Photovoltaic flexible support structure health assessment method and system

Abstract

The invention belongs to the technical field of structure health monitoring and safety assessment, and discloses a photovoltaic flexible support structure health assessment method and system. Firstly, acceleration signals, inclination angle data and cable force data of the photovoltaic flexible support are collected; extracting inherent frequencies, vibration modes and modal damping ratios of the first three orders of the structure based on a random subspace recognition method, analyzing inclination angle data to obtain static displacement or dynamic rotation angles of key nodes, obtaining a cable force distribution state by combining cable force data, and integrating to form a dynamic feature vector; inputting the vector into a pre-trained structure health assessment model, and outputting a damage position and degree; and when the damage degree exceeds a preset threshold value, a report containing the position of the damaged component, the damage degree and the disposal suggestion is automatically generated. Through cooperation of multi-source data fusion and an intelligent algorithm, real-time, accurate and automatic evaluation of structural health and accurate damage positioning are realized, a scientific basis is provided for preventive maintenance, safe operation of a photovoltaic power station is effectively guaranteed, and the operation and maintenance cost is reduced.

Description

Technical Field

[0001] This invention belongs to the field of structural health monitoring and safety assessment technology, and relates to a method and system for assessing the structural health of photovoltaic flexible support structures. Background Technology

[0002] Flexible photovoltaic (PV) supports are widely used in large-scale PV power plants in complex terrains due to their advantages such as large span, strong adaptability to terrain, low steel consumption, and good economic efficiency. These structures typically consist of prestressed cables, supporting structures, and connectors, and belong to a highly flexible structural system, making them sensitive to dynamic loads such as wind and snow loads. During long-term service, due to material degradation, prestress relaxation, joint loosening, and the effects of extreme load events, the structure may experience cumulative damage, thereby affecting its safety and durability.

[0003] Currently, the structural condition assessment of photovoltaic flexible supports relies heavily on periodic manual inspections or simple cable stress checks. Manual inspections are highly subjective and difficult to detect internal structural damage and performance degradation; while simple cable stress monitoring cannot comprehensively reflect the overall mechanical state of the structure, especially in terms of damage location and remaining life prediction capabilities.

[0004] Therefore, there is an urgent need to develop a method that can assess the health status of photovoltaic flexible support structures in real time, accurately and automatically, in order to ensure the safe operation of power plants, reduce operation and maintenance costs, and provide a scientific basis for preventive maintenance. Summary of the Invention

[0005] The purpose of this invention is to solve the problems in the prior art and provide a method and system for health assessment of photovoltaic flexible support structures.

[0006] To achieve the above objectives, the present invention employs the following technical solution: The present invention proposes a method for health assessment of photovoltaic flexible support structures, comprising the following steps: Acquire acceleration signals, tilt angle data, and cable force data of the photovoltaic flexible support structure; Based on the random subspace identification method, the first three natural frequencies and mode shapes of the photovoltaic flexible support structure are extracted; the modal damping ratio of the photovoltaic flexible support structure is calculated; the tilt angle data is analyzed to obtain the static displacement or dynamic rotation angle of the key nodes; the cable force distribution state is obtained based on the cable force data; and the above parameters are used as dynamic feature vectors characterizing the health status of the photovoltaic flexible support structure. The dynamic feature vector is input into a pre-trained structural health assessment model, which outputs the location and extent of damage to the photovoltaic flexible support structure. When the extent of damage exceeds a preset threshold, a report containing the location of the damaged component, the extent of damage, and treatment recommendations is generated, thus realizing the health assessment of the photovoltaic flexible support structure.

[0007] This invention proposes a health assessment system for photovoltaic flexible support structures, comprising: The data acquisition module is used to acquire acceleration signals, tilt angle data, and cable force data of the photovoltaic flexible support structure. The data processing module is used to extract the first three natural frequencies and mode shapes of the photovoltaic flexible support structure based on the random subspace identification method; calculate the modal damping ratio of the photovoltaic flexible support structure; analyze the tilt angle data to obtain the static displacement or dynamic rotation angle of key nodes; obtain the cable force distribution state based on the cable force data; and use the above parameters as dynamic feature vectors characterizing the health state of the photovoltaic flexible support structure. The structural health assessment module is used to input dynamic feature vectors into a pre-trained structural health assessment model and output the damage location and damage degree of the photovoltaic flexible support structure. When the damage degree exceeds a preset threshold, a report containing the location of the damaged component, the damage degree, and treatment suggestions is generated to realize the structural health assessment of the photovoltaic flexible support structure.

[0008] Compared with the prior art, the present invention has the following beneficial effects: This invention proposes a health assessment method for photovoltaic flexible support structures. Firstly, addressing the limitations of traditional methods that only monitor cable forces and lack comprehensive data, it simultaneously collects acceleration signals, tilt angle data, and cable force data. This covers three core data categories: dynamic response (acceleration), static attitude (tilt angle), and key forces (cable forces), enabling a holistic understanding of the structural state and providing a data foundation for comprehensive assessment, avoiding assessment biases caused by single data points. Secondly, addressing the difficulty in capturing internal damage and inaccurate feature extraction, it employs a random subspace identification method to extract the first three natural frequencies, mode shapes, and modal damping ratios. Combined with tilt angle data analysis of key node displacement / rotation and cable force distribution, these parameters are integrated into a dynamic feature vector. This method can accurately capture changes in mechanical characteristics caused by internal damage such as material degradation and node loosening, compensating for the limitations of manual inspection in perceiving internal conditions. To address the issues of insufficient real-time automated assessment, damage location, and treatment capabilities, a dynamic feature vector is input into a pre-trained health assessment model to intelligently output the location and extent of damage without human intervention, achieving real-time automated assessment. Simultaneously, a damage threshold is set, and when it is exceeded, a report containing the location, extent, and treatment recommendations is automatically generated. This solves the pain points of traditional methods, such as vague damage location and inability to support preventive maintenance. Ultimately, it enables accurate assessment and early warning of the health of photovoltaic flexible support structures, ensuring the safe operation of power plants. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart of the photovoltaic flexible support structure health assessment method of the present invention.

[0011] Figure 2 This is a detailed flowchart of the photovoltaic flexible support structure health assessment method of the present invention.

[0012] Figure 3 This is a diagram of the photovoltaic flexible support structure health assessment system of the present invention. Detailed Implementation

[0013] The present invention will now be described in further detail with reference to the accompanying drawings: Example 1 The photovoltaic flexible support structure health assessment method of this invention collects various response data of the structure in real time through a deployed sensor network. The data is then processed on the server side, and a random subspace identification method based on singular value decomposition (SVD) is used to extract dynamic characteristic parameters such as the structure's natural frequency, mode shape, and damping ratio from the acceleration signal, forming feature vectors representing the structure's current state. These feature vectors are input into a pre-trained neural network model for intelligent diagnosis and assessment, thereby identifying damage and triggering early warnings. Finally, all data and assessment results are stored in a database, and predictive maintenance is achieved through long-term trend analysis. In this entire process, SVD serves as the core algorithm for accurately extracting modal features from noisy data, while the neural network, as the assessment model, uses these features for intelligent judgment. The two are interconnected, forming an intelligent analysis chain from data to decision.

[0014] This invention proposes a method for health assessment of photovoltaic flexible support structures, such as... Figure 1 As shown, it includes the following steps: Step 1: Obtain the acceleration signal, tilt angle data, and cable force data of the photovoltaic flexible support structure; Step 2: Based on the random subspace identification method, extract the first three natural frequencies and mode shapes of the photovoltaic flexible support structure; calculate the modal damping ratio of the photovoltaic flexible support structure; analyze the tilt angle data to obtain the static displacement or dynamic rotation angle of the key nodes; obtain the cable force distribution state based on the cable force data; use the above parameters as dynamic feature vectors characterizing the health state of the photovoltaic flexible support structure. The random subspace identification method extracts the first three natural frequencies and mode shapes of the photovoltaic flexible support structure, specifically as follows: The preprocessed multi-channel acceleration signals are organized into a Hankel matrix, the Hankel matrix is ​​projected, and then singular value decomposition is performed. The magnitude of the singular values ​​obtained after decomposition directly corresponds to the order of the structure. Determine the system order n, and select the system order based on the magnitude of the singular values; take the first n singular values ​​and their corresponding singular vectors to construct an extended observable matrix and state sequence, establish a discrete state-space model, and solve the system matrix A and output matrix C of the discrete state-space model by least squares estimation; Eigenvalue decomposition is performed on the state matrix A to convert the discrete-time eigenvalues ​​into continuous-time eigenvalues, and the natural frequencies of each mode are calculated. f k The output matrix C is combined with the eigenvectors of the state matrix A to obtain the mode shape of the structure. F k ; Calculate the damping ratio based on continuous-time eigenvalues x k ; Repeat the above for a series of system orders, and for each order, calculate a set of modal parameters ( f k , x k , F k Plot the stability diagram in the frequency order coordinate system, and label each set of modal parameters as modal points on the stability diagram; By using frequency stability criteria, mode shape correlation criteria, and damping ratio stability criteria, the modal points that form vertical stable rods at different orders in the stability diagram are identified, and the first three natural frequencies and corresponding mode shapes of the photovoltaic flexible support structure are determined.

[0015] The process involves projecting the Hankel matrix and then performing singular value decomposition. The magnitude of the singular values ​​obtained after decomposition directly corresponds to the order of the structure. Specifically: Projecting the Hankel matrix:

[0016] For projection matrix O i Perform SVD decomposition:

[0017] in, O i Represents the projection matrix. + Indicates Moore-Penrose pseudoinverse; Hp H indicates the past part. f Indicates the future part; U Σ is a left singular vector matrix; Σ is a singular value diagonal matrix: Σ = diag(σ1, σ2, ..., σ 2i · l ), and σ1≥ σ2≥ ... ≥ 0; V It is a right singular vector matrix.

[0018] The process involves taking the first n singular values ​​and their corresponding singular vectors to construct an extended observable matrix and state sequence, establishing a discrete state-space model, and then solving for the system matrix A and output matrix C of the discrete state-space model using least squares estimation. Specifically: Take the first n singular values ​​and their corresponding singular vectors: ; The extended observable matrix is: ; The state sequence is as follows: ; Establish a discrete state-space model: ; Solve for the system matrix A and output matrix C using least squares estimation:

[0019] in, and It is a delayed version of the state sequence. Y i | i It is a specific block of the output matrix; It is the matrix of the first n left singular vectors. Singular value diagonal matrix; It is the matrix of the first n right singular vectors. It is the first k The system state vector at time t. It is the first k The system state vector at time +1 It is the first k The observed output vector at time t. It is observation noise. It is system noise.

[0020] The state matrix A is decomposed into eigenvalues, converting the discrete-time eigenvalues ​​into continuous-time eigenvalues, and the natural frequencies of each mode are calculated. The output matrix C is combined with the eigenvectors of the state matrix A to obtain the mode shapes of the structure. F k ; Calculate the damping ratio based on continuous-time eigenvalues x k Specifically: Perform eigenvalue decomposition on the state matrix A: ; The obtained discrete-time eigenvalues ​​are converted into continuous-time eigenvalues, and then the natural frequency of each mode is calculated: No. k The first natural frequency is: ;in, , ; The output matrix C is combined with the eigenvectors of the state matrix A to obtain the mode shapes of the structure. The mode shapes are then normalized and stored as vectors. The mode shape vectors are obtained from the output matrix C and the eigenvectors Ψ. ; Damping ratio calculated based on continuous-time eigenvalues ; Where Λ is the eigenvalue diagonal matrix; Ψ is the eigenvector matrix; and |·| denotes taking the modulus of a complex number. ψ k It is the first k There are eigenvectors, and Re(·) represents taking the real part.

[0021] The frequency stability criterion, mode shape correlation criterion, and damping ratio stability criterion are as follows: Frequency stability: | f n - f n+Δn | / f n < e f ; Mode shape correlation: MAC( F n , F n+Δn )> e MAC ; Damping ratio stability: | x n - x n+Δn | / ξ n < e ξ ; in, , f n These are the modal frequencies calculated for system order n. f n+Δn For the same modal frequency calculated at system order n+Δn, F nThe modal shape vectors are calculated when the system order is n. F n+Δn This is the corresponding mode shape vector calculated when the system order is n+Δn. e MAC This is the modal correlation threshold. e f This is the frequency stability tolerance threshold. e ξ This is the damping ratio stability tolerance threshold. x n The modal damping ratio is calculated for system order n. x n+Δn This is the corresponding damping ratio calculated when the system order is n+Δn. Let i be the mode shape vector of the i-th mode. Let be the mode shape vector of the j-th mode.

[0022] Step 3: Input the dynamic feature vector into the pre-trained structural health assessment model to output the damage location and damage level of the photovoltaic flexible support structure; when the damage level exceeds the preset threshold, generate a report containing the location of the damaged component, the damage level and treatment suggestions to realize the health assessment of the photovoltaic flexible support structure.

[0023] The trained structural health assessment model is specifically as follows: A reference finite element model of a photovoltaic flexible support is established, and the reference finite element model is corrected by modal parameters measured by on-site environmental excitation tests. Multiple working conditions and damage states are simulated on the modified baseline finite element model to generate sample data containing input load, dynamic feature vector, and damage state labels. A deep neural network was used to train the sample data to obtain a trained structural health assessment model.

[0024] When the damage exceeds a preset threshold, a report is generated that includes the location of the damaged component, the extent of the damage, and recommended treatment methods. Standardize the real-time feature vectors: ; Damage prediction: ; Health rating: ; when Health Score ≥90, all i If the value is less than 0.1, then it is normal. When 80≤ Health Score <90, or any iIf ∈ [0.1, 0.2), then pay attention; When 70≤ Health Score <80, or any i If the range is ∈ [0.2, 0.3), then an early warning is issued; when Health Score <70, or any i If the value is ≥0.3, an alarm will be triggered; in, These are the predicted damage values ​​for critical components. These are the mean and standard deviation of the training set, respectively. i Let be the predicted damage value for the i-th component; When the warning or alarm level is reached, the system automatically sends SMS / email to the relevant maintenance personnel, pops up an alert window on the monitoring platform, records the event log, and initiates the emergency response process.

[0025] like Figure 2 The method is described in detail below: S1, System Deployment and Data Acquisition: A triaxial accelerometer and a biaxial tilt sensor are deployed at the top of the support rod to monitor the overall vibration and tilt of the structure. Cable force sensors are installed on the load-bearing cables. Strain gauges are deployed in the mid-span area of ​​the cable net and at key connection nodes. An anemometer and wind pressure sensor are also installed at the top of the mast to monitor environmental wind loads. All sensors form a monitoring network via IoT nodes, transmitting the collected response data, including acceleration, tilt, strain, cable force, and wind speed, to the central processing server in real time.

[0026] S2, Dynamic Feature Parameter Extraction: After receiving data, the server performs preprocessing. A random subspace identification method is used to perform modal analysis on the acceleration signal, extracting the structure's current first three natural frequencies and mode shapes. Tilt angle data is analyzed to obtain the static displacement or dynamic rotation angle of key nodes. Readings from all cable force sensors are integrated to obtain the cable force distribution. The modal damping ratio of the structure is calculated. These parameters are used as dynamic feature vectors characterizing the current structural health state.

[0027] Modal parameter extraction based on random subspace identification (SSI) is a key step in obtaining the global dynamic fingerprint of a structure from acceleration signals.

[0028] S2.1, Constructing the Hankel Data Matrix: The preprocessed multi-channel acceleration signals are organized into a large Hankel matrix. The matrix transforms the time series data into a specific matrix form that can reveal the internal state of the system.

[0029] Assume there is lThere are 1 acceleration sensor channel, and each channel collects data. N There are 10 data points. The sampling time interval is 1000. Δt The preprocessed multi-channel acceleration data y k ∈R l (k = 1, 2, ..., N) are organized into a block Hankel matrix:

[0030] in: i These are user-defined parameters (usually 2-3 times the maximum order of the model). j =N-2 i +1 is the column number of the matrix; the matrix is ​​divided into two parts: the past part H. p (forward i (Line) and future part H f (back i (rows); the dimension of the matrix is ​​2. i · l ×j.

[0031] S2.2, Projection and Singular Value Decomposition (SVD): The Hankel matrix is ​​projected, followed by singular value decomposition (SVD). The magnitude of the singular values ​​obtained after decomposition directly corresponds to the order of the system (i.e., the structure), with larger singular values ​​representing the dominant system modes.

[0032] The projection operation, orthogonal projection of future outputs to past outputs, is the core of SSI.

[0033] in + This indicates the Moore-Penrose pseudo-inverse.

[0034] Singular Value Decomposition (SVD): For projection matrices... O i Perform SVD decomposition:

[0035] in: U Σ is a left singular vector matrix (orthogonal matrix); Σ is a singular value diagonal matrix: Σ = diag(σ1, σ2,..., σ...). 2i · l ), and σ1≥ σ2≥ ... ≥ 0; V It is a right singular vector matrix (orthogonal matrix), σ 2i · l It is the 2i*lth singular value.

[0036] S2.3, Establish the state-space model: Use the results of SVD to identify the state-space model of the system.

[0037] Determine the system order n, and choose the system order based on the magnitude of the singular values. Ideally, the first n singular values ​​are relatively large, and the subsequent singular values ​​drop sharply to near zero.

[0038] Extracting the state sequence: Take the first n singular values ​​and their corresponding singular vectors:

[0039] The extended observable matrix is:

[0040] The state sequence is estimated as follows:

[0041] State-space model: Establish a discrete state-space model:

[0042] Solve for the system matrix A and output matrix C using least squares estimation:

[0043] in and It is a delayed version of the state sequence. Y i | i It is a specific block of the output matrix.

[0044] S2.4, Solve for the modal parameters: Eigenvalue decomposition: Perform eigenvalue decomposition on the state matrix A:

[0045] Where Λ is the eigenvalue diagonal matrix; Ψ is the eigenvector matrix.

[0046] Calculate the natural frequencies: Perform eigenvalue decomposition on the state matrix A. Convert the obtained discrete-time eigenvalues ​​into continuous-time eigenvalues, and then calculate the natural frequencies of each mode.

[0047] Discrete-time eigenvalues l k With continuous time eigenvalues m k The relationship is:

[0048] get:

[0049] No. k The first natural frequency is:

[0050] Where |·| represents taking the modulus of a complex number.

[0051] Calculating mode shapes: Combining the output matrix C with the eigenvectors of the state matrix A yields the mode shapes of the structure, i.e., the forms of vibration of the structure at various frequencies. The mode shapes are then normalized and stored as vectors. The mode shape vectors are directly obtained from the output matrix C and the eigenvectors Ψ.

[0052] in ψ k It is the first k 1 eigenvector.

[0053] The relationship between the real and imaginary parts of the eigenvalues ​​and the damping ratio is as follows:

[0054] Where Re(·) denotes taking the real part.

[0055] S2.5, Stability graph analysis to identify true modes: used to distinguish between true physical modes and computational noise.

[0056] Generating a stable graph: for a series of systems of order n=n min , n min +Δn, ..., n max Repeat steps S2.2-S2.4 to calculate a set of modal parameters for each order n. f k , x k , F k ).

[0057] Plotting a stability diagram: In a frequency-order coordinate system, each calculated mode is represented by a point, and the shape or color of the point can indicate the stability of that mode.

[0058] Stability criterion: Criteria for automatically or manually identifying stable modes typically include: Frequency stability: | f n - f n+Δn | / f n < ef (generally e f =0.01-0.05) Mode shape correlation: MAC( F n , F n+Δn )> e MAC (generally e MAC = 0.8-0.9) Damping ratio stability: | x n - x n+Δn | / ξ n < e ξ (generally e ξ = 0.1-0.2) The formula for calculating MAC (Modal Assurance Criterion) is as follows:

[0059] True mode identification: In the stability diagram, the true physical modes appear as vertical "stabilizing bars" at different orders, while computational noise or spurious modes are randomly distributed. By identifying the stabilizing bars, the true first three natural frequencies and corresponding mode shapes of the structure can be determined.

[0060] The calculated frequencies and mode shapes at different orders are plotted as stability diagrams. Frequency points that remain stable at different orders, forming vertical stabilizers, are considered the true physical modes of the structure. Randomly occurring, unstable points are considered computational noise or spurious modes. Operators or automated identification algorithms analyze the stability diagrams to ultimately determine the structure's true first three natural frequencies and corresponding mode shapes.

[0061] S3, Health Status Assessment and Early Warning: The dynamic feature vector obtained in step S2 is input into the pre-trained structural health assessment model. The model construction process is as follows: S3.1 Establish a benchmark model. Based on the design drawings of the photovoltaic flexible support, establish its finite element model, and correct the model through modal parameters measured by on-site environmental excitation tests to ensure that the model is highly consistent with the actual structure.

[0062] Initial finite element modeling Geometric Modeling: Based on the design drawings, a precise 3D geometric model is created in ANSYS / ABAQUS, including all structural components such as the mast, load-bearing cables, stabilizing cables, and connection nodes. Material Property Definition, Element Type Selection, and Boundary Condition Constraints: Constraints are defined according to the foundation design, including fixed connections at the mast base and hinged connections at the cable ends.

[0063] Model Revision and Validation Field testing: Obtain the modal parameters of the actual structure, including fundamental frequency, mode shape and damping ratio, through environmental excitation testing.

[0064] Objective function construction

[0065] Where θ is the parameter vector to be corrected, w is the weighting coefficient, and MAC is the modal guarantee criterion. The natural frequency (Hz) is obtained from finite element analysis. The actual natural frequency (Hz) obtained from field testing. As a modal guarantee criterion, it measures the correlation between two vibration modes. is the stiffness parameter (N / m) in the finite element model. The design stiffness parameters (N / m) are given.

[0066] Parameter correction: The objective function J(θ) is minimized using an optimization algorithm, and key parameters are corrected accordingly. Boundary spring stiffness: k boundary Initial pretension of the cable: T initial Node connection stiffness: k joint Material damping coefficient: ξ material .

[0067] Verification Standards Frequency error: |f FEM -f test | / f test <3% MAC value: MAC(Φ) FEM ,Φ test )>0.9 Static deformation matching: Under the same load, the error between calculated deformation and measured deformation is <5%. S3.2 generates sample data and simulates various working conditions on the corrected baseline model, including wind loads with different wind speeds and directions; it also simulates various damage states, such as the loss of cable force in a main cable, the reduction of stiffness at the mast bottom nodes, and support settlement. Through extensive simulation calculations, several sets of sample data are generated, each labeled with "Input Load - Structural Response (converted to characteristic parameters) - Damage State".

[0068] Load case simulation Wind load simulation: Davenport spectrum is used to simulate random wind fields.

[0069]

[0070] in, It is the wind power spectral density (m² / s), representing the distribution of wind energy at different frequencies. It is surface roughness. U 10 It is the wind speed at a height of 10m. f g The characteristic frequency is denoted as .

[0071] Temperature load: Simulates the thermal stress caused by temperature changes from -20°C to 50°C.

[0072] Definition of loss condition: Cable stress loss:

[0073] Stiffness reduction:

[0074] Support settlement:

[0075] Sample data generation Feature parameter extraction: Perform transient dynamic analysis for each operating condition to extract feature vectors.

[0076] Damage label definition

[0077] in, D overall ∈[0,1] represents the overall stiffness reduction factor. D cablei ∈[0,1] is the first i The cable force loss rate of the root cable, D masti ∈[0,1] is the first i mast stiffness reduction rate, D jointi For the first i The stiffness reduction rate of the node.

[0078] Generate several sets of sample data, construct a dataset, and divide the dataset into a training set, a validation set, and a test set.

[0079] S3.3, Training and Evaluation Model: A deep neural network was selected as the machine learning algorithm. The sample data was divided into a training set and a test set. The neural network was trained using the training set, adjusting its weights and biases to enable it to accurately predict the location and extent of structural damage, as well as other health status indicators, based on the input feature parameters. The test set was used to verify the model's accuracy and generalization ability.

[0080] Deep Neural Network Design Network Architecture: Input layer: n_input nodes Hidden layers: 3-5 fully connected layers, each with 256-512 nodes, using the ReLU activation function. Output layer: n_output nodes, using the Sigmoid activation function.

[0081] The loss function uses a weighted mean squared error loss:

[0082] in, For the first j The weights of each output component are assigned, with higher weights given to key components. For the first i The first sample j The true value of each damage label, The total number of training samples, For the first i The first sample j Predicted values ​​for each damage label.

[0083] Optimizer: The Adam optimizer is used with an initial learning rate of lr=0.001 and a learning rate decay strategy.

[0084] Model training stops when the validation set loss no longer decreases for 10 consecutive epochs.

[0085] The model evaluation metrics include mean absolute error, coefficient of determination, accuracy, and F1 score.

[0086] S3.4 Real-time assessment and early warning: When the damage index output by the model exceeds the threshold, the system automatically triggers an alarm mechanism, notifying maintenance personnel via SMS, email, etc. The report generation module automatically generates an inspection report based on the model output, including the location of the damaged component, the degree of damage, possible causes, and recommended remedial measures.

[0087] Online evaluation, feature vector preprocessing, and the same standardization process used on the real-time feature vectors as on the training set.

[0088]

[0089] in, These are the mean and standard deviation of the training set, respectively.

[0090] Damage prediction:

[0091] The output consists of predicted values ​​for each damage index.

[0092] Comprehensive health status assessment, health score:

[0093] in This represents the predicted damage value for critical components.

[0094] Alarm level: normal: Health Score ≥90, all i <0.1 Attention: 80≤ Health Score <90, or any i ∈[0.1,0.2) Warning: 70≤ Health Score <80, or any i ∈[0.2,0.3) Call the police: Health Score <70, or any i ≥0.3 Alarm Trigger: When the warning or alarm level is reached, the system automatically sends SMS / email to relevant maintenance personnel, pops up an alert window on the monitoring platform, records the event log, and initiates the emergency response process.

[0095] S4, a long-term performance tracking system, stores all monitoring data, extracted feature parameters, and assessment results in a structural health status database. Maintenance personnel can access historical data at any time, and the system also provides trend analysis capabilities. By analyzing the monthly average of the structural fundamental frequency over the past year, a curve showing its change over time is plotted. If a slow but continuous downward trend in frequency is observed, it indicates potential cumulative damage to the structure, requiring further inspection or reinforcement considerations, thus enabling condition-based predictive maintenance.

[0096] Example 2 This invention proposes a health assessment system for photovoltaic flexible support structures, such as... Figure 3 As shown, it includes: The data acquisition module is used to acquire acceleration signals, tilt angle data, and cable force data of the photovoltaic flexible support structure. The data processing module is used to extract the first three natural frequencies and mode shapes of the photovoltaic flexible support structure based on the random subspace identification method; calculate the modal damping ratio of the photovoltaic flexible support structure; analyze the tilt angle data to obtain the static displacement or dynamic rotation angle of key nodes; obtain the cable force distribution state based on the cable force data; and use the above parameters as dynamic feature vectors characterizing the health state of the photovoltaic flexible support structure. The structural health assessment module is used to input dynamic feature vectors into a pre-trained structural health assessment model and output the damage location and damage degree of the photovoltaic flexible support structure. When the damage degree exceeds a preset threshold, a report containing the location of the damaged component, the damage degree, and treatment suggestions is generated to realize the structural health assessment of the photovoltaic flexible support structure.

[0097] When the damage exceeds a preset threshold, a report is generated that includes the location of the damaged component, the extent of the damage, and recommended treatment methods. Standardize the real-time feature vectors: ; Damage prediction: ; Health rating: ; when Health Score ≥90, all i If the value is less than 0.1, then it is normal. When 80≤ Health Score <90, or any i If ∈ [0.1, 0.2), then pay attention; When 70≤ Health Score <80, or any i If the range is ∈ [0.2, 0.3), then an early warning is issued; when Health Score <70, or any i If the value is ≥0.3, an alarm will be triggered; in, These are the predicted damage values ​​for critical components. These are the mean and standard deviation of the training set, respectively. i Let be the predicted damage value for the i-th component; When the warning or alarm level is reached, the system automatically sends SMS / email to the relevant maintenance personnel, pops up an alert window on the monitoring platform, records the event log, and initiates the emergency response process.

[0098] Example 3 This invention discloses a photovoltaic flexible support structure health assessment system. Through real-time monitoring, dynamic analysis, status assessment, and long-term tracking, it achieves a comprehensive understanding of the structure's health status and enables predictive maintenance, effectively ensuring the safety, stability, and long-term operation of photovoltaic power plants. Specifically, it includes: The sensor monitoring module is used to collect response data and environmental data of the photovoltaic flexible support structure; The data transmission and processing module is used to receive the response data and environmental data, and extract dynamic feature vectors representing the structural health status from them; A health assessment and early warning module is used to assess the structural health status based on the dynamic feature vector and issue an early warning. It also includes a long-term performance tracking database module, which stores all data, parameters and evaluation results, and provides trend analysis functions; The sensor monitoring module includes a triaxial acceleration sensor and a biaxial tilt sensor installed on the top of the support rod, cable force sensors installed on the key main cables and side cables, strain gauges installed in the mid-span area of ​​the cable net and at key connection nodes, and an anemometer and a wind pressure sensor installed on the top of the mast. The sensors in the sensor monitoring module form a monitoring network through IoT nodes, and transmit the collected data to the data transmission and processing module in real time.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for health assessment of photovoltaic flexible support structures, characterized in that, Includes the following steps: Acquire acceleration signals, tilt angle data, and cable force data of the photovoltaic flexible support structure; Based on the random subspace identification method, the first three natural frequencies and mode shapes of the photovoltaic flexible support structure are extracted; the modal damping ratio of the photovoltaic flexible support structure is calculated; and the tilt angle data is analyzed to obtain the static displacement or dynamic rotation angle of the key nodes. The cable force distribution state is obtained based on cable force data; the above parameters are used as dynamic feature vectors characterizing the health status of photovoltaic flexible support structures. The dynamic feature vector is input into a pre-trained structural health assessment model, which outputs the location and extent of damage to the photovoltaic flexible support structure. When the extent of damage exceeds a preset threshold, a report containing the location of the damaged component, the extent of damage, and treatment recommendations is generated, thus realizing the health assessment of the photovoltaic flexible support structure.

2. The method for health assessment of photovoltaic flexible support structures according to claim 1, characterized in that, The random subspace identification method extracts the first three natural frequencies and mode shapes of the photovoltaic flexible support structure, specifically as follows: The preprocessed multi-channel acceleration signals are organized into a Hankel matrix, the Hankel matrix is ​​projected, and then singular value decomposition is performed. The magnitude of the singular values ​​obtained after decomposition directly corresponds to the order of the structure. Determine the system order n, and select the system order based on the magnitude of the singular values; take the first n singular values ​​and their corresponding singular vectors to construct an extended observable matrix and state sequence, establish a discrete state-space model, and solve the system matrix A and output matrix C of the discrete state-space model by least squares estimation; Eigenvalue decomposition is performed on the state matrix A to convert the discrete-time eigenvalues ​​into continuous-time eigenvalues, and the natural frequencies of each mode are calculated. f k The output matrix C is combined with the eigenvectors of the state matrix A to obtain the mode shape of the structure. Φ k ; Calculate the damping ratio based on continuous-time eigenvalues ξ k ; Repeat the above for a series of system orders, and for each order, calculate a set of modal parameters ( f k , ξ k , Φ k Plot the stability diagram in the frequency order coordinate system, and label each set of modal parameters as modal points on the stability diagram; By using frequency stability criteria, mode shape correlation criteria, and damping ratio stability criteria, the modal points that form vertical stable rods at different orders in the stability diagram are identified, and the first three natural frequencies and corresponding mode shapes of the photovoltaic flexible support structure are determined.

3. The method for health assessment of photovoltaic flexible support structures according to claim 2, characterized in that, The process involves projecting the Hankel matrix and then performing singular value decomposition. The magnitude of the singular values ​​obtained after decomposition directly corresponds to the order of the structure. Specifically: Projecting the Hankel matrix: For projection matrix O i Perform SVD decomposition: in, O i Represents the projection matrix. + Indicates Moore-Penrose pseudoinverse; H p H indicates the past part. f Indicates the future part; U Σ is a left singular vector matrix; Σ is a singular value diagonal matrix: Σ = diag(σ1, σ2, ..., σ 2i · l ), and σ1 ≥ σ2 ≥ ... ≥ 0; V It is a right singular vector matrix.

4. The method for health assessment of photovoltaic flexible support structures according to claim 2, characterized in that, The process involves taking the first n singular values ​​and their corresponding singular vectors to construct an extended observable matrix and state sequence, establishing a discrete state-space model, and then solving for the system matrix A and output matrix C of the discrete state-space model using least squares estimation. Specifically: Take the first n singular values ​​and their corresponding singular vectors: ; The extended observable matrix is: ; The state sequence is as follows: ; Establish a discrete state-space model: ; Solve for the system matrix A and output matrix C using least squares estimation: in, and It is a delayed version of the state sequence. Y i | i It is a specific block of the output matrix; It is the matrix of the first n left singular vectors. Singular value diagonal matrix; It is the matrix of the first n right singular vectors. It is the first k The system state vector at time t. It is the first k The system state vector at time +1 It is the first k The observed output vector at time t. It is observation noise. It is system noise.

5. The method for health assessment of photovoltaic flexible support structures according to claim 2, characterized in that, The state matrix A is decomposed into eigenvalues, converting the discrete-time eigenvalues ​​into continuous-time eigenvalues, and the natural frequencies of each mode are calculated. The output matrix C is combined with the eigenvectors of the state matrix A to obtain the mode shapes of the structure. Φ k ; Calculate the damping ratio based on continuous-time eigenvalues ξ k Specifically: Perform eigenvalue decomposition on the state matrix A: ; The obtained discrete-time eigenvalues ​​are converted into continuous-time eigenvalues, and then the natural frequency of each mode is calculated: No. k The first natural frequency is: ;in, , ; The output matrix C is combined with the eigenvectors of the state matrix A to obtain the mode shapes of the structure. The mode shapes are then normalized and stored as vectors. The mode shape vectors are obtained from the output matrix C and the eigenvectors Ψ. ; Damping ratio calculated based on continuous-time eigenvalues ; Where Λ is the eigenvalue diagonal matrix; Ψ is the eigenvector matrix; and |·| denotes taking the modulus of a complex number. ψ k It is the first k There are eigenvectors, and Re(·) represents taking the real part.

6. The method for health assessment of photovoltaic flexible support structures according to claim 2, characterized in that, The frequency stability criterion, mode shape correlation criterion, and damping ratio stability criterion are as follows: Frequency stability: | f n - f n+Δn | / f n < ε f ; Mode shape correlation: MAC( Φ n , Φ n+Δn ) > ε MAC ; Damping ratio stability: | ξ n - ξ n+Δn | / ξ n < ε ξ ; in, , f n These are the modal frequencies calculated for system order n. f n+Δn For the same modal frequency calculated at system order n+Δn, Φ n The modal shape vectors are calculated when the system order is n. Φ n+Δn This is the corresponding mode shape vector calculated when the system order is n+Δn. ε MAC This is the modal correlation threshold. ε f This is the frequency stability tolerance threshold. ε ξ This is the damping ratio stability tolerance threshold. ξ n The modal damping ratio is calculated for system order n. ξ n+Δn This is the corresponding damping ratio calculated when the system order is n+Δn. Let i be the mode shape vector of the i-th mode. Let be the mode shape vector of the j-th mode.

7. The method for health assessment of photovoltaic flexible support structures according to claim 1, characterized in that, The trained structural health assessment model is specifically as follows: A reference finite element model of a photovoltaic flexible support is established, and the reference finite element model is corrected by modal parameters measured by on-site environmental excitation tests. Multiple working conditions and damage states are simulated on the modified baseline finite element model to generate sample data containing input load, dynamic feature vector, and damage state labels. A deep neural network was used to train the sample data to obtain a trained structural health assessment model.

8. The method for health assessment of photovoltaic flexible support structures according to claim 1, characterized in that, When the damage exceeds a preset threshold, a report is generated that includes the location of the damaged component, the extent of the damage, and recommended treatment methods. Standardize the real-time feature vectors: ; Damage prediction: ; Health rating: ; when Health Score ≥90, all i If the value is less than 0.1, then it is normal. When 80≤ Health Score <90, or any i If ∈ [0.1, 0.2), then pay attention; When 70≤ Health Score <80, or any i If the range is ∈ [0.2, 0.3), then an early warning is issued; when Health Score <70, or any i If the value is ≥0.3, an alarm will be triggered; in, These are the predicted damage values ​​for critical components. These are the mean and standard deviation of the training set, respectively. i Let be the predicted damage value for the i-th component; When the warning or alarm level is reached, the system automatically sends SMS / email to the relevant maintenance personnel, pops up an alert window on the monitoring platform, records the event log, and initiates the emergency response process.

9. A health assessment system for photovoltaic flexible support structures, characterized in that, include: The data acquisition module is used to acquire acceleration signals, tilt angle data, and cable force data of the photovoltaic flexible support structure. The data processing module is used to extract the first three natural frequencies and mode shapes of the photovoltaic flexible support structure based on the random subspace identification method. Calculate the modal damping ratio of the photovoltaic flexible support structure; analyze the tilt angle data to obtain the static displacement or dynamic rotation angle of the key nodes; The cable force distribution state is obtained based on cable force data; the above parameters are used as dynamic feature vectors characterizing the health status of photovoltaic flexible support structures. The structural health assessment module is used to input dynamic feature vectors into a pre-trained structural health assessment model and output the damage location and damage degree of the photovoltaic flexible support structure. When the damage degree exceeds a preset threshold, a report containing the location of the damaged component, the damage degree, and treatment suggestions is generated to realize the structural health assessment of the photovoltaic flexible support structure.

10. The photovoltaic flexible support structure health assessment system according to claim 9, characterized in that, When the damage exceeds a preset threshold, a report is generated that includes the location of the damaged component, the extent of the damage, and recommended treatment methods. Standardize the real-time feature vectors: ; Damage prediction: ; Health rating: ; when Health Score ≥90, all i If the value is less than 0.1, then it is normal. When 80≤ Health Score <90, or any i If ∈ [0.1, 0.2), then pay attention; When 70≤ Health Score <80, or any i If the range is ∈ [0.2, 0.3), then an early warning is issued; when Health Score <70, or any i If the value is ≥0.3, an alarm will be triggered; in, These are the predicted damage values ​​for critical components. These are the mean and standard deviation of the training set, respectively. i Let be the predicted damage value for the i-th component; When the warning or alarm level is reached, the system automatically sends SMS / email to the relevant maintenance personnel, pops up an alert window on the monitoring platform, records the event log, and initiates the emergency response process.

Images