A photovoltaic intelligent operation and maintenance system and method
By combining edge computing and cloud collaboration with blockchain evidence storage, a closed-loop update of the photovoltaic power station model is achieved, solving the problem that static models cannot adapt to changes in equipment and environment, and improving the accuracy and efficiency of operation and maintenance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DAJIAN (HANGZHOU) DIGITAL TECHNOLOGY CO LTD
- Filing Date
- 2026-03-10
- Publication Date
- 2026-06-05
AI Technical Summary
In the fault diagnosis and operation and maintenance management of existing photovoltaic power plants, the statically deployed machine learning models cannot be iteratively updated according to on-site feedback and data changes, resulting in a decrease in model accuracy and frequent false alarms and missed alarms.
A smart photovoltaic operation and maintenance system is constructed by collecting and preprocessing data in real time through edge computing nodes, building and deploying a baseline model in the cloud, performing anomaly detection at the edge, conducting attribution analysis in the cloud, and using blockchain evidence storage to drive the generation of operation and maintenance work orders, and updating the model in a closed loop based on on-site feedback.
This improved the long-term accuracy of the model, reduced the false alarm and false negative rates, ensured the transparency and traceability of the operation and maintenance process, and built an efficient and accurate photovoltaic smart operation and maintenance system.
Smart Images
Figure CN122155691A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of photovoltaic operation and maintenance technology, and more specifically, to a photovoltaic intelligent operation and maintenance system and method. Background Technology
[0002] As a crucial component of renewable energy, photovoltaic power plants are typically built in remote areas with open spaces and abundant sunshine. The equipment is exposed to the outdoors for extended periods, making it susceptible to various factors such as weather changes, component aging, and dust accumulation, which can lead to decreased power generation efficiency or even system shutdowns. Therefore, efficient and precise operation and maintenance management of photovoltaic power plants is key to ensuring their stable operation.
[0003] In existing technologies, some solutions attempt to introduce machine learning models for fault diagnosis and operation and maintenance management of photovoltaic power plants. These typically employ a "cloud-edge" separation model, where a diagnostic model is trained in the cloud using historical data and then deployed to the edge for real-time monitoring. However, this static deployment model has significant drawbacks: once trained in the cloud, the model remains fixed and cannot be iteratively updated based on actual on-site feedback and new data. As photovoltaic power plants operate over time, equipment performance degrades, and environmental conditions change dynamically. Static models struggle to adapt to this long-term evolution, leading to a gradual decline in model accuracy and frequent false alarms and missed alarms. Therefore, how to construct a smart photovoltaic operation and maintenance system that can continuously optimize the model based on on-site operation and maintenance feedback and achieve closed-loop model updates has become a pressing issue for the industry. Summary of the Invention
[0004] content This application provides a photovoltaic intelligent operation and maintenance system and method, which can build a photovoltaic intelligent operation and maintenance system that can continuously optimize the model based on on-site operation and maintenance feedback and realize closed-loop update of the model.
[0005] Firstly, this application provides a photovoltaic smart operation and maintenance method, comprising the following steps: Edge computing nodes collect operational and environmental data from photovoltaic strings, preprocess the operational and environmental data, and then transmit the preprocessed data to the cloud. The cloud-based system trains the preprocessed data to construct a baseline model of power generation behavior that characterizes the normal operation of the photovoltaic power station, and then distributes and deploys the baseline model to the edge computing node. The edge computing node calls the baseline model deployed locally to perform anomaly comparison on the real-time collected current running data and environmental data. When the data is detected to deviate from the baseline model by more than a preset threshold, the identified abnormal data is uploaded to the cloud. The cloud receives the abnormal data, performs attribution analysis on the abnormal data, and generates an explanation report that includes the fault type and the contribution of key influencing factors. The explanation report includes the attribution results of this anomaly. The cloud package and upload the explanation report and abnormal data to the blockchain node for evidence storage. The blockchain node automatically triggers the work order generation process according to the rules set in the pre-deployed smart contract, pushes the attribution results to the operation and maintenance work order management system, generates an operation and maintenance work order containing the attribution results, and dispatches the work order to the operation and maintenance terminal. The cloud receives the on-site processing results from the operation and maintenance terminal, corrects the parameters of the power generation behavior baseline model based on the on-site processing results, and then sends the corrected model back to the edge computing node to complete the closed-loop update.
[0006] In some embodiments, correcting the parameters of the power generation behavior baseline model based on the on-site processing results further includes: The cloud receives the on-site processing results fed back by the operation and maintenance terminal. The on-site processing results include fault type labels and corresponding abnormal data segments. The abnormal data segments with fault type labels are stored in the fault diagnosis knowledge base for optimizing and updating the fault attribution analysis model. Based on the newly added and confirmed fault-free operation data and environmental data, the cloud performs incremental training or parameter iteration on the power generation behavior baseline model and distributes the iteratively updated baseline model to the edge computing nodes.
[0007] In some embodiments, a blockchain node automatically triggers a work order generation process according to rules set in a pre-deployed smart contract, pushes the attribution results to the operation and maintenance work order management system, generates an operation and maintenance work order containing the attribution results, and dispatches the work order to the operation and maintenance terminal, specifically including: The cloud platform uploads the explanation report and abnormal data to the blockchain node for evidence storage. After receiving the evidence storage data and completing the hash verification and on-chain evidence storage, the smart contract on the blockchain node triggers an event according to preset rules, and pushes the attribution results and abnormal data to the external operation and maintenance work order management system through an oracle. The maintenance work order management system generates formal maintenance work orders based on the received information, combined with personnel status, spare parts inventory, and geographical location information, and dispatches them to the designated maintenance terminal.
[0008] In some embodiments, the edge computing node invokes the locally deployed baseline model to perform anomaly comparisons on the real-time collected current operating data and environmental data, specifically including: Edge computing nodes collect current environmental data in real time and input it into the locally deployed power generation behavior baseline model to obtain the baseline operating value that the string should achieve under the current environmental conditions; The edge computing node compares the real-time collected current operating data with the baseline operating value; If the deviation between the actual operating data and the baseline operating value exceeds a preset threshold, the data is determined to be abnormal.
[0009] In some embodiments, the cloud receives the abnormal data, performs attribution analysis on the abnormal data, and generates an explanatory report including the fault type and the contribution of key influencing factors, specifically including: The cloud receives the abnormal data and simultaneously retrieves the baseline reference value of the string under the same operating conditions, as well as the real-time operating data of other strings in the same photovoltaic power station that are operating normally under the same environmental conditions. The abnormal data, baseline values, and real-time operating data of normal sequences are used as inputs to call the fault attribution analysis model for diagnosis, generating an explanatory report that includes the fault type and the contribution of key influencing factors.
[0010] In some embodiments, the preprocessed data is used in the cloud to train a baseline model of power generation behavior characterizing the normal operation of a photovoltaic power plant, which further includes: The cloud cleans the received preprocessed data, removing data labels that include equipment failure, communication interruption, or shutdown maintenance status, and selecting data that only represents the fault-free operation of the equipment as the training sample set.
[0011] In some embodiments, the edge computing node collects operational data and environmental data from the photovoltaic string, preprocesses the operational data and environmental data, and then transmits the preprocessed data to the cloud, specifically including: Edge computing nodes perform cleaning and normalization preprocessing on the collected raw data; The preprocessed full data is cached on edge nodes, and based on network status and data importance, some or all of the preprocessed data is transmitted to the cloud during off-peak periods. At the same time, real-time anomaly comparison is performed locally, and only the identified abnormal data and related environmental context information are uploaded in real time.
[0012] Secondly, this application provides a photovoltaic intelligent operation and maintenance system, which includes: The data acquisition module, deployed on an edge computing node, is used to collect operating data and environmental data of the photovoltaic string. After preprocessing the operating data and environmental data, the preprocessed data is transmitted to the cloud. The processing module is used to train the preprocessed data in the cloud, construct a baseline model of power generation behavior that characterizes the normal operation of the photovoltaic power station, and distribute and deploy the baseline model to the edge computing node. The processing module is also used for edge computing nodes to call the locally deployed baseline model, perform anomaly comparison on the real-time collected current running data and environmental data, and upload the identified abnormal data to the cloud when the data deviates from the baseline model by more than a preset threshold. The processing module is also used to receive the abnormal data in the cloud, perform attribution analysis on the abnormal data, and generate an explanatory report that includes the fault type and the contribution of key influencing factors. The processing module is also used to package and upload the explanation report and abnormal data to the blockchain node for evidence storage in the cloud. The blockchain node automatically triggers the work order generation process according to the rules set in the pre-deployed smart contract, pushes the attribution results to the operation and maintenance work order management system, generates an operation and maintenance work order containing the attribution results, and dispatches the work order to the operation and maintenance terminal. The execution module is used to receive the on-site processing results from the operation and maintenance terminal in the cloud, correct the parameters of the power generation behavior baseline model according to the on-site processing results, and then send the corrected model back to the edge computing node to complete the closed-loop update.
[0013] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-mentioned photovoltaic smart operation and maintenance method.
[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described photovoltaic smart operation and maintenance method.
[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: This application utilizes edge computing nodes to collect and preprocess data in real time, constructs and distributes a baseline model of power generation behavior in the cloud, and achieves intelligent anomaly detection at the edge. When an anomaly occurs, the cloud performs attribution analysis and uses blockchain evidence storage to drive the generation of automated operation and maintenance work orders. Simultaneously, based on the on-site processing results fed back from the operation and maintenance terminal, the baseline model undergoes closed-loop correction and iterative updates. This solution solves the problem that static models cannot be continuously optimized based on on-site feedback. By updating the model in a closed loop to adapt to equipment degradation and environmental changes, it improves the long-term accuracy of the model, reduces false alarms and missed alarms, and ensures data immutability and transparent and traceable operation and maintenance processes through blockchain. Ultimately, it constructs an efficient and accurate intelligent photovoltaic operation and maintenance system. Attached Figure Description
[0016] Figure 1 This is an exemplary flowchart of a photovoltaic smart operation and maintenance method according to some embodiments of this application; Figure 2This is an exemplary flowchart illustrating the construction of a baseline model of power generation behavior characterizing the normal operating state of a photovoltaic power plant, based on some embodiments of this application. Figure 3 This is a schematic diagram of the structure of a photovoltaic smart operation and maintenance system according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a computer device for implementing a photovoltaic smart operation and maintenance method according to some embodiments of this application. Detailed Implementation
[0017] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific implementation methods. (Reference) Figure 1 The figure is an exemplary flowchart of a photovoltaic smart operation and maintenance method according to some embodiments of this application. The method mainly includes the following steps: In step 101, the edge computing node collects the operating data and environmental data of the photovoltaic string, preprocesses the operating data and environmental data, and transmits the preprocessed data to the cloud.
[0018] The operational data of the photovoltaic (PV) string may include real-time indicators such as current, voltage, and power, while environmental data may include parameters such as light intensity, temperature, and humidity. In some embodiments, the edge computing node collects the operational and environmental data of the PV string, preprocesses the data, and then transmits the preprocessed data to the cloud. Specifically, this can be achieved in the following manner: Edge computing nodes perform cleaning and normalization preprocessing on the collected raw data. Specifically, data cleaning can be done by removing noise points and filling missing values, while standardizing the data to a uniform dimension. Normalization can be done by mean-standard deviation normalization or maximum absolute value normalization. This application does not make specific limitations on this. The preprocessed full data is cached on edge nodes, and based on network status and data importance, some or all of the preprocessed data is transmitted to the cloud during off-peak periods. At the same time, real-time anomaly comparison is performed locally, and only the identified abnormal data and related environmental context information are uploaded in real time.
[0019] It should be noted that off-peak periods can be dynamically determined by monitoring network bandwidth, for example, by uploading in batches at night or during periods of low load, in order to reduce transmission latency and costs.
[0020] In step 102, the cloud trains the preprocessed data to construct a baseline model of power generation behavior that characterizes the normal operation of the photovoltaic power station, and then distributes and deploys the baseline model to the edge computing node.
[0021] The power generation behavior baseline model refers to a model built based on machine learning algorithms (such as random forests or neural networks) to predict the normal power generation behavior of photovoltaic strings under given environmental conditions. In some embodiments, the preprocessed data is trained in the cloud to construct a power generation behavior baseline model characterizing the normal operating state of the photovoltaic power station, which may further include: The cloud cleans the received preprocessed data, removing data labels that include equipment failure, communication interruption, or shutdown maintenance status, and selecting data that only represents the equipment's fault-free operation status as the training sample set. In practice, this can be achieved through label filtering (such as excluding data labeled "fault" or "maintenance") to ensure that the training set only contains data that is operating normally, thus avoiding model bias.
[0022] When building the model, the cloud can use historical preprocessed data for model training. For example, it can use environmental data as input and predicted normal operation data of photovoltaic strings as output to build a baseline for power generation behavior. Subsequently, the trained model is distributed to edge computing nodes through a secure channel for local deployment.
[0023] For specific implementation, refer to Figure 2 The cloud-based training of the preprocessed data to construct a baseline model of power generation behavior characterizing the normal operation of a photovoltaic power station can be achieved in the following manner: First, in step 1021, data preparation and filtering are performed, which involves receiving cleaned and normalized preprocessed data from the edge nodes. The preprocessed data includes photovoltaic operation data: current, voltage, and power; environmental data: light intensity, ambient temperature, module temperature, and humidity; and time characteristics: time period, season, etc. Normal operation samples are filtered out, removing data from abnormal periods such as faults, maintenance, and communication interruptions. For data from extreme weather periods such as typhoons and blizzards, only abnormal data related to equipment faults are removed, retaining only the normal operation data of fault-free equipment. Only stable and fault-free operation data of the photovoltaic power station is retained as the training set, and only normal samples are retained through data filtering. Then, the data is divided, that is, the normal samples are divided into a training set and a validation set according to time or random methods. For example, 70%-80% of the normal samples are divided into the training set, and 20%-30% of the normal samples are divided into the validation set.
[0024] Then, in step 1022, feature engineering is performed, where the basic features directly use the normalized original monitoring features (such as real-time irradiance, module temperature, and output power); then, core derived features of the photovoltaic scenario are constructed (such as irradiance, power ratio, temperature correction coefficient, and hourly power generation trend) to improve the model's ability to represent normal behavior; then, feature screening is performed, that is, redundant features are eliminated through correlation analysis and variance analysis, such as eliminating irrelevant environmental parameters that are not significantly related to power, thereby reducing model complexity.
[0025] Next, in step 1023, model selection and initialization are performed, initializing model parameters (such as the depth of the tree model, learning rate, number of layers in the neural network, and number of neurons). In specific implementation, models that adapt to the nonlinear operating characteristics of photovoltaics are preferred. Among them, the mainstream in industry are gradient boosting trees (XGBoost / LightGBM) or neural networks (MLP / CNN). It should be noted that gradient boosting trees are suitable for small samples, easy to interpret, and suitable for quickly building baselines; while neural networks are suitable for multi-dimensional and highly dynamic photovoltaic data, with higher fitting accuracy. The choice can be made according to the actual situation, and no specific limitation is made here. In step 1024, model training is performed, where the training objective is to predict the "normal output power / current" (i.e., baseline value) of a photovoltaic power station using environmental features (e.g., light intensity, temperature) as input. The training process iteratively optimizes the model using training set data, using the "error between predicted value and actual normal value (e.g., MAE, RMSE)" as the loss function to minimize the error. The model performance is monitored using a validation set to avoid overfitting (e.g., training is stopped when the validation set error increases continuously).
[0026] In step 1025, model verification and calibration are performed, which means verifying the model with independent normal operating data: if the error between the predicted baseline value and the actual value is within the engineering allowable range (e.g., ≤5%), the model is usable; the specific implementation of the model also requires dynamic calibration, that is, updating the model parameters periodically with new normal data according to the season / equipment aging degree to adapt to the degradation characteristics of photovoltaic modules.
[0027] In step 1026, the baseline model is deployed, which means that the trained and validated power generation behavior baseline model is distributed to the edge computing nodes of the photovoltaic power station in the form of a lightweight deployment package through a secure communication channel. After the edge nodes complete the local loading and environment adaptation of the model, it is deployed to the local inference engine of the edge nodes. The deviation threshold is dynamically set according to the operating years of the power station, the degree of equipment aging, and seasonal characteristics. The threshold range is 5%-10% of the baseline operating value. The threshold can be iteratively optimized synchronously based on the false alarms and missed alarms fed back by operation and maintenance. When the deviation between the real-time power generation data and the predicted value of the locally deployed baseline model exceeds the set threshold, it is judged as an anomaly and an operation and maintenance warning is triggered.
[0028] In some embodiments, the baseline model is distributed and deployed to the edge computing node, i.e., the cloud has completed the training and verification of the power generation behavior baseline model. The model is distributed to the edge computing node of the photovoltaic power station in the form of a lightweight deployment package through a secure communication channel. After the edge node completes the local loading and environment adaptation of the model, it is deployed to the local inference engine so that it can be called independently to realize local anomaly comparison and analysis of real-time collected data.
[0029] In step 103, the edge computing node calls the locally deployed baseline model to perform anomaly comparison on the real-time collected current running data and environmental data. When the data is detected to deviate from the baseline model by more than a preset threshold, the identified abnormal data is uploaded to the cloud.
[0030] In some embodiments, the edge computing node invokes the locally deployed baseline model to perform anomaly comparisons on the real-time collected current operating data and environmental data. Specifically, this can be done in the following ways: Edge computing nodes collect real-time environmental data and input it into the locally deployed power generation behavior baseline model to obtain the baseline operating value that the string should achieve under the current environmental conditions. That is, the edge computing nodes continuously collect on-site environmental data such as light, temperature, and humidity, and input this data into the locally deployed power generation behavior baseline model. Based on the learned normal operation rules, the model calculates and outputs the normal operation value that the photovoltaic string should achieve under the environmental conditions, i.e., the baseline operating value. The edge computing node compares the real-time collected current operating data with the baseline operating value. That is, the edge computing node compares the real-time collected actual operating data of the photovoltaic string with the baseline operating value calculated by the baseline model to clarify the deviation between the two. If the deviation between the actual operating data and the baseline operating value exceeds the preset threshold, the data is determined to be abnormal. In practice, the preset threshold can be dynamically adjusted based on historical experience, for example, set to 5%-10% of the baseline value, in order to balance sensitivity and false alarm rate.
[0031] In step 104, the cloud receives the abnormal data, performs attribution analysis on the abnormal data, and generates an explanatory report that includes the fault type and the contribution of key influencing factors.
[0032] In some embodiments, the cloud receives the abnormal data, performs attribution analysis on the abnormal data, and generates an explanatory report containing the fault type and the contribution of key influencing factors. Specifically, this can be done in the following ways: The cloud receives the abnormal data and simultaneously retrieves the baseline reference value of the string under the same operating conditions, as well as the real-time operating data of other strings in the same photovoltaic power station that are operating normally under the same environmental conditions. That is, after the cloud receives the abnormal data uploaded by the edge node, it will synchronously retrieve the baseline reference value of the photovoltaic string under the same operating conditions, and at the same time extract the real-time operating data of other strings in the same photovoltaic power station that are operating normally under the same environmental conditions, to prepare data support for subsequent attribution analysis. This will not be elaborated here. The abnormal data, baseline values, and real-time operating data of normal photovoltaic (PV) modules are used as inputs to call a fault attribution analysis model for diagnosis. This generates an explanatory report containing the fault type and the contribution of key influencing factors. The fault attribution analysis model determines the fault type of PV equipment malfunctions, quantifies the contribution of each influencing factor, and outputs the fault cause analysis results. Specifically, the fault attribution analysis model can use SHAP (SHapley Additive exPlanations) or LIME methods to calculate the contribution of each influencing factor (such as dust cover or module aging). A classification model is then used to determine the PV fault type. This classification model can be, for example, an existing classification model such as a random forest or support vector machine. Finally, an explanatory report containing the fault type and the contribution of key influencing factors can be output, such as "Fault type: Module shading, Contribution: Light factor 40%, Temperature factor 30%". This is just an example.
[0033] In step 105, the cloud packages and uploads the explanation report and abnormal data to the blockchain node for evidence storage. The blockchain node automatically triggers the work order generation process according to the rules set in the pre-deployed smart contract, pushes the attribution results to the operation and maintenance work order management system, generates an operation and maintenance work order containing the attribution results, and dispatches the work order to the operation and maintenance terminal.
[0034] In some embodiments, a blockchain node automatically triggers a work order generation process according to rules set in a pre-deployed smart contract, pushes the attribution results to the operation and maintenance work order management system, generates an operation and maintenance work order containing the attribution results, and dispatches the work order to the operation and maintenance terminal. Specifically, this can be done in the following ways: The cloud platform uploads the explanation report and abnormal data to the blockchain node for evidence storage. After receiving the evidence data and completing hash verification and on-chain evidence storage, the smart contract on the blockchain node triggers an event according to preset rules and pushes the attribution results and abnormal data to the external operation and maintenance work order management system through the off-chain trusted service interface. If it is necessary to achieve two-way trusted interaction between on-chain and off-chain, the off-chain data such as the work order execution results and processing completion status of the operation and maintenance work order management system can be trustedally stored on the blockchain through a blockchain oracle. That is, after the smart contract of the blockchain node receives the evidence data of abnormal data and attribution analysis results, it completes the data authenticity verification and tamper-proof evidence storage, and triggers the corresponding event according to the pre-set rules to realize the linkage between the blockchain and the work order system. The maintenance work order management system generates formal maintenance work orders based on the received information, combined with personnel status, spare parts inventory, and geographical location information, and dispatches them to the designated maintenance terminal. The smart contract can preset rules such as "if the fault type is a component fault, it will be dispatched to the nearest maintenance personnel first". This is only an example and is not intended to limit the specific application. Among them, the rule-triggered event is a pre-set triggering rule related to photovoltaic fault operation and maintenance in the smart contract. When the chain receives the matching evidence data and completes the verification, the corresponding preset event is automatically triggered, which starts the linkage process for subsequent data push and work order generation.
[0035] In step 106, the cloud receives the on-site processing results from the operation and maintenance terminal, corrects the parameters of the power generation behavior baseline model based on the on-site processing results, and then sends the corrected model back to the edge computing node to complete the closed-loop update.
[0036] In some embodiments, correcting the parameters of the power generation behavior baseline model based on the on-site processing results further includes: The cloud receives the on-site processing results from the maintenance terminal, which include fault type labels and corresponding abnormal data segments. The abnormal data segments with fault type labels are stored in the fault diagnosis knowledge base. Based on the newly added labeled data, the fault attribution analysis model is incrementally trained and its parameters optimized to improve the accuracy of fault diagnosis and attribution analysis. Furthermore, based on newly added and confirmed fault-free operating data and environmental data, the cloud performs incremental training or parameter iteration on the power generation behavior baseline model and distributes the iteratively updated baseline model to edge computing nodes. Incremental training can be achieved through… Online learning algorithms are used, such as fine-tuning parameters on the basis of the original model, to adapt to equipment degradation. That is, if the on-site handling result is a confirmed fault and repair is completed: abnormal data segments with fault type labels are stored in the fault diagnosis knowledge base, and the fault attribution analysis model is incrementally trained and the parameters are optimized based on the newly added labeled data to improve the accuracy of fault diagnosis and attribution analysis; at the same time, the fault-free operation data after fault repair is included in the training set of the power generation behavior baseline model for incremental training, and the core parameters of the model such as equipment aging coefficient and environmental correlation weight are adjusted in a synchronous manner to adapt to the natural degradation characteristics of photovoltaic modules. If the on-site processing result is a false alarm by the model (data deviates from the baseline but the equipment is not faulty): include the confirmed fault-free operating data into the training set, fine-tune the environmental feature weights and anomaly detection thresholds in the corresponding scenario of the model, and reduce the false alarm rate of similar scenarios. If the on-site processing result is a model miss (the equipment has a fault but the model does not recognize it): supplement and label the abnormal data segments corresponding to the fault type, store them in the fault diagnosis knowledge base, optimize the fault feature extraction logic and anomaly judgment sensitivity of the model, and supplement them to the training set to complete the model iteration; and, based on the newly added and confirmed fault-free operating data and environmental data, perform incremental training or parameter iteration on the power generation behavior baseline model in the cloud, and distribute the iteratively updated baseline model to the edge computing nodes. The incremental training can adopt online learning algorithms, such as fine-tuning the parameters on the basis of the original model to adapt to equipment degradation.
[0037] In practice, after on-site fault repair, maintenance personnel upload the results via their terminals, including confirmation that the fault has been eliminated, the actual replacement parts, and operational records. The cloud system first receives and verifies this feedback to confirm that the fault has been completely resolved. Based on the verified results, the cloud-based power generation behavior baseline model performs targeted parameter backtracking and optimization. For example, if the power degradation is found to be caused by component aging, the model adjusts the aging coefficient; if it is a misjudgment of the environment, the environmental correlation weights are corrected. This step uses real maintenance data to "correct" and "strengthen" the model, making it more closely reflect the actual operation of the power plant. Simultaneously, abnormal data segments with fault type labels are stored in the fault diagnosis knowledge base. Based on the newly added labeled data, incremental training and parameter optimization are performed on the fault attribution analysis model to improve the accuracy of fault diagnosis. The optimized model, after parameter correction, is repackaged and securely deployed to the edge computing nodes of the photovoltaic power plant. The edge nodes unload the old model and load and replace it with the latest version. Thus, the entire process forms a complete data loop: from anomaly detection at the edge nodes, to cloud-based analysis and attribution, work order dispatch, on-site handling and feedback, and finally, feedback and model updates. This enables the power generation behavior baseline model to continuously learn and evolve, accurately adapting to dynamic characteristics such as aging power plant equipment and environmental changes over the long term, significantly improving the accuracy of anomaly detection and the efficiency of intelligent operation and maintenance.
[0038] In another aspect, in some embodiments, this application provides a photovoltaic intelligent operation and maintenance system, referring to... Figure 3 The figure shows a photovoltaic intelligent operation and maintenance system 302 and an execution module 303 according to some embodiments of this application, which are described below: The acquisition module 301, in this application, is deployed on an edge computing node to collect the operating data and environmental data of the photovoltaic string. After preprocessing the operating data and environmental data, the preprocessed data is transmitted to the cloud. Processing module 302, in this application, is mainly used to train the preprocessed data in the cloud, construct a baseline model of power generation behavior that characterizes the normal operation state of the photovoltaic power station, and distribute and deploy the baseline model to the edge computing node; The processing module 302 is also used for the edge computing node to call the locally deployed baseline model, perform anomaly comparison on the real-time collected current running data and environmental data, and upload the identified abnormal data to the cloud when the data deviates from the baseline model by more than a preset threshold. In addition, the processing module 302 described in this application is also used to receive the abnormal data in the cloud, perform attribution analysis on the abnormal data, and generate an explanatory report containing the fault type and the contribution of key influencing factors. In addition, the processing module 302 described in this application is also used to package and upload the explanation report and abnormal data to the blockchain node for evidence storage in the cloud. The blockchain node automatically triggers and generates an operation and maintenance work order containing attribution results according to the rules set in the pre-deployed smart contract, and dispatches the work order to the operation and maintenance terminal. The execution module 303 in this application is mainly used to receive the field processing results fed back by the operation and maintenance terminal in the cloud, correct the parameters of the power generation behavior baseline model according to the field processing results, and then send the corrected model back to the edge computing node to complete the closed-loop update.
[0039] In addition, this application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described photovoltaic smart operation and maintenance method.
[0040] In some embodiments, reference Figure 4 The figure is a schematic diagram of the structure of a computer device for implementing a photovoltaic intelligent operation and maintenance method according to some embodiments of this application. The method in the above embodiments can be implemented through... Figure 4 The computer device shown is used to implement this, and the computer device 400 includes at least one processor 401, a communication bus 402, a memory 403, and at least one communication interface 404.
[0041] The processor 401 may be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices used to control the execution of the photovoltaic smart operation and maintenance method in this application.
[0042] The communication bus 402 may include a path for transmitting information between the aforementioned components.
[0043] The memory 403 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or it may be an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), a magnetic disk or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. The memory 403 may exist independently and be connected to the processor 401 via a communication bus 402. The memory 403 may also be integrated with the processor 401.
[0044] The memory 403 stores program code for executing the scheme of this application, and its execution is controlled by the processor 401. The processor 401 executes the program code stored in the memory 403. The program code may include one or more software modules. In the above embodiments, the construction of the power generation behavior baseline model can be achieved by the processor 401 and one or more software modules in the program code in the memory 403.
[0045] Communication interface 404 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area network (WLAN), etc.
[0046] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).
[0047] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.
[0048] In addition, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described photovoltaic smart operation and maintenance method.
[0049] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.
[0050] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of the invention. Therefore, if these modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.
Claims
1. A photovoltaic intelligent operation and maintenance method, characterized in that, Includes the following steps: Edge computing nodes collect operational and environmental data from photovoltaic strings, preprocess the operational and environmental data, and then transmit the preprocessed data to the cloud. The cloud-based system trains the preprocessed data to construct a baseline model of power generation behavior that characterizes the normal operation of the photovoltaic power station, and then distributes and deploys the baseline model to the edge computing node. The edge computing node calls the baseline model deployed locally to perform anomaly comparison on the real-time collected current running data and environmental data. When the data is detected to deviate from the baseline model by more than a preset threshold, the identified abnormal data is uploaded to the cloud. The cloud receives the abnormal data, performs attribution analysis on the abnormal data, and generates an explanation report that includes the fault type and the contribution of key influencing factors. The explanation report includes the attribution results of this anomaly. The cloud package and upload the explanation report and abnormal data to the blockchain node for evidence storage. The blockchain node automatically triggers the work order generation process according to the rules set in the pre-deployed smart contract, pushes the attribution results to the operation and maintenance work order management system, generates an operation and maintenance work order containing the attribution results, and dispatches the work order to the operation and maintenance terminal. The cloud receives the on-site processing results from the operation and maintenance terminal, corrects the parameters of the power generation behavior baseline model based on the on-site processing results, and then sends the corrected model back to the edge computing node to complete the closed-loop update.
2. The photovoltaic intelligent operation and maintenance method according to claim 1, characterized in that, Based on the on-site processing results, the parameters of the power generation behavior baseline model are corrected, further including: The cloud receives the on-site processing results fed back by the operation and maintenance terminal. The on-site processing results include fault type labels and corresponding abnormal data segments. The abnormal data segments with fault type labels are stored in the fault diagnosis knowledge base for optimizing and updating the fault attribution analysis model. Based on the newly added and confirmed fault-free operation data and environmental data, the cloud performs incremental training or parameter iteration on the power generation behavior baseline model and distributes the iteratively updated baseline model to the edge computing nodes.
3. The photovoltaic intelligent operation and maintenance method according to claim 1, characterized in that, According to the rules set in the pre-deployed smart contract, the blockchain node automatically triggers the work order generation process, pushes the attribution results to the operation and maintenance work order management system, generates an operation and maintenance work order containing the attribution results, and dispatches the work order to the operation and maintenance terminal, specifically including: The cloud platform uploads the explanation report and abnormal data to the blockchain node for evidence storage. After receiving the evidence storage data and completing the hash verification and on-chain evidence storage, the smart contract on the blockchain node triggers an event according to preset rules, and pushes the attribution results and abnormal data to the external operation and maintenance work order management system through an oracle. The maintenance work order management system generates formal maintenance work orders based on the received information, combined with personnel status, spare parts inventory, and geographical location information, and dispatches them to the designated maintenance terminal.
4. The photovoltaic intelligent operation and maintenance method according to claim 1, characterized in that, The edge computing node invokes the locally deployed baseline model to perform anomaly comparisons on the real-time collected current operating data and environmental data, specifically including: Edge computing nodes collect current environmental data in real time and input it into the locally deployed power generation behavior baseline model to obtain the baseline operating value that the string should achieve under the current environmental conditions; The edge computing node compares the real-time collected current operating data with the baseline operating value; If the deviation between the actual operating data and the baseline operating value exceeds a preset threshold, the data is determined to be abnormal.
5. The photovoltaic intelligent operation and maintenance method according to claim 1, characterized in that, The cloud receives the abnormal data, performs attribution analysis on the abnormal data, and generates an explanatory report that includes the fault type and the contribution of key influencing factors, specifically including: The cloud receives the abnormal data and simultaneously retrieves the baseline reference value of the string under the same operating conditions, as well as the real-time operating data of other strings in the same photovoltaic power station that are operating normally under the same environmental conditions. The abnormal data, baseline values, and real-time operating data of normal sequences are used as inputs to call the fault attribution analysis model for diagnosis, generating an explanatory report that includes the fault type and the contribution of key influencing factors.
6. The photovoltaic intelligent operation and maintenance method according to claim 1, characterized in that, The cloud-based system trains the preprocessed data to construct a baseline model of power generation behavior characterizing the normal operation of a photovoltaic power plant. This process also includes: The cloud cleans the received preprocessed data, removing data labels that include equipment failure, communication interruption, or shutdown maintenance status, and selecting data that only represents the fault-free operation of the equipment as the training sample set.
7. The photovoltaic intelligent operation and maintenance method according to claim 1, characterized in that, Edge computing nodes collect operational and environmental data from photovoltaic (PV) modules. After preprocessing the operational and environmental data, the preprocessed data is transmitted to the cloud. Specifically, this includes: Edge computing nodes perform cleaning and normalization preprocessing on the collected raw data; The preprocessed full data is cached on edge nodes, and based on network status and data importance, some or all of the preprocessed data is transmitted to the cloud during off-peak periods. At the same time, real-time anomaly comparison is performed locally, and only the identified abnormal data and related environmental context information are uploaded in real time.
8. A photovoltaic intelligent operation and maintenance system, characterized in that, include: The data acquisition module, deployed on an edge computing node, is used to collect operating data and environmental data of the photovoltaic string. After preprocessing the operating data and environmental data, the preprocessed data is transmitted to the cloud. The processing module is used to train the preprocessed data in the cloud, construct a baseline model of power generation behavior that characterizes the normal operation of the photovoltaic power station, and distribute and deploy the baseline model to the edge computing node. The processing module is also used for edge computing nodes to call the locally deployed baseline model, perform anomaly comparison on the real-time collected current running data and environmental data, and upload the identified abnormal data to the cloud when the data deviates from the baseline model by more than a preset threshold. The processing module is also used to receive the abnormal data in the cloud, perform attribution analysis on the abnormal data, and generate an explanatory report that includes the fault type and the contribution of key influencing factors. The processing module is also used to package and upload the explanation report and abnormal data to the blockchain node for evidence storage in the cloud. The blockchain node automatically triggers the work order generation process according to the rules set in the pre-deployed smart contract, pushes the attribution results to the operation and maintenance work order management system, generates an operation and maintenance work order containing the attribution results, and dispatches the work order to the operation and maintenance terminal. The execution module is used to receive the on-site processing results from the operation and maintenance terminal in the cloud, correct the parameters of the power generation behavior baseline model according to the on-site processing results, and then send the corrected model back to the edge computing node to complete the closed-loop update.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the photovoltaic smart operation and maintenance method according to any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the photovoltaic smart operation and maintenance method according to any one of claims 1 to 7.