Solar photovoltaic power generation assembly matching device with wind resistance damping function

Abstract

This invention relates to a supporting device for solar photovoltaic power generation modules with wind resistance and vibration reduction functions, belonging to the field of photovoltaic power generation technology. The device includes: a multi-source sensing acquisition module for real-time acquisition of raw data from the area where the photovoltaic module is located; a data preprocessing and feature extraction module for obtaining standardized real-time feature vectors; a wind load and response prediction module for predicting wind pressure time histories and module vibration response envelopes within a preset future time period; a vibration risk quantification module for real-time calculation of a comprehensive vibration risk index; a vibration reduction control decision module for generating control commands for adjustable dampers and / or support tilt adjustment mechanisms; and an execution and closed-loop feedback module for executing control commands and continuously acquiring vibration acceleration data after execution, feeding it back to the vibration risk quantification module. This device can effectively reduce wind-induced vibration damage to photovoltaic modules, improving the system's wind resistance reliability and operational lifespan.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation technology, and in particular to a solar photovoltaic power generation module supporting device with wind resistance and shock absorption function. Background Technology

[0002] Currently, solar photovoltaic (PV) power generation systems typically use fixed brackets or tracking brackets to install and support PV modules, and improve the system's wind resistance through structural reinforcement, damping devices, or mechanical limiting devices. Some more intelligent systems also incorporate basic monitoring software to collect wind speed and vibration data, and use simple threshold judgments to trigger alarms or emergency shutdown controls.

[0003] However, existing software systems mostly rely on static thresholds for control, lacking the ability to dynamically model and predict wind load changes and structural responses, making it difficult to adjust control strategies in a timely and accurate manner. Furthermore, most systems fail to achieve closed-loop linkage between vibration data and control actuators, resulting in delayed or unstable responses to vibration reduction measures. They are unable to make optimized decisions based on the actual risk state of the components, thus remaining at risk of component microcracks, support loosening, or even overturning under strong winds or wind-induced vibration scenarios. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a solar photovoltaic power generation module supporting device with wind-resistant and vibration-damping functions that can effectively reduce the damage of wind-induced vibration to photovoltaic modules and improve the wind resistance reliability and service life of the system.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: This invention provides a solar photovoltaic power generation module supporting device with wind resistance and shock absorption function, the device comprising: The multi-source sensor acquisition module is used to acquire raw data of the area where the photovoltaic module is located in real time. The raw data includes wind speed, wind direction, turbulence intensity, vibration acceleration of module surface, module tilt angle and strain data. The data preprocessing and feature extraction module is used to perform noise reduction, time synchronization and feature extraction on the raw data output by the multi-source sensing acquisition module to obtain a standardized real-time feature vector. The wind load and response prediction module has a built-in time series prediction model based on physical information enhancement, which is used to predict the wind pressure time history and component vibration response envelope within a preset period of time based on the historical feature vector sequence and the real-time feature vector. The vibration risk quantification module is used to calculate the comprehensive vibration risk index in real time based on the predicted wind pressure time history and component vibration response envelope within the preset future time period, combined with the material cumulative damage model. The vibration reduction control decision module is used to generate control commands for the adjustable damper and / or the support tilt adjustment mechanism based on the comparison results between the comprehensive vibration risk index and the preset multi-level risk threshold. The execution and closed-loop feedback module is used to execute the control commands and continuously acquire the vibration acceleration data after execution, and feed it back to the vibration risk quantification module to dynamically correct the risk index and control strategy.

[0006] Furthermore, to achieve the above objectives, the present invention also proposes an electronic device, comprising: a memory for storing computer software programs; and a processor for reading and executing the computer software programs, thereby realizing the control method for a solar photovoltaic power generation module supporting device with wind resistance and vibration reduction function as described above.

[0007] Furthermore, to achieve the above objectives, the present invention also proposes a non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements a control method for a solar photovoltaic power generation module supporting device with wind resistance and vibration reduction function as described above.

[0008] The beneficial effects of this invention are: (1) The present invention avoids the lag of traditional static threshold control by using the wind load and response prediction module, and can adjust the damping and attitude in advance before the arrival of strong winds to achieve preventive shock reduction.

[0009] (2) The vibration risk quantification module and the execution module of the present invention form a closed loop. The continuous adjustable control of the damper and the support is driven by the risk index. The vibration reduction response is fast and the effect is stable, which significantly reduces the vibration amplitude of the components.

[0010] (3) The present invention introduces a damping attenuation integral term and a stress over-limit penalty term, which can truly reflect the cumulative impact of wind vibration history on the structure and prevent fatigue failure caused by frequent low-to-medium intensity vibrations, which is a blind spot missed by the traditional threshold method.

[0011] In summary, the present invention aims to provide a solar photovoltaic power generation module supporting device with wind resistance and vibration reduction function. By integrating multi-source sensor data, establishing a dynamic prediction model of wind load and structural response, and using the vibration risk index as the decision basis, it realizes closed-loop linkage control of adjustable damping mechanism and / or support posture, thereby effectively reducing the damage of wind-induced vibration to photovoltaic modules and improving the system's wind resistance reliability and service life. Attached Figure Description

[0012] Figure 1 A scene diagram illustrating a control method for a solar photovoltaic power generation module with wind resistance and shock absorption function provided by the present invention; Figure 2This is a schematic diagram of a solar photovoltaic power generation module supporting device with wind resistance and shock absorption function provided by the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0015] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0016] Please see Figure 1 , Figure 1 This is a scene diagram illustrating a control method for a solar photovoltaic power generation module with wind resistance and vibration reduction function provided by the present invention. (See diagram for example.) Figure 1 As shown, the terminal and server are connected via a network, such as a wired or wireless network. The terminal can include, but is not limited to, portable devices such as mobile phones and tablets with various network platform applications installed, as well as fixed terminals such as computers, kiosks, and advertising machines. The server provides users with various business services, including service push servers and user recommendation servers.

[0017] It should be noted that, Figure 1The scenario diagram illustrating a control method for a solar photovoltaic power generation module with wind resistance and vibration reduction function is merely an example. The terminals, servers, and application scenarios described in this embodiment are for the purpose of more clearly illustrating the technical solutions of this embodiment and do not constitute a limitation on the technical solutions provided by this embodiment. As those skilled in the art will know, with the evolution of devices and the emergence of new business scenarios, the technical solutions provided by this embodiment are also applicable to similar technical problems.

[0018] The terminal can be used for: The multi-source sensor acquisition module is used to acquire raw data of the area where the photovoltaic module is located in real time. The raw data includes wind speed, wind direction, turbulence intensity, vibration acceleration of module surface, module tilt angle and strain data. The data preprocessing and feature extraction module is used to perform noise reduction, time synchronization and feature extraction on the raw data output by the multi-source sensor acquisition module to obtain a normalized real-time feature vector. The wind load and response prediction module has a built-in time series prediction model based on physical information enhancement, which is used to predict the wind pressure time history and component vibration response envelope within a preset period of time based on historical feature vector sequences and real-time feature vectors. The vibration risk quantification module is used to calculate the comprehensive vibration risk index in real time based on the predicted wind pressure time history and component vibration response envelope within a preset future period, combined with the material cumulative damage model. The vibration control decision module is used to generate control commands for the adjustable damper and / or the support tilt adjustment mechanism based on the comparison results of the comprehensive vibration risk index and the preset multi-level risk thresholds. The execution and closed-loop feedback module is used to execute control commands and continuously acquire vibration acceleration data after execution, feeding it back to the vibration risk quantification module to dynamically correct the risk index and control strategy.

[0019] Please see Figure 2 , Figure 2 This is a schematic diagram of a solar photovoltaic power generation module supporting device with wind resistance and shock absorption function provided by the present invention.

[0020] like Figure 2 As shown in the figure, a solar photovoltaic power generation module supporting device with wind resistance and shock absorption function proposed in this embodiment of the invention includes: The multi-source sensor acquisition module 201 is used to acquire raw data of the area where the photovoltaic module is located in real time. The raw data includes wind speed, wind direction, turbulence intensity, vibration acceleration of the module surface, module tilt angle and strain data.

[0021] In some embodiments, this module acquires various physical quantities in the area where the photovoltaic module is located in real time using a high-frequency sampling method. Specifically, wind speed and direction are acquired using an ultrasonic anemometer mounted on the module frame or support, and turbulence intensity is calculated using the high-frequency pulsation component of the three-dimensional ultrasonic anemometer; vibration acceleration is collected using a triaxial accelerometer attached to the module backsheet; the tilt angle of the support is read in real time using a tilt sensor; and strain is acquired using strain gauges attached to the module frame or key welds of the support. All sensors are timestamped according to a unified time base, forming a time-stamped raw multi-source sensor data stream for subsequent modules to access.

[0022] In some embodiments, the multi-source sensing acquisition module 201 can also be used for: Based on wind speed and direction, and combined with the current tilt angle of the photovoltaic module, determine the incoming flow angle and relative wind speed acting on the module surface; Based on the incoming angle of attack and relative wind speed, combined with the air density and aerodynamic coefficient mapping table, the instantaneous wind pressure distribution of each zone on the component surface is determined; Based on the instantaneous wind pressure distribution of each zone on the component surface, the equivalent wind pressure resultant force application point and wind pressure resultant force time history of the component are calculated, and the wind pressure resultant force time history is output as part of the raw data.

[0023] The ultrasonic anemometer in the multi-source sensor acquisition module outputs a scalar value of wind speed and azimuth angle of wind direction at a first sampling frequency, while the tilt sensor outputs the tilt angle of the photovoltaic module support at a second sampling frequency.

[0024] The wind pressure distribution calculation unit synchronously captures the wind speed, wind direction, and support tilt angle data of the nearest neighbors in time, and performs the following calculations: First, based on the geometric relationship between the wind direction azimuth and the support tilt angle, the wind speed vector is decomposed by projection in the component coordinate system to obtain the normal wind speed component perpendicular to the component surface and the tangential wind speed component parallel to the component surface.

[0025] Then, based on the normal wind speed component and the tangential wind speed component, the incoming angle of attack of the airflow relative to the component surface is calculated. The incoming angle of attack is defined as the angle between the incoming direction of the airflow and the component surface, and its magnitude is obtained by arctangent calculation of the ratio of the normal wind speed component to the tangential wind speed component.

[0026] Meanwhile, the relative wind speed is obtained by combining the normal wind speed component and the tangential wind speed component. This relative wind speed characterizes the actual impact velocity of the airflow relative to the surface of the component under the current tilted attitude.

[0027] Through the above process, the wind pressure distribution calculation unit determines the incoming flow angle and relative wind speed acting on the component surface at the current moment.

[0028] The surface of the photovoltaic module is pre-divided into multiple zones, each with a unique zone identifier. The aerodynamic coefficient mapping table uses "incoming flow angle of attack - relative wind speed" as the index dimension to store the aerodynamic pressure coefficient of each zone under different operating conditions.

[0029] The wind pressure distribution calculation unit performs a two-dimensional lookup operation in the aerodynamic coefficient mapping table based on the incoming flow angle of attack and relative wind speed. If the current incoming flow angle of attack and relative wind speed fall exactly on a discrete node of the mapping table, the aerodynamic pressure coefficient of the corresponding zone is directly read; if the current incoming flow angle of attack and / or relative wind speed do not fall on a discrete node, the aerodynamic pressure coefficient interpolation of each zone under the current operating condition is calculated based on the aerodynamic pressure coefficients of adjacent nodes using bilinear interpolation or cubic spline interpolation methods.

[0030] After obtaining the aerodynamic pressure coefficient of each zone, the wind pressure distribution calculation unit determines the air density based on the current ambient temperature, humidity and atmospheric pressure; then, for each zone, the instantaneous wind pressure of that zone is calculated: the air density, the square of the relative wind speed and the aerodynamic pressure coefficient of that zone are multiplied together and then multiplied by half to obtain the instantaneous wind pressure value of that zone.

[0031] The instantaneous wind pressure value of each zone corresponds one-to-one with the zone identifier, together forming the instantaneous wind pressure distribution on the surface of the component at the current moment.

[0032] After obtaining the instantaneous wind pressure distribution, the wind pressure distribution calculation unit further synthesizes the equivalent wind pressure resultant force and the point of application of the resultant force.

[0033] Specifically, for each zone, the instantaneous wind pressure value of that zone is multiplied by the effective area of ​​that zone to obtain the normal wind pressure vector of that zone, and the point of application is the geometric center coordinates of that zone.

[0034] Traverse all partitions and sum the normal wind pressure vectors of all partitions to obtain the equivalent wind pressure resultant force acting on the entire component surface. The direction of this resultant force is the dominant direction of the normal wind pressure, and its magnitude is the magnitude of the equivalent wind pressure resultant force.

[0035] Simultaneously, based on the magnitude of the normal wind pressure on each zone and the coordinates of its point of application, the point of application of the equivalent wind pressure resultant force is calculated using the torque equivalence principle: that is, taking a preset reference point on the surface of the component (e.g., the geometric center of the component) as the origin, the wind pressure torque of all zones is synthesized on each coordinate axis, and the total torque component is divided by the magnitude of the equivalent wind pressure resultant force to obtain the offset coordinates of the point of application of the resultant force relative to the reference point, thereby determining the point of application of the equivalent wind pressure resultant force.

[0036] The wind pressure distribution calculation unit repeats the aforementioned steps according to the update rate of the first sampling frequency to obtain a set of equivalent wind pressure resultant force magnitudes and points of application that change over time, namely the wind pressure resultant force time history.

[0037] The wind pressure resultant force time history, along with data such as wind speed, wind direction, turbulence intensity, vibration acceleration, tilt angle, and strain, are used as raw data output by the multi-source sensor acquisition module for use by the data preprocessing and feature extraction module.

[0038] It should be noted that, in some preferred embodiments, the construction of the above-mentioned aerodynamic coefficient mapping table adopts an aerodynamic coefficient prediction model based on a deep neural network to replace the traditional discrete lookup table and interpolation method, so as to obtain a more continuous and refined wind pressure distribution.

[0039] The construction and training process of this aerodynamic coefficient prediction model is as follows: The aerodynamic coefficient prediction model employs a fully connected deep neural network structure, comprising one input layer, three hidden layers, and one output layer. The input layer has three neurons, corresponding to the incoming angle of attack, relative wind speed, and zone identifier encoding, respectively. The first and second hidden layers each have 128 neurons, using the ReLU activation function to introduce nonlinear expressive power. The third hidden layer has 64 neurons, also using the ReLU activation function. The output layer has one neuron, outputting the predicted aerodynamic pressure coefficient for the corresponding zone. To prevent overfitting, Dropout layers are placed after the first and second hidden layers, with a Dropout ratio of 0.2.

[0040] Training data was obtained through computational fluid dynamics simulation combined with wind tunnel testing. Different combinations of incoming flow angles (range 0° to 90°, step size 2°) and relative wind speeds (range 5 m / s to 40 m / s, step size 2 m / s) were selected to calculate the steady-state values ​​of the aerodynamic pressure coefficients for each zone on the component surface under each condition, forming a sample set. The sample set was then randomly divided into a training set, a validation set, and a test set in an 8:1:1 ratio.

[0041] During training, the incoming air angle of attack, relative wind speed, and zone identifiers from the training set are encoded as input features, and the corresponding aerodynamic pressure coefficients are used as labels, which are then input into the aerodynamic coefficient prediction model. Mean squared error is used as the loss function, and the Adam optimizer is employed. The initial learning rate is set to 0.001, and the batch size is set to 64. The model is trained iteratively for 300 epochs on the training set. After each epoch, the validation loss is calculated on the validation set. When the validation loss no longer decreases for 20 consecutive epochs, an early stopping strategy is implemented, and the model weights with the minimum validation loss are saved as the final model.

[0042] In actual operation, the wind pressure distribution calculation unit of the multi-source sensor acquisition module loads the weights of the trained aerodynamic coefficient prediction model, inputs the real-time calculated incoming flow angle of attack, relative wind speed, and the queried zone identifier into the model, and the model outputs the predicted aerodynamic pressure coefficient value of the zone under the current operating conditions after one forward propagation. After traversing all zones, the complete instantaneous wind pressure distribution on the component surface is obtained, and then the resultant wind pressure force is calculated according to the aforementioned time history.

[0043] The above aerodynamic coefficient prediction model realizes a continuous and rapid mapping from the incoming flow angle of attack and relative wind speed to the wind pressure of each zone, avoiding the quantization error of discrete table lookup and the accuracy loss caused by interpolation, making the wind pressure distribution calculation more accurate and efficient.

[0044] The data preprocessing and feature extraction module 202 is used to perform noise reduction, time synchronization and feature extraction on the raw data output by the multi-source sensor acquisition module to obtain a standardized real-time feature vector.

[0045] In some embodiments, this module receives the aforementioned time-stamped raw data and processes it according to the following steps: First, it selects an appropriate wavelet basis and decomposition level based on the noise characteristics of each channel to perform wavelet threshold denoising, eliminating electromagnetic interference and accidental impact spikes to obtain denoised data; then, it performs cross-channel synchronous interpolation and alignment based on the timestamp to form a synchronous sequence with a unified sampling rate; then, it calculates the root mean square value and peak factor from the synchronous sequence of vibration acceleration as time-domain features, and extracts the dominant frequency point and spectral kurtosis as frequency-domain features through fast Fourier transform; at the same time, it extracts instantaneous mean, rate of change and other time-domain statistics from the synchronous sequences of wind speed, wind direction, tilt angle and strain, and finally forms a multi-dimensional normalized real-time feature vector, which is output in time sequence.

[0046] In some embodiments, the data preprocessing and feature extraction module 202 can also be used for: Based on the signal characteristics of each sensing channel in the original data, the corresponding wavelet basis function and decomposition level are selected, and wavelet threshold denoising is performed on each channel signal to obtain the denoised signal. Based on the unified timestamp carried on each channel signal, the noise-reduced signal is time-aligned and resampled to obtain a synchronous sampling sequence; Based on the synchronous sampling sequence, extract time-domain and frequency-domain features; Based on time-domain and frequency-domain features, a normalized real-time feature vector is generated by combining them.

[0047] Specifically, the data preprocessing and feature extraction module can include a signal characteristic analysis subunit and a wavelet noise reduction subunit.

[0048] The signal characteristic analysis subunit first performs short-time stationarity analysis on the signals from each sensor channel output by the multi-source sensor acquisition module. Specifically, for the signal sequence of each sensor channel, the root mean square value and zero-crossing rate of the signal within the window are calculated using a preset time window length and sliding step size, and the instantaneous signal-to-noise ratio and main noise type of the channel signal are determined accordingly.

[0049] Based on the instantaneous signal-to-noise ratio characteristics, the signal characteristic analysis subunit determines the selection strategy for wavelet basis functions for each channel: for channels containing obvious impact components and transient responses, such as vibration acceleration signals and strain signals, Daubechies wavelets with tight support and sufficient vanishing moment order are selected to accurately capture the local abrupt change characteristics of the signal; for channels with relatively gentle changes, such as wind speed and turbulence intensity, Symlet wavelets with good symmetry and smoothness are selected to obtain more regular subband decomposition in the frequency domain; for tilt signals, because they change slowly and low-frequency components dominate, Haar wavelets or lower-order Daubechies wavelets are selected.

[0050] Simultaneously, the signal characteristic analysis subunit determines the number of wavelet decomposition levels based on the sampling frequency and distribution of the main frequency components of each channel signal. The determination method is as follows: A Fast Fourier Transform is performed on the signal to obtain the power spectral density distribution, identifying the frequency range where the signal power dominates; the number of decomposition levels is set such that the frequency range corresponding to the approximation coefficients of the Nth level exactly covers this dominant frequency range, thus ensuring that useful low-frequency approximation information of the signal is retained while removing high-frequency noise detail coefficients. Generally, the number of decomposition levels for wind speed and acceleration signals is 3 to 5, and for tilt angle and strain signals, it is 4 to 6.

[0051] The wavelet denoising subunit performs wavelet decomposition on the signal of each channel based on the wavelet basis functions and decomposition levels determined above, obtaining the approximation coefficients and detail coefficients for each level. Then, using a Bayesian threshold estimation method, an adaptive threshold is calculated for each level's detail coefficients based on the estimated noise level of that level. Coefficients with absolute values ​​less than the adaptive threshold are set to zero, while coefficients with absolute values ​​greater than or equal to the threshold undergo soft thresholding. Finally, wavelet reconstruction is performed using the processed detail coefficients and the retained approximation coefficients to obtain the denoised signal.

[0052] As a result, each sensing channel obtains a noise-reduced signal with the same sampling rate as the original signal, but with high-frequency noise effectively suppressed.

[0053] Since different sensors in the multi-source sensing acquisition module may have different hardware sampling frequencies and data output delays, the data preprocessing and feature extraction module further performs time alignment and resampling operations.

[0054] Specifically, the module first extracts the timestamp sequence of the denoised signal from each channel, and uses the channel with the highest sampling frequency in the timestamp sequence as the reference clock to determine a unified sampling time grid. The unified sampling time grid covers the common effective time period of all channels, with a step size of the preset target resampling period.

[0055] Then, for the denoised signal sequence of each channel, interpolation resampling is performed at each grid point of the unified sampling time grid using the signal value at the original timestamp of that channel. Cubic spline interpolation is used to ensure the smoothness of the signal during the sampling rate conversion.

[0056] Through this operation, the denoised signals from all sensor channels are converted into synchronous sampling sequences with the same time coordinates and data lengths. Each time step corresponds to a set of multidimensional synchronous sampling values, including denoised vibration acceleration, wind speed, wind direction, turbulence intensity, tilt angle, and strain.

[0057] After obtaining the synchronous sampling sequence, the data preprocessing and feature extraction module performs window-by-window feature extraction on the synchronous sampling sequence using a preset sliding time window. Each sliding time window covers sampling points of several consecutive time steps, and adjacent windows have a preset overlap rate.

[0058] For time-domain feature extraction, the synchronous sampling sequence of vibration acceleration signal is the main target: First, the arithmetic mean, root mean square (RMS) value, and peak factor of the vibration acceleration signal sequence within the current sliding time window are calculated as the first set of time-domain features. The RMS value reflects the average level of vibration energy, and the peak factor reflects the impact degree and distribution characteristics of the vibration signal.

[0059] Secondly, the average value of the absolute value of the first difference of the vibration acceleration signal sequence within the current sliding time window is calculated to obtain the vibration acceleration change rate characteristic, which is used to characterize the degree of transient change in vibration response.

[0060] Next, the maximum cross-correlation coefficients between each pair of triaxial vibration acceleration signals are calculated to obtain spatial correlation characteristics, which are used to reflect the degree of vibration coupling of the component in different directions.

[0061] Meanwhile, the mean, standard deviation, and trend slope of the wind speed signal sequence and the strain signal sequence are extracted within the same sliding time window, respectively, as the second set of time-domain features to characterize the statistical and changing trends of environmental load and structural response.

[0062] For frequency domain feature extraction, the synchronous sampling sequence of vibration acceleration signal is the main target: First, the vibration acceleration signal sequence within the current sliding time window is windowed, and the Hanning window is selected as the window function to suppress spectral leakage; then, a fast Fourier transform is performed to obtain a one-sided power spectral density estimate.

[0063] Secondly, from the power spectral density estimation, the frequency points corresponding to several peaks with the highest amplitude are identified by the spectral peak search algorithm, and the peak frequency with the largest amplitude in the low frequency band is selected as the vibration main frequency feature; at the same time, the spectral centroid of the power spectral density is calculated, that is, the ratio of the power spectral density weighted sum with frequency to the total power spectral density, to obtain the spectral centroid feature, which reflects the concentrated position of vibration energy in the frequency domain.

[0064] Next, the spectral kurtosis of the power spectral density is calculated, which is the mean of the sum of the fourth power of the difference between the power spectral density values ​​at each frequency point and the mean, divided by the fourth power of the standard deviation. This spectral kurtosis characteristic is used to characterize the significance of non-Gaussian components and potential impact features in vibration signals.

[0065] For the synchronous sampling sequence of strain signals, the same fast Fourier transform and peak search are performed within the same sliding time window to extract the strain dominant frequency features as supplementary features reflecting the frequency domain characteristics of the dynamic changes in structural stress.

[0066] After completing the calculation of each feature, the data preprocessing and feature extraction module combines all time-domain and frequency-domain features extracted within the current sliding time window to form a multidimensional original feature vector. This multidimensional original feature vector includes at least: root mean square value of vibration acceleration, peak factor, rate of change of vibration acceleration, triaxial spatial correlation coefficient, mean wind speed, standard deviation of wind speed, slope of wind speed trend, mean strain, slope of strain trend, dominant vibration frequency, centroid of spectrum, kurtosis of spectrum, and dominant strain frequency.

[0067] Then, based on the preset normalization parameters for each dimension of features, the multidimensional original feature vector is subjected to maximum and minimum value normalization or Z-score normalization processing, mapping the feature values ​​of each dimension to a unified numerical range, thereby obtaining the normalized real-time feature vector.

[0068] The normalized real-time feature vector is continuously output step by step as the sliding time window moves, forming a real-time feature vector stream, which is directly used as the input to the wind load and response prediction module.

[0069] In some preferred embodiments, in order to improve the relevance of feature extraction to subsequent wind load and response prediction tasks, a feature refinement model based on an autoencoder network is introduced to perform dimensionality reduction and redundancy removal on the normalized real-time feature vectors and output refined feature vectors.

[0070] The structure and training process of the feature refinement model are as follows: The feature refinement model employs an undercomplete autoencoder structure. The encoder consists of an input layer, two hidden layers, and a bottleneck layer: the number of neurons in the input layer is equal to the dimension of the normalized real-time feature vector; the first hidden layer has half the number of neurons of this dimension and uses the ReLU activation function; the second hidden layer has one-quarter the number of neurons of this dimension and also uses the ReLU activation function; the number of neurons in the bottleneck layer is a preset compact dimension. The decoder is a mirror image of the encoder, with two hidden layers symmetrical to the encoder, and the output layer has the same number of neurons as the input layer, used to reconstruct the input feature vector. The encoder and decoder are fully connected through the bottleneck layer.

[0071] The training data utilizes a large set of standardized real-time feature vector samples collected on-site or generated through simulation, requiring no manual annotation. During training, the standardized real-time feature vectors are used as the input and output labels of the autoencoder. Mean squared error is used as the reconstruction loss function, and the Adam optimizer is employed. The initial learning rate is set to 0.0005, the batch size to 128, and training is iteratively performed for 200 epochs on the sample set. An early stopping strategy is used to prevent overfitting. After training, the encoder portion of the autoencoder, i.e., the network structure and parameters from the input layer to the bottleneck layer, is extracted.

[0072] In real-time operation, the data preprocessing and feature extraction module inputs the normalized real-time feature vector into the encoder. After one forward propagation, a compact and refined feature vector is obtained at the bottleneck layer. This refined feature vector removes redundant information and noise interference from the original features, retaining the intrinsic features most relevant to wind load and vibration response prediction, and is used as the real-time input to the wind load and response prediction module.

[0073] In some embodiments, the data preprocessing and feature extraction module 202 can also be used for: Based on the vibration acceleration signal in the synchronous sampling sequence, calculate the root mean square value and peak factor of vibration acceleration as at least part of the time-domain characteristics; Based on the vibration acceleration signal in the synchronous sampling sequence, the power spectral density is obtained by fast Fourier transform, and the dominant frequency and spectral kurtosis are extracted from the power spectral density as at least part of the frequency domain features.

[0074] The wind load and response prediction module 203 has a built-in time series prediction model based on physical information enhancement, which is used to predict the wind pressure time history and component vibration response envelope within a preset period of time based on historical feature vector sequences and real-time feature vectors.

[0075] In some embodiments, the module incorporates a physically-enhanced Long Short-Term Memory (LSTM) network. Its operation involves: first, concatenating the real-time feature vectors from the current moment and several consecutive time steps prior to it, output by the data preprocessing module, into a historical feature vector sequence in chronological order; then, using this sequence as input to the LSTM network, which, based on learned spatiotemporal mapping relationships, outputs the equivalent wind pressure prediction value for each time step within a preset future period (e.g., 30 seconds to 2 minutes), forming the wind pressure time history; simultaneously, it outputs the vibration acceleration envelope prediction value corresponding to each time step, forming the component vibration response envelope. During offline training, in addition to minimizing the error between the predicted and measured values, the LSTM network also incorporates the residual dynamic equations of the component-support system as a physical constraint, ensuring that the prediction results conform to the physical laws of structural vibration.

[0076] In some embodiments, the data preprocessing and feature extraction module 203 can also be used for: Based on the historical feature vector sequence, construct the input sequence, which contains the real-time feature vectors of the current time and multiple consecutive time steps before it; Based on the input sequence, the equivalent wind pressure prediction value at each time step within a preset period is predicted through a physical information-enhanced long short-term memory network, thus forming the wind pressure time history; Based on the input sequence and wind pressure time history, the component vibration acceleration envelope value at each time step within a preset period is predicted synchronously through a physical information-enhanced long short-term memory network, thus forming the component vibration response envelope.

[0077] The wind load and response prediction module includes an input buffer that stores real-time feature vectors from the most recent time steps in a first-in, first-out manner. Each real-time feature vector corresponds to a time step and carries all normalized features extracted from multi-source sensor data within that time step.

[0078] In each prediction period, this module extracts real-time feature vectors from the input buffer in chronological order, including the current time and several consecutive time steps prior, to form an input sequence. Specifically, if the sequence time window length is set to T time steps, the input sequence contains all real-time feature vectors numbered from (t-T+1) to t, where t represents the current time. Each real-time feature vector maintains its temporal position in the input sequence, and the vectors are arranged in ascending order along the time axis, forming a two-dimensional input tensor of dimension T×M, where M is the feature dimension of the real-time feature vectors.

[0079] The input sequence preserves the temporal evolution information of environmental loads and structural responses in the multi-source feature space, including the recent change trajectories of wind speed, wind direction, turbulence intensity, vibration acceleration characteristics, tilt angle characteristics, and strain characteristics, providing a temporal context for subsequent time series prediction.

[0080] The wind load and response prediction module feeds the input sequence into the encoder part of a physical information-enhanced long short-term memory network.

[0081] The physically enhanced long short-term memory (LSTM) network employs an encoder-dual decoder architecture. The encoder, composed of stacked LSM units, performs temporal encoding on the input sequence. Specifically, the encoder contains two LSM layers, each with H hidden state units. The first LSM layer receives the input sequence, processes it step-by-step, and outputs a hidden state vector at each time step. The hidden state vector at the last time step serves as the temporal summary representation of the input sequence for that layer. The second LSM layer receives the hidden state sequence output from the first layer, processes it step-by-step, and uses the hidden state vector at the last time step as the context vector for the entire encoder. This context vector compressively encodes the key temporal patterns and features related to the dynamic evolution of wind load and vibration response contained in the input sequence.

[0082] The context vector is then fed into the wind pressure decoder. The wind pressure decoder consists of a long short-term memory (LSM) layer and a fully connected output layer. The LSM receives the context vector as its initial hidden state and, at each prediction time step within a preset future time period, receives the prediction output of the previous prediction time step as part of the input for the current step, progressively generating the future wind pressure prediction sequence through iterative expansion. At each prediction time step, the LSM outputs a hidden state vector, which is mapped to a one-dimensional equivalent wind pressure prediction value via the fully connected output layer. The fully connected output layer employs a linear activation function.

[0083] By iteratively expanding N prediction time steps within a preset future period, the wind pressure decoder sequentially outputs N equivalent wind pressure prediction values, which, in chronological order, constitute the wind pressure time history. This wind pressure time history characterizes the predicted evolution of the magnitude of the equivalent wind pressure resultant force acting on the surface of the photovoltaic module within the next N time steps.

[0084] While obtaining the wind pressure time history, the wind load and response prediction module simultaneously predicts the component vibration response envelope through the second decoder.

[0085] Specifically, the context vector output by the encoder is simultaneously fed into the vibration response decoder. The structure of the vibration response decoder is similar to that of the wind pressure decoder, consisting of a long short-term memory (LSM) layer and a fully connected output layer. However, the number of hidden units in its LSM can be independently set to adapt to the dynamic characteristics of the vibration response prediction task.

[0086] The long short-term memory unit of the vibration response decoder receives the context vector as the initial hidden state. At each prediction time step within a preset future period, it receives not only the predicted vibration acceleration envelope value output from the previous prediction time step, but also the equivalent wind pressure prediction value output from the wind pressure decoder at the same prediction time step as external input. This cross-decoder connection establishes a physical causal relationship between wind pressure and vibration response, enabling the prediction of vibration response to use the wind pressure prediction result as the dynamic excitation condition, thereby ensuring the physical consistency of the two in terms of time sequence.

[0087] At each prediction time step, the long short-term memory unit of the vibration response decoder fuses the vibration output from the previous time step and the wind pressure input from the current time step, updates the hidden state, and outputs the hidden state vector to the fully connected output layer. The fully connected output layer maps this hidden state vector to a predicted vibration acceleration envelope value. The fully connected output layer uses a linear activation function.

[0088] By iteratively expanding N prediction time steps within a preset future period, the vibration response decoder sequentially outputs N predicted vibration acceleration envelope values, which, in chronological order, constitute the component's vibration response envelope. This component vibration response envelope characterizes the evolution trend of the upper limit of the vibration acceleration amplitude of the photovoltaic module under predicted wind pressure excitation within the next N time steps.

[0089] The training of the aforementioned physically-enhanced Long Short-Term Memory (LSTM) network does not rely solely on data-driven error backpropagation. Instead, it jointly optimizes data loss and physical constraint loss, enabling the network to learn input-output mappings that simultaneously fit the patterns of measured data and the physical laws of structural dynamics. The specific training process is as follows: The training dataset was obtained through field measurements and / or numerical simulations. For different operating conditions of the photovoltaic module-support system, multiple sets of continuous time-series samples were collected or generated. Each set of samples included: an input sequence (a real-time feature vector sequence over T consecutive time steps), a wind pressure label sequence (corresponding to the measured or simulated equivalent wind pressure over the next N time steps), and a vibration acceleration envelope label sequence (corresponding to the measured or simulated vibration acceleration envelope over the next N time steps). All samples were randomly divided into training, validation, and test sets according to a preset ratio.

[0090] Random weight initialization was performed on the two long short-term memory layers in the encoder, the long short-term memory layers in the wind pressure decoder and vibration response decoder, and their respective output fully connected layers. The weights of the long short-term memory layers were initialized using Xavier uniform initialization, the weights of the fully connected layers were initialized using He normal initialization, and the bias term was initialized to zero.

[0091] In each iteration of training, a batch of samples is randomly drawn from the training set. For each sample, the input sequence is fed into the encoder, and the predicted wind pressure time history and component vibration response envelope are generated according to the forward propagation process.

[0092] Then, the data loss term and the physical constraint loss term are calculated separately.

[0093] The data loss term consists of two parts: the wind pressure prediction loss is obtained by comparing the predicted wind pressure time history with the wind pressure label sequence and calculating the mean square error of both; the vibration response prediction loss is obtained by comparing the predicted component vibration response envelope with the vibration acceleration envelope label sequence and calculating the mean square error of both. The wind pressure prediction loss and the vibration response prediction loss are weighted and summed to form the total data loss term.

[0094] The physical constraint loss term is calculated as follows: Based on the structural dynamics equations of the photovoltaic module-support system, a physical relationship is established between the predicted wind pressure time history and the module vibration response envelope. Specifically, the module-support system is simplified into an equivalent single-degree-of-freedom vibration system, whose equations of motion include mass, damping, and stiffness parameters, as well as an external excitation wind pressure term. At each future prediction time step, the residuals on both sides of the equations of motion are calculated based on the predicted wind pressure value, the predicted vibration acceleration envelope output by the vibration response decoder, and the estimated values ​​of the corresponding velocity and displacement magnitudes. The sum of squares or absolute values ​​of the residuals from all future time steps is taken as the physical constraint loss term. The estimated values ​​of the velocity and displacement magnitudes are approximated by performing a first and second integral on the predicted vibration acceleration envelope.

[0095] The final joint loss function is obtained by weighted summation of the total data loss term and the physical constraint loss term, with the weight coefficients determined based on the tuning results on the validation set.

[0096] The joint loss function is backpropagated along the network to calculate the gradients of the weights and biases of each layer. The Adam optimizer is used to update the network parameters based on the gradients. The learning rate is initially set to 0.001, and a cosine annealing learning rate decay strategy is used to gradually reduce the learning rate during training to improve convergence stability.

[0097] The forward propagation, loss calculation, and backpropagation are repeated until the preset maximum number of iterations is reached. After every few iterations, the joint validation loss is calculated on the validation set. When the joint validation loss no longer decreases for several consecutive iterations, an early stopping mechanism is triggered to terminate training, and the model with the minimum joint validation loss is saved as the final deployed physical information-enhanced long short-term memory network.

[0098] In actual operation, the wind load and response prediction module loads the trained network parameters, inputs the real-time constructed input sequence into the encoder, and outputs the predicted wind pressure time history and component vibration response envelope through a forward propagation of the wind pressure decoder and vibration response decoder, without the need for online calculation of physical constraint loss terms.

[0099] The training of Long Short-Term Memory (LSTM) networks includes: Based on the predicted wind pressure time history and component vibration response envelope, and combined with the structural dynamics equations of the photovoltaic module-support system, the physical residuals are calculated. Based on the physical residuals and data prediction errors, a joint loss function is constructed to enable the network prediction results to simultaneously fit the measured data and the physical laws of structural dynamics.

[0100] Before constructing physical constraints, an equivalent dynamic model of the photovoltaic module-support system is first established. In this embodiment, considering that the main vibration mode of the module under wind load is bending vibration perpendicular to the module surface, the module-support system is simplified into an equivalent single-degree-of-freedom mass-spring-damped system. The parameters of this equivalent system are pre-calibrated through field modal testing or finite element analysis.

[0101] The equation of motion for an equivalent single-degree-of-freedom system is expressed as follows: at any given moment, the sum of the product of the system's mass and vibration acceleration, the product of the damping coefficient and vibration velocity, and the product of the stiffness coefficient and vibration displacement equals the external excitation force acting on the system at that moment. The external excitation force is determined by multiplying the equivalent wind pressure acting on the component surface by the effective wind-receiving area; the mass, damping coefficient, and stiffness coefficient correspond to the system's equivalent modal mass, equivalent modal damping, and equivalent modal stiffness, respectively.

[0102] For discrete time steps, the equations of motion are discretized in the time domain with sampling periods. Assume that at the time corresponding to the k-th prediction time step, there exists a recursive relationship between the vibration acceleration, velocity, and displacement, determined by numerical integration. This embodiment employs the Newmark-β method for time-domain discretization. This method transforms the equations of motion into algebraic equations containing only unknown accelerations by introducing recursive expressions for velocity and displacement.

[0103] Specifically, the velocity at step k can be recursively derived from the velocity at step (k-1), the acceleration at step (k-1), and the acceleration at step k, combined with the time step and Newmark parameters; the displacement at step k can be recursively derived from the displacement at step (k-1), the velocity at step (k-1), the acceleration at step (k-1), and the acceleration at step k, combined with the time step and Newmark parameters. Substituting the recursive expressions for velocity and displacement into the equations of motion yields a discrete equation of motion, with the acceleration at step k as the unknown, expressed by the state variables of the previous step and the external excitation force of the current step.

[0104] During training, for each training sample, the wind load and response prediction module outputs prediction results including wind pressure time histories and component vibration response envelopes for the next N prediction time steps. Each element of the wind pressure time histories is the equivalent wind pressure prediction value for the corresponding prediction time step; each element of the component vibration response envelope is the vibration acceleration envelope prediction value for the corresponding prediction time step, which in this embodiment serves as the system's vibration acceleration estimate for that prediction time step.

[0105] The calculation of physical residuals is performed according to the following steps: For each prediction time step, the equivalent wind pressure prediction value for that step is multiplied by the effective wind-receiving area of ​​the component to obtain the external excitation force acting on the equivalent single-degree-of-freedom system. By iterating through N future prediction time steps, a sequence of external excitation forces is obtained.

[0106] The initial state is either the actual vibration state of the time step preceding the prediction start time, or the state at the end of the previous prediction period. Assume the vibration acceleration, velocity, and displacement at the initial time are known.

[0107] Starting from the first prediction time step, using the recursive relationship of the Newmark-β method, the velocity and displacement estimates for the k-th step are recursively calculated based on the velocity and displacement at the (k-1)-th step, the acceleration at the (k-1)-th step, and the predicted vibration acceleration envelope value at the current k-th step. The specific recursive formula is as follows: First, according to the displacement recursion of the Newmark-β method, the displacement estimate at step k is expressed as the sum of the displacement value at step (k-1), the product of the velocity value at step (k-1) and the time step, and a weighted combination of the predicted acceleration values ​​at steps (k-1) and (k-1), and the acceleration envelope predictions at step k. The weighting coefficients used in this weighted combination are determined by the Newmark parameter β and the square of the time step.

[0108] Secondly, according to the velocity recursion of the Newmark-β method, the velocity estimate at step k is expressed as the sum of the velocity value at step (k-1) and the weighted combination of the acceleration predictions at steps (k-1) and (k-1), and the acceleration envelope predictions at step k. The weighting coefficients used in the weighted combination are determined by the Newmark parameter γ and the time step.

[0109] Through the above recursive process, the velocity estimation sequence and displacement estimation sequence for the next N predicted time steps are calculated one by one.

[0110] For each prediction time step, the sum of the following three factors is obtained: the product of the equivalent mass and the predicted acceleration envelope at the k-th step; the product of the equivalent damping coefficient and the estimated velocity at the k-th step; and the product of the equivalent stiffness coefficient and the estimated displacement at the k-th step. This sum of internal forces is then compared with the external excitation force at the k-th step, and the difference between the two is the instantaneous physical residual for that prediction time step.

[0111] The instantaneous physical residual reflects whether the wind pressure and vibration response predicted by the network at the current time step satisfies the structural dynamics equations of motion. If the instantaneous physical residual is zero, it indicates that the prediction results fully conform to the dynamic equations; if the instantaneous physical residual is non-zero, it indicates that the prediction results deviate at the level of physical laws.

[0112] By iterating through N future prediction time steps, N instantaneous physical residuals are obtained, forming an instantaneous physical residual sequence.

[0113] The N instantaneous physical residuals are aggregated to obtain the final physical residual term, which is used to construct the loss function. The aggregation can be performed using one of the following two methods: Method 1: Square the N instantaneous physical residuals, then sum them to obtain a scalar value as the physical residual term.

[0114] Method 2: Square the N instantaneous physical residuals, calculate the average, and then take the square root to obtain a root mean square physical residual term.

[0115] This embodiment preferably adopts Method 1 to emphasize the total magnitude of the physical residuals while maintaining the simplicity of gradient calculation.

[0116] The joint loss function is a weighted combination of the data prediction error term and the physical residual term, which enables the network to minimize both the prediction bias of the labeled data and the degree of violation of physical laws during the training process.

[0117] The calculation method for wind pressure prediction loss is as follows: Iterate through N future prediction time steps, compare the equivalent wind pressure prediction value of each prediction time step with the corresponding measured or simulated wind pressure value in the wind pressure label sequence, and calculate the square of the difference between the two; sum or average the N squared differences to obtain the wind pressure prediction loss value.

[0118] The method for calculating the vibration response prediction loss is as follows: Iterate through the next N prediction time steps, compare the predicted value of the vibration acceleration envelope at each prediction time step with the corresponding measured or simulated envelope value in the vibration acceleration envelope label sequence, and calculate the square of the difference between the two; sum or average the N squared differences to obtain the vibration response prediction loss value.

[0119] The wind pressure prediction loss value and the vibration response prediction loss value are multiplied by their respective preset weighting coefficients and then summed to obtain the total data prediction error term. This error term prompts the network output to approximate the true label in the training data.

[0120] The final joint loss function is formed by adding the product of the physical residual term and the physical constraint weights to the total data prediction error term. The specific expression is: Joint loss = Total data prediction error term + λ × Physical residual term.

[0121] Here, λ is the physical constraint weight coefficient, a preset hyperparameter used to adjust the relative importance of physical constraints during training. A larger λ makes the model more strictly follow the dynamic equations, but may reduce the fitting accuracy to the labeled data; a smaller λ allows the model to have greater freedom in data fitting. The value of λ is determined by performing a grid search on the validation set or by gradient-based hyperparameter optimization methods. In this embodiment, λ is initially set to 0.1 and can be dynamically adjusted according to changes in the validation loss during training.

[0122] In each iteration of training, forward propagation calculates the predicted wind pressure time history and component vibration response envelope, and then calculates the physical residual term according to the aforementioned process. During backpropagation, the gradient of the joint loss function with respect to the network parameters includes both the gradient of the data prediction error with respect to the parameters and the gradient of the physical residual term with respect to the parameters.

[0123] Specifically, the gradient of the physical residual term with respect to the vibration response decoder parameters is propagated through the following chain: the physical residual depends on the predicted vibration acceleration envelope and the velocity and displacement estimates derived from it. Therefore, a small change in the predicted vibration acceleration envelope will simultaneously affect the instantaneous physical residual and the velocity and displacement estimates generated by the recursive relationship in subsequent time steps, thus affecting the physical residual in subsequent time steps. During backpropagation, the gradient is automatically calculated along this dependency chain, enabling the vibration response decoder to learn not only how to fit the labeled data but also how to generate a vibration response sequence that satisfies a dynamic causal relationship with the predicted wind pressure.

[0124] Meanwhile, the gradient transmission of the physical residual term to the wind pressure decoder parameters is achieved through the external excitation force term: the equivalent wind pressure prediction value directly affects the external excitation force, and thus affects the residual of the motion equation; the wind pressure decoder is therefore constrained so that its predicted wind pressure time history is not only close to the label in amplitude, but also maintains dynamic consistency with the vibration response in terms of time sequence change pattern.

[0125] By embedding the dynamic equations as soft constraints into the loss function, the physically-enhanced long short-term memory network can still utilize known physical knowledge to regulate its input-output mapping space under conditions of scarce data or high noise, suppress non-physical overfitting, and improve the model's generalization ability and prediction reliability for unseen conditions.

[0126] The vibration risk quantification module 204 is used to calculate the comprehensive vibration risk index in real time based on the predicted wind pressure time history and component vibration response envelope within a preset future period, combined with the material cumulative damage model.

[0127] In some embodiments, this module calculates a comprehensive vibration risk index in real time based on the predicted future wind pressure time history and the component vibration response envelope. Specifically, this includes: traversing each time step in the future time history and calculating the equivalent stress amplitude of key points of the component within that step based on the wind pressure and acceleration envelope values; statistically analyzing the degree to which the equivalent stress amplitude exceeds the allowable fatigue stress of the material to obtain a stress over-limit penalty term; simultaneously, starting from time zero, substituting the equivalent wind pressure and corresponding acceleration amplitudes of each past step into a cumulative integral function containing exponential damping decay to obtain a wind-induced vibration cumulative damage term. This decay term reflects the dissipation effect of the structure's own damping on historical energy input, gradually weakening the wind load influence at earlier times; finally, weighted summing of the cumulative damage term and the stress over-limit penalty term yields the comprehensive vibration risk index at the current moment. The larger this index, the higher the risk of future wind-induced vibration causing cumulative damage and instantaneous failure to the structure.

[0128] In some embodiments, the vibration risk quantification module 204 can also be used for: Based on the predicted wind pressure time history, determine the wind pressure loading sequence that represents the equivalent wind pressure over time. Based on the wind pressure loading sequence and the component vibration response envelope, the equivalent stress amplitude sequence at key points of the component is determined; Based on the equivalent stress amplitude sequence and the allowable fatigue stress of the component material, the stress over-limit penalty term is determined in real time. Based on the equivalent wind pressure at each past moment in the wind pressure loading sequence, the acceleration amplitude at the corresponding moment in the component vibration response envelope, and the system damping attenuation term, the cumulative damage term of wind-induced vibration is determined by integral summation. The comprehensive vibration risk index is obtained by weighted summation of the cumulative damage term caused by wind-induced vibration and the stress over-limit penalty term.

[0129] The vibration risk quantification module includes a wind pressure loading sequence construction sub-unit, an equivalent stress calculation sub-unit, a stress over-limit penalty calculation sub-unit, a wind-induced vibration cumulative damage calculation sub-unit, and a risk index synthesis sub-unit. Each sub-unit is executed sequentially according to the following steps.

[0130] The wind pressure loading sequence construction sub-unit receives the wind pressure time history output by the wind load and response prediction module. This wind pressure time history contains a sequence of equivalent wind pressure prediction values ​​for the next N prediction time steps. Each equivalent wind pressure prediction value represents the magnitude of the wind pressure resultant force acting on the equivalent resultant force point on the surface of the photovoltaic module within the corresponding time step.

[0131] The wind pressure loading sequence construction subunit arranges the equivalent wind pressure prediction values ​​for each prediction time step in the wind pressure time history in chronological order, and multiplies them by the effective wind-receiving area of ​​the component at each time step to obtain the normal wind load sequence acting on the component surface. This normal wind load sequence corresponds one-to-one with the prediction time step, constituting the wind pressure loading sequence.

[0132] Meanwhile, the wind pressure loading sequence construction subunit also converts the predicted time step index corresponding to the wind pressure time history into a time offset with the current time as the zero point, providing a time reference for subsequent cumulative damage calculation.

[0133] The equivalent stress calculation subunit receives the wind pressure loading sequence output by the wind pressure loading sequence construction subunit, as well as the component vibration response envelope output by the wind load and response prediction module.

[0134] In this embodiment, the key points of the component are pre-calibrated using finite element analysis and / or field strain gauge measurement data, and selected as one or more locations with the highest stress concentration and shortest fatigue life under the combined action of wind load and vibration. Typical key points include the edge of the bolt hole connecting the component frame and the bracket, the heat-affected zone of the weld at the root of the bracket column, etc.

[0135] For each future prediction time step, the equivalent stress calculation sub-unit calculates the equivalent stress magnitude at the key point in the following manner: First, based on the normal wind load in the wind pressure loading sequence corresponding to the predicted time step, and the lever arm length of the normal wind load relative to the key point, the bending moment acting on the key point is calculated. This lever arm length is obtained by projecting the spatial vector from the resultant force application point to the key point in a direction perpendicular to the bending moment plane. When the wind load and response prediction module only outputs the magnitude of the equivalent wind pressure resultant force, the resultant force application point is taken from a pre-calibrated typical location; when the multi-source sensor acquisition module has provided a real-time resultant force application point, this real-time resultant force application point is used.

[0136] Secondly, based on the predicted vibration acceleration envelope value in the component vibration response envelope corresponding to the predicted time step, the inertial force acting on the key point is calculated. Specifically, the equivalent mass of the structural sub-block where the key point is located is multiplied by the predicted vibration acceleration envelope value to obtain the dynamic inertial force of the sub-block; then, based on the lever arm length of the dynamic inertial force relative to the key point, the additional bending moment generated by the vibration inertial force is calculated.

[0137] Next, the wind load bending moment and the vibration inertial force bending moment are vector-synthesized to obtain the total bending moment amplitude at the key point. Based on the total bending moment amplitude, the section modulus of the cross section at the key point, and the cross section geometry parameters, the equivalent stress amplitude at the key point for this prediction time step is calculated. The calculation method is to divide the total bending moment amplitude by the section modulus of the cross section and then multiply it by the stress concentration factor, which is determined by the local geometry and material properties of the key point through tables or empirical formulas.

[0138] By iterating through N future prediction time steps, the equivalent stress amplitude is calculated for each prediction time step to obtain the equivalent stress amplitude sequence. This equivalent stress amplitude sequence characterizes the evolution trend of the stress level at key points over time within a predetermined future period.

[0139] The stress over-limit penalty calculation subunit receives the equivalent stress amplitude sequence and the pre-stored allowable fatigue stress value of the component material at the corresponding stress ratio. The allowable fatigue stress value is determined based on the stress-life curve of the material at the critical point of the component, taking into account a preset fatigue safety factor.

[0140] For each equivalent stress amplitude in the equivalent stress amplitude sequence, the stress over-limit penalty calculation sub-unit performs the following over-limit judgment and penalty calculation operations: The equivalent stress amplitude is compared with the allowable fatigue stress value. If the equivalent stress amplitude is less than or equal to the allowable fatigue stress value, the corresponding single-step penalty value is zero; if the equivalent stress amplitude is greater than the allowable fatigue stress value, the excess amount of the equivalent stress amplitude over the allowable fatigue stress value is calculated, the excess amount is divided by the allowable fatigue stress value to obtain the relative over-limit ratio, and the square of the relative over-limit ratio or the first power term is taken to obtain the single-step penalty value for this prediction time step.

[0141] After iterating through all prediction time steps, the single-step penalty values ​​of each step are summed to obtain the total stress over-limit penalty. Further, this total stress over-limit penalty is multiplied by a preset stress over-limit penalty gain coefficient to obtain the stress over-limit penalty term.

[0142] The stress overload penalty term has the following characteristics: when the equivalent stress of all predicted time steps is within the allowable range of the material, the term is zero and does not introduce additional risk; when stress overload occurs at any time step, the term increases sharply with the magnitude and duration of the overload, thus reflecting the serious threat of instantaneous overload to structural safety in the comprehensive risk index.

[0143] The wind-induced vibration cumulative damage calculation subunit receives the wind pressure loading sequence and its time offset output by the wind pressure loading sequence construction subunit, as well as the acceleration amplitude of the corresponding time step in the component vibration response envelope output by the wind load and response prediction module.

[0144] This sub-unit is designed based on the following physical understanding: the damage caused by wind-induced vibration to the photovoltaic module-support system has a time-cumulative effect. The wind pressure impact and vibration energy from the past do not disappear immediately, but gradually decay under the action of structural damping; the contribution of energy input from earlier times to the current risk state decays exponentially over time.

[0145] Therefore, the wind-induced vibration cumulative damage calculation sub-unit implements integral summation calculation with damping attenuation in a discrete time step recursive manner. The specific process is as follows: The initial value of the cumulative damage register D is set to zero. Simultaneously, the calculation starting point is set to the beginning of a sufficiently long decay memory time window traversing back in the historical direction from the current moment, or from the system startup moment. In this embodiment, the length of the decay memory time window is taken as the time length corresponding to several first-order oscillation cycles of the system. This length ensures that historical events exceeding this window decay to a negligible level.

[0146] Starting from the beginning of the decaying memory time window, recursively calculate each time step until the last historical time step corresponding to the current moment. For each historical time step, perform the following recursive operation: First, the equivalent wind pressure value and vibration acceleration amplitude corresponding to this historical time step are read. The equivalent wind pressure value comes from the recorded value in the wind pressure loading sequence that was actually recorded or predicted for this historical time step; the vibration acceleration amplitude comes from the recorded value of the component vibration response envelope that was actually measured or predicted for this historical time step.

[0147] Secondly, the instantaneous damage contribution value for this historical time step is calculated. This is done by multiplying the equivalent wind pressure value for that step by a coupling amplification factor that includes the influence of vibration acceleration amplitude, thus obtaining the wind-vibration coupled energy term for that step. The coupling amplification factor is expressed as a unit value plus the absolute value of the ratio of vibration acceleration amplitude to gravitational acceleration, multiplied by an acceleration-wind pressure coupling amplification coefficient. This coupling amplification coefficient is determined through wind tunnel testing or field calibration, and its physical meaning is: when a component is simultaneously subjected to wind pressure and vibration, the additional aerodynamic effects and structural inertial effects caused by vibration amplify the actual destructive capacity of the wind pressure.

[0148] Next, the instantaneous damage contribution value is accumulated into the cumulative damage register D. However, before accumulation, the current stored value of the cumulative damage register D is multiplied by a damping attenuation factor, causing the old accumulated value in the register to undergo exponential decay over one time step. The damping attenuation factor is obtained by taking the negative of the exponential function of the product of the system's equivalent damping ratio, the system's first-order natural angular frequency, and the time step.

[0149] This recursive structure of "attenuation first, then accumulation" ensures that the contribution of each new damage step is fully recorded, while the historical contribution gradually "fades away" over time, and its attenuation rate is determined by the damping characteristics of the system itself.

[0150] After the recursive calculation covers all historical time steps, the value stored in the cumulative damage register D is the original cumulative damage amount that takes into account the attenuation effect of the complete wind load vibration history. Multiplying this original cumulative damage amount by the wind-induced vibration cumulative damage weighting coefficient yields the wind-induced vibration cumulative damage term.

[0151] The risk index synthesis sub-unit receives the stress over-limit penalty term and the output wind-induced vibration cumulative damage term, and synthesizes the comprehensive vibration risk index through a weighted summation method.

[0152] Specifically, the comprehensive vibration risk index R is calculated by directly adding the cumulative damage term of wind-induced vibration and the stress over-limit penalty term. The weighting coefficients of each term are already included in the cumulative damage weighting coefficient of wind-induced vibration and the stress over-limit penalty gain coefficient, respectively, without the need for additional weighting factors.

[0153] The resulting comprehensive vibration risk index R is a scalar value, and its magnitude comprehensively reflects the superposition of the following two types of risk information: first, from the perspective of time accumulation, it represents the residual damage energy of past and current wind-induced vibration events under the damping attenuation of the system; second, from the perspective of instantaneous stress, it represents the degree of danger that the stress at key points may exceed the allowable range of the material in the future prediction period.

[0154] The comprehensive vibration risk index R is output to the vibration reduction control decision module, serving as the basis for subsequent multi-level threshold comparisons and control command generation. When the comprehensive vibration risk index R is low, it indicates that the structural stress and cumulative damage are within a safe range in the current and future periods; as R gradually increases, it indicates that the structure faces increased wind-induced vibration risk, requiring corresponding increases in damping and / or adjustments to the support posture to reduce the risk level.

[0155] In some embodiments, the vibration risk quantification module 204 includes a comprehensive vibration risk index. The following formula can be used for calculation:

[0156] in, For the current moment The comprehensive vibration risk index is dimensionless. The higher the value, the higher the combined risk of cumulative wind-induced vibration damage and instantaneous stress failure faced by the system at that moment. It is directly output to the vibration reduction control decision module, serving as the sole decision variable for multi-level threshold comparison and risk level determination. Its dimensionless nature allows different photovoltaic module-support systems to be evaluated using a unified risk threshold system, requiring only system-level calibration of each weight coefficient.

[0157] for The equivalent wind pressure acting on the component surface at all times is calculated in real time based on wind speed, wind direction, air density, and aerodynamic coefficients. The real-time calculation process for this equivalent wind pressure is as follows: the ultrasonic anemometer in the multi-source sensor acquisition module acquires the scalar value of wind speed and the azimuth angle of wind direction at time τ; the tilt angle of the support at time τ is acquired by the tilt sensor; the incoming flow angle and relative wind speed are determined based on the geometric relationship between wind speed, wind direction, and support tilt angle; based on the incoming flow angle and relative wind speed, the aerodynamic pressure coefficients of each zone on the component surface are obtained by looking up a table or interpolation in the aerodynamic coefficient mapping table; then, the aerodynamic pressure coefficients of each zone are multiplied by the square of the air density and the relative wind speed, and then weighted and synthesized to obtain the equivalent wind pressure acting on the entire component surface. Air density is calculated in real time based on the ambient temperature, humidity, and atmospheric pressure at time τ. It is the fundamental excitation quantity of the integrand in the cumulative damage term of wind-induced vibration, and its magnitude directly determines the intensity of wind-borne energy input at each historical moment.

[0158] for The measured vibration acceleration amplitude on the surface of the component at any given time. This refers to gravitational acceleration. Specifically, the triaxial accelerometer outputs vibration acceleration components in the normal and two tangential directions of the component surface, respectively, and takes the vibration acceleration of the normal component as... Or, the vector sum of the three-axis acceleration components can be used as... This value characterizes the intensity of vibration of the component at time τ and can directly reflect the level of dynamic response of the structure caused by wind load.

[0159] The acceleration-wind pressure coupling amplification factor is determined based on wind tunnel tests or field calibration. This factor characterizes the amplification of the actual destructive power of wind pressure caused by the additional aeroelastic effects and structural inertial effects caused by vibration when the module is simultaneously subjected to wind pressure and vibration. The physical mechanism is as follows: when the module vibrates under wind load, the instantaneous velocity of the module surface relative to the incoming flow changes, resulting in a dynamic difference between the actual aerodynamic load and the wind pressure predicted based on static assumptions. Simultaneously, when the vibration inertial force and wind pressure are superimposed in the same direction, the actual stress at key points will be greater than the stress calculation result considering only static wind pressure. By applying forced vibrations of different frequencies and amplitudes to scale-down or full-size photovoltaic module models in wind tunnel tests, and simultaneously measuring the actual wind pressure distribution and stress at key points on the module surface, the dynamic wind pressure or dynamic stress is compared with the values ​​under the corresponding static conditions, and statistical analysis is performed to obtain the coupling amplification factor. The calibration value. Under field conditions, the measured wind pressure, vibration acceleration, and strain data from strong wind events can also be used to back-estimate the values ​​using system identification methods. .generally It is a constant greater than zero, with a value between 0.05 and 0.5. The specific value is related to the geometry of the component, the stiffness of the support, and the aerodynamic shape.

[0160] This is the equivalent damping ratio of the photovoltaic module-support system. The damping ratio is the system's first natural frequency, both of which are obtained in advance through modal analysis. This parameter characterizes the rate at which the amplitude decays during free vibration. The larger the damping ratio, the stronger the system's ability to dissipate vibrational energy, and the faster the vibration decays. The value range is usually between 0.01 and 0.10, with steel structure supports tending towards the lower limit and systems with more dampers or connectors tending towards the upper limit.

[0161] The damping ratio is obtained during system installation and commissioning or regular maintenance through on-site hammer-based modal testing. Specifically, accelerometers are installed on the component surface or at key locations on the support frame, and a force hammer is used to apply pulse excitation to collect the free decay vibration response signal. The decay signal is then analyzed using the logarithmic subtraction method or modal parameter identification based on the frequency response function to extract the damping ratio corresponding to the first-order mode of the system. Alternatively, it can be obtained through working modal analysis, using vibration response data under natural wind load excitation for random subspace identification.

[0162] This is the damping attenuation term, characterizing the residual effect of energy applied in the past at the current time. This term is the core time-weighted mechanism in the wind-induced vibration cumulative damage term; its function is to assign a time interval from that historical time τ to the energy input at the current time. The weighting coefficient increases and decays exponentially.

[0163] When the historical moment τ is very close to the current moment t, that is... When I was very young, The value is close to 1, meaning that the energy contribution at that historical moment is almost fully preserved; as the time interval... As the exponent increases, the value of the exponent term changes according to... and The jointly determined time constant decreases, and the energy contribution of that historical moment is gradually "forgotten". The decay rate depends on and The product of: The larger the value, the faster the system damping dissipates; The higher the value, the shorter the system's oscillation period and the faster the energy decays. Therefore, this damping attenuation term allows the formula to distinguish the different contributions of recent strong wind events and long-term dissipated events to the current risk state, giving the formula "memory" capabilities while avoiding infinite accumulation.

[0164] The cumulative damage weighting coefficient for wind-induced vibration. This is the stress over-limit penalty weighting coefficient.

[0165] This is the equivalent fatigue stress of the component calculated based on vibration response and wind pressure, expressed in megapascals (MPA). The calculation process is as follows: Based on... Equivalent wind pressure at any moment Calculate the wind load bending moment acting on the key point based on the location of the resultant force application point; vibration acceleration amplitude at time 1 Calculate the inertial bending moment acting on the key point by taking the equivalent mass of the structural sub-block containing the key point; after vector synthesis of the wind load bending moment and the inertial bending moment, divide by the section modulus of the section at the key point and multiply by the stress concentration factor to obtain... When considering multiaxial stress states, the von Mises equivalent stress or the maximum principal stress can be used as the basis for determining the stress state. .

[0166] This represents the allowable fatigue stress of the material, measured in megapascals (MPa). The method for determining this value is as follows: based on the stress-life curve of the material at key points, under preset fatigue life design requirements and a safety factor, extract the allowable stress amplitude or the maximum allowable stress under the corresponding stress ratio conditions. For example, for aluminum alloy frame materials, if the design requires an infinite lifespan, the fatigue limit of the material is divided by the fatigue safety factor; if the design is for a finite lifespan, the fatigue strength at the target number of cycles is divided by the safety factor. In the formula, it serves as a comparison benchmark when... When the stress exceeds this benchmark, the stress over-limit penalty term is activated.

[0167] This formula accumulates the historical wind pressure effect through integration and uses a damping attenuation term to characterize the system's energy dissipation. It also introduces an acceleration coupling term and a stress over-limit penalty term, which solves the problem that traditional threshold methods cannot reflect vibration memory effects and cumulative damage.

[0168] The vibration reduction control decision module 205 is used to generate control commands for the adjustable damper and / or the support tilt adjustment mechanism based on the comparison results of the comprehensive vibration risk index and the preset multi-level risk threshold.

[0169] In some embodiments, this module calculates a comprehensive vibration risk index in real time based on the predicted future wind pressure time history and the component vibration response envelope. Specifically, this includes: traversing each time step in the future time history and calculating the equivalent stress amplitude of key points of the component within that step based on the wind pressure and acceleration envelope values; statistically analyzing the degree to which the equivalent stress amplitude exceeds the allowable fatigue stress of the material to obtain a stress over-limit penalty term; simultaneously, starting from time zero, substituting the equivalent wind pressure and corresponding acceleration amplitudes of each past step into a cumulative integral function containing exponential damping decay to obtain a wind-induced vibration cumulative damage term. This decay term reflects the dissipation effect of the structure's own damping on historical energy input, gradually weakening the wind load influence at earlier times; finally, weighted summing of the cumulative damage term and the stress over-limit penalty term yields the comprehensive vibration risk index at the current moment. The larger this index, the higher the risk of future wind-induced vibration causing cumulative damage and instantaneous failure to the structure.

[0170] In some embodiments, the vibration damping control decision module 205 can also be used for: Based on the comparison between the comprehensive vibration risk index and the first risk threshold, and when the comprehensive vibration risk index is less than the first risk threshold, a low-damping control command is generated to put the adjustable damper at the preset lowest damping level and keep the support at the power generation tilt angle. Based on the comparison results of the comprehensive vibration risk index with the first risk threshold and the second risk threshold, and when the comprehensive vibration risk index is greater than or equal to the first risk threshold and less than the second risk threshold, a medium damping control command and an inclination fine-tuning command are generated to adjust the adjustable damper to the preset medium damping level and to rotate the bracket inclination angle by a first preset angle in the direction of reducing the windward side. Based on the comparison between the comprehensive vibration risk index and the second risk threshold, and when the comprehensive vibration risk index is greater than or equal to the second risk threshold, a high-damping control command and a safe posture command are generated, causing the adjustable damper to be adjusted to the preset highest damping level, and the support tilt angle to the minimum windward posture or the flat posture.

[0171] The vibration reduction control decision module includes a threshold storage unit, a comparison and judgment unit, an instruction generation unit, and a status management unit. These units work together and execute the following steps.

[0172] The threshold storage unit stores at least two preset risk thresholds, namely a first risk threshold and a second risk threshold, wherein the first risk threshold is lower than the second risk threshold. These two thresholds divide the range of the comprehensive vibration risk index into three risk level ranges: low risk range, medium risk range, and high risk range.

[0173] The first and second risk thresholds are determined as follows: First, during the offline debugging or historical operation data accumulation phase of the system, the time series of the comprehensive vibration risk index of the photovoltaic module-support system under different wind conditions is collected, and the corresponding measured stress data and vibration amplitude data of key points of the module are recorded simultaneously.

[0174] Secondly, based on the allowable fatigue stress of the component materials and the preset fatigue safety factor, the upper limit of the acceptable stress level under long-term safe operation conditions is determined. The comprehensive vibration risk index for the time period corresponding to when the measured stress data is lower than this upper limit of the acceptable stress level is statistically analyzed, and the preset upper quantile of this statistical distribution is taken as the benchmark value of the first risk threshold.

[0175] Next, the allowable limit of instantaneous stress is determined by multiplying the yield strength or ultimate strength of the component material by the instantaneous safety factor. The comprehensive vibration risk index for the time period corresponding to when the measured stress data reaches or exceeds this allowable limit of instantaneous stress is statistically analyzed, and the preset lower quantile of this statistical distribution is taken as the benchmark value for the second risk threshold.

[0176] Once the baseline values ​​for the first and second risk thresholds are determined, they can be adaptively adjusted during actual operation based on seasonal characteristics, component aging status, and maintenance feedback.

[0177] During real-time operation, the comparison and judgment unit obtains the latest calculated comprehensive vibration risk index R from the vibration risk quantification module at a preset control cycle.

[0178] The comparison and judgment unit first compares the comprehensive vibration risk index R with the first risk threshold. When R is determined to be less than the first risk threshold, the system is confirmed to be in a low-risk state.

[0179] Under low-risk conditions, the instruction generation unit generates control instructions in the following manner: The instruction generation unit queries a pre-stored damping level mapping table, which maps different risk levels to the damping levels of the adjustable damper. When the risk level is determined to be low risk, the corresponding damping level is the preset lowest damping level. This lowest damping level corresponds to the minimum excitation current or minimum operating voltage of the adjustable damper, enabling the adjustable damper to provide structural damping, allowing the components to vibrate freely under light wind conditions to release residual stress, while avoiding unnecessary energy consumption and damper wear. The instruction generation unit generates a low-damping control instruction containing the excitation current or voltage value corresponding to this lowest damping level.

[0180] The instruction generation unit does not generate tilt adjustment instructions under low-risk conditions; alternatively, it generates a support tilt angle maintenance instruction, which instructs the support tilt adjustment mechanism to maintain the currently preset power generation tilt angle. The power generation tilt angle is determined based on the photovoltaic maximum power point tracking strategy, typically the angle that minimizes the angle of solar incidence, to ensure that power generation efficiency is not reduced due to unnecessary tilt adjustments.

[0181] The state management unit marks the current control state as "low-damping-generation mode" and records the current timestamp and the corresponding comprehensive vibration risk index value for hysteresis judgment during subsequent control state switching.

[0182] When the comparison and judgment unit determines that the comprehensive vibration risk index R is greater than or equal to the first risk threshold and less than the second risk threshold, it confirms that the current system is in a medium-risk state.

[0183] Under medium-risk conditions, the risk of wind-induced vibration in the system has increased significantly, but has not yet reached a level that endangers structural safety. At this point, it is necessary to take appropriate control measures while ensuring structural safety and considering power generation benefits. The command generation unit generates control commands in the following manner: The instruction generation unit queries the damping level mapping table to obtain the medium damping level corresponding to the medium-risk state. This medium damping level corresponds to the medium excitation current or operating voltage of the adjustable damper. At this level, the adjustable damper provides a damping force significantly higher than the lowest damping level, effectively dissipating wind-induced vibration energy and suppressing further increases in component vibration amplitude, while still maintaining a certain damping adjustment margin compared to the highest damping level. The instruction generation unit generates a medium damping control instruction containing the excitation current or voltage value corresponding to this medium damping level.

[0184] The instruction generation unit simultaneously generates a fine-tuning instruction for the support tilt angle. Specifically, the instruction generation unit calculates the amount of tilt angle change that needs to be adjusted. The calculation of the tilt angle change can be determined based on the relative position of the comprehensive vibration risk index between the first risk threshold and the second risk threshold, using a preset proportional mapping relationship.

[0185] The risk level proportional factor is obtained by subtracting the first risk threshold from the comprehensive vibration risk index R, and then dividing the result by the difference between the second and first risk thresholds. This factor ranges from 0 to 1. The risk level proportional factor is then multiplied by a preset maximum fine-tuning angle to obtain the current required tilt angle change. The maximum fine-tuning angle is a pre-calibrated first preset angle, which in this embodiment ranges from 10 to 15 degrees.

[0186] The command generation unit adds this tilt angle change to the current support tilt angle setting to obtain the target support tilt angle, and changes the tilt angle in the direction of reducing the windward side, that is, the support rotates towards the leveling direction. At the same time, the command generation unit limits the target support tilt angle to ensure that it is not lower than the minimum power generation tilt angle. That is, under medium risk conditions, the support is not adjusted to a completely level position to retain at least some power generation capacity.

[0187] The instruction generation unit generates an inclination fine-tuning instruction that includes the inclination angle of the aforementioned target support.

[0188] The status management unit marks the current control state as "medium damping-fine adjustment mode" and records the state switching timestamp and the comprehensive vibration risk index value at the time of switching.

[0189] When the comparison and judgment unit determines that the comprehensive vibration risk index R is greater than or equal to the second risk threshold, it confirms that the current system is in a high-risk state.

[0190] Under high-risk conditions, wind-induced vibration poses an extremely serious threat to structural safety. Without strong measures, it could potentially cause microcracks in components, plastic deformation of the support structure, or even complete collapse within a short period. In this situation, the control strategy prioritizes structural safety above all else, even at the expense of sacrificing all or most of the power generation revenue. The command generation unit generates control commands in the following manner: The instruction generation unit queries the damping level mapping table to obtain the highest damping level corresponding to the high-risk state. This highest damping level corresponds to the maximum excitation current or the maximum operating voltage of the adjustable damper, causing the adjustable damper to output the maximum damping force within its rated range, thereby dissipating wind-induced vibration energy to the maximum extent and quickly suppressing the component vibration amplitude within a safe range. The instruction generation unit generates a high-damping control instruction containing the excitation current or voltage value corresponding to this highest damping level.

[0191] The command generation unit simultaneously generates a safety posture command. The safety posture command instructs the support tilt adjustment mechanism to quickly adjust the support tilt angle to the minimum windward posture or the fully level posture.

[0192] Minimum windward orientation refers to the bracket tilt angle at which the component surface is essentially parallel to the prevailing wind direction. At this angle, the incoming airflow angle to the component surface approaches zero degrees, and the wind pressure acting on the component surface is reduced to the lowest level. The fully leveled orientation refers to the bracket tilt angle being adjusted to the minimum angle position allowed by the mechanical travel, with the component surface nearly horizontal. At this position, the windward projected area of ​​the component is minimized, and the wind load is also minimized.

[0193] The selection strategy for the minimum windward attitude and the leveling attitude is as follows: When the wind direction data provided by the multi-source sensor acquisition module is reliable and the tilt angle corresponding to the minimum windward attitude is within the allowable range of the support's mechanical travel, the minimum windward attitude is selected first to reduce wind load while preventing the component from being in an extreme leveling position; when the wind direction changes drastically, it is difficult to accurately track the minimum windward attitude in real time, or the minimum windward attitude exceeds the mechanical travel range, the leveling attitude is selected as the default safe position. The instruction generation unit generates a safe attitude instruction containing the target safe tilt angle.

[0194] In high-risk situations, the instruction generation unit can also optionally generate an alarm trigger signal and send a high-risk alarm notification to the power plant monitoring system or operation and maintenance personnel through the communication interface, prompting them to pay attention to the on-site situation or prepare for subsequent inspections.

[0195] The status management unit marks the current control state as "high damping-safe mode", records the status switching timestamp, the peak value of the comprehensive vibration risk index that triggers high risk, and the current wind speed and direction data for post-event analysis and structural health assessment.

[0196] To avoid frequent changes in control state when the comprehensive vibration risk index fluctuates around the threshold, the state management unit also introduces a hysteresis control mechanism when performing the multi-level threshold comparison.

[0197] Specifically, when the system is currently in a low-risk state and needs to switch to a medium-risk state, the aforementioned first risk threshold is used as the upward trigger threshold; while when the system is currently in a medium-risk state and needs to recover to a low-risk state, a downward trigger threshold lower than the first risk threshold is used, i.e., the first risk recovery threshold. The first risk recovery threshold is a preset percentage of the first risk threshold, for example, 70% to 80% of the first risk threshold.

[0198] Similarly, the threshold for triggering a switch from a medium-risk state to a high-risk state is the second risk threshold, while the threshold for triggering a switch from a high-risk state to a medium-risk state is the second risk recovery threshold, which is lower than the second risk threshold. The second risk recovery threshold is a preset proportion of the second risk threshold.

[0199] Through the aforementioned hysteresis control, the state management unit only performs state switching after the comprehensive vibration risk index has continuously changed and stably crossed the corresponding trigger threshold, preventing the state from oscillating frequently at the threshold boundary and ensuring the stability of control commands and the smooth operation of the actuator.

[0200] The command generation unit encapsulates the generated control commands according to a preset communication protocol and outputs them simultaneously or sequentially to the execution and closed-loop feedback module via the device's internal communication bus or interface. Specifically, low-damping, medium-damping, or high-damping control commands are sent to the drive circuit of the adjustable damper; tilt angle fine-tuning commands or safety posture commands are sent to the motor driver of the support tilt angle adjustment mechanism; and alarm signals are sent to the upper-level monitoring system.

[0201] The output timestamps of each command are uniformly marked to ensure that the damping adjustment action of the adjustable damper and the tilt angle adjustment action of the support are coordinated in time, so as to avoid additional dynamic loads caused by asynchronous actions.

[0202] The execution and closed-loop feedback module 206 is used to execute control commands and continuously acquire vibration acceleration data after execution, and feed it back to the vibration risk quantification module to dynamically correct the risk index and control strategy.

[0203] In some embodiments, this module is responsible for converting decision commands into physical actions and performing closed-loop corrections. The implementation steps are as follows: First, the control commands are converted into adjustable damper drive current (or voltage) signals and support tilt motor pulse / direction signals to drive the corresponding actuators. Next, the vibration acceleration data newly acquired by the multi-source sensor acquisition module after the action is continuously acquired. Then, the residual vibration amplitude after execution is compared with the preset safe amplitude. If the vibration amplitude has not yet dropped to a safe level, an incremental compensation command is issued to the damper driver to add an additional damping level on the basis of the original damping level and finely adjust the tilt angle until the vibration amplitude converges to the safe range. At the same time, the vibration acceleration data after execution is fed back to the vibration risk quantification module in real time, so that it recalculates the risk index, forming a continuous closed loop of "perception-prediction-decision-execution-feedback" to ensure that the device always operates in a risk-controllable state.

[0204] In some embodiments, the execution and closed-loop feedback module 206 can also be used for: According to the control command, the first adjustment signal is output to the drive circuit of the adjustable damper to change the damping coefficient of the adjustable damper. According to the control command, a second adjustment signal is output to the motor driver of the support tilt adjustment mechanism to change the support tilt angle; The residual vibration amplitude is determined based on the vibration acceleration data continuously acquired by the multi-source sensor acquisition module after the control command is executed. Based on the comparison between the residual vibration amplitude and the preset safe amplitude, when the residual vibration amplitude has not dropped to within the preset safe amplitude, a damping increment compensation command is generated, causing the adjustable damper to add one level of damping based on the current damping level.

[0205] In some embodiments, the device may further include a digital twin verification module for: A cloud-based virtual model is constructed based on the structural parameters, material parameters, and boundary conditions of the photovoltaic module-support system. Based on the wind pressure time history and component vibration response envelope output by the wind load and response prediction module, the cloud virtual model is driven to perform synchronous simulation to obtain the virtual vibration response; Based on the virtual vibration response, the control effect of the candidate control commands generated by the vibration reduction control decision module is pre-verified. Based on the pre-validation results, the optimal control command that minimizes the virtual vibration response amplitude and stabilizes the adjustment process is selected from multiple candidate control commands, and the optimal control command is sent to the execution and closed-loop feedback module.

[0206] The virtual model construction and synchronization subunit constructs a cloud-based virtual model that corresponds one-to-one with the physical photovoltaic module-support system on a cloud server or edge computing node.

[0207] The virtual model construction and synchronization subunit reads the structural parameters of the photovoltaic module-support system from a preset structural parameter database. These structural parameters include: the geometric dimensions, thickness, weight, and center of gravity of the photovoltaic module; the cross-sectional shape, dimensions, length, and node connection methods of the support columns, beams, and diagonal braces; the type, quantity, and spatial coordinates of the connectors between the module and the support; the installation position, travel range, and damping force-excitation current characteristic curve of the adjustable damper; and the transmission ratio, travel range, and response speed characteristics of the support tilt adjustment mechanism.

[0208] The virtual model construction and synchronization subunit reads the material parameters of each structural component from a preset material parameter database. These material parameters include: the elastic modulus, Poisson's ratio, density, yield strength, fatigue limit, and stress-life curve parameters of the component glass, frame aluminum alloy, and support steel or aluminum alloy materials.

[0209] The virtual model construction and synchronization subunit sets the boundary conditions of the cloud-based virtual model based on the actual installation status of the physical system. These boundary conditions include: the constraint type at the bottom of the support column, such as fixed constraint, hinged constraint, or elastic foundation constraint; the overlap or gap conditions between adjacent photovoltaic modules; and the rotational pair constraint relationship of the module tilt adjustment mechanism.

[0210] The virtual model construction and synchronization sub-units perform finite element mesh generation on the aforementioned geometric model. At key points, such as the edges of bolt holes connecting the component frame and support, the heat-affected zone of welds at the base of the support columns, and stress concentration areas like the adjustable damper mounting brackets, the mesh is locally refined. In large-area component panel areas far from key points, a coarser mesh is used to balance computational efficiency. After mesh generation, a complete finite element model is generated, including node coordinates, element topology, and material properties.

[0211] The virtual model construction and synchronization sub-unit performs modal analysis on the finite element model, calculating the first few natural frequencies and corresponding mode shapes of the system. The calculated natural frequencies are compared with the measured natural frequencies obtained through on-site hammer-impact modal testing. The connection stiffness and boundary constraint parameters in the model are adjusted to ensure that the relative error between the simulated and measured values ​​of the first few natural frequencies is less than a preset convergence threshold. This completes the model verification, ensuring that the cloud-based virtual model accurately reflects the dynamic characteristics of the physical system.

[0212] During device operation, the virtual model construction and synchronization subunit also periodically receives component tilt angle, key point strain and ambient temperature data from the multi-source sensor acquisition module of the physical system, dynamically updates the support tilt angle attitude and material temperature related parameters of the cloud virtual model, so that the virtual model and the physical system are synchronized in real time or near real time.

[0213] The virtual excitation loading subunit receives the wind pressure time history and component vibration response envelope output by the wind load and response prediction module, and transforms them into input excitation for the cloud virtual model.

[0214] The virtual excitation loading subunit decomposes the equivalent resultant wind pressure in the wind pressure time history, combined with the real-time coordinates of the resultant force application point provided by the multi-source sensor acquisition module, or using pre-calibrated typical resultant force application point locations, into normal pressure distributions acting on each finite element node or partition of the component surface. The decomposition method adopts spatial weighting consistent with the aerodynamic coefficient mapping table. When the resultant force application point changes, the normal pressure distribution of each node or partition is redistributed proportionally to maintain the resultant force magnitude and application point location unchanged.

[0215] In some embodiments, the vibration response envelope can be directly used as the amplitude constraint of the surface acceleration boundary condition of the component in the virtual model to drive the virtual model to perform transient response analysis; in other embodiments, the vibration response envelope is used to verify the upper or lower bound of the vibration response result output by the simulation.

[0216] The virtual response simulation subunit employs the direct integration method to perform transient dynamic simulations on the cloud-based virtual model over a predetermined future time period. The integration method used is either the Newmark-β method or the Wilson-θ method, with the time step consistent with the prediction time step output by the wind load and response prediction module.

[0217] At each simulation time step, the virtual response simulation sub-unit performs the following iterations: read the wind pressure load applied to each node at the current time step; calculate the node displacement increment at the current time step by solving the equivalent stiffness equation based on the node displacement, velocity, and acceleration of the previous time step and the load of the current step; update the node displacement, velocity, and acceleration; extract the equivalent stress, node vibration acceleration, and displacement envelope at key points and record them as the virtual vibration response at that time step.

[0218] After completing the simulation of all future prediction time steps, the virtual response simulation subunit outputs a virtual vibration response sequence, which includes at least the vibration acceleration response time history, vibration displacement response time history, and equivalent stress response time history at key points.

[0219] The control effect evaluation subunit receives multiple candidate control commands provided by the vibration reduction control decision module before generating the final control command, and performs pre-verification using a cloud-based virtual model.

[0220] In some embodiments, when the comprehensive vibration risk index crosses a certain threshold, the vibration reduction control decision module does not directly generate a single, definitive control command. Instead, it generates a set of candidate control commands for the digital twin verification module to select the best. The set of candidate control commands includes combinations of various damping levels and various tilt adjustment amounts. For example, under medium-risk conditions, candidate commands may include combinations such as: medium damping level with 10-degree tilt adjustment, medium damping level with 12-degree tilt adjustment, and medium-high damping level with 8-degree tilt adjustment; under high-risk conditions, candidate commands may include: highest damping level with a level posture, highest damping level with a minimum upwind posture, and highest damping level with a certain tilt transition posture.

[0221] The control effect evaluation subunit maps each candidate control command to corresponding damping parameters and tilt attitude parameters in the cloud-based virtual model. For damping parameters, based on the damping level in the candidate command, it queries the damping force-excitation current characteristic curve of the adjustable damper to obtain the damping force or equivalent damping coefficient output by the adjustable damper under that excitation, and updates the damping unit parameters corresponding to the adjustable damper in the virtual model with this equivalent damping coefficient. For tilt attitude parameters, based on the target tilt angle in the candidate command, it updates the current rotation angle of the rotating joint of the support tilt adjustment mechanism in the virtual model, and simultaneously updates the spatial coordinates of each node of the component and the direction and distribution of the wind pressure load.

[0222] After updating the virtual model parameters, the virtual response simulation subunit uses the same wind pressure time history excitation to re-execute the transient dynamics simulation, obtaining the virtual vibration response of the system under the candidate control command. The virtual vibration response also includes at least the vibration acceleration response time history, vibration displacement response time history, and equivalent stress response time history at key points.

[0223] The control effect evaluation subunit executes the above "parameter mapping-resimulation" process for each group of candidate control commands to obtain the virtual vibration response sequence corresponding to each candidate command, which serves as the basis for subsequent command selection.

[0224] The instruction selection subunit quantifies and evaluates the virtual vibration response of each candidate instruction output by the control effect evaluation subunit, and selects the optimal control instruction.

[0225] For each candidate command, the virtual vibration response of the subunit is selected and evaluated using at least the following metrics: The first indicator is the vibration response amplitude indicator, which is the maximum absolute value or root mean square value of the virtual vibration acceleration response time history within a preset future time period. The smaller this indicator is, the better the vibration reduction and suppression effect.

[0226] The second indicator is the stability index of the adjustment process. It is the ratio of the root mean square value of the virtual vibration acceleration response time history in the second half of the preset calculation window to the root mean square value in the first half of the window. The closer the ratio is to 1 and the less than 1, the more stable the response tends to be. If the ratio oscillates violently or is much greater than 1, it indicates that the adjustment process has caused additional transient disturbances.

[0227] The third indicator is the stress safety margin indicator, which takes the maximum value in the virtual equivalent stress response time history and compares it with the allowable stress of the material to calculate the minimum safety margin. The larger the margin, the safer the structure.

[0228] In some embodiments, a fourth indicator may be included as a power generation loss indicator, which estimates the power generation loss caused by tilt adjustment based on the deviation between the target tilt angle and the optimal power generation tilt angle, as well as the expected duration of the high-risk state.

[0229] The instruction selection subunit selects the optimal control instruction from all candidate instructions according to the preset multi-objective evaluation rules.

[0230] One comprehensive selection rule is as follows: First, candidate commands that do not meet the hard constraint of stress safety margin are excluded, that is, candidate commands whose equivalent stress at any time step in the virtual equivalent stress response time history exceeds the preset safety upper limit are directly eliminated; among the candidate commands that meet the safety constraint, several candidate commands with the smallest vibration response amplitude index are selected first; when the vibration response amplitude indexes of multiple candidate commands are similar, the candidate command with the optimal stability index of the regulation process is selected; if there are still multiple similar candidate commands, the one with the lowest power generation loss index is further considered as the optimal control command.

[0231] Another comprehensive selection rule is to construct a weighted evaluation function, normalize the above indicators, and sum them by weight. The control command with the highest total score is selected as the optimal control command. The weight coefficients of each indicator are dynamically adjusted according to the risk level of the system: under high-risk conditions, the weights of vibration response amplitude and stress safety margin indicators are significantly increased; under medium-risk conditions, the weights of regulation process stability and power generation loss indicators are moderately increased.

[0232] The optimal control command selection subunit encapsulates the selected command according to a preset communication protocol and sends it to the execution and closed-loop feedback module via the communication link between the device and the cloud. Upon receiving the optimal control command, the execution and closed-loop feedback module drives the adjustable damper and the support tilt adjustment mechanism to perform the corresponding actions.

[0233] Meanwhile, the instruction selection and issuance subunit records the evaluation results of each candidate instruction, the selected optimal instruction parameters, and the corresponding virtual vibration response in the digital twin pre-verification process into the historical database for subsequent offline analysis, control strategy iterative optimization, and further refinement of the digital twin model.

[0234] Through the aforementioned digital twin verification module, the device of the present invention can simulate the effects of multiple candidate control strategies in parallel in the cloud without actually changing the state of the physical system. After quantitative evaluation, only the optimal control command is issued and executed, thereby avoiding structural impact, power generation loss and equipment wear caused by improper or suboptimal control actions on the physical system, and significantly improving the reliability and intelligence of control decisions.

[0235] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0236] Those skilled in the art will understand that embodiments of the present invention can be provided as apparatus, methods, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0237] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0238] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0239] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0240] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0241] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A solar photovoltaic power generation assembly complete device with wind resistance damping function, characterized in that, The device includes: The multi-source sensor acquisition module is used to acquire raw data of the area where the photovoltaic module is located in real time. The raw data includes wind speed, wind direction, turbulence intensity, vibration acceleration of module surface, module tilt angle and strain data. The data preprocessing and feature extraction module is used to perform noise reduction, time synchronization and feature extraction on the raw data output by the multi-source sensing acquisition module to obtain a standardized real-time feature vector. The wind load and response prediction module has a built-in time series prediction model based on physical information enhancement, which is used to predict the wind pressure time history and component vibration response envelope within a preset period of time based on the historical feature vector sequence and the real-time feature vector. The vibration risk quantification module is used to calculate the comprehensive vibration risk index in real time based on the predicted wind pressure time history and component vibration response envelope within the preset future time period, combined with the material cumulative damage model. The vibration reduction control decision module is used to generate control commands for the adjustable damper and / or the support tilt adjustment mechanism based on the comparison results between the comprehensive vibration risk index and the preset multi-level risk threshold. The execution and closed-loop feedback module is used to execute the control commands and continuously acquire the vibration acceleration data after execution, and feed it back to the vibration risk quantification module to dynamically correct the risk index and control strategy.

2. The solar photovoltaic power generation assembly complete device with wind resistance and shock absorption functions according to claim 1, characterized in that, The multi-source sensing acquisition module is also used for: Based on the wind speed and wind direction, and combined with the current tilt angle of the photovoltaic module, determine the incoming flow angle and relative wind speed acting on the module surface; Based on the incoming angle of attack and the relative wind speed, and combined with the air density and aerodynamic coefficient mapping table, the instantaneous wind pressure distribution of each zone on the component surface is determined; Based on the instantaneous wind pressure distribution of each zone on the surface of the component, the equivalent wind pressure resultant force application point and wind pressure resultant force time history of the component are calculated, and the wind pressure resultant force time history is output as part of the original data. 3.The solar photovoltaic power generation assembly complete device with wind resistance and shock absorption functions according to claim 1, characterized in that, The data preprocessing and feature extraction module is also used for: Based on the signal characteristics of each sensing channel in the original data, the corresponding wavelet basis function and decomposition level are selected, and wavelet threshold denoising is performed on each channel signal to obtain the denoised signal. Based on the unified timestamp carried on each channel signal, the denoised signal is time-aligned and resampled to obtain a synchronous sampling sequence; Based on the synchronous sampling sequence, extract time-domain features and frequency-domain features; The normalized real-time feature vector is generated by combining the time-domain features and the frequency-domain features.

4. The solar photovoltaic power generation assembly complete device with wind resistance and shock absorption functions according to claim 3, characterized in that, The data preprocessing and feature extraction module is also used for: Based on the vibration acceleration signal in the synchronous sampling sequence, calculate the root mean square value and peak factor of the vibration acceleration, as at least a part of the time-domain features; Based on the vibration acceleration signal in the synchronous sampling sequence, the power spectral density is obtained by fast Fourier transform, and the dominant frequency and spectral kurtosis are extracted from the power spectral density as at least a part of the frequency domain features. 5.The solar photovoltaic power generation assembly complete device with wind resistance and shock absorption functions according to claim 1, wherein, The wind load and response prediction module is also used for: Based on the historical feature vector sequence, an input sequence is constructed, the input sequence containing the real-time feature vectors of the current time and multiple consecutive time steps prior to it; Based on the input sequence, the equivalent wind pressure prediction value for each time step within a preset time period is predicted through a physical information-enhanced long short-term memory network, thus forming the wind pressure time history; Based on the input sequence and the wind pressure time history, the component vibration acceleration envelope value at each time step within a preset future period is synchronously predicted through the physical information-enhanced long short-term memory network, thus forming the component vibration response envelope.

6. The solar photovoltaic power generation module supporting device with wind resistance and shock absorption function according to claim 5, characterized in that, The physically enhanced long short-term memory network in the wind load and response prediction module includes the following during training: Based on the predicted wind pressure time history and the component vibration response envelope, and combined with the structural dynamics equations of the photovoltaic module-support system, the physical residuals are calculated. Based on the physical residuals and data prediction errors, a joint loss function is constructed to enable the network prediction results to simultaneously fit the measured data and the physical laws of structural dynamics.

7. The solar photovoltaic power generation module supporting device with wind resistance and shock absorption function according to claim 1, characterized in that, The vibration risk quantification module is also used for: Based on the predicted wind pressure time history, determine the wind pressure loading sequence that represents the equivalent wind pressure over time. Based on the wind pressure loading sequence and the component vibration response envelope, determine the equivalent stress amplitude sequence at key points of the component; Based on the equivalent stress amplitude sequence and the allowable fatigue stress of the component material, the stress over-limit penalty term is determined in real time. Based on the equivalent wind pressure at each past moment in the wind pressure loading sequence, the acceleration amplitude at the corresponding moment in the component vibration response envelope, and the system damping attenuation term, the cumulative damage term of wind-induced vibration is determined by integral summation. The comprehensive vibration risk index is obtained by weighted summation of the cumulative damage term caused by wind-induced vibration and the stress over-limit penalty term.

8. The solar photovoltaic power generation module supporting device with wind resistance and shock absorption function according to claim 1, characterized in that, The vibration reduction control decision module is also used for: Based on the comparison result between the comprehensive vibration risk index and the first risk threshold, and when the comprehensive vibration risk index is less than the first risk threshold, a low-damping control command is generated to make the adjustable damper at the preset lowest damping level and keep the support at the power generation tilt angle. Based on the comparison results of the comprehensive vibration risk index with the first risk threshold and the second risk threshold, and when the comprehensive vibration risk index is greater than or equal to the first risk threshold and less than the second risk threshold, a medium damping control command and an inclination fine-tuning command are generated to adjust the adjustable damper to a preset medium damping level and to rotate the bracket inclination angle by a first preset angle in the direction of reducing the windward side. Based on the comparison result between the comprehensive vibration risk index and the second risk threshold, and when the comprehensive vibration risk index is greater than or equal to the second risk threshold, a high damping control command and a safe posture command are generated, so that the adjustable damper is adjusted to the preset highest damping level, and the bracket tilt angle is adjusted to the minimum windward posture or the flat posture.

9. The solar photovoltaic power generation module supporting device with wind resistance and shock absorption function according to claim 1, characterized in that, The execution and closed-loop feedback module is also used for: According to the control command, a first adjustment signal is output to the drive circuit of the adjustable damper to change the damping coefficient of the adjustable damper; According to the control command, a second adjustment signal is output to the motor driver of the support tilt adjustment mechanism to change the support tilt angle; The residual vibration amplitude is determined based on the vibration acceleration data continuously acquired by the multi-source sensor acquisition module after the control command is executed. Based on the comparison between the residual vibration amplitude and the preset safety amplitude, when the residual vibration amplitude does not drop below the preset safety amplitude, a damping increment compensation command is generated, causing the adjustable damper to increase the damping by one level based on the current damping level.

10. The solar photovoltaic power generation module supporting device with wind resistance and shock absorption function according to any one of claims 1 to 9, characterized in that, It also includes a digital twin verification module for: A cloud-based virtual model is constructed based on the structural parameters, material parameters, and boundary conditions of the photovoltaic module-support system. Based on the wind pressure time history and the component vibration response envelope output by the wind load and response prediction module, the cloud virtual model is driven to perform synchronous simulation to obtain the virtual vibration response; Based on the virtual vibration response, the control effect of the candidate control commands generated by the vibration reduction control decision module is pre-verified. Based on the pre-verification results, the optimal control command that minimizes the virtual vibration response amplitude and stabilizes the adjustment process is selected from multiple candidate control commands, and the optimal control command is sent to the execution and closed-loop feedback module.

Images