domain, if the fault discriminator's current fault discriminator's result of discriminating the coordinated protection domain indicates that no fault occurs in the coordinated protection domain of the second fault discriminator, then after a preset discrimination time has elapsed, the first fault discriminator's result of discriminating the fault of its protection domain is obtained again, and the second fault discriminator's result of discriminating the fault of its coordinated protection domain is obtained again; If the fault determination result of the first fault determiner obtained again indicates that a fault occurs in its protection domain and the fault determination result of the second fault determiner obtained again indicates that a fault occurs in its collaborative protection domain, it is determined that the fault occurring in the protection domain corresponding to the first fault determiner is not misjudged; If the fault determination result of the first fault determiner obtained again indicates that a fault occurs in its protection domain and the fault determination result of the second fault determiner obtained again indicates that no fault occurs in its collaborative protection domain, determining that a fault misjudgment occurs in the protection domain corresponding to the first fault determiner; If the fault determination result of the first fault determiner obtained again indicates that no fault occurs in its protection domain and the fault determination result of the second fault determiner obtained again indicates that a fault occurs in its collaborative protection domain, it is determined that a fault misjudgment occurs in the protection domain corresponding to the first fault determiner.

7. The station domain protection method according to claim 5 or 6, characterized in that: The method further comprises: When the fault occurring in the protection domain corresponding to the first fault discriminator is not misjudged, after a preset judgment time, the fault type and location of the fault occurring in the protection domain corresponding to the first fault discriminator are again determined collaboratively based on the fault information judged by the first fault discriminator and the fault information judged by the second fault discriminator.

8. The station domain protection method according to claim 5 or 6, characterized in that: The preset determination time includes eight minutes and three cycles.

9. The station domain protection method according to claim 7, characterized in that: The method further comprises: When the fault occurring in the protection domain corresponding to the first fault discriminator is not misjudged and after the preset judgment time has passed, the fault occurring in the protection domain corresponding to the first fault discriminator is again judged collaboratively based on the fault information judged by the first fault discriminator and the fault information of any fault discriminator in the collaborative protection domain except the first fault discriminator, and the preset judgment time is reduced, and the adjusted preset judgment time is greater than or equal to one eighth of a cycle, so as to accelerate the exit from the judgment state.

Images