TW201437476A - Wind power fault prediction system and method thereof - Google Patents

Abstract

A wind power fault prediction system applies to a wind power generator is provided and includes a sensing module, a chaotic synchronization detection module, a gray prediction modules and a diagnostic module. The sensing module senses a features of the wind power generator. The chaotic synchronization detection module is connected to the sensing module, and captures the features to establish a chaotic error scatter plot. The gray prediction module is connected to the chaotic synchronization detection module, and predicts a gravity trajectory of the chaotic error scatter plot. Diagnostic module is connected to the gray prediction module, and diagnoses a predicted error point of the wind power generator based on the trajectory of the chaotic error scatter plot. Therefore, the present invention can effectively predict the trend of the wind power generator failure.

Description

Wind power fault prediction system and method thereof

The invention relates to a wind power fault prediction system, in particular to a wind power fault prediction system for reducing fault data feature extraction.

In the operation process of large-scale wind power generation equipment, in the face of ever-changing wind conditions, the reliability and availability of the wind turbine generator are difficult to maintain due to the long-term load with severe load, and the power system and electronic control are covered. Problems such as sensors, generators, and fan blades. Once the wind power generation device fails, it is difficult for maintenance personnel to find the cause of the failure, which makes maintenance difficult. Therefore, by designing a set of equipment operation records, effectively predicting the occurrence of faults, it is possible to prevent or repair or adjust the operation mode in advance, and prevent the occurrence of wind power generation equipment accidents and ensure the safe and stable operation of the power generation system, thereby improving its management efficiency. And reducing operating and maintenance costs has become one of the most important projects in wind power system research.

The general fault diagnosis of wind power generation equipment is divided into traditional diagnosis methods, mathematical processing methods and intelligent diagnosis methods. The traditional diagnostic methods include vibration analysis, oil quality analysis, infrared temperature measurement, and noise. Sound and acoustic analysis, non-destructive detection, etc.; mathematical diagnosis methods include spectrum analysis, axial trajectory analysis, wavelet analysis and short-time Fourier analysis, and other signal processing methods, and the concept of intelligent diagnostic analysis such as hidden mark Theory, Bayesian theory and neural networks.

However, the above-mentioned wisdom diagnosis analysis needs to establish a huge data database, and it takes a long time to diagnose and analyze the fault category of the wind power generation device. Therefore, how to reduce the data required for diagnosis and accurately prevent wind power generation equipment accidents, ensure the safe and stable operation of wind power generation equipment and reduce maintenance costs, a more complete fault diagnosis system for wind power generation equipment is needed.

One embodiment of the present invention provides a wind power failure prediction system for a wind power generation device. The wind power failure prediction system includes a sensing module, a chaotic synchronization detection module, and a gray prediction. Module and a diagnostic module. The sensing module senses one of the data values of the wind power generator. The chaotic synchronization detection module is connected to the sensing module, and the chaotic synchronization detection module takes data values and establishes a chaotic error scatter map. The gray prediction module is connected to the chaotic synchronization detection module, and the gray prediction module predicts the center of gravity trajectory of the chaotic error scatter diagram based on the chaotic error scatter plot. The diagnostic module is connected to the gray prediction module, and the diagnostic module diagnoses one of the predicted failure points of the wind power generation device according to the gravity center trajectory of the chaotic error scatter map. Thereby, the present invention can effectively predict the trend of failure of the wind power generation device.

According to an embodiment of the invention, the chaotic synchronization detecting module uses a chaotic dynamic error equation to identify data values and establish chaotic errors. Scatter map. The wind power generation device comprises a gear box and a generator, and the data value can be a vibration value of one of the gear boxes, an oil temperature of the gear box or a vibration value of the generator. The predicted point of failure may be a gearbox or generator of the wind power plant.

One embodiment of a method aspect of the present invention provides a wind power failure prediction method for a wind power generation device. The wind power failure prediction method includes the following steps: sensing a data value of one of the wind power generation devices; and extracting data Value and establish a chaotic error scatter plot; predict the center of gravity trajectory of the chaotic error scatter plot; and diagnose the fault point of one of the wind turbines.

According to another embodiment of the present invention, the step of diagnosing a predicted fault point of the wind power generator utilizes an extension theory. The step of extracting the data values and establishing a chaotic error scatter plot uses a chaotic dynamic error equation to identify the data values. The data value may be one of the vibration values of one of the generators of the wind power generation device, one of the vibration values of one of the gearboxes of the wind power generator, and one of the oil temperatures of the gearbox of the wind power generator. The predicted point of failure can be a gearbox or generator of the wind turbine.

The center of gravity of the chaotic error scatter map established by the chaotic synchronization detection module of the present invention is characterized by fault diagnosis, and the range of data values can be effectively set, thereby reducing the number of data values.

100‧‧‧Wind power plant

110‧‧‧data value

200‧‧‧Wind power failure prediction system

210‧‧‧Sense Module

220‧‧‧Chaotic Synchronous Detection Module

221‧‧‧Chaotic error scatter plot

230‧‧‧ Gray Prediction Module

231‧‧‧ center of gravity track

240‧‧‧Diagnostic Module

241‧‧‧Predicting the point of failure

300‧‧‧Steps

310‧‧‧Steps

320‧‧‧Steps

330‧‧‧Steps

1 is a block diagram of a wind power failure prediction system according to an embodiment of the present invention.

Figure 2 is a diagram showing the chaotic error of prediction and actual measurement according to an embodiment of the present invention. The trajectory of the center of gravity of the difference scatter plot.

FIG. 3 is a flow chart showing a method for predicting wind power failure according to an embodiment of the present invention.

Please refer to FIG. 1 , which is a block diagram of a wind power failure prediction system according to an embodiment of the present invention. The wind power failure prediction system 200 is applied to a wind power generation device 100. The wind power generation device failure prediction system 200 includes a sensing module 210, a chaotic synchronization detection module 220, a gray prediction module 230, and a diagnostic module 240.

The sensing module 210 is configured to sense one of the data values 110 of the wind power generating device 100.

The chaotic synchronization detection module 220 is connected to the sensing module 210, and the chaotic synchronization detection module 220 captures the data value 110 and establishes a chaotic error scatter diagram 221 .

The gray prediction module 230 is connected to the chaotic synchronization detection module 220, and the gray prediction module 230 predicts a gravity center trajectory 231 of the chaotic error dispersion map 221 according to the chaotic error scatter map 221 .

The diagnostic module 240 is coupled to the gray prediction module 230. The diagnostic module 240 diagnoses one of the predicted fault points 241 of the wind power generation device 100 based on the gravity center trajectory 231 of the chaotic error scatter map 221 .

The operation of the wind power failure prediction system 200 described above is as follows: the sensing module 210 senses the data value 110 of the wind power generation device 100, such as the gearbox oil temperature of the wind power generation device 100 (not shown), the gear of the wind power generation device 100. Box vibration value (not shown) or generator of wind power generator 100 Vibration value (not shown). The chaotic synchronization detection module 220 establishes a chaotic error scatter plot 221 by the principle of chaotic synchronization.

Chaotic synchronization is a theory that uses a chaotic system signal to control another chaotic system signal and finally synchronize the two chaotic system signals. Generally speaking, two synchronous chaotic systems are called the main chaotic system and the servant chaotic system respectively. When the initial value of the main chaotic system and the servant chaotic system are different, the trajectory of the two chaotic systems will be different, and usually The controller will be used to track the main chaotic system at the back end of the servant chaotic system, and the controller can make the two chaotic systems equal the trajectory motion at the same time. This tracking state is chaotic synchronization such as (1-1): Where i =1, 2,..., n , X _{Slave , i} represents the servant chaotic system, X _{Master , i} represents the main chaotic system. In this embodiment, the chaotic synchronization method is used to detect the chaotic trajectory of the signal of the wind power generation device 100, and the main chaotic system and the servant chaotic system are respectively (1-2) and (1-3):

Where F _i (i = 1, 2, ..., n) belongs to a nonlinear function, and the equations (1-2) and (1-3) form an error state such as (1-4), and the dynamic error Such as (1-5):

Wherein G _i ( i =1, 2, . . . , n ) is a nonlinear equation. However, in the present embodiment, the motion trajectory of the attractor using the chaotic phenomenon, that is, the chaotic dynamic error equation is used to identify the power generating device 100. The basis of the state; the dynamic error needs to be multiple data, and the data mode is expressed as (1-6): Where j=1, 2, 3, ..., n-1.

Accordingly, the chaotic synchronization detection module 220 performs an accumulation operation of the chaotic trajectory on the data value 110 captured by the sensing module 210 as a state basis for identifying the wind power generation device 100. Since the gearbox oil temperature of the wind power generation device 100, the gearbox vibration value of the wind power generation device 100, or the generator vibration value of the wind power generation device 100 can be used as the main characteristic value input chaotic dynamic error equation, the three main features are chaotically synchronized. After the detection module 220 forms the chaotic error scatter plot 221, the center of gravity of the chaotic error scatter plot 221 is then used as the feature value. However, the scatter plot formed by one main feature has two centers of gravity, and the X-axis of the center of gravity must be recorded separately. The value of the Y-axis, so there are a total of twelve sub-features.

Please refer to the attached figure 1 for the chaotic error scatter diagram of the gearbox oil temperature under different blade failures, (a) for the normal state, (b) for one blade failure, and (c) for two blade failures. Figure 2 shows the chaotic error scatter plot of the gearbox oil temperature under different oil leakage conditions, (a) is the normal state, (b) is the oil leakage 30%, (c) is the oil leakage 50%, (d) is Oil leakage is 70%, and (e) is 90% oil leakage. Figure 3 shows the chaotic error scatter plot of the gearbox vibration under different blade failures, (a) for the normal state, (b) for one blade failure, and (c) for two blade failures. Figure 4 shows the chaotic error scatter diagram of the vibration of the gearbox under different oil leakage conditions, (a) is the normal state, (b) is the oil leakage 30%, (c) is the oil leakage 50%, and (d) is the leakage 70% oil, (e) is 90% oil leakage. Figure 5 shows the chaotic error scatter plot of generator vibration under different blade faults, (a) is the normal state, (b) is a blade fault, and (c) is a two-blade fault. Figure 6 shows the chaotic error scatter diagram of generator vibration under different oil leakage conditions, (a) is the normal state, and (b) is the leakage. 30% oil, (c) is 50% oil leakage, (d) is 70% oil leakage, and (e) is 90% oil leakage. The triangle marks shown in the attached figures 1 to 6 indicate the center of gravity of the x-axis positive and negative domains for each chaotic error scatter plot.

The next prediction method uses the gray prediction module 230 to predict the change of the center of gravity trajectory 231 of the chaotic error scatter plot 221 according to the twelve sub-features of the chaotic error scatter plot 221, wherein the gray prediction module 230 adopts a gray theory. First, we must use the grey theory to define the formula (1-7), which is the differential equation of the gray model (1, 1): Where t is the self-variable, a and b are the pending parameters of the gray model (a is the development coefficient and b is the grey control variable), and x ⁽¹⁾ is the value generated by the accumulation; this gray prediction modeling can be divided into parametric and matrix There are two types, and this embodiment adopts a matrix type to establish the gray model. The modeling steps are as follows: The first prediction system uses (1-7) to draw a set of series as the original series; After completing a cycle, each time a number of columns in the data will be retrieved as a new original sequence, where (1-8) is the original sequence (x ⁽⁰⁾ ) modeled by the gray prediction.

x ⁽⁰⁾ = ( x ⁽⁰⁾ (1), x ⁽⁰⁾ (2),..., x ⁽⁰⁾ ( k )) (1-8). Where k =1, 2, 3,..., m ; m N. Then, the original data that is disorderly and has no regularity is generated by the accumulation process, and a new and more regular sequence can be obtained, which accumulates the expression; as shown in (1-9) Number of columns (x ⁽¹⁾ ): Where L N. The mean generation can adjust the weight value α of the data, so the predicted weight value α will change according to different root trajectories; as shown in (1-10): z ⁽¹⁾ ( L ) = α * x ¹ ( L ) +(1- α )* x ⁽¹⁾ ( L -1) (1-10). Where L >1; L N. To obtain the gray model (1,1) to determine the two values of the coefficients a and b, this step uses the least square method (Least Square) to calculate the range of two values; as shown in (1-11): x ⁽⁰⁾ ( L )+ az ⁽¹⁾ ( L )= b (1-11). Where L N. The equation (1-11) is converted into a matrix equation of (1-12), and the parameters Y, B and â are represented by (1-13), (1-14) and (1-15), respectively:

The values of the parameters a and b of the formula (1-15) can be obtained from the equations (1-16) and (1-17):

After the parameters a and b are solved, the generated predicted values are accumulated, as shown in equation (1-18): Where L N. The value of the gray prediction is obtained by Inverse Accumulated Generating Operation (IAGO); as shown in (1-19).

After receiving the above values, the diagnostic module 240 diagnoses the predicted fault point of the wind power generator 100 using the principle of the extension theory. The extension theory uses the matter-element model and the extension set to quantify the relationship between the things and the degree of correlation, and can easily describe the information of things. The matter-element theory and extension set are the two cores of extension theory. The matter-element theory is to study the extension of matter-element and the transformation property of matter-element. The extension set is the possibility of change of things, and the transformation of matter-element is Solve the contradiction of the problem and transform the relationship between matter and quantity. The extension set is the extension model and fuzzy mathematics to deal with the contradiction and incompatibility, and expand the fuzzy set range from <0,1> to <-∞,∞>. To describe the matter element of a thing in detail, it is necessary to have the basic three elements to be the object element composed of the thing name N (Name), the thing feature C (Characteristic) and the feature quantity value V (Value); The mathematical expression of the matter element is represented by the formula (1-20): R = ( N , C , V ) (1-20). The specific evaluation step of the extension theory is represented by the following steps: firstly, the range of the extension classical domain and the section is set, as shown by the formulas (1-21) and (1-22):

Where k = 1, 2, 3, ..., m , p = 1, 2, 3, ..., m . Enter the object to be tested, as shown in (1-23): Where t represents the number of objects to be tested, and q is the name of the thing, as in the formula (1-20), n represents the number of features, and x represents the range of feature values. The weighting factor of each feature is determined according to the importance, as shown in (1-24): Calculate the correlation value between the data to be tested and each level set, as shown in (1-25): The definition of the eigenvalue x _i for each classical domain R _k is as shown in (1-26), and the eigenvalue x _i is defined for the distance of the region R _P as shown in (1-27):

Calculate the relevance of the input matter to each category: Where k =1, 2, 3,..., m . By normalizing the degree of association, by calculating the relative value of the degree of association of each level set, the present embodiment uses the normalized equation of the formula (1-28) to make the correlation values of each level set fall at <1, -1. >Between, (1-28) is as follows: If When it is equal to 1, it can be judged that the object to be tested belongs to the rth category, and the correlation of other categories is measured by the degree of association. Thereby, the current fault point condition of the wind power generator 100 and the predicted fault point condition are determined.

Please refer to FIG. 2, which is a diagram showing the trajectory of the center of gravity of the prediction and actual measurement according to an embodiment of the present invention. This embodiment is directed to the center of gravity trajectory predicted by the gearbox vibration value of the gear accelerator of the wind power generator 100 at different oil levels. The gearbox vibration value data of the normal, oil leakage 30%, oil leakage 50% and oil leakage 70% condition is used to predict the 90% leakage of the gear accelerator. This figure is (1-5). The numerical value of the calculation result proves the feasibility and accuracy of the gray prediction situation.

In the test of fault diagnosis and prediction of wind power generation equipment, the chaotic scatter diagram formed by 10000 data of each feature (gearbox vibration value, generator vibration value, gearbox oil temperature, etc.) in the database is taken as a reference, and each setting is set. The classic domain and the local area under the condition, and then use the additional 5000 data in the database as the test. The test proves that the accuracy rate of this embodiment is as high as 98.8%. As shown in Table 1, the fault diagnosis method for this embodiment is compared. The accuracy comparison table.

The computing layer of the multi-layer neural network I is 9 layers, the multi-layer neural network has 10 layers, and the multi-layer neural network III has 11 layers. The input layer of each multi-layer neural network represents 12 sub-features of the X-axis and Y-axis values of the center of gravity, so it is selected as 12, and the output layer represents 7 fault categories for wind turbine failure.

The actual measurement of the forecasting part is predicted and tested with the data of 5000 data in the database. After the identification of the diagnosis method proposed by the research institute, the accuracy of the predicted fault and the actual fault is more than 90%. Table 2 shows a comparison table of the accuracy of each method's fault prediction.

Table 2. Comparison table of accuracy rates for various failure prediction methods

Referring to FIG. 3, a wind power failure prediction process according to an embodiment of the present invention is applied to a wind power generation device. The wind power failure prediction process includes the following steps: Step 300: sensing one of the wind power generation devices Data value; step 310 retrieves the data value and establishes a chaotic error scatter plot; step 320 predicts a center of gravity trajectory of the chaotic error scatter plot; and step 330 diagnoses one of the wind power generator predictive fault points.

It can be seen from the above embodiments of the present invention that the application of the present invention has the following advantages:

1. The present invention proposes a preliminary acquisition of a wind power generation device signal, uses a chaotic synchronization detection mode to obtain a chaotic error distribution, uses the center of gravity in the chaotic error distribution map as a feature value, and diagnoses the fault category with a center of gravity, thereby reducing features. Take the number and shorten the time of the program operation.

2. The invention predicts the characteristic value of the next cycle by the grey theory, and has predicted the fault by this value, and can detect the potential symptom before the failure of the wind power generation device, and can stop the operation of the wind power generation device early, thereby reducing the damage and Maintenance costs.

100‧‧‧Wind power plant

110‧‧‧data value

200‧‧‧Wind power failure prediction system

210‧‧‧Sense Module

220‧‧‧Chaotic Synchronous Detection Module

221‧‧‧Chaotic error scatter plot

230‧‧‧ Gray Prediction Module

231‧‧‧ center of gravity track

240‧‧‧Diagnostic Module

241‧‧‧Predicting the point of failure

Claims (10)

A wind power fault prediction system is applied to a wind power generation device, the wind power failure prediction system includes: a sensing module that senses a data value of the wind power generation device; and a chaotic synchronization detection module that connects the sensing The module, the chaotic synchronization detection module takes the data value and establishes a chaotic error scatter diagram; a gray prediction module is connected to the chaotic synchronization detection module, and the gray prediction module predicts the chaos according to the chaotic error scatter diagram a gravity trajectory of the error scatter plot; and a diagnostic module coupled to the gray prediction module, the diagnostic module diagnosing one of the wind power generation devices to predict a fault point according to the gravity trajectory of the chaotic error scatter plot. The wind power failure prediction system of claim 1, wherein the chaotic synchronization detection module uses a chaotic dynamic error equation to identify the data value and establish the chaotic error scatter map. The wind power failure prediction system of claim 1, wherein the wind power generation device further comprises a generator, and the data value is a vibration value of the generator. The wind power fault prediction system of claim 3, wherein the wind turbine further comprises a gearbox, the data value being a vibration value of the gearbox or an oil temperature of the gearbox. The wind power failure prediction system of claim 4, wherein the prediction The point of failure is the gearbox or the generator of the wind power plant. A wind power failure prediction method is applied to a wind power generation device, the wind power failure prediction method comprising the steps of: sensing a data value of one of the wind power generation devices; extracting the data value and establishing a chaotic error scatter map; predicting the a gravity center trajectory of the chaotic error scatter plot; and diagnosing one of the wind power generation devices to predict a fault point. The wind power failure prediction method of claim 6, wherein the step of diagnosing the current fault point of the wind power generation device and the predicted fault point utilizes an extension theory. The wind power failure prediction method of claim 6, wherein the step of extracting the data value and establishing a chaotic error scatter plot identifies the data value using a chaotic dynamic error equation. The wind power failure prediction method of claim 6, wherein the data value is a vibration value of one of the generators of the wind power generation device. The wind power failure prediction method of claim 9, wherein the data value comprises a vibration value of one of the gearboxes of the wind power generation device or an oil temperature of the gearbox.