An intelligent fault diagnosis method and device for a photovoltaic device

Abstract

The application discloses an intelligent fault diagnosis method and device for a photovoltaic device, obtains operation data of the photovoltaic device, processes the operation data by using a preset multi-modal model to generate target labeling results corresponding to each operation data and target confidence degrees corresponding to each target labeling result, determines that the operation data is to-be-verified data when the target confidence degree corresponding to any target labeling result exceeds a target confidence degree threshold value and the target confidence degrees corresponding to the remaining target labeling results are lower than the target confidence degree threshold value, analyzes the to-be-verified data by using a preset cross-validation algorithm, and determines that the photovoltaic device is in a fault state when at least two target labeling results point to the same fault type and the corresponding target confidence degrees satisfy a preset confirmation condition. The complementarity of different modal data is brought into play, false positives or false negatives caused by judgment errors of a single data source are avoided, and the diagnosis stability and accuracy of the system in a complex environment are significantly improved.

Description

Technical Field

[0001] This application relates to the field of large language models, and in particular to an intelligent fault diagnosis method and device for photovoltaic equipment. Background Technology

[0002] As a core component of the clean energy transition, photovoltaic power generation faces severe challenges in its long-term stable operation. Under the continuous action of complex natural environments (such as sudden temperature changes, strong ultraviolet radiation, and mechanical stress), photovoltaic modules are prone to malfunctions such as hot spot effects, microcracks in solar cells, and burnout of junction boxes, leading to a significant decrease in power generation efficiency or even safety accidents.

[0003] Existing technologies for diagnosing photovoltaic equipment fault conditions rely on single data sources, making them ill-suited for multi-dimensional fault scenarios. For example, electrical diagnostics based on current-voltage curve analysis are susceptible to interference from light fluctuations and shading, resulting in a high false alarm rate. Infrared thermal imaging detection based on computer vision, on the other hand, heavily depends on ideal lighting conditions and equipment compatibility, leading to a surge in missed detections in scenarios such as dawn / dusk, rain / fog, or densely packed modules. Furthermore, existing technologies generally lack collaborative verification mechanisms for multi-modal data; electrical signals, visual images, and manual inspection texts are fragmented, failing to form a comprehensive basis for judgment. Summary of the Invention

[0004] This application provides an intelligent fault diagnosis method and apparatus for photovoltaic equipment, which overcomes the problem that existing technologies generally lack collaborative verification mechanisms for multimodal data.

[0005] Firstly, this application provides an intelligent fault diagnosis method for photovoltaic equipment, including:

[0006] Obtain operational data from photovoltaic equipment;

[0007] The running data is processed using a pre-defined multimodal model to generate target annotation results for each running data point, as well as target confidence scores for each target annotation result.

[0008] When the target confidence score corresponding to any target annotation result exceeds the target confidence score threshold, and the target confidence scores corresponding to the remaining target annotation results are lower than the target confidence score threshold, the running data is determined to be data to be verified.

[0009] The data to be verified is analyzed using a preset cross-validation algorithm. When at least two target annotation results point to the same fault type and the corresponding target confidence scores meet the preset confirmation conditions, the photovoltaic equipment is determined to be in a fault state.

[0010] Secondly, this application provides an intelligent fault diagnosis device for photovoltaic equipment, comprising:

[0011] The operating data determination module is configured to acquire operating data of photovoltaic equipment;

[0012] The target confidence determination module is configured to process the running data using a preset multimodal model to generate target annotation results corresponding to each running data, and target confidence corresponding to each target annotation result;

[0013] The module for determining data to be verified is configured to determine the running data as data to be verified when the target confidence level corresponding to any target annotation result exceeds the target confidence level threshold, and the target confidence level corresponding to the remaining target annotation results is lower than the target confidence level threshold.

[0014] The fault status determination module is configured to analyze the data to be verified using a preset cross-validation algorithm. When at least two target annotation results point to the same fault type and the corresponding target confidence scores meet the preset confirmation conditions, the photovoltaic equipment is determined to be in a fault state.

[0015] Thirdly, this application provides a readable medium including executable instructions, which, when executed by a processor of an electronic device, cause the electronic device to perform any of the methods described in the first aspect.

[0016] Fourthly, this application provides an electronic device including a processor and a memory storing execution instructions, wherein when the processor executes the execution instructions stored in the memory, the processor performs the method as described in any of the first aspects.

[0017] This application provides an intelligent fault diagnosis method and device for photovoltaic equipment. By establishing a spatiotemporal alignment mechanism and cross-validation logic for multimodal data and constructing a continuously evolving optimization sample library, it enables the leap from passive response maintenance to proactive predictive maintenance in photovoltaic fault diagnosis.

[0018] The further effects of the aforementioned non-conventional preferred method will be explained below in conjunction with specific embodiments. Attached Figure Description

[0019] To more clearly illustrate the embodiments of this application or the existing technical solutions, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating an intelligent fault diagnosis method for photovoltaic equipment provided in an embodiment of this application;

[0021] Figure 2 A flowchart illustrating another intelligent fault diagnosis method for photovoltaic equipment provided in an embodiment of this application;

[0022] Figure 3 A flowchart illustrating another intelligent fault diagnosis method for photovoltaic equipment provided in an embodiment of this application;

[0023] Figure 4 This is a flowchart illustrating another intelligent fault diagnosis method for photovoltaic equipment provided in an embodiment of this application;

[0024] Figure 5 This is a schematic diagram of the structure of an intelligent fault diagnosis device for photovoltaic equipment provided in an embodiment of this application;

[0025] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] As a core component of the clean energy transition, photovoltaic power generation faces severe challenges in its long-term stable operation. Under the continuous action of complex natural environments (such as sudden temperature changes, strong ultraviolet radiation, and mechanical stress), photovoltaic modules are prone to malfunctions such as hot spot effects, microcracks in solar cells, and burnout of junction boxes, leading to a significant decrease in power generation efficiency or even safety accidents.

[0028] Existing technologies for diagnosing photovoltaic equipment fault conditions rely on single data sources, making them ill-suited for multi-dimensional fault scenarios. For example, electrical diagnostics based on current-voltage curve analysis are susceptible to interference from light fluctuations and shading, resulting in a high false alarm rate. Infrared thermal imaging detection based on computer vision, on the other hand, heavily depends on ideal lighting conditions and equipment compatibility, leading to a surge in missed detections in scenarios such as dawn / dusk, rain / fog, or densely packed modules. Furthermore, existing technologies generally lack collaborative verification mechanisms for multi-modal data; electrical signals, visual images, and manual inspection texts are fragmented, failing to form a comprehensive basis for judgment.

[0029] To address this issue, this application proposes an intelligent fault diagnosis method for photovoltaic equipment, aiming to solve the problem that existing technologies generally lack collaborative verification mechanisms for multimodal data. In this embodiment, the intelligent fault diagnosis method for photovoltaic equipment includes:

[0030] Step 101: Obtain the operating data of the photovoltaic equipment.

[0031] The system collects electrical signal data from photovoltaic (PV) equipment via PV inverters; collects visual image data from PV equipment via cameras; collects inspection text data using smart terminals; and determines operational data based on the electrical signal data, visual image data, and inspection text data.

[0032] To comprehensively reflect the operating status of photovoltaic equipment under different conditions, the system needs to collect multi-source heterogeneous information, including electrical characteristics, visual characteristics, and on-site semantic descriptions. Specifically, the system collects electrical signal data from photovoltaic inverters deployed in the photovoltaic module system in real time. This data typically includes parameters such as voltage, current, and power, and reflects the electrical performance status of the equipment during actual power generation.

[0033] The system also acquires visual image data via cameras deployed on the ground or drones. This data typically includes infrared thermal images, visible light photographs, or video clips. Infrared images can effectively expose problems such as hot spots and overheated junction boxes, while visible light images can identify visual defects that may affect light absorption efficiency, such as glass cracks, surface stains, and obstructions. The image acquisition is time-aligned with the electrical signals collected by the inverter, enabling the system to perform accurate cross-modal matching in subsequent analysis.

[0034] Furthermore, manual inspections still play an indispensable role in routine maintenance. Maintenance personnel record various anomalies or operational information observed on-site using smart terminals. This text data contains subjective descriptions of unstructured information such as equipment appearance, operating environment, sounds, or odors, often providing supplementary details that electrical signals and images cannot easily cover. To achieve standardized processing, this text information is transformed into a structured format using natural language processing technology, thus forming a standardized input with electrical signal and image data.

[0035] The system uses the device's unique identifier and timestamp as the associated index to fuse electrical signal data, visual image data, and inspection text data, and then determines them as a one-time operation data sample.

[0036] Step 102: Process the running data using a preset multimodal model to generate target annotation results corresponding to each running data, and target confidence scores corresponding to each target annotation result.

[0037] After data acquisition is complete, it will be input into a pre-defined multimodal model for analysis and processing. This multimodal model consists of multiple sub-models with specific task capabilities, each responsible for feature extraction and semantic understanding of different types of data. Through collaborative work between models, the system can not only fully explore potential fault features in each modality of data, but also perform semantic alignment between different information sources, enabling multi-faceted verification of fault judgment results.

[0038] For electrical signal data, the system inputs it into a time-series analysis model for processing; visual image data is fed into a visual model built on a deep convolutional neural network for processing; and inspection text data is input into a natural language processing model for semantic understanding and entity recognition. The system unifies the outputs of the three types of models to form a labeled dataset containing a triplet of "data type - label - target confidence".

[0039] Step 103: When the target confidence level corresponding to any target annotation result exceeds the target confidence level threshold, and the target confidence level corresponding to the remaining target annotation results is lower than the target confidence level threshold, the running data is determined to be the data to be verified.

[0040] To avoid false alarms or missed alarms due to misjudgment of a single modality, the system performs confidence comparison and consistency analysis on the judgment results of each modality after completing the initial annotation of the multimodal model. Each model output includes not only a predicted label for the fault type but also a quantitative confidence value, reflecting the model's degree of trust in that judgment. The system pre-sets a uniform confidence threshold to determine whether the model is "confident" that its annotation results are sufficiently reliable.

[0041] In actual operation, if the confidence level of a certain modal model (such as the electrical signal model) exceeds the preset threshold, it indicates that it has a high confidence intention in judging a certain fault. At the same time, the confidence levels of other modal models (such as the visual model and the text model) do not reach the threshold, which means that they do not support the same fault conclusion in the current sample, or that there is still uncertainty in their judgment of the current state. This state of "single modality prominence and other modalities silence" indicates that there is inconsistency in the judgments between modalities at the information level in the operational data, which belongs to the marginal sample with potential risks but for which a consensus has not yet been reached.

[0042] In this situation, the system does not immediately classify the data as faulty or non-faulty; instead, it marks the data as data to be verified. This mechanism effectively prevents premature fault identification due to environmental sensitivity of a particular modal model or model overfitting. At the same time, it retains potential fault clues, creating opportunities for subsequent cross-validation and manual review.

[0043] Step 104: Analyze the data to be verified using a preset cross-validation algorithm. When at least two target annotation results point to the same fault type and the corresponding target confidence scores meet the preset confirmation conditions, determine that the photovoltaic equipment is in a fault state.

[0044] After initial screening and marking of the operational data as data to be validated, the system will initiate a preset cross-validation algorithm to conduct in-depth analysis of these samples with discrepancies in intermodal judgments. The goal of the cross-validation process is to identify data that, although not consistently recognized by all models, exhibit significant fault symptoms in some models through mutual verification of information between different modalities, thereby achieving robust identification of potential faults.

[0045] The cross-validation algorithm first performs semantic aggregation and type comparison on the annotation results of each modal model. Even if some modal models do not reach the initial confidence threshold, if at least two modal models output target annotation results pointing to the same fault type—for example, both the electrical signal and text models classify it as "component hot spot," or both the image and electrical signal models identify it as "junction box burnout"—this constitutes a cross-modal fault consistency event. Based on this, the system further verifies whether the confidence levels corresponding to these consistency judgments meet preset confirmation conditions.

[0046] The confirmation criteria are not based solely on exceeding a confidence threshold, but rather on a combination of the characteristics and weights of each model, employing computational strategies such as weighted averaging, confidence compensation, or strategic fusion. For example, visual models typically have higher weights on sunny days, while electrical signal models are relatively more reliable on cloudy days or at night. If two or more annotation results point to the same fault type, and their combined weighted confidence exceeds the system's set confirmation threshold, then the photovoltaic equipment can be determined to be in a fault state.

[0047] As can be seen from the above technical solutions, the beneficial effects of this embodiment are:

[0048] This application provides an intelligent fault diagnosis method for photovoltaic (PV) equipment. The method acquires the PV equipment's operational data; processes the operational data using a preset multimodal model to generate target labeling results for each operational data point, and target confidence scores for each target labeling result; when the target confidence score for any target labeling result exceeds a target confidence score threshold, and the target confidence scores for the remaining target labeling results are below the target confidence score threshold, the operational data is determined to be data to be verified; a preset cross-validation algorithm is used to analyze the data to be verified, and when at least two target labeling results point to the same fault type, and the corresponding target confidence scores meet preset confirmation conditions, the PV equipment is determined to be in a fault state. This method leverages the complementarity of different modal data, avoids false alarms or missed alarms caused by errors in judgment from a single data source, and significantly improves the diagnostic stability and accuracy of the system in complex environments.

[0049] Figure 1 The above is only a basic embodiment of an intelligent fault diagnosis method for photovoltaic equipment according to this application. With certain optimizations and extensions, other preferred embodiments of the intelligent fault diagnosis method for photovoltaic equipment can be obtained.

[0050] like Figure 2 The image shows another specific embodiment of an intelligent fault diagnosis method for photovoltaic equipment according to this application.

[0051] In this embodiment, a smart fault diagnosis method for photovoltaic equipment includes the following steps:

[0052] Step 201: Obtain the operating data of the photovoltaic equipment.

[0053] Step 202: Process the running data using a preset multimodal model to generate target annotation results corresponding to each running data, and target confidence scores corresponding to each target annotation result.

[0054] Step 203: Process electrical signal data using a time series model to generate the first annotation result corresponding to the photovoltaic device, and the first confidence level corresponding to the first annotation result.

[0055] The system processes the electrical signal data acquired from the photovoltaic inverter using a structured time-series analysis model. This model is typically built on Transformer, LSTM, or other deep time-series network architectures and can capture the dynamic characteristics of signals such as voltage, current, and power over time.

[0056] By modeling and analyzing historical waveform data, the model can identify abnormal patterns caused by component short circuits, open circuits, power attenuation, etc. The model output includes a first annotation result to indicate whether a fault exists and its type, and generates a corresponding first confidence value to quantify the reliability of the model's judgment.

[0057] Step 204: Process the visual image data using a visual model to generate a second annotation result corresponding to the photovoltaic device, and a second confidence level corresponding to the second annotation result.

[0058] For visual image data acquired by drones, fixed cameras, or other visualization acquisition devices, the system inputs it into a visual model constructed by a deep convolutional neural network. This model uses image enhancement, feature extraction, and multi-scale fusion to detect and locate typical faults in the images, such as hot spots, hidden cracks, occlusions, and stains.

[0059] The model outputs a second annotation result, identifying whether there are visible structural or thermodynamic anomalies in the current image, and provides a second confidence score, measuring the model's confidence in the judgment result under the current image conditions. This type of model is particularly suitable for the identification and auxiliary verification of non-electrical faults.

[0060] Step 205: Process the inspection text data using a text model to generate the third annotation result corresponding to the photovoltaic equipment, and the third confidence level corresponding to the third annotation result.

[0061] For inspection text data submitted via mobile terminals by maintenance personnel, the system employs a natural language processing model for intelligent parsing. This model, based on BERT or similar pre-trained language models, combines named entity recognition and syntactic analysis techniques to semantically extract and classify key fault terms, equipment locations, and environmental descriptions within the text content.

[0062] The model outputs a third annotation result, clarifying whether the text involves specific fault information and its type, and also generates a third confidence score to reflect the model's level of confidence in understanding and classifying the text content. This modality is of significant value in supplementing human observation results and providing semantic verification evidence.

[0063] Step 206: When the target confidence level corresponding to any target annotation result exceeds the target confidence level threshold, and the target confidence level corresponding to the remaining target annotation results is lower than the target confidence level threshold, the running data is determined to be the data to be verified.

[0064] Step 207: Analyze the data to be verified using a preset cross-validation algorithm. When at least two target annotation results point to the same fault type and the corresponding target confidence scores meet the preset confirmation conditions, determine that the photovoltaic equipment is in a fault state.

[0065] As can be seen from the above technical solution, the beneficial effects of this embodiment are: the system obtains annotation results and confidence levels corresponding to electrical signals, images, and text data, thus laying the foundation for subsequent modal fusion, cross-validation, and final fault diagnosis. This modeling approach with clear division of labor and collaborative complementarity significantly improves the performance and reliability of the diagnostic system in various scenarios.

[0066] like Figure 3 The image shows another specific embodiment of an intelligent fault diagnosis method for photovoltaic equipment according to this application. This embodiment is further described based on the foregoing embodiments.

[0067] In this embodiment, a smart fault diagnosis method for photovoltaic equipment includes the following steps:

[0068] Step 301: Obtain the operating data of the photovoltaic equipment.

[0069] Step 302: Process the running data using a preset multimodal model to generate target annotation results corresponding to each running data, and target confidence scores corresponding to each target annotation result.

[0070] Step 303: When the target confidence level corresponding to any target annotation result exceeds the target confidence level threshold, and the target confidence level corresponding to the remaining target annotation results is lower than the target confidence level threshold, the running data is determined to be the data to be verified.

[0071] Step 304: Analyze the data to be verified using a preset cross-validation algorithm. When at least two target annotation results point to the same fault type and the corresponding target confidence scores meet the preset confirmation conditions, determine that the photovoltaic equipment is in a fault state.

[0072] Step 305: Calculate a weighted average of the confidence scores of each target according to the corresponding model weight coefficients to generate a weighted confidence score.

[0073] During each diagnostic process, the system acquires confidence scores from the time-series model, visual model, and text model. Because these three models differ in their perceived objects, input characteristics, and susceptibility to environmental interference, the system does not simply average these confidence scores. Instead, it assigns a weighting coefficient to each model. These weighting coefficients reflect the system's assessment of the diagnostic reliability of each modality in a specific scenario, and are typically derived from historical model performance statistics, actual deployment conditions, or dynamically adjusted by operational strategies.

[0074] For example, in data collected under clear weather conditions, the visual model may have a relatively high weight due to its better image quality. However, in cloudy, rainy, foggy, or nighttime inspection scenarios, the visual model may be subject to greater interference, and the system will appropriately reduce its weight while increasing the influence of the electrical signal model or text model. This mechanism of dynamically adjusting weights based on the environment significantly enhances the adaptability and robustness of the model fusion results.

[0075] Step 306: When the weighted confidence level exceeds the preset confirmation threshold, determine that the target confidence level meets the preset confirmation conditions.

[0076] After obtaining the confidence scores and corresponding weight coefficients of each model, the system will integrate them using a weighted average mathematical method. Specifically, the confidence score of each model is multiplied by its weight and then summed, and the sum is normalized by the total weights to finally output the weighted confidence score.

[0077] After the weighted confidence score is calculated, the system compares it with the confirmation threshold. If the weighted confidence score is higher than the threshold, it indicates that multiple models have reached significant consistency in their analysis of the current data, and they all demonstrate high confidence in their respective dimensions. In this case, the system determines that the weighted result meets the preset confirmation criteria, and therefore the diagnostic result corresponding to the operational data can be considered highly reliable, and the fault type it points to can be formally confirmed.

[0078] As can be seen from the above technical solutions, the beneficial effects of this embodiment are: the weighted confidence mechanism effectively solves the impact of the performance differences of different modal models under specific environments, and avoids misleading the final diagnosis due to overfitting or fluctuations in input quality of a certain modality.

[0079] like Figure 4 The image shows another specific embodiment of an intelligent fault diagnosis method for photovoltaic equipment according to this application. This embodiment is further described based on the foregoing embodiments.

[0080] Step 401: Obtain the operating data of the photovoltaic equipment.

[0081] Step 402: Process the running data using a preset multimodal model to generate target annotation results corresponding to each running data, and target confidence scores corresponding to each target annotation result.

[0082] Step 403: When the target confidence level corresponding to any target annotation result exceeds the target confidence level threshold, and the target confidence level corresponding to the remaining target annotation results is lower than the target confidence level threshold, the running data is determined to be the data to be verified.

[0083] Step 404: Analyze the data to be verified using a preset cross-validation algorithm. When at least two target annotation results point to the same fault type and the corresponding target confidence scores meet the preset confirmation conditions, determine that the photovoltaic equipment is in a fault state.

[0084] Step 405: When the target annotation results point to different fault types, and / or the target confidence corresponding to the target annotation results of the same fault type does not meet the confirmation conditions, the running data is marked as data to be reviewed.

[0085] When the target annotation results output by different models point to different fault types, it indicates a significant discrepancy in the system's interpretation of the current operating data. For example, the time-series model determines it as a short-circuit fault based on the IV curve, while the visual model identifies a hot spot defect. This inconsistency in fault type indicates a lack of effective coordination among the current multimodal information, which may be due to varying degrees of interference in each data modality or the existence of fuzzy features in the data itself, making it difficult for the system to make reliable automatic judgments.

[0086] Even when the target annotation results point to the same fault type, if their corresponding target confidence scores do not reach the preset confirmation criteria (i.e., the weighted confidence score does not reach the preset confirmation threshold), the authenticity of the fault cannot be directly determined. Since confidence score is a crucial indicator reflecting the reliability of the model's judgment, diagnoses made with low confidence scores are highly likely to lead to misjudgments or omissions. Therefore, even when the model outputs are semantically consistent, if the confidence score is insufficient, the system will still mark them as data requiring review.

[0087] Step 406: Detect the data to be reviewed to generate correction results, and store the correction results in the tuning sample library corresponding to the photovoltaic equipment.

[0088] The data is verified through manual inspection, secondary measurement using high-precision sensors, or expert-assisted analysis, and correction labels are generated based on the on-site confirmation results. These correction results include the accurate fault type, the corresponding timestamp and device ID, and possible reasons for misjudgment. This information not only has immediate operational and maintenance value, but more importantly, it is recorded by the system and stored in the corresponding optimization sample library for the photovoltaic equipment.

[0089] The establishment and use of the optimization sample library marks a step forward for the system from static diagnosis to dynamic evolution. These corrected samples serve as high-value training feedback, driving the system to relearn key error patterns and fine-tune parameters.

[0090] The parameters of the multimodal model are updated using corrected labeled samples from the tuning sample library; the parameter update includes fine-tuning training of at least one of the temporal model, visual model and text model.

[0091] When the multimodal diagnostic system encounters data requiring verification during operation, the corrected labeled samples generated after confirmation by manual or auxiliary detection methods will be automatically stored in the optimization sample library. Compared with ordinary training samples, these corrected labeled samples usually have higher information density and error targeting, and are therefore given higher weight in subsequent model training. The system will periodically or as needed call these samples to filter and process them, thereby providing the model with a high-quality source of updated data.

[0092] The parameter update process does not operate solely on the entire model system, but rather possesses targeted and differentiated processing capabilities. The system selectively fine-tunes at least one of the time-series model, visual model, or text model based on the source modality and error distribution of different samples.

[0093] For example, if a batch of corrected samples reflects a concentration of misjudgments in electrical signal identification, the system will focus on updating the time-series model that processes electrical signal data. This will be achieved by retraining the encoding structure, optimizing the loss function, or adjusting the learning rate to improve the accuracy of identifying dynamic signal anomalies. Similarly, if the errors in the samples mainly originate from image discrimination—for example, blurred hotspot boundaries or unrecognized dirt interference—the system will focus on fine-tuning the convolutional layers or feature extractors in the visual model to enhance its sensitivity to edge information and temperature gradient changes.

[0094] Fine-tuning of the text model focuses more on language understanding and context modeling capabilities. Especially when inspection texts contain ambiguous descriptions or non-standard terminology, the system will fine-tune language models such as BERT to better adapt them to the language characteristics of the photovoltaic operation and maintenance field. Through this mechanism, the model can not only correct specific cases of previous identification failures but also summarize general feature patterns, improving overall generalization ability.

[0095] The target annotation results and target confidence scores are constructed as input vectors and fed into the natural language processing model; a fault diagnosis report is generated using the preset semantic templates and logical rules in the natural language processing model.

[0096] The system structurally integrates the target annotation results generated by each model and their corresponding confidence values, constructing an input vector in a unified format. This vector not only contains the fault type and its source model, but also the confidence score for each type of diagnosis, serving as a quantitative basis for the reliability of the diagnosis.

[0097] This set of input vectors is then fed into a natural language processing model, typically a large language model based on the Transformer architecture, such as BERT or T5, and processed in conjunction with predefined semantic templates and logical rules. The semantic templates in the system are designed to cover common fault types, location of occurrence, time information, abnormal related parameters, and handling suggestions. Logical rules are used to determine information completeness, fault severity, and cross-verification between multimodal information, ensuring that the generated report content is semantically coherent and logically closed-loop.

[0098] In the actual generation process, the natural language processing model maps the input vector to the filling slots of the corresponding semantic template and selects the optimal expression path based on the fault context. The final generated diagnostic report can be presented in a combination of text and graphics, including natural language descriptions, confidence score lists, fault areas marked in CV images, IV curve waveforms, and other information, and is packaged into a visual interface and pushed to the operation and maintenance platform.

[0099] like Figure 5 The image shown is a specific embodiment of an intelligent fault diagnosis device for photovoltaic equipment according to this application. This embodiment describes an intelligent fault diagnosis device for photovoltaic equipment, specifically used for performing... Figures 1-4 A physical device for intelligent fault diagnosis of photovoltaic equipment is provided. Its technical solution is essentially the same as the above embodiments, and the corresponding descriptions in the above embodiments also apply to this embodiment. This embodiment of an intelligent fault diagnosis device for photovoltaic equipment includes:

[0100] The operating data determination module 501 is configured to acquire the operating data of the photovoltaic equipment.

[0101] The target confidence determination module 502 is configured to process the running data using a preset multimodal model to generate target annotation results corresponding to each running data and target confidence corresponding to each target annotation result;

[0102] The data to be verified module 503 is configured to determine the running data as data to be verified when the target confidence corresponding to any target annotation result exceeds the target confidence threshold and the target confidence corresponding to the other target annotation results is lower than the target confidence threshold.

[0103] The fault state determination module 504 is configured to analyze the data to be verified using a preset cross-validation algorithm. When at least two target annotation results point to the same fault type and the corresponding target confidence scores meet the preset confirmation conditions, the photovoltaic equipment is determined to be in a fault state.

[0104] Figure 6This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include RAM, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk storage device. Of course, the electronic device may also include other hardware required for other services.

[0105] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0106] Memory is used to store instructions for execution. Specifically, instructions for execution are computer programs that can be executed. Memory can include main memory and non-volatile memory, and it provides the processor with execution instructions and data.

[0107] In one possible implementation, the processor reads the corresponding execution instructions from non-volatile memory into main memory and then executes them. Alternatively, it may obtain the corresponding execution instructions from other devices to logically form an intelligent fault diagnosis device for photovoltaic equipment. The processor executes the execution instructions stored in the memory to implement the intelligent fault diagnosis method for photovoltaic equipment provided in any embodiment of this application.

[0108] The above is as stated in this application. Figure 5The method for intelligent fault diagnosis of photovoltaic equipment provided in the illustrated embodiment can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor.

[0109] The steps of the method disclosed in the embodiments of this application can be directly manifested as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0110] This application also proposes a readable medium storing execution instructions. When these instructions are executed by a processor of an electronic device, the electronic device can perform an intelligent fault diagnosis method for photovoltaic equipment provided in any embodiment of this application, specifically for executing, as... Figure 1 or Figure 2 or Figure 3 or Figure 4 The method shown.

[0111] The electronic devices in the foregoing embodiments may be computers.

[0112] Those skilled in the art will understand that the embodiments of this application can be provided as methods or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or a combination of software and hardware.

[0113] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0114] It should also be noted that 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 process, method, article, or apparatus. Unless otherwise specified, 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.

[0115] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for intelligent fault diagnosis of photovoltaic equipment, characterized in that, include: Acquiring operational data of photovoltaic (PV) equipment includes: collecting electrical signal data of the PV equipment via a PV inverter; collecting visual image data of the PV equipment via a camera; collecting inspection text data using a smart terminal; and determining the operational data based on the electrical signal data, the visual image data, and the inspection text data. The process utilizes a preset multimodal model to process the operational data, generating target annotation results corresponding to each set of operational data, and target confidence scores corresponding to each target annotation result. This includes: processing the electrical signal data using a time-series model to generate a first annotation result corresponding to the photovoltaic device, and a first confidence score corresponding to the first annotation result; processing the visual image data using a visual model to generate a second annotation result corresponding to the photovoltaic device, and a second confidence score corresponding to the second annotation result; and processing the inspection text data using a text model to generate a third annotation result corresponding to the photovoltaic device, and a third confidence score corresponding to the third annotation result. When the target confidence score corresponding to any of the target annotation results exceeds the target confidence score threshold, and the target confidence scores corresponding to the remaining target annotation results are lower than the target confidence score threshold, the running data is determined to be data to be verified. The data to be verified is analyzed using a preset cross-validation algorithm. When at least two target annotation results point to the same fault type and the corresponding target confidence scores meet preset confirmation conditions, the photovoltaic equipment is determined to be in a fault state. The preset confirmation conditions for target confidence scores to meet include: weighting each target confidence score according to the corresponding model weight coefficient to generate a weighted confidence score; when the weighted confidence score exceeds a preset confirmation threshold, the target confidence score is determined to meet the preset confirmation conditions.

2. The method according to claim 1, characterized in that, The condition that at least two target annotation results point to the same fault type, and the corresponding target confidence scores meet the preset confirmation conditions, further includes: When the target annotation results point to different fault types, and / or the target confidence corresponding to the target annotation results of the same fault type does not meet the confirmation condition, the running data is marked as data to be reviewed. The data to be reviewed is detected to generate a correction result, and the correction result is stored in the tuning sample library corresponding to the photovoltaic equipment.

3. The method according to claim 2, characterized in that, Also includes: The parameters of the multimodal model are updated using the corrected labeled samples in the optimization sample library; The parameter update includes fine-tuning training of at least one of the time series model, the visual model, and the text model.

4. The method according to any one of claims 1-3, characterized in that, Also includes: The target annotation results and the target confidence scores are constructed into an input vector and then input into a natural language processing model. A fault diagnosis report is generated using the preset semantic templates and logical rules in the natural language processing model.

5. An intelligent fault diagnosis device for photovoltaic equipment, characterized in that, include: The operation data determination module is configured to acquire operation data of photovoltaic equipment, including: acquiring electrical signal data of the photovoltaic equipment through a photovoltaic inverter; acquiring visual image data of the photovoltaic equipment through a camera; acquiring inspection text data using a smart terminal; and determining the operation data based on the electrical signal data, the visual image data, and the inspection text data. The target confidence determination module is configured to process the operating data using a preset multimodal model to generate target labeling results corresponding to each set of operating data, and target confidence levels corresponding to each target labeling result. This includes: processing the electrical signal data using a time-series model to generate a first labeling result corresponding to the photovoltaic device, and a first confidence level corresponding to the first labeling result; processing the visual image data using a visual model to generate a second labeling result corresponding to the photovoltaic device, and a second confidence level corresponding to the second labeling result; and processing the inspection text data using a text model to generate a third labeling result corresponding to the photovoltaic device, and a third confidence level corresponding to the third labeling result. The data to be verified module is configured to determine the running data as data to be verified when the target confidence level corresponding to any of the target annotation results exceeds the target confidence level threshold, and the target confidence level corresponding to the remaining target annotation results is lower than the target confidence level threshold. The fault state determination module is configured to analyze the data to be verified using a preset cross-validation algorithm. When at least two target annotation results point to the same fault type and the corresponding target confidence scores meet preset confirmation conditions, the photovoltaic equipment is determined to be in a fault state. The target confidence scores meeting the preset confirmation conditions include: weighting each target confidence score according to the corresponding model weight coefficient to generate a weighted confidence score; when the weighted confidence score exceeds a preset confirmation threshold, the target confidence score is determined to meet the preset confirmation conditions.

6. A computer-readable storage medium storing a computer program, characterized in that, The computer program is used to execute the intelligent fault diagnosis method for photovoltaic equipment as described in any one of claims 1-4.

7. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the intelligent fault diagnosis method for photovoltaic equipment as described in any one of claims 1-4.

Images