Fan blade intelligent monitoring and optimizing method based on deep learning

Abstract

The invention discloses an intelligent fan blade monitoring and optimizing method based on deep learning, and relates to the technical field of wind power generation, and the method comprises the steps: calculating a fatigue accumulation value corresponding to strain data in each historical period; predicting the current fatigue state of each local area according to the fatigue state prediction model, and generating a plurality of candidate load adjustment strategies through a pre-training reinforcement learning model according to the current fatigue state; determining short-term trend change characteristics of fatigue of the fan blade, screening and determining a target load adjusting strategy according to the difference between the short-term trend change characteristics and the current fatigue state, and generating an active load control signal to actively adjust the local area load of the fan blade; according to the method, through deep learning and multi-scale convolution feature extraction, high-precision prediction and active load adjustment of the fatigue state of the fan blade are realized, so that the service life of the fan blade is remarkably prolonged, and the operation safety of the fan blade is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a method for intelligent monitoring and optimization of wind turbine blades based on deep learning. Background Art

[0002] As a vital component of renewable energy, wind turbines' operational reliability and safety are crucial to ensuring electricity supply. During wind power generation, wind turbine blades, as core components, are subjected to complex aerodynamic loads and environmental influences over long periods of time, making them susceptible to fatigue damage. This damage can accumulate, especially at high wind speeds and during frequent starts and stops, ultimately leading to blade breakage or performance degradation. Therefore, monitoring the fatigue status of wind turbine blades and performing timely maintenance and optimization are crucial measures to improve wind turbine operational efficiency and safety.

[0003] Currently, wind turbine blade monitoring methods primarily include systems based on vibration and strain sensors and image processing technologies. However, existing monitoring systems still suffer from insufficient accuracy, unstable predictions, and an inability to accurately adjust loads in real time. These issues prevent existing technologies from fully addressing the real-time monitoring, prediction, and proactive adjustment of wind blade fatigue damage.

[0004] In response to the above technical problems, the existing technology still lacks an effective solution that can monitor the fatigue status of wind turbine blades in real time with high precision and low power consumption, and perform intelligent load adjustment based on the monitoring results. Summary of the Invention

[0005] The purpose of the present invention is to provide a system and method for intelligent monitoring and optimization of wind turbine blades based on deep learning to solve the problems in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solutions: The present invention provides a method for intelligent monitoring and optimization of wind turbine blades based on deep learning, comprising: S101: Based on the historical strain data of the local area of ​​the wind turbine blade, the fatigue accumulation value corresponding to the strain data in each historical period is calculated in combination with the fatigue accumulation rate calculation method, and the fatigue state prediction model is trained using the historical strain data as input and the fatigue accumulation value as output; S102: Predicting a current fatigue state of each local area according to the fatigue state prediction model, and generating a plurality of candidate load adjustment strategies through a pre-trained reinforcement learning model based on the current fatigue state; S103: collecting real-time vibration data of the wind turbine blades, determining short-term trend variation characteristics of wind turbine blade fatigue, and screening and determining a target load adjustment strategy from the plurality of candidate load adjustment strategies based on a difference between the short-term trend variation characteristics and a current fatigue state; S104: According to the determined target load adjustment strategy, an active load control signal is generated by the wind turbine active load control system to actively adjust the load of a local area of ​​the wind turbine blade.

[0007] In the above technical solution, the technical effects and advantages provided by the present invention are: The present invention achieves high-precision prediction of the fatigue state of wind turbine blades through a fatigue state prediction model based on deep learning; wherein, by adopting a multi-scale convolution feature extraction structure, combined with historical strain data and fatigue accumulation values, the strain characteristics of each local area of ​​the wind turbine blade can be fully captured, thereby obtaining more accurate fatigue prediction results; this improvement significantly improves the accuracy of fatigue monitoring in the existing technology, avoids the potential risks caused by insufficient strain data processing and low prediction accuracy in traditional monitoring methods, and ensures the safety and stability of wind turbine blades in long-term operation.

[0008] The present invention can realize active load regulation of wind turbine blades by constructing a load regulation strategy based on deep reinforcement learning. Specifically, by designing the reinforcement learning state space, action space and reward function, and training a pre-trained reinforcement learning model based on historical fatigue data, the system can generate multiple candidate load regulation strategies, and respond quickly according to the current fatigue state of the wind turbine blades to select the most appropriate regulation strategy. This innovation can significantly improve the service life of wind turbine blades, reduce damage or performance degradation caused by excessive fatigue, and improve the economy and stability of wind turbine generator sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0010] Figure 1 This is a flow chart of a method for intelligent monitoring and optimization of wind turbine blades based on deep learning according to the present invention; Figure 2 This is a framework diagram of a deep learning-based intelligent monitoring and optimization system for wind turbine blades in the present invention. DETAILED DESCRIPTION

[0011] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be more comprehensive and complete, and will fully convey the concepts of the example embodiments to those skilled in the art. The accompanying drawings are merely schematic illustrations of the disclosure and are not necessarily drawn to scale. Identical reference numerals in the figures indicate identical or similar parts, and thus any repetitive description thereof will be omitted.

[0012] In addition, the described features, structures or characteristics can be combined in one or more example embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the example embodiments disclosed in this application. However, those skilled in the art will appreciate that the technical solutions disclosed in this application can be practiced while omitting one or more of the specific details, or other methods, components, steps, etc. can be adopted. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring the various aspects disclosed in this application.

[0013] Example 1 like Figure 1 As shown, this embodiment discloses a method for intelligent monitoring and optimization of wind turbine blades based on deep learning, including: S101: Based on the historical strain data of the local area of ​​the wind turbine blade, the fatigue accumulation value corresponding to the strain data in each historical period is calculated in combination with the fatigue accumulation rate calculation method, and the fatigue state prediction model is trained using the historical strain data as input and the fatigue accumulation value as output; In implementation, the training method of the fatigue state prediction model includes: Collect historical strain data of a local area of ​​the wind turbine blade, segment the historical strain data according to a fixed time length, and obtain strain data samples corresponding to multiple historical time periods; It should be noted that the local area of ​​the wind turbine blade in this embodiment is a typical high-incidence fatigue area determined through finite element simulation analysis of the blade structure and statistical analysis of actual operation monitoring data, specifically including but not limited to: the connection area between the blade root and the hub; the connection area between the blade main beam and the skin; the cross-sectional area with the maximum bending moment in the middle section of the blade; and the area near the leading edge of the blade tip.

[0014] It should be further explained that, first, multiple high-precision fiber Bragg grating strain sensors (FBG sensors) are installed on the surface of each local area to continuously collect historical strain data of the wind turbine blades in real time during long-term operation. The historical strain data includes the following three specific contents: Strain amplitude: the microscopic tensile or compressive deformation of the material surface in a local area of ​​the wind blade; Strain frequency: The number of strain cycles extracted by the rainflow counting method per unit time, indicating the number of strain cycles experienced by a local area of ​​the wind turbine blade; Strain waveform signal: A two-dimensional time series data sequence consisting of sampling time points (unit: seconds) and corresponding strain amplitudes.

[0015] Subsequently, to facilitate data processing and model training, the long-term historical strain data are divided into multiple independent historical period strain data samples with a fixed time length (such as 60 minutes) as the unit to form a model training data set, where each strain data sample is in the two-dimensional data form of "number of time series data points × number of measurement points".

[0016] According to Miner's fatigue damage criterion and the fatigue SN curve of the wind turbine blade material, the fatigue cumulative value corresponding to the strain data sample in each historical period is calculated respectively; In the specific implementation, in order to obtain the target output of model training, that is, the fatigue accumulation value in each historical period, the classic Miner fatigue damage criterion is combined with the fatigue SN curve of the wind turbine blade material for calculation. The specific process is fully disclosed and clear as follows: Firstly, the rain flow counting method is used to perform cycle identification on the strain waveform signal in each historical period, and the strain amplitude and actual cycle number of each cycle are extracted.

[0017] Secondly, the fatigue accumulation value is calculated based on the Miner fatigue damage criterion calculation formula: ; Where: is the fatigue accumulation value (dimensionless), which is used to indicate the degree of fatigue damage in the local area of ​​the blade. The D value ranges from 0 to 1. When D=1, it means that the fatigue life of the material has been completely exhausted. is the actual number of occurrences of the i-th level strain cycle, obtained by the rainflow counting method; is the number of fatigue life cycles corresponding to the strain amplitude obtained based on the material fatigue SN curve; is the total number of all fatigue cycle levels, determined by the rainflow counting method.

[0018] It should be further explained that the fatigue SN curve of wind turbine blade material refers to the strain-life relationship curve obtained through fatigue testing of blade material (such as glass fiber reinforced composite material, GFRP) under different strain amplitude conditions. Its specific mathematical expression is as follows: ; Where: N is the strain amplitude The number of fatigue life cycles under the condition of fatigue life is , C and m are material parameters obtained through a large number of material fatigue tests; The fatigue performance parameters of materials determined through a large number of material fatigue tests, such as the fatigue performance parameters of GFRP materials ; Construct a multi-scale convolution feature extraction structure, perform convolution operations on strain data samples of each historical period, and extract the initial fatigue features corresponding to each scale; Specifically, the specific construction method of the multi-scale convolutional feature extraction structure includes: Construct multiple parallel convolution branches with different convolution kernel sizes. In each parallel convolution branch, set up dilated convolution layer, depthwise separable convolution layer and attention mechanism module in sequence. In specific implementation, this embodiment constructs a multi-scale convolution feature extraction structure for extracting multi-scale fatigue features from strain data samples in each historical period. It includes multiple parallel convolution branches, each corresponding to a convolution kernel size (for example, 3×3, 5×5, 7×7), and the input data dimension is 6000 time points × 8 measurement points (10-minute sampling, 10Hz sampling frequency). The specific branch structure is: Branch 1 (3×3 kernel): input 8 channels → dilated convolution (d=1) → output 64 channels → depthwise separable convolution → output 64 channels → CA module → output 64 channels; Branch 2 (5×5 kernel): input 8 channels → dilated convolution (d=2) → output 128 channels → depthwise separable convolution → output 128 channels → CA module → output 128 channels; Branch 3 (7×7 kernel): Input 8 channels → Dilated convolution (d=4) → Output 256 channels → Depthwise separable convolution → Output 256 channels → CA module → Output 256 channels.

[0019] In each parallel convolution branch, the initial convolution features of the corresponding scale are obtained by dilating the convolution layer, and then the initial convolution features are further convolved using the depthwise separable convolution layer to obtain the refined fatigue features of the corresponding scale; This embodiment uses dilated convolution for initial feature extraction, where the dilated convolution of each convolution branch is specifically defined as: ; Where: is the convolution output feature; Input for historical period strain data samples; is the convolution kernel weight; is the expansion factor, which takes values ​​of 1, 2, and 4 to obtain different receptive field characteristics; are the convolution kernel sizes, which are 3×3, 5×5, and 7×7 respectively.

[0020] In the specific implementation, in order to further extract refined fatigue features, each convolution branch sets a depth-wise separable convolution layer, which includes two steps of "depthwise convolution" and "point-wise convolution": Among them, the depth convolution operation formula is: ; Where: is the convolution output feature; is the cth channel corresponding to the input feature; is the depth convolution kernel; M and N are the depth convolution kernel sizes, specifically 3×3; point-by-point convolution uses a 1×1 convolution kernel to fuse the feature channels: ; Where: is the point-by-point convolution output feature; C is the number of input feature channels; is the point-by-point convolution kernel weight.

[0021] For the refined fatigue features obtained in each parallel convolution branch, the attention weight of the feature channel is calculated through the attention mechanism module. Based on the attention weight, each channel of the refined fatigue feature is weighted and adjusted to obtain the attention feature of each convolution branch. To strengthen the fatigue feature representation of key channels, a channel attention (CA) module is set in each parallel convolution branch, which is disclosed as follows: First, the channel features are spatially compressed through global average pooling (GAP) to obtain the global average of each channel; Then, two fully connected layers (FC) are used to perform channel feature mapping, and finally the attention weights of each feature channel are obtained through the sigmoid activation function: ; Where: is the channel attention weight; is global average pooling; is a fully connected mapping; is the corrected linear unit function; is the Sigmoid function; is the refined feature channel of the input; After obtaining the channel attention weight, the refined fatigue feature is multiplied channel by channel to obtain the attention feature of each convolution branch, which is the multi-scale fatigue feature.

[0022] The fusion weight is determined based on the correlation between fatigue features at different scales. The fused fatigue features of strain data from each historical period are obtained by fusion. The fused fatigue features are used as the model input, and the fatigue accumulation value is used as the model output to train and obtain the fatigue state prediction model. Specifically, the fusion weight is determined according to the correlation between fatigue features at different scales, and the fusion fatigue features of strain data at each historical period are obtained by fusion, including: According to the correlation between fatigue characteristics at each scale, the corresponding correlation coefficients were calculated respectively; In specific implementations, this embodiment determines the fusion weight and performs feature fusion based on the correlation between fatigue features at each scale. Specifically, based on the correlation between fatigue features at multiple scales (i.e., attention features of different convolution branches), the Pearson correlation coefficient calculation method is used to calculate the correlation coefficient between fatigue features at each scale: ; Where: is the Pearson correlation coefficient between the j-th scale fatigue characteristic and the fatigue accumulation value; is the fatigue eigenvalue of the i-th sample at the j-th scale (compressed into a scalar through global average pooling); is the mean of fatigue characteristics of all samples at the jth scale; is the fatigue accumulation value of the i-th sample; is the mean of the cumulative fatigue values ​​of all samples; is the total number of samples.

[0023] According to the numerical value of the calculated correlation coefficient, the fusion weight corresponding to each scale fatigue feature is determined; Specifically, the fusion weights corresponding to fatigue features at each scale are obtained according to the correlation coefficients calculated above, and the fusion weights are determined using the Softmax normalization method: ; Where: is the fusion weight corresponding to the j-th scale fatigue feature; is the correlation coefficient between the fatigue feature of the jth scale and the target output (fatigue accumulation value); K is the number of scales (for example, 3 scales).

[0024] The fatigue features of different scales are weightedly fused according to the determined fusion weights to obtain the fused fatigue features of the strain data of each historical period; Finally, fatigue features of different scales are weightedly fused according to the above fusion weights to obtain the fused fatigue features corresponding to the strain data of each historical period: ; Where: is the fatigue characteristic after fusion; is the j-th scale fatigue feature.

[0025] S102: Predicting a current fatigue state of each local area according to the fatigue state prediction model, and generating a plurality of candidate load adjustment strategies through a pre-trained reinforcement learning model based on the current fatigue state; In implementation, the training method of the reinforcement learning model includes: Based on the historical fatigue state data of local areas of the wind turbine blades, the changes in the cumulative fatigue values ​​of each local area before and after the implementation of the historical load adjustment strategy are determined (the state after the implementation of the historical load adjustment strategy is obtained through simulation with OpenFAST wind turbine simulation software), and the state space and action space of the reinforcement learning model are constructed; In the specific implementation, we first determine the changes in the fatigue accumulation values ​​of each local area before and after the implementation of the historical load adjustment strategy based on the historical fatigue state data collected from each local area of ​​the wind turbine blade in different historical periods. Based on this, we construct the state space and action space of the reinforcement learning model, as follows: The historical fatigue status data specifically includes the fatigue accumulation value of each local area during the historical operation process; The state space is defined as a state vector consisting of the fatigue accumulation values ​​of each local area, for example: ; in, They represent the fatigue accumulation values ​​corresponding to the blade root, blade tip, blade edge and blade center at time t respectively; The action space is defined as the different parameter combinations of the implemented load adjustment strategy, for example: the adjustment amplitude of the wind turbine blade angle of attack (e.g. etc.), yaw angle adjustment range (e.g. etc.), pitch angle variation range (e.g. ) and speed adjustment range (e.g. rated speed ), the action space is expressed as: ; in: is the attack angle adjustment amplitude corresponding to the i-th strategy; is the yaw angle adjustment range; is the pitch angle variation range; is the speed adjustment range; It should be noted that the angle of attack adjustment range The aerodynamic load distribution function is converted into the local area load change. The conversion formula is as follows: ; Where: is the load variation in region j; is the regional aerodynamic coefficient (e.g., blade root: 0.8, blade tip: 1.2); is the lift curve slope (example value 0.12 / deg); is the equivalent area (leaf root: 2.5 , Tip: 0.8 ), is the air density; is the wind speed; Based on the magnitude of the fatigue cumulative value change, the fatigue damage expansion rate of each local area is determined, and the reward function of the reinforcement learning model is constructed; Specifically, the method for constructing the reward function of the reinforcement learning model includes: Based on the fatigue accumulation values ​​of each local area before and after the implementation of the historical load adjustment strategy, the change in the fatigue accumulation rate after the implementation of the load adjustment strategy is calculated relative to that before the implementation, and the change is used as the basic reward value of each local area; In the specific implementation, based on the fatigue accumulation values ​​of each local area before and after the implementation of the historical load adjustment strategy, the change in the fatigue accumulation rate after the implementation of the load adjustment strategy relative to that before implementation is calculated and defined as the basic reward value, specifically: First, define the fatigue accumulation rate: ; Then calculate the basic reward value: ; Where: is the fatigue accumulation value of the ith local area at time t; The time interval before and after the policy is implemented (such as 10 minutes). is the basic reward value corresponding to the i-th local area; The period before the implementation of the strategy [t- , t] fatigue accumulation rate: ; The period before the implementation of the strategy [t, t+ ]’s fatigue accumulation rate: .

[0026] Calculate the fatigue sensitivity coefficient corresponding to each local area based on the difference between the current fatigue cumulative value of each local area and the set fatigue safety threshold; Specifically, the method for determining the fatigue sensitivity coefficient includes: Determine the fatigue margin corresponding to each local area based on the difference between the fatigue limit value of the material in the local area of ​​the wind turbine blade and the current fatigue accumulation value; In specific implementation, the fatigue margin is determined based on the difference between the current fatigue accumulation value (a dimensionless value ranging from 0 to 1) and the material fatigue damage limit (i.e., the damage limit value 1 when the fatigue life is exhausted). The dimensionless definition is as follows: ; Where: is the fatigue margin of the i-th local area; is the fatigue limit value; is the current fatigue accumulation value of the i-th local area.

[0027] Calculate the initial sensitivity coefficient corresponding to each local area according to the size of the fatigue margin; Furthermore, the initial sensitivity coefficient corresponding to each local area is calculated according to the determined fatigue margin size, for example, using an exponential function as follows: ; Where: is the initial sensitivity coefficient; To prevent zero constant, the value range is 0.01, The upper limit of the sensitivity coefficient, for example, the value is 10.

[0028] Determine the dynamic amplification factor of the fatigue state according to the wind speed level and load level when the blade is running; Further, according to the wind speed level (e.g., level I to V) and load level (e.g., light load, medium load, and heavy load) when the blade is running, the dynamic amplification factor is determined through experiments. The specific calculation formula is, for example: ; Where: is the dynamic amplification factor; 、 is the coefficient obtained based on actual wind turbine test data, and the values ​​of wind speed level and load level are obtained through real-time sensor data (e.g. 0.1, 0.3); 、 They are numerical representations of wind speed level and load level respectively; Multiply the dynamic amplification factor and the initial sensitivity coefficient one by one to obtain the fatigue sensitivity coefficient of each local area; Finally, the dynamic amplification factor is multiplied by the initial sensitivity coefficient one by one to obtain the fatigue sensitivity coefficient: ; Perform weighted calculations one by one based on the basic reward value and the fatigue sensitivity coefficient to obtain a comprehensive reward value, and construct a reward function of the reinforcement learning model based on the comprehensive reward value; The comprehensive reward value is obtained by weighting the above basic reward value and fatigue sensitivity coefficient one by one, and the reward function of the reinforcement learning model is constructed as follows: .

[0029] The constructed state space, action space, and reward function are used as inputs to the reinforcement learning model, and the deep reinforcement learning algorithm is used to train the reinforcement learning model to obtain a pre-trained reinforcement learning model for generating candidate load adjustment strategies in real time. In the specific implementation, the constructed state space, action space and reward function are used as input, and a deep reinforcement learning algorithm (such as the DDPG algorithm) is used to train the reinforcement learning model. Specifically, the state space vector is used as input; the action space vector is used as output; the comprehensive reward value is used as the feedback evaluation value; Iterative training is performed until convergence to obtain a pre-trained reinforcement learning model for generating candidate load adjustment strategies in real time.

[0030] S103: collecting real-time vibration data of the wind turbine blades, determining short-term trend variation characteristics of wind turbine blade fatigue, and screening and determining a target load adjustment strategy from the plurality of candidate load adjustment strategies based on a difference between the short-term trend variation characteristics and a current fatigue state; In implementation, the method for screening and determining the target load adjustment strategy includes: Collect real-time vibration data of wind turbine blades and determine the short-term trend change characteristics of wind turbine blade fatigue based on the real-time vibration data; In the specific implementation, multiple three-axis vibration sensors installed at different positions on the surface of the wind turbine blades are first used to collect real-time vibration data of local areas of the wind turbine blades (blade root, blade tip, blade edge and blade center). The real-time vibration data specifically includes the vibration time domain waveform signal data and the corresponding real-time sampling frequency, vibration amplitude, and vibration acceleration data, and is sent to the control center through a wireless transmission module (ZigBee protocol).

[0031] The method for determining the short-term trend change characteristics of wind turbine blade fatigue specifically includes: Real-time collection of vibration data from a local area of ​​the fan blades, and the corresponding frequency domain characteristic parameters obtained through the fast Fourier transform method; In the specific implementation, the vibration time domain waveform data of the local area of ​​the wind turbine blade collected in real time is converted into a frequency domain signal through the fast Fourier transform (FFT) method to obtain the frequency domain characteristic parameters, including frequency spectrum lines, spectrum amplitudes and vibration energy spectra.

[0032] For example, the calculation formula of the fast Fourier transform is: ; Where: is the frequency domain spectrum; is the time domain data point of the nth real-time sampling; N is the number of sampling points in each sampling period; k is the discrete frequency index of the spectrum.

[0033] Extract the current vibration main frequency, spectrum amplitude and corresponding vibration energy of the local area of ​​the blade based on the frequency domain characteristic parameters, and calculate and obtain the real-time vibration characteristic value; Furthermore, the main vibration frequency of the local area of ​​the blade at the current moment (i.e., the frequency corresponding to the highest amplitude in the spectrum), the corresponding spectrum amplitude, and the corresponding vibration energy value are extracted based on the frequency domain characteristic parameters, and the real-time vibration characteristic value is further calculated.

[0034] In a specific implementation, the real-time vibration characteristic value is: ; Where: is the real-time vibration characteristic value at the current moment; is the maximum spectrum amplitude at the current moment; The main vibration frequency at the current moment; is the vibration energy value corresponding to the current moment; is the corresponding reference amplitude, frequency, and energy (determined according to experimental data, such as the historical mean, maximum value, or standard value).

[0035] Calculating a vibration characteristic value change rate within a current period based on the real-time vibration characteristic value and the historical vibration characteristic value of a local area of ​​the wind turbine blade, and determining the vibration characteristic value change rate as a short-term trend change characteristic of wind turbine blade fatigue; It should be noted that the vibration characteristic value change rate is calculated based on the difference between the real-time vibration characteristic value and the historical vibration characteristic value of the local area of ​​the blade (the vibration characteristic value of the previous period). The calculation formula is as follows: ; Where: is the rate of change of the vibration characteristic value in the current period; is the real-time vibration characteristic value at the current moment; is the historical vibration characteristic value at the previous moment; The vibration characteristic value update period (for example, 5 minutes or 10 minutes); determining a difference level based on a numerical difference between the short-term trend change characteristic and the current fatigue state predicted by the fatigue state prediction model; In a specific implementation, the difference level is specifically determined based on the numerical difference between the determined short-term trend change characteristics and the current fatigue state predicted by the aforementioned real-time fatigue state prediction model: Specifically, the calculation formula for the numerical difference is: ; ; Where: is the difference value at the current moment; The fatigue accumulation value corresponding to the current fatigue state predicted by the real-time fatigue state prediction model; Indicates the current fatigue accumulation value predicted by the fatigue status prediction model; It should be noted that to ensure the comparability between the vibration characteristic value change rate and the fatigue cumulative value change rate, this embodiment is based on a large amount of historical experimental data fitting and establishes a linear mapping relationship between the two to convert the vibration characteristic change rate into the fatigue damage change rate, as shown below: First, calculate the fatigue damage change rate corresponding to the vibration characteristic change: ; Among them, the parameters 、 Determined by fitting historical data, for example 、 .

[0036] Then calculate the numerical difference: ; For example, according to the size of the numerical difference, the difference level is set as follows: Low difference level (Level I): If ,The difference level is defined as “low”, indicating that the short-term changes are basically consistent with the predicted state, and the blade fatigue develops steadily; Medium difference level (Level II): If ,The difference level is defined as “medium”, which indicates that there is a significant difference in blade fatigue and the ,strategy adjustment effort should be increased; High difference level (Level III): If ,The difference level is defined as “high”, indicating that the fatigue state is severely out of the prediction and requires urgent load adjustment intervention.

[0037] For each candidate load regulation strategy, the implementation effect of the candidate load regulation strategy is evaluated according to the pre-built multi-factor fuzzy weight comprehensive evaluation system to obtain the strategy evaluation value corresponding to each candidate strategy; The specific construction method of the multi-factor fuzzy weight comprehensive evaluation system includes: Based on the historical implementation records of the load adjustment strategy, the fatigue state data of the local area of ​​the blade before and after the implementation of the strategy is collected, and multiple evaluation indicators corresponding to the implementation effect of the load adjustment strategy are determined; First, based on the historical records of load regulation strategy implementation, the fatigue status data of the local area of ​​the wind turbine blades before and after the implementation of the strategy (including the fatigue accumulation value change rate, vibration amplitude change rate, and fatigue damage expansion rate) are collected, and these data are used as multiple evaluation indicators for evaluating the implementation effect of the load regulation strategy.

[0038] Constructing a fuzzy evaluation factor set, wherein the fuzzy evaluation factor set includes the fatigue damage expansion rate, fatigue cumulative value change rate and vibration amplitude change rate of the local area of ​​the blade; The hierarchical analysis method is used to calculate the pairwise comparison matrix between each evaluation index, and the initial weight is modified after consistency test to obtain the final weight corresponding to each evaluation index; It should be noted that the pairwise comparison matrix in the AHP is constructed as follows: ; For example, the pairwise comparison matrix (illustrative) given by experts is: ; The consistency of the matrix is ​​tested by the eigenvalue method. , requiring the consistency ratio (CR) ≤ 0.1, otherwise the matrix is ​​modified and the weights are recalculated.

[0039] Determine the membership corresponding to each evaluation indicator based on the fuzzy membership function, and perform fuzzy matrix operation on the final weight and membership to construct the multi-factor fuzzy weight comprehensive evaluation system; It can be understood that the fuzzy membership function is a trapezoidal or triangular fuzzy membership function.

[0040] For example, the trapezoidal fuzzy membership function of the fatigue cumulative value change rate is defined as: ; in, is the fatigue cumulative value change rate (%), and a, b, c, and d are characteristic parameter values ​​of the membership function. In this embodiment, after calculating the cumulative distribution function CDF of the fatigue cumulative value change rate index by statistically analyzing a large amount of historical load adjustment test data, the 5%, 25%, 75%, and 95% quantiles of the CDF are taken as example values ​​of 1%, 3%, 6%, and 8%, respectively. The specific parameters of the trapezoidal fuzzy membership function are determined as follows: .

[0041] Determine the priority coefficient corresponding to each candidate load regulation strategy according to the difference level, and perform weighted calculation using the priority coefficient and the corresponding strategy evaluation value to obtain a comprehensive priority score, and determine the candidate load regulation strategy with the highest comprehensive priority score as the target load regulation strategy; Furthermore, the priority coefficient corresponding to the difference level can be adjusted according to the blade safety requirements and actual control effect, for example: If the wind turbine blades are in a condition with a low fatigue safety margin, the priority factor of the difference level can be appropriately increased to facilitate a timely and rapid response (as shown in the example in Table 1 below); Table 1: Coefficient assignment table

[0042] If the fatigue safety margin is high, the priority factor can be appropriately lowered to reduce the frequency of load adjustment actions and extend the service life of the system.

[0043] For example, the comprehensive priority score calculation formula is: ; Where: Score the comprehensive priority of candidate strategy i; is the priority coefficient corresponding to strategy i; is the strategy evaluation value corresponding to candidate strategy i; Represents a normalization function, such as Min-Max or Softmax normalization.

[0044] The candidate load adjustment strategy with the highest comprehensive priority score is determined as the target load adjustment strategy, thereby ensuring that the load adjustment strategy selection process is reasonable, efficient and accurate.

[0045] S104: generating an active load control signal through the wind turbine active load control system according to the determined target load adjustment strategy, so as to actively adjust the load of a local area of ​​the wind turbine blade; It should be noted that the purpose of implementing this step is to generate precise control instructions based on the optimal load adjustment strategy selected in the previous step, so as to effectively adjust the load state of the local area of ​​the wind turbine blade in real time, thereby suppressing the expansion of blade fatigue damage and extending the service life of the blade.

[0046] In implementation, the active load control signal generation method includes: load adjustment amplitude and load adjustment direction of each blade local area according to the determined target load adjustment strategy; It should be noted that the target load adjustment strategy includes a set of precise load adjustment instructions for different blade local areas, and the instruction set specifically includes the following contents: Load adjustment amplitude: The absolute change or relative percentage change in the load of the local area of ​​the blade to be adjusted, specifically expressed in units of N·m or %. In specific implementation, a dynamic mapping relationship is established between the overall load change of global adjustment parameters (such as angle of attack and yaw angle) and the load change of each local area through the local aerodynamic coefficient matrix calculated based on CFD simulation. The specific calculation formula is: ;in, The overall load change caused by the adjustment of parameters such as angle of attack and yaw angle; is the aerodynamic characteristic coefficient of the local area of ​​the i-th blade, which is obtained from the data table obtained by CFD simulation calculation or wind tunnel test; Load adjustment direction: load increase or load decrease.

[0047] For example, the current load of a certain area near the root area of ​​the wind turbine blade is , the target load adjustment strategy indicates that it needs to be reduced by 5%, so the load adjustment range is: ; The load adjustment direction is to reduce the load.

[0048] The specific calculation formula is as follows: ; Where: is the load adjustment amplitude of the local area of ​​the i-th blade; is the current load value of the local area of ​​the i-th blade; is the load adjustment percentage of the local area of ​​the i-th blade. A positive value represents an increase in the load, and a negative value represents a decrease in the load.

[0049] It should be noted that the data table formed by the load adjustment instruction data of each local area is shown in Table 2 below (for example):

[0050] Determine the execution sequence corresponding to the target load adjustment strategy based on the current operating condition data of the wind turbine; It should be noted that the operating condition data includes but is not limited to wind speed level, wind direction angle, fan operating speed, current blade angle position, load level and operating environment temperature.

[0051] In a specific implementation, the above working condition data is first collected, and then the specific execution timing of each load adjustment instruction is calculated according to the real-time status of the current working condition data and the target load adjustment strategy requirements.

[0052] The execution timing is determined as follows: Setting preset working condition thresholds, including wind speed threshold (such as rated wind speed), speed threshold (such as rated speed), and blade position angle threshold; Based on the real-time collected operating data, the load regulation strategy is started when the wind turbine operating condition meets the preset threshold. It should be noted that to avoid execution difficulties caused by multiple condition coupling, the execution conditions must meet the following priority order: When the wind speed meets the threshold condition, it will be executed first; If the wind speed is not satisfied, but the rotation speed and blade position are satisfied, it can also be executed; If only the blade position satisfies the requirement, the judgment is delayed until the next period; If the delay time exceeds 30 minutes, regardless of whether the working conditions meet the threshold conditions, the strategy is forced to be executed in the next sampling period. At this time, the adjustment amplitude during execution is reduced to 50% of the original value (to prevent sudden working condition shocks).

[0053] For example, the fan operating condition data table is as follows Table 3 (example):

[0054] When the working condition meets the conditions, the moment is marked as the execution moment of the load regulation strategy ; In addition, it should be noted that the duration of the load regulation strategy is exemplarily set as: starting time Once confirmed, the strategy implementation cycle is 10 minutes; Right now: ; Where: For the current moment, It is the delay time determined according to the working condition data.

[0055] The determined load adjustment amplitude, load adjustment direction and execution sequence are used as inputs of the active load control system, and an active load control signal is obtained by calculation through the active load control system; It should be noted that the active load control system described in this step is a real-time control system installed in the fan control cabinet, which specifically includes: a controller mainboard; a data interface module; a control algorithm module; and a signal output module.

[0056] In a specific implementation, the load adjustment amplitude, load adjustment direction and execution timing of each local area determined in the previous step are input into the controller mainboard through the data interface module of the active load control system in the data input format for real-time calculation; Specifically, the active load control system implements the PID control algorithm calculation according to the control algorithm module of the following formula: ; Where: is the output value of the active load control signal, in voltage (V) or duty cycle (%); It is the absolute difference between the load adjustment target value and the current actual load, in N·m. express The specific calculation formula for the error between the load adjustment target and the actual load at the moment is: First determine the specific value of the target load: ; Then calculate the error value: ; Where: is the load adjustment percentage, a positive value indicates an increase in load, a negative value indicates a decrease in load, is the current actual load; are the proportional coefficient, integral coefficient and differential coefficient of the PID controller respectively.

[0057] It should be noted that the initial setting values ​​of the PID parameters (see Table 4 below for examples) are:

[0058] In addition, the active load control signal output is in the form of a PWM pulse width modulation signal. The PWM frequency is 1kHz, and the duty cycle (D) is calculated as follows: ; Where: is the maximum output control voltage, a typical value is 10V; for example: The calculated result is 5V, which corresponds to a PWM duty cycle of 50%.

[0059] Sending active load control signals to the active control actuator of the wind turbine to drive the local area load of the wind turbine blades to implement active adjustment; It should be noted that the active control actuator in this step is an active control actuator installed at the root or inside of the wind turbine blade, specifically: a hydraulic brake control device; or an electric active control device (such as a servo motor or an electric push rod).

[0060] In a specific implementation, the active load control system outputs the active load control signal to the drive unit in the active control actuator through the PWM signal output interface.

[0061] It should be noted that the drive unit includes a signal receiving module (PWM signal input port) and a motor drive module (such as a PWM driver or a hydraulic pump controller).

[0062] It should be further explained that the drive unit converts the received PWM duty cycle signal into a specific load adjustment execution action, including the execution amplitude (such as the motor rotation angle or the hydraulic cylinder piston displacement) and the action execution speed.

[0063] The quantitative relationship of action execution (example) is as follows: When the PWM duty cycle is 50%, the displacement amplitude executed by the active control actuator is 50% of the rated displacement; When the PWM duty cycle is 100%, the displacement amplitude performed by the active control actuator reaches the rated displacement.

[0064] The relationship between the displacement and the PWM duty cycle is shown in Table 5 below: Table 5: Relationship between execution displacement and PWM duty cycle

[0065] Finally, it should be noted that when the active control actuator performs the load adjustment action, it collects the actuator's displacement feedback signal in real time to form a closed-loop control to ensure the accurate implementation of the load adjustment action in the local area of ​​the wind turbine blade and correct the error in the load control process in real time.

[0066] Example 2 like Figure 2 As shown, the parts not described in detail in this embodiment are as shown in Example 1. This embodiment discloses a wind turbine blade intelligent monitoring and optimization system based on deep learning, including: The data prediction module 201 is used to calculate the fatigue accumulation value corresponding to the strain data of each historical period based on the historical strain data of the local area of ​​the wind turbine blade in combination with the fatigue accumulation rate calculation method, and train a fatigue state prediction model using the historical strain data as input and the fatigue accumulation value as output; a strategy generation module 202 for predicting the current fatigue state of each local area according to the fatigue state prediction model, and generating a plurality of candidate load adjustment strategies based on the current fatigue state through a pre-trained reinforcement learning model; a strategy screening module 203 for collecting real-time vibration data of the wind turbine blades, determining short-term trend variation characteristics of wind turbine blade fatigue, and screening and determining a target load adjustment strategy from the plurality of candidate load adjustment strategies based on a difference between the short-term trend variation characteristics and the current fatigue state; The blade adjustment module 204 is configured to generate an active load control signal through the wind turbine active load control system according to the determined target load adjustment strategy, so as to actively adjust the load of a local area of ​​the wind turbine blade.

[0067] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters, weights and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0068] The above embodiments can be implemented in whole or in part via software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are fully or partially performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium accessible by a computer, or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, or magnetic tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0069] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. A method for intelligent monitoring and optimization of wind turbine blades based on deep learning, characterized in that: include: S101: Based on the historical strain data of the local area of ​​the wind turbine blade, the fatigue accumulation value corresponding to the strain data in each historical period is calculated in combination with the fatigue accumulation rate calculation method, and the fatigue state prediction model is trained using the historical strain data as input and the fatigue accumulation value as output; S102: Predicting a current fatigue state of each local area according to the fatigue state prediction model, and generating a plurality of candidate load adjustment strategies through a pre-trained reinforcement learning model based on the current fatigue state; S103: collecting real-time vibration data of the wind turbine blades, determining short-term trend variation characteristics of wind turbine blade fatigue, and screening and determining a target load adjustment strategy from the plurality of candidate load adjustment strategies based on a difference between the short-term trend variation characteristics and a current fatigue state; S104: According to the determined target load adjustment strategy, an active load control signal is generated by the wind turbine active load control system to actively adjust the load of a local area of ​​the wind turbine blade.

2. The method for intelligent monitoring and optimization of wind turbine blades based on deep learning according to claim 1, characterized in that: The training method of the fatigue state prediction model includes: Collect historical strain data of a local area of ​​the wind turbine blade, segment the historical strain data according to a fixed time length, and obtain strain data samples corresponding to multiple historical time periods; According to Miner's fatigue damage criterion and the fatigue SN curve of the wind turbine blade material, the fatigue cumulative value corresponding to the strain data sample in each historical period is calculated respectively; Construct a multi-scale convolution feature extraction structure, perform convolution operations on strain data samples of each historical period, and extract the initial fatigue features corresponding to each scale; The fusion weight is determined according to the correlation between fatigue features at different scales. The fused fatigue features of strain data of each historical period are obtained by fusion. The fused fatigue features are used as the model input, and the fatigue accumulation value is used as the model output to train and obtain the fatigue state prediction model.

3. The method for intelligent monitoring and optimization of wind turbine blades based on deep learning according to claim 2, characterized in that: The specific construction method of the multi-scale convolutional feature extraction structure includes: Construct multiple parallel convolution branches with different convolution kernel sizes. In each parallel convolution branch, set up dilated convolution layer, depthwise separable convolution layer and attention mechanism module in sequence. In each parallel convolution branch, the initial convolution features of the corresponding scale are obtained by dilating the convolution layer, and then the initial convolution features are further convolved using the depthwise separable convolution layer to obtain the refined fatigue features of the corresponding scale; For the refined fatigue features obtained in each parallel convolution branch, the attention weight of the feature channel is calculated through the attention mechanism module, and each channel of the refined fatigue feature is weighted and adjusted according to the attention weight to obtain the attention feature of each convolution branch.

4. The method for intelligent monitoring and optimization of wind turbine blades based on deep learning according to claim 3 is characterized in that: The fusion weight is determined based on the correlation between fatigue features at different scales, and the fusion fatigue features of strain data of each historical period are obtained by fusion, including: According to the correlation between fatigue characteristics at each scale, the corresponding correlation coefficients were calculated respectively; According to the numerical value of the calculated correlation coefficient, the fusion weight corresponding to each scale fatigue feature is determined; The fatigue characteristics of different scales are weightedly fused according to the determined fusion weights to obtain the fused fatigue characteristics of the strain data in each historical period.

5. The method for intelligent monitoring and optimization of wind turbine blades based on deep learning according to claim 4 is characterized in that: The training method of the reinforcement learning model includes: Based on the historical fatigue state data of the local area of ​​the wind turbine blade, the changes in the fatigue accumulation value of each local area before and after the implementation of the historical load adjustment strategy are determined, and the state space and action space of the reinforcement learning model are constructed; Based on the magnitude of the fatigue cumulative value change, the fatigue damage expansion rate of each local area is determined, and the reward function of the reinforcement learning model is constructed; The constructed state space, action space and reward function are used as the input of the reinforcement learning model, and the deep reinforcement learning algorithm is used to train the reinforcement learning model to obtain a pre-trained reinforcement learning model for real-time generation of candidate load adjustment strategies.

6. The method for intelligent monitoring and optimization of wind turbine blades based on deep learning according to claim 5, characterized in that: The method for constructing the reward function of the reinforcement learning model includes: Based on the fatigue accumulation values ​​of each local area before and after the implementation of the historical load adjustment strategy, the change in the fatigue accumulation rate after the implementation of the load adjustment strategy is calculated relative to that before the implementation, and the change is used as the basic reward value of each local area; Calculate the fatigue sensitivity coefficient corresponding to each local area based on the difference between the current fatigue cumulative value of each local area and the set fatigue safety threshold; A weighted calculation is performed one by one according to the basic reward value and the fatigue sensitivity coefficient to obtain a comprehensive reward value, and a reward function of the reinforcement learning model is constructed based on the comprehensive reward value.

7. The method for intelligent monitoring and optimization of wind turbine blades based on deep learning according to claim 6, characterized in that: The method for determining the fatigue sensitivity coefficient includes: Determine the fatigue margin corresponding to each local area based on the difference between the fatigue limit value of the material in the local area of ​​the wind turbine blade and the current fatigue accumulation value; Calculate the initial sensitivity coefficient corresponding to each local area according to the size of the fatigue margin; Determine the dynamic amplification factor of the fatigue state according to the wind speed level and load level when the blade is running; The dynamic amplification factor is multiplied by the initial sensitivity coefficient one by one to obtain the fatigue sensitivity coefficient of each local area.

8. The method for intelligent monitoring and optimization of wind turbine blades based on deep learning according to claim 7, characterized in that: The method for screening and determining the target load adjustment strategy includes: Collect real-time vibration data of wind turbine blades and determine the short-term trend change characteristics of wind turbine blade fatigue based on the real-time vibration data; determining a difference level based on a numerical difference between the short-term trend change characteristic and the current fatigue state predicted by the fatigue state prediction model; For each candidate load regulation strategy, the implementation effect of the candidate load regulation strategy is evaluated according to the pre-built multi-factor fuzzy weight comprehensive evaluation system to obtain the strategy evaluation value corresponding to each candidate strategy; The priority coefficient corresponding to each candidate load regulation strategy is determined according to the difference level, and the priority coefficient and the corresponding strategy evaluation value are weighted to obtain a comprehensive priority score, and the candidate load regulation strategy with the highest comprehensive priority score is determined as the target load regulation strategy.

9. The method for intelligent monitoring and optimization of wind turbine blades based on deep learning according to claim 8, characterized in that: The method for determining the short-term trend change characteristics of fan blade fatigue specifically includes: Real-time collection of vibration data from a local area of ​​the fan blades, and the corresponding frequency domain characteristic parameters obtained through the fast Fourier transform method; The vibration characteristic value change rate in the current period is calculated based on the real-time vibration characteristic value and the historical vibration characteristic value of the local area of ​​the wind blade, and the vibration characteristic value change rate is determined as the short-term trend change characteristic of the wind blade fatigue.

10. The method for intelligent monitoring and optimization of wind turbine blades based on deep learning according to claim 9, characterized in that: The specific construction method of the multi-factor fuzzy weight comprehensive evaluation system includes: Based on the historical implementation records of the load adjustment strategy, the fatigue state data of the local area of ​​the blade before and after the implementation of the strategy is collected, and multiple evaluation indicators corresponding to the implementation effect of the load adjustment strategy are determined; Constructing a fuzzy evaluation factor set, wherein the fuzzy evaluation factor set includes the fatigue damage expansion rate, fatigue cumulative value change rate and vibration amplitude change rate of the local area of ​​the blade; The initial weight of each evaluation indicator is determined according to the expert scoring method, and the pairwise comparison matrix between each evaluation indicator is calculated using the hierarchical analysis method. After consistency test, the initial weight is modified to obtain the final weight corresponding to each evaluation indicator; The membership degree corresponding to each evaluation index is determined based on the fuzzy membership function, and the final weight and membership degree are subjected to fuzzy matrix operation to construct the multi-factor fuzzy weight comprehensive evaluation system.

Images