A corrosion early warning system and method for fixed photovoltaic bracket

Abstract

The present invention relates to the field of photovoltaic support technology, and more particularly to a corrosion early warning system and method for fixed photovoltaic supports. The system comprises: a data acquisition module for acquiring multimodal data of the photovoltaic support; a feature extraction module for determining a comprehensive corrosion feature based on the features of the multimodal data, and determining an adjustment method based on the difference between a scoring threshold and a score when the quality of the enhanced comprehensive corrosion feature is determined to be low; a feature processing module for generating a corrosion distribution map based on the enhanced comprehensive corrosion feature, constructing a spatiotemporal corrosion model based on historical corrosion data and corrosion severity values, and generating a corrosion thermogram based on the corrosion severity and radiation trend; and an early warning management module for determining the scope of an early warning area based on the corrosion distribution map and the corrosion thermogram, and determining an adjustment method if the warning area is unqualified after comparison with the predicted early warning area. The present invention improves the accuracy of corrosion early warning for fixed photovoltaic supports.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic supports, and in particular to a corrosion early warning system and method for a fixed photovoltaic support. Background Art

[0002] Photovoltaic power generation, as a clean energy technology, has been widely adopted worldwide in recent years. Fixed photovoltaic mounting systems are a crucial component of photovoltaic power generation systems, supporting and securing photovoltaic modules to ensure they receive solar radiation at optimal angles. However, in vast plains or deserts, fixed photovoltaic mounting systems are exposed to harsh natural environments for long periods of time, facing erosion from corrosive factors such as high temperatures and wind and sand. This can cause the mounting materials to gradually age and corrode, affecting their structural strength and service life.

[0003] Chinese Patent Application Publication No. CN113489454A discloses a corrosion monitoring system and method for photovoltaic brackets. The present invention utilizes a corrosion detection module to perform real-time corrosion detection on corrosion-sensitive locations of the photovoltaic bracket, obtaining a detection value that is then sent to an early warning module. The early warning module obtains the initial state value of the photovoltaic bracket when it is uncorroded, calculates the difference between the detection value and the initial state value, and compares the difference with a preset alarm threshold. When the difference is greater than or equal to the preset alarm threshold, the early warning module generates an early warning signal based on the corrosion-sensitive location and issues an alarm based on the early warning signal. This embodiment detects the corrosion state in real time, and when the difference between the detection value and the initial state value is greater than or equal to the preset alarm threshold, promptly notifies and alerts maintenance personnel to inspect and repair the corrosion-sensitive location, thereby improving the service life of the photovoltaic bracket and the safety of the power plant. However, the prior art suffers from the following problems: The prior art only uses the difference between the detection value and the initial state value to compare with the preset alarm threshold for corrosion early warning, lacking analysis of the complexity of photovoltaic bracket corrosion factors and accurate early warning based on the temporal and spatial evolution patterns. This results in insufficient accuracy in photovoltaic bracket corrosion early warning. Summary of the Invention

[0004] To this end, the present invention provides a corrosion warning system and method for a fixed photovoltaic bracket, which is used to overcome the problem in the prior art that corrosion warning is only performed by comparing the difference between the detection value and the initial state value with a preset alarm threshold, lacks analysis of the complexity of the corrosion factors of the photovoltaic bracket and accurate warning of the temporal and spatial evolution laws, thereby resulting in insufficient accuracy of the corrosion warning of the photovoltaic bracket.

[0005] To achieve the above objectives, the present invention provides a corrosion early warning system for a fixed photovoltaic support, comprising:

[0006] A data acquisition module for acquiring multimodal data of the photovoltaic support, comprising a multispectral camera disposed on the surface of the photovoltaic support for collecting visible light image data and infrared spectrum image data, an ultrasonic sensor disposed at a key support point of the photovoltaic support for collecting ultrasonic echo signal data, and an environmental sensor disposed on the upper portion of the photovoltaic support for collecting environmental data;

[0007] a feature extraction module connected to the data acquisition module, configured to determine a comprehensive corrosion feature based on the features of the multimodal data, and, if the quality of the enhanced comprehensive corrosion feature is determined to be low, determine, based on the difference between the score threshold and the score, whether to increase the weight of the generator loss function or increase the dimension of the random noise vector;

[0008] a feature processing module connected to the feature extraction module, configured to generate a corrosion distribution map based on the enhanced comprehensive corrosion features, construct a spatiotemporal corrosion model based on historical corrosion data and corrosion severity values, and generate a corrosion thermodynamic map based on the corrosion severity and radiation trend;

[0009] An early warning management module is connected to the feature processing module and is used to determine the early warning area range based on the corrosion distribution map and the corrosion thermodynamic map, and adjust the corrosion degree threshold or the early warning area range according to the unqualified comparison result between the early warning area range and the predicted early warning area range.

[0010] Furthermore, the feature extraction module includes:

[0011] A multimodal feature fusion unit, for extracting features of the multimodal data respectively and fusing them to obtain a comprehensive corrosion feature after dimensionality reduction;

[0012] A feature enhancement unit is connected to the multimodal feature fusion unit and is used to combine the dimensionality-reduced comprehensive corrosion feature and the enhanced comprehensive corrosion feature into an enhanced feature data set.

[0013] Furthermore, the multimodal feature fusion unit includes:

[0014] The image feature extraction subunit is used to convert the color images of the visible light image and the infrared spectrum image into grayscale images through a gray level co-occurrence matrix and extract texture features of the grayscale images;

[0015] The ultrasonic feature extraction subunit is connected to the image feature extraction subunit and is used to extract ultrasonic echo features related to the corrosion area of ​​the photovoltaic support based on the collected ultrasonic echo signals;

[0016] The environmental feature extraction subunit is connected to the ultrasonic feature extraction subunit and is used to construct an environmental feature vector based on environmental data to determine the impact of environmental conditions on photovoltaic support corrosion;

[0017] The feature fusion subunit is connected to the environmental feature extraction subunit and is used to fuse the texture feature, the ultrasonic echo feature and the environmental feature vector to generate a comprehensive corrosion feature.

[0018] Furthermore, the feature fusion subunit performs weighted processing on the comprehensive corrosion feature to obtain a weighted comprehensive corrosion feature, and performs dimensionality reduction processing on the weighted comprehensive corrosion feature according to a principal component analysis method to obtain a dimensionality reduced comprehensive corrosion feature.

[0019] Furthermore, the feature enhancement unit determines that the quality of the enhanced comprehensive corrosion feature is low based on the comparison result that the discriminator's score for the enhanced comprehensive corrosion feature is less than a score threshold, and determines to increase the weight of the generator loss function by a preset weight coefficient based on the comparison result that the difference between the score threshold and the score is less than or equal to a preset difference.

[0020] Furthermore, the feature enhancement unit determines to increase the dimension of the random noise vector by a preset noise vector adjustment coefficient based on a comparison result that the difference between the score threshold and the score is greater than a preset difference.

[0021] Furthermore, the feature processing module includes:

[0022] The corrosion distribution map generating unit is used to convert discrete corrosion detection points into a continuous corrosion distribution map based on the corrosion degree value;

[0023] The spatiotemporal corrosion model unit is used to construct a spatiotemporal corrosion model by combining historical corrosion data and the corrosion degree value;

[0024] The dynamic heat map generation unit is used to generate a corrosion heat map according to the corrosion degree and radiation trend.

[0025] Furthermore, the warning management module determines that the warning area range is unqualified based on the comparison result that the warning area range is smaller than the predicted warning area range of the spatiotemporal corrosion model, and determines to reduce the corrosion degree threshold with the corresponding preset threshold adjustment coefficient based on the comparison result of the ratio of the warning area range to the predicted warning area range and the preset ratio.

[0026] Furthermore, the warning management module determines that the warning area range is unqualified based on the fact that the warning area range is larger than the predicted warning area range of the spatiotemporal corrosion model, and determines to increase the warning area range by the corresponding preset range adjustment coefficient based on the comparison result of the relative difference between the warning area range and the predicted warning area range and the preset relative difference.

[0027] Another aspect of the present invention provides a corrosion early warning method for a fixed photovoltaic support, comprising:

[0028] Obtain multimodal data of photovoltaic brackets;

[0029] Based on the characteristics of multimodal data, the comprehensive corrosion characteristics are determined, and after weighting the comprehensive corrosion characteristics, dimensionality reduction is performed according to the principal component analysis method to obtain the comprehensive corrosion characteristics after dimensionality reduction;

[0030] Under the condition of low quality of enhanced comprehensive corrosion characteristics, the adjustment method is determined according to the difference between the scoring threshold and the score;

[0031] Generate corrosion distribution maps based on enhanced comprehensive corrosion features, build spatiotemporal corrosion models based on historical corrosion data and corrosion severity values, and generate corrosion thermodynamic maps based on corrosion severity and radiation trends.

[0032] The scope of the early warning area is determined based on the corrosion distribution map and the corrosion thermodynamic map, and the adjustment method is determined based on the unqualified comparison result with the predicted early warning area.

[0033] Compared with the prior art, the beneficial effect of the present invention lies in that the present invention obtains multimodal data of photovoltaic brackets through the data acquisition module, the feature extraction module determines the comprehensive corrosion characteristics, the feature processing module generates the corrosion distribution map, the spatiotemporal corrosion model and the corrosion thermodynamic map, the early warning management module determines the early warning area, and the multimodal data is integrated to perform corrosion feature extraction and spatiotemporal modeling, thereby improving the accuracy of corrosion detection. The generated corrosion thermodynamic map intuitively displays the corrosion situation, improves the response capability to corrosion early warning, and thus improves the accuracy of corrosion early warning of fixed photovoltaic brackets.

[0034] Furthermore, the present invention comprehensively extracts the corrosion features of photovoltaic brackets through a multimodal feature fusion unit and a feature enhancement unit. The image features use the grayscale co-occurrence matrix to extract texture features, the ultrasonic features construct feature vectors based on the echo signals, and the environmental features construct feature vectors based on the environmental data. The features of multiple data sources are integrated to improve the comprehensiveness and accuracy of corrosion feature extraction. The feature extraction methods for different data sources take into account the characteristics of each data and the sensitivity to the impact of corrosion, thereby achieving more refined corrosion monitoring and evaluation.

[0035] Furthermore, the present invention highlights important data features through weighted summation and then uses principal component analysis for dimensionality reduction processing, and then uses generative adversarial networks to generate enhanced comprehensive corrosion feature samples to solve the problems of sample imbalance and poor model generalization ability. The feature data set is further optimized through dimensionality reduction and feature enhancement technology, the expression ability of corrosion features and the generalization performance of the model are improved, and accurate and reliable technical support is provided for corrosion monitoring and early warning of photovoltaic brackets.

[0036] Furthermore, the present invention generates a continuous corrosion distribution map through interpolation, builds a spatiotemporal corrosion model based on historical corrosion data to predict corrosion trends, and uses a thermal diffusion model to dynamically generate a corrosion thermogram, providing intuitive corrosion distribution and trend predictions. The radiation trend of corrosion is displayed through a thermogram, providing comprehensive and detailed technical support for corrosion monitoring and early warning of photovoltaic brackets, and improving the ability to promptly detect and deal with potential corrosion problems.

[0037] Furthermore, the present invention compares the corrosion distribution map, corrosion thermodynamic map and predicted warning area range of the spatiotemporal corrosion model through the early warning management module, and adjusts the warning area range and corrosion degree threshold according to the comparison results, thereby ensuring the accuracy and rationality of the warning area and improving the precision and response efficiency of corrosion early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a structural diagram of a corrosion early warning system for a fixed photovoltaic support according to an embodiment of the present invention;

[0039] Figure 2 This is a flow chart of a corrosion early warning method for a fixed photovoltaic support according to an embodiment of the present invention;

[0040] Figure 3 Schematic diagram of the structure of a multimodal feature fusion unit according to an embodiment of the present invention;

[0041] Figure 4 Schematic diagram of the structure of the feature processing module of an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0043] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0044] It should be pointed out that the data in this embodiment are obtained by comprehensive analysis and evaluation of the historical test data and the corresponding historical test results of the three months before this test. It can be understood by those skilled in the art that the present invention can determine the above parameters for a single item by selecting the value with the highest proportion as the preset standard parameter based on the data distribution, using weighted summation to use the obtained value as the preset standard parameter, substituting each historical data into a specific formula and using the value obtained by the formula as the preset standard parameter or other selection methods, as long as the present invention can clearly define the different specific situations in the single determination process through the obtained values.

[0045] See also Figures 1-4 As shown, Figure 1 This is a structural diagram of a corrosion early warning system for a fixed photovoltaic support according to an embodiment of the present invention; Figure 2 This is a flow chart of a corrosion early warning method for a fixed photovoltaic support according to an embodiment of the present invention; Figure 3 Schematic diagram of the structure of a multimodal feature fusion unit according to an embodiment of the present invention; Figure 4 Schematic diagram of the structure of the feature processing module of an embodiment of the present invention.

[0046] The embodiment of the present invention is a corrosion early warning system for a fixed photovoltaic support, comprising:

[0047] A data acquisition module is used to acquire multimodal data of the photovoltaic support, wherein the multimodal data includes visible light image data, infrared spectrum data, ultrasonic echo signal data, and environmental data. The module comprises a multispectral camera disposed on the surface of the photovoltaic support to collect visible light image data and infrared spectrum image data; an ultrasonic sensor disposed at a key support point of the photovoltaic support to collect ultrasonic echo signal data; and an environmental sensor disposed on the upper portion of the photovoltaic support to collect the environmental data.

[0048] a feature extraction module connected to the data acquisition module, configured to determine a comprehensive corrosion feature based on the features of the multimodal data, and, if the quality of the enhanced comprehensive corrosion feature is determined to be low, determine, based on the difference between the score threshold and the score, whether to increase the weight of the generator loss function or increase the dimension of the random noise vector;

[0049] a feature processing module connected to the feature extraction module, configured to generate a corrosion distribution map based on the enhanced comprehensive corrosion features, construct a spatiotemporal corrosion model based on historical corrosion data and corrosion severity values, and generate a corrosion thermodynamic map based on the corrosion severity and radiation trend;

[0050] An early warning management module is connected to the feature processing module and is used to determine the early warning area range based on the corrosion distribution map and the corrosion thermodynamic map, and adjust the corrosion degree threshold or the early warning area range according to the unqualified comparison result between the early warning area range and the predicted early warning area range.

[0051] In the embodiment of the present invention, the key support points are corners and connection nodes of the photovoltaic bracket.

[0052] In an embodiment of the present invention, the environmental data includes temperature, humidity, wind speed and light intensity.

[0053] In an embodiment of the present invention, the multispectral camera collects images at a frequency of 5 frames per second; the ultrasonic sensor collects data at a frequency of 10 times per second; and the environmental sensor collects data at a frequency of 10 times per second.

[0054] Please refer to Figure 3 As shown, the feature extraction module includes a multimodal feature fusion unit and a feature enhancement unit;

[0055] The multimodal feature fusion unit includes an image feature extraction subunit, an ultrasonic feature extraction subunit, an environmental feature extraction subunit and a feature fusion subunit.

[0056] It's understandable that multimodal feature fusion can combine the strengths of different data sources to improve the comprehensiveness and accuracy of corrosion signatures. Conventional feature extraction methods typically rely on a single data source and fail to fully reflect the complex characteristics of corrosion. Therefore, this embodiment uses multimodal feature fusion to integrate image texture features, ultrasonic echo features, and environmental features to generate a more discriminative comprehensive corrosion signature.

[0057] The image feature extraction subunit is used to extract texture features of visible light images and infrared spectrum images through gray level co-occurrence matrix (GLCM).

[0058] Specifically, the grayscale representation of converting the color images of the visible light image and the infrared spectrum image into grayscale images is:

[0059]

[0060] Where, represents the pixel value of the grayscale image, They represent the red, green, and blue channel values ​​in the color image, respectively. The coefficients 0.299, 0.587, and 0.114 represent the weights used to weight the red, green, and blue channels, respectively. The weight coefficients are determined based on the human eye's perception of the intensity of light of different wavelengths.

[0061] Specifically, under the condition that the image feature extraction subunit obtains the pixel value of the grayscale image, the grayscale co-occurrence matrix is ​​calculated by calculating the co-occurrence probability of the grayscale values ​​of the pixel pairs in the grayscale image in a specific direction and distance to obtain the texture feature of the grayscale image, which is expressed as:

[0062]

[0063] Where, P is the direction and distance Above, the gray value is The pixel and gray value are The probability of pixels appearing simultaneously is represents the pixel value of the grayscale image, , Respectively represent the coordinates of the first pixel and the second pixel in the grayscale image. These two pixels appear in pairs to form a pixel pair. and Represent the grayscale values ​​in the grayscale image, , L-1 represents the number of gray levels of the grayscale image, that is, the total number of grayscale values ​​in the grayscale image, Represents the distance between pixel pairs, usually Euclidean distance, Indicates the direction angle between pixel pairs, usually 0°, 45°, 90°, 135°, etc.

[0064] After obtaining the grayscale co-occurrence matrix, the texture features of the grayscale image are extracted. The texture features generally include the contrast used to determine the clarity and depth of the grayscale image, the energy used to determine the uniformity of the grayscale distribution and the coarseness of the texture of the grayscale image, the entropy used to determine the complexity of the texture of the grayscale image, and the correlation used to determine the linear dependence of the grayscale values ​​in the grayscale image. The calculation process of the above texture features is a conventional process and will not be repeated here.

[0065] Please refer to Figure 3 As shown, the ultrasonic feature extraction subunit is used to extract features related to the corrosion area of ​​the photovoltaic bracket based on the collected ultrasonic echo signal.

[0066] Specifically, the ultrasonic echo feature vector is constructed according to the echo time (TOF), echo amplitude (A), and frequency change (Δf) of the ultrasonic echo signal, which is expressed as:

[0067]

[0068] Where, is the ultrasonic echo eigenvector, Echo time, which indicates the time difference between the ultrasonic wave emission and the echo signal reception. is the echo amplitude, Indicates the amplitude of the ultrasonic emission signal, Indicates the attenuation coefficient of the photovoltaic bracket, The distance that the ultrasonic wave propagates. is the frequency change, Indicates the frequency of the transmitted signal, Indicates the frequency of the received signal.

[0069] In the above scheme, by constructing the ultrasonic echo feature vector, the depth of the corrosion area, the change in acoustic impedance and the frequency characteristics can be inferred, thereby determining the degree and type of corrosion.

[0070] Furthermore, the environmental feature extraction subunit is used to determine the impact of environmental conditions on the corrosion of the photovoltaic support based on the environmental data.

[0071] Specifically, the environmental feature vector is constructed based on the temperature feature, humidity feature, wind speed feature, and light intensity feature in the environmental data, which is expressed as:

[0072]

[0073] Where, is the environmental feature vector, is the temperature characteristic, is the humidity characteristic, is the wind speed characteristic, is the light intensity characteristic, They are the average temperature, average humidity, average wind speed, and average light intensity. are the standard deviation of temperature, the standard deviation of humidity, the standard deviation of wind speed, and the standard deviation of light intensity. They are the maximum temperature, maximum humidity, maximum wind speed, and maximum light intensity. They are the minimum temperature, minimum humidity, minimum wind speed, and minimum light intensity.

[0074] In the above scheme, the influence of environmental conditions on corrosion is characterized by constructing an environmental feature vector.

[0075] Please refer to Figure 3 As shown in FIG, the feature fusion subunit fuses the image texture features, ultrasonic echo features and environmental features to generate comprehensive corrosion features.

[0076] It can be understood that by combining the features of different modes and utilizing the complementary information of each mode, the accuracy and robustness of corrosion detection can be improved.

[0077] Specifically, the grayscale image texture features, ultrasonic echo feature vectors, and environmental feature vectors are directly connected together to form a high-dimensional feature vector, namely the comprehensive corrosion feature, which is expressed as:

[0078]

[0079] Where, For comprehensive corrosion characteristics, is the grayscale image texture feature, is the ultrasonic echo eigenvector, is the environmental feature vector.

[0080] It can be understood that directly connecting the grayscale image texture features, ultrasonic echo feature vectors, and environmental feature vectors together can retain the original information of each modal feature and avoid information loss.

[0081] According to the importance of grayscale image texture features, ultrasonic echo feature vectors and environmental feature vectors, a weighted summation is performed on the grayscale image texture features, ultrasonic echo feature vectors and environmental feature vectors to highlight the features of important data and suppress the features of unimportant data.

[0082] Specifically, the generated comprehensive corrosion features are weighted and expressed as:

[0083]

[0084] Where, is the weighted comprehensive corrosion feature, Represents the texture features of grayscale images, represents the ultrasonic echo feature vector, Represents the weighting coefficient of the environmental feature vector.

[0085] It is understandable that , the value range of each weight coefficient is The specific value of each weight coefficient is determined according to the average value of the ratio of the accuracy of each single modal feature to the accuracy of the comprehensive corrosion feature in several historical corrosion tests.

[0086] The comprehensive corrosion features after weighted fusion are high-dimensional and contain a lot of redundant information. Therefore, in order to reduce the computational complexity and improve the model performance, the weighted comprehensive corrosion features are subjected to dimensionality reduction processing. Since the corrosion areas of the photovoltaic brackets are not labeled, the principal component analysis method is used to map the high-dimensional data to a low-dimensional space through linear transformation, retaining the direction with the largest variance in the data. The calculation process of dimensionality reduction using the principal component analysis method is a conventional process and will not be repeated here.

[0087] It is understandable that dimensionality reduction of the comprehensive corrosion features after weighted fusion can reduce computational complexity, remove redundant information and improve model performance, but there may be problems of information loss and reduced interpretability. Therefore, the feature enhancement unit generates enhanced comprehensive corrosion feature samples through generative adversarial networks (GANs), which can solve the problems of sample imbalance and poor model generalization ability.

[0088] It should be noted that the generative adversarial network consists of two neural networks, including a generator for generating synthetic data similar to real data, and a discriminator for distinguishing real data from synthetic data generated by the generator.

[0089] In the embodiment of the present invention, the comprehensive corrosion characteristics after dimension reduction of the input are set The dimension is ,in for The number of samples, for The feature dimension of ; set the input random noise vector The dimension is ,in For the dimension of the noise vector, set the enhanced comprehensive corrosion feature generated by the generator The dimension is The above setting process is a conventional process and will not be repeated here.

[0090] Specifically, the generator receives a random noise vector As input, random noise Sampling from a uniform distribution or a normal distribution, the generator transforms random noise into Mapped to synthetic feature samples, expressed as:

[0091]

[0092] Where, To enhance the comprehensive corrosion characteristics, represents a generator, represents a random noise vector, Represents the synthetic feature samples generated by the generator.

[0093] It should be noted that the fully connected neural network uses two fully connected hidden layers with 256 and 128 neurons respectively. Each hidden layer is followed by an activation function ReLU. It uses one fully connected output layer with 50 neurons and an activation function Tanh.

[0094] Specifically, the generator loss function for generator training is calculated as:

[0095]

[0096] Where, is the generator loss function, represents a random noise vector The probability distribution of Represents the discriminator's response to the synthetic feature sample The score of , that is, the score of the discriminator on the enhanced comprehensive corrosion feature, that is, the probability that the discriminator believes that the generated sample is a real sample.

[0097] It can be understood that if the discriminator's score for the synthetic feature sample is less than the score threshold, it is determined that the distribution difference is large and the quality of the synthetic feature sample is low; if the discriminator's score for the synthetic feature sample is greater than or equal to the score threshold, it is determined that the distribution difference is small and the quality of the synthetic feature sample is high.

[0098] The synthetic feature sample generation process is adjusted according to the comparison result of the difference between the scoring threshold and the score and the preset difference. If the difference between the scoring threshold and the score is less than or equal to the preset difference, the weight of the generator loss function is increased; if the difference between the scoring threshold and the score is greater than the preset difference, the random noise vector is increased by the preset noise vector adjustment coefficient 2. dimension.

[0099] Specifically, the weight of the generator loss function is increased, expressed as:

[0100]

[0101] Where, is the preset weight coefficient, The value range is , those skilled in the art can adjust the value according to actual needs.

[0102] After dimensionality reduction, the comprehensive corrosion features and enhanced comprehensive corrosion characteristics Input the discriminator to measure whether the discriminator can correctly distinguish between real data and synthetic data.

[0103] Specifically, the discriminator loss function is calculated as:

[0104]

[0105] Where, is the discriminator loss function, Represents the comprehensive corrosion characteristics after dimensionality reduction , Represents the distribution of comprehensive corrosion features after dimensionality reduction.

[0106] It should be noted that the generator and the discriminator are continuously optimized through adversarial training, and the generator is minimized. , the discriminator minimizes , the training process is carried out alternately.

[0107] After dimensionality reduction, the comprehensive corrosion features and enhanced comprehensive corrosion characteristics Combined to form an enhanced feature dataset, expressed as:

[0108]

[0109] In the above scheme, enhanced comprehensive corrosion feature samples are generated by generative adversarial networks to solve the problems of sample imbalance and poor model generalization ability.

[0110] Please refer to Figure 4 As shown, the feature processing module includes a corrosion distribution map generation unit, a spatiotemporal corrosion model unit, and a dynamic thermal map generation unit;

[0111] The corrosion distribution map generation unit is used to convert discrete corrosion detection points into a continuous corrosion distribution map, which intuitively displays the location, degree and distribution of the corrosion area.

[0112] It is understood that spatial interpolation methods can leverage the spatial distribution information of discrete corrosion detection points to improve the continuity and accuracy of corrosion distribution maps. Conventional interpolation methods typically rely on a single spatial relationship and are unable to fully reflect the complex spatial distribution characteristics of corrosion. Therefore, this embodiment uses a combination of multiple interpolation methods and exploits the spatial autocorrelation of Kriging interpolation to generate a more accurate and robust continuous corrosion distribution map.

[0113] Specifically, the enhanced feature dataset is used to estimate the degree of corrosion at different locations using Kriging interpolation, which is expressed as:

[0114]

[0115] Where, For location The corrosion degree value at Indicates the The weight coefficient of the known points, Indicates the The eigenvalues ​​of the enhanced feature dataset of known points.

[0116] It should be noted that The variance function is calculated by fitting the experimental variance function. The calculation process of the above fitting experiment is a conventional process and will not be repeated here.

[0117] For each interpolation point , calculate its corrosion degree value , a continuous corrosion distribution map is generated, and the corrosion degree is graded. If the corrosion degree value of the corrosion location is less than or equal to the first corrosion degree threshold Q1, the corrosion location is determined to be slightly corroded; if the corrosion degree value of the corrosion location is greater than the first corrosion degree threshold and less than or equal to the second corrosion degree threshold Q2, the corrosion location is determined to be moderately corroded; if the corrosion degree value of the corrosion location is greater than the second corrosion degree threshold, the corrosion location is determined to be severely corroded.

[0118] It should be noted that the first corrosion degree threshold and the second corrosion degree threshold are determined according to the mean ± standard deviation method, which is expressed as:

[0119]

[0120] Where Q1 and Q2 represent the first corrosion threshold and the second corrosion threshold, respectively. All corrosion severity values The mean and standard deviation of Q1 are 0.25 and Q2 are 0.75.

[0121] In the above scheme, a continuous corrosion distribution map is generated by applying the Kriging interpolation method and utilizing the enhanced comprehensive corrosion features.

[0122] The spatiotemporal corrosion model unit is used to combine historical corrosion data and current detection results to construct a spatiotemporal corrosion model and predict corrosion development trends.

[0123] It is understood that spatiotemporal modeling can leverage the temporal and spatial evolution of corrosion data to improve the accuracy and reliability of corrosion trend predictions. Conventional prediction methods typically rely solely on a single time series or spatial distribution, making it difficult to fully reflect the complex temporal and spatial evolution of corrosion. Therefore, this embodiment, through the comprehensive application of spatiotemporal modeling methods, leverages the temporal dependence of time series analysis and the spatial autocorrelation of spatial interpolation to generate more accurate and robust corrosion trend prediction results.

[0124] The corrosion degree value at the location Align the historical corrosion data in time and space to ensure data consistency. The above alignment process is a conventional process and will not be repeated here. The historical corrosion data includes a timestamp. , spatial location and corrosion severity values .

[0125] Specifically, the time series model is used to model the time evolution of the corrosion degree, which can be expressed as:

[0126]

[0127] Where, For in time and location The corrosion degree value at It is the time series model ARIMA or LSTM, which is not limited to specific ones. is the time window size, is a random error, set to have a mean of 0 and a variance of Gaussian noise.

[0128] Specifically, the spatial interpolation method is used to calculate the value of each time point. , the Kriging interpolation method is used to estimate the corrosion degree at different locations, which is expressed as

[0129]

[0130] Where, Indicates the location and time The predicted corrosion degree value at Indicates the first The weight coefficient of the known points, Indicates the actual corrosion extent value at a known location and time.

[0131] Specifically, the time series model and the spatial interpolation method are combined to construct a spatiotemporal erosion model, which is expressed as:

[0132]

[0133] Where, represents the spatiotemporal corrosion model, It is a spatial interpolation method, namely the Kriging interpolation method.

[0134] The constructed spatiotemporal corrosion model is trained using historical data and the model performance is evaluated through cross-validation or holdout method. The above evaluation process is a routine process and will not be repeated here.

[0135] Specifically, the trained spatiotemporal corrosion model is used to predict the degree of corrosion at future time points:

[0136]

[0137] Where, Expressed as the predicted corrosion degree value at a future time point.

[0138] In the above scheme, a spatiotemporal corrosion model is constructed by combining the time series model and the spatial interpolation method to predict the evolution trend of corrosion in time and space, thereby improving the accuracy and reliability of photovoltaic bracket corrosion warning.

[0139] The dynamic heat map generation unit is used to dynamically generate a corrosion heat map according to the corrosion degree and the radiation trend to the surrounding area.

[0140] It is understood that the heatmap generation method can utilize the spatial distribution information and radiation trends of the corrosion degree to enhance the intuitiveness and dynamics of the corrosion heatmap. Conventional heatmap generation methods typically rely on a single corrosion degree value, which fails to fully reflect the radiation characteristics and development trends of corrosion in space. Therefore, this embodiment uses the comprehensive application of the heat diffusion model to utilize the spatial distribution and radiation trends of the corrosion degree to generate a more intuitive corrosion heatmap.

[0141] Specifically, using time and location The corrosion degree value at Calculate the thermal value, expressed as:

[0142]

[0143] Where, is the thermal value, They represent the maximum and minimum corrosion degree values ​​respectively.

[0144] Specifically, the thermal diffusion model is used to simulate the radiation trend of corrosion in space, which can be expressed as:

[0145]

[0146] Where, represents the diffusion coefficient, with a value of 0.1. Respectively represent the thermal value in and The second-order partial derivative in the direction.

[0147] Specifically, the radiation trend and thermal value of corrosion in space are combined to obtain the iterative formula of the heat diffusion equation, which is expressed as:

[0148]

[0149] Where, is the thermal value of the iteration, Indicates the time of growth.

[0150] The updated color value Mapped onto the heat map, a dynamic heat map is generated, which is expressed as:

[0151]

[0152] Where, is the updated color value, Indicates thermal value The corresponding color value is Indicates thermal value The corresponding color value.

[0153] It should be noted that in order to meet the visualization requirements, the color gradient from green to red is set. The corresponding color is green. The corresponding color is red.

[0154] In the above scheme, by combining the corrosion degree value and the heat diffusion model, the location, degree and radiation trend of the corrosion area are displayed, which improves the dynamics and intuitiveness of the corrosion heat map.

[0155] Specifically, the warning management module determines the warning area range based on the corrosion distribution map and the corrosion thermodynamic map. If the warning area range is smaller than the predicted warning area range of the space-time corrosion model, it is determined that the warning area range is unqualified; if the warning area range is equal to the predicted warning area range of the space-time corrosion model, it is determined that the warning area range is qualified; if the warning area range is larger than the predicted warning area range of the space-time corrosion model, it is determined that the warning area range is unqualified.

[0156] Specifically, under the condition that the warning area range is determined to be unqualified based on the comparison result that the warning area range is smaller than the predicted warning area range of the spatiotemporal corrosion model, if the comparison result of the ratio of the warning area range to the predicted warning area range is less than or equal to the preset ratio, it is determined that the first corrosion degree threshold is reduced by a first preset threshold adjustment coefficient of 0.97; if the comparison result of the ratio of the warning area range to the predicted warning area range is greater than the preset ratio, it is determined that the first corrosion degree threshold is reduced by a second preset threshold adjustment coefficient of 0.93.

[0157] The default ratio is 0.5.

[0158] Specifically, under the condition that the warning area range is determined to be unqualified based on the comparison result that the warning area range is larger than the predicted warning area range of the spatiotemporal corrosion model, if the relative difference between the warning area range and the predicted warning area range is less than or equal to the preset relative difference, the warning area range is increased by a first preset range adjustment coefficient of 1.3; if the relative difference between the warning area range and the predicted warning area range is greater than the preset relative difference, the warning area range is increased by a second preset range adjustment coefficient of 1.5.

[0159] The default relative difference value is 0.65.

[0160] Another embodiment of the present invention further provides a corrosion early warning method for a fixed photovoltaic support, comprising:

[0161] Step S1, obtaining multimodal data of a photovoltaic bracket;

[0162] Step S2, determining a comprehensive corrosion feature based on the features of the multimodal data, performing a weighted process on the comprehensive corrosion feature, and then performing a dimensionality reduction process using a principal component analysis method to obtain a dimensionality-reduced comprehensive corrosion feature;

[0163] Step S3, based on the condition that the quality of the enhanced comprehensive corrosion feature is low, determining an adjustment method according to the difference between the score threshold and the score;

[0164] Step S4: generating a corrosion distribution map based on the enhanced comprehensive corrosion characteristics, building a spatiotemporal corrosion model based on historical corrosion data and corrosion severity values, and generating a corrosion thermodynamic map based on the corrosion severity and radiation trend;

[0165] Step S5: determining the warning area based on the corrosion distribution map and the corrosion thermodynamic map, and determining an adjustment method based on a comparison result with the predicted unqualified warning area.

[0166] The above method can be implemented on any carrier.

[0167] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

[0168] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A corrosion warning system for fixed photovoltaic brackets, characterized in that: include: A data acquisition module for acquiring multimodal data of the photovoltaic support, comprising a multispectral camera disposed on the surface of the photovoltaic support for collecting visible light image data and infrared spectrum image data, an ultrasonic sensor disposed at a key support point of the photovoltaic support for collecting ultrasonic echo signal data, and an environmental sensor disposed on the upper portion of the photovoltaic support for collecting environmental data; a feature extraction module connected to the data acquisition module, configured to determine a comprehensive corrosion feature based on features of the multimodal data, and determine, in a feature enhancement unit, that the quality of the enhanced comprehensive corrosion feature is low based on a comparison result in which the discriminator's score for the enhanced comprehensive corrosion feature is less than a score threshold, and determine, based on a comparison result in which the difference between the score threshold and the score is less than or equal to a preset difference, to increase the weight of a generator loss function by a preset weight coefficient, or, based on a comparison result in which the difference between the score threshold and the score is greater than a preset difference, to increase the dimension of a random noise vector by a preset noise vector adjustment coefficient; a feature processing module connected to the feature extraction module, configured to generate a corrosion distribution map based on the enhanced comprehensive corrosion features, construct a spatiotemporal corrosion model based on historical corrosion data and corrosion severity values, and generate a corrosion thermodynamic map based on the corrosion severity and radiation trend; An early warning management module is connected to the feature processing module, and is used to determine the scope of the early warning area based on the corrosion distribution map and the corrosion thermodynamic map, and determine that the scope of the early warning area is unqualified based on the comparison result that the scope of the early warning area is smaller than the predicted early warning area of ​​the spatiotemporal corrosion model, and determine to reduce the corrosion degree threshold by a corresponding preset threshold adjustment coefficient based on the comparison result of the ratio of the early warning area to the predicted early warning area with a preset ratio, or determine to increase the scope of the early warning area by a corresponding preset range adjustment coefficient based on the comparison result of the relative difference between the early warning area and the predicted early warning area with a preset relative difference.

2. The corrosion warning system for a fixed photovoltaic support according to claim 1, characterized in that: The feature extraction module includes: A multimodal feature fusion unit, for extracting features of the multimodal data respectively and fusing them to obtain a comprehensive corrosion feature after dimensionality reduction; A feature enhancement unit is connected to the multimodal feature fusion unit and is used to combine the dimensionality-reduced comprehensive corrosion feature and the enhanced comprehensive corrosion feature into an enhanced feature data set.

3. The corrosion warning system for a fixed photovoltaic support according to claim 2, characterized in that: The multimodal feature fusion unit includes: An image feature extraction subunit, configured to convert the color images of the visible light image and the infrared spectrum image into grayscale images through a gray-level co-occurrence matrix and extract texture features of the grayscale images; An ultrasonic feature extraction subunit, connected to the image feature extraction subunit, for extracting ultrasonic echo features related to the corrosion area of ​​the photovoltaic support based on the collected ultrasonic echo signals; An environmental feature extraction subunit, connected to the ultrasonic feature extraction subunit, is used to construct an environmental feature vector based on environmental data to determine the impact of environmental conditions on photovoltaic support corrosion; A feature fusion subunit is connected to the environmental feature extraction subunit and is used to fuse the texture feature, the ultrasonic echo feature and the environmental feature vector to generate a comprehensive corrosion feature.

4. The corrosion warning system for a fixed photovoltaic support according to claim 3, characterized in that: The feature fusion subunit performs weighted processing on the comprehensive corrosion feature to obtain a weighted comprehensive corrosion feature, and performs dimensionality reduction processing on the weighted comprehensive corrosion feature according to a principal component analysis method to obtain a dimensionality reduced comprehensive corrosion feature.

5. The corrosion warning system for a fixed photovoltaic support according to claim 1, characterized in that: The feature processing module includes: The corrosion distribution map generating unit is used to convert the discrete corrosion detection points into a continuous corrosion distribution map based on the corrosion degree value; The spatiotemporal corrosion model unit is used to construct a spatiotemporal corrosion model by combining historical corrosion data and the corrosion degree value; The dynamic heat map generation unit is used to generate a corrosion heat map according to the corrosion degree and radiation trend.

6. A corrosion warning method for a fixed photovoltaic support, using the corrosion warning system for a fixed photovoltaic support according to any one of claims 1 to 5, characterized in that: include: Obtain multimodal data of photovoltaic brackets; Based on the characteristics of multimodal data, the comprehensive corrosion characteristics are determined, and after weighting the comprehensive corrosion characteristics, dimensionality reduction is performed according to the principal component analysis method to obtain the comprehensive corrosion characteristics after dimensionality reduction; Under the condition of low quality of enhanced comprehensive corrosion characteristics, the adjustment method is determined according to the difference between the scoring threshold and the score; Generate corrosion distribution maps based on enhanced comprehensive corrosion features, build spatiotemporal corrosion models based on historical corrosion data and corrosion severity values, and generate corrosion thermodynamic maps based on corrosion severity and radiation trends. The scope of the early warning area is determined based on the corrosion distribution map and the corrosion thermodynamic map, and the adjustment method is determined based on the unqualified comparison result with the predicted early warning area.

Images