Sensor-based online monitoring system for wind turbine tower fastening bolts

Abstract

The present invention relates to the field of data processing technology, and in particular to a sensor-based online monitoring system for tightening bolts of wind turbine towers. The system comprises a processor and a memory, wherein the processor executes a computer program in the memory to implement the following steps: obtaining a multidimensional monitoring data sequence of any bolt on the wind turbine tower at a current sampling moment; dividing the multidimensional monitoring data sequence into subsequences and a current subsequence, and obtaining a temperature and wind pressure influence coefficient between every two subsequences; obtaining a regression weight of each temperature and wind pressure influence coefficient based on the time interval between the sampling moment of each subsequence and the initial sampling moment, constructing a weighted loss function, training a support vector regression model to obtain an optimal prediction model, and obtaining a predicted value; and issuing an abnormal warning for any bolt based on the difference between the temperature and wind pressure influence coefficient between the current subsequence and each subsequence and its predicted value, thereby improving the accuracy of the bolt tightening status monitoring result.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a sensor-based online monitoring system for fastening bolts of a wind turbine tower. Background Art

[0002] With the rapid development of wind power technology, the installed scale and operating life of wind turbines are growing, and structural safety and operational stability are receiving increasing attention. As a key structure supporting the entire wind turbine system, the wind turbine tower's connections, particularly the preload of its fastening bolts, are directly related to the equipment's structural integrity and operational safety.

[0003] With the rapid development of sensor technology, wireless communications, and the Internet of Things (IoT), sensor-based structural health monitoring systems have become an important tool for the intelligent operation and maintenance of wind turbine equipment. Existing monitoring systems install micro-sensors such as strain gauges and piezoelectric sensors at key tower connections to collect real-time data on bolt force and vibration, enabling monitoring of bolt tightening status. However, wind turbine tower bolts are typically located at a high altitude and subject to variable environmental loads. The pressure on the bolts may be affected by environmental interference factors such as temperature and wind load, resulting in a decrease in the accuracy and stability of the monitoring system's bolt tightening status results, making it prone to false alarms or missed alarms.

[0004] Therefore, how to reduce the impact of interference factors in the environment and accurately obtain the bolt tightening status has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, an embodiment of the present invention provides a sensor-based online monitoring system for wind turbine tower fastening bolts to solve the problem of how to reduce the influence of interference factors in the environment and accurately obtain the bolt fastening status.

[0006] An embodiment of the present invention provides a sensor-based online monitoring system for wind turbine tower fastening bolts, comprising a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the following method is implemented:

[0007] For any bolt on the wind turbine tower, obtain multidimensional monitoring data of the bolt at each sampling time to obtain a multidimensional monitoring data sequence at the current sampling time, wherein each multidimensional monitoring data includes pressure data, temperature data, and wind load data;

[0008] Using a sliding window of preset length and step size, the multidimensional monitoring data sequence is divided into at least two subsequences excluding the multidimensional monitoring data at the current sampling time, and a current subsequence including the multidimensional monitoring data at the current sampling time, and based on the fluctuation characteristics of the multidimensional monitoring data in each of the two subsequences, the temperature and wind pressure influence coefficient and the temperature change index between each of the two subsequences are obtained;

[0009] Obtaining a regression weight for each temperature and wind pressure influence coefficient based on the time interval between the sampling moment of each subsequence and the initial sampling moment of the multidimensional monitoring data sequence, constructing a weighted loss function based on the regression weight of each temperature and wind pressure influence coefficient, and training a support vector regression model based on the weighted loss function to obtain an optimal prediction model, wherein the input of the optimal prediction model is the temperature change index and the output is the temperature and wind pressure influence coefficient;

[0010] The temperature and wind pressure influence coefficient between the current subsequence and each subsequence is obtained and recorded as a true coefficient. The predicted value of each true coefficient is obtained using the optimal prediction model. Based on the difference between each true coefficient and its predicted value, an abnormal warning is issued for the tightening state of any bolt.

[0011] Preferably, obtaining the temperature and wind pressure influence coefficient and the temperature change index between each two subsequences based on the fluctuation characteristics of the multidimensional monitoring data in each two subsequences includes:

[0012] For any two subsequences, respectively obtain the temperature mean values ​​of the arbitrary two subsequences, and obtain the temperature change index between the arbitrary two subsequences according to the difference between the temperature mean values ​​of the arbitrary two subsequences;

[0013] The temperature and wind pressure influence coefficient between the arbitrary two subsequences is obtained according to the temperature change indexes of the arbitrary two subsequences and the fluctuation characteristics of the multi-dimensional monitoring data in the arbitrary two subsequences.

[0014] Preferably, obtaining the temperature and wind pressure influence coefficient between the arbitrary two subsequences based on the temperature change indexes of the arbitrary two subsequences and the fluctuation characteristics of the multidimensional monitoring data in the arbitrary two subsequences includes:

[0015] For any subsequence of the arbitrary two subsequences, obtaining a wind pressure response coefficient of the subsequence according to a difference between a fluctuation characteristic of the pressure data and a fluctuation characteristic of the wind load data in the subsequence;

[0016] Obtaining the wind pressure response coefficient of each subsequence in the arbitrary two subsequences respectively; when the temperature change index between the arbitrary two subsequences is 0, obtaining the difference between the wind pressure response coefficients of the arbitrary two subsequences to obtain the temperature-wind pressure influence coefficient between the arbitrary two subsequences;

[0017] When the temperature change index between the arbitrary two subsequences is not 0, the ratio of the difference between the wind pressure response coefficients of the arbitrary two subsequences and the temperature change index between the arbitrary two subsequences is obtained to obtain the temperature and wind pressure influence coefficient between the arbitrary two subsequences.

[0018] Preferably, obtaining the wind pressure response coefficient of any subsequence according to the difference between the fluctuation characteristics of the pressure data and the fluctuation characteristics of the wind load data in any subsequence includes:

[0019] Obtaining a pressure change degree of any subsequence according to a fluctuation characteristic of the pressure data in any subsequence;

[0020] Obtaining a degree of change in the wind load of any subsequence according to a fluctuation characteristic of the wind load data in any subsequence;

[0021] The difference between the pressure variation degree and the wind load variation degree of any subsequence is substituted into an exponential function with a natural constant as a base to obtain the wind pressure response coefficient of any subsequence.

[0022] Preferably, obtaining the pressure change degree of any subsequence according to the fluctuation characteristics of the pressure data in any subsequence includes:

[0023] Performing a fast Fourier transform on the pressure data in any subsequence to obtain a pressure spectrum of the pressure data in any subsequence, wherein the abscissa of the pressure spectrum is frequency and the ordinate is amplitude;

[0024] The product of each frequency in the pressure spectrum and its corresponding amplitude is obtained respectively, and the accumulated value of the product is obtained as the pressure change degree of any subsequence.

[0025] Preferably, obtaining the degree of change of the wind load of any subsequence according to the fluctuation characteristics of the wind load data in any subsequence includes:

[0026] Performing a fast Fourier transform on the wind load data in any subsequence to obtain a wind load spectrum diagram of the wind load data in any subsequence, wherein the abscissa of the wind load spectrum diagram is frequency and the ordinate is amplitude;

[0027] The product of each frequency in the wind load spectrum diagram and its corresponding amplitude is obtained respectively, and the accumulated value of the corresponding products is used as the wind load change degree of any subsequence.

[0028] Preferably, obtaining the regression weight of each temperature and wind pressure influence coefficient according to the time interval between the sampling moment of each subsequence and the initial sampling moment of the multidimensional monitoring data sequence includes:

[0029] The sampling time of the first multidimensional monitoring data in each subsequence is recorded as the sampling time of each subsequence;

[0030] For any temperature and wind pressure influence coefficient, the two subsequences corresponding to the said temperature and wind pressure influence coefficient are recorded as target subsequences, and the time interval between the sampling time of each target subsequence and the initial sampling time of the multidimensional monitoring data sequence is obtained respectively. The inverse of the product of the time intervals is obtained as the accuracy of the said temperature and wind pressure influence coefficient;

[0031] The cumulative value of the accuracy of all temperature and wind pressure influence coefficients is obtained to obtain the total accuracy, and the ratio of the accuracy of any temperature and wind pressure influence coefficient to the total accuracy is obtained as the regression weight of any temperature and wind pressure influence coefficient.

[0032] Preferably, the weighted loss function is constructed according to the regression weight of each temperature and wind pressure influence coefficient, including:

[0033]

[0034] Among them, M is the weighted loss function; is the sth temperature and wind pressure influence coefficient; is the predicted value of the sth temperature and wind pressure influence coefficient; is the regression weight of the sth temperature and wind pressure influence coefficient; m is the number of temperature and wind pressure influence coefficients.

[0035] Preferably, the abnormal warning of the tightening state of any bolt according to the difference between each true coefficient and its predicted value includes:

[0036] A preset number of true coefficients are obtained from all true coefficients and recorded as reference coefficients. The degree of bolt loosening abnormality at the current sampling moment is obtained based on the difference between each reference coefficient and its predicted value.

[0037] If the abnormal degree of bolt loosening is greater than a preset abnormal degree threshold of bolt loosening, an abnormal warning is issued for the tightening state of any bolt at the current sampling moment.

[0038] Preferably, obtaining the abnormal degree of bolt loosening at the current sampling moment according to the difference between each reference coefficient and its predicted value includes:

[0039] For any reference coefficient, obtaining the perfect square difference between the reference coefficient and its predicted value to obtain an error value of the reference coefficient;

[0040] Obtaining a time interval between a sampling moment of a reference subsequence corresponding to any reference coefficient and an initial sampling moment of the multidimensional monitoring data sequence, and using the reciprocal of the time interval as a reference weight of an error value of any reference coefficient;

[0041] Obtain the weighted average of the error values ​​of all reference coefficients to obtain the degree of deviation of the multi-dimensional monitoring data at the current moment;

[0042] In the process of training the support vector regression model based on the weighted loss function, the minimum value of the weighted loss function is obtained, and the ratio of the deviation degree of the multidimensional monitoring data at the current moment to the minimum value of the weighted loss function is linearly normalized to obtain the abnormal degree of bolt loosening at the current sampling moment.

[0043] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0044] The present invention targets any bolt on a wind turbine tower, obtains multidimensional monitoring data of the bolt at each sampling moment, and obtains a multidimensional monitoring data sequence at the current sampling moment, wherein each of the multidimensional monitoring data includes pressure data, temperature data, and wind load data; uses a sliding window with a preset length and step size to divide the multidimensional monitoring data sequence into at least two subsequences that do not include the multidimensional monitoring data at the current sampling moment, and a current subsequence that includes the multidimensional monitoring data at the current sampling moment; obtains the temperature-wind pressure influence coefficient and the temperature change index between each two subsequences based on the fluctuation characteristics of the multidimensional monitoring data in each of the two subsequences; and obtains the temperature-wind pressure influence coefficient and the temperature change index between each of the two subsequences based on the sampling moment of each subsequence and the The time interval between the initial sampling moments of the multidimensional monitoring data sequence is used to obtain the regression weight of each temperature and wind pressure influence coefficient. According to the regression weight of each temperature and wind pressure influence coefficient, a weighted loss function is constructed. Based on the weighted loss function, the support vector regression model is trained to obtain the optimal prediction model, wherein the input of the optimal prediction model is the temperature change index, and the output is the temperature and wind pressure influence coefficient; the temperature and wind pressure influence coefficient between the current subsequence and each subsequence is obtained, recorded as the true coefficient, and the predicted value of each true coefficient is obtained by using the optimal prediction model. According to the difference between each true coefficient and its predicted value, an abnormal warning is issued for the tightening state of any bolt. The present invention first obtains the temperature and wind pressure influence coefficient between each two subsequences based on the fluctuation characteristics of the multidimensional monitoring data in each two subsequences, then obtains the regression weight of each temperature and wind pressure influence coefficient based on the time interval between the sampling moment of each subsequence and the initial sampling moment of the multidimensional monitoring data sequence, constructs a weighted loss function, and improves the accuracy of prediction based on the optimal prediction model obtained based on the weighted loss function. Finally, the true coefficient and predicted value between the current subsequence and each subsequence are obtained, so that according to the difference between each true coefficient and its predicted value, a more accurate bolt status can be obtained, thereby improving the accuracy of state monitoring of the wind turbine tower bolt tightening. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 This is a flow chart of a sensor-based online monitoring method for wind turbine tower fastening bolts provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0047] The embodiments of the present disclosure are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present disclosure, but should not be understood as limiting the present disclosure.

[0048] It should be noted that the terms "first," "second," and the like in the specification of the present disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0049] In order to illustrate the technical solution of the present invention, specific embodiments are provided below.

[0050] An embodiment of the present invention provides a sensor-based online monitoring system for fastening bolts of wind turbine towers, comprising a processor and a memory, wherein the processor executes a computer program in the memory to implement a sensor-based online monitoring method for fastening bolts of wind turbine towers, such as Figure 1 As shown, the method includes the following steps:

[0051] Step S101: For any bolt on the wind turbine tower, obtain multidimensional monitoring data of the bolt at each sampling moment to obtain a multidimensional monitoring data sequence at the current sampling moment, wherein each multidimensional monitoring data includes pressure data, temperature data and wind load data.

[0052] The wind turbine tower is a key load-bearing component that supports the upper structure of a wind turbine generator (including the nacelle, generator, blades, etc.). Its main function is to reliably transfer the mechanical load generated by wind energy conversion to the foundation, ensuring the stable operation of the wind turbine under different wind speeds and operating conditions. The connection and fixation of the wind turbine tower to the foundation or other components usually requires bolts. However, the bolts fixed in the wind turbine tower structure may loosen, weaken, or even break due to factors such as vibration and impact caused by long-term wind loads, thermal expansion and contraction caused by temperature changes, material fatigue, insufficient installation preload, or corrosion and wear. This can affect the safety and stability of the wind turbine's overall structure. Therefore, it is necessary to monitor the tightening status of the bolts in the wind turbine tower.

[0053] With the rapid development of sensor technology, wireless communications, and the Internet of Things (IoT), sensor-based structural health monitoring systems have become an important tool for the intelligent operation and maintenance of wind turbine equipment. Existing monitoring systems install micro-sensors such as strain gauges and piezoelectric sensors at key tower connections to collect real-time data on bolt force and vibration, enabling monitoring of bolt tightening status. However, wind turbine tower bolts are typically located at a high altitude and subject to variable environmental loads. The pressure on the bolts may be affected by environmental interference factors such as temperature and wind load, resulting in a decrease in the accuracy and stability of the monitoring system's bolt tightening status results, making it prone to false alarms or missed alarms.

[0054] Therefore, this embodiment divides the collected multidimensional monitoring data sequence into multiple subsequences and a current subsequence, obtains the temperature and wind pressure influence coefficient and temperature change index between every two subsequences, and then obtains the regression weight of each temperature and wind pressure influence coefficient, constructs a weighted loss function, and based on the weighted loss function, trains the support vector regression model to obtain the optimal prediction model, and finally obtains the true coefficient and predicted value between the current subsequence and each subsequence. According to the difference between each true coefficient and its predicted value, an abnormal warning is issued for the tightening status of the bolt, thereby improving the accuracy of monitoring the tightening status of the wind turbine tower bolts.

[0055] Because the pressure changes observed during wind turbine tower bolt monitoring are primarily influenced by external environmental factors such as wind load and temperature, when the wind turbine tower vibrates under strong winds, the bolts at the connection points will be subjected to periodic or impact loads caused by structural vibrations, causing their stress state to fluctuate. At the same time, due to the thermal expansion and contraction characteristics of the materials of the bolts and their connecting components, in environments with large temperature fluctuations, even under the same wind load conditions, the actual pressure on the bolts may show different changes due to differences in thermal stress. Therefore, fluctuations in the pressure data on the bolts not only reflect the wind load response of the wind turbine structure, but are also significantly affected by changes in ambient temperature.

[0056] Therefore, this embodiment targets any bolt on the wind turbine tower and obtains the pressure data and temperature data of any bolt at each sampling moment through the smart gasket between the bolt and the object. The wind load is mainly related to the wind speed. In order to simplify the analysis, it is approximately considered in engineering analysis that the wind load is proportional to the square of the wind speed, that is, the square of the wind speed can indirectly reflect the magnitude of the wind load on the wind turbine tower. Therefore, this embodiment obtains the wind speed data at each sampling moment, and uses the square of the wind speed data at each sampling moment as the wind load data at each sampling moment. The pressure data, temperature data and wind load data at each sampling moment are combined into multi-dimensional monitoring data. In this embodiment, the acquisition frequency is set to once per second, and multi-dimensional monitoring data within a week including the current sampling moment is obtained to form a multi-dimensional monitoring data sequence for subsequent analysis of the tightening status of any bolt. There is no restriction here and it can be set according to the specific implementation scenario.

[0057] Step S102: Using a sliding window of preset length and step size, the multidimensional monitoring data sequence is divided into at least two subsequences excluding the multidimensional monitoring data at the current sampling moment, and a current subsequence including the multidimensional monitoring data at the current sampling moment. Based on the fluctuation characteristics of the multidimensional monitoring data in each of the two subsequences, the temperature and wind pressure influence coefficient and the temperature change index between each of the two subsequences are obtained.

[0058] After obtaining the multidimensional monitoring data sequence, a sliding window with a preset length of ten minutes and a step size of five minutes is constructed in the multidimensional monitoring data sequence (that is, the sliding window includes 600 multidimensional monitoring data, and the step size of the sliding window is 300 multidimensional monitoring data). There is no restriction here. According to the specific implementation scenario setting, the sliding window can be used to slide on the multidimensional monitoring data sequence to divide the multidimensional monitoring data sequence into at least two subsequences that do not include the multidimensional monitoring data at the current sampling moment, and a current subsequence that includes the multidimensional monitoring data at the current sampling moment.

[0059] The change in the pressure on the bolt is related to the wind load on the wind turbine tower it fixes. Therefore, for any subsequence in the multidimensional monitoring data sequence, the wind pressure response coefficient of any subsequence can be obtained based on the difference in the fluctuation characteristics of the pressure data and the fluctuation characteristics of the wind load data in any subsequence, reflecting the degree of influence of the wind load data on the pressure data in any subsequence.

[0060] Among them, according to the difference between the fluctuation characteristics of the pressure data and the fluctuation characteristics of the wind load data in any subsequence, the method for obtaining the wind pressure response coefficient of any subsequence is as follows:

[0061] (1) Obtaining the pressure change degree of any subsequence according to the fluctuation characteristics of the pressure data in any subsequence.

[0062] Specifically, a fast Fourier transform is performed on the pressure data in any subsequence to obtain a pressure spectrum of the pressure data in any subsequence, wherein the abscissa of the pressure spectrum is frequency and the ordinate is amplitude. The fast Fourier transform belongs to the prior art and will not be described in detail here.

[0063] The product of each frequency in the pressure spectrum and its corresponding amplitude is obtained respectively, and the accumulated value of the product is obtained as the pressure change degree of any subsequence.

[0064] In one embodiment, the Take the subsequence as an example to obtain the pressure spectrum. The calculation formula for the pressure change degree of a subsequence is:

[0065]

[0066] in, For the The degree of pressure change of each subsequence; is the i-th frequency in the pressure spectrum; is the amplitude corresponding to the i-th frequency in the pressure spectrum; is the number of frequencies in the pressure spectrum.

[0067] It should be noted that, in the spectrum diagram, when the amplitude corresponding to a certain frequency point is larger, that is, the contribution of this frequency point to the original signal is greater, the overall change frequency of the original signal is closer to this frequency point. If the frequency of this frequency point is larger at this time, it means that the fluctuation frequency of the original signal, that is, its degree of change may be greater. Therefore, The larger the The overall change frequency of pressure data in the subsequence is The closer, The larger the The greater the fluctuation frequency of the pressure data of a subsequence, The bigger it is.

[0068] (2) Obtaining the degree of change of the wind load of any subsequence according to the fluctuation characteristics of the wind load data in any subsequence.

[0069] Specifically, performing a fast Fourier transform on the wind load data in any subsequence to obtain a wind load spectrum diagram of the wind load data in any subsequence, wherein the abscissa of the wind load spectrum diagram is frequency and the ordinate is amplitude;

[0070] The product of each frequency in the wind load spectrum diagram and its corresponding amplitude is obtained respectively, and the accumulated value of the corresponding products is used as the wind load change degree of any subsequence.

[0071] In one embodiment, according to the above The method for obtaining the pressure change degree of the subsequence is to obtain the The degree of wind load variation of each subsequence.

[0072] (3) Substitute the difference between the pressure change degree and the wind load change degree of any subsequence into an exponential function with a natural constant as the base to obtain the wind pressure response coefficient of any subsequence.

[0073] In one embodiment, the For example, the subsequence The calculation formula of the wind pressure response coefficient of a subsequence is:

[0074]

[0075] in, For the The wind pressure response coefficient of each subsequence; For the The degree of pressure change of each subsequence; For the The degree of wind load variation of each subsequence; e is a natural constant.

[0076] It should be noted that For the The difference between the pressure change degree of each subsequence and the wind load change degree, The larger the The pressure variation of each subsequence is greater than that of the wind load. At this time, the fluctuation of wind load data has a greater impact on the pressure data. The bigger it is.

[0077] According to the above A method for obtaining the wind pressure response coefficient of each subsequence is provided to obtain the wind pressure response coefficient of each subsequence.

[0078] When the temperature of the bolt remains unchanged, the wind pressure response coefficient corresponding to the bolt may be in a certain stable state. However, since the wind turbine tower is located in a natural environment, the temperature to which the bolt is subjected will also change with changes in the natural environment, so the wind pressure response coefficient corresponding to the bolt will also change.

[0079] Therefore, for any two subsequences, the temperature-wind pressure influence coefficient between any two subsequences can be obtained based on the fluctuation characteristics of the temperature data in any two subsequences and the wind pressure response coefficients of any two subsequences, reflecting the degree of influence of the change in the temperature data of any two subsequences on the wind pressure response coefficient. Then, the support vector regression model is trained according to the temperature-wind pressure influence coefficients of every two subsequences to obtain the optimal prediction model.

[0080] The method for obtaining the temperature-wind pressure influence coefficient between any two subsequences is as follows, based on the fluctuation characteristics of the temperature data in any two subsequences and the wind pressure response coefficients of any two subsequences:

[0081] (1) According to the fluctuation characteristics of the temperature data in any two subsequences, the temperature change index between any two subsequences is obtained.

[0082] Specifically, the temperature mean values ​​of the arbitrary two subsequences are respectively obtained, and the temperature change index between the arbitrary two subsequences is obtained according to the difference between the temperature mean values ​​of the arbitrary two subsequences.

[0083] In one embodiment, the Taking the bth subsequence and the bth subsequence as an example, the The calculation formula for the temperature change index between the bth subsequence and the bth subsequence is:

[0084]

[0085] in, For the The temperature change index between the bth subsequence and the bth subsequence; For the The mean temperature of the subsequences; is the mean temperature of the b-th subsequence.

[0086] (2) According to the temperature change index of any two subsequences and the wind pressure response coefficient of any two subsequences, the temperature and wind pressure influence coefficient between the any two subsequences is obtained.

[0087] Specifically, when the temperature change index between any two subsequences is 0, the difference in wind pressure response coefficients between the any two subsequences is obtained to obtain the temperature-wind pressure influence coefficient between the any two subsequences;

[0088] When the temperature change index between the arbitrary two subsequences is not 0, the ratio of the difference between the wind pressure response coefficients of the arbitrary two subsequences and the temperature change index between the arbitrary two subsequences is obtained to obtain the temperature and wind pressure influence coefficient between the arbitrary two subsequences.

[0089] In one embodiment, the Taking the bth subsequence and the bth subsequence as an example, the The calculation formula for the temperature and wind pressure influence coefficient between the bth subsequence and the bth subsequence is:

[0090]

[0091] in, For the The temperature and wind pressure influence coefficient between the bth subsequence and the bth subsequence; For the The wind pressure response coefficient of each subsequence; is the wind pressure response coefficient of the bth subsequence; For the The temperature change index between the bth subsequence and the bth subsequence.

[0092] It should be noted that Indicates the The change in wind pressure response coefficient between the bth subsequence and the bth subsequence is hour, The larger the The greater the impact of the change in temperature data between the first and second subsequences on the wind pressure response coefficient, the greater the impact of the change in temperature data between the first and second subsequences on the wind pressure response coefficient. When The temperature data does not change between the subsequence b and the subsequence b. Theoretically, hour, But if hour, , it may indicate that The bolts are loose during the time period corresponding to the bth subsequence or the bth subsequence.

[0093] According to the above A method for obtaining the temperature and wind pressure influence coefficient and temperature change index between the bth subsequence and the bth subsequence is used to obtain the temperature and wind pressure influence coefficient and temperature change index between every two subsequences.

[0094] At this point, the temperature and wind pressure influence coefficient and temperature change index between every two subsequences are obtained.

[0095] Step S103: Obtain the regression weight of each temperature and wind pressure influence coefficient according to the time interval between the sampling time of each subsequence and the initial sampling time of the multidimensional monitoring data sequence; construct a weighted loss function according to the regression weight of each temperature and wind pressure influence coefficient; and train the support vector regression model based on the weighted loss function to obtain the optimal prediction model, wherein the input of the optimal prediction model is the temperature change index, and the output is the temperature and wind pressure influence coefficient.

[0096] After obtaining the temperature-wind pressure influence coefficient and temperature change index between each two subsequences, a loss function can be constructed based on the temperature-wind pressure influence coefficient and temperature change index between each two subsequences, and the support vector regression model can be trained to obtain the optimal prediction model and the predicted value of the temperature-wind pressure influence coefficient. Then, based on the difference between each temperature-wind pressure influence coefficient and its predicted value, an abnormal warning of the tightening status of the bolts can be issued, thereby improving the accuracy of monitoring the tightening status of the wind turbine tower bolts.

[0097] Since in the support vector regression model, the weight of each data in the regression process is the same, and the possibility of loosening of the bolt during use is strongly related to time, the earlier the multidimensional monitoring data is obtained, the better the tightening state of the bolt may be, and the more accurate the temperature and wind pressure influence coefficient obtained. Therefore, the sampling time of the first multidimensional monitoring data in each subsequence can be recorded as the sampling time of each subsequence. According to the time interval between the sampling time of each subsequence and the initial sampling time of the multidimensional monitoring data sequence, the regression weight of each temperature and wind pressure influence coefficient is obtained, and then the optimal prediction model is constructed according to the regression weight of each temperature and wind pressure influence coefficient.

[0098] Among them, the method for obtaining the regression weight of each temperature and wind pressure influence coefficient according to the time interval between the sampling time of each subsequence and the initial sampling time of the multidimensional monitoring data sequence is as follows:

[0099] For any temperature and wind pressure influence coefficient, the two subsequences corresponding to the said temperature and wind pressure influence coefficient are recorded as target subsequences, and the time interval between the sampling time of each target subsequence and the initial sampling time of the multidimensional monitoring data sequence is obtained respectively. The inverse of the product of the time intervals is obtained as the accuracy of the said temperature and wind pressure influence coefficient;

[0100] The cumulative value of the accuracy of all temperature and wind pressure influence coefficients is obtained to obtain the total accuracy, and the ratio of the accuracy of any temperature and wind pressure influence coefficient to the total accuracy is obtained as the regression weight of any temperature and wind pressure influence coefficient.

[0101] In one embodiment, taking the sth temperature and wind pressure response coefficient as an example, the calculation formula for the regression weight of the sth temperature and wind pressure influence coefficient is:

[0102]

[0103] in, is the regression weight of the sth temperature and wind pressure influence coefficient; is the time interval between the sampling time of the first subsequence in the two subsequences corresponding to the s-th temperature and wind pressure response coefficient and the initial sampling time of the multidimensional monitoring data sequence; is the time interval between the sampling time of the second subsequence in the two subsequences corresponding to the sth temperature and wind pressure response coefficient and the initial sampling time of the multidimensional monitoring data sequence; m is the number of temperature and wind pressure response coefficients.

[0104] It should be noted that and The smaller it is, the closer the sampling times of the two subsequences corresponding to the s-th temperature and wind pressure response coefficient are to the initial sampling time of the multidimensional monitoring data sequence, the smaller the possibility of bolt loosening is, and the more accurate the s-th temperature and wind pressure response coefficient is. The bigger it is.

[0105] Similarly, the regression weight of each temperature and wind pressure influence coefficient is obtained.

[0106] Furthermore, a weighted loss function is constructed based on the regression weight of each temperature and wind pressure influence coefficient:

[0107]

[0108] Among them, M is the weighted loss function; is the sth temperature and wind pressure influence coefficient; is the predicted value of the sth temperature and wind pressure influence coefficient; is the regression weight of the sth temperature and wind pressure influence coefficient; m is the number of temperature and wind pressure influence coefficients.

[0109] After obtaining the weighted loss function, the support vector regression model is trained based on the weighted loss function. Since in the process of training the support vector regression model, the smaller the loss function, the higher the accuracy of the obtained prediction model, so in the process of training the support vector regression model based on the weighted loss function, when the difference between two adjacent M is less than the difference threshold, it is confirmed that the support vector regression model has reached the optimal state, and the minimum value of the two adjacent M is taken as the minimum value of the weighted loss function. Based on the minimum value of the weighted loss function, the optimal prediction model is obtained. The input of the optimal prediction model is the temperature change index, and the output is the temperature and wind pressure influence coefficient. The process of training the support vector regression model to the optimal prediction model belongs to the existing technology and will not be repeated here. In this embodiment, the difference threshold is set to 0.03 based on the empirical value. There is no restriction here and it can be set according to the specific implementation scenario.

[0110] At this point, the optimal prediction model is obtained.

[0111] Step S104: Obtain the temperature and wind pressure influence coefficient between the current subsequence and each subsequence, record it as the true coefficient, use the optimal prediction model to obtain the predicted value of each true coefficient, and issue an abnormal warning for the tightening state of any bolt based on the difference between each true coefficient and its predicted value.

[0112] Furthermore, according to the above The method for obtaining the temperature and wind pressure influence coefficient between the bth subsequence and the bth subsequence is to obtain the temperature and wind pressure influence coefficient between the current subsequence and each subsequence respectively, and record the temperature and wind pressure influence coefficient between the current subsequence and each subsequence as the true coefficient; according to the above The temperature change index between the bth subsequence and the bth subsequence is obtained, and the temperature change index between the current subsequence and each subsequence is obtained respectively; then the temperature change index between the current subsequence and each subsequence is input into the optimal prediction model to obtain the predicted value of each true coefficient, and then according to the difference between each true coefficient and its predicted value, an abnormal warning is issued for the tightening state of any bolt. Using the optimal prediction model to obtain the predicted value of each temperature and wind pressure influence coefficient belongs to the existing technology and will not be repeated here.

[0113] Among them, the method of providing abnormal warning for the tightening state of any bolt based on the difference between each true coefficient and its predicted value is as follows:

[0114] (1) During the use of bolts, the earlier the multi-dimensional monitoring data is obtained, the better the tightening state of the bolts may be, and the more accurate the temperature and wind pressure influence coefficients obtained are. Therefore, a preset number of real coefficients are obtained from all the real coefficients and recorded as reference coefficients. In this embodiment, the preset number is set to 20, which is not limited here and can be set according to the specific implementation scenario. According to the difference between each reference coefficient and its predicted value, the abnormal degree of bolt loosening at the current sampling moment is obtained.

[0115] Specifically, for any reference coefficient, a perfect square difference between the reference coefficient and its predicted value is obtained to obtain an error value of the reference coefficient;

[0116] Obtaining a time interval between a sampling moment of a reference subsequence corresponding to any reference coefficient and an initial sampling moment of the multidimensional monitoring data sequence, and using the reciprocal of the time interval as a reference weight of an error value of any reference coefficient;

[0117] Obtain the weighted average of the error values ​​of all reference coefficients to obtain the degree of deviation of the multi-dimensional monitoring data at the current moment;

[0118] In the process of training the support vector regression model based on the weighted loss function, the minimum value of the weighted loss function is obtained, and the ratio of the deviation degree of the multidimensional monitoring data at the current moment to the minimum value of the weighted loss function is linearly normalized to obtain the abnormal degree of bolt loosening at the current sampling moment.

[0119] In one embodiment, the calculation formula for the abnormal degree of bolt loosening at the current sampling moment is:

[0120]

[0121] Among them, G is the abnormal degree of bolt loosening at the current sampling moment; is the jth reference coefficient; is the predicted value of the jth reference coefficient; is the time interval between the sampling time of the reference subsequence corresponding to the j-th reference coefficient and the initial sampling time of the multidimensional monitoring data sequence; is the minimum value of the weighting function; Z is the number of reference subsequences (that is, the number of reference coefficients); is the normalization function.

[0122] It should be noted that is the reference weight of the j-th reference coefficient, The smaller it is, the higher the accuracy of the j-th reference coefficient is. The larger it is, the more accurate G is; is the error value of the jth reference coefficient, The larger it is, the greater the difference between the jth reference coefficient and its predicted value, the more likely the bolt at the current sampling moment is abnormal, and the larger G is; is the deviation degree of the multi-dimensional monitoring data at the current moment, The bigger, The larger it is, the more the multi-dimensional monitoring data at the current sampling moment deviates from the normal state, the greater the probability that the bolt is loose at the current sampling moment, and the larger G is.

[0123] (2) Provide an abnormal warning of the tightening status of any bolt based on the abnormal degree of bolt loosening.

[0124] According to the empirical value, the threshold value of the abnormal degree of bolt loosening is set to 0.2. There is no restriction here and it can be set according to the specific implementation scenario. If the abnormal degree of bolt loosening is greater than 0.2, it is confirmed that the bolt is loose at this time, which may affect the stability of the wind turbine. At the current sampling moment, an abnormal warning is issued for the tightening status of any bolt, and the warning information is sent to the relevant operation and maintenance management personnel, so that the operation and maintenance management personnel can tighten and maintain the wind turbine tower bolts that may be loose in time, thereby improving the stability of the wind turbine tower fixation and reducing the probability of accidents and potential safety hazards.

[0125] In summary, the embodiment of the present invention obtains the multidimensional monitoring data of any bolt on the wind turbine tower at each sampling moment, and obtains a multidimensional monitoring data sequence at the current sampling moment, wherein each of the multidimensional monitoring data includes pressure data, temperature data and wind load data; uses a sliding window with a preset length and step size to divide the multidimensional monitoring data sequence into at least two subsequences that do not include the multidimensional monitoring data at the current sampling moment, and a current subsequence that includes the multidimensional monitoring data at the current sampling moment; according to the fluctuation characteristics of the multidimensional monitoring data in each of the two subsequences, the temperature-wind pressure influence coefficient and the temperature change index between each of the two subsequences are obtained; according to the sampling time of each subsequence The time interval between the current subsequence and the initial sampling moment of the multidimensional monitoring data sequence is used to obtain the regression weight of each temperature and wind pressure influence coefficient, and a weighted loss function is constructed according to the regression weight of each temperature and wind pressure influence coefficient. Based on the weighted loss function, the support vector regression model is trained to obtain the optimal prediction model, wherein the input of the optimal prediction model is the temperature change index, and the output is the temperature and wind pressure influence coefficient; the temperature and wind pressure influence coefficient between the current subsequence and each subsequence is obtained, recorded as the true coefficient, and the predicted value of each true coefficient is obtained by using the optimal prediction model. According to the difference between each true coefficient and its predicted value, an abnormal warning is issued for the tightening state of any bolt. The embodiment of the present invention first obtains the temperature and wind pressure influence coefficient between each two subsequences based on the fluctuation characteristics of the multidimensional monitoring data in each two subsequences, and then obtains the regression weight of each temperature and wind pressure influence coefficient based on the time interval between the sampling moment of each subsequence and the initial sampling moment of the multidimensional monitoring data sequence, constructs a weighted loss function, and improves the accuracy of prediction based on the optimal prediction model obtained by the weighted loss function. Finally, the true coefficient and predicted value between the current subsequence and each subsequence are obtained, so that according to the difference between each true coefficient and its predicted value, a more accurate bolt status can be obtained, thereby improving the accuracy of state monitoring of the wind turbine tower bolt tightening.

[0126] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A sensor-based online monitoring system for wind turbine tower fastening bolts, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the following method is implemented: For any bolt on the wind turbine tower, obtain multidimensional monitoring data of the bolt at each sampling time to obtain a multidimensional monitoring data sequence at the current sampling time, wherein each multidimensional monitoring data includes pressure data, temperature data, and wind load data; Using a sliding window of preset length and step size, the multidimensional monitoring data sequence is divided into at least two subsequences excluding the multidimensional monitoring data at the current sampling time, and a current subsequence including the multidimensional monitoring data at the current sampling time, and based on the fluctuation characteristics of the multidimensional monitoring data in each of the two subsequences, the temperature and wind pressure influence coefficient and the temperature change index between each of the two subsequences are obtained; Obtaining a regression weight for each temperature and wind pressure influence coefficient based on the time interval between the sampling moment of each subsequence and the initial sampling moment of the multidimensional monitoring data sequence, constructing a weighted loss function based on the regression weight of each temperature and wind pressure influence coefficient, and training a support vector regression model based on the weighted loss function to obtain an optimal prediction model, wherein the input of the optimal prediction model is the temperature change index and the output is the temperature and wind pressure influence coefficient; The temperature and wind pressure influence coefficient between the current subsequence and each subsequence is obtained and recorded as a true coefficient. The predicted value of each true coefficient is obtained using the optimal prediction model. Based on the difference between each true coefficient and its predicted value, an abnormal warning is issued for the tightening state of any bolt.

2. The sensor-based online monitoring system for wind turbine tower fastening bolts according to claim 1 is characterized in that: The method of obtaining the temperature and wind pressure influence coefficient and the temperature change index between each two subsequences based on the fluctuation characteristics of the multidimensional monitoring data in each two subsequences includes: For any two subsequences, respectively obtain the temperature mean values ​​of the arbitrary two subsequences, and obtain the temperature change index between the arbitrary two subsequences according to the difference between the temperature mean values ​​of the arbitrary two subsequences; The temperature and wind pressure influence coefficient between the arbitrary two subsequences is obtained according to the temperature change indexes of the arbitrary two subsequences and the fluctuation characteristics of the multi-dimensional monitoring data in the arbitrary two subsequences.

3. The sensor-based online monitoring system for wind turbine tower fastening bolts according to claim 2, characterized in that: The obtaining of the temperature and wind pressure influence coefficient between the arbitrary two subsequences according to the temperature change indexes of the arbitrary two subsequences and the fluctuation characteristics of the multidimensional monitoring data in the arbitrary two subsequences includes: For any subsequence of the arbitrary two subsequences, obtaining a wind pressure response coefficient of the subsequence according to a difference between a fluctuation characteristic of the pressure data and a fluctuation characteristic of the wind load data in the subsequence; Obtaining the wind pressure response coefficient of each subsequence in the arbitrary two subsequences respectively; when the temperature change index between the arbitrary two subsequences is 0, obtaining the difference between the wind pressure response coefficients of the arbitrary two subsequences to obtain the temperature-wind pressure influence coefficient between the arbitrary two subsequences; When the temperature change index between the arbitrary two subsequences is not 0, the ratio of the difference between the wind pressure response coefficients of the arbitrary two subsequences and the temperature change index between the arbitrary two subsequences is obtained to obtain the temperature and wind pressure influence coefficient between the arbitrary two subsequences.

4. The sensor-based online monitoring system for wind turbine tower fastening bolts according to claim 3 is characterized in that: The obtaining of the wind pressure response coefficient of any subsequence according to the difference between the fluctuation characteristics of the pressure data and the fluctuation characteristics of the wind load data in any subsequence includes: Obtaining a pressure change degree of any subsequence according to a fluctuation characteristic of the pressure data in any subsequence; Obtaining a degree of change in the wind load of any subsequence according to a fluctuation characteristic of the wind load data in any subsequence; The difference between the pressure variation degree and the wind load variation degree of any subsequence is substituted into an exponential function with a natural constant as a base to obtain the wind pressure response coefficient of any subsequence.

5. The sensor-based online monitoring system for wind turbine tower fastening bolts according to claim 4 is characterized in that: The obtaining, based on the fluctuation characteristics of the pressure data in any subsequence, the pressure change degree of any subsequence includes: Performing a fast Fourier transform on the pressure data in any subsequence to obtain a pressure spectrum of the pressure data in any subsequence, wherein the abscissa of the pressure spectrum is frequency and the ordinate is amplitude; The product of each frequency in the pressure spectrum and its corresponding amplitude is obtained respectively, and the accumulated value of the product is obtained as the pressure change degree of any subsequence.

6. The sensor-based online monitoring system for wind turbine tower fastening bolts according to claim 4, characterized in that: The obtaining, based on the fluctuation characteristics of the wind load data in any subsequence, the degree of change of the wind load in any subsequence includes: Performing a fast Fourier transform on the wind load data in any subsequence to obtain a wind load spectrum diagram of the wind load data in any subsequence, wherein the abscissa of the wind load spectrum diagram is frequency and the ordinate is amplitude; The product of each frequency in the wind load spectrum diagram and its corresponding amplitude is obtained respectively, and the accumulated value of the corresponding products is used as the wind load change degree of any subsequence.

7. The sensor-based online monitoring system for wind turbine tower fastening bolts according to claim 1, characterized in that: The step of obtaining the regression weight of each temperature and wind pressure influence coefficient according to the time interval between the sampling moment of each subsequence and the initial sampling moment of the multidimensional monitoring data sequence includes: The sampling time of the first multidimensional monitoring data in each subsequence is recorded as the sampling time of each subsequence; For any temperature and wind pressure influence coefficient, the two subsequences corresponding to the said temperature and wind pressure influence coefficient are recorded as target subsequences, and the time interval between the sampling time of each target subsequence and the initial sampling time of the multidimensional monitoring data sequence is obtained respectively. The inverse of the product of the time intervals is obtained as the accuracy of the said temperature and wind pressure influence coefficient; The cumulative value of the accuracy of all temperature and wind pressure influence coefficients is obtained to obtain the total accuracy, and the ratio of the accuracy of any temperature and wind pressure influence coefficient to the total accuracy is obtained as the regression weight of any temperature and wind pressure influence coefficient.

8. The sensor-based online monitoring system for wind turbine tower fastening bolts according to claim 1, characterized in that: The weighted loss function is constructed based on the regression weight of each temperature and wind pressure influence coefficient, including: Among them, M is the weighted loss function; is the sth temperature and wind pressure influence coefficient; is the predicted value of the sth temperature and wind pressure influence coefficient; is the regression weight of the sth temperature and wind pressure influence coefficient; m is the number of temperature and wind pressure influence coefficients.

9. The sensor-based online monitoring system for wind turbine tower fastening bolts according to claim 7, characterized in that: The step of providing an abnormal warning for the tightening state of any bolt according to the difference between each true coefficient and its predicted value includes: A preset number of true coefficients are obtained from all true coefficients and recorded as reference coefficients. The degree of bolt loosening abnormality at the current sampling moment is obtained based on the difference between each reference coefficient and its predicted value. If the abnormal degree of bolt loosening is greater than a preset abnormal degree threshold of bolt loosening, an abnormal warning is issued for the tightening state of any bolt at the current sampling moment.

10. The sensor-based online monitoring system for wind turbine tower fastening bolts according to claim 9, characterized in that: The abnormal degree of bolt loosening at the current sampling moment is obtained based on the difference between each reference coefficient and its predicted value, including: For any reference coefficient, obtaining the perfect square difference between the reference coefficient and its predicted value to obtain an error value of the reference coefficient; Obtaining a time interval between a sampling moment of a reference subsequence corresponding to any reference coefficient and an initial sampling moment of the multidimensional monitoring data sequence, and using the reciprocal of the time interval as a reference weight of an error value of any reference coefficient; Obtain the weighted average of the error values ​​of all reference coefficients to obtain the degree of deviation of the multi-dimensional monitoring data at the current moment; In the process of training the support vector regression model based on the weighted loss function, the minimum value of the weighted loss function is obtained, and the ratio of the deviation degree of the multidimensional monitoring data at the current moment to the minimum value of the weighted loss function is linearly normalized to obtain the abnormal degree of bolt loosening at the current sampling moment.

Images