A method for diagnosing a turn-to-turn short circuit of a rotor winding of a turbo-generator based on least square optimization

Abstract

The application discloses a kind of least square method optimization-based steam turbine generator rotor winding interturn short circuit diagnosis method, comprising the following steps: S1.collection steam turbine generator under normal operating condition Historical data;S2.sliding window method is selected from training sample in historical data, forms training sample set;S3.based on training sample set, least square method is used to optimize and solve to keep ladder reactance, and the optimized keep ladder reactance value is obtained;S4.using optimized keep ladder reactance value, combined with ASA phasor method calculates excitation current;S5.based on sliding window, the relative deviation mean of excitation current is calculated, and multi-level fault criterion is constructed according to historical normal data;S6.real-time acquisition operating data, repeat steps S4 to S5, based on optimized keep ladder reactance value and multi-level fault criterion, realize the online diagnosis and early warning of rotor winding interturn short circuit.The application can realize high-precision, high-reliability online diagnosis and early warning to steam turbine generator rotor winding interturn short circuit fault.

Description

Technical Field

[0001] This invention relates to the field of motor fault diagnosis technology, specifically to a method for diagnosing inter-turn short circuits in the rotor windings of a steam turbine generator based on least squares optimization. Background Technology

[0002] Large steam turbine generators are core equipment in power systems, and inter-turn short-circuit faults in their rotor excitation windings are a major hidden danger threatening the safe and stable operation of the unit. This fault can lead to increased rotor vibration, magnetic field asymmetry, and localized overheating, and in severe cases, may cause rotor grounding or even rotor swirl accidents, resulting in huge economic losses. Therefore, achieving early and accurate diagnosis of inter-turn short circuits has significant engineering value.

[0003] Currently, the diagnostic methods widely used in engineering mainly rely on the monitoring and analysis of excitation current. Their technical approaches can be divided into two main categories: intelligent algorithm-driven and mechanism model-based.

[0004] Intelligent algorithm-driven methods include Multivariate State Assessment (MSET), Support Vector Regression (SVR), Particle Swarm Optimization (PSO), and various neural networks (such as backpropagation neural networks and deep learning networks). These methods learn the complex nonlinear relationship between operating parameters and excitation current through data modeling. Their advantage lies in not requiring in-depth knowledge of motor electromagnetic design and their strong ability to capture certain nonlinear characteristics. However, these methods have the following significant drawbacks: (1) Poor interpretability: The diagnostic conclusions are difficult to be directly related to the physical fault mechanism, which is not conducive to the decision-making of maintenance personnel; (2) Weak generalization ability: It relies heavily on the completeness and representativeness of the training data. When the operating conditions of the unit exceed the range of the training data, the model performance drops sharply. (3) Insufficient anti-interference ability: It is sensitive to data quality and is easily affected by random noise and transient interference.

[0005] Mechanistic model-based computational methods, represented by the ASA phasor method recommended by the Institute of Electrical and Electronics Engineers (IEEE), are also the standard method in my country's national standard GB / T 1029-2021, "Test Methods for Three-Phase Synchronous Motors." This method is based on the classic phasor diagram of synchronous motors, using no-load and short-circuit characteristic curves, combined with fixed Boldt reactance parameters, to calculate the theoretical excitation current under load. Similar methods include the Swedish method, the Boldt reactance method directly utilizing the no-load characteristic curve, and numerical calculation methods based on finite element analysis (FEA). The advantages of mechanistic models lie in their clear physical meaning, high reliability, and wide applicability.

[0006] However, existing mechanistic methods such as the ASA phasor method have the following core problems when applied to online diagnosis of inter-turn short circuits: (1) The problem of fixed values ​​for the elevator reactance: The saturation reactance is a parameter related to the motor's saturation level, rotor structure, and air gap magnetic field distribution. Manufacturers typically obtain a fixed recommended value through limited type tests (such as no-load and short-circuit tests). However, in reality, this value can drift with unit aging, operating points (such as different loads and power factors), cooling conditions (such as hydrogen pressure and temperature), and even individual differences. Using a fixed value to calculate the excitation current inevitably introduces systematic errors, which may be misjudged as faults during fluctuations under normal operating conditions, while being masked by noise during early minor faults.

[0007] (2) Poor adaptability to flexible operating conditions: In modern power systems, thermal power units frequently need to participate in flexible operations such as deep peak shaving and rapid start-up and shutdown, resulting in large and rapid fluctuations in excitation current. Fixed-parameter mechanistic models and simple criteria are insufficient to accurately distinguish between normal operating condition changes and the manifestation of fault characteristics in such dynamic processes.

[0008] Therefore, the industry urgently needs a new diagnostic method that can inherit the advantages of physical interpretability and high reliability of mechanistic models, while overcoming their shortcomings such as fixed parameters, insufficient accuracy, and low level of intelligence. Summary of the Invention

[0009] The technical problem to be solved by the present invention is to provide a method for diagnosing inter-turn short circuits in turbine generator rotor windings based on least squares optimization, which can achieve high-precision and high-reliability online diagnosis and early warning of inter-turn short circuit faults in turbine generator rotor windings.

[0010] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows.

[0011] A method for diagnosing inter-turn short circuits in the rotor winding of a steam turbine generator based on least squares optimization includes the following steps: S1. Collect historical data of the steam turbine generator under normal operating conditions; S2. Use the sliding window method to select training samples from historical data to form a training sample set; S3. Based on the training sample set, the least squares method is used to optimize the Bold reactance and obtain the optimized Bold reactance value; S4. Calculate the excitation current using the optimized padding reactance value and the ASA phasor method; S5. Calculate the average relative deviation of the excitation current based on the sliding window, and construct a multi-level fault criterion based on historical normal data; S6. Collect operating data in real time, repeat steps S4 to S5, and realize online diagnosis and early warning of inter-turn short circuits in rotor windings based on optimized protection reactance value and multi-level fault criteria.

[0012] Preferably, in step S1, the historical data includes three-phase voltage, three-phase current, active power, reactive power, power factor, and excitation current.

[0013] Preferably, in step S2, selecting training samples from historical data using the sliding window method specifically includes: S21. Set the sliding window length L, where L is the number of data points included and represents the size of the time window; S22. Set the sliding step size S, where S represents the number of data points the sliding window moves each time; S23. Apply a sliding window to a historical normal data sequence, moving S data points each time, with each window containing L consecutive data points; S24. Standardize the collected variable data falling into each sliding window, and treat the data at each time point in the window as an independent training sample; each training sample contains the voltage, current, power factor angle and measured excitation current data at that time point; S25. Merge all training samples, or select representative samples from all training samples through cluster analysis to form the final training sample set.

[0014] Preferably, in step S3, the least squares method is used to optimize the solution of the surge reactance based on the training sample set, specifically including: S31. Based on the recommended value of the elevator protection reactance provided by the manufacturer, preset the optimization search range of the elevator protection reactance; S32. Uniformly discretize the optimization search range into... n The candidate values ​​are evenly spaced to form the candidate vector of the surge reactance; S33. For each candidate value in the candidate vector of the buffer reactance, the theoretical excitation current is calculated for each training sample in the training sample set using the ASA phasor method; S34. Calculate the relative deviation of the excitation current for each training sample under the corresponding candidate value of the paving reactance, and form a relative deviation matrix; S35. For each candidate value of the protection reactance, based on the least squares optimization criterion, calculate the sum of squares of the relative deviations of the excitation current of all training samples corresponding to it, so as to minimize the overall deviation between the theoretical calculation value and the actual measurement value, and form a vector of the sum of squares of deviations. S36. Find the minimum value in the vector of squared deviations. The candidate value of the Bold reactance corresponding to the minimum value is the optimized Bold reactance value.

[0015] Preferably, in step S33, for each candidate value in the candidate vector of the paving reactance, the theoretical excitation current is calculated for each training sample in the training sample set using the ASA phasor method. The formula is as follows: in, The calculated air gap potential; is the stator phase voltage of the generator, and is the measured voltage value in the training sample; is the stator phase current of the generator, and is the measured current value in the training sample; For the first Candidate values ​​for the elevator reactance; The power factor angle; Based on the calculated air gap potential By referring to the "air gap potential-excitation current" relationship curve obtained in advance through no-load test, the theoretical excitation current value required to maintain the air gap magnetic field under the current operating conditions can be obtained. ; In step S34, the formula for calculating the relative deviation of the excitation current is: in, Indicates the first j The training sample, using the first training sample The relative deviation of the excitation current obtained when calculating the candidate values ​​of the individual protection reactance; Indicates the first j The training sample, using the first training sample When considering candidate values ​​for the step-up reactance, the theoretical excitation current value is calculated using the ASA phasor method. Indicates the first j The measured excitation current corresponding to each training sample; In step S35, the formula for calculating the sum of squared deviations is: in, Indicates the first The sum of squared deviations corresponding to the candidate values ​​of the elevator reactance; Obtain the vector of squared deviations . Preferably, in step S4, calculating the excitation current using the optimized hold reactance value includes: S41. Obtain real-time operating data at the current moment, including three-phase voltage, three-phase current, active power, reactive power, power factor, and excitation current; S42. Calculate the current power factor angle based on real-time operating data; S43. Substitute the optimized paving reactance value, real-time operating data, and power factor angle into the ASA phasor method calculation formula to calculate the theoretical excitation current at the current moment.

[0016] Preferably, step S5 specifically includes: S51. Set the length of the diagnostic window, which is used in the online diagnostic phase and includes the most recent diagnostic window. Data at each point in time; S52. Calculate the relative deviation of the single-point excitation current at each time point within the diagnostic window. The calculation formula is as follows: in, for The theoretical excitation current is calculated based on the optimized protection reactance value at all times; for The excitation current measured at all times; S53. Calculate the average relative deviation of excitation current within the diagnostic window. The calculation formula is as follows: S54. Based on the historical normal data, the statistical characteristics of the relative deviation mean sequence of excitation current are statistically analyzed, including the mean μ and the standard deviation σ. S55. Based on statistical characteristics, set multi-level fault judgment thresholds, including early warning thresholds. α and fault threshold β .

[0017] Preferably, step S6 specifically includes: S61. Real-time acquisition of turbine generator operating data; S62. Based on steps S4 and S51 to S53, calculate the average relative deviation of the excitation current within the current diagnostic window in real time. ; S63. Will Compare with a preset threshold: like ≤ α The turbine generator rotor winding was determined to be in normal operating condition. like α < ≤ β The system determines that an abnormal trend has occurred in the rotor winding of the steam turbine generator and issues an early warning signal. like > β The system determines that an inter-turn short circuit fault has occurred in the rotor winding of the steam turbine generator and issues a fault alarm signal. S64. Record the fault diagnosis results for each time period and generate a diagnosis report. The diagnosis report includes the time period and the diagnosis window. Value range and corresponding diagnostic results.

[0018] Preferably, step S6 is followed by the following step: S7. Periodic update mechanism; specifically including: S71. Regularly collect new normal operation data; S72. Re-execute steps S2 to S3 using the new normal operating data to update the optimized elevator reactance value; S73. Using the updated ballast reactance value, recalculate the mean relative deviation of excitation current in historical normal data, and update statistical features and thresholds; S74. Apply the updated padding reactance value and threshold to subsequent online diagnostics.

[0019] Due to the adoption of the above technical solutions, the technical progress achieved by this invention is as follows.

[0020] This invention achieves a revolution in diagnostic mechanisms and a breakthrough in computational accuracy: For the first time, least squares parameter optimization and the classic ASA phasor method are deeply integrated, creatively resolving the contradiction between the "static fixation" of the key parameter—the susceptibility test reactance—in traditional mechanistic models and the actual "dynamic operation" of the unit. Through a data-driven approach, the optimal susceptibility test reactance value is "tailor-made" for each unit, achieving unprecedented levels of computational accuracy in the mechanistic model. In practical application verification, the average relative deviation of the calculated excitation current value was optimized from -0.77% to +0.07%, while the deviation fluctuation range was reduced by approximately 40%. This breakthrough in accuracy lays a solid computational foundation for the subsequent accurate identification and diagnosis of faults, fundamentally improving the core reliability of the diagnostic method.

[0021] This invention achieves a significant improvement in the robustness and reliability of fault identification: To address the shortcomings of traditional single-point criteria, which are susceptible to noise and transient operating condition fluctuations, this invention proposes a trend-based statistical criterion based on the sliding window mean. This criterion effectively filters out random interference by smoothing and integrating time-series data, and extracts the persistent and trend-based changes caused by inter-turn short-circuit faults. This method significantly enhances the noise immunity and operating condition adaptability of the diagnostic system, maintaining high reliability even under complex operating conditions such as peak shaving and variable loads. Field tests show that this method significantly reduces the false alarm rate and false negative rate, achieving excellent performance with zero false alarms in continuous testing over several years.

[0022] This invention enhances the system's intelligence and adaptability: The diagnostic system constructed using this method possesses self-learning and adaptive capabilities. On one hand, the core parameter, the protection surge reactance, can be periodically re-optimized based on newly acquired normal operating data; on the other hand, the diagnostic threshold can be dynamically adjusted based on the statistical characteristics of the unit's current health status. This mechanism enables the diagnostic system to continuously and autonomously maintain optimal diagnostic performance as the unit ages, component performance drifts, and operating conditions change, demonstrating a high level of intelligence and overcoming the drawback of traditional fixed-parameter models that become outdated upon service.

[0023] The early warning capability of this invention is greatly improved: By employing trend-based criteria and dynamic thresholds, this invention achieves early warning of inter-turn short-circuit faults. In actual fault case tests, compared to the traditional ASA method, this invention can capture early signs of faults several days in advance, providing maintenance personnel with a valuable early warning window, making preventive maintenance and planned repairs possible, and effectively avoiding unplanned downtime and major equipment damage that may result from the escalation of faults.

[0024] This invention has strong engineering applicability and low implementation and promotion costs: This invention is based entirely on standard operating data (three-phase voltage, current, power, excitation current, etc.) that can be collected by existing power plant monitoring systems (such as DCS), without the need for any expensive new sensors or detection devices, nor relying on complex internal electromagnetic design parameters of the motor. The algorithm has clear logic and a high degree of modularity, and can be easily integrated into existing power plant-level monitoring information systems (SIS), distributed control systems (DCS), or new equipment health management platforms in software form. Implementation and modification costs are low, deployment is convenient, and it has good engineering practicality and broad prospects for widespread application.

[0025] This invention offers significant overall benefits and has remarkable practical value: In summary, this invention, through the fusion of mechanisms and data, the combination of static and dynamic approaches, and the complementarity of single-point and trend analyses, forms a highly accurate, robust, and adaptive intelligent diagnostic system. It not only improves the diagnostic accuracy of individual faults but also provides a reliable technical tool for predictive maintenance and intelligent operation and maintenance of power generation equipment through a systematic architecture, demonstrating significant technological advancement and practical economic value. Attached Figure Description

[0026] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0028] A method for diagnosing inter-turn short circuits in the rotor winding of a steam turbine generator based on least squares optimization, combined with... Figure 1 As shown, it includes two stages: offline training and parameter optimization, and online real-time diagnosis and early warning. Specifically, it includes the following steps: S1. Collect historical data of the steam turbine generator under normal operating conditions.

[0029] Historical data includes three-phase voltage, three-phase current, active power, reactive power, power factor, and excitation current.

[0030] S2. Use the sliding window method to select training samples from historical data to form a training sample set.

[0031] The sliding window method for selecting training samples from historical data specifically includes: S21. Set the sliding window length L, where L is the number of data points included, representing the size of the time window.

[0032] S22. Set the sliding step size S, where S represents the number of data points the sliding window moves each time.

[0033] S23. Apply a sliding window to the historical normal data sequence, moving S data points each time, with each window containing L consecutive data points.

[0034] S24. Standardize the collected variable data falling into each sliding window, and treat the data at each time point in the window as an independent training sample; each training sample contains the voltage, current, power factor angle and measured excitation current data at that time point.

[0035] Specifically, the power factor angle is calculated from active power and reactive power or through the power factor.

[0036] S25. Merge all training samples, or select representative samples from all training samples through cluster analysis to form the final training sample set.

[0037] S3. Based on the training sample set, the least squares method is used to optimize the Pt reactance and obtain the optimized Pt reactance value.

[0038] The least squares method is used to optimize the solution of the surge reactance based on the training sample set, specifically including: S31. Based on the recommended value of the protection reactance provided by the manufacturer, preset the optimization search range of the protection reactance.

[0039] S32. Uniformly discretize the optimization search range into... n The candidate values ​​are evenly spaced to form the candidate vector of the ballast reactance.

[0040] S33. For each candidate value in the candidate vector of the purlite reactance, calculate the theoretical excitation current for each training sample in the training sample set using the ASA phasor method.

[0041] Specifically, for each candidate value in the candidate vector of the paternity reactance, the theoretical excitation current is calculated for each training sample in the training sample set using the ASA phasor method. The formula is as follows: in, The calculated air gap potential; is the stator phase voltage of the generator, and is the measured voltage value in the training sample; is the stator phase current of the generator, and is the measured current value in the training sample; For the first Candidate values ​​for the elevator reactance; The power factor angle; Based on the calculated air gap potential By referring to the "air gap potential-excitation current" relationship curve (i.e., the no-load characteristic curve) obtained in advance through the no-load test, the theoretical excitation current value required to maintain the air gap magnetic field under the current operating conditions can be obtained. .

[0042] S34. Calculate the relative deviation of the excitation current for each training sample under the corresponding candidate value of the padding reactance, and form a relative deviation matrix.

[0043] Specifically, the formula for calculating the relative deviation of the excitation current is: in, Indicates the first j The training sample, using the first training sample The relative deviation of the excitation current obtained when calculating the candidate values ​​of the individual protection reactance; Indicates the first j The training sample, using the first training sample When considering candidate values ​​for the step-up reactance, the theoretical excitation current value is calculated using the ASA phasor method. Indicates the first j The measured excitation current corresponding to each training sample.

[0044] S35. For each candidate value of the ballast reactance, based on the least squares optimization criterion, calculate the sum of squares of the relative deviations of the excitation current of all training samples corresponding to it, so as to minimize the overall deviation between the theoretical calculation value and the actual measurement value, and form a vector of the sum of squares of deviations.

[0045] Specifically, the formula for calculating the sum of squared deviations is: in, Indicates the first The sum of squared deviations corresponding to the candidate values ​​of the elevator reactance; Obtain the vector of squared deviations .

[0046] S36. Find the minimum value in the vector of squared deviations. The candidate value of the Bold reactance corresponding to the minimum value is the optimized Bold reactance value.

[0047] S4. Calculate the excitation current using the optimized plutonium reactance value and the ASA phasor method.

[0048] Specifically, calculating the excitation current using the optimized plenum reactance value includes: S41. Obtain real-time operating data at the current moment, including three-phase voltage, three-phase current, active power, reactive power, power factor, and excitation current; S42. Calculate the current power factor angle based on real-time operating data; S43. Substitute the optimized paving reactance value, real-time operating data, and power factor angle into the ASA phasor method calculation formula to calculate the theoretical excitation current at the current moment.

[0049] S5. Calculate the average relative deviation of the excitation current based on a sliding window, and construct multi-level fault criteria based on historical normal data. This includes the following steps: S51. Set the length of the diagnostic window, which is used in the online diagnostic phase and includes the most recent diagnostic window. Data at each point in time; S52. Calculate the relative deviation of the single-point excitation current at each time point within the diagnostic window. The calculation formula is as follows: in, for The theoretical excitation current is calculated based on the optimized protection reactance value at all times; for The excitation current measured at all times; S53. Calculate the average relative deviation of excitation current within the diagnostic window. The calculation formula is as follows: S54. Based on the historical normal data, the statistical characteristics of the relative deviation mean sequence of excitation current are statistically analyzed, including the mean μ and the standard deviation σ. S55. Based on statistical characteristics, set multi-level fault judgment thresholds, including early warning thresholds. α and fault threshold β .

[0050] S6. Real-time acquisition of operational data, repeating steps S4 to S5, based on optimized protection reactance values ​​and multi-level fault criteria, to achieve online diagnosis and early warning of inter-turn short circuits in the rotor windings. Specifically, this includes the following steps: S61. Real-time acquisition of turbine generator operating data; S62. Based on steps S4 and S51 to S53, calculate the average relative deviation of the excitation current within the current diagnostic window in real time. ; S63. Will Compare with a preset threshold: like ≤ α The turbine generator rotor winding was determined to be in normal operating condition. like α < ≤ β The system determines that an abnormal trend has occurred in the rotor winding of the steam turbine generator and issues an early warning signal. like > β The system determines that an inter-turn short circuit fault has occurred in the rotor winding of the steam turbine generator and issues a fault alarm signal. S64. Record the fault diagnosis results for each time period and generate a diagnosis report. The diagnosis report includes the time period and the diagnosis window. Value range and corresponding diagnostic results.

[0051] S7. Periodic update mechanism. This includes the following steps: S71. Regularly collect new normal operation data; S72. Re-execute steps S2 to S3 using the new normal operating data to update the optimized elevator reactance value; S73. Using the updated ballast reactance value, recalculate the mean relative deviation of excitation current in historical normal data, and update statistical features and thresholds; S74. Apply the updated paternity reactance value and threshold to subsequent online diagnostics. Example 1

[0052] I. Implementation Background and Data Preparation (1) Target unit and data source Unit model: QFSN-300-2 steam turbine generator; Rated parameters: power 300MW, voltage 20kV, rated excitation current: 2575A; Data source: Power plant DCS system; Data collection time range: May 2017 (normal), May 2018 (normal), January 2019 (normal), January 2021 (fault), January 2022 (fault). Data collected: three-phase voltage, three-phase current, active power, reactive power, power factor, and excitation current; Data processing: Extract start-up and shutdown status data, retain continuous operation segments, and sample at 1-minute intervals.

[0053] (2) Selection of training samples Define a time window of length L. Slide this window across the historical normal data sequence, with each slide step being S. The collected variable data (including voltage, current, power factor angle, and measured excitation current) falling within each window constitute a training sample. Select all training samples or filter representative training samples through clustering to form the final training sample set.

[0054] In this example, a sliding window method is used to select training samples from the normal data of May 2017. The window length L is 30 data points (i.e., 30 minutes, one point per minute), and the sliding step size S is 10 points. The final training sample set is obtained. ,in, Indicates the first j Stator phase voltage of each training sample; Indicates the first j Stator phase current of each training sample; Indicates the first j The power factor angle of each training sample; Indicates the first j Measured excitation current of each training sample; m =1500 groups, which is the total number of training samples.

[0055] II. Optimization process of elevator protection reactance (1) Setting the candidate range of elevator reactor Based on the recommended values ​​of the elevator reactance provided by the manufacturer Centered on this, a reasonable optimization search range is set based on engineering experience, and this optimization search range is discretized into... The candidate values ​​are evenly spaced to form the candidate vector of the ballast reactance.

[0056] Recommended value for manufacturer type in this example The optimized search range is set as follows: Divided into equal parts The candidate values ​​constitute the candidate vector of the elevator reactance:

[0057] (2) Excitation current calculation and deviation assessment For each candidate value, substitute it into the ASA phasor method calculation formula, and apply it to the training sample set. DFor each training sample, the theoretical excitation current is calculated. Then, the relative deviation of the excitation current under that training sample is calculated. After traversing all training samples and candidate values, a relative deviation matrix is ​​obtained.

[0058] In this example, for each candidate value in the candidate vector of the paternity reactance, the theoretical excitation current is calculated for each training sample in the training sample set using the ASA phasor method. The formula is as follows: Then, based on the calculated air gap potential By referring to the "air gap potential-excitation current" relationship curve obtained in advance through no-load test, the theoretical excitation current value required to maintain the air gap magnetic field under the current operating conditions can be obtained. .

[0059] Calculate the first Relative deviation of excitation current for each candidate value under each training sample: Obtain the relative deviation matrix .

[0060] (3) Optimization solution using the least squares method The goal of the least squares method is to minimize the sum of squares of the relative biases across all training samples. In this example, the formula for calculating the sum of squares of the relative biases for this column is as follows: Obtain the vector of squared deviations

[0061] (4) Determination of optimal protection reactance The minimum value in the objective function value vector is the corresponding candidate value of the purse sewer reactance, which is the optimal solution.

[0062] In this example, the index corresponding to the minimum value is taken. : The optimization result in this example is: The optimized protection reactance value, compared to the recommended value of 0.165895Ω, is the protection reactance that best matches the theoretical calculation with the actual historical operating data of the unit in the least squares sense.

[0063] III. Optimization Calculation and Verification of Excitation Current Use the optimized ballast reactance value The data for May 2017 were recalculated, and the results are shown in Table 1.

[0064] Table 1 Comparison of elevator reactance values ​​before and after optimization

[0065] Conclusion: After optimization, the bias is closer to zero and the distribution is more concentrated, resulting in a significant improvement in calculation accuracy.

[0066] IV. Implementation of Inter-turn Short Circuit Fault Diagnosis (1) Construction of sliding window mean criterion During the online diagnostic phase, a real-time sliding diagnostic window is set, the length of which can be the same as or different from the sliding window used in the offline training phase. At any given time, the diagnostic window contains the most recent... Data at various time points. Calculate the mean relative deviation of excitation current within the diagnostic window. . It is a smoothed indicator that reflects the average level of recent deviation. In this example, the diagnostic window length... One point (30 minutes).

[0067] Formula for calculating the average relative deviation of excitation current within the diagnostic window: in, ,for The relative deviation of the excitation current at a single point at any given time.

[0068] (2) Threshold setting and fault diagnosis The threshold is not fixed. During the offline training phase, it is adjusted based on historical normal data for each window. The value series is analyzed, and its mean and standard deviation are calculated. Early warning thresholds and fault thresholds can be set. During online operation, the mean and standard deviation can be updated periodically (e.g., quarterly) with new normal data to achieve adaptive thresholding. In this example, training data from May 2017 is used to determine the thresholds under normal conditions. The distribution range is:

[0069] Set the fault alarm threshold as follows: Early warning threshold: Fault threshold:

[0070] (3) Fault diagnosis results Tests were conducted on data from May 2018 (normal), January 2021 (faulty), and January 2022 (faulty): Table 2 Fault diagnosis results for each time period

[0071] (4) Comparison with traditional methods Table 3 Comparison results with traditional methods

Claims

1. A method for diagnosing inter-turn short circuits in the rotor winding of a steam turbine generator based on least squares optimization, characterized in that: Includes the following steps: S1. Collect historical data of the steam turbine generator under normal operating conditions; S2. Use the sliding window method to select training samples from historical data to form a training sample set; S3. Based on the training sample set, the least squares method is used to optimize the Bold reactance and obtain the optimized Bold reactance value; S4. Calculate the excitation current using the optimized padding reactance value and the ASA phasor method; S5. Calculate the average relative deviation of the excitation current based on the sliding window, and construct a multi-level fault criterion based on historical normal data; S6. Collect operating data in real time, repeat steps S4 to S5, and realize online diagnosis and early warning of inter-turn short circuits in rotor windings based on optimized protection reactance value and multi-level fault criteria.

2. The method for diagnosing inter-turn short circuits in the rotor winding of a steam turbine generator based on least squares optimization as described in claim 1, characterized in that: In step S1, the historical data includes three-phase voltage, three-phase current, active power, reactive power, power factor, and excitation current.

3. The method for diagnosing inter-turn short circuits in the rotor winding of a steam turbine generator based on least squares optimization as described in claim 1, characterized in that: In step S2, selecting training samples from historical data using the sliding window method specifically includes: S21. Set the sliding window length L, where L is the number of data points included and represents the size of the time window; S22. Set the sliding step size S, where S represents the number of data points the sliding window moves each time; S23. Apply a sliding window to a historical normal data sequence, moving S data points each time, with each window containing L consecutive data points; S24. Standardize the collected variable data falling into each sliding window, and treat the data at each time point in the window as an independent training sample; each training sample contains the voltage, current, power factor angle and measured excitation current data at that time point; S25. Merge all training samples, or select representative samples from all training samples through cluster analysis to form the final training sample set.

4. The method for diagnosing inter-turn short circuits in the rotor winding of a steam turbine generator based on least squares optimization as described in claim 1, characterized in that: In step S3, the least squares method is used to optimize the solution of the surge reactance based on the training sample set, specifically including: S31. Based on the recommended value of the elevator protection reactance provided by the manufacturer, preset the optimization search range of the elevator protection reactance; S32. Uniformly discretize the optimization search range into... n The candidate values ​​are evenly spaced to form the candidate vector of the surge reactance; S33. For each candidate value in the candidate vector of the buffer reactance, the theoretical excitation current is calculated for each training sample in the training sample set using the ASA phasor method; S34. Calculate the relative deviation of the excitation current for each training sample under the corresponding candidate value of the paving reactance, and form a relative deviation matrix; S35. For each candidate value of the protection reactance, based on the least squares optimization criterion, calculate the sum of squares of the relative deviations of the excitation current of all training samples corresponding to it, so as to minimize the overall deviation between the theoretical calculation value and the actual measurement value, and form a vector of the sum of squares of deviations. S36. Find the minimum value in the vector of squared deviations. The candidate value of the Bold reactance corresponding to the minimum value is the optimized Bold reactance value.

5. The method for diagnosing inter-turn short circuits in the rotor winding of a steam turbine generator based on least squares optimization according to claim 4, characterized in that: In step S33, for each candidate value in the candidate vector of the paving reactance, the theoretical excitation current is calculated for each training sample in the training sample set using the ASA phasor method. The formula is as follows: in, The calculated air gap potential; is the stator phase voltage of the generator, and is the measured voltage value in the training sample; is the stator phase current of the generator, and is the measured current value in the training sample; For the first Candidate values ​​for the elevator reactance; The power factor angle; Based on the calculated air gap potential By referring to the "air gap potential-excitation current" relationship curve obtained in advance through no-load test, the theoretical excitation current value required to maintain the air gap magnetic field under the current operating conditions can be obtained. ; In step S34, the formula for calculating the relative deviation of the excitation current is: in, Indicates the first j The training sample, using the first training sample The relative deviation of the excitation current obtained when calculating the candidate values ​​of the individual protection reactance; Indicates the first j The training sample, using the first training sample When considering candidate values ​​for the step-up reactance, the theoretical excitation current value is calculated using the ASA phasor method. Indicates the first j The measured excitation current corresponding to each training sample; In step S35, the formula for calculating the sum of squared deviations is: in, Indicates the first The sum of squared deviations corresponding to the candidate values ​​of the elevator reactance; Obtain the vector of squared deviations .

6. The method for diagnosing inter-turn short circuits in the rotor winding of a steam turbine generator based on least squares optimization according to claim 1, characterized in that: In step S4, calculating the excitation current using the optimized padding reactance value includes: S41. Obtain real-time operating data at the current moment, including three-phase voltage, three-phase current, active power, reactive power, power factor, and excitation current; S42. Calculate the current power factor angle based on real-time operating data; S43. Substitute the optimized paving reactance value, real-time operating data, and power factor angle into the ASA phasor method calculation formula to calculate the theoretical excitation current at the current moment.

7. The method for diagnosing inter-turn short circuits in the rotor winding of a steam turbine generator based on least squares optimization according to claim 1, characterized in that: Step S5 specifically includes: S51. Set the length of the diagnostic window, which is used in the online diagnostic phase and includes the most recent diagnostic window. Data at each point in time; S52. Calculate the relative deviation of the single-point excitation current at each time point within the diagnostic window. The calculation formula is as follows: in, for The theoretical excitation current is calculated based on the optimized protection reactance value at all times; for The excitation current measured at all times; S53. Calculate the average relative deviation of excitation current within the diagnostic window. The calculation formula is as follows: S54. Based on the historical normal data, the mean sequence of excitation current relative deviation is statistically analyzed, including the mean μ and the standard deviation σ. S55. Based on statistical characteristics, set multi-level fault judgment thresholds, including early warning thresholds. α and fault threshold β .

8. The method for diagnosing inter-turn short circuits in the rotor winding of a steam turbine generator based on least squares optimization according to claim 7, characterized in that: Step S6 specifically includes: S61. Real-time acquisition of turbine generator operating data; S62. Based on steps S4 and S51 to S53, calculate the average relative deviation of the excitation current within the current diagnostic window in real time. ; S63. Will Compare with a preset threshold: like ≤ α The turbine generator rotor winding was determined to be in normal operating condition. like α < ≤ β The system determines that an abnormal trend has occurred in the rotor winding of the steam turbine generator and issues an early warning signal. like > β The system determines that an inter-turn short circuit fault has occurred in the rotor winding of the steam turbine generator and issues a fault alarm signal. S64. Record the fault diagnosis results for each time period and generate a diagnosis report. The diagnosis report includes the time period and the diagnosis window. Value range and corresponding diagnostic results.

9. The method for diagnosing inter-turn short circuits in the rotor winding of a steam turbine generator based on least squares optimization according to claim 1, characterized in that: The step S6 is followed by the following step: S7. Periodic update mechanism; specifically including: S71. Regularly collect new normal operation data; S72. Re-execute steps S2 to S3 using the new normal operating data to update the optimized elevator reactance value; S73. Using the updated ballast reactance value, recalculate the mean relative deviation of excitation current in historical normal data, and update statistical features and thresholds; S74. Apply the updated paternity reactance value and threshold to subsequent online diagnostics.

Images