A method for dividing the vibration region of a hydro-turbine based on vibration amplitude weighted K-means clustering
By using K-means clustering based on vibration amplitude, the vibration zone of the turbine can be quickly identified, which solves the difficulties of dynamic identification and adjustment in traditional methods. This achieves efficient and flexible vibration zone division, ensuring unit safety and equipment life.
Patent Information
- Application Number
- CN202610821980.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-09
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-06-09
AI Technical Summary
Traditional methods are difficult to quickly and dynamically identify the vibration zone of a hydro turbine, which makes it impossible for the power station to update in a timely manner under frequently changing load conditions, affecting the unit's adjustment flexibility and equipment safety. In addition, intelligent models are costly and cannot be adjusted in real time.
A weighted K-means clustering method based on vibration amplitude is adopted. By acquiring the vibration data of the turbine unit, the relative vibration value and gradient feature value are calculated to construct a weighted feature space. The K-means clustering algorithm is then used to divide the turbine vibration region.
It enables rapid and objective vibration zone division, provides clear data-driven standards, allows for flexible adjustment of feature weights to adapt to power plant requirements, ensures unit safety, and lays the foundation for subsequent models.
Smart Images

Figure CN122364976B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hydraulic turbine vibration technology, and in particular to a method for dividing hydraulic turbine vibration regions based on vibration amplitude weighted K-means clustering. Background Technology
[0002] The integrated development of wind, solar, and hydropower has shifted hydropower from a traditional, stable generation mode to one that frequently and rapidly participates in grid peak shaving, frequency regulation, and backup response. This shift results in generating units frequently traversing various load conditions, leading to highly transient vibration characteristics. If vibration zones cannot be accurately and dynamically identified and delineated, power plants will not only sacrifice regulation flexibility and economic value due to conservative operation, but may also experience component fatigue and damage due to units operating within vibration zones, threatening the safety of the power plant.
[0003] Currently, the dynamic identification of vibration zones mainly faces the following problems: traditional fine division methods rely heavily on a large amount of historical stability test data, have long analysis cycles, high subjectivity in manual interpretation, and are difficult to update in a timely manner according to changes in the random group state; traditional methods lack real-time prediction and dynamic tracking capabilities and cannot effectively capture the minute vibration characteristic changes of the unit caused by state drift, component wear, etc.; using intelligent models for division is costly, time-consuming to train, and the model cannot be adjusted in real time according to changes in the random group.
[0004] Therefore, a method for dividing the vibration region of a hydro-turbine based on vibration amplitude weighted K-means clustering is proposed to solve the above problems. Summary of the Invention
[0005] The purpose of this invention is to provide a method for dividing the vibration region of a hydro turbine based on vibration amplitude weighted K-means clustering, so as to solve the problems mentioned in the background art.
[0006] This invention is achieved through the following technical solution: A method for dividing the vibration region of a hydro-turbine based on vibration amplitude weighted K-means clustering includes the following steps: Step S1: Obtain vibration data of the turbine unit under multiple steady-state operating conditions. The data shall include at least the power corresponding to each operating condition and the vibration amplitude of multiple measuring points. Step S2: Based on the preset vibration allowable standard, convert the measured vibration amplitude of each measuring point into a dimensionless relative vibration value; Step S3: For each working condition, calculate the maximum relative vibration value, the average relative vibration value, and the vibration gradient characteristic value under that working condition based on the relative vibration values of all measuring points. The vibration gradient characteristic value is the maximum or average value of the gradient value of the vibration of each measuring point as a function of power. Step S4: Standardize the maximum relative vibration value, average relative vibration value, and vibration gradient characteristic value under each working condition to eliminate the differences in dimensions and orders of magnitude between different characteristics; Step S5: Assign weight coefficients to the three standardized features respectively, and construct a weighted feature space based on the weighted distance metric; Step S6: In the weighted feature space, a clustering algorithm is used to divide all operating points into three clusters, corresponding to the stable operating area, the restricted operating area, and the vibration operating area, respectively. Step S7: Calculate the average relative vibration value of the operating points contained in each cluster, and identify each cluster as a vibration zone, a restricted zone, and a stable zone in order of the average value from high to low, thereby completing the division of the turbine operating area.
[0007] Furthermore, the raw data obtained in step S1 is used to form a matrix. D m×(n+2) ,in m For operating conditions, n To represent the number of measurement points, the first column of the matrix is the operating condition number, the second column is the power value, and the third column is... n+2 The vibration amplitude values at each measuring point are listed below.
[0008] Furthermore, the formula for converting the measured vibration amplitude into a relative vibration value in step S2 is as follows:
[0009] The maximum allowable value is derived from the vibration allowable limit of the corresponding measuring point in the national or industry standard.
[0010] Furthermore, in step S3: The formula for calculating the maximum relative vibration value is:
[0011] in, Represents the index of operating points. Represents the measurement point index. For the first Under the first working condition The relative vibration values at each measuring point; The formula for calculating the average relative vibration value is:
[0012] The vibration gradient eigenvalue is calculated as follows: First, calculate the eigenvalue of each measuring point. Vibration gradient values between adjacent operating conditions:
[0013] in, For the first The power values at each operating point are then taken, and all measuring points are recorded. The maximum or average value of the gradient is taken as the value of this working condition. The vibration gradient characteristic value; for boundary conditions, forward difference or backward difference is used for calculation.
[0014] Furthermore, the standardization process in step S4 adopts... Standardization specifically includes: Calculate the sample mean for each feature: ; Calculate the sample standard deviation for each feature: ; Calculate the standardized eigenvalues: ; in, The total number of operating points. For the first The first working condition point 1 eigenvalue, =1,2,3 correspond to the maximum relative vibration value, the average relative vibration value, and the vibration gradient characteristic value, respectively.
[0015] Furthermore, the construction of the weighted feature space in step S5 specifically involves defining weight coefficients. , , These correspond to the standardized maximum relative vibration value, average relative vibration value, and vibration gradient characteristic value, respectively; the weighted Euclidean distance between the characteristic vectors of any two operating points is defined as...
[0016] in, , , These are the standardized feature values.
[0017] Furthermore, the clustering algorithm in step S6 is K-means clustering, specifically including: Three cluster centers are randomly initialized in the weighted feature space; Calculate the weighted Euclidean distance from the feature vector of each operating point to the center of each cluster, and assign the operating point to the nearest cluster; Recalculate the characteristic mean of all operating points in each cluster, and use it as the updated cluster center; Repeat the above allocation and update steps until the change in the position of the cluster center is less than the preset threshold or the maximum number of iterations is reached.
[0018] Furthermore, the formula for calculating the weighted Euclidean distance is as follows:
[0019] in, For the first Standardized weighted feature vectors of each working condition point For the first The center vector of each cluster.
[0020] Furthermore, the robustness processing mechanism in step S6 includes: repeatedly running the clustering algorithm using different random seeds, selecting the clustering result with the smallest sum of squared errors within the cluster from the multiple running results as the final partition, and detecting whether there is an empty cluster during each clustering process. If so, the cluster center is reinitialized.
[0021] Furthermore, the method for calculating the average relative vibration value of each cluster in step S7 is as follows: for each operating point within a cluster, take the average relative vibration value calculated in step S3. Then, for all operating points within that cluster... Calculate the arithmetic mean to obtain the average relative vibration value of the cluster; sort the clusters by value from high to low, identify the cluster with the highest average value as the vibration region, the cluster with the second highest average value as the restricted region, and the cluster with the lowest average value as the stable region.
[0022] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: 1. This invention directly extracts features from the stability data monitored in real time and performs cluster analysis using three quantified features, avoiding the high subjectivity of traditional methods and providing a clear and objective data-driven standard for the division of vibration areas.
[0023] 2. The method proposed in this invention has a clear process and is computationally efficient. Through standardized numbers, feature calculation and cluster analysis, it can quickly output the results of unit vibration zone division without going through a lengthy model training and optimization cycle.
[0024] 3. This invention introduces a feature weighting step before clustering, which allows for flexible adjustment of the weights of different features according to the actual situation of the power plant, so that the partitioning results can meet the actual needs of the power plant.
[0025] 4. This invention can quickly and automatically divide operating conditions into stable zones, restricted zones, and vibration zones. The vibration zone division results of this invention have clear engineering guidance significance. On the one hand, it can be intuitively used to guide scheduling operations, avoid high-risk operating conditions, and ensure unit safety and equipment lifespan. On the other hand, the division results provided can serve as high-quality, labeled initial samples, laying the foundation for building more refined prediction models in the future. Attached Figure Description
[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0027] Figure 1 This is a schematic diagram of the overall process of a method for dividing the vibration region of a water turbine based on vibration amplitude weighted K-means clustering, which is provided by the present invention.
[0028] Figure 2 This is a schematic diagram of the clustering results for the vibration zone division of a power station in an embodiment of the present invention.
[0029] Figure 3 This is a schematic diagram of the preliminary prediction and division of the vibration zone of a power station in an embodiment of the present invention.
[0030] Figure 4 This is a schematic diagram of the vibration zone division results when changing the clustering weights in an embodiment of the present invention.
[0031] Figure 5 This is a schematic diagram of the vibration zone division results after clustering data from three water heads at a power station in an embodiment of the present invention. Detailed Implementation
[0032] To make the objectives, technical solutions, and advantages of the present invention more apparent, exemplary embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of the present invention, and not all of the embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein. Based on the embodiments of the present invention described herein, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of the present invention.
[0033] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.
[0034] It should be understood that the invention can be embodied in various forms and should not be construed as being limited to the embodiments set forth herein. Rather, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
[0035] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising” and / or “including,” when used in this specification, identify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.
[0036] To fully understand this invention, a detailed structure will be presented in the following description to illustrate the technical solution proposed by this invention. Optional embodiments of the invention are described in detail below; however, in addition to these detailed descriptions, the invention may have other embodiments.
[0037] This embodiment uses field stability test data from a mixed-flow turbine unit at a power station at a head of 112.48 meters as an example to illustrate the method of the present invention in detail. Based on long-term operating experience, the power station initially divided the unit's operating area into two roughly defined zones: a prohibited operating zone (0MW-450MW) and a stable operating zone (450MW-600MW). However, this division is too coarse and cannot reflect the differences in the actual vibration characteristics of the unit within the 300MW-450MW range, thus limiting its guidance for dispatching operations. Furthermore, each measuring point only has a single data point under each operating condition, insufficient to train complex intelligent models such as neural networks or support vector machines. Therefore, this embodiment employs the method of the present invention to quickly and objectively refine the vibration zone division based on a small amount of measured data.
[0038] See Figures 1-5 A method for dividing the vibration region of a hydro-turbine based on vibration amplitude weighted K-means clustering includes the following steps: Step S1: Obtain vibration data of the turbine unit under multiple steady-state operating conditions. The data shall include at least the power corresponding to each operating condition and the vibration amplitude of multiple measuring points. The raw data obtained in step S1 forms a matrix. D m×(n+2) ,in m For operating conditions, n To represent the number of measurement points, the first column of the matrix is the operating condition number, the second column is the power value, and the third column is... n+2 The vibration amplitude values at each measuring point are listed below.
[0039] For example, vibration data of the unit was collected during stable operation under 20 different power conditions. The power conditions were: 30MW, 60MW, 90MW, 120MW, 150MW, 180MW, 210MW, 240MW, 270MW, 300MW, 330MW, 360MW, 390MW, 420MW, 450MW, 480MW, 510MW, 540MW, 570MW, and 600MW. Under each condition, the peak vibration (unit: μm) was measured at the following 10 measuring points: upper guide X, upper guide Y, water guide X, water guide Y, upper frame horizontal, upper frame vertical, lower frame horizontal, lower frame vertical, and top cover + Y horizontal. The raw data are shown in Table 1.
[0040] Table 1
[0041] The above data is organized into an original data matrix. The first column is the operating condition number, the second column is the power value, and the third to 12th columns are the vibration peak values at each measuring point.
[0042] Step S2: Based on the preset vibration allowable standard, convert the measured vibration amplitude of each measuring point into a dimensionless relative vibration value; The formula for converting the measured vibration amplitude into a relative vibration value in step S2 is as follows:
[0043] The maximum allowable value is derived from the vibration allowable limit of the corresponding measuring point in the national or industry standard.
[0044] For example, based on the vibration allowable limits of the corresponding measuring points in the national standards GB / T 15468 Basic Technical Conditions for Hydropower Turbines and GB / T 17189 Field Test Procedures for Vibration and Pulsation of Hydraulic Machinery, the standard allowable maximum values for each measuring point are determined as follows: the allowable limit for horizontal and vertical movement of the upper frame is 110 μm, the allowable limit for horizontal and vertical movement of the lower frame is 100 μm, the allowable limit for horizontal movement of the top cover is 120 μm, and the allowable limit for runout of the upper guide and water guide bearings is 200 μm (because some runout values in the original data exceed 200 μm, 240 μm is actually taken as the benchmark according to the procedure. This embodiment is a demonstration method, and the measured maximum value of each measuring point is used as the normalization benchmark, but the standard method should be based on the value specified in the national standard). For the sake of simplicity, this embodiment uniformly uses the maximum measured value of each measuring point in all operating conditions as the standard allowable maximum value (the national standard value should be used in actual applications). Taking a horizontal measuring point (56μm) on the rack under 30MW operating conditions as an example, the relative vibration value = 56 / 110 ≈ 0.509. The same conversion is performed on all measuring points under all operating conditions to obtain the relative vibration value matrix. .
[0045] Step S3: For each working condition, calculate the maximum relative vibration value, the average relative vibration value, and the vibration gradient characteristic value under that working condition based on the relative vibration values of all measuring points. The vibration gradient characteristic value is the maximum or average value of the gradient value of the vibration of each measuring point as a function of power. In step S3: The formula for calculating the maximum relative vibration value is:
[0046] in, Represents the index of operating points. Represents the measurement point index. For the first Under the first working condition The relative vibration values at each measuring point; The formula for calculating the average relative vibration value is:
[0047] The vibration gradient eigenvalue is calculated as follows: First, calculate the eigenvalue of each measuring point. Vibration gradient values between adjacent operating conditions:
[0048] in, For the first The power values at each operating point are then taken, and all measuring points are recorded. The maximum or average value of the gradient is taken as the value of this working condition. The vibration gradient characteristic value; for boundary conditions, forward difference or backward difference is used for calculation.
[0049] For example, for each working condition Based on the relative vibration values of 10 measuring points Calculate three features: (1) Maximum relative vibration value
[0050] For example, under the 30MW operating condition, the largest relative value among all measuring points is the water conductance Y (1.325), therefore... =1.325.
[0051] (2) Average relative vibration value
[0052] Under 30MW operating conditions, the sum of the relative values at the 10 measuring points is approximately: 0.7+0.695+1.225+1.325+0.509+0.6+0.22+0.6+0.317=6.191, with an average value of 0.619.
[0053] (3) Vibration gradient characteristic value First, calculate the value for each measuring point. Vibration gradient values between adjacent operating conditions:
[0054] in For the first Power values for each operating condition (unit: MW).
[0055] For boundary conditions Forward difference is used; Backward difference is used.
[0056] Then, take all measuring points. The maximum value of the gradient (the largest absolute value, reflecting the most drastic vibration change) is taken as the value for this working condition. Vibration gradient eigenvalues The reason for choosing the maximum value rather than the average value is that the division of the vibration zone of the water turbine focuses more on local abrupt changes, while the average value may mask dangerous changes at individual measuring points.
[0057] For example, calculate the gradient of the water conductance Y measurement point between 30MW and 60MW operating conditions: The absolute value is 0.00617.
[0058] For each working condition, calculate the absolute value of the gradient at all measuring points, and take the maximum value as the gradient for that working condition. Calculations show that the gradient eigenvalues for each working condition are typically on the order of magnitude of... The magnitude is much smaller than and (Both are usually between 0 and 1.5).
[0059] The eigenvalue matrix of 20 working conditions was obtained. Each row corresponds to a working condition, and the three columns are respectively , , .
[0060] Step S4: Standardize the maximum relative vibration value, average relative vibration value, and vibration gradient characteristic value under each working condition to eliminate the differences in dimensions and orders of magnitude between different characteristics; The standardization process in step S4 adopts... Standardization specifically includes: Calculate the sample mean for each feature: ; Calculate the sample standard deviation for each feature: ; Calculate the standardized eigenvalues: ; in, The total number of operating points. For the first The first working condition point 1 eigenvalue, =1,2,3 correspond to the maximum relative vibration value, the average relative vibration value, and the vibration gradient characteristic value, respectively.
[0061] For example, since the vibration gradient feature value is much smaller than the first two features, direct clustering would cause the gradient feature to be ignored. Therefore, clustering is performed on the three columns separately. standardization: Calculate the sample mean for each column. and sample standard deviation ( =1,2,3): ,
[0062] Then calculate the standardized eigenvalues: .
[0063] After standardization, the mean of all three features is 0, and the standard deviation is 1, eliminating differences in dimensions and orders of magnitude. This yields the standardized feature matrix. .
[0064] Step S5: Assign weight coefficients to the three standardized features respectively, and construct a weighted feature space based on the weighted distance metric; The construction of the weighted feature space in step S5 specifically involves defining weight coefficients. , , These correspond to the standardized maximum relative vibration value, average relative vibration value, and vibration gradient characteristic value, respectively; the weighted Euclidean distance between the characteristic vectors of any two operating points is defined as...
[0065] in, , , These are the standardized feature values.
[0066] For example, an equal-weight scheme is adopted, that is = = =1 indicates that the maximum vibration, average vibration, and vibration gradient are given equal importance. The impact of weight adjustment on the results will be shown later. A weighted feature matrix is constructed. This is equivalent to direct use. Perform clustering.
[0067] Step S6: In the weighted feature space, a clustering algorithm is used to divide all operating points into three clusters, corresponding to the stable operating area, the restricted operating area, and the vibration operating area, respectively. The clustering algorithm in step S6 is K-means clustering, specifically including: Three cluster centers are randomly initialized in the weighted feature space; Calculate the weighted Euclidean distance from the feature vector of each operating point to the center of each cluster, and assign the operating point to the nearest cluster; Recalculate the characteristic mean of all operating points in each cluster, and use it as the updated cluster center; Repeat the above allocation and update steps until the change in the position of the cluster center is less than the preset threshold or the maximum number of iterations is reached.
[0068] The formula for calculating the weighted Euclidean distance is:
[0069] in, For the first Standardized weighted feature vectors of each working condition point For the first The center vector of each cluster.
[0070] The robust processing mechanism in step S6 includes: repeatedly running the clustering algorithm using different random seeds, selecting the clustering result with the smallest sum of squared errors within the cluster from the multiple running results as the final partition, and detecting whether there is an empty cluster during each clustering process. If there is, the cluster center is reinitialized.
[0071] For example, we set the number of clusters K=3, corresponding to the stable region, the restricted region, and the oscillating region, respectively. The algorithm flow is as follows: (1) Randomly initialize three cluster centers in the three-dimensional weighted feature space. , , .
[0072] (2) For each operating point Calculate the weighted Euclidean distance from each cluster center: .
[0073] (3) Assign the operating point to the nearest cluster.
[0074] (4) After the allocation is completed, the mean of all points in each cluster is recalculated and used as the new cluster center.
[0075] (5) Repeat steps (2)-(4) until the change in cluster center is less than the preset threshold (e.g. (or reach the maximum number of iterations (set to 100 in this embodiment).
[0076] To prevent getting trapped in local optima, a random seed of 100 is set, and the process is repeated 10 times, selecting the result with the smallest sum of squared errors (SSE) within the cluster. Simultaneously, in each iteration, the existence of an empty cluster is checked; if one exists, the cluster center is reinitialized.
[0077] Clustering results as follows Figure 2 As shown. Figure 2 This is a three-dimensional scatter plot, with the coordinate axes representing the standardized maximum relative vibration value, average relative vibration value, and vibration gradient characteristic value, respectively. The plot uses different shapes (or colors) to mark three clusters of operating points, showing that the operating points are naturally separated in space.
[0078] Step S7: Calculate the average relative vibration value of the operating points contained in each cluster, and identify each cluster as a vibration zone, a restricted zone, and a stable zone in order of the average value from high to low, thereby completing the division of the turbine operating area.
[0079] The method for calculating the average relative vibration value of each cluster in step S7 is as follows: For each operating point within a cluster, take the average relative vibration value calculated in step S3. Then, for all operating points within that cluster... Calculate the arithmetic mean to obtain the average relative vibration value of the cluster; sort the clusters by value from high to low, identify the cluster with the highest average value as the vibration region, the cluster with the second highest average value as the restricted region, and the cluster with the lowest average value as the stable region.
[0080] For example, calculate the average relative vibration value of all operating points within each cluster (i.e., the average vibration value of each operating point). The average value within each cluster). The calculated average relative vibration values of the three clusters are as follows: Cluster A: Average =0.812 (highest) Cluster B: Average =0.465 (Medium) Cluster C: Average =0.287 (lowest) Based on the order from highest to lowest, cluster A is identified as the vibration region, cluster B as the restricted region, and cluster C as the stable region.
[0081] The original power operating conditions are labeled according to their respective regions, resulting in the following divisions: Vibration zones (8 operating conditions): 30MW, 60MW, 90MW, 120MW, 150MW, 180MW, 390MW, 420MW.
[0082] Restricted areas (2 operating conditions): 300MW, 330MW.
[0083] Stable range (10 operating conditions): 210MW, 240MW, 270MW, 450MW, 480MW, 510MW, 540MW, 570MW, 600MW.
[0084] The segmentation results are presented in the form of a line chart, such as... Figure 3 As shown. Figure 3 The horizontal axis represents power (MW), and the vertical axis represents the average relative vibration value under each operating condition. The vibration zone (red), restricted zone (yellow), and stable zone (green) are marked with different background colors. It can be seen that the operating points within the vibration zone... All values are above 0.6, the stable zone values are below 0.4, and the restricted zone values are between 0.4 and 0.6. This classification is much more refined than the original binary classification of "0-450MW / 450-600MW", and can guide operators to avoid the unexpected vibration zone of 390-420MW (which is located within the originally considered stable zone) during actual scheduling, thereby effectively protecting the units.
[0085] In practical engineering, different power plants may have varying degrees of concern regarding different characteristics. For example, one power plant may be more concerned with the average vibration level of the unit (reflecting long-term fatigue damage) and less sensitive to instantaneous maximum vibration (which may be caused by accidental factors). Therefore, the weighting coefficient is adjusted to... =0.5 (maximum vibration weight reduced) =1.5 (average vibration weight increased) =1 (gradient invariant). Repeat steps S5-S7.
[0086] After re-clustering, the cluster affiliation of each working condition changed. The revised partitioning results are shown in Table 2.
[0087] Table 2
[0088] As shown in Table 2, after weight adjustment, the 120-150MW range was excluded from the vibration zone due to its relatively low average vibration value, while the 300-330MW range was included in the vibration zone due to its higher average vibration value. This result is as follows... Figure 4 As shown, this demonstrates that the present invention can flexibly adapt to the individual needs of power plants and provide customized operating area divisions.
[0089] To verify the adaptability of this invention under different operating heads, stability test data of the unit at two other heads (105 meters and 120 meters) were collected (data format is the same as in Table 1, but not listed due to space limitations). Clustering was performed according to the above steps (equal weight scheme) to obtain the vibration zone boundaries at the three heads as shown below. Figure 5 As shown, the values correspond to water heads of 105 meters, 112.48 meters, and 120 meters, respectively. The horizontal axis represents power (MW), and the vertical axis represents the average relative vibration value. The shaded area represents the vibration zone. The results show that at a water head of 105 meters, the vibration zone is mainly concentrated in the 0-350MW and 420-500MW range; at a water head of 112.48 meters (in this embodiment), the vibration zone is 0-200MW and 390-420MW; at a water head of 120 meters, the vibration zone shrinks to 0-150MW, and a new small-scale vibration zone appears above 500MW.
[0090] The results show that the method of the present invention can quickly and objectively output the vibration zone distribution under different water heads, providing a reliable basis for multi-condition scheduling of power plants, and does not require separate model training for each water head.
[0091] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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 method for dividing the vibration region of a hydro-turbine based on vibration amplitude weighted K-means clustering, characterized in that, Includes the following steps: Step S1: Obtain vibration data of the turbine unit under multiple steady-state operating conditions. The data shall include at least the power corresponding to each operating condition and the vibration amplitude of multiple measuring points. Step S2: Based on the preset vibration allowable standard, convert the measured vibration amplitude of each measuring point into a dimensionless relative vibration value; Step S3: For each working condition, calculate the maximum relative vibration value, the average relative vibration value, and the vibration gradient characteristic value under that working condition based on the relative vibration values of all measuring points. The vibration gradient characteristic value is the maximum or average value of the gradient value of the vibration of each measuring point as a function of power. The formula for calculating the maximum relative vibration value is as follows: in, Represents the index of operating points. Represents the measurement point index. For the first Under the first working condition The relative vibration values at each measuring point; The formula for calculating the average relative vibration value is: The vibration gradient eigenvalue is calculated as follows: First, calculate the eigenvalue of each measuring point. Vibration gradient values between adjacent operating conditions: in, For the first The power values at each operating point are then taken, and all measuring points are recorded. The maximum or average value of the gradient is taken as the value of this working condition. The vibration gradient characteristic value; for boundary condition points, forward difference or backward difference is used for calculation; Step S4: Standardize the maximum relative vibration value, average relative vibration value, and vibration gradient characteristic value under each working condition to eliminate the differences in dimensions and orders of magnitude between different characteristics; Step S5: Assign weight coefficients to the three standardized features respectively, and construct a weighted feature space based on the weighted distance metric; Specifically, constructing the weighted feature space involves defining the weight coefficients. , , These correspond to the standardized maximum relative vibration value, average relative vibration value, and vibration gradient characteristic value, respectively; the weighted Euclidean distance between the characteristic vectors of any two operating points is defined as... in, , , These are the standardized eigenvalues; Step S6: In the weighted feature space, a clustering algorithm is used to divide all operating points into three clusters, corresponding to the stable region, the restricted region, and the vibration region, respectively; The clustering algorithm used is K-means clustering, which specifically includes: Three cluster centers are randomly initialized in the weighted feature space; Calculate the weighted Euclidean distance from the feature vector of each operating point to the center of each cluster, and assign the operating point to the nearest cluster; Recalculate the characteristic mean of all operating points in each cluster, and use it as the updated cluster center; Repeat the above allocation and update steps until the change in the position of the cluster center is less than the preset threshold or the maximum number of iterations is reached; Step S7: Calculate the average relative vibration value of the operating points contained in each cluster, and identify each cluster as a vibration zone, a restricted zone, and a stable zone in order of the average value from high to low, thereby completing the division of the turbine operating area.
2. The method according to claim 1, characterized in that, The raw data obtained in step S1 forms a matrix. D m×(n+2) ,in m For operating conditions, n To represent the number of measurement points, the first column of the matrix is the operating condition number, the second column is the power value, and the third column is... n+2 The vibration amplitude values at each measuring point are listed below.
3. The method according to claim 1, characterized in that, The formula for converting the measured vibration amplitude into a relative vibration value in step S2 is as follows: The maximum allowable value is derived from the vibration allowable limit of the corresponding measuring point in the national or industry standard.
4. The method according to claim 1, characterized in that, The standardization process in step S4 adopts... Standardization specifically includes: Calculate the sample mean for each feature: ; Calculate the sample standard deviation for each feature: ; Calculate the standardized eigenvalues: ; in, The total number of operating points. For the first The first working condition point 1 eigenvalue, =1,2,3 correspond to the maximum relative vibration value, the average relative vibration value, and the vibration gradient characteristic value, respectively.
5. The method according to claim 1, characterized in that, The formula for calculating the weighted Euclidean distance is: in, For the first Standardized weighted feature vectors of each working condition point For the first The center vector of each cluster.
6. The method according to claim 1, characterized in that, The robust processing mechanism in step S6 includes: repeatedly running the clustering algorithm using different random seeds, selecting the clustering result with the smallest sum of squared errors within the cluster from the multiple running results as the final partition, and detecting whether there is an empty cluster during each clustering process. If there is, the cluster center is reinitialized.
7. The method according to claim 1, characterized in that, The method for calculating the average relative vibration value of each cluster in step S7 is as follows: For each operating point within a cluster, take the average relative vibration value calculated in step S3. Then, for all operating points within that cluster... Calculate the arithmetic mean to obtain the average relative vibration value of the cluster; sort the clusters by value from high to low, identify the cluster with the highest average value as the vibration zone, the cluster with the second highest average value as the restricted zone, and the cluster with the lowest average value as the stable zone.
Citation Information
Patent Citations
K-means three-dimensional clustering algorithm-based wind pressure coefficient rapid partitioning method and system and storage medium
CN112487720A
Method for dividing operation vibration area of giant water-turbine generator set
CN118820786A
