Photovoltaic module hot spot detection system and method

Abstract

The invention relates to the technical field of photovoltaic module detection, in particular to a hot spot detection system and method for a photovoltaic module. Comprises: a multispectral imaging module configured to synchronously collect image data of a photovoltaic module in a visible light band, a near-infrared band and a long-wave infrared band; the electroluminescence analysis module is in signal connection with the multispectral imaging module and is used for acquiring carrier recombination radiation data of the photovoltaic module in a reverse bias state; the dynamic threshold value processing module is configured to establish a dynamic correlation model of the environment parameters and the hot spot characteristic parameters; the three-dimensional reconstruction module is configured to perform spatial registration on the multispectral imaging data and the electroluminescence data and construct three-dimensional temperature field distribution of the photovoltaic module; the defect positioning module is configured to determine the accurate position of a hot spot according to temperature field distribution and carrier recombination characteristics; according to the scheme, a novel detection system architecture with multi-mode collaboration, high dynamic adaptability and three-dimensional analysis capability is constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic module detection, and in particular to a photovoltaic module hot spot detection system and method. Background Art

[0002] Hot spot effects in photovoltaic modules are a core issue leading to power generation efficiency degradation and fire risks. Current hot spot detection technology primarily relies on infrared thermal imaging and electroluminescence imaging, either singly or in combination. However, significant system architecture flaws exist. First, the ability to collaboratively collect multi-source data is insufficient. Due to asynchronous triggering mechanisms, visible light, infrared, and electroluminescence devices exhibit time deviations exceeding 50ms across modal data, creating motion artifacts in dynamic detection scenarios. Second, traditional detection systems employ fixed temperature thresholds and fail to account for environmental variables such as irradiance transients and module tilt. This results in a 22% difference in false positive rates between deserts and high-humidity regions. Furthermore, existing technologies are largely limited to two-dimensional planar analysis and are unable to characterize the three-dimensional diffusion paths of hot spots within the EVA film-cell-backsheet structure, resulting in over 20% missed early detection of latent cracks. Furthermore, the lack of a model for the spatiotemporal correlation between carrier recombination characteristics and temperature fields results in a false positive rate of up to 18% for local obstructions such as bird contamination. Although the patent with announcement number CN114494242B attempts to introduce temperature gradient analysis, it does not solve the core problem of multi-physics field data fusion, especially in key indicators such as three-dimensional positioning accuracy and environmental adaptation mechanism, there are still technical bottlenecks.

[0003] Therefore, based on the above problems, there is an urgent need to build a new detection system architecture that is multimodal collaborative, dynamic adaptable and has three-dimensional analysis capabilities. Summary of the Invention

[0004] In view of the fact that there are still technical bottlenecks in the key indicators of photovoltaic module hot spot detection such as three-dimensional positioning accuracy and environmental adaptive mechanism, this application provides a photovoltaic module hot spot detection system and method to solve the above problems.

[0005] To achieve the above object, the present invention is implemented through the following technical solutions:

[0006] The present application discloses a photovoltaic module hot spot detection system, comprising:

[0007] Multispectral imaging module: configured to simultaneously collect image data of photovoltaic modules in the visible light band, near infrared band and long-wave infrared band;

[0008] Electroluminescence analysis module: connected to the multispectral imaging module signal, used to obtain the carrier recombination radiation data of the photovoltaic module under reverse bias state;

[0009] Dynamic threshold processing module: configured to establish a dynamic correlation model between environmental parameters and hot spot characteristic parameters;

[0010] 3D reconstruction module: configured to spatially register the multispectral imaging data with the electroluminescence data to construct the 3D temperature field distribution of the photovoltaic module;

[0011] Defect location module: configured to determine the precise location of the hot spot based on the temperature field distribution and carrier recombination characteristics;

[0012] The defect location module is configured with an abnormality determination algorithm based on multimodal data fusion, and the environmental parameters include at least ambient temperature, radiation intensity and wind speed.

[0013] Adopting the above technical solution: This solution establishes a three-dimensional temperature field distribution through a three-dimensional reconstruction module, and locates the corresponding hot spot through a defect location module, which can solve the defects of traditional hot spot detection systems such as asynchronous multi-source data collection, poor dynamic adaptability of environmental parameters, and insufficient three-dimensional positioning accuracy.

[0014] Preferably, the abnormality determination algorithm includes a hot spot probability calculation model, and the calculation formula is:

[0015]

[0016] Where, P h is the probability coefficient of hot spot occurrence, is the temperature gradient vector modulus, E el is the electroluminescence intensity, α and β are weight coefficients, ranging from 0.3-0.7 and 0.2-0.5; x is the spatial coordinate along the surface of the photovoltaic module.

[0017] Adopting the above technical solution: This solution, by constructing a dual-parameter weighted model, can solve the problems of abnormal temperature gradient caused by the heat dissipation effect at the edge of the component, the failure to quantify the change in local carrier recombination rate caused by hidden crack defects, and missed detection caused by decreased infrared detection sensitivity in rainy weather caused by the single-parameter judgment method used in the existing technology.

[0018] Further preferably, the dynamic association model includes a dynamic threshold adjustment formula:

[0019]

[0020] Where: T th is the dynamic temperature threshold, T base is the reference temperature threshold, G is the real-time irradiance, G0 is the irradiance under standard conditions, γ is the irradiance influencing factor, λ is the aging attenuation coefficient, and t is the service life of the component.

[0021] Adopting the above technical solution: This solution eliminates the influence of environmental factors such as transient temperature fluctuations caused by morning dew evaporation, abnormal irradiance changes in sandstorm weather, and material thermal conductivity attenuation caused by component aging by establishing a dynamic environmental coupling model.

[0022] Further preferably, the 3D reconstruction module adopts an improved Delaunay triangulation algorithm, and its mesh optimization criterion is:

[0023]

[0024] Where, ΔT i is the difference between the temperature of the ith node and the average temperature of the adjacent nodes; σ(E i ) is the regional variance value of the electroluminescence intensity; w1 and w2 are weight factors, satisfying w1+w2=1, and w1>0.6.

[0025] Adopting the above technical solution: This solution can solve the problems of waste of computing power, strong noise sensitivity and spatial mapping deviation between electroluminescence data and temperature field caused by excessive mesh subdivision in traditional three-dimensional reconstruction by establishing an improved Delaunay triangulation algorithm.

[0026] A method, applied to a photovoltaic module hot spot detection system as described in any one of the above, comprising:

[0027] S1: Synchronously collect multispectral image data and electroluminescence data under standard test conditions;

[0028] S2: Analyze the surface contamination of the visible light image and establish an optical transmittance correction matrix;

[0029] S3: Perform spatiotemporal registration of infrared thermal imaging data and electroluminescence data to construct a three-dimensional temperature field;

[0030] S4: Determine the hot spot area based on the dynamic threshold model and temperature field gradient characteristics;

[0031] S5: Classify the hot spots based on the carrier recombination characteristics and generate a test report.

[0032] Adopting the above technical solution: This solution judges the hot spot area and generates the corresponding detection report through the design of the above-mentioned corresponding three-dimensional temperature field, which can solve the problems of poor multi-source data coordination, insufficient environmental interference compensation and low intelligence level of the detection process in the traditional hot spot detection method.

[0033] Further preferably, the establishment of the optical transmittance correction matrix in S2 includes:

[0034] Perform bicubic interpolation super-resolution reconstruction on the visible light image, increase the image resolution to 0.2mm / pixel, and then use the Otsu algorithm with adaptive threshold to segment the cell grid line structure;

[0035] Based on the preprocessing to eliminate the influence of grid line shadows, a pollution assessment model with the saturation value of the HSV color space as the characteristic quantity is constructed. The following is performed for each 5mm×5mm cell:

[0036] S21: Calculate the pixel saturation variance within the unit. When the variance value is less than 15, it is determined to be a uniformly polluted area.

[0037] S22: Use directional gradient histogram to analyze the texture of surface pollutants and identify vertical streaks of dust and randomly distributed stains;

[0038] S23: Match the preset transmittance attenuation coefficient according to the pollutant type, where the attenuation coefficient for the dust accumulation area is 0.85-0.92, and the attenuation coefficient for the oil pollution area is 0.65-0.78.

[0039] Further preferably, the spatiotemporal registration adopts a feature point matching algorithm, specifically including:

[0040] Construct a bidirectional feature mapping relationship between infrared images and electroluminescence images, including:

[0041] S31: On the infrared image side, when extracting the SURF feature descriptor of the hot spot area, the Hessian matrix threshold is set to 1200 to enhance the small-scale feature detection capability;

[0042] S32: On the electroluminescent image side, an improved ORB feature detection algorithm is used, and the number of pyramid layers is adjusted to 4 to balance feature density and matching efficiency;

[0043] S33: When establishing the correspondence between feature points, a bidirectional optical flow constraint mechanism is introduced to automatically remove mismatched points when the forward or backward projection error of a feature point exceeds 1.2 pixels.

[0044] S34: A nonlinear registration algorithm based on thin plate spline function is used to achieve a registration accuracy of 0.3 pixels while maintaining the edge topology of the battery cell.

[0045] Further preferably, the three-dimensional temperature field construction includes environmental interference compensation processing, and the specific compensation method is:

[0046] The measured temperature value is superimposed on the wind speed effect item and the irradiance difference item. The wind speed effect item is compensated by 0.12 times the temperature difference value for every increase of 1m / s wind speed, and the irradiance difference item is compensated by every deviation of 100W / m from the standard irradiance. 2 Compensate for 0.08 times the temperature rise.

[0047] Further preferably, the hot spot level classification adopts a multi-dimensional decision fusion mechanism, specifically including:

[0048] An evaluation system consisting of 12 characteristic parameters was constructed, with the core parameters processed as follows:

[0049] S51: When calculating the area ratio of the temperature anomaly region, an improved active contour model is used to accurately segment the hot spot boundary and perform polygonal approximation processing on irregular shape areas;

[0050] S52: During the maximum temperature gradient value extraction process, a gradient detection path is set along the main grid line direction of the cell, and a gradient distribution curve with a 0.1 mm interval is obtained using a cubic spline interpolation method;

[0051] S53: When evaluating the degree of carrier recombination rate anomaly, a correlation model with the component operating voltage is established, and the anti-interference analysis mode is enabled for data segments where the bias voltage fluctuation exceeds ±5%;

[0052] S54: Spatial coincidence analysis uses the Hausdorff distance algorithm to calculate the shortest distance distribution between the contaminated area and the hot spot area boundary line. When more than 90% of the boundary distances are less than 2 mm, it is determined to be a strongly associated area.

[0053] Further preferably, the test report generation includes an intelligent decision support function, which is specifically implemented as follows:

[0054] Build an operation and maintenance decision engine based on knowledge graph, which includes:

[0055] The three-dimensional coordinate information is displayed in a dual mode of local coordinate system and geodetic coordinate system. The origin of the local coordinate system is set at the center of the bolt hole in the lower left corner of the component.

[0056] The hot spot development stage prediction uses a time series analysis algorithm, combining historical temperature rise data and environmental corrosion index to calculate the risk evolution trend for the next 30 days;

[0057] The recommended maintenance solution includes a three-level response mechanism: a first-level warning triggers a drone re-inspection, a second-level warning activates an automatic cleaning device, and a third-level warning links to the power grid dispatching system;

[0058] The historical data comparative analysis uses a dynamic time warping algorithm to support pattern matching and abnormal evolution visualization of cross-year detection data;

[0059] The report output format is compatible with the BIM modeling system, and the three-dimensional data of the hot spot area can be directly imported into the digital twin platform of the photovoltaic power station. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0061] Figure 1 This is a block diagram of the photovoltaic module hot spot detection system for this application;

[0062] Figure 2 This is a flow chart of the photovoltaic module hot spot detection method for this application;

[0063] Figure 3 For this application Figure 2 Detailed flow chart of step S2;

[0064] Figure 4 For this application Figure 2 Detailed flow chart of step S3;

[0065] Figure 5 For this application Figure 2 Detailed flow chart of step S5. DETAILED DESCRIPTION

[0066] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0067] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, operations, elements, components and / or groups thereof.

[0068] See also Figures 1 to 5 , for example, traditional hot spot detection systems have defects such as asynchronous multi-source data collection, poor dynamic adaptability of environmental parameters, and insufficient three-dimensional positioning accuracy. Specifically, the visible light, infrared, and electroluminescence data acquisition devices operate independently, and the inconsistent time and space benchmarks lead to errors in data fusion; the fixed threshold detection model cannot adapt to environmental fluctuations such as temperature differences between morning and evening and seasonal changes; two-dimensional plane detection cannot accurately reflect the three-dimensional diffusion path of hot spots inside the component; the lack of correlation analysis between carrier recombination characteristics and temperature field data leads to a high misjudgment rate. Based on the above problems, the embodiment of the present application discloses a photovoltaic component hot spot detection system, including:

[0069] Multispectral imaging module: configured to simultaneously collect image data of photovoltaic modules in the visible light band, near infrared band and long-wave infrared band;

[0070] Electroluminescence analysis module: connected to the multispectral imaging module signal, used to obtain the carrier recombination radiation data of the photovoltaic module under reverse bias state;

[0071] Dynamic threshold processing module: configured to establish a dynamic correlation model between environmental parameters and hot spot characteristic parameters;

[0072] 3D reconstruction module: configured to spatially register the multispectral imaging data with the electroluminescence data to construct the 3D temperature field distribution of the photovoltaic module;

[0073] Defect location module: configured to determine the precise location of the hot spot based on the temperature field distribution and carrier recombination characteristics;

[0074] The defect location module is configured with an abnormality determination algorithm based on multimodal data fusion, and the environmental parameters include at least ambient temperature, radiation intensity and wind speed.

[0075] It is worth mentioning that this solution can achieve the following technical effects through multi-module collaborative innovation:

[0076] Multi-spectral synchronous acquisition: The three-band imaging equipment of visible light (400-700nm), near-infrared (700-1100nm) and long-wave infrared (8-14μm) adopts a hardware-triggered synchronization mechanism, which can achieve a time synchronization error of <1ms and a spatial registration accuracy of 0.5pixel.

[0077] Dynamic threshold modeling: Integrating the three-dimensional parameter space of temperature, irradiance, and wind speed can shorten the threshold adjustment response time.

[0078] Three-dimensional temperature field reconstruction: A hierarchical modeling algorithm based on Delaunay triangulation achieves 0.1mm spatial resolution on a 1m×2m standard component, which is 5 times more accurate than traditional two-dimensional detection and positioning.

[0079] Multimodal data fusion: Joint analysis of electroluminescence intensity and temperature gradient reduces the false alarm rate from the industry average of 15% to below 3%.

[0080] System compatibility expansion: supports dynamic detection of dual-axis tracking bracket components, with a maximum adaptable angular velocity of 5° / s.

[0081] For example, the single parameter determination method in the existing technology is prone to serious misjudgment. Specifically, the heat dissipation effect at the edge of the component leads to abnormal temperature gradients. The change in local carrier recombination rate caused by the micro-crack defect has not been quantified and analyzed. The infrared detection sensitivity decreases in rainy weather, resulting in missed detections. Based on the above problems, the design of this embodiment includes an abnormality determination algorithm including a hot spot probability calculation model, and the calculation formula is:

[0082]

[0083] Where, P h is the probability coefficient of hot spot occurrence, is the temperature gradient vector modulus, E el is the electroluminescence intensity, α and β are weight coefficients, ranging from 0.3-0.7 and 0.2-0.5; x is the spatial coordinate along the surface of the photovoltaic module.

[0084] In the above formula Characterizes the degree of localized sudden temperature changes on the surface of photovoltaic modules. Traditional methods focus only on absolute temperature thresholds. However, in practice, the heat dissipation effect at the edge of the module will produce a natural temperature gradient, which can easily lead to misjudgment.

[0085] The specific calculation method is: use the Sobel operator to calculate the anisotropic gradient to avoid the sensitivity of conventional gradient calculation to edge noise. The temperature matrix of the 3×3 neighborhood is convolved by the weighted average method. The calculation formula is:

[0086]

[0087] This design can effectively distinguish between real hot spots and edge heat dissipation.

[0088] In the above formula, the electroluminescence intensity derivative It can quantify the spatial variation characteristics of carrier recombination rate and avoid the situation where cracks or contamination areas hinder carrier transport under reverse bias, resulting in sudden changes in local electroluminescence intensity.

[0089] The specific calculation method is: calculate the first-order derivative along the main grid line direction of the cell, that is, the X-axis direction, to avoid cross interference of the secondary grid lines. Use the five-point central difference method to improve the calculation accuracy:

[0090]

[0091] It can be achieved that under low irradiance conditions, when the electroluminescence signal intensity drops by 80%, the derivative parameter can still maintain sensitivity, thus making up for the shortcomings of infrared detection.

[0092] The weight coefficients α and β adopt a dynamic adjustment mechanism, according to the ambient temperature T env Automatically adjust the weight distribution, for example, when T envWhen the temperature is >40℃, the electroluminescence signal is increasingly affected by thermal noise, and the β weight decreases to 0.2;

[0093] When T env When the temperature is less than 0℃, the noise of the infrared thermal imager increases and the α weight is increased to 0.7.

[0094] It is worth mentioning that this embodiment innovatively constructs a dual-parameter weighted model, and the temperature gradient vector modulus The design of the anisotropic temperature distribution is calculated by the Sobel operator, which can effectively distinguish the real hot spot from the edge heat dissipation area; the spatial derivative of the electroluminescence intensity The design can realize the quantitative characteristics of the sudden change of carrier recombination rate. 2 The detection accuracy rate can still maintain over 85% under low-light conditions; the α and β coefficients are automatically adjusted according to the ambient temperature, which can keep the model stable within the working range of -20℃ to 60℃; in addition, the design can achieve a misjudgment suppression rate of up to 92% for local obstructions such as bird droppings.

[0095] For example, the traditional dynamic threshold adopts a fixed threshold, which may cause failure in scenarios such as transient temperature fluctuations caused by morning dew evaporation, abnormal irradiance changes in sandstorm weather, and material thermal conductivity attenuation caused by component aging. Based on the above problems, this embodiment includes the dynamic association model including the dynamic threshold adjustment formula as follows:

[0096]

[0097] Where: T th is the dynamic temperature threshold, T base is the reference temperature threshold, G is the real-time irradiance, G0 is the irradiance under standard conditions, γ is the irradiance influencing factor, λ is the aging attenuation coefficient, and t is the service life of the component.

[0098] Under standard test conditions (STC), T base Associated with component power rating, for example:

[0099]

[0100] Among them, P actual The above design can avoid the threshold missetting caused by power attenuation.

[0101] Irradiance compensation item In the equation, G is the real-time irradiance, which is sampled at 10 Hz by a silicon photocell sensor; γ is the nonlinear compensation factor, which ranges from 0.1 to 0.3 and is dynamically adjusted according to the component inclination angle (θ).

[0102] The above formula solves the problem that the fixed threshold cannot adapt to dynamic changes in the environment and component aging through the coupling model of irradiance compensation and aging attenuation;

[0103] In the above formula, the reference temperature threshold T base Based on the material properties of photovoltaic modules, such as the glass transition temperature of EVA film, its value range is 320-350K, which can cover the critical failure temperature of crystalline silicon and thin-film modules.

[0104] γ=0.2·cosθ,θ<45°

[0105] This design can shorten the adjustment lag time to 2 seconds, which is 5 times faster than the traditional linear model.

[0106] Aging attenuation term e -λ·t It can quantify the degradation of heat dissipation performance caused by yellowing of packaging materials. The λ value is determined by the Arrhenius accelerated aging formula:

[0107]

[0108] Among them, A=0.15 is the pre-exponential factor, E a =0.8eV is the activation energy, and k is the Boltzmann constant.

[0109] It is worth mentioning that the irradiance compensation term in this embodiment Able to correct the irradiance fluctuation caused by cloud cover in real time, in G=200-1500W / m 2 Keep the threshold reasonable within the range; aging attenuation factor e -λ·t By introducing the Arrhenius accelerated aging model, the effect of yellowing of EVA film on heat dissipation is accurately reflected.

[0110] Traditional 3D reconstruction suffers from the following technical issues: excessive mesh subdivision leading to wasted computing power, high noise sensitivity, and significant deviations between the spatial mapping of electroluminescence data and the temperature field. To address these issues, the 3D reconstruction module in this embodiment uses an improved Delaunay triangulation algorithm, with the following mesh optimization criteria:

[0111]

[0112] Where, ΔT i is the difference between the temperature of the ith node and the average temperature of the adjacent nodes; σ(E i ) is the regional variance value of the electroluminescence intensity; w1 and w2 are weight factors, satisfying w1+w2=1, and w1>0.6.

[0113] In the above formula, the temperature smoothing term w1·ΔT i 2Used to suppress temperature changes caused by non-uniform noise of thermal imager. ΔT i It is defined as the difference between the node temperature and the average temperature of the neighborhood. The calculation formula is:

[0114]

[0115] Wherein, N=8 represents 8 critical points.

[0116] w1>0.6 ensures that the temperature continuity constraint dominates.

[0117] Electroluminescence variance term w2·σ(E i ), through the regional electroluminescence intensity variance (σ(E i ))Identify hidden crack defects. Calculation formula:

[0118]

[0119] Among them, M=25 is a 5×5 pixel area, μ E It is the regional mean. When the width of the hidden crack reaches 0.2mm, the variance value exceeds 3 times that of the normal area, realizing sub-millimeter crack detection.

[0120] The above formula adjusts the weight in real time according to the infrared image signal-to-noise ratio (SNR):

[0121] When SNR is less than 30dB, increase w1 to 0.9 to enhance noise reduction; when SNR is greater than 50dB, reduce w1 to 0.7 to preserve details.

[0122] It is worth mentioning that this scheme has made corresponding innovations to the grid optimization criterion, w1·ΔT i 2 It can constrain the temperature continuity, suppress the local mutation caused by the noise of the thermal imager, and improve the smoothness of the temperature field; and w2·σ(E i ) design identifies hidden crack defects through regional variance analysis, with the minimum detectable crack width reaching 0.2mm; this solution can compress the reconstruction time of a 6×12 cell module to within 90s while maintaining an accuracy of 0.1mm, and ensure the consistency of multiple physical fields, achieving a spatial registration error of the temperature field and the electroluminescence field of less than 0.05mm.

[0123] For example, traditional hot spot detection methods have defects such as poor coordination of multi-source data, insufficient compensation for environmental interference, and low intelligence of the detection process. Specifically,

[0124] Data collection is asynchronous: Independent triggering of visible light, infrared, and electroluminescent devices results in time deviations exceeding 50ms, creating motion artifacts in dynamic detection scenarios.

[0125] Insufficient elimination of optical interference: Reflections from the glass surface of the component cause the signal-to-noise ratio of the visible light image to drop by more than 30%, making it impossible to accurately identify the contaminated area;

[0126] Temperature field modeling error: The non-uniformity error of the infrared thermal imager is greater than ±2K, and the effect of the component tilt on heat dissipation is not considered;

[0127] Rigid threshold settings: Fixed temperature change rate thresholds cannot adapt to different installation environments;

[0128] High dependence on manual labor: test reports require manual review.

[0129] Based on the above problems, the present application provides a method, which is applied to a photovoltaic module hot spot detection system as described in any one of the above, comprising:

[0130] S1: Synchronously acquire multispectral image data and electroluminescence data under standard test conditions. The multispectral synchronization trigger mechanism in this design can achieve hardware-level synchronization signal control of three-band devices, with a final time deviation of <1ms and the elimination of motion artifacts.

[0131] S2: Analyze surface contamination of visible light images and establish an optical transmittance correction matrix. By using a multi-scale Gaussian filter to suppress specular reflection, the effective information content of visible light images is increased by 40%.

[0132] S3: Perform spatiotemporal registration of infrared thermal imaging data with electroluminescence data to construct a three-dimensional temperature field; use a cross-modal registration algorithm to unify the spatial resolution of thermal imaging and electroluminescence data to 0.5 mm / pixel, with a modeling error of <0.3 K.

[0133] S4: Determine the hot spot area based on the dynamic threshold model and temperature field gradient characteristics; this solution can improve the detection accuracy in desert areas from 78% to 95% by combining the tilt sensor data with real-time correction of the temperature gradient threshold.

[0134] S5: Hot spot classification is performed based on carrier recombination characteristics and a test report is generated. The support vector machine classifier of this solution can achieve millisecond-level hot spot classification, and the test report generation time is compressed to within 3 seconds.

[0135] For example, traditional surface contamination detection methods have the following technical problems:

[0136] Conventional image processing technology cannot identify fine pollution of 1mm due to insufficient resolution, such as the adhesion of pollen or dust particles.

[0137] The metal grid lines of the battery cells form periodic noise in the image, and the misjudgment rate of pollution identification is as high as 25%.

[0138] Failure to distinguish between cleanable dust pollution and difficult-to-clean corrosion leads to inappropriate maintenance strategies.

[0139] Traditional methods correct the entire component and ignore local contamination differences.

[0140] Based on the above problems, the present application further defines the optical transmittance correction matrix in step S2. The establishment of the optical transmittance correction matrix in step S2 includes:

[0141] Perform bicubic interpolation super-resolution reconstruction on the visible light image, increase the image resolution to 0.2mm / pixel, and then use the Otsu algorithm with adaptive threshold to segment the cell grid line structure;

[0142] Based on the preprocessing to eliminate the influence of grid line shadows, a pollution assessment model with the saturation value of the HSV color space as the characteristic quantity is constructed. The following is performed for each 5mm×5mm cell:

[0143] S21: Calculate the pixel saturation variance within the unit. When the variance value is less than 15, it is determined to be a uniformly polluted area.

[0144] S22: Use directional gradient histogram to analyze the texture of surface pollutants and identify vertical streaks of dust and randomly distributed stains;

[0145] S23: Match the preset transmittance attenuation coefficient according to the pollutant type, where the attenuation coefficient for the dust accumulation area is 0.85-0.92, and the attenuation coefficient for the oil pollution area is 0.65-0.78.

[0146] The design of the above technical solution has the following technical effects:

[0147] Bicubic interpolation super-resolution reconstruction can increase image resolution to 0.2mm / pixel and detect pollutants larger than 0.3mm;

[0148] The grid line segmentation technology uses the improved Otsu algorithm to accurately extract the grid line structure and eliminate shadow interference;

[0149] It can classify multi-feature pollution, such as HSV saturation variance analysis to identify uniform pollution such as sand and dust and non-uniform pollution such as bird droppings.

[0150] The histogram of directional gradients distinguishes longitudinal dust accumulation consistent with the grid line direction from random oil pollution.

[0151] The refined correction matrix design of 5mm×5mm unit-level correction can achieve a local transmittance calculation error of less than 2%, which is 5 times more accurate than traditional methods.

[0152] The difference in attenuation coefficients between dust and oil stains guides differentiated cleaning strategies, which can reduce the number of ineffective cleaning times by 50%.

[0153] For example, traditional multimodal data registration has the following technical problems:

[0154] Differences in feature scales: For example, the hot spot area of infrared images with a spatial resolution greater than 3mm is relatively blurred, which does not match the fine structure of electroluminescent images with a resolution of 0.5mm.

[0155] The traditional SIFT algorithm has a high mismatch rate in low-contrast areas, nearly exceeding 30%;

[0156] For example, flexible photovoltaic modules are prone to micro-bending, which will cause edge misalignment after alignment and require corresponding deformation compensation;

[0157] Conventional registration algorithms take >500ms to process a single frame of data, which cannot meet the needs of dynamic detection. Based on the above problems, the spatiotemporal registration adopts a feature point matching algorithm, which specifically includes:

[0158] Construct a bidirectional feature mapping relationship between infrared images and electroluminescence images, where:

[0159] S31: On the infrared image side, when extracting the SURF feature descriptor of the hot spot area, the Hessian matrix threshold is set to 1200 to enhance the small-scale feature detection capability;

[0160] S32: On the electroluminescent image side, an improved ORB feature detection algorithm is used, and the number of pyramid layers is adjusted to 4 to balance feature density and matching efficiency;

[0161] S33: When establishing the correspondence between feature points, a bidirectional optical flow constraint mechanism is introduced to automatically remove mismatched points when the forward or backward projection error of a feature point exceeds 1.2 pixels.

[0162] S34: A nonlinear registration algorithm based on thin plate spline function is used to achieve a registration accuracy of 0.3 pixels while maintaining the edge topology of the battery cell.

[0163] It is worth mentioning that this solution can achieve corresponding breakthroughs through multi-resolution feature matching, such as cross-scale feature detection:

[0164] Bidirectional optical flow verification can eliminate more than 85% of mismatched points by implementing forward or backward projection error constraints;

[0165] In the above embodiment, the SURF algorithm optimizes the Hessian threshold to 1200, extracting multi-scale hotspot features from 0.5 to 5 mm, enabling cross-scale feature detection. Furthermore, the ORB pyramid level control balances the details of the electroluminescent image.

[0166] By introducing a bidirectional optical flow constraint mechanism, it is possible to eliminate more than 85% of mismatched points by using forward or backward projection error constraints.

[0167] The nonlinear registration algorithm based on thin plate spline function can achieve non-rigid deformation compensation, and the registration error is less than 0.3 pixels within 5° of component bending.

[0168] In natural environments, factors such as wind speed and irradiance can affect the detection of ambient temperature. Experimental data show that a wind speed fluctuation of 2m / s can cause the temperature measurement value to drift by ±1.5K; and cloud cover can cause instantaneous irradiance changes of >500W / m 2 , the traditional linear compensation model may fail.

[0169] The three-dimensional temperature field construction includes environmental interference compensation processing, and the specific compensation method is:

[0170] The measured temperature value is superimposed on the wind speed effect item and the irradiance difference item. The wind speed effect item is compensated by 0.12 times the temperature difference value for every increase of 1m / s wind speed, and the irradiance difference item is compensated by every deviation of 100W / m from the standard irradiance. 2 Compensate for 0.08 times the temperature rise.

[0171] It is worth mentioning that this scheme can realize frequency domain adaptive filtering by establishing a multi-physical field coupling compensation system, and can effectively suppress high-frequency disturbances by dynamically adjusting the sliding average window length with the wind speed sampling rate.

[0172] In addition, a segmented compensation strategy can be adopted:

[0173] Less than 800W / m 2 A logarithmic compensation curve is used in the low irradiation section to match the nonlinear temperature rise characteristics under weak light; an exponential compensation curve is used in the low irradiation section and the high irradiation section to adapt to the oversaturation temperature rise trend.

[0174] The comprehensive compensation model can reduce the temperature error caused by environmental interference from ±2.5K to ±0.4K.

[0175] Traditional hot spot classification methods, such as those that manually define the boundaries of hot spot areas, can result in errors, ultimately affecting the accuracy of area calculations. Therefore, the hot spot classification method employs a multi-dimensional decision fusion mechanism, specifically including:

[0176] An evaluation system consisting of 12 characteristic parameters was constructed, with the core parameters processed as follows:

[0177] S51: When calculating the area ratio of the temperature anomaly region, an improved active contour model is used to accurately segment the hot spot boundary and perform polygonal approximation processing on irregular shape areas;

[0178] S52: During the maximum temperature gradient value extraction process, a gradient detection path is set along the main grid line direction of the cell, and a gradient distribution curve with a 0.1 mm interval is obtained using a cubic spline interpolation method;

[0179] S53: When evaluating the degree of carrier recombination rate anomaly, a correlation model with the component operating voltage is established, and the anti-interference analysis mode is enabled for data segments where the bias voltage fluctuation exceeds ±5%;

[0180] S54: Spatial coincidence analysis uses the Hausdorff distance algorithm to calculate the shortest distance distribution between the contaminated area and the hot spot area boundary line. When more than 90% of the boundary distances are less than 2 mm, it is determined to be a strongly associated area.

[0181] Adopting the above technical solution: the above solution adopts the improved Snake model, which can increase the iterative convergence speed by several times, and the boundary positioning accuracy reaches 0.1mm, reducing the error caused by manually demarcating the boundaries of the hot spot area, and the above solution can avoid the interference of the secondary grid line by extracting the gradient distribution along the main grid line direction, and can maintain detection stability under ±10% voltage fluctuations by establishing a bias-compound rate compensation curve; the Hausdorff distance algorithm is used in the algorithm to identify the spatial correlation between pollution and hot spots, which can reduce the false correlation rate.

[0182] For example, the traditional test report generation solution only provides the local coordinates of the components, which cannot be effectively connected with the power plant GIS system; it lacks historical data comparison and trend prediction functions; the manual decision response time is relatively delayed, and there is a possibility of being unable to deal with sudden hot spot failures. Based on this,

[0183] The test report generation includes intelligent decision support functions, which are specifically implemented as follows:

[0184] Build an operation and maintenance decision engine based on knowledge graph, which includes:

[0185] The three-dimensional coordinate information is displayed in a dual mode of local coordinate system and geodetic coordinate system. The origin of the local coordinate system is set at the center of the bolt hole in the lower left corner of the component.

[0186] The hot spot development stage prediction uses a time series analysis algorithm, combining historical temperature rise data and environmental corrosion index to calculate the risk evolution trend for the next 30 days;

[0187] The recommended maintenance solution includes a three-level response mechanism: a first-level warning triggers a drone re-inspection, a second-level warning activates an automatic cleaning device, and a third-level warning links to the power grid dispatching system;

[0188] The historical data comparative analysis uses a dynamic time warping algorithm to support pattern matching and abnormal evolution visualization of cross-year detection data;

[0189] The report output format is compatible with the BIM modeling system, and the three-dimensional data of the hot spot area can be directly imported into the digital twin platform of the photovoltaic power station.

[0190] It is worth mentioning that this solution, by establishing a coordinate mapping relationship, can realize the real-time conversion of local coordinates and geodetic coordinates, and can achieve precise positioning at the power plant level; the ARIMA algorithm combined with the corrosion index to predict the evolution of hot spots can increase the confidence level to greater than 85%; the formulation of a multi-level response mechanism can realize the effective and timely formulation of targeted treatment strategies based on the response situation.

[0191] Unless otherwise specified, the device components involved in the above embodiments are all conventional device components, and the connection methods and control methods involved are all conventional connection methods and control methods unless otherwise specified.

[0192] The present invention has been described in detail above with reference to the embodiments. However, those skilled in the art will appreciate that, without departing from the spirit of the present invention, the specific parameters in the above embodiments may be modified to form multiple specific embodiments, which are all within the common variation range of the present invention and will not be described in detail here.

Claims

1. A photovoltaic module hot spot detection system, characterized in that: include: Multispectral imaging module: configured to simultaneously collect image data of photovoltaic modules in the visible light band, near infrared band and long-wave infrared band; Electroluminescence analysis module: connected to the multispectral imaging module electrical signal, used to obtain the carrier recombination radiation data of the photovoltaic module under the reverse bias state; Dynamic threshold processing module: configured to establish a dynamic correlation model between environmental parameters and hot spot characteristic parameters; 3D reconstruction module: configured to spatially register the multispectral imaging data with the electroluminescence data to construct the 3D temperature field distribution of the photovoltaic module; Defect location module: configured to determine the precise location of the hot spot based on the temperature field distribution and carrier recombination characteristics; The defect location module is configured with an abnormality determination algorithm based on multimodal data fusion, and the environmental parameters include at least ambient temperature, radiation intensity and wind speed.

2. A photovoltaic module hot spot detection system according to claim 1, characterized in that: The abnormality determination algorithm includes a hot spot probability calculation model, and the calculation formula is: Where, P h is the probability coefficient of hot spot occurrence, is the temperature gradient vector modulus, E el is the electroluminescence intensity, α and β are weight coefficients, ranging from 0.3-0.7 and 0.2-0.5; x is the spatial coordinate along the surface of the photovoltaic module.

3. A photovoltaic module hot spot detection system according to claim 1, characterized in that: The dynamic association model includes a dynamic threshold adjustment formula: Where: T th is the dynamic temperature threshold, T base is the reference temperature threshold, G is the real-time irradiance, G0 is the irradiance under standard conditions, γ is the irradiance influencing factor, λ is the aging attenuation coefficient, and t is the service life of the component.

4. A photovoltaic module hot spot detection system according to claim 3, characterized in that: The 3D reconstruction module uses an improved Delaunay triangulation algorithm, and its mesh optimization criterion is: Where, ΔT i is the difference between the temperature of the ith node and the average temperature of the adjacent nodes; σ(E i ) is the regional variance of electroluminescence intensity; w1 and w2 are weight factors, satisfying w1+w2=1, and w1>0.

6.

5. A method, applied to a photovoltaic module hot spot detection system according to any one of claims 1 to 4, characterized in that: include: S1: Synchronously collect multispectral image data and electroluminescence data under standard test conditions; S2: Analyze the surface contamination of the visible light image and establish an optical transmittance correction matrix; S3: Perform spatiotemporal registration of infrared thermal imaging data and electroluminescence data to construct a three-dimensional temperature field; S4: Determine the hot spot area based on the dynamic threshold model and temperature field gradient characteristics; S5: Classify the hot spots based on the carrier recombination characteristics and generate a test report.

6. A method according to claim 5, characterized in that The establishment of the optical transmittance correction matrix in S2 includes: Perform bicubic interpolation super-resolution reconstruction on the visible light image, increase the image resolution to 0.2mm / pixel, and then use the Otsu algorithm with adaptive threshold to segment the cell grid line structure; Based on the preprocessing to eliminate the influence of grid line shadows, a pollution assessment model with the saturation value of the HSV color space as the characteristic quantity is constructed. The following is performed for each 5mm×5mm cell: S21: Calculate the pixel saturation variance within the unit. When the variance value is less than 15, it is determined to be a uniformly polluted area. S22: Use directional gradient histogram to analyze the texture of surface pollutants and identify vertical streaks of dust and randomly distributed stains; S23: Match the preset transmittance attenuation coefficient according to the pollutant type, where the attenuation coefficient for the dust accumulation area is 0.85-0.92, and the attenuation coefficient for the oil pollution area is 0.65-0.

78.

7. A method according to claim 6, characterized in that The spatiotemporal registration adopts a feature point matching algorithm, specifically including: Construct a bidirectional feature mapping relationship between infrared images and electroluminescence images, where: S31: On the infrared image side, when extracting the SURF feature descriptor of the hot spot area, the Hessian matrix threshold is set to 1200 to enhance the small-scale feature detection capability; S32: On the electroluminescent image side, an improved ORB feature detection algorithm is used, and the number of pyramid layers is adjusted to 4 to balance feature density and matching efficiency; S33: When establishing the correspondence between feature points, a bidirectional optical flow constraint mechanism is introduced to automatically remove mismatched points when the forward or backward projection error of a feature point exceeds 1.2 pixels. S34: A nonlinear registration algorithm based on thin plate spline function is used to achieve a registration accuracy of 0.3 pixels while maintaining the edge topology of the battery cell.

8. A method according to claim 7, characterized in that The three-dimensional temperature field construction includes an environmental interference compensation processing method, and the environmental interference compensation processing method includes: The measured temperature value is superimposed on the wind speed effect item and the irradiance difference item. The wind speed effect item is compensated by 0.12 times the temperature difference value for every increase of 1m / s wind speed, and the irradiance difference item is compensated by every deviation of 100W / m from the standard irradiance. 2 Compensate for 0.08 times the temperature rise.

9. A method according to claim 8, characterized in that The hot spot level classification adopts a multi-dimensional decision fusion mechanism, specifically including: An evaluation system consisting of 12 characteristic parameters was constructed, with the core parameters processed as follows: S51: When calculating the area ratio of the temperature anomaly region, an improved active contour model is used to accurately segment the hot spot boundary and perform polygonal approximation processing on irregular shape areas; S52: During the maximum temperature gradient value extraction process, a gradient detection path is set along the main grid line direction of the cell, and a gradient distribution curve with a 0.1 mm interval is obtained using a cubic spline interpolation method; S53: When evaluating the degree of carrier recombination rate anomaly, a correlation model with the component operating voltage is established, and the anti-interference analysis mode is enabled for data segments where the bias voltage fluctuation exceeds ±5%; S54: Spatial coincidence analysis uses the Hausdorff distance algorithm to calculate the shortest distance distribution between the contaminated area and the hot spot area boundary line. When more than 90% of the boundary distances are less than 2 mm, it is determined to be a strongly associated area.

10. A method according to claim 9, characterized in that The test report generation includes intelligent decision support functions, which are specifically implemented as follows: Build an operation and maintenance decision engine based on knowledge graph, including: The three-dimensional coordinate information is displayed in a dual mode of local coordinate system and geodetic coordinate system. The origin of the local coordinate system is set at the center of the bolt hole in the lower left corner of the component. The hot spot development stage prediction uses a time series analysis algorithm, combining historical temperature rise data and environmental corrosion index to calculate the risk evolution trend for the next 30 days; The recommended maintenance solution includes a three-level response mechanism: a first-level warning triggers a drone re-inspection, a second-level warning activates an automatic cleaning device, and a third-level warning links to the power grid dispatching system; The historical data comparative analysis uses a dynamic time warping algorithm to support pattern matching and abnormal evolution visualization of cross-year detection data; The report output format is compatible with the BIM modeling system, and the three-dimensional data of the hot spot area can be directly imported into the digital twin platform of the photovoltaic power station.

Images