Fault diagnosis method for excitation power unit of hydraulic generator hybrid driven by data model

Abstract

A fault diagnosis method for an excitation power unit of a hydraulic generator hybrid driven by a data model comprises the following steps: firstly, acquiring operation data of the excitation power unit, eliminating abnormal data by utilizing a Raida criterion, and constructing a fault data training set; secondly, according to the fault condition of the thyristor in the three-phase rectifier bridge, the output voltage waveform is divided into a plurality of fault categories, and the peak occurrence time is extracted as a feature; thirdly, comparing a fault waveform collected in real time with the fault waveform classification library, identifying a fault category and calling a corresponding training set; and then, calculating the Euclidean distance between the real-time fault waveform and the training sample by adopting a K-proximity algorithm, selecting K samples with the minimum distance, and determining a final prediction result in combination with a voting method and a regression method. According to the method, data driving and an intelligent algorithm are combined, so that the accuracy and efficiency of fault diagnosis are remarkably improved, and reliable guarantee is provided for stable operation of a hydro-generator excitation system.

Description

Technical Field

[0001] The present invention belongs to the field of hydropower, and particularly relates to a fault diagnosis method for the excitation power unit of a hydrogenerator driven by a hybrid data model. Background Art

[0002] The water turbine is one of the core devices of the hydropower generation system. By converting the potential energy or kinetic energy of water into mechanical energy, it provides power output for the generator set. As a way to utilize clean and renewable energy, hydropower generation is widely used in various power systems around the world. According to its design and working principle, water turbines can be divided into various types such as Francis turbines, Kaplan turbines, and tubular turbines, which are suitable for different head and flow conditions.

[0003] The excitation system is an important part of the hydrogenerator set. Its main function is to provide a DC excitation current for the rotor of the generator, thereby generating a magnetic field and realizing the efficient conversion and output of electrical energy. Modern excitation systems usually adopt fully controlled excitation devices, which rectify alternating current into adjustable direct current through power units and adjust the voltage and reactive power output by the generator in real time to meet the dynamic requirements of power grid operation. The operating performance of the excitation system not only affects the stability of the generator but also plays an important role in the regulation ability and power quality of the power grid.

[0004] During the operation of the hydrogenerator, the excitation system needs to coordinate with the dynamic characteristics of the water turbine to cope with complex working conditions such as load fluctuations and power grid faults. However, due to the fact that its power unit is vulnerable to multiple factors such as electrical shocks and temperature changes during long-term operation under high voltage and high load, it is prone to failure, which directly threatens the operation stability of the unit and the reliability of the power grid. And the three-phase rectifier bridge failure is the main cause of faults in the excitation unit. When a fault occurs, it is necessary to timely and accurately identify the position of the faulty thyristor. Especially in the initial stage of the fault, the failure range may be expanded or the equipment damage may be aggravated due to the lag in identification. At the same time, traditional methods have a low degree of real-time monitoring of operation data and utilization of historical data, and lack intelligent diagnosis means, which are difficult to meet the requirements of modern hydropower stations for efficient equipment maintenance and precise management. Therefore, there is an urgent need for a fault diagnosis method for the excitation power unit of a hydrogenerator based on intelligent algorithms. Summary of the Invention The technical problem to be solved by the present invention is to provide a fault diagnosis method for the excitation power unit of a hydrogenerator driven by a hybrid data model. By collecting operation data, extracting multi-dimensional features, and combining historical data to construct an intelligent diagnosis model, it realizes accurate judgment of fault types, in-depth analysis of fault causes, and rapid positioning of fault positions, thereby improving the operation reliability and maintenance efficiency of the excitation system of the hydrogenerator and providing a solid guarantee for the stable operation of the power grid.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is: A data model hybrid driven hydraulic generator excitation power unit fault diagnosis method, the steps are: S1. Collect the operating data of the excitation power unit, use the Raida criterion to eliminate abnormal data, and save the processed data as a fault data training set; S2. According to the fault conditions of the thyristors in the three-phase rectifier bridge, the output voltage waveform is divided into multiple fault categories, and the peak appearance time of the waveform under each fault category is extracted as a feature; S3, comparing the real-time collected fault waveform with the fault waveform classification library; identifying the fault category according to the waveform characteristics; calling the corresponding fault data training set from the database according to the identified fault category; the waveform characteristics include the number and continuity of waveforms; S4, comparing the peak appearance time of the real-time fault waveform with the peak appearance time of each sample in the training set, and calculating the Euclidean distance; based on the calculated Euclidean distance, selecting K training samples with the smallest Euclidean distance; Voting method: Count the fault categories with the highest frequency among the K nearest neighbor samples as the prediction result; Regression method: Calculate the average value of the peak occurrence time of the K nearest neighbor samples as the prediction result; The final prediction result is determined based on the historical accuracy of the voting method and the regression method: if the prediction results of the two methods are consistent, the results are directly output; if they are inconsistent, the method with a higher weight coefficient is selected as the final result; S5, output fault type and location of specific faulty thyristor; S6. According to the diagnosis results in actual applications, dynamically adjust the K value of the KNN algorithm, add new fault data to the training set, update the fault data training library, and improve the generalization ability of the model. Preferably, in S1, collecting the operating data of the excitation power unit includes: Collect the real-time voltage waveform output by the three-phase rectifier bridge; Record the peak time of the waveform in each cycle.

[0006] Preferably, in S1, using the Raida criterion to eliminate abnormal data includes: Calculate the mean and standard deviation of the peak occurrence time for each fault category; Abnormal data that do not meet the Raida criterion are eliminated, that is, abnormal data with peak time exceeding the range of the mean ± 3 times the standard deviation are eliminated.

[0007] Preferably, in S2, the fault categories include: Fault category 1: Single thyristor fault, the output voltage waveform is 4 consecutive waveforms in one cycle; Fault category 2: Two thyristors in the same phase are faulty, and the output voltage waveform is two discontinuous waveforms within one cycle; Fault category 3: Two thyristors of the same pole are faulty, and the output voltage waveform is two continuous waveforms within one cycle; Fault category 4: Two thyristors of different phases and different poles are faulty, and the output voltage waveform is three continuous waveforms within one cycle; Fault category 5: Two thyristors in the same phase and one thyristor in the other two phases are faulty, and the output voltage waveform is one waveform within one cycle; Fault category 6: All three thyristors of the same pole are faulty, or two thyristors in each of the different poles are faulty, and the output voltage waveform has no waveform.

[0008] Preferably, in S1, the data is processed and trained using the 3σ criterion, and the specific process is as follows; Abnormal data is removed using the 3σ criterion. First, record the time data of the peak appearance in this case , i = 1, …, 6; j = 1, …n, where i refers to the serial number of the fault category and j refers to the corresponding data serial number; calculate the average peak appearance time of the same fault category i : (1); Calculate the standard deviation of the peak appearance time under the fault category i : (2); The 3σ criterion stipulates that for the recorded data , it needs to satisfy to be marked as normal data, otherwise it is defined as an incorrect record or noisy data with errors. For the data that does not meet the above conditions is removed, and the data after screening is recorded as the fault data training set : (3); where i refers to the serial number of the fault situation classification, j refers to the jth fault record corresponding to the fault situation i, and dmn should record the data of the faulty thyristor under the fault situation n V i and the peak appearance time in each of its cycles .

[0009] Preferably, the specific steps of S4 are as follows: (1) First, according to the output voltage waveform, compare it with the fault waveform to identify the fault category. After identifying the fault category, call the fault data training set of this fault category in the database ; (2) Denote the occurrence time of the fault waveform to be verified as X test , for each training sample corresponding to the peak occurrence time and the test sample X test , calculate the Euclidean distance between them: (4); (3) Calculate the distance between the test sample and each sample in the training set. According to the calculated distance, sort the training samples by distance and select the K training samples with smaller distances, that is, the samples in the training set that best match the fault situation to be predicted. The value of K should be a number not less than 1. When K = 1, the model may be easily affected by noise. When the value of K is too large, the model will be too smooth, resulting in underfitting. Therefore, the value of K should not be greater than the square root of the number of samples in the dataset, that is, K≈ , m is the total number of samples in the dataset; The K samples are the "nearest neighbors" of the test sample, which are defined as y p , p = 1, 2,.., K; (4) Use the voting method to vote on the prediction sample and the K nearest neighbor results, and select the category with the highest frequency as the prediction result. The prediction result obtained by the voting method is y test1 ; (5); Use the regression method to average the K nearest neighbor results, and thus obtain the prediction result. The prediction result obtained by the regression method is y test2 ; (6); Preferably, in step (4), to ensure accuracy, when using the voting method and the regression method for prediction, the weight method should be used to determine the final output result. When using this algorithm, the result weight of the voting method prediction should be configured as α, and the result weight of the regression method prediction should be β; For α and β, which are the accuracies of the two methods in practical applications, each time they are used, their accuracies should be recorded and updated in real time; When using the voting method, the number of predictions is 50 times, and the number of prediction errors is 5 times. Then the weight coefficient α of the voting method should be 0.9; When using the regression method, the number of predictions is 50 times, and the number of prediction errors is 3 times. Then the weight coefficient β of the voting method should be 0.94; Then when the prediction results of the two methods are the same, take this result as the output of the prediction; When the two are inconsistent, the method with the larger weight coefficient should be selected as the output result.

[0010] A fault diagnosis system for the excitation power unit of a hydro-generator driven by a hybrid data model adopts the fault diagnosis method for the excitation power unit of a hydro-generator driven by the hybrid data model as described above.

[0011] A computer device, characterized in that it includes: One or more processors; The processor is used to store one or more programs; When the one or more programs are executed by the one or more processors, the fault diagnosis method for the excitation power unit of a hydro-generator driven by the hybrid data model as described above is implemented.

[0012] A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed, the fault diagnosis method for the excitation power unit of a hydro-generator driven by the hybrid data model as described above is implemented.

[0013] The present invention can achieve the following beneficial effects: The present invention can accurately identify and classify the faulty thyristors of the rectifier bridge, significantly improving the diagnosis efficiency and accuracy. By collecting the operation data of the excitation power unit and constructing a high-precision diagnosis model, the present invention can not only quickly capture abnormal signals at the initial stage of the fault to provide early warnings, but also deeply analyze the fault types and fault causes, so as to quickly locate the fault occurrence position, effectively shorten the troubleshooting time, and improve the maintenance efficiency. It comprehensively guarantees the stable operation of the excitation system of the hydro-generator. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below with reference to the drawings and embodiments: Figure 1 is a working schematic diagram of a hydro-generator set and a power rectifier unit; Figure 2 is a working schematic diagram of a three-phase rectifier bridge of a generator and a power rectifier unit; Figure 3 is a waveform schematic diagram of different fault conditions; Figure 4 is a flowchart for eliminating abnormal data by the 3σ criterion; Figure 5 is a flowchart for training data by the KNN algorithm. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] As Figure 1 shown, when the water turbine is working, it outputs mechanical power P m to the rotor of the generator, and the rotor generates a rotating magnetic field that can drive the stator of the generator to work, causing the stator to generate voltage U 1to the power rectifier unit, which rectifies the voltage U 1 and outputs the rectified result to obtain U 2 , and then inputs it into the power grid through a transformer for grid connection.

[0016] As Figure 2 shown, u a , u b , u c are respectively the three-phase voltages generated by the generator, V 1 , V 2 , V 3 , V 4 , V 5 , V 6 are the thyristors of the three-phase rectifier bridge, where V 1 , V 3 , V 5 have their cathodes connected to each other, called the common cathode group; V 2 , V 4 , V 6 have their anodes connected to each other, called the common anode group. Figure 2 On the right is the waveform diagram of u a , u b , u c within one cycle of the rectifier bridge; when the three-phase rectifier bridge is working properly, the thyristors with higher phase voltages in the common cathode group will conduct, and the thyristors with lower phase voltages in the common anode group will conduct; based on this, one cycle is divided into 6 different working intervals: ① The common cathode group V 1 conducts, and the common anode group V 4 conducts; ② The common cathode group V 1 conducts, and the common anode group V 6 conducts; ③ The common cathode group V 5 conducts, and the common anode group V 6 conducts; ④ The common cathode group V 5 conducts, and the common anode groupV 2 Conductivity; ⑤ Common cathode group V 3 Conduction, common anode group V 2 Conductivity; ⑥ Common cathode group V 3 Conduction, common anode group V 4 Conductivity; The preferred solution is Figures 1 to 5 As shown, a method for diagnosing faults of a hydro-generator excitation power unit driven by a data model hybrid comprises the following steps: S1. Collect the operating data of the excitation power unit, use the Raida criterion to eliminate abnormal data, and save the processed data as a fault data training set; S2. According to the fault conditions of the thyristors in the three-phase rectifier bridge, the output voltage waveform is divided into multiple fault categories, and the peak appearance time of the waveform under each fault category is extracted as a feature; S3, comparing the real-time collected fault waveform with the fault waveform classification library; identifying the fault category according to the waveform characteristics; calling the corresponding fault data training set from the database according to the identified fault category; the waveform characteristics include the number and continuity of waveforms; S4, comparing the peak appearance time of the real-time fault waveform with the peak appearance time of each sample in the training set, and calculating the Euclidean distance; based on the calculated Euclidean distance, selecting K training samples with the smallest Euclidean distance; Voting method: Count the fault categories with the highest frequency among the K nearest neighbor samples as the prediction result; Regression method: Calculate the average value of the peak occurrence time of the K nearest neighbor samples as the prediction result; The final prediction result is determined based on the historical accuracy of the voting method and the regression method: if the prediction results of the two methods are consistent, the results are directly output; if they are inconsistent, the method with a higher weight coefficient is selected as the final result; S5, output fault type and location of specific faulty thyristor; The system provides comprehensive fault diagnosis information to operation and maintenance personnel by outputting specific fault categories, faulty thyristor locations, fault impact analysis, and maintenance recommendations. At the same time, through visual output and historical data analysis, the practicality and operability of fault diagnosis are further improved, providing strong support for the efficient maintenance and stable operation of the turbine generator excitation power unit.

[0017] S6. According to the diagnosis results in actual applications, dynamically adjust the K value of the KNN algorithm, add new fault data to the training set, update the fault data training library, and improve the generalization ability of the model. Preferably, in S1, collecting the operation data of the excitation power unit includes: Collecting the real-time voltage waveforms output by the three-phase rectifier bridge; Recording the peak occurrence time of the waveforms within each period.

[0018] Preferably, in S1, using the 3σ criterion to eliminate abnormal data includes: Calculating the average value and standard deviation of the peak occurrence time under each fault category; Eliminating the abnormal data that does not conform to the 3σ criterion, that is, eliminating the abnormal data whose peak time exceeds the range of the average value ± 3 times the standard deviation.

[0019] Preferably, in S2, the fault categories include: The waveform diagrams of different fault conditions are as shown in Figure 3 as follows, where Figure 3 Waveform I is the voltage waveform output when the rectifier bridge is working normally U 2 ; when one or more of the 6 thyristors fail and do not conduct, the working interval where they are located will not output U 2 , therefore, the fault categories of the thyristors and their rectified voltage waveforms U 2 can be classified: Fault category (1): V 1 to V 6 One thyristor among the thyristors fails, for example V 1 an open circuit occurs; the output voltage U 2 is 4 consecutive waveforms within one period, corresponding to Figure 3 Waveform II; Fault category (2): V 1 to V 6 Two thyristors in the same phase are open-circuited, for example V 1 and V 4 are open-circuited; the output voltage U 2 is 2 discontinuous waveforms within one period, corresponding to Figure 3 Waveform III; Fault category (3): V 1 to V 6 Two thyristors of the same pole are open-circuited, for example V1 and V 3 Circuit breaker; output voltage U 2 For two consecutive waveforms in one cycle, corresponding to Figure 3 Waveform IV; Fault category (4): V 1 arrive V 6 Two thyristors with different phases and poles are disconnected, for example V 1 and V 6 Circuit breaker; output voltage U 2 For 3 consecutive waveforms in one cycle, corresponding to Figure 3 Waveform V; Fault category (5): V 1 arrive V 6 The two thyristors in the same phase and one thyristor in the other two phases are disconnected at the same time, for example V 1 , V 4 and V 6 Circuit breaker; output voltage U 2 is a waveform in one cycle, corresponding to Figure 3 Waveform VI; Fault category (6): V 1 arrive V 6 All three thyristors in the same pole fail, or two thyristors in different poles fail, for example V 1 , V 3 and V 5 Failure, or V 1 , V 3 , V 2 and V 4 Fault; Output voltage U 2 No waveform.

[0020] Preferably, in S1, the Raida criterion is used to process and train the data, and the specific process is as follows: The Raida criterion is used to eliminate abnormal data. The flowchart of the Raida criterion for eliminating abnormal data is as follows:Figure 4 as shown

[0021] First, record the peak appearance time data in this case , i = 1, …, 6; j = 1, … n, where i refers to the serial number of the fault category and j refers to the corresponding data serial number; calculate the average peak appearance time of the same fault category i : (1); Calculate the standard deviation of the peak appearance time under the fault category i : (2); The 3σ criterion stipulates that for the recorded data , it is necessary to satisfy to be marked as normal data, otherwise it is defined as an incorrect record or noisy data with errors. For the data that does not meet the above conditions perform rejection processing, and record the data after screening as the fault data training set : (3); where i refers to the serial number of the fault situation classification, j refers to the jth fault record corresponding to the fault situation i, and dmn should record the data of the faulty thyristor V i and the peak appearance time in each cycle .

[0022] Preferably, the specific steps of S4 are as follows: The process of using the KNN algorithm to judge the faulty thyristor is as follows: (1) First, according to the output voltage waveform, compare it with Figure 3 the fault waveforms Ⅰ - Ⅵ in to identify the fault category. After identifying the fault category, call the fault data training set of this fault category in the database . For example, if the fault is waveform Ⅵ, then call the training set

[0023] (2) Denote the appearance time of the fault waveform to be verified as X test , and for each training sample corresponding peak appearance time and the test sample X test , calculate the Euclidean distance between them: (4) (3) Calculate the distance between the test sample and each sample in the training set. Sort the training samples according to the calculated distance. Select the K training samples with smaller distances, that is, the samples in the training set that most conform to the fault condition to be predicted. The value of K should be a number not less than 1. When K = 1, the model may be easily affected by noise. When the value of K is too large, the model may be too smooth, resulting in underfitting. Therefore, in practical engineering applications, the value of K should not be greater than the square root of the number of samples in the dataset, that is, K≈ , where m is the total number of samples in the dataset. These K samples are considered the "nearest neighbors" of the test sample and are defined as y p , (p = 1, 2,.., K). For the selection of the result, a method combining voting and regression is adopted.

[0024] (4) The voting method is to vote on the prediction sample and the K nearest neighbor results, and select the category with the highest frequency as the prediction result. The prediction result obtained by the voting method is y test1 .

[0025] (5) The regression method is to average the K nearest neighbor results, and thus obtain the prediction result. The prediction result obtained by the voting method is y test2 .

[0026] (6) To ensure accuracy, when using the voting method and the regression method for prediction, the weight method should be used to determine the final output result. When using this algorithm, the result weight of the voting method prediction should be configured as α, and the result weight of the regression method prediction should be β. For α and β, they should be the accuracies of the two methods in practical applications. Each time it is used, their accuracies should be recorded and updated in real time; for example, when using the voting method, the number of predictions is 50 times, and the number of prediction errors is 5 times, then the weight coefficient α of the voting method should be 0.9; when using the regression method, the number of predictions is 50 times, and the number of prediction errors is 3 times, then the weight coefficient β of the voting method should be 0.94. Then when the results predicted by the two methods are the same, this result is used as the output of the prediction; when the two are inconsistent, the method with the larger weight coefficient should be selected as the output result.

[0027] A fault diagnosis system for the excitation power unit of a hydro-generator driven by a data model hybrid drive adopts the above-mentioned fault diagnosis method for the excitation power unit of a hydro-generator driven by a data model hybrid drive.

[0028] A computer device, characterized in that it includes: One or more processors; The processor described above is used to store one or more programs; When the one or more programs are executed by the one or more processors, a fault diagnosis method for the excitation power unit of a hydrogenerator driven by a hybrid data model as described above is implemented.

[0029] A computer-readable storage medium stores a computer program, and when the computer program is executed, a fault diagnosis method for the excitation power unit of a hydrogenerator driven by a hybrid data model as described above is implemented.

[0030] The present invention collects the operating data of the output voltage of the excitation system, uses the waveform arrival time, and combines machine learning algorithms to achieve accurate diagnosis of different fault types. By fitting the mapping relationship between fault characteristics and operating states, this method improves the accuracy of fault diagnosis. This module can effectively avoid misdiagnosis and missed diagnosis, thereby ensuring the operating reliability of the unit and significantly improving the maintenance efficiency and regulation ability of the system.

[0031] The above embodiments are only the preferred technical solutions of the present invention and should not be regarded as limitations on the present invention. The protection scope of the present invention should be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A data model hybrid driven hydraulic generator excitation power unit fault diagnosis method, characterized in that The following steps are involved: S1. Collect the operating data of the excitation power unit, use the Raida criterion to eliminate abnormal data, and save the processed data as a fault data training set; S2. According to the fault conditions of the thyristors in the three-phase rectifier bridge, the output voltage waveform is divided into multiple fault categories, and the peak appearance time of the waveform under each fault category is extracted as a feature; S3, comparing the fault waveform collected in real time with the fault waveform classification library; Identify the fault type based on waveform characteristics; According to the identified fault category, the corresponding fault data training set is called from the database; The waveform characteristics include the number and continuity of waveforms; S4, comparing the peak appearance time of the real-time fault waveform with the peak appearance time of each sample in the training set, and calculating the Euclidean distance; According to the calculated Euclidean distance, select K training samples with the smallest Euclidean distance; Voting method: Count the fault categories with the highest frequency among the K nearest neighbor samples as the prediction result; Regression method: Calculate the average value of the peak occurrence time of the K nearest neighbor samples as the prediction result; The final prediction result is determined based on the historical accuracy of the voting method and the regression method: if the prediction results of the two methods are consistent, the result is directly output; If they are inconsistent, the method with a higher weight coefficient is selected as the final result; S5, output fault type and location of specific faulty thyristor; S6. According to the diagnosis results in actual applications, dynamically adjust the K value of the KNN algorithm, add new fault data to the training set, update the fault data training library, and improve the generalization ability of the model.

2. The method for diagnosing faults of a hydraulic generator excitation power unit driven by a data model hybrid according to claim 1 is characterized in that: In S1, the operation data of the excitation power unit are collected including: Collect the real-time voltage waveform output by the three-phase rectifier bridge; Record the peak time of the waveform in each cycle.

3. The method for diagnosing faults of a hydraulic generator excitation power unit driven by a data model hybrid according to claim 1 is characterized in that: In S1, the use of the Raida criterion to remove abnormal data includes: Calculate the mean and standard deviation of the peak occurrence time for each fault category; Abnormal data that do not meet the Raida criterion are eliminated, that is, abnormal data with peak time exceeding the range of the mean ± 3 times the standard deviation are eliminated.

4. The method for diagnosing faults of a hydraulic generator excitation power unit driven by a data model hybrid according to claim 1 is characterized in that: In S2, the fault categories include: Fault category 1: Single thyristor fault, the output voltage waveform is 4 consecutive waveforms in one cycle; Fault category 2: Two thyristors in the same phase fail, and the output voltage waveform is two discontinuous waveforms in one cycle; Fault category 3: Two thyristors at the same pole fail, and the output voltage waveform is 2 consecutive waveforms in one cycle; Fault category 4: Two thyristors with different phases and different poles fail, and the output voltage waveform is 3 consecutive waveforms in one cycle; Fault category 5: Two thyristors in the same phase or one thyristor in the other two phases fails, and the output voltage waveform is 1 waveform in one cycle; Fault Category 6: All three thyristors in the same pole fail, or two thyristors in different poles fail, and the output voltage waveform has no waveform.

5. The method for diagnosing faults of a hydraulic generator excitation power unit driven by a data model hybrid according to claim 3 is characterized by: In S1, the Raida criterion is used to process and train the data. The specific process is as follows; Use the Raida criterion to eliminate abnormal data. First, record the peak time data in this case. , i=1,…,6; j=1,…n, where i refers to the fault category number and j refers to the corresponding data number; Calculate the average peak occurrence time of the same fault category i : (1)； Calculate the standard deviation of the peak occurrence time under fault category i : (2)； The Raida Guidelines stipulate that the data recorded , need to meet Only then can it be marked as normal data, otherwise it is defined as erroneous records or noise data with errors. For data that does not meet the above conditions Perform elimination processing and record the data after screening as the fault data training set : (3)； Where i refers to the fault classification number, j refers to the jth fault record corresponding to fault condition i, and dm n should record the faulty thyristor under fault condition n. V i Data and the peak time in each cycle .

6. The method for diagnosing faults of a hydraulic generator excitation power unit driven by a data model hybrid according to claim 5 is characterized in that: The specific steps of S4 are: (1) First, the output voltage waveform is compared with the fault waveform to identify the fault category. After the fault category is identified, the fault data training set of the fault category is called in the database. ; (2) The time when the fault waveform to be verified occurs is recorded as X test , for each training sample The corresponding peak time and test sample X test , calculate the Euclidean distance between them: (4)； (3) Calculate the distance between the test sample and each sample in the training set. According to the calculated distance, sort the training samples by distance and select K training samples with smaller distances, that is, the samples in the training set that best match the fault condition to be predicted. The K value should be a number not less than 1. When K=1, the model may be easily affected by noise. When the K value is too large, the model will be too smooth, resulting in underfitting. Therefore, the K value should not be greater than the square root of the number of samples in the data set, that is, K≈ , m is the total number of samples in the data set; K samples are the "nearest neighbors" of the test sample, which are defined as y p , p=1,2,..,K; (4) The prediction sample and the K nearest neighbor results are voted by the voting method, and the category with the highest frequency is selected as the prediction result. The prediction result obtained by the voting method is y test1 ; (5)； The regression method is used to average the K nearest neighbor results to obtain the prediction result. The prediction result obtained by the regression method is y test2 ; (6)。 7. The method for diagnosing faults of a hydraulic generator excitation power unit driven by a data model hybrid according to claim 1 is characterized by: In step (4), to ensure accuracy, when using the voting method and regression method for prediction, the weight method should be used to determine the final output result. When using this algorithm, the weight of the result predicted by the voting method should be configured as α, and the weight of the result predicted by the regression method should be configured as β; α and β are the accuracy of the two methods in practical applications. Each time they are used, their accuracy should be recorded and updated in real time; When the voting method is used, the number of predictions is 50 and the predictions are wrong 5 times, then the weight coefficient α of the voting method should be 0.9; when the regression method is used, the number of predictions is 50 and the predictions are wrong 3 times, then the weight coefficient β of the voting method should be 0.94; then when the results predicted by the two methods are consistent, the result is output as the predicted result; When the two are inconsistent, the method with a larger weight coefficient should be selected as the output result.

8. A data model hybrid driven hydraulic generator excitation power unit fault diagnosis system, characterized by: A data model hybrid driven hydraulic generator excitation power unit fault diagnosis method according to any one of claims 1 to 7 is adopted.

9. A computer device, characterized in that: include: one or more processors; The processor is used to store one or more programs; When the one or more programs are executed by the one or more processors, a data model hybrid driven hydro-generator excitation power unit fault diagnosis method as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed, a data model hybrid driven hydro-generator excitation power unit fault diagnosis method as described in any one of claims 1 to 7 is implemented.

Images