Photovoltaic module fault prediction method and system based on AI

Abstract

The invention discloses an AI-based photovoltaic module fault prediction method and system in the field of photovoltaic technology, and the system comprises a data collection unit, a communication unit, a data preprocessing unit, a dynamic feature modeling unit, a fault prediction unit, and an early warning and operation unit. The data acquisition module comprises an electrical parameter acquisition module, a thermal infrared forming module, a visible light image module and an environmental data acquisition module, the communication unit is externally connected with a cloud collaboration module, and the cloud collaboration module is provided with an edge computing node. By constructing a time sequence convolutional network and introducing an attention mechanism, the system can capture dynamic changes of electrical parameters, infrared thermal imaging and visible light images in real time, and effectively adapt to a complex and changeable outdoor environment. The system integrates electrical parameters, infrared thermal imaging, visible light images and environmental data, and improves the accuracy and reliability of fault prediction by fully utilizing complementarity of different modal data through weighted fusion and deep fusion methods.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic technology, specifically to an AI-based method and system for predicting photovoltaic module failures. Background Technology

[0002] As the core equipment of photovoltaic power plants, photovoltaic modules are exposed to complex outdoor environments for extended periods, enduring various natural factors such as thermal cycling, ultraviolet radiation, wind and sand erosion, and humidity changes. These environmental factors can lead to a variety of failures in photovoltaic modules, including but not limited to microcracks, hot spots, and PID (potential-induced degradation) effects. Traditional fault detection methods mainly rely on manual inspections, which are not only inefficient but also prone to missing early-stage faults, causing them to worsen and affecting the overall performance and lifespan of the photovoltaic power plant.

[0003] With the rapid development of AI technology, its application in fault diagnosis is becoming increasingly widespread. However, existing AI fault diagnosis technologies are mostly limited to static data analysis and lack adaptability to dynamic environmental changes. For example, the performance of photovoltaic modules is affected in real time by environmental factors such as light intensity, temperature, and humidity, and traditional AI models often cannot effectively capture these dynamic changes, resulting in limited prediction accuracy. In addition, photovoltaic module faults are diverse, and the diagnosis of compound faults (i.e., two or more faults occurring simultaneously in the same device or system) is particularly difficult; existing methods have insufficient ability to identify compound faults. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based method and system for predicting photovoltaic module failures, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an AI-based photovoltaic module fault prediction system, comprising a data acquisition unit, a communication unit, a data preprocessing unit, a dynamic feature modeling unit, a fault prediction unit, and an early warning and maintenance unit. The data acquisition module includes an electrical parameter acquisition module, a thermal infrared forming module, a visible light image module, and an environmental data acquisition module. The communication unit is externally connected to a cloud collaboration module, which is equipped with an edge computing node. The edge computing node enables localized preprocessing and real-time analysis of data, combined with a cloud-based AI model for complex fault diagnosis, reducing data transmission latency and improving system response speed. The data preprocessing unit includes a data denoising module, a threshold segmentation module, and a parameter synchronization module. The dynamic feature modeling unit includes a temporal convolutional network module, which performs multi-scale feature extraction on electrical parameters, captures dynamic changes such as power attenuation and voltage fluctuations, introduces an attention mechanism, and performs weighted fusion of features from infrared thermal imaging and visible light images to highlight fault-related areas. Combined with environmental data, an environment-performance correlation model is established to dynamically adjust the fault threshold. The fault prediction unit constructs a Transformer architecture to perform sequence modeling on the fused multimodal features, predicts the probability and type of fault occurrence, and dynamically optimizes the parameters of the prediction model through reinforcement learning algorithms to improve adaptability in complex scenarios. The early warning and operation and maintenance unit includes an early warning information sending module and a daily operation and maintenance module.

[0006] As a further aspect of the present invention: the electrical parameter acquisition module in the data acquisition unit is set at the electrical equipment and transmission circuit node, and collects the voltage, current, power and IV curve data of the photovoltaic module in real time through multiple sensors. The infrared thermoforming unit includes multiple infrared cameras, which capture the surface temperature distribution of the module and identify hot spot faults.

[0007] As a further aspect of the present invention: the visible light spectrum module in the data acquisition unit collects images of the component's appearance through multiple drone patrol routes and fixed cameras, and detects cracks, stains and defects. The environmental data module establishes a signal connection with the weather station through the communication module and acquires meteorological data, including light intensity, temperature, humidity and wind speed, and analyzes the impact of environmental factors on the component's performance.

[0008] As a further aspect of the present invention: the data preprocessing unit normalizes the electrical parameters to eliminate dimensional differences, performs temperature threshold segmentation on the infrared image to extract abnormal temperature rise areas, performs noise reduction, enhancement, and target detection on the visible light image to locate defects such as cracks and stains, and performs time series alignment on the environmental data for synchronous analysis with the electrical parameters.

[0009] As a further aspect of the present invention: the dynamic feature modeling unit constructs a temporal convolutional network to extract features from electrical parameters at multiple scales, captures dynamic changes in power attenuation and voltage fluctuations, and introduces an attention mechanism to perform weighted fusion of features from infrared thermal imaging and visible light images, highlighting fault-related areas. Combined with environmental data, an environment-performance correlation model is established to dynamically adjust the fault threshold.

[0010] An AI-based method for predicting photovoltaic module failures includes the aforementioned failure prediction system and the following steps: S100: Acquires detection data through pre-set sensors and drone patrols, integrates the data, and transmits the integrated data to the communication unit; S200: The communication unit receives the data transmitted in step S100, transmits the data to the cloud collaboration module for data storage and archiving, and transmits the data to the data preprocessing module at the same time. S300: Performs preliminary processing on the received data, including data normalization, data denoising, data augmentation, and data time series alignment, and transmits the preprocessed data to the next step; S400: After receiving the data processed in step S300, dynamic modeling is performed, and a temporal convolutional network is constructed. By combining the data model with real-time detection data, dynamic features of electrical parameters are extracted, an attention fusion model is constructed, infrared and visible light image features are integrated, environmental data is combined, an environment-performance correlation model is established, the fault threshold is dynamically adjusted, and the data processed by the model is transmitted to the next step. S500: Through the Transformer architecture, sequence modeling is performed on the fused features to predict the fault type and probability of occurrence.

[0011] Compared with existing technologies, the beneficial effects of this invention are as follows: By constructing a temporal convolutional network (TCN) and introducing an attention mechanism, the system can capture the dynamic changes of electrical parameters, infrared thermal imaging, and visible light images in real time, effectively adapting to complex and ever-changing outdoor environments. The system integrates electrical parameters, infrared thermal imaging, visible light images, and environmental data, and through weighted fusion and deep fusion methods, fully utilizes the complementarity of different modal data, improving the accuracy and reliability of fault prediction.

[0012] For complex faults, the system employs a multi-model fusion prediction strategy, combining the TCN model, attention fusion model, and environment-performance correlation model to effectively identify and differentiate different types of faults, improving the comprehensiveness and accuracy of fault diagnosis. The system is equipped with an early warning and maintenance unit, capable of sending real-time warning information and providing detailed maintenance suggestions to help maintenance personnel respond and handle faults promptly, reducing downtime and repair costs. Through cloud collaboration modules and edge computing nodes, the system achieves localized data preprocessing and real-time analysis, reducing data transmission latency, improving system response speed, and ensuring data security and privacy. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the structure of an AI-based photovoltaic module fault prediction system according to the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0016] In the description of this invention, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0017] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "configuration" should be interpreted broadly. For example, they can refer to a fixed connection or configuration, a detachable connection or configuration, or an integral connection or configuration. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0018] Please see Figure 1 In this embodiment of the invention, an AI-based photovoltaic module fault prediction system is provided: Electrical parameter acquisition: High-precision voltage and current sensors are installed at the output end of the photovoltaic modules to collect voltage and current data in real time. Simultaneously, a power analyzer is used to obtain the real-time power of the modules, and the IV curve data of the modules under different light intensities is recorded. To ensure data accuracy and completeness, the sampling frequency is set to once per second, and the data recording interval is 1 minute.

[0019] Infrared thermal imaging acquisition: Infrared cameras are strategically deployed within the photovoltaic power station to ensure coverage of all photovoltaic modules. The cameras acquire infrared thermal imaging data of the module surfaces every 10 minutes, recording the temperature distribution on the module surfaces. During the acquisition process, care must be taken to avoid the influence of direct sunlight on the infrared cameras to ensure the accuracy of temperature measurements.

[0020] Visible light image acquisition: High-definition cameras mounted on drones or fixed on brackets are used to periodically acquire visible light images of the photovoltaic modules. The drone inspection cycle can be set according to the power plant scale and actual needs, generally once a week; fixed cameras can acquire images in real time, but to reduce data storage, they can be set to save one image every 30 minutes. The acquired images must include the complete appearance of the modules in order to accurately detect defects such as cracks and stains.

[0021] Environmental data acquisition: A weather station is installed within the photovoltaic power station to collect environmental data such as sunlight intensity, temperature, humidity, and wind speed in real time. The weather station's data acquisition frequency is consistent with the electrical parameter acquisition frequency, both being once per second, to achieve data time synchronization.

[0022] Data synchronization processing: Due to the different acquisition frequencies of different types of data, time alignment processing is required. Using the timestamp of electrical parameter data as a reference, interpolation processing is performed on infrared thermal imaging data, visible light image data, and environmental data to ensure that all data are consistent in time, providing a foundation for subsequent data fusion and analysis.

[0023] Data preprocessing Electrical parameter preprocessing Normalization: To eliminate differences in dimensions and numerical ranges between different electrical parameters, a minimum-maximum normalization method is used to map parameters such as voltage, current, and power to the [0, 1] interval. The formula is:

[0024] Where x is the original data, Xmin and Xmar are the minimum and maximum values ​​of the parameter, respectively, and Xnorm is the normalized data.

[0025] Outlier handling: The 3σ criterion is used to detect outliers in electrical parameters. Data exceeding the mean ± 3 standard deviations are considered outliers and removed. Then, linear interpolation is used to fill in the removed outliers to ensure data continuity.

[0026] Infrared thermal imaging preprocessing Temperature calibration: Due to measurement errors in infrared cameras and the influence of environmental factors, temperature calibration is required for the acquired infrared thermal imaging data. This is achieved by taking images in front of a standard blackbody radiation source at a known temperature, establishing a mapping relationship between infrared image pixel values ​​and the actual temperature, and then performing temperature correction on subsequently acquired infrared images.

[0027] Image segmentation: A threshold-based segmentation method is employed. A temperature threshold is set based on the normal operating temperature range of the component, and regions in the infrared image with temperatures exceeding the threshold are segmented as suspected hotspot areas. Simultaneously, morphological processing is performed on the segmented image to remove noise and small interference areas, improving the accuracy of hotspot detection.

[0028] Visible light image preprocessing Denoising: Median filtering is used to denoise the visible light image, removing interference factors such as salt-and-pepper noise. Median filtering replaces the gray value of each pixel in the image with the median of the gray values ​​of its neighboring pixels, effectively preserving the image's edge information.

[0029] Image enhancement: Histogram equalization is used to enhance the denoised image, improving its contrast and making defects such as cracks and stains more clearly visible. Histogram equalization redistributes the gray values ​​of image pixels to make the image's histogram more uniformly distributed.

[0030] Target detection: Deep learning-based target detection algorithms (such as the YOLO series algorithms) are used to detect targets in the enhanced visible light image, identify photovoltaic modules in the image, and locate the location and size of defects such as cracks and stains.

[0031] Environmental data preprocessing Data smoothing: A moving average filtering algorithm is used to smooth the environmental data and remove high-frequency noise. The moving average filter takes the average value of data within a certain time window as the current data value; the window size can be adjusted according to actual needs.

[0032] Feature extraction: Features related to photovoltaic module performance, such as the rate of change of irradiance and the daily variation of temperature, are extracted from the smoothed environmental data. These features reflect the impact of environmental factors on module performance and provide auxiliary information for subsequent fault prediction.

[0033] Multimodal data fusion Feature extraction and selection Electrical parameter feature extraction: Short-time Fourier transform (STFT) is used to perform time-frequency analysis on the normalized electrical parameters to extract dynamic features such as power attenuation and voltage fluctuation. Simultaneously, statistical characteristics of the electrical parameters, such as mean, variance, and peak value, are calculated to reflect the steady-state performance of the components.

[0034] Infrared thermal imaging feature extraction: For the segmented suspected hot spot areas, features such as area, mean temperature, and temperature standard deviation are extracted. These features can reflect the severity and development trend of the hot spots.

[0035] Visible light image feature extraction: Extract the length, width, and shape features of cracks, as well as the area and color features of stains from the target detection results. These features can intuitively reflect the appearance defects of the components.

[0036] Environmental data feature extraction: The extracted environmental features are combined with the aforementioned electrical, infrared, and visible light features to form a multidimensional feature vector. To reduce the feature dimensionality and improve the model's training efficiency and generalization ability, Principal Component Analysis (PCA) is used to reduce the dimensionality of the multidimensional feature vector, and principal components with a cumulative contribution rate of over 90% are selected as the final feature vector.

[0037] Feature fusion method Weighted fusion: Based on the importance of different types of features to fault prediction, corresponding weights are assigned to each feature. Weight values ​​can be determined through expert experience or data-driven methods (such as entropy weighting). The weighted feature vectors are then summed to obtain the fused feature vector.

[0038] Deep fusion: Deep learning models (such as multilayer perceptrons, MLPs) are used to deeply fuse the dimensionality-reduced feature vectors. Feature vectors from different modalities are taken as input, and through multi-layer nonlinear transformations, deep-level correlations between features are uncovered, outputting a fused high-level feature representation. Deep fusion can better capture the complex relationships between data and improve the accuracy of fault prediction.

[0039] Dynamic Feature Modeling and Fault Prediction Dynamic feature modeling Temporal Convolutional Network (TCN) Modeling: A TCN model is constructed to model the dynamic characteristics of electrical parameters. The TCN model employs dilated convolutions and residual connections, effectively capturing long-range dependencies in time-series data. The fused feature vectors are input into the TCN model in chronological order, and the model's parameters are adjusted through training to enable it to accurately predict component performance trends.

[0040] Attention Mechanism Fusion Modeling: For the fusion features of infrared thermal imaging and visible light images, an attention mechanism is introduced for modeling. The attention mechanism can automatically learn the importance of different feature regions, assigning greater weight to important regions, thereby improving the model's focus on fault-related areas. The attention mechanism is combined with a convolutional neural network (CNN) to construct an attention fusion model, which further extracts and analyzes the fused image features.

[0041] Environment-Performance Correlation Modeling: An environment-performance correlation model is established by incorporating environmental characteristics. Using the Support Vector Regression (SVR) algorithm, with environmental characteristics as input and component performance metrics (such as power and efficiency) as output, the model is trained to establish the mapping relationship between environmental factors and component performance. This model can predict component performance under normal conditions based on current environmental data, providing a reference benchmark for fault prediction.

[0042] Fault prediction Multi-model fusion prediction: The prediction results of the TCN model, the attention fusion model, and the environment-performance correlation model are fused. A weighted average method is used to assign appropriate weights according to the prediction accuracy and reliability of each model to obtain the final fault prediction result.

[0043] Fault type identification: Based on the prediction results and preset fault thresholds, determine whether a component has failed and the type of fault. For example, when the hot spot feature exceeds the threshold, it is identified as a hot spot fault; when the crack feature exceeds the threshold, it is identified as a microcrack fault, etc. Simultaneously, graded warnings are issued based on the severity of the fault, providing detailed maintenance recommendations for operations and maintenance personnel.

[0044] Reinforcement learning optimization: Reinforcement learning algorithms (such as Deep Q-Network, DQN) are used to dynamically optimize the fault prediction model. The fault prediction process is viewed as a Markov decision process, with prediction accuracy and false alarm rate as reward functions. Through continuous interaction and learning with the environment, the model parameters are adjusted so that the model can maintain high prediction performance in different scenarios.

[0045] The Transformer architecture is a neural network architecture based on the self-attention mechanism. Its core advantages lie in its ability to process sequential data in parallel, capture long-range dependencies, and maintain a unified structure that is easy to extend. Its core architecture is as follows: The Transformer architecture mainly consists of four parts: the input part (input-output embedding and positional encoding), a multi-layer encoder, a multi-layer decoder, and the output part (output linear layer and Softmax). It adopts an Encoder-Decoder architecture, but the encoder and decoder parts can be used independently to construct language models, corresponding to the Encoder-Only model and the Decoder-Only model, respectively.

[0046] The input section includes: Source text embedding layer: Converts the numerical representation of words in the source text into vector representation to capture the relationships between words.

[0047] Position encoder: Generates a position vector for each position in the input sequence so that the model can understand the positional information in the sequence. Since the Transformer lacks temporal awareness, positional information needs to be injected through position encoding. Commonly used formulas are sine encoding and cosine encoding.

[0048] Target text embedding layer (used in the decoder): Converts the lexical numerical representations in the target text into vector representations.

[0049] The encoder includes: Encoder stack: Consists of N encoder layers stacked together (N is 6 in the paper, but other numbers can be tried). Each encoder layer has the same structure but does not share weights.

[0050] Encoder layer structure: Each encoder layer consists of two sub-layer connection structures.

[0051] The first sub-layer: Multi-Head Attention, is used to capture the relationships and contextual information between words in the sequence.

[0052] The second sub-layer, the Feed-Forward Network, employs a location-independent computation method, meaning that the transformation operations at each time step are identical and executed in parallel.

[0053] Each sublayer is followed by a normalization layer (LayerNorm) and a residual connection (ResidualConnection), which together are called the Add&Norm operation. The residual connection is used to prevent network degradation, and LayerNorm is used to normalize the activation values ​​of each layer.

[0054] The decoder includes: Decoder stack: Consists of N decoder layers stacked together (the same number of encoder layers). Each decoder layer has the same structure but does not share weights.

[0055] Decoder layer structure: Each decoder layer consists of a three-layer interconnection structure.

[0056] The first sub-layer is a masked multi-head self-attention sub-layer, which is used to ensure that only the generated positions are seen during decoding, thus avoiding information leakage.

[0057] The second sub-layer, the multi-head attention sub-layer (Encoder-Decoder Attention), is used to model the association between the input sequence and the currently generated word.

[0058] The third sub-layer: the feedforward fully connected sub-layer, is the same as the feedforward fully connected sub-layer in the encoder.

[0059] Each sub-layer is followed by a normalization layer and a residual connection, collectively known as the Add & Norm operation.

[0060] The output includes: Linear layer: Transforms the vector output by the decoder into the final output dimension.

[0061] Softmax layer: Converts the output of the linear layer into a probability distribution for final prediction.

[0062] The method for handling complex faults is as follows: A compound fault refers to the phenomenon where two or more faults occur simultaneously in the same device or system. Its core characteristics include: The types of faults are diverse: faults may involve different components or subsystems (such as cracks in mechanical bearings and broken gear teeth, short circuits and grounding faults in power systems).

[0063] Spatial correlation: Faults may occur in the same location (such as multiple damages to the same component) or in different locations (such as faults in different lines in a power system), but it is important to distinguish between "compound faults" and "multiple faults" (the latter usually refers to independent faults that are spatially dispersed).

[0064] Coupling: Fault features may interfere with, superimpose or cancel each other out, leading to the ambiguity of a single fault feature (such as the mixing of multiple fault frequency components in a vibration signal).

[0065] The working principle of this invention is as follows: Based on multimodal data fusion and deep learning technology, it integrates electrical parameters, infrared thermal imaging, visible light images, and environmental data to achieve accurate prediction of photovoltaic module faults. The system first collects multi-source heterogeneous data through sensors and drones, and performs preliminary processing to eliminate noise and outliers.

[0066] Subsequently, weighted fusion and deep fusion methods were used to effectively integrate the data features of different modalities, forming a multi-dimensional feature vector that comprehensively reflects the state of photovoltaic modules.

[0067] In the dynamic feature modeling stage, the system constructs a temporal convolutional network (TCN) model to capture the dynamic changes of electrical parameters; at the same time, an attention mechanism is introduced to enhance the attention to fault-related regions in infrared thermal imaging and visible light images.

[0068] By combining environmental data, an environment-performance correlation model is established, and the fault threshold is dynamically adjusted to adapt to the fault prediction needs under different environmental conditions.

[0069] Finally, in the fault prediction stage, the system uses the Transformer architecture to perform sequence modeling on the fused multimodal features to predict the probability and type of fault occurrence.

[0070] By dynamically optimizing the parameters of the prediction model using reinforcement learning algorithms, the model's adaptability and robustness in complex scenarios can be improved.

[0071] The early warning and operation and maintenance unit promptly sends early warning information based on the forecast results and provides operation and maintenance suggestions to help operation and maintenance personnel respond quickly and handle faults, ensuring the safe and stable operation of photovoltaic power plants.

[0072] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0073] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. An AI-based photovoltaic module fault prediction system, comprising a data acquisition unit, a communication unit, a data preprocessing unit, a dynamic feature modeling unit, a fault prediction unit, and an early warning and maintenance unit, characterized in that: The data acquisition module includes an electrical parameter acquisition module, a thermal infrared shaping module, a visible light image module, and an environmental data acquisition module. The communication unit is externally connected to a cloud collaboration module. The data preprocessing unit includes a data denoising module, a threshold segmentation module, and a parameter synchronization module. The dynamic feature modeling unit includes a temporal convolutional network construction module. The temporal convolutional network performs multi-scale feature extraction on electrical parameters, captures dynamic changes such as power attenuation and voltage fluctuations, introduces an attention mechanism, and performs weighted fusion of features from infrared thermal imaging and visible light images to highlight fault-related areas. Combined with environmental data, an environment-performance correlation model is established, and the fault threshold is dynamically adjusted. The fault prediction unit constructs a Transformer architecture to perform sequence modeling on the fused multimodal features, predicts the probability and type of fault occurrence, and dynamically optimizes the parameters of the prediction model through reinforcement learning algorithms to improve adaptability in complex scenarios. The early warning and operation and maintenance unit includes an early warning information sending module and a daily operation and maintenance module.

2. The AI-based photovoltaic module fault prediction system according to claim 1, characterized in that: The electrical parameter acquisition module in the data acquisition unit is located at the nodes of electrical equipment and transmission circuits. It collects the voltage, current, power and IV curve data of the photovoltaic module in real time through multiple sensors. The infrared thermoforming unit includes multiple infrared cameras, which capture the surface temperature distribution of the module and identify hot spot faults.

3. The AI-based photovoltaic module fault prediction system according to claim 1, characterized in that: The visible light spectrum module in the data acquisition unit collects images of the component's appearance through multiple drone patrol routes and fixed cameras, detecting cracks, stains, and defects. The environmental data module establishes a signal connection with a weather station through a communication module and acquires meteorological data, including light intensity, temperature, humidity, and wind speed, to analyze the impact of environmental factors on the component's performance.

4. The AI-based photovoltaic module fault prediction system according to claim 1, characterized in that: The data preprocessing unit normalizes electrical parameters to eliminate dimensional differences, performs temperature threshold segmentation on infrared images to extract abnormal temperature rise areas, performs noise reduction, enhancement, and target detection on visible light images to locate defects such as cracks and stains, and performs time series alignment on environmental data for synchronous analysis with electrical parameters.

5. The AI-based photovoltaic module fault prediction system according to claim 1, characterized in that: The dynamic feature modeling unit extracts features from electrical parameters at multiple scales by constructing a temporal convolutional network, capturing dynamic changes in power attenuation and voltage fluctuations. It also introduces an attention mechanism to weightedly fuse features from infrared thermal imaging and visible light images, highlighting fault-related areas. Combined with environmental data, it establishes an environment-performance correlation model and dynamically adjusts the fault threshold.

6. The AI-based photovoltaic module fault prediction method according to claim 1, characterized in that: Includes the fault prediction system as described in claims 1-5, and the following steps: S100: Acquires detection data through pre-set sensors and drone patrols, integrates the data, and transmits the integrated data to the communication unit; S200: The communication unit receives the data transmitted in step S100, transmits the data to the cloud collaboration module for data storage and archiving, and transmits the data to the data preprocessing module at the same time. S300: Performs preliminary processing on the received data, including data normalization, data denoising, data augmentation, and data time series alignment, and transmits the preprocessed data to the next step; S400: After receiving the data processed in step S300, dynamic modeling is performed, and a temporal convolutional network is constructed. By combining the data model with real-time detection data, dynamic features of electrical parameters are extracted, an attention fusion model is constructed, infrared and visible light image features are integrated, environmental data is combined, an environment-performance correlation model is established, the fault threshold is dynamically adjusted, and the data processed by the model is transmitted to the next step. S500: Through the Transformer architecture, sequence modeling is performed on the fused features to predict the fault type and probability of occurrence.

Images