Wind turbine generator gearbox fault early warning system based on time sequence analysis

Abstract

The invention relates to the technical field of wind turbine generator gearbox parameter data processing, in particular to a wind turbine generator gearbox fault early warning system based on time sequence analysis. The system comprises a training matrix construction module used for constructing a training matrix; the importance degree calculation module is used for calculating the importance degree of the parameter data at the moment according to the characteristics of the parameter data at the moment in the set step length in the corresponding original sequences; the state irrelevance calculation module is used for calculating the state irrelevance of the parameter data at the moment according to the difference between the parameter data at one moment and the parameter data at other moments in the set step length and the difference between the sampling moments; the data screening module is used for screening and constructing a process memory matrix based on the data retention degree of the parameter data; and the fault early warning module is used for predicting parameter data at the next moment, obtaining an early warning threshold value and performing early warning. According to the invention, the components of the process memory matrix are optimized, and the accuracy of early warning of the gearbox is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of parameter data processing of wind turbine gearboxes, and particularly to a fault warning system for wind turbine gearboxes based on time series analysis. Background Art

[0002] A wind turbine, i.e., a wind power generation unit, is a system that converts wind energy into electrical energy, and is mostly installed at windy places such as mountains, wildernesses, and beaches. The wind power gearbox is an important mechanical component in a wind turbine, and its main function is to transmit the power generated by the wind wheel under the action of wind to the generator and make it obtain the corresponding rotational speed. Since the installation environment of the wind turbine is relatively harsh and the gearbox is installed in a narrow space at the top of the tower, it is very difficult to repair once a failure occurs. Therefore, it is necessary to give a fault warning to the wind turbine gearbox.

[0003] Traditionally, various physical quantities such as vibration, temperature, rotational speed, etc. of the running state of the gearbox are collected through sensors, and thresholds are set for these data for warning. However, the thresholds of this warning method are fixed and lack a comprehensive analysis of the running state of the gearbox. The NSET (Nonlinear State Estimation) prediction model compares the current running data with the generated historical running states by synthesizing the time series data of multiple parameters, calculates and compares the similarities between multiple state variables, so as to give a fault warning. This model has the characteristics of considering multi-dimensional data for early warning and being fast and reliable. Therefore, when using the NSET warning model to predict the fault state of the gearbox, the construction of the process memory matrix in the NEST prediction model is particularly important. Traditionally, a fixed-step sampling method is used to perform equidistant sampling from the training matrix, and the data that meet the conditions are added to the process matrix. Due to the influence of the change of wind speed in the natural environment, the running state of the wind turbine fluctuates continuously. This method of constructing the process matrix has a large randomness and is prone to the loss of some key running state data. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a fault warning system for wind turbine gearboxes based on time series analysis, and the specific technical solution adopted is as follows: An embodiment of the present invention provides a fault warning system for wind turbine gearboxes based on time series analysis. The system includes: A training matrix construction module, configured to obtain the historical parameter data of the wind turbine gearbox and construct a training matrix; An importance calculation module, configured to sample from the training matrix using a set step length, and calculate the importance of the parameter data at a moment within the set step length according to the adjacent parameter data of the parameter data at each moment in their respective original sequences; A state non - correlation calculation module, which is used to calculate the state non - correlation of the parameter data at a certain moment according to the differences between the parameter data at one moment and other moments within a set step length and the differences between sampling moments; A data screening module, which is used to obtain the data retention degree of the parameter data at a certain moment according to the importance degree and state non - correlation of the parameter data at a certain moment within a set step length; and use the data retention degree to screen the parameter data at each moment within each set step length and form a process memory matrix; A fault warning module, which is used to predict the parameter data at the next moment based on the process memory matrix; obtain a warning threshold, and conduct a warning on the operating state of the gearbox of the wind turbine at the next moment according to the warning threshold.

[0005] Preferably, obtain the historical parameter data of the gearbox of the wind turbine and construct a training matrix, including: Pre - process the historical parameter data of the gearbox of the wind turbine, and construct an observation matrix using the pre - processed historical parameter data; construct a training matrix based on the observation matrix, and the parameter data includes gearbox temperature, converter temperature, engine active power, and vibration data.

[0006] Preferably, calculate the importance degree of the parameter data at a certain moment according to the differences between the parameter data at a certain moment and other moments within a set step length and the differences between sampling moments, including: Obtain the mean value of a preset number of parameter data adjacent to the left of a certain parameter data at a certain moment in its corresponding original sequence within a set step length, denoted as the left mean value; obtain the mean value of a preset number of parameter data adjacent to the right of a certain parameter data at a certain moment in its corresponding original sequence within a set step length, denoted as the right mean value; respectively obtain the absolute values of the differences between the parameter data and the corresponding left mean value and right mean value, and take the maximum value among them as the left - right maximum difference of the parameter data. Average and normalize the left - right maximum differences of the parameter data at a certain moment to obtain the first distribution feature of the parameter data at this moment; Obtain the number of normal parameter data among a preset number of parameter data adjacent to the left and a preset number of parameter data adjacent to the right of the parameter data in its corresponding original sequence; average and normalize the number of normal parameter data corresponding to each parameter data at a certain moment to obtain the second distribution feature of the parameter data at this moment; Average the first distribution feature and the second distribution feature to obtain the importance degree of the parameter data at this moment.

[0007] Preferably, calculate the state non - correlation of the parameter data at a certain moment according to the differences between the parameter data at a certain moment and other moments within a set step length and the differences between sampling moments, including: Calculate the Euclidean distance between the parameter data at a moment within a set step length and the parameter data at other moments, and take the average to obtain the distance difference of the parameter data at this moment; calculate the average of the absolute values of the differences between the sampling moments of the parameter data at a moment within a set step length and the sampling moments of the parameter data at other moments to obtain the timing difference of the parameter data at this moment; take the average of the normalized distance difference and the normalized timing difference to obtain the state irrelevance of the data difference at this moment.

[0008] Preferably, obtain the data retention degree of the parameter data at a moment according to the importance degree and state irrelevance of the parameter data at a moment within a set step length, including: Weightedly sum the importance degree and state irrelevance of the parameter data at a moment within a set step length to obtain the retention degree of the parameter data at this moment.

[0009] Preferably, use the data retention degree to screen the parameter data at each moment within each set step length and form a process memory matrix, including: When the retention degree of the parameter data at a moment in a set step length in the gearbox training matrix is greater than or equal to the retention threshold, retain the parameter data at this moment, and add the parameter data at this moment to the process memory matrix to obtain the process memory matrix.

[0010] Preferably, predict the parameter data at the next moment based on the process memory matrix, including: Use the NSET prediction model to predict the parameter data at the next moment based on the vector composed of the process memory matrix and the parameter data at the current moment.

[0011] Preferably, obtain the warning threshold, including: Obtain the parameter data at each moment in the training matrix, and perform prediction to obtain the predicted parameter data at each moment; subtract the parameter data at a moment from the predicted parameter data at this moment to obtain the residual sequence corresponding to this moment; obtain the mean value of the residuals of the residual sequence at a moment and the root mean square error , obtain the corresponding lower limit value and upper limit value at this moment, the lower limit value is , the upper limit value is , then take the mean value of the lower limit value and upper limit value corresponding to each moment respectively to obtain the average lower limit value and average upper limit value, and the range between the average lower limit value and average upper limit value is the warning threshold.

[0012] Preferably, give a warning about the operating state of the gearbox of the wind turbine at the next moment according to the warning threshold, including: Subtract the true parameter data at the next moment from the predicted parameter data to obtain the residual sequence corresponding to the next moment; obtain the lower limit value and the upper limit value corresponding to the next moment according to the residual sequence at the next moment, compare them with the warning threshold, and issue a warning if the warning threshold is exceeded.

[0013] The embodiments of the present invention at least have the following beneficial effects: First, based on the historical parameter data of the wind turbine gearbox, a training matrix is constructed. Further, samples are taken from the training matrix using a set step size. Then, the characteristics of the parameter data at each moment within each set step size in the original data sequence are analyzed, and thus the importance of the parameter data at each moment within each set step size is obtained. Next, the correlation between the parameter data at each moment within the set step size and the parameter data at other moments is analyzed, and thus the state non-correlation corresponding to the parameter data at each moment is obtained. By combining the two, taking into account the importance and uniqueness of the parameter data, the retention degree of the parameter data at each moment within the set step size is obtained, so that the selected parameter data has a low redundancy degree and a high integrity, achieving the purpose of optimizing the construction of the process memory matrix, making the parameter data vectors at each moment in the obtained process memory matrix have less redundancy and more obvious features, making the prediction more accurate, and improving the reliability and accuracy of fault warning. Description of the Drawings

[0014] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0015] Figure 1 It is a system block diagram of a fault warning system for a wind turbine gearbox based on time series analysis provided by an embodiment of the present invention. Detailed Embodiments

[0016] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following, in combination with the drawings and preferred embodiments, details the specific embodiments, structures, features, and effects of a fault warning system for a wind turbine gearbox based on time series analysis proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0018] The following specifically describes the specific solution of a fault warning system for a wind turbine gearbox based on time series analysis provided by the present invention in conjunction with the accompanying drawings.

[0019] Embodiment: The main application scenario of the present invention is as follows: Traditionally, the fault monitoring and warning of wind turbine gearboxes mainly collect various parameter data during the operation of the gearbox through sensors, such as vibration, temperature, speed data, etc., and then analyze each parameter data and set thresholds. When the fault risk reaches the preset threshold, the system triggers an alarm. The present invention optimizes the NEST prediction model and then monitors the faults of the wind turbine gearbox.

[0020] Please refer to Figure 1 , which shows a system block diagram of a fault warning system for a wind turbine gearbox based on time series analysis provided by an embodiment of the present invention. The system includes the following modules: A training matrix construction module, configured to obtain historical parameter data of the wind turbine gearbox and construct a training matrix.

[0021] Since the NSET prediction model is required to predict the operating state of the gearbox, various types of historical data during the operation of the gearbox need to be obtained. These data are all time series data, including various types of data such as gearbox temperature, converter temperature, engine active power, and vibration data. These data are recorded as parameter data, and the historical parameter data during the operation of the wind turbine gearbox is preprocessed, including denoising processing, missing value filling, and outlier removal.

[0022] An observation matrix P is constructed using the preprocessed historical parameter data. The observation matrix is specifically: , where b is the number of data in different time periods during monitoring, n represents the number of types of parameter data collected at one moment, that is, the number of parameter data collected at one moment. One column in the observation matrix is various parameters at one moment, which is called an observation vector.

[0023] Furthermore, a training matrix K is constructed according to the constructed observation matrix. The training matrix K is specifically: ,

[0024] The training matrix K is composed of k observation vectors selected from the observation matrix P. Each observation vector is generated when the gearbox is in a normal working state. That is, the above training matrix K is the matrix obtained by removing all parameter data in the abnormal state from the observation matrix P. n is the number of selected parameter data. In this application, n is 4.

[0025] An importance calculation module is used to sample from the training matrix using a set step size, and calculate the importance of the parameter data at a certain moment according to the adjacent parameter data of the parameter data at each moment within the set step size in their respective original sequences.

[0026] When the traditional NSET prediction model is used to predict the operating state of the gearbox, the construction of the process memory matrix is to select the observation vectors that meet the threshold conditions in each step size interval of the training matrix K with a fixed step size, and then add the observation vectors that meet the conditions to the matrix. Affected by the change of wind speed in the natural environment, the operating state of the wind turbine keeps fluctuating. This method cannot accurately reflect the normal operating characteristics of the wind turbine. Therefore, it is necessary to analyze the data characteristics of the parameters at each moment within the fixed step size interval, and then obtain the retention degree of the parameter data at each moment within the step size interval; the greater the retention degree of the parameter data at a certain moment within the step size interval, it means that the parameter data at this moment is more important and is the data under a relatively special normal state. At this time, the parameter data at this moment is added to the process memory matrix, and then the state of the wind turbine gearbox at the next moment is warned.

[0027] The higher the importance of the parameter data at a certain moment within each step size in the training matrix of the gearbox, the more important the parameter data at this moment is for the state monitoring of the wind turbine gearbox, that is, the data at this moment has a greater impact on the model prediction result. At this time, the parameter data at this moment should be retained. Therefore, the importance of the parameter data at each moment is obtained by analyzing the characteristics of the parameter data at each moment in their respective original sequences.

[0028] To eliminate the influence of dimensions, the obtained training matrix is first normalized, and the normalized training matrix is denoted as , assuming that the original fixed-step sampling method sets the step size to h when sampling the training matrix .

[0029] Sample from the training matrix using the set step size, and calculate the importance of the parameter data at a certain moment according to the adjacent parameter data of the parameter data at each moment within the set step size in their respective original sequences.

[0030] Specifically, calculate the mean of a preset number of parameter data adjacent to the left of a parameter data at a certain moment within a set step length in its corresponding original sequence, denoted as the left mean; calculate the mean of a preset number of parameter data adjacent to the right of a parameter data at a certain moment within a set step length in its corresponding original sequence, denoted as the right mean; calculate the absolute values of the differences between the parameter data and the corresponding left mean and right mean respectively, and take the maximum value as the left-right maximum difference of the parameter data. Average and normalize the left-right maximum differences of the parameter data at a certain moment to obtain the first distribution feature of the parameter data at this moment; obtain the number of normal parameter data among the preset number of parameter data adjacent to the left and the preset number of parameter data adjacent to the right of the parameter data in its corresponding original sequence; average and normalize the number of normal parameter data corresponding to each parameter data at a certain moment to obtain the second distribution feature of the parameter data at this moment; average the first distribution feature and the second distribution feature to obtain the importance degree of the parameter data at this moment.

[0031] The calculation model for the degree of parameter data at a certain moment within a set step length is specifically as follows: Among them, represents the importance degree of the parameter data at the j-th moment in the i-th set step length, represents the normalization operation, M represents the number of parameter data in a certain moment, m represents the m-th parameter in a certain moment, is the parameter data adjacent to one side of the m-th parameter data at the j-th moment in the i-th set step length in its own original data sequence, k represents the preset number, and the reference value is 10. The implementer can adjust it according to the amount of data; The absolute value of the difference between the m-th parameter data at the j-th moment in the i-th set step length and its corresponding left mean, represents the absolute value of the difference between the m-th parameter data at the j-th moment in the i-th set step length and its corresponding right mean, represents taking the maximum value of these two; represents the first distribution feature. The larger this value is, the more it indicates that the parameter data at the j-th moment in the i-th set step length all show turning point situations in the original data sequence. Also, because the training data is the operation data of the gearbox under normal conditions selected from the original observation matrix, when the parameter data at this moment has the characteristics of turning points, it indicates that the importance degree of the parameter data at this moment is greater; is the number of the left and right adjacent parameter data of the m-th parameter data at the j-th moment in the i-th set step length in its own original data sequence, and the number of parameter data adjacent to both sides is equal, Indicates the number of normal parameter data among the preset number of parameter data adjacent to the left and the preset number of parameter data adjacent to the right of the m-th parameter data in its corresponding original sequence. The larger this number is, the more stable the overall data state of the neighborhood where the m-th parameter data is located in the original data sequence. At this time, the corresponding importance degree is greater. Indicates the second distribution feature. The larger this value is, the higher the importance degree of the parameter data at a certain moment. Thus, the importance degree of the parameter data at each moment within each step size can be calculated and obtained.

[0032] The state non-correlation calculation module is used to calculate the state non-correlation of the parameter data at a certain moment according to the differences between the parameter data at a certain moment and the parameter data at other moments within the set step size and the differences between the sampling moments.

[0033] The stronger the state non-correlation between the parameter data at a certain moment within the set step size and the remaining data means the stronger the uniqueness of the parameter data at this moment, that is, the parameter data at this moment is the normal operation data obtained by the gearbox under specific conditions. The parameter data at this moment reflects the operating state of the gearbox under characteristic conditions. At this time, the data at this moment should be retained; Therefore, the retention degree of the parameter data at the current moment is obtained by comprehensively considering the importance degree of the parameter data at each moment within each step size and its state non-correlation with the remaining data.

[0034] Calculating the state non-correlation of the parameter data at a certain moment according to the differences between the parameter data at a certain moment and the parameter data at other moments within the set step size and the differences between the sampling moments includes: Calculating the Euclidean distance between the parameter data at a certain moment and the parameter data at other moments within the set step size and taking the average to obtain the distance difference of the parameter data at this moment; calculating the absolute value of the difference between the sampling moment of the parameter data at a certain moment and the sampling moments of the parameter data at other moments within the set step size and taking the average to obtain the time series difference of the parameter data at this moment; taking the average of the normalized distance difference and the normalized time series difference to obtain the state non-correlation of the data difference at this moment.

[0035] The calculation model for the state non-correlation between the parameter data at the j-th moment in the i-th set step size in the gearbox training matrix and the remaining data is: , Where, Represents the state non-correlation of the parameter data at the j-th moment in the i-th set step size. I represents the number of data entries within a set step size, that is, how many moments of data are there within a set step size. Represents the Euclidean distance between the parameter data at the j-th moment and the parameter data at the k-th moment in the i-th step size. represents the Euclidean distance operation symbol. The Euclidean distance has an intuitive physical meaning and is a basic operator for measuring the similarity of two vectors. When calculating the Euclidean distance, the parameter data at a certain moment can be regarded as a vector. represents the mean of the sum of the Euclidean distances between the parameter data at the j-th moment within the i-th step length in the gearbox training matrix and the parameter data at the remaining moments, that is, the distance difference. The greater this distance difference, the greater the difference between the parameter data at the two moments, that is, the greater the non-correlation. and represents the sampling moments of the parameter data at the j-th moment and the k-th moment within the i-th set step length. The sampling moment is also the time stamp of the parameter data at the j-th moment and the k-th moment on their respective original sequences. represents the absolute value of the difference between the two sampling moments. represents the mean of the absolute values of the differences between the sampling moment of the parameter data at the j-th moment and the sampling moments corresponding to the parameter data at other moments within the set step length, that is, the time series difference at the j-th moment. The larger this value, the greater the time span between the parameter data at the j-th moment corresponding to the original data sequence and the parameter data corresponding to the remaining parameter data on the original data sequence. At this time, the possibility of non-correlation between the parameter data at the j-th moment and the parameter data at the remaining moments is greater.

[0036] Thus, the state non-correlation of the parameter data at each moment within each set step length can be obtained.

[0037] The data screening module is used to obtain the data retention degree of the parameter data at a certain moment according to the importance and state non-correlation of the parameter data at a certain moment within the set step length; use the data retention degree to screen the parameter data at each moment within each set step length and form a process memory matrix.

[0038] After obtaining the importance and state non-correlation of the parameter data at a certain moment within the set step length, it is necessary to comprehensively consider the two, and then obtain the importance of the parameter data at this moment. Specifically, the importance and state non-correlation of the parameter data at a certain moment within the set step length are weighted and summed to obtain the retention degree of the parameter data at this moment.

[0039] The calculation formula is:

[0040] Among them, is the retention degree of the parameter data at the j-th moment within the i-th set step length in the gearbox training matrix. is the importance of the parameter data at the j-th moment within the i-th set step length in the gearbox training matrix. The state irrelevance of the parameter data at the j-th moment within the i-th set step length in the gearbox training matrix. Since the importance level is more important for the subsequent establishment of the prediction model in the state irrelevance, , , the implementer can adjust the weight.

[0041] The greater the retention degree of a certain parameter data in the gearbox training matrix, it means that the parameter data at this moment is more important and has a lower similarity with the remaining parameter data. At this time, as a relatively unique existence, this parameter data can well reflect the state of the data to be predicted.

[0042] Regarding the retention degree of the parameter data at each moment within each set step length obtained, a lower retention degree means that the parameter data at this moment has a stronger redundancy. At this time, the parameter data at this moment should not be selected for the construction of the process memory matrix; a higher retention degree means that the parameter data at this moment has a stronger importance and a greater irrelevance with the remaining data. At this time, the parameter data at this moment should be selected for the construction of the process matrix. Therefore, this solution stipulates that when the retention degree of the parameter data at a moment in a set step length in the gearbox training matrix is greater than or equal to the retention threshold, the parameter data at this moment is retained, and the parameter data at this moment is added to the process memory matrix, thereby obtaining the process memory matrix D. The reference value of the retention threshold is 0.65, and the implementer can adjust it according to the actual situation.

[0043] A fault warning module, used to predict the parameter data at the next moment based on the process memory matrix; obtain a warning threshold, and perform a warning on the operating state of the gearbox of the wind turbine at the next moment according to the warning threshold.

[0044] After obtaining the process memory matrix, according to the existing NSET prediction model and the vector composed of the parameter data at the current moment, the vector composed of the predicted parameter data at the next moment can be obtained, that is, the parameter data at the next moment. The specific prediction model is: , where, The vector composed of the predicted parameter data of the gearbox at the next moment, is the process memory matrix constructed in this application, is the vector composed of the parameter data of the gearbox at the current moment, is the Euclidean distance operator of the two vectors. Thus, the parameter data at the next moment can be predicted using the NSET prediction model based on the process memory matrix and the vector composed of the parameter data at the current moment.

[0045] Further, residual is used for early warning analysis. Specifically, parameter data at each moment in the training matrix is obtained and prediction is carried out to obtain predicted parameter data at each moment; the parameter data at one moment is subtracted from the predicted parameter data at that moment to obtain the residual sequence corresponding to that moment; the mean value of the residual of the residual sequence at one moment is calculated. and root mean square error , obtaining the lower limit value and upper limit value corresponding to that moment. The lower limit value is , and the upper limit value is . Then, the mean values of the lower limit values and upper limit values corresponding to each moment are calculated respectively to obtain the average lower limit value and average upper limit value. The range between the average lower limit value and average upper limit value is the early warning threshold. It should be noted that the early warning threshold can be continuously updated as the historical data increases continuously to ensure real-time adaptation to environmental changes. Specifically, a sliding window is established and slides with a fixed step length to obtain the parameter data and predicted parameter data at each moment when the gearbox is in normal operation within the sliding window, and then the early warning threshold is updated. Among them, the parameter data at each moment in the training matrix are all parameter data in the normal state.

[0046] Since it is necessary to judge whether the residual sequence at the next moment exceeds the early warning threshold, and the residual sequence is a vector and difficult to compare with a single threshold, it is necessary to calculate the residual sequence using the calculation method of the early warning threshold, and compare the calculation result with the threshold to conduct fault early warning.

[0047] According to the early warning threshold, the running state of the gearbox of the wind turbine at the next moment is warned. Specifically, the real parameter data at the next moment is subtracted from the predicted parameter data at the next moment to obtain the residual sequence corresponding to the next moment; the lower limit value and upper limit value corresponding to the next moment are obtained according to the residual sequence at the next moment and compared with the early warning threshold. If it exceeds the early warning threshold, a warning is issued.

[0048] In summary, according to the above method, the prediction result is judged for abnormality and fault early warning is carried out. In this application, the NSET prediction model is used to predict the state of the gearbox of the wind turbine. In the construction of the prediction model, the key step of constructing the process memory matrix is optimized, making up for the shortcomings that when the traditional process memory matrix is obtained by sampling and screening the gearbox training data matrix based on a fixed step length, it is affected by the change of wind speed in the natural environment, the running state of the wind turbine fluctuates continuously, and there is a large randomness in the construction of the process memory matrix, which is likely to cause the loss of some key running state data. In this application, the process memory matrix is constructed according to the own characteristics of the parameter data at each moment, making the parameter data vectors at each moment in the obtained process matrix have less redundancy and more obvious characteristics, and thus making the prediction more accurate and increasing the reliability and accuracy of fault early warning.

[0049] It should be noted that: the above order of the embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. In addition, the specific embodiments of this specification have been described. Moreover, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0050] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and the key point of each embodiment is to illustrate the differences from other embodiments.

[0051] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A wind turbine gearbox fault warning system based on time series analysis, characterized in that: The system includes: A training matrix building module is used to obtain historical parameter data of the wind turbine gearbox and build a training matrix; The importance calculation module is used to sample from the training matrix using a set step length, and calculate the importance of the parameter data at a moment according to the adjacent parameter data in the corresponding original sequence of each parameter data at a moment within the set step length; A state non-correlation calculation module, used to calculate the state non-correlation of the parameter data at a moment according to the difference between the parameter data at other moments within a set step length and the difference between the sampling moments; A data screening module is used to obtain the data retention degree of the parameter data at a moment in a set step length according to the importance degree and state non-correlation of the parameter data at that moment; the parameter data at each moment in each set step length is screened by using the data retention degree and a process memory matrix is ​​formed; The fault warning module is used to predict the parameter data at the next moment based on the process memory matrix; obtain the warning threshold, and issue a warning on the gearbox operating status of the wind turbine at the next moment according to the warning threshold.

2. A wind turbine gearbox fault early warning system based on time series analysis according to claim 1, characterized in that: The step of obtaining historical parameter data of the wind turbine gearbox and constructing a training matrix includes: The historical parameter data of the wind turbine gearbox are preprocessed, and an observation matrix is ​​constructed using the preprocessed historical parameter data; a training matrix is ​​constructed based on the observation matrix, wherein the parameter data include gearbox temperature, converter temperature, engine active power and vibration data.

3. The wind turbine gearbox fault early warning system based on time series analysis according to claim 1 is characterized in that: The step of calculating the importance of the parameter data at a moment within the set step size according to the adjacent parameter data in the corresponding original sequence includes: The mean of a preset number of parameter data adjacent to the left side of a parameter data at a moment within a set step length in the original sequence corresponding to the parameter data is obtained, and recorded as the left mean; the mean of a preset number of parameter data adjacent to the right side of a parameter data at a moment within the set step length in the original sequence corresponding to the parameter data is obtained, and recorded as the right mean; the absolute values ​​of the differences between the parameter data and the corresponding left mean and right mean are respectively obtained, and the maximum value is taken as the left-right maximum difference of the parameter data, and the left-right maximum differences of each parameter data at a moment are averaged and normalized to obtain the first distribution feature of the parameter data at the moment; Obtaining the number of normal parameter data in a preset number of parameter data adjacent to the left and a preset number of parameter data adjacent to the right in the original sequence corresponding to the parameter data itself; averaging and normalizing the number of normal parameter data corresponding to each parameter data at a moment to obtain a second distribution feature of the parameter data at the moment; The first distribution feature and the second distribution feature are averaged to obtain the importance of the parameter data at that moment.

4. The wind turbine gearbox fault early warning system based on time series analysis according to claim 1 is characterized in that: The state non-correlation of the parameter data at a moment is calculated according to the difference between the parameter data at one moment and other moments within the set step length and the difference between the sampling moments, including: The Euclidean distance between the parameter data at one moment within the set step size and the parameter data at other moments is calculated and averaged to obtain the distance difference of the parameter data at that moment; the absolute value of the difference between the sampling moment of the parameter data at one moment within the set step size and the sampling moment of the parameter data at other moments is calculated and averaged to obtain the time series difference of the parameter data at that moment; the normalized distance difference and the normalized time series difference are averaged to obtain the state non-correlation of the data difference at that moment.

5. The wind turbine gearbox fault early warning system based on time series analysis according to claim 1 is characterized in that: The step of obtaining the data retention degree of the parameter data at a moment according to the importance degree and state non-correlation of the parameter data at a moment within the set step size includes: The importance and state irrelevance of the parameter data at a moment within the set step are weighted and summed to obtain the retention degree of the parameter data at that moment.

6. The wind turbine gearbox fault early warning system based on time series analysis according to claim 1 is characterized in that: The method of using the data retention degree to screen the parameter data at each moment within each set step length and form a process memory matrix includes: When the retention degree of parameter data at a moment in a set step in the gearbox training matrix is ​​greater than or equal to the retention threshold, the parameter data at that moment is retained and added to the process memory matrix to obtain the process memory matrix.

7. The wind turbine gearbox fault early warning system based on time series analysis according to claim 1 is characterized in that: The predicting of parameter data at the next moment based on the process memory matrix includes: The NSET prediction model is used to predict the parameter data at the next moment based on the vector composed of the process memory matrix and the parameter data at the current moment.

8. The wind turbine gearbox fault early warning system based on time series analysis according to claim 1 is characterized in that: The obtaining of the warning threshold comprises: Obtain the parameter data at each moment in the training matrix, and perform prediction to obtain the predicted parameter data at each moment; subtract the parameter data at a moment from the predicted parameter data to obtain the residual sequence corresponding to that moment; calculate the residual mean of the residual sequence at a moment and RMS error , get the upper and lower limits corresponding to the moment, the lower limit is , the upper limit is , and then calculate the average of the lower limit value and the upper limit value corresponding to each moment to obtain the average lower limit value and the average upper limit value. The range between the average lower limit value and the average upper limit value is the warning threshold.

9. The wind turbine gearbox fault early warning system based on time series analysis according to claim 1, characterized in that: The step of giving an early warning of the gearbox operating state of the wind turbine generator set at the next moment according to the early warning threshold comprises: Subtract the actual parameter data of the next moment from the predicted parameter data to obtain the residual sequence corresponding to the next moment; obtain the lower limit and upper limit corresponding to the next moment based on the residual sequence of the next moment, and compare them with the warning threshold. If the warning threshold is exceeded, a warning is issued.

Images