A Steam Turbine Condition Monitoring and Early Warning Method Based on MREMD-DS Fusion

Abstract

This invention discloses a turbine condition monitoring and early warning method based on MREMD-DS fusion, belonging to the field of energy and power equipment condition monitoring technology. The method includes: collecting and normalizing multi-parameter operating data of the turbine; decomposing the signal using an improved MREMD method, optimizing and filtering IMF components and reducing noise; calculating dynamic positive and negative ideal solutions; calculating distance relationship evidence and grey relational evidence; fusing multi-source evidence using D-S evidence theory; calculating dynamic stability scores and early warning thresholds; generating visualization results and triggering early warnings. This invention improves analysis accuracy through improved signal decomposition methods, enhances monitoring sensitivity through dynamic ideal solution calculations, and improves assessment reliability through evidence fusion, achieving real-time, accurate monitoring and early warning of the turbine's operating status.

Description

Technical Field

[0001] This invention relates to the field of energy and power equipment condition monitoring and fault early warning technology, and in particular to a steam turbine condition monitoring and early warning method based on MREMD-DS fusion. Background Technology

[0002] As a core component of energy power systems such as thermal power and nuclear power, the operational safety and stability of steam turbines directly affect the reliability and economy of the entire power generation system. Traditional steam turbine condition monitoring methods mainly rely on threshold alarms and expert experience, which have shortcomings such as alarm lag, high false alarm rate, and inability to provide early warnings. With the development of sensor technology and big data analytics, data-driven condition monitoring methods have gradually become a research hotspot.

[0003] Existing turbine condition monitoring methods mainly suffer from the following problems:

[0004] 1. Single-parameter threshold alarm methods are poorly adaptable to complex operating conditions involving multiple coupled parameters, and are difficult to fully reflect the overall status of the equipment;

[0005] 2. Traditional signal processing methods have limited ability to analyze non-stationary and nonlinear vibration signals;

[0006] 3. There is a lack of effective multi-source information fusion mechanisms, and the correlation information between various monitoring parameters has not been fully utilized;

[0007] 4. The warning thresholds are mostly fixed values, which cannot adapt to the dynamic changes in the operating status of the equipment. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention proposes a turbine condition monitoring and early warning method based on MREMD-DS fusion. This monitoring and early warning method improves the accuracy of analysis through an improved signal decomposition method, enhances the monitoring sensitivity through dynamic ideal solution calculation, and improves the reliability of assessment through evidence fusion, thereby realizing real-time and accurate monitoring and early warning of the turbine operating status.

[0009] A turbine condition monitoring and early warning method based on MREMD-DS fusion, the specific process of which is as follows:

[0010] S1, collect multi-parameter operating data of the steam turbine and perform normalization preprocessing;

[0011] S2 uses an improved MREMD method to decompose the signal, optimizes the selection of IMF components, and performs adaptive noise reduction;

[0012] S3, calculate the dynamic positive and negative ideal solutions;

[0013] S4, calculate distance relationship evidence and grey relational evidence;

[0014] S5 employs the DS evidence theory to integrate multi-source evidence;

[0015] S6, calculate the dynamic stability score and early warning threshold;

[0016] S7 generates visualizations and triggers alerts.

[0017] As a preferred embodiment of the above technical solution, in step S2, the improved MREMD method is as follows:

[0018] Boundary effects are addressed by using mirror-symmetric endpoint extension;

[0019] The envelope is calculated using spline interpolation at both the maxima and minima.

[0020] Set the convergence criteria to a normalized standard deviation less than 0.001 or to reach the maximum number of iterations.

[0021] As a preferred embodiment of the above technical solution, in step S2, the condition for optimizing the screening of IMF components is:

[0022] Correlation coefficient threshold: corr thr = max(0.1, mean corr - 0.5*std corr );

[0023] Root mean square error threshold: rmse thr = min(0.1, mean rmse + 0.5*std rmse );

[0024] Where, mean corr std corr The mean and standard deviation of the correlation coefficient. rmse std rmse Here are the mean and standard deviation of RMSE.

[0025] As a preferred embodiment of the above technical solution, in step S3, the calculation formulas for the dynamic positive and negative ideal solutions are as follows:

[0026] Positive ideal solution: F plus = min(0.8, win _mean + 2*win std );

[0027] Negative ideal solution: F minus = max(0.2, win _mean - 2*win std );

[0028] Among them, win _mean The mean of the data within the time window, win std The standard deviation is denoted as .

[0029] As a preferred embodiment of the above technical solution, in step S4, the formula for calculating the improved distance relationship between the monitoring data and the positive and negative ideal solutions is as follows:

[0030] H plus =-d plus ×(1-α·S plus );

[0031] H minus =-d minus ×(1-α·S minus );

[0032] Where, d plus d minus S is the Euclidean distance. plus S minus Here, α represents the trend similarity, and α is the adjustment coefficient.

[0033] As a preferred embodiment of the above technical solution, in step S4, the improved grey relational degree calculation formula between the monitoring data and the positive and negative ideal solutions is as follows:

[0034] G plus =1 / (1+mean(Δ plus ));

[0035] G minus =1 / (1+mean(Δ minus ));

[0036] Where, Δ plus Δ minus This represents the combined slope difference.

[0037] As a preferred embodiment of the above technical solution, in step S4, the formula for calculating the comprehensive slope difference Δ in the calculation of the grey relational evidence is as follows:

[0038] ;

[0039] in, The difference in slope is the first order. This is the difference in second-order slope.

[0040] As a preferred embodiment of the above technical solution, in step S5, the DS evidence theory fusion includes:

[0041] When the evidence is highly consistent, a weighted average fusion is used: T = mean(evidence) set );

[0042] When the consistency of evidence is low, Dempster's combination rule is used for fusion: T = Π(evidence) set ) / (1 -conflict);

[0043] Among them, evidence set Let be the set of evidence, and conflict be the conflict coefficient.

[0044] As a preferred embodiment of the above technical solution, in step S6, the formula for calculating the stability score is:

[0045] Score i =100-P v -P t - P p ;

[0046] Among them, P v To deduct points for volatility, P t Deduct points for trend instability, P p Points will be deducted for extreme values.

[0047] As a preferred embodiment of the above technical solution, in step S6, the calculation formula for the dynamic early warning threshold is:

[0048] Warning threshold = max(50, local _mean - 2×local std );

[0049] Among them, local mean The mean score for the local window. std The standard deviation is denoted as .

[0050] The beneficial effects of this invention are as follows:

[0051] This monitoring and early warning method improves the accuracy of analysis through an improved signal decomposition method, enhances the monitoring sensitivity through dynamic ideal solution calculation, and improves the reliability of assessment through evidence fusion. It realizes real-time and accurate monitoring and early warning of the turbine operating status, and provides quantitative status assessment results to support preventive maintenance decisions. Attached Figure Description

[0052] Figure 1 This is a flowchart of the monitoring and early warning method of the present invention.

[0053] Figure 2 This is a comparison chart of signal noise reduction effects in embodiments of the present invention.

[0054] Figure 3This is a graph showing the dynamic positive and negative ideal solutions in an embodiment of the present invention.

[0055] Figure 4 This is a heatmap showing the distribution of evaluation evidence in an embodiment of the present invention.

[0056] Figure 5 This is a diagram of the stability score distribution and early warning system in an embodiment of the present invention.

[0057] Figure 6 This is a diagram of the stability score distribution and early warning system in an embodiment of the present invention. Detailed Implementation

[0058] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0059] A turbine condition monitoring and early warning method based on MREMD-DS fusion, the specific process of which is as follows:

[0060] S1, collect multi-parameter operating data of the steam turbine and perform normalization preprocessing;

[0061] S2 uses an improved MREMD method to decompose the signal, optimizes the selection of IMF components, and performs adaptive noise reduction;

[0062] S3, calculate the dynamic positive and negative ideal solutions;

[0063] S4, calculate distance relationship evidence and grey relational evidence;

[0064] S5 employs the DS evidence theory to integrate multi-source evidence;

[0065] S6, calculate the dynamic stability score and early warning threshold;

[0066] S7 generates visualizations and triggers alerts.

[0067] In this embodiment, the turbine data collected in step S1 includes speed, condenser vacuum, main steam pressure, main steam temperature, reheat steam pressure, reheat steam temperature, feedwater temperature, exhaust steam pressure, shaft vibration, and oil temperature parameters.

[0068] In this embodiment, in step S2, the improved MREMD method is as follows:

[0069] Boundary effects are addressed by using mirror-symmetric endpoint extension;

[0070] The envelope is calculated using spline interpolation at both the maxima and minima.

[0071] Set the convergence criteria to a normalized standard deviation less than 0.001 or to reach the maximum number of iterations.

[0072] In this embodiment, in step S2, the condition for optimizing the screening of IMF components is:

[0073] Correlation coefficient threshold: corr thr = max(0.1, mean corr - 0.5*std corr );

[0074] Root mean square error threshold: rmse thr = min(0.1, mean rmse + 0.5*std rmse );

[0075] Where, mean corr std corr The mean and standard deviation of the correlation coefficient. rmse std rmse Here are the mean and standard deviation of RMSE.

[0076] In this embodiment, in step S3, the calculation formulas for the dynamic positive and negative ideal solutions are as follows:

[0077] Positive ideal solution: F plus = min(0.8, win _mean + 2*win std );

[0078] Negative ideal solution: F minus = max(0.2, win _mean - 2*win std );

[0079] Among them, win _mean The mean of the data within the time window, win std The standard deviation is denoted as .

[0080] In this embodiment, in step S4, the formula for calculating the improved distance relationship between the monitoring data and the positive and negative ideal solutions is as follows:

[0081] H plus =-d plus ×(1-α·S plus );

[0082] H minus =-d minus ×(1-α·S minus );

[0083] Where, dplus d minus S is the Euclidean distance. plus S minus Here, α represents the trend similarity, and α is the adjustment coefficient.

[0084] In this embodiment, in step S4, the improved grey relational degree calculation formula for the monitoring data and the positive and negative ideal solutions is as follows:

[0085] G plus =1 / (1+mean(Δ plus ));

[0086] G minus =1 / (1+mean(Δ minus ));

[0087] Where, Δ plus Δ minus This represents the combined slope difference.

[0088] In this embodiment, in step S4, the formula for calculating the comprehensive slope difference Δ in the calculation of the grey relational evidence is as follows:

[0089] ;

[0090] in, The difference in slope is the first order. This is the difference in second-order slope.

[0091] In this embodiment, step S5, the DS evidence theory fusion includes:

[0092] When the evidence is highly consistent, a weighted average fusion is used: T = mean(evidence) set );

[0093] When the consistency of evidence is low, Dempster's combination rule is used for fusion: T = Π(evidence) set ) / (1 -conflict);

[0094] Among them, evidence set Let be the set of evidence, and conflict be the conflict coefficient.

[0095] In this embodiment, in step S6, the formula for calculating the stability score is:

[0096] Score i =100-P v -P t - P p ;

[0097] Among them, P vTo deduct points for volatility, P t Deduct points for trend instability, P p Points will be deducted for extreme values.

[0098] As a preferred embodiment of the above technical solution, in step S6, the calculation formula for the dynamic early warning threshold is:

[0099] Warning threshold = max(50, local _mean - 2×local std );

[0100] Among them, local mean The mean score for the local window. std The standard deviation is denoted as .

[0101] Taking the actual operating data of a steam turbine unit as an example, the specific implementation process of the present invention is explained.

[0102] 1. Data Acquisition and Preprocessing

[0103] Ten key parameter data points were collected during 940 seconds of continuous operation of the steam turbine, with a sampling frequency of 1 Hz. The collected raw data were normalized to eliminate the influence of dimensions.

[0104] 2. MREMD Decomposition and Noise Reduction

[0105] An improved MREMD decomposition was performed on the time series of each parameter, with a maximum number of iterations set to 6. The number of IMF components obtained from the decomposition ranged from 3 to 7. An optimized dynamic threshold was used to filter effective IMF components, and the number of filtered IMF components accounted for 60%-80% of the total decomposition. Adaptive wavelet denoising was performed on the selected IMF components, and the residual of the denoised signal conformed to a normal distribution, verifying the denoising effect.

[0106] 3. Calculation of Dynamic Ideal Solution

[0107] The time window length was set to 30 seconds, resulting in 32 time windows. The positive and negative ideal solutions for each parameter within each window were calculated. The results show that the positive ideal solution F... plus It fluctuates within the range of [0.2, 0.8], with a negative ideal solution F. minus The fluctuation within the range of [0.2, 0.7] is consistent with actual engineering conditions.

[0108] 4. Evaluation of evidence calculation

[0109] Calculate the evidence H of the distance relationship between each window. plus H minus Grey relational evidence G plus G minusThe calculation process considered the data's changing trends, using correlation coefficients to measure trend similarity. Evidence values ​​were normalized to the [0,1] interval for easier subsequent fusion.

[0110] 5. DS Evidence Fusion

[0111] The Dempster-Shafer rule was used to fuse two types of evidence. When the two types of evidence were highly consistent, a weighted average was used; when the consistency was low, the Dempster combination rule was used to handle evidence conflicts. The combined weight T of each parameter after fusion reflects the importance of each parameter in the overall state assessment.

[0112] 6. Dynamic scoring and early warning

[0113] Based on three indicators—volatility, trend stability, and extreme value ratio—a stability score for each parameter is calculated. Then, the overall stability score of the equipment is calculated using a comprehensive weight T. Finally, a dynamic early warning threshold is calculated using a moving window statistical method. In this embodiment, the average score is 78.3 points, the standard deviation is 8.2, and the early warning threshold is set at 61.9 points. A total of four alarm windows were detected, representing an alarm rate of 12.5%.

[0114] 7. Generate visualization results

[0115] like Figures 2 to 6 As shown. Among them, Figure 2 This is a comparison chart of signal noise reduction effects. Figure 3 This is a graph showing the dynamic changes of positive and negative ideal solutions. Figure 4 To evaluate the evidence distribution heatmap, Figure 5 This is a diagram showing the stability score distribution and early warning system. Figure 6 This is a diagram showing the stability score distribution and early warning system.

[0116] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A turbine condition monitoring and early warning method based on MREMD-DS fusion, characterized in that: The specific process is as follows: S1, collect multi-parameter operating data of the steam turbine and perform normalization preprocessing; S2 uses an improved MREMD method to decompose the signal, optimizes the selection of IMF components, and performs adaptive noise reduction; S3, calculate the dynamic positive and negative ideal solutions; S4, calculate distance relationship evidence and grey relational evidence; S5 employs the DS evidence theory to integrate multi-source evidence; S6, calculate the dynamic stability score and early warning threshold; S7 generates visualizations and triggers alerts.

2. The turbine condition monitoring and early warning method based on MREMD-DS fusion according to claim 1, characterized in that: In step S2, the improved MREMD method is as follows: Boundary effects are addressed by using mirror-symmetric endpoint extension; The envelope is calculated using spline interpolation at both the maxima and minima. Set the convergence criteria to a normalized standard deviation less than 0.001 or to reach the maximum number of iterations.

3. The turbine condition monitoring and early warning method based on MREMD-DS fusion according to claim 2, characterized in that: In step S2, the conditions for optimizing the screening of IMF components are: Correlation coefficient threshold: corr thr = max(0.1, mean corr - 0.5*std corr ); Root mean square error threshold: rmse thr = min(0.1, mean rmse + 0.5*std rmse ); Where, mean corr std corr The mean and standard deviation of the correlation coefficient. rmse std rmse Here are the mean and standard deviation of RMSE.

4. The turbine condition monitoring and early warning method based on MREMD-DS fusion according to claim 1, characterized in that: In step S3, the calculation formulas for the dynamic positive and negative ideal solutions are as follows: Positive ideal solution: F plus = min(0.8, win _mean + 2*win std ); Negative ideal solution: F minus = max(0.2, win _mean - 2*win std ); Among them, win _mean The mean of the data within the time window, win std The standard deviation is denoted as .

5. The turbine condition monitoring and early warning method based on MREMD-DS fusion according to claim 1, characterized in that: In step S4, the formula for calculating the improved distance relationship between the monitoring data and the positive and negative ideal solutions is as follows: H plus =-d plus ×(1-α·S plus ); H minus =-d minus ×(1-α·S minus ); Where, d plus d minus Let S be the Euclidean distance. plus S minus Here, α represents the trend similarity, and α is the adjustment coefficient.

6. The turbine condition monitoring and early warning method based on MREMD-DS fusion according to claim 5, characterized in that: In step S4, the improved grey relational degree calculation formula between the monitoring data and the positive and negative ideal solutions is as follows: G plus =1 / (1+mean(Δ plus )); G minus =1 / (1+mean(Δ minus )); Where, Δ plus Δ minus This represents the combined slope difference.

7. The turbine condition monitoring and early warning method based on MREMD-DS fusion according to claim 6, characterized in that: In the calculation of the grey relational evidence, the formula for calculating the comprehensive slope difference Δ is as follows: ； in, The difference in slope is the first order. This represents the difference in second-order slopes.

8. The turbine condition monitoring and early warning method based on MREMD-DS fusion according to claim 1, characterized in that: In step S5, the DS evidence theory fusion includes: When the evidence is highly consistent, a weighted average fusion is used: T = mean(evidence) set ); When the consistency of evidence is low, Dempster's combination rule is used for fusion: T = Π(evidence) set ) / (1 -conflict); Among them, evidence set Let be the set of evidence, and conflict be the conflict coefficient.

9. The turbine condition monitoring and early warning method based on MREMD-DS fusion according to claim 1, characterized in that: In step S6, the formula for calculating the stability score is: Score i =100-P v -P t - P p ; Among them, P v To deduct points for volatility, P t Deduct points for trend instability, P p Points will be deducted for extreme values.

10. A turbine condition monitoring and early warning method based on MREMD-DS fusion according to claim 9, characterized in that: In step S6, the formula for calculating the dynamic early warning threshold is: Warning threshold = max(50, local _mean - 2×local std ); Among them, local mean The mean score for the local window. std The standard deviation is denoted as .

Images