A wind turbine tower intelligent state monitoring system with autonomous learning characteristics

Abstract

The method discloses a monitoring and diagnosis system for a wind turbine tower. The method collects information of the wind turbine tower from multiple dimensions, constructs a system architecture with information collection, feature extraction, intelligent diagnosis and autonomous learning functions, generates a complete set of diagnosis standards for the operation state of the wind turbine tower, improves the timeliness and accuracy of the fault diagnosis of the wind turbine tower, reduces the influence of wind load on the fault diagnosis of the tower, and realizes the adaptive diagnosis strategy of the wind turbine tower.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation technology and relates to a method for monitoring the condition of wind turbine towers. Background Technology

[0002] With the expansion of the new energy market and the continuous growth of wind power installed capacity, wind power projects are gradually developing in inland Class III and IV regions. Wind turbine towers and blades are becoming increasingly taller and longer, leading to frequent major accidents such as tower tilting and collapse, posing a serious threat to the safe production of power generation companies. The tower is a crucial component of the wind power industry chain, supporting the entire wind turbine and absorbing its vibrations. Wind turbine tower damage has attracted significant attention both domestically and internationally, making tower damage monitoring methods a hot research topic in the engineering field in recent years. However, existing tower monitoring methods still have some problems. First, there is a lack of relatively mature judgment standards for diagnosing tower conditions. Second, single-factor failures may be influenced by other factors, making single-factor monitoring unsuitable for timely detection of tower faults. Third, the degree of tower fault presentation varies under different wind loads, hindering fault diagnosis. Finally, current tower maintenance operates in a fault-triggered mode, preventing timely and effective updates to tower diagnostic methods, which lack self-learning capabilities. Summary of the Invention

[0003] This method proposes a monitoring and diagnosis system for wind turbine towers. By monitoring the bolt status, tower status, and wind load status of the wind turbine, the system constructs an architecture with information acquisition, feature extraction, intelligent diagnosis, and autonomous learning functions. This improves the timeliness and accuracy of wind turbine tower fault diagnosis, reduces the impact of wind load on tower fault diagnosis, and realizes an adaptive diagnosis strategy for wind turbine towers.

[0004] As shown in Figure 1, this system consists of four layers: wind field information acquisition layer, feature database, intelligent diagnosis layer, and autonomous learning layer.

[0005] The wind farm information acquisition layer is responsible for collecting bolt status information, wind load information, and tower status information of wind turbine generators. Bolt status information includes the preload of bolts fixing the first, second, third, and fourth floors of the tower; wind load information includes the wind vibration coefficient at height z, wind pressure height variation coefficient, wind load shape coefficient, and basic wind pressure; tower status information includes the average tower verticality over a period of time, the average effective value of nacelle sway over a period of time, and the average effective value of tower base tilt angle over a period of time.

[0006] The feature database is responsible for extracting features from the collected bolt and tower status information to obtain feature factors. The bolt status feature fusion module fuses the collected bolt status information to obtain bolt feature factors, which are then input as new data into the intelligent diagnostic layer. Similarly, the tower status feature fusion module fuses the collected tower status information to obtain tower feature factors, which are also input as new data into the intelligent diagnostic layer.

[0007] The intelligent diagnostic layer is responsible for extracting diagnostic factors for the operating status of wind turbine towers and generating diagnostic standards for the operating status of wind turbine towers. First, the collected wind load information is input into the tower dynamics feature estimation module to obtain the theoretical bolt characteristic function and tower characteristic function under the same wind load information. Second, the theoretical bolt characteristic function is input into the bolt self-diagnosis module, and correlation analysis is performed with the bolt characteristic factors to obtain bolt statistical characteristic parameters. A bolt state diagnosis interval is constructed, and coordinate transformation is performed on the bolt state diagnosis interval to obtain bolt state diagnosis factors. The bolt state diagnosis factors are then input into the state factor module. Third, the theoretical tower characteristic function is input into the tower self-diagnosis module, and correlation analysis is performed with the tower characteristic factors to obtain tower statistical characteristic parameters. A tower state diagnosis interval is constructed, and coordinate transformation is performed on the tower state diagnosis interval to obtain tower state diagnosis factors. The tower state diagnosis factors are then input into the state factor module. Finally, the state factor module fuses the obtained bolt state diagnosis factors and tower state diagnosis factors to obtain wind turbine tower operating state diagnosis factors, generating tower state diagnosis standards for the entire wind farm. These diagnostic standards are then input as new data into the autonomous learning layer.

[0008] The autonomous learning layer is responsible for autonomously iteratively learning the diagnostic standards for the operating status of wind turbine towers. Based on the standards obtained from the intelligent diagnostic layer, the tower status is evaluated, and the on-site maintenance feedback results are compared with the diagnostic standards. First, the status diagnosis interval error is calculated, and digital feedback is achieved by finding the error boundary value. Second, the statistical characteristic parameters of the status change factors are calculated, and status feedback is achieved by updating the status change factors. The results of the digital feedback and status feedback are input into the intelligent diagnostic layer to realize the iterative learning of the diagnostic system. Attached Figure Description

[0009] Figure 1. Intelligent condition monitoring system for wind turbine towers with autonomous learning capabilities. Detailed Implementation

[0010] The wind farm information acquisition layer collects bolt status information, wind load status information, and tower status information for wind turbine towers. Bolt status information includes the preload of bolts fixing the first-floor tower (F1), the second-floor tower (F2), the third-floor tower (F3), and the fourth-floor tower (F4). Tower status information includes the average verticality value (b1) of the tower within time t, the average effective value of nacelle sway (b2) within time t, and the average effective value of tower base tilt angle (b3) within time t. The wind load status information is presented in the following format:

[0011] C = β z μ s μ z ω0

[0012] In the formula

[0013] C is the wind load factor;

[0014] β z Let z be the wind vibration coefficient at height z;

[0015] μ s This is the wind pressure height variation coefficient;

[0016] μ z This is the wind load shape coefficient;

[0017] ω0 is the basic wind pressure, approximately 600 N / m. 2 .

[0018] Bolt status information and tower status information are input into the feature database, and feature fusion is performed on the bolt status information. The specific form of this fusion is as follows:

[0019]

[0020] In the formula

[0021] A is the bolt characteristic factor;

[0022] f represents the data obtained after fusing the bolt status information of the s types;

[0023] ω 1s The weight of the s-th type of bolt state information data;

[0024] σ1 2 The variance after fusing the bolt status information data of type s;

[0025] x1 is the actual value after fusing the bolt status information data of type s.

[0026] The variance σ after determining the data fusion of the s-th type of bolt state information is... 1s 2When the mean of the s-th bolt state information is used as the unbiased estimate of the fused data, the variance of the estimated s-th bolt state information data is denoted as . ω 1s Calculate using the following formula:

[0027]

[0028] The fusion of tower status information is specifically represented as follows:

[0029]

[0030] In the formula

[0031] B is the tower characteristic factor;

[0032] b1 is the average verticality of the tower within time t, b2 is the average effective value of the nacelle sway within time t, and b3 is the average effective value of the tower base tilt angle within time t.

[0033] d1 is the radius of the center of the tower cylinder;

[0034] d2 is the radius of the lower center of the tower;

[0035] h is the height of the tower;

[0036] b represents the data obtained after fusing the s types of tower state information;

[0037] ω 2s The weight of the s-th tower state information data;

[0038] σ2 2 The variance of the fused tower condition information data of type s;

[0039] x2 is the true value after fusing the s types of tower status information data.

[0040] The variance σ after determining the s-th type of tower state information data fusion is... 2s 2 When the mean of the s-th tower state information is used as the unbiased estimate of the fused data, the variance of the estimated s-th tower state information data is denoted as . ω 2s Calculate using the following formula:

[0041]

[0042] By inputting the wind load factor into the tower dynamics characteristic estimation, the bolt dynamics equation f(x) and the tower dynamics equation g(x) are obtained. Under the same wind load factor C, the bolt factor A has a theoretical value. Right now The tower factor B has a theoretical value. Right now Factors A and B obtained from the feature database were input into the intelligent diagnostic layer for experimentation. The results are as follows:

[0043]

[0044] In the formula

[0045] A i Let A be the i-th factor;

[0046] B j Let B be the j-th factor;

[0047]

[0048] σ a 2 For factor A i The variance;

[0049] σ b 2 For factor B j The variance;

[0050] Based on the obtained statistical characteristic parameter μ a σ a 2 Statistical characteristic parameter μ b σ b 2 We obtain three state intervals:

[0051]

[0052] In the formula

[0053] Q1 represents interval one, Q2 represents interval two, and Q3 represents interval three;

[0054] l represents factors a and b;

[0055] p l1 This represents the probability parameter of factor l in the normal state.

[0056] p l2 This represents the probability parameter of the warning state for factor l;

[0057] p l3 This represents the probability parameter of alarm status for factor l;

[0058] When l = a, μ is the bolt characteristic factor. l =μ a , σ l 2 =σ a 2 p l1 =p a1p l2 =p a2 p l3 =p a3 When A i When ∈Q1, the bolt characteristic factor is in a normal state, when A i When ∈Q2, the bolt characteristic factor is in an early warning state, when A i When ∈Q3, the bolt characteristic factor is in an alarm state;

[0059] When l = b, μ is the characteristic factor of the tower. l =μ b , σ l 2 =σ b 2 p l1 =p b1 p l2 =p b2 p l3 =p b3 When B j When ∈Q1, the tower characteristic factor is in a normal state, when B j When ∈Q2, the tower characteristic factor is in an early warning state, when B j When ∈Q3, the tower characteristic factor is in an alarm state.

[0060] Based on the obtained state interval, the state diagnostic factor P is also obtained. l :

[0061]

[0062] In the formula

[0063] When l = a, the bolt condition diagnostic factor P is... a ;

[0064] When l = b, the tower condition diagnostic factor P is... b .

[0065] The obtained state diagnostic factors are simulated and integrated to obtain the operating state diagnostic factor P:

[0066]

[0067] In the formula

[0068] p 11 The probability of maintaining the tower's normal condition;

[0069] p 12 This represents the probability that the tower's status changes from normal to warning.

[0070] p 13 This represents the probability that the tower's status changes from normal to alarm.

[0071] p 22 The probability of maintaining an early warning for the tower's condition;

[0072] p 23 This represents the probability that the tower's status changes from a warning to an alarm.

[0073] The obtained state transition factor P is used as the tower's operating state diagnostic factor. If the rate of change of the operating diagnostic factor exceeds M%, the tower's state is considered to have changed.

[0074] The autonomous learning layer is responsible for evaluating and verifying the obtained tower condition diagnosis criteria, collecting feedback results from on-site maintenance, and outputting the obtained condition diagnosis factors and factor change rates as tower condition diagnosis criteria if the judgment is accurate; if the judgment is inaccurate, the intelligent diagnosis layer is iteratively updated through condition feedback and digital feedback.

[0075] The status feedback implementation process is as follows:

[0076] G1: Collect the rate of change of diagnostic factors with inaccurate judgments (M) g There are m in total;

[0077] G2: Calculate the mean μ of the rate of change of diagnostic factors. g ,Right now G3: Calculate the variance σ of the rate of change of diagnostic factors. g 2 ,Right now G4: Based on the feedback from the site, let M′=M+σ g 2 Or M′=M-σ g 2 This is to determine the rate of change of new diagnostic factors, thereby enabling state feedback.

[0078] The digital feedback implementation process is as follows:

[0079] F1: Collect parameters A that are not accurately judged. i B j ,

[0080] F2: Randomly select from A i Three centers are selected from the data, denoted as μ. f (f = 1, 2, 3);

[0081] F3: Calculate A i objective function F4: Repeat step F2 until the minimum value of the objective function is found. F5: The center values ​​of each cluster are obtained through iterative calculation, and the boundary values ​​H are obtained through reverse calculation. avv = 1, 2, 3, that is, v = 1 indicates the normal state, v = 2 indicates the warning state, and v = 3 indicates the alarm state;

[0082] F6: Based on boundary value H av Iterative state probability parameters, i.e.

[0083] F7: Similarly, parameter B j Steps F2-F6 are performed to obtain the probability parameters after boundary value iteration, thereby realizing digital feedback.

Claims

1. A wind turbine tower intelligent state monitoring system with autonomous learning characteristics, the system comprising four levels, namely a wind farm information acquisition layer, a feature database, an intelligent diagnosis layer and an autonomous learning layer, the wind farm information acquisition layer acquiring bolt state information, wind load state information and tower state information of the wind turbine tower, the bolt state information being first layer tower fixing bolt pretightening force , second layer tower fixing bolt pretightening force , third layer tower fixing bolt pretightening force , and fourth layer tower fixing bolt pretightening force , the tower state information being average value of tower perpendicularity in a time period , being average value of nacelle sway effective value in a time period , being average value of tower foundation inclination effective value in a time period , and the wind load state information having the following specific forms: In the formula wind load factor; Cz is the wind vibration coefficient at height z; where: h - height of the wind pressure variation coefficient; Cm is the shape coefficient for wind load; For the basic wind pressure, take 600 , The bolt state information and the tower tube state information are input to a feature database, the bolt state information is fused, and a specific performance form is as follows: In the formula bolt characteristic factor; For A bolt state information fused data; For the first weight of the bolt status information data; For A bolt state information data fusion variance; For A bolt state information data fused true value, In determining the variance of the bolt state information data after fusion , the mean of the first bolt state information is used as the unbiased estimate of the data after fusion, and the variance of the first bolt state information data is estimated to be , where ​ The following was calculated: ， The tower tube state information is fused, and a specific performance form is as follows: In the formula Tower cylinder characteristic factor; is the average of the tower verticality within the time, is the average of the nacelle sway RMS within the time, is the average of the tower tilt RMS within the time; R is the radius of the circle on the tower drum; The radius of the lower center of the tower; Ht is the tower height; For A tower state information fused data; For the first weight of the tower state information data; For A variance of tower state information data after fusion; For A true value after data fusion of tower state information In determining the variance of the fused data of the tower state information data , the mean value of the tower state information is used as the unbiased estimate of the fused data, and the variance of the estimated tower state information data is denoted as ,​​​ The following formula was used for the calculation: ， The wind load factor is input to the tower drum dynamic characteristic estimation to obtain the bolt dynamic equation and the tower drum dynamic equation Under the same wind load factor , the bolt factor has a theoretical value , that is , the tower drum factor has a theoretical value , that is The factors obtained from the characteristic database , are input to the intelligent diagnosis layer, and experiments are performed, which are shown in the following forms: In the formula for the i-th factor ; for the jth factor ; ； for the factor of variance; for the factor of variance; According to the statistical characteristic parameters obtained , , the statistical characteristic parameters , three state intervals are obtained: In the formula represents interval one, represents interval two, represents interval three; representative factor , ; representative factor normal state probability parameter factor of representation warning state probability parameter factor of representation alarm state probability parameter is a bolt characteristic factor, , , , , when the bolt characteristic factor is in a normal state, when the bolt characteristic factor is in a pre-warning state, and when the bolt characteristic factor is in an alarm state. Time is the characteristic factor of the tower. , , , , when At that time, the tower characteristic factor was in a normal state. At that time, the tower characteristic factor is in an early warning state. At that time, the tower characteristic factor is in alarm mode. According to the obtained state interval, a state diagnosis factor is obtained : In the formula time is a bolt condition diagnostic factor ; time is a tower state diagnostic factor , The obtained state diagnosis factor is simulated and integrated to obtain an operation state diagnosis factor P: In the formula Probability that the tower state remains normal; the probability of the tower state being converted from normal to pre-alarm; Ptrans = probability of transition from normal to alarm for tower status; a probability of a tower section state remaining alert; the probability of the tower state being converted from a pre-alarm to an alarm, The state transition factor obtained As the operating state diagnosis factor of the tower drum, if the operating diagnosis factor change rate exceeds %, it is considered that the state of the tower drum has changed, The autonomous learning layer is responsible for state evaluation and inspection on the obtained tower tube state diagnosis standard, collects feedback results of on-site maintenance, and if the judgment is accurate, the obtained state diagnosis factor and the factor change rate are output as the tower tube state diagnosis standard; if the judgment is not accurate, the intelligent diagnosis layer is iteratively updated through state feedback and digital feedback, The state feedback implementation process is as follows: G1: Collecting a rate of change of a diagnostic factor whose judgment is inaccurate , collectively ; G2: Calculate the mean of the diagnostic factor change rate i.e. ; G3: Calculate variance of diagnostic factor change rate i.e. ; G4: According to the field feedback, let or is the new diagnostic factor change rate, and then realize state feedback, The digital feedback implementation process is as follows: F1 : Collect inaccurate parameters of judgment , , ; F2: randomly select 3 centers from , denoted as ; F3: Compute Objective function ; F4: repeat step F2 until the minimum value of the objective function is found, i.e. ; F5: the center value of each cluster is obtained by iterative calculation, and the boundary value is obtained by back calculation , , i.e. normal state, warning state, alarm state; F6: According to the boundary value The iteration state probability parameter, i.e. , , ; F7: The parameters are updated in the same way The boundary value iteration probability parameters are obtained by performing steps F2-F6, and digital feedback is realized.

Images