Photovoltaic module fault image recognition system based on infrared thermal imaging analysis

Abstract

The invention provides a photovoltaic module fault image recognition system based on infrared thermal imaging analysis, and relates to the field of image recognition, comprising: an asynchronous image acquisition module calculates a response delay value based on component spatial distribution, a shielding state and a thermal capacity coefficient, and generates a dynamic scheduling path to control an infrared thermal imager to capture component images of different thermal response processes; the thermal anomaly trend analysis and recognition module constructs a time difference thermal response matrix, recognizes continuous multi-frame candidate temperature rise areas with the temperature rising continuously, and builds a temperature trend prediction model to judge thermal anomaly; the response behavior intervention and disturbance execution module calculates a disturbance influence factor, decides to apply disturbance to a target or an adjacent component and triggers image acquisition; and the routing inspection scheduling feedback updating module dynamically adjusts the priority and frequency of the acquisition path according to the thermal anomaly judgment result and the intervention data.

Description

Technical Field

[0001] This invention relates to the field of image recognition, specifically to a photovoltaic module fault image recognition system based on infrared thermal imaging analysis. Background Technology

[0002] Photovoltaic power generation, as an important form of clean energy, is developing rapidly. Its large-scale application places higher demands on operation and maintenance efficiency and safety. Thermal failures of photovoltaic modules, such as hot spots, are common hidden dangers that can lead to power loss or even fire risks. Infrared thermal imaging technology, due to its non-contact and visualization advantages, has become an effective means of monitoring the thermal state of modules. Accurately identifying thermal anomalies is crucial to ensuring the safe and economical operation of power plants.

[0003] Existing technical solutions often employ infrared thermal imagers to synchronously scan photovoltaic arrays along a predetermined path; acquire thermal image sequences of the modules; identify overheated areas by analyzing the temperature distribution of a single frame image or setting a fixed threshold; some solutions combine multiple frames of images for simple differential or temperature statistics; and determine whether there are abnormal heat points in the modules. These methods rely on a preset scanning sequence and fixed time intervals for image acquisition.

[0004] The shortcomings of the existing solution are as follows: due to differences in component characteristics and environmental changes, it is difficult to capture images of different components in comparable thermal response stages within the same time window during synchronous acquisition, which affects the accuracy of status judgment; the identification of thermal anomalies is mostly based on static thresholds or simple time series analysis; it is susceptible to instantaneous interference, leading to misjudgment and missed judgment; and the intervention decision lacks dynamic evaluation of the correlation between components and the feasibility of implementation. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides a photovoltaic module fault image recognition system based on infrared thermal imaging analysis. This system solves the problems mentioned in the background: differences in module characteristics and environmental changes make it difficult to capture images of different modules in comparable thermal response stages within the same time window, affecting the accuracy of status judgment; thermal anomaly identification is mostly based on static thresholds or simple time-series analysis; it is susceptible to instantaneous interference leading to misjudgments and omissions; and intervention decisions lack dynamic assessment of inter-module correlations and implementation feasibility.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solution: a photovoltaic module fault image recognition system based on infrared thermal imaging analysis, comprising an asynchronous image acquisition module, which calculates the response delay value based on the spatial distribution, shading state and structural thermal capacity coefficient of the photovoltaic module, and generates a dynamic image acquisition scheduling path accordingly, controls the infrared thermal imager to acquire images of modules in different thermal response processes within the same time window, and records the acquisition time and spatial coordinates of the modules.

[0009] The thermal anomaly trend analysis and identification module receives the component image and its time and space coordinates, calculates the time difference between consecutive image frames, constructs a time difference thermal response matrix, identifies candidate heating regions where the temperature continues to rise for more than three consecutive frames, establishes a temperature trend prediction model, predicts the rate of temperature change, and determines the thermal anomaly state.

[0010] The response behavior intervention and disturbance execution module receives the thermal anomaly status and component spatial coordinates, selects the target or nearby components to apply disturbance based on the disturbance influence factor, and triggers the asynchronous image acquisition module to acquire corresponding images within a set time period;

[0011] The inspection scheduling feedback update module receives thermal anomaly judgment results and intervention data, and dynamically adjusts the priority and acquisition frequency of the dynamic image acquisition scheduling path.

[0012] Preferably, the asynchronous image acquisition module first uses the spatial distribution characteristics of the photovoltaic modules as input information, and obtains the actual arrangement position of each module in the photovoltaic power station by constructing a two-dimensional matrix coordinate mapping model of the modules. Based on this, it further combines real-time acquired images of the shading status on-site, and classifies the shading status identifier of each module according to the illumination angle and the distribution pattern of the shadows cast by the shading objects. The shading status identifier is expressed using a hierarchical numerical system: level 0 for completely unshaded areas, level 1 for slightly shaded areas, level 2 for intermittently shaded areas, level 3 for continuously shaded areas, and so on, forming a hierarchical structure. The system identifies the shading level and, in addition, retrieves the structural material type data stored in the component structure database. Based on the structural material type and a preset heat capacity parameter table, it obtains the structural heat capacity coefficient, which uses heat capacity per unit area as the measurement basis. The system uses the component's spatial distribution location, shading status identifier, and structural heat capacity coefficient as joint inputs, which are then standardized and input into the response delay estimation function. The response delay estimation function calculates the response delay value for each component, in seconds, based on the time difference between the time required for the component to absorb external heat energy and the time required to release heat. This response delay value is used to determine the component's response time when receiving equal amounts of external heat energy. The image after thermal excitation reveals the time offset required for a discernible thermal response. After calculating the response delay values ​​of all components, they are sorted according to their magnitude, and a preset sorting priority rule is introduced. Components with response delay value differences exceeding 3 seconds are set as high-priority imaging objects. By constructing an acquisition sequence table based on the response delay value sorting results, combined with the maximum acquisition frame rate of the infrared thermal imager and the acquisition interval time per frame, a dynamic image acquisition scheduling path is formed, defining the absolute timestamp and acquisition sequence number of each component image acquisition. After the scheduling path is generated, the asynchronous image acquisition module strictly controls the infrared thermal imager. Image acquisition actions are executed according to the acquisition scheduling path to ensure that components in different thermal response processes are captured and imaged within the set acquisition time window, thereby achieving standardized comparison of the thermal states of different components at the same moment. During the acquisition process, the absolute time information of each image acquisition and the spatial coordinates of the currently focused component are also recorded in real time. The time information is timed in seconds, and the spatial coordinates are stored using two-dimensional positioning information of the station reference system. The above operations together ensure that the subsequent thermal anomaly identification module has an accurate time and space registration basis when constructing the time difference thermal response matrix, forming a highly reliable image data input structure.

[0013] Preferably, after receiving the photovoltaic module image data acquired by the asynchronous image acquisition module, the thermal anomaly trend analysis and identification module first sorts all images on the time axis according to the recorded acquisition time information and manages the images by frame. Then, it extracts the spatial coordinates of the components corresponding to the images and constructs a time difference matrix with component positions as indices and acquisition time differences between image frames as element values. The time difference matrix is ​​constructed by calculating the difference in acquisition time between two adjacent frames at the same spatial coordinates, in seconds. The time differences between all adjacent image frames are recorded and combined into matrix units. Simultaneously, pixel-level analysis is performed on the temperature distribution in each frame, mapping the image temperature data into a two-dimensional array structure, where each pixel contains a temperature value and a position index. For three or more consecutive images, difference judgment processing is performed. For any pixel, if the temperature value corresponding to it in the three frames continuously shows an upward trend, the location of the pixel is marked as a heating state and added to the preliminary list of candidate heating regions. At this time, a spatial adjacency merging rule is introduced. The process involves clustering and merging spatially continuous or adjacent heated pixels to form regional candidate heated region units, with the boundaries of these regions determined by temperature gradient descent curves. After extracting the candidate heated regions, temperature records at the same spatial coordinates in historical image frames are retrieved to form a temperature change sequence of the candidate heated regions at multiple time points, which is the temperature change trajectory. This trajectory records the thermal response development trend during the component's thermal evolution and serves as the core input for subsequent prediction model establishment. Each trajectory is arranged in frame time order and includes temperature values ​​and corresponding time labels. To ensure analysis accuracy, the temperature trajectories are smoothed, abnormal jump values ​​are removed, and the trajectories are resampled at a standard step size of 1 second to generate consistent input data. Finally, the processed candidate heated region list is stored one-to-one with the corresponding historical temperature change trajectories to form a candidate heated region database that can be further analyzed and identified. This database will serve as a key reference when establishing temperature trend prediction models, ensuring that the identified thermal anomaly regions exhibit the physical characteristics of real heated evolution.

[0014] Preferably, after extracting candidate heating areas and their historical temperature change trajectories, the thermal anomaly trend analysis and identification module proceeds to the temperature trend prediction model establishment stage. First, the temperature value sequence of the candidate heating areas in different image frames is used as input samples, and feature fusion is performed by combining it with ambient temperature data acquired synchronously with the image acquisition time. The ambient temperature data is provided by meteorological monitoring equipment deployed at the photovoltaic power station, recording three indicators: current light intensity, air temperature, and humidity. The air temperature is standardized to a sampling frequency matching the image acquisition time and used as one of the time series inputs to the model. Furthermore, to improve the predictive model's adaptability to individual component differences, background temperature data of the current components is introduced. The background temperature is defined as the stable operating temperature of the target component under no external interference, and is obtained by calculating the average value from the first twenty frames of images where no heating state was identified. The temperature change trajectory, synchronous ambient temperature sequence, and background temperature values ​​are uniformly mapped into a multi-dimensional feature tensor structure, which serves as the model training input. Subsequently, the embedded time-series prediction algorithm framework is invoked. A temperature trend prediction model is constructed using a unidirectional recurrent neural network. The input sequence length is set to 10 frames, and the corresponding predicted output length is set to the temperature change sequence within the next 5 seconds. During training, the model uses historical temperature change trajectories and known results as supervised learning targets, and optimizes the loss function to minimize prediction error. After completing model training and predicting the current candidate heating region, the predicted temperature sequence and the calculated temperature change rate are output. The temperature change rate is obtained by differencing the first and last temperature values ​​in the output sequence and dividing by the time interval, with the unit being degrees Celsius per second. If the temperature change rate is greater than 3 degrees Celsius per second, and the temperature of the candidate heating region continues to rise for more than 5 frames, the system determines this region to be in a thermal anomaly state and outputs its spatial coordinates along with the predicted temperature information. The determination criteria are designed to fully exclude misjudgments caused by short-term erroneous data collection, ensuring that each region determined to be in a thermal anomaly state has a clear and continuous temperature rise trend that exceeds the normal background change range, conforming to the typical performance characteristics of photovoltaic module fault thermal response.

[0015] Preferably, after receiving a candidate heating area identified as a thermal anomaly by the thermal anomaly trend analysis and identification module, the response behavior intervention and disturbance execution module first locates the target module based on the corresponding photovoltaic module spatial coordinates, and then enters the disturbance feasibility determination process. The core of the disturbance feasibility determination process is the calculation of the disturbance impact factor, which is determined by four specific parameters: the physical location parameter of the module, the connection structure parameter of adjacent modules, the historical number of successful interventions, and the remaining energy parameter of the current intervention equipment. Among them, the physical location parameter is obtained based on the row and column number of the module and its circuit topology position in the area. If the module is located in the center area of ​​the site, it is given a higher response priority value. The connection structure parameter of adjacent modules is queried through a pre-built module connection relationship matrix. The module connection relationship matrix reflects the structural coupling strength between any two modules. The higher the strength value, the more obvious the heat diffusion effect after the disturbance is applied. The historical number of successful interventions parameter records the number of times that the intervention operation on the module has successfully induced an effective thermal response. The credibility of the current intervention is evaluated by weighting the cumulative number of successful interventions. The remaining energy parameter of the intervention equipment is determined by the actual deployment location. The mobile disturbance device in the photovoltaic field collects data, the values ​​of which reflect the current power or operable count of the equipment, ensuring that the system does not execute disturbance actions beyond the equipment's capabilities. These four parameters are uniformly converted into standardized values ​​and then input into the disturbance influence factor function to generate the disturbance influence factor. If the disturbance influence factor value is greater than the set feasible threshold of 3.5, the system determines that the conditions for executing a disturbance are met. At this time, the control intervention device applies a disturbance operation to the target component, the disturbance form being a directional thermal excitation or a thermal response caused by short-term physical contact, with the disturbance execution time controlled within 1 second. Immediately afterwards, a command is sent to the asynchronous... The image acquisition module continuously acquires images of the target component from 5 seconds before the disturbance is applied to 10 seconds after the disturbance ends. The image acquisition frequency is one frame per second, for a total of 15 frames. All image acquisition times are strictly recorded with timestamps and associated with the disturbance time points to ensure accurate analysis of the disturbance response effect based on the timeline. Through the above method, the system realizes direct intervention testing of the thermal anomaly state of the target component, providing real-time response data under disturbance excitation for the thermal anomaly trend analysis module, which helps to further verify the accuracy and robustness of the preliminary judgment results.

[0016] Preferably, when the response behavior intervention and disturbance execution module determines that the disturbance impact factor is less than the set feasible threshold of 3.5 after completing the calculation of the disturbance impact factor, the system will not directly apply disturbance to the target component, but will instead enter the neighboring component disturbance substitution process. First, the component connection structure database is called to filter neighboring components that have a structural connection relationship with the target component and whose structural connection strength value is greater than 5. Based on this, the similarity parameter of illumination conditions is further analyzed by comparing the illumination intensity and occlusion status identifier received by the target component and neighboring components at the same time point, expressed as a percentage. If the similarity is greater than 85%, the illumination conditions are considered similar. Neighboring components that meet the above two conditions will be selected as disturbance targets. Then, the intervention device is controlled to apply a disturbance operation to the selected neighboring components. The disturbance method is the same as that of the target component, and the intervention time does not exceed 1 second. Three seconds after the disturbance is applied, i.e., the third set time, a collection command is sent to the asynchronous image acquisition module, ordering it to collect the infrared image data of the disturbed neighboring components. The collection operation is executed continuously for 5 frames, 1 frame per second, and the precise timestamp and component empty space are recorded during the collection. The image data is transmitted to the thermal anomaly trend analysis and recognition module for disturbance response simulation. Temperature change sequences are extracted from the disturbance response image, and a thermal response propagation path model is constructed by combining the structural connection strength between the target component and neighboring components, the similarity of illumination conditions, and the component's material information. The target component, as the inference object without direct disturbance, has its thermal response simulated and predicted using a mapping function. Based on the disturbance thermal response data of neighboring components, a reverse thermal propagation path reconstruction operation is performed to generate a potential temperature response curve of the target component under the current disturbance environment. The potential temperature response curve expresses the thermal diffusion result with the trend of temperature change over time. If the temperature change rate in the simulated and predicted temperature response curve exceeds a set risk threshold of 2 degrees Celsius per second, the target component is confirmed to be in a thermal anomaly state, and relevant judgment information is recorded. This achieves indirect inference and recognition of the thermal response state of the target component under conditions where direct intervention is not possible, through logical substitution based on structural thermal conduction and illumination condition similarity. This ensures that the entire recognition system has the capability to detect thermal anomalies and achieve intervention coverage under complex spatial distributions.

[0017] Preferably, when the thermal anomaly trend analysis and identification module receives image data from a neighboring component after disturbance, it first performs disturbance response feature extraction on the image data, including three aspects: First, disturbance diffusion speed calculation, which is based on the temperature increase of the same pixel area between image frames divided by the frame time difference, in degrees Celsius per second, to measure the speed change trend of thermal response after disturbance excitation; Second, thermal response expansion direction determination, which identifies the main direction and range of thermal diffusion by vectorizing the temperature difference change trend of adjacent pixel areas; Third, response amplitude boundary calculation, which extracts the area of ​​continuous pixel blocks in the response region whose temperature value exceeds the background temperature by more than 2 degrees Celsius, to determine the scale of the thermally affected area; The above three types of disturbance response feature information are used as disturbance feature vector input to the fitting modeling process; Further, the structural connection strength parameters, illumination condition similarity parameters, and component material parameters stored in the database are called and jointly input with the disturbance feature vector, and a disturbance response fitting function is generated by constructing a multi-factor linear fitting model; This function is used to describe the heat conduction relationship when the disturbance propagates from the neighboring component to the target component, and thereby generates a thermal response mapping relationship model. Next, without relying on the current image data of the target component, the system uses the aforementioned thermal conduction mapping relationship to project the disturbance response characteristics of neighboring components to the target component location, performing a trend path inversion operation. During this operation, a fitted potential temperature response curve is output to simulate the possible thermal response path of the target component under the same disturbance conditions. The system then analyzes the temperature change rate of the simulated curve. If the temperature increase rate exceeds 3 degrees Celsius per second and lasts for more than 5 seconds within any time interval of the curve, the system determines that the target component has a thermal anomaly and generates an alarm marker. After the asynchronous image acquisition module acquires the actual image of the target component, the thermal anomaly trend analysis and recognition module extracts the temperature change trajectory from the actual image and compares it with the previously simulated potential temperature response curve. It calculates the temperature deviation between the two at each time point and outputs the error value between the predicted and actual temperature values. All error values ​​are summarized to form disturbance response prediction residual data, serving as a basic indicator for evaluating the system's simulation accuracy. This residual data is further used to update the dynamic image acquisition scheduling path strategy, improving the overall prediction accuracy and response sensitivity of the recognition system.

[0018] Preferably, when the response behavior intervention and disturbance execution module applies a disturbance to the target component based on the disturbance impact factor being greater than the set feasible threshold of 3.5, the system synchronously controls the asynchronous image acquisition module to continuously acquire images of the target component at a frequency of one frame per second during the time interval from 5 seconds before the disturbance to 10 seconds after the disturbance. The obtained image data includes the state image before the disturbance and the response image after the disturbance, totaling 15 frames. After receiving the above image data, the thermal anomaly trend analysis and identification module first performs image frame difference analysis to extract the pixel temperature difference values ​​of the same spatial coordinate points in the images before and after the disturbance. The temperature difference calculation results are represented in the form of a two-dimensional difference map matrix, where the difference value of each pixel represents the temperature change amplitude during the time interval before and after the disturbance. Subsequently, based on the temperature change map matrix, three types of key disturbance response features are extracted: the total temperature change triggered by the disturbance, i.e., the average temperature change of all pixels with positive differences in the region; the hotspot expansion path, i.e., the path trajectory of the high-temperature pixel block formed in the image from the center to the edge, whose morphological features are obtained through connected component identification technology; and the area features of the affected area, i.e., the temperature... The area covered by a contiguous region of pixels with a temperature difference exceeding 2 degrees Celsius is measured in pixels. These three types of disturbance response features are combined into a disturbance response feature vector, which serves as the disturbance response expression information of the target component. The system internally stores a standard database of disturbance response feature vectors of fault-free components, constructed by collecting disturbance response samples of components under normal conditions over a long period of time. The feature vector format is consistent with the above. The similarity between the disturbance response feature vector of the current target component and the standard vector in the database is calculated using the Euclidean distance similarity analysis algorithm, outputting a similarity value between 0 and 1. If the similarity value is less than the set matching threshold of 0.75, the system confirms that the thermal response of the target component is significantly different from the fault-free state, determines that it has a thermal anomaly state, and outputs the spatial coordinates and the judgment level. The entire process realizes an anomaly detection method based on the comparison and analysis of thermal response data and standard models under disturbance intervention. It does not rely on a single temperature value for judgment, but rather on the difference between the behavior pattern of the thermal diffusion process and the historical normal pattern for intelligent identification, effectively improving the stability and accuracy of thermal anomaly judgment.

[0019] Preferably, after receiving the thermal anomaly status determination result output by the thermal anomaly trend analysis and identification module, the inspection scheduling feedback update module immediately calls the recorded target area spatial coordinates and its thermal anomaly level information as the first input parameter for scheduling feedback update. Next, it synchronously receives disturbance response prediction residual data, which is the difference between the predicted temperature value generated during the disturbance response simulation prediction process and the temperature value in the actual acquired image. The system converts the residual data into a disturbance response residual index matrix, where each unit represents the error amplitude of a component. First, the thermal anomaly status determination results are classified, marking components with anomalies as high-risk nodes. Then, the target area status information and the disturbance response residual index matrix are jointly analyzed. In the joint analysis, if a component is confirmed to have a thermal anomaly status, or its disturbance response prediction residual data exceeds a preset residual threshold of 1.5 degrees Celsius, or its similarity calculation result is lower than a set matching threshold of 0.75, the system determines it as a high-priority inspection object. The inspection scheduling feedback update module then updates the module based on the above determination results. The dynamic image acquisition scheduling path is updated by adjusting the acquisition priority of target components in the scheduling path to the top five and increasing their image acquisition frequency to 6 frames per minute. Simultaneously, components without abnormal states, with disturbance response prediction residuals below 1 degree Celsius and similarity calculation results above 0.85, are given lower priority, their acquisition frequency is reduced to 2 frames per minute, and their acquisition sequence number in the scheduling path is moved to the end. The previous scheduling path record is retained during the update process to compare acquisition result trends and verify the effectiveness of the dynamic path adjustment. After the update, the inspection scheduling feedback update module sends the newly generated dynamic image acquisition scheduling path to the asynchronous image acquisition module, replacing the current execution path. This ensures that the system's inspection plan responds in real-time to thermal anomaly changes, improving the timeliness and coverage of key component image acquisition. Through these mechanisms, the system implements a scheduling path optimization scheme driven by both recognition results and actual response residuals, significantly improving the overall image acquisition system's adaptability and fault detection efficiency.

[0020] Preferably, after receiving the thermal anomaly status results determined by the thermal anomaly trend analysis and identification module and the intervention execution status feedback from the response behavior intervention and disturbance execution module, the inspection scheduling feedback update module immediately initiates the dynamic image acquisition scheduling path update process. Based on the current state of the target component, the priority order and acquisition frequency of the component in the acquisition path are reordered by comprehensively considering three data indicators: its thermal anomaly status level, the disturbance response prediction residual amplitude, and the intervention execution result. First, the components are divided into three levels according to the thermal anomaly status level: Level 1 indicates a temperature change rate greater than 3 degrees Celsius per second and lasting for more than 5 frames; Level 2 indicates a temperature change rate between 2 and 3 degrees Celsius per second and lasting for 3 to 5 frames; Level 3 indicates a rate less than 2 degrees Celsius per second but with a potential warming trend. For Level 1 components, the system sets their priority order to the top 5 in the acquisition path and the acquisition frequency to 10 frames per minute; for Level 2 components, the priority is set between 6 and 15, and the acquisition frequency is 6 frames per minute; for Level 3 components, the priority is set after 16, and the acquisition frequency is 4 frames per minute. If the disturbance response prediction residual of a component exceeds 1.5 degrees Celsius, its priority will be automatically increased to the top 10, regardless of its thermal anomaly level, and the acquisition frequency will be increased by 2 frames per minute. Furthermore, if the feedback from the response behavior intervention and disturbance execution module indicates that a component has successfully elicited typical thermal response characteristics in the past three interventions, the system will mark the component as a high-sensitivity component and increase its acquisition frequency to 12 frames per minute to monitor subtle thermal changes. Conversely, if a component does not show any signs of temperature anomaly in five consecutive image acquisitions, the disturbance response prediction residual will be... Components consistently below 0.5 degrees Celsius and with a similarity greater than 0.9 will be classified as low-risk components, their acquisition priority will be downgraded to the lowest position, and the acquisition frequency will be adjusted to 2 frames per minute. After priority reordering and frequency updates are completed, the newly generated scheduling path will be sent to the asynchronous image acquisition module through the high-speed bus interface, and the current acquisition task plan will be refreshed in real time to ensure that the system's image acquisition resources are always preferentially allocated to high-risk components, achieving the optimal image coverage strategy under limited acquisition capabilities. At the same time, a closed-loop feedback mechanism for thermal anomaly status will be formed to improve the coverage and response efficiency of system fault identification.

[0021] Preferably, data interaction in the system is based on time synchronization identifiers, forming a complete closed-loop control process. The entire system starts with an asynchronous image acquisition module, which executes image acquisition tasks on the components according to a dynamically generated image acquisition scheduling path and records the timestamp and spatial coordinate information of each acquisition, sending the acquired images together with the thermal anomaly trend analysis and identification module. The image sequence is sorted temporally and analyzed for temperature, generating a time difference thermal response matrix and identifying candidate heating regions. At the same time, the historical temperature trajectory of the heating region is extracted and combined with the current ambient temperature and component background temperature to construct a temperature trend prediction. The model states that if the predicted temperature change rate is greater than 3 degrees Celsius per second and the heating duration exceeds 5 frames, the corresponding area is identified as a thermal anomaly. Data such as the thermal anomaly label, component coordinates, and predicted temperature sequence are then transferred to the response behavior intervention and disturbance execution module. Based on the target component's spatial location, the connection strength of adjacent structures, historical intervention results, and the remaining energy of the intervention equipment, the disturbance impact factor is calculated. If the impact factor is higher than 3.5, a disturbance is applied to the target component, and the asynchronous image acquisition module is notified to acquire images of the target component within the first 5 seconds to the last 10 seconds. If the impact factor is lower than 3.5, the selected structure is then used instead. Neighboring components with a connection strength higher than 5 and a similarity in illumination conditions higher than 85% are subjected to perturbation, and images are acquired 3 seconds after the perturbation. All perturbation images are then sent to the thermal anomaly trend analysis and identification module. Based on the perturbation images, perturbation response features are extracted and a perturbation response fitting model is constructed. A potential thermal response path for the target component is generated through structural coupling mapping, and thermal anomaly state simulation inference is performed. When the asynchronous image acquisition module subsequently acquires the actual image of the target component, the simulation results are further compared with the real images to generate perturbation response prediction residual data. The thermal anomaly state determination results and residual data are received by the inspection scheduling feedback update module and used as input to perform dynamic image acquisition scheduling path update operations. The acquisition priority is reordered and the frequency allocation is adjusted to form an updated acquisition task plan, which is then fed back to the asynchronous image acquisition module to complete a complete closed-loop process. Through the above process, the system achieves strict time synchronization and data consistency in all aspects of data acquisition, state identification, intervention execution, simulation inference, and scheduling feedback. This ensures that every thermal anomaly is fully captured and a dynamic response is made based on high-reliability data, effectively improving the accuracy, stability, and intelligence level of the photovoltaic module fault identification system.

[0022] (III) Beneficial Effects

[0023] This invention provides a photovoltaic module fault image recognition system based on infrared thermal imaging analysis. It has the following beneficial effects:

[0024] 1. This invention calculates the response delay value of each component through an asynchronous image acquisition module and generates a dynamic image acquisition scheduling path accordingly, specifying the order and timestamp of the infrared thermal imager's image acquisition; ensuring that components in different thermal response processes can have their thermal images captured within the same time window; effectively solving the problem of image distortion or omission of key thermal information caused by differences in component thermal response in traditional sequential acquisition; providing accurate time-aligned basic image data for subsequent precise thermal anomaly identification; and significantly improving the system's ability to capture the true thermal distribution state of photovoltaic modules.

[0025] 2. The response behavior intervention and disturbance execution module of this invention makes intelligent decisions on the object to be disturbed based on disturbance influencing factors; and triggers the asynchronous image acquisition module to acquire images of the target or nearby components before and after the disturbance; the thermal anomaly trend analysis and identification module uses these images to perform disturbance response analysis or fit and infer the thermal state of the target component; the inspection scheduling feedback update module integrates the thermal anomaly judgment results, prediction errors and intervention execution status; dynamically adjusts the priority and frequency of subsequent image acquisition; and greatly improves the system's active detection, verification and identification accuracy and overall inspection efficiency for potential thermal anomalies under complex working conditions. Detailed Implementation

[0026] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Example 1:

[0028] This invention provides a photovoltaic module fault image recognition system based on infrared thermal imaging analysis. The system, at a photovoltaic power station operation and maintenance site, initiates a 24 / 7 inspection task. First, after system startup, the asynchronous image acquisition module loads the dynamic image acquisition scheduling path generated by the most recent inspection scheduling feedback update module. The path records the spatial coordinate number, acquisition priority order, and corresponding acquisition frequency setting for each photovoltaic module. Under unattended operation, the asynchronous image acquisition module sequentially performs infrared image acquisition operations on each module according to the scheduling path, recording a timestamp accurate to milliseconds and module coordinate information at each acquisition. Then, the acquired infrared thermal image data and temporal and spatial information are transmitted via a data bus to a thermal anomaly trend analysis and recognition module.

[0029] After receiving the data, the thermal anomaly trend analysis and identification module first performs time series rearrangement, arranging multiple frames of images of the same component acquired at different consecutive times in ascending order of timestamps to form a time-series frame column. Then, it performs pixel-level temperature matrix parsing on each frame to extract the temperature distribution values ​​on the component surface and calculates the corresponding temperature change rate at each time. If the temperature change rate of a certain area is detected to be higher than 3 degrees Celsius per second for 5 consecutive frames and the area of ​​the area exceeds 50 pixels, this area is marked as a candidate thermal anomaly area and enters the next step of trend prediction processing. The trend prediction processing calls the component's historical temperature change trajectory data and the currently acquired real-time temperature data as input, and uses the temperature change gradient to estimate the temperature trend curve for the next 5 seconds. If the predicted curve shows that the temperature will continue to rise in the next 5 seconds and the change range is not less than 5 degrees Celsius, the system confirms that the component has a thermal anomaly tendency and sends the marking results, component coordinates, predicted curve, and anomaly level information to the response behavior intervention and disturbance execution module.

[0030] After receiving the data, the response behavior intervention and disturbance execution module queries the structural connection strength and historical disturbance records of the component, and reads the energy reserves of the currently available disturbance devices. It calculates the disturbance impact factor; if the calculated result is higher than 3.5, it directly implements a disturbance operation on the target component. At this time, the disturbance device applies a rapid local heating signal of a preset intensity to the component surface. The disturbance signal automatically stops after 2 seconds. Simultaneously, the asynchronous image acquisition module is notified to continuously acquire infrared images of the target component at a frequency of 1 frame per second within a time range of 5 seconds before the disturbance and 10 seconds after the disturbance. After acquisition, all images are transmitted to the thermal anomaly trend analysis and identification module. Frame difference calculation is performed on the images before and after the disturbance to obtain a pixel-level temperature difference matrix. Three types of feature values ​​are extracted from this matrix: total temperature change, hotspot expansion path, and affected area. These are combined into a disturbance response feature vector. Then, the disturbance response feature vector is compared with the standard fault-free disturbance response feature vector stored in the system database using Euclidean distance similarity. The similarity is calculated, and if the similarity value is less than 0.75, the component's thermal response is determined to be significantly different from the fault-free state, confirming the existence of a thermal anomaly. At the same time, disturbance response prediction residual data is generated. The residual data is calculated from the difference between the simulated predicted temperature value and the actual collected temperature value. After receiving the thermal anomaly determination result and residual data, the inspection scheduling feedback update module prioritizes and adjusts the acquisition frequency of the dynamic image acquisition scheduling path. High-risk components are prioritized to the top 5 and the acquisition frequency is increased to 10 frames per minute. Low-risk components are prioritized to the bottom and the acquisition frequency is adjusted to 2 frames per minute. The updated scheduling path is sent to the asynchronous image acquisition module through the bus to replace the current execution plan, thus forming a complete closed-loop process of thermal anomaly detection, intervention, response analysis, and scheduling optimization. All data interactions in the entire process have a unified time stamp to ensure data consistency in each link of acquisition, analysis, intervention, and feedback, ensuring that the thermal anomaly state is detected in a timely manner and continuously tracked and monitored.

[0031] Example 2:

[0032] This embodiment, based on Embodiment 1, differs primarily in the implementation strategies for response behavior intervention and disturbance execution, as well as the priority adjustment method for the scheduling feedback update stage. Firstly, at the task initiation stage, the asynchronous image acquisition module still loads the latest dynamic image acquisition scheduling path provided by the inspection scheduling feedback update module. However, in this embodiment, the initial allocation of priority and frequency within the path is sorted according to the weighted scores of the disturbance response prediction residual and the thermal anomaly status level from the previous inspection. The weighting coefficients are 0.6 for residual and 0.4 for anomaly level. This approach ensures that the scheduling path already reflects early attention to high-residual components from the initial stage. The asynchronous image acquisition module executes infrared image acquisition operations according to the path sequence, recording component coordinates and timestamps during each acquisition and simultaneously acquiring the ambient temperature data for subsequent analysis. The acquired data is transmitted to the thermal anomaly trend analysis and identification module via a high-speed data bus.

[0033] After receiving image data, the thermal anomaly trend analysis and identification module performs temporal rearrangement and pixel temperature parsing. Simultaneously, it uses real-time ambient temperature as a correction factor in the temperature change rate calculation. If four consecutive frames show a temperature change rate exceeding 2.8 degrees Celsius per second and the hotspot area is larger than 60 pixels, the region is designated as a candidate thermal anomaly region for trend prediction analysis. This trend prediction uses an improved linear regression combined with an exponential smoothing model to estimate the temperature trend over the next 5 seconds. If the prediction shows that the temperature will continue to rise with an increase greater than 4 degrees Celsius, the thermal anomaly determination result and related data are sent to the response behavior intervention and disturbance execution module.

[0034] In this embodiment, the response behavior intervention and disturbance execution module introduces the difference in illumination intensity between neighboring components as a correction parameter when calculating the disturbance impact factor. The illumination difference is calculated using the illumination intensity data collected in real time by the component backplane sensor. If the disturbance impact factor plus the illumination correction value is higher than 3.2, the component with the highest adjacent structural connection strength and the closest illumination conditions is preferentially selected for disturbance, and the target component is indirectly affected through the thermal coupling effect of neighboring components. The disturbance device applies a mild heating signal to the surface of the selected component for 3 seconds. After the intervention, the asynchronous image acquisition module continuously acquires images of the target component and the disturbed component at a frequency of 1 frame per second within a time range of 3 seconds to 15 seconds after the disturbance. Infrared images of the components are captured and transmitted to the thermal anomaly trend analysis and identification module for disturbance response feature extraction and analysis. During the analysis, the system establishes a dual-channel disturbance response model to fit the temperature change curves of the target component and the disturbed component respectively. Then, the similarity and time delay information between the two curves are used to infer the propagation path and coupling strength of the thermal anomaly. If the inference results show that the temperature response of the target component is highly consistent with the temperature curve of the disturbed component with a time delay of no more than 2 seconds and the similarity is less than 0.78, it is determined that the target component has a thermal anomaly caused by structural coupling, and residual data is generated. The residual data and the thermal anomaly determination results are simultaneously transmitted to the inspection scheduling feedback update module.

[0035] In this embodiment, a dynamic frequency adjustment algorithm is used when updating the scheduling path. For components with residuals exceeding 1.8 degrees Celsius, regardless of their current priority, they are directly inserted into the first 3 positions of the path, and the acquisition frequency is set to 12 frames per minute. For components with residuals below 0.4 degrees Celsius for two consecutive inspections, they are directly moved to the end, and the acquisition frequency is adjusted to 1 frame per minute. The updated scheduling path immediately covers the currently executing task, ensuring that acquisition resources are concentrated on high-risk objects, thereby forming a priority control closed loop driven by residuals. The entire inspection process, from acquisition, analysis, intervention, re-analysis to scheduling optimization, is interconnected. Each link uses timestamps as the basis for data matching, ensuring that all processing processes correspond precisely in the time dimension. This allows the system to efficiently and accurately identify and track thermal anomalies even in complex component coupling scenarios.

[0036] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A photovoltaic module fault image recognition system based on infrared thermal imaging analysis, characterized in that, include: The asynchronous image acquisition module calculates the response delay value based on the spatial distribution, shading status, and structural thermal capacity coefficient of the photovoltaic modules, and generates a dynamic image acquisition scheduling path accordingly. It controls the infrared thermal imager to acquire images of modules in different thermal response processes within the same time window, and records the acquisition time and spatial coordinates of the modules. The thermal anomaly trend analysis and identification module receives the component image and its time and space coordinates, calculates the time difference between consecutive image frames, constructs a time difference thermal response matrix, identifies candidate heating regions where the temperature continues to rise for more than three consecutive frames, establishes a temperature trend prediction model, predicts the rate of temperature change, and determines the thermal anomaly state. The response behavior intervention and disturbance execution module receives the thermal anomaly status and component spatial coordinates, selects the target or nearby components to apply disturbance based on the disturbance influence factor, and triggers the asynchronous image acquisition module to acquire corresponding images within a set time period; The inspection scheduling feedback update module receives thermal anomaly judgment results and intervention data, and dynamically adjusts the priority and acquisition frequency of the dynamic image acquisition scheduling path.

2. The photovoltaic module fault image recognition system based on infrared thermal imaging analysis according to claim 1, characterized in that: The asynchronous image acquisition module analyzes the arrangement of photovoltaic modules in the spatial structure, the shading conditions affected by sunlight, and the differences in structural heat capacity coefficient caused by the construction materials to calculate a set of different response delay values ​​for different modules. During the acquisition process, the asynchronous image acquisition module does not acquire images sequentially according to the physical adjacency order, but dynamically constructs a dynamic image acquisition scheduling path based on the response delay values. The dynamic construction is to arrange the acquisition order of the modules according to the magnitude of the response delay values ​​and a preset sorting rule, so that the modules with response delay value differences exceeding a preset difference threshold are imaged first. This allows for the capture of image thermal states at different response stages within the same acquisition time window; simultaneously, the asynchronous image acquisition module records the time of each image acquisition and the corresponding component spatial coordinates for each acquisition node.

3. The photovoltaic module fault image recognition system based on infrared thermal imaging analysis according to claim 1, characterized in that: The thermal anomaly trend analysis and identification module establishes a time difference thermal response matrix based on the image received by the asynchronous image acquisition module, the acquisition time and spatial coordinates, and selects pixel regions with continuously increasing temperature values ​​in three or more consecutive frames as candidate heating regions. For the candidate heating regions, the thermal anomaly trend analysis and identification module further extracts the historical response trajectory of the region in different frames, and combines the historical ambient temperature data in historical inspections with the current component background temperature to establish a temperature trend prediction model. After establishing a temperature trend prediction model, the thermal anomaly trend analysis and identification module outputs the predicted temperature value and temperature change rate along with the thermal anomaly state.

4. The photovoltaic module fault image recognition system based on infrared thermal imaging analysis according to claim 1, characterized in that: After receiving the thermal anomaly status from the thermal anomaly trend analysis and identification module, the response behavior intervention and disturbance execution module first determines whether disturbance operation on the target area is permitted based on the disturbance impact factor. The disturbance impact factor is composed of the physical location of the component, the degree of coupling between adjacent components, the number of successful historical interventions, and the current remaining energy of the device. If several factors are deemed infeasible, the response behavior intervention and disturbance execution module will not directly disturb the target component. Instead, it will select neighboring components whose structural connection strength with the target component exceeds a set strength threshold and whose similarity in illumination conditions exceeds a set similarity threshold to apply the disturbance. The asynchronous image acquisition module will then acquire images of neighboring components at a third set time after the disturbance is applied to capture the thermal response changes during the disturbance propagation process. The images will then be used by the thermal anomaly trend analysis and identification module to fit the thermal response path that the target area may present under the same intervention scenario, thereby simulating and calculating the temperature response curve of the target component, realizing thermal state inference when the target area image is not directly acquired. If the rate of change of the temperature response curve exceeds a set risk threshold, the thermal anomaly status of the target component will be confirmed.

5. The photovoltaic module fault image recognition system based on infrared thermal imaging analysis according to claim 4, characterized in that: After receiving images acquired from neighboring components, the thermal anomaly trend analysis and identification module first extracts disturbance response feature information, including disturbance diffusion velocity, thermal response expansion direction, and response amplitude boundary. It then constructs a disturbance response fitting model by combining structural connection strength parameters, illumination condition similarity parameters, and component material information between the target region and neighboring regions. Using this model, the module substitutes the disturbance response features from neighboring regions into the thermal conduction mapping relationship between the target component and neighboring components, constructed based on the structural connection strength parameters, illumination condition similarity parameters, and component material information. Finally, it performs a trend path inversion operation. This inversion operation does not depend on the current image of the target region but outputs its potential response curve based on the disturbance simulation mapping. If the response curve shows that the temperature change rate of the target area exceeds the preset risk threshold, then the target component is determined to be in a thermal anomaly state. When the target component image is subsequently acquired, the thermal anomaly trend analysis and identification module compares the temperature response curve with the temperature response of the target component in the actual acquired image, calculates the deviation between the predicted temperature value and the actual temperature value, and generates disturbance response prediction residual data.

6. The photovoltaic module fault image recognition system based on infrared thermal imaging analysis according to claim 5, characterized in that: The thermal anomaly trend analysis and identification module compares the predicted response curve output by the fitted model with the subsequent actual collected temperature response data of the target component, records the error offset between the predicted value and the true value, and constructs a disturbance response residual index matrix. The matrix reflects the fitting stability and the misjudgment risk level. Subsequently, the inspection scheduling feedback update module receives the disturbance response residual index matrix, which reflects the fitting stability and the misjudgment risk level, and uses it as an input factor to update the dynamic image acquisition scheduling path. During the update process, the inspection scheduling feedback update module adjusts the priority order and acquisition frequency of each component in the dynamic image acquisition scheduling path according to the residual magnitude. For components whose disturbance response prediction residual data exceeds the preset residual threshold, the acquisition frequency is increased, and for components whose disturbance response prediction residual data is lower than the preset residual threshold, the acquisition frequency is increased.

7. The photovoltaic module fault image recognition system based on infrared thermal imaging analysis according to claim 1, characterized in that: When the disturbance impact factor is higher than a set feasible threshold, the response behavior intervention and disturbance execution module applies a disturbance to the target component and triggers the asynchronous image acquisition module to acquire images of the target component from a first set time before the disturbance to a second set time after the disturbance. The thermal anomaly trend analysis and identification module performs frame difference analysis to calculate the temperature difference between the same coordinate points in the images before and after the disturbance, extracts the temperature change, hotspot expansion path, and affected area features triggered by the disturbance, and forms a disturbance response feature vector; the disturbance response feature vector is then compared with a preset or historically learned standard fault-free component disturbance response feature vector for similarity calculation; if the similarity is lower than a set matching threshold, the target component is determined to have a thermal anomaly state.

8. The photovoltaic module fault image recognition system based on infrared thermal imaging analysis according to claim 1, characterized in that: The inspection scheduling feedback update module receives the thermal anomaly status judgment result output by the thermal anomaly trend analysis and identification module, and performs joint analysis on the target area status information and the disturbance response residual index matrix. For components with thermal anomaly status, components with disturbance response prediction residual data exceeding the set residual threshold, or components with similarity calculation results lower than the set matching threshold, their priority and acquisition frequency in the dynamic image acquisition scheduling path are increased. For components with no detected thermal anomaly status and disturbance response prediction residual data lower than the set residual threshold or similarity calculation results higher than the set matching threshold, their priority and acquisition frequency in the dynamic image acquisition scheduling path are reduced.