A method and device for identifying abnormal state of wind turbine generators based on neural network

Abstract

The present invention relates to a method for identifying abnormal states of wind turbines based on a neural network, comprising: S1, continuously sampling operating variables of each wind turbine through a sliding window to obtain a plurality of time series data; S2, performing data preprocessing on each time series data; S3, pre-building and training a prediction model; inputting each time series data into the prediction model to obtain a corresponding prediction interval; S4, calculating the proportion of each time series data in the prediction interval; if the proportion exceeds an expected proportion, it is considered that a monitoring object corresponding to the time series data has an abnormality; if the proportion does not exceed the expected proportion, it is considered that the monitoring object corresponding to the time series data operates normally; S5, updating the window size according to the number and distribution characteristics of sampling points in the window and the proportion of the number of sampling points falling into the prediction interval, and returning to step S1.

Description

Technical Field

[0001] The invention relates to a method and equipment for identifying abnormal states of wind turbine generator sets based on neural networks, and belongs to the field of wind turbine generator set fault detection. Background Art

[0002] Wind turbine condition monitoring technology collects extensive information from various parts and subsystems of a wind turbine, enabling operators and maintenance personnel to take appropriate measures before a turbine shuts down or fails. Currently, Supervisory Control and Data Acquisition (SCADA) systems are widely used industrial software. Wind turbines typically include sensors that collect physical quantities such as temperature and pressure, which then transmit this information to SCADA systems. This eliminates the need for additional equipment and allows for a wide range of applications. However, operational data from different turbines exhibits varying sampling periods and probability distribution characteristics. As turbine performance degrades, data distribution becomes increasingly complex, placing high demands on the versatility, accuracy, computational efficiency, stability, and engineering applicability of data cleaning methods.

[0003] Existing technology standardizes monitoring data by averaging, aggregating, or decomposing data of different attributes at a consistent time interval (10 minutes is used). This standardized data is then mapped into the standardized data structure of the analysis software. Proper data cleaning and standardization are crucial to the accuracy of data analysis results. For more information, see "Refined Analysis of Wind Farm and Wind Turbine Operational Data" by Li Yuewen and Zhang Lan.

[0004] In summary, a more reliable monitoring data processing method is needed to improve the accuracy of abnormal state identification.

[0005] Patent CN106503794A "A method for predicting the remaining life of a wind turbine gearbox" discloses: obtaining historical data on the oil temperature of the wind turbine gearbox, selecting data samples, which are the gearbox oil temperature data collected at intervals at the beginning of the first maintenance cycle of the gearbox, and using a BP neural network to predict the future oil temperature of the gearbox, thereby deriving the remaining life of the gearbox corresponding to different oil temperatures of the gearbox in the first maintenance cycle. Summary of the Invention

[0006] In order to overcome the problems existing in the prior art, the present invention designs a method for identifying abnormal states of wind turbines based on a neural network. According to the number of sampling points in the window, the distribution characteristics and the proportion of the number of sampling points falling into the prediction interval, the window size is dynamically adjusted to improve the accuracy of the prediction results, thereby improving the accuracy of abnormal state identification.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] Technical solution 1:

[0009] A method for identifying abnormal states of wind turbines based on a neural network comprises the following steps:

[0010] S1. Continuously sample the operating variables of each wind turbine through a sliding window to obtain a number of time series data;

[0011] S2. Preprocess the time series data.

[0012] S3. Pre-build and train a prediction model; input each time series data into the prediction model to obtain the corresponding prediction interval;

[0013] S4. Calculate the proportion of the time series data in the prediction interval; if the proportion exceeds the expected proportion, it is considered that the monitoring object corresponding to the time series data is abnormal; if the proportion does not exceed the expected proportion, it is considered that the monitoring object corresponding to the time series data is operating normally;

[0014] S5. Update the window size according to the number of sampling points in the window, distribution characteristics, and the proportion of sampling points falling into the prediction interval, and return to step S1.

[0015] Furthermore, the update window size is expressed as follows:

[0016]

[0017]

[0018] Where L represents the window size; T b represents the sampling period reference value; T0 represents the actual value of the sampling period; p represents the expected value of the ratio of the number of sampling points falling into the prediction interval to the total number of sampling points; E represents the allowable error; μ represents the true value of the ratio of the number of sampling points falling into the prediction interval to the total number of sampling points; N s Indicates the number of sampling points in the window.

[0019] Furthermore, the data preprocessing of each time series data includes:

[0020] Normalize the time series data and express it in the following formula:

[0021]

[0022] Where x i 、x new Respectively represent the values ​​of the i-th sampling point in the sequence before and after normalization; x max 、x minRepresent the maximum and minimum values ​​in the sequence respectively.

[0023] Furthermore, the data preprocessing of each time series data includes:

[0024] Taking the orthogonal wavelet basis as the decomposition basis, determine the decomposition layer k; perform wavelet transform on the time series data to obtain k groups of wavelet coefficients; set the wavelet coefficients that do not exceed the threshold to zero, and reduce the coefficients that exceed the threshold to make them close to zero; reconstruct the time series data through inverse wavelet transform of the wavelet coefficients to obtain the denoised time series data.

[0025] Furthermore, the corresponding prediction interval is obtained, including:

[0026] The preprocessed time series data is input into the trained prediction model to obtain multiple prediction values; the prediction interval is constructed with the minimum value of the prediction value as the lower bound of the interval and the maximum value as the upper bound of the interval.

[0027] Technical solution 2:

[0028] A device for identifying abnormal conditions of a wind turbine generator system based on a neural network includes a processor and a memory for storing executable instructions of the processor; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the following steps:

[0029] S1. Continuously sample the operating variables of each wind turbine through a sliding window to obtain a number of time series data;

[0030] S2. Preprocess the time series data.

[0031] S3. Pre-build and train a prediction model; input each time series data into the prediction model to obtain the corresponding prediction interval;

[0032] S4. Calculate the proportion of the time series data in the prediction interval; if the proportion exceeds the expected proportion, it is considered that the monitoring object corresponding to the time series data is abnormal; if the proportion does not exceed the expected proportion, it is considered that the monitoring object corresponding to the time series data is operating normally;

[0033] S5. Update the window size according to the number of sampling points in the window, distribution characteristics, and the proportion of sampling points falling into the prediction interval, and return to step S1.

[0034] Furthermore, the update window size is expressed as follows:

[0035]

[0036]

[0037] Where L represents the window size; T brepresents the sampling period reference value; T0 represents the actual value of the sampling period; p represents the expected value of the ratio of the number of sampling points falling into the prediction interval to the total number of sampling points; E represents the allowable error; μ represents the true value of the ratio of the number of sampling points falling into the prediction interval to the total number of sampling points; N s Indicates the number of sampling points in the window.

[0038] Furthermore, the data preprocessing of each time series data includes:

[0039] Normalize the time series data and express it in the following formula:

[0040]

[0041] Where x i 、x new Respectively represent the values ​​of the i-th sampling point in the sequence before and after normalization; x max 、x min Represent the maximum and minimum values ​​in the sequence respectively.

[0042] Furthermore, the data preprocessing of each time series data includes:

[0043] Taking the orthogonal wavelet basis as the decomposition basis, determine the decomposition layer k; perform wavelet transform on the time series data to obtain k groups of wavelet coefficients; set the wavelet coefficients that do not exceed the threshold to zero, and reduce the coefficients that exceed the threshold to make them close to zero; reconstruct the time series data through inverse wavelet transform of the wavelet coefficients to obtain the denoised time series data.

[0044] Furthermore, the corresponding prediction interval is obtained, including:

[0045] The preprocessed time series data is input into the trained prediction model to obtain multiple prediction values; the prediction interval is constructed with the minimum value of the prediction value as the lower bound of the interval and the maximum value as the upper bound of the interval.

[0046] Compared with the prior art, the present invention has the following characteristics and beneficial effects:

[0047] Taking into account the characteristics of different wind turbine operating data such as different sampling periods and differentiated probability distribution characteristics, the present invention dynamically adjusts the window size according to the number of sampling points in the window, the distribution characteristics and the proportion of the number of sampling points falling into the prediction interval, thereby improving the accuracy of the prediction results and thus improving the accuracy of abnormal state identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be described in more detail below with reference to the embodiments.

[0050] Example 1

[0051] like Figure 1 As shown, a method for identifying abnormal state of a wind turbine generator system based on a neural network includes the following steps:

[0052] S1. Continuously sample the operating variables of each wind turbine using a sliding window to obtain time series data. Specifically, the operating variable is the gearbox oil temperature. Initially, technicians preset an initial window size based on experience.

[0053] S2. Perform data preprocessing on each time series data, such as data normalization, noise reduction, missing value processing, etc.

[0054] S3. Pre-build and train the prediction model. Input each time series data into the prediction model to obtain the corresponding prediction interval.

[0055] S4. Calculate the proportion of the time series data in the prediction interval; if the proportion exceeds the expected proportion, it is considered that the monitoring object corresponding to the time series data is abnormal; if the proportion does not exceed the expected proportion, it is considered that the monitoring object corresponding to the time series data is operating normally.

[0056] S5. Continue to sample the operating variables of each wind turbine using a sliding window to determine whether the monitored object is operating normally. Considering that the operating data of different wind turbines have different sampling periods and different probability distribution characteristics, the window size is dynamically adjusted based on the number of sampling points in the window, the distribution characteristics, and the proportion of sampling points falling into the prediction interval. The formula is expressed as:

[0057]

[0058]

[0059] Where L represents the window size; T b represents the sampling period reference value; T0 represents the actual value of the sampling period; p represents the expected value of the ratio of the number of sampling points falling within the prediction interval to the total number of sampling points, which is set to 0.9 in this embodiment; E represents the allowable error, which is set to 0.1 in this embodiment; μ represents the true value of the ratio of the number of sampling points falling within the prediction interval to the total number of sampling points, which is determined by the latest prediction result; N s Indicates the number of sampling points in the window.

[0060] Example 2

[0061] Perform data preprocessing on each time series data, specifically:

[0062] Normalize the time series data and express it in the following formula:

[0063]

[0064] Where x i 、x new Respectively represent the values ​​of the i-th sampling point in the sequence before and after normalization; x max 、x min Represent the maximum and minimum values ​​in the sequence respectively.

[0065] The denoising function is achieved by using wavelet transform. The specific steps include: first, selecting a suitable orthogonal wavelet basis as the decomposition basis, and then selecting the number k of layers to decompose the time series data; using wavelet transform on the time series data containing noise, decomposing it into k layers, and obtaining k groups of wavelet coefficients; performing threshold processing on the decomposed wavelet coefficients, setting the wavelet coefficients with smaller values ​​to zero, and at the same time reducing the larger wavelet coefficients to a certain extent to make them close to zero; using the wavelet coefficients after threshold processing to reconstruct the time series data through inverse wavelet transform to obtain denoised time series data.

[0066] Example 3

[0067] Build a prediction model based on a BP neural network. Initialize the model weights and thresholds, divide a set of time series data into input data and target data, and perform forward iterative calculations to calculate the predicted output value. Based on the difference between the BP neural network output value and the target data, the weights and thresholds are adjusted. Based on the error between the predicted output value and the target data, the weights and thresholds are continuously adjusted through backpropagation using gradient descent until the error between the predicted output value and the target data meets the required error.

[0068] The normalized time series data is input into the trained BP neural network model to obtain a prediction sequence including multiple prediction values. The minimum value among the prediction values ​​is used as the lower bound of the interval, and the maximum value is used as the upper bound of the interval.

[0069] It should be noted that the above-mentioned neural network-based wind turbine abnormal state identification device is also used to achieve the above-mentioned Figure 1 The method steps corresponding to each embodiment of the neural network-based wind turbine abnormal state identification method shown are not repeated in this application.

[0070] It should be noted that the functional units / modules in the various embodiments of the present invention may be integrated into a single processing unit / module, each unit / module may exist physically separately, or two or more units / modules may be integrated into a single unit / module. The aforementioned integrated units / modules may be implemented in the form of hardware or software functional units / modules.

[0071] Through the description of the above embodiments, it will be clear to those skilled in the art that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a processor, a controller, a microcontroller, a microprocessor, other electronic units designed to implement the functions described herein, or a combination thereof. For software implementation, part or all of the processes of the embodiments can be completed by instructing the relevant hardware through a computer program. During implementation, the above program can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. The storage medium can be any available medium that can be accessed by a computer. Computer-readable media can include, but are not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should analyze that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for identifying abnormal state of wind turbine generator system based on neural network, characterized in that: The following steps are involved: S1. Continuously sample the operating variables of each wind turbine through a sliding window to obtain a number of time series data; S2. Preprocess the time series data. S3. Pre-build and train a prediction model; input each time series data into the prediction model to obtain the corresponding prediction interval; S4. Calculate the proportion of each time series data in the prediction interval; if the proportion exceeds the expected proportion, it is considered that the monitoring object corresponding to the time series data is abnormal; if the proportion does not exceed the expected proportion, it is considered that the monitoring object corresponding to the time series data is operating normally; S5. Update the window size according to the number of sampling points in the window, distribution characteristics, and the proportion of sampling points falling into the prediction interval, and return to step S1.

2. The method for identifying abnormal state of a wind turbine generator system based on a neural network according to claim 1, characterized in that: The update window size is expressed as follows: Where L represents the window size; T b represents the sampling period reference value; T0 represents the actual value of the sampling period; p represents the expected value of the ratio of the number of sampling points falling into the prediction interval to the total number of sampling points; E represents the allowable error; μ represents the true value of the ratio of the number of sampling points falling into the prediction interval to the total number of sampling points; N s Indicates the number of sampling points in the window.

3. The method for identifying abnormal state of a wind turbine generator system based on a neural network according to claim 1, characterized in that: The data preprocessing of each time series data includes: Normalize the time series data and express it in the following formula: Where x i 、x new Respectively represent the values ​​of the i-th sampling point in the sequence before and after normalization; x max 、x min Represent the maximum and minimum values ​​in the sequence respectively.

4. The method for identifying abnormal state of a wind turbine generator system based on a neural network according to claim 1, characterized in that: The data preprocessing of each time series data includes: Taking the orthogonal wavelet basis as the decomposition basis, determine the decomposition layer k; perform wavelet transform on the time series data to obtain k groups of wavelet coefficients; set the wavelet coefficients that do not exceed the threshold to zero, and reduce the coefficients that exceed the threshold to make them close to zero; reconstruct the time series data through inverse wavelet transform of the wavelet coefficients to obtain the denoised time series data.

5. The method for identifying abnormal state of a wind turbine generator system based on a neural network according to claim 1, characterized in that: Get the corresponding prediction interval, including: The preprocessed time series data is input into the trained prediction model to obtain multiple prediction values; the prediction interval is constructed with the minimum value of the prediction value as the lower bound of the interval and the maximum value as the upper bound of the interval.

6. A wind turbine abnormal state identification device based on neural network, characterized in that: The system comprises a processor and a memory for storing instructions executable by the processor; the processor is configured to read the instructions from the memory and execute the instructions to implement the following steps: S1. Continuously sample the operating variables of each wind turbine through a sliding window to obtain a number of time series data; S2. Preprocess the time series data. S3. Pre-build and train a prediction model; input each time series data into the prediction model to obtain the corresponding prediction interval; S4. Calculate the proportion of the time series data in the prediction interval; if the proportion exceeds the expected proportion, it is considered that the monitoring object corresponding to the time series data is abnormal; if the proportion does not exceed the expected proportion, it is considered that the monitoring object corresponding to the time series data is operating normally; S5. Update the window size according to the number of sampling points in the window, distribution characteristics, and the proportion of sampling points falling into the prediction interval, and return to step S1.

7. The neural network-based wind turbine abnormal state identification device according to claim 6, characterized in that: The update window size is expressed as follows: Where L represents the window size; T b represents the sampling period reference value; T0 represents the actual value of the sampling period; p represents the expected value of the ratio of the number of sampling points falling into the prediction interval to the total number of sampling points; E represents the allowable error; μ represents the true value of the ratio of the number of sampling points falling into the prediction interval to the total number of sampling points; N s Indicates the number of sampling points in the window.

8. The neural network-based wind turbine abnormal state identification device according to claim 6, characterized in that: The data preprocessing of each time series data includes: Normalize the time series data and express it in the following formula: Where x i 、x new Respectively represent the values ​​of the i-th sampling point in the sequence before and after normalization; x max 、x min Represent the maximum and minimum values ​​in the sequence respectively.

9. The neural network-based wind turbine abnormal state identification device according to claim 6, characterized in that: The data preprocessing of each time series data includes: Taking the orthogonal wavelet basis as the decomposition basis, determine the decomposition layer k; perform wavelet transform on the time series data to obtain k groups of wavelet coefficients; set the wavelet coefficients that do not exceed the threshold to zero, and reduce the coefficients that exceed the threshold to make them close to zero; reconstruct the time series data through inverse wavelet transform of the wavelet coefficients to obtain the denoised time series data.

10. The wind turbine abnormal state identification device based on neural network according to claim 6, characterized in that: Get the corresponding prediction interval, including: The preprocessed time series data is input into the trained prediction model to obtain multiple prediction values; the prediction interval is constructed with the minimum value of the prediction value as the lower bound of the interval and the maximum value as the upper bound of the interval.

Images