A Smart Integrated Operation and Maintenance Method and System for Photovoltaic Power Plants
By combining physical mechanisms and data-driven models, a hybrid diagnostic method has been developed to address the issues of inconsistent data collection and low fault diagnosis accuracy in photovoltaic power plant operation and maintenance. This has enabled intelligent integrated operation and maintenance of photovoltaic power plants, improving both efficiency and safety.
Patent Information
- Application Number
- CN202611097941.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-23
- Publication Date
- 2026-11-17
AI Technical Summary
Existing photovoltaic power plant operation and maintenance technologies suffer from problems such as inconsistent data collection, low accuracy of fault diagnosis, system silos, and insufficient security protection, making it difficult to achieve refined and automated management and control.
A hybrid model combining physical mechanisms and data-driven models is used for data fusion and fault diagnosis. Graph neural networks are used to analyze the mutual influence between devices to achieve self-healing control and closed-loop operation and maintenance management. A redundant communication network and a national cryptographic algorithm encryption system are also constructed.
It improves data quality and fault diagnosis accuracy, realizes closed-loop management from string-level monitoring to inverter control, reduces operation and maintenance costs and manual intervention, and enhances system integration, collaboration and security protection capabilities.
Smart Images

Figure CN123121982A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of photovoltaic power plant operation and maintenance technology, and in particular to an intelligent integrated operation and maintenance method and system for photovoltaic power plants. Background Technology
[0002] With the large-scale development of the photovoltaic industry, the installed capacity of ground-mounted centralized photovoltaic and distributed photovoltaic power stations continues to climb, leading to a significant increase in the volume of power station operation and maintenance. Traditional operation and maintenance models based on manual inspections and separate monitoring platforms are no longer adequate for the refined and automated management needs of large-scale power stations. Currently, photovoltaic operation and maintenance related technical solutions in the industry generally suffer from multi-dimensional technical deficiencies, hindering the improvement of power station power generation efficiency and operation and maintenance management levels.
[0003] The existing photovoltaic (PV) operation and maintenance sensing and data acquisition system has significant shortcomings: PV power plants are mostly deployed in complex operating conditions such as deserts, coastlines, and mountains. Dust, salt spray, and extreme temperature and humidity can cause measurement drift, data distortion, or even long-term failure of sensors such as irradiance, temperature, current, and voltage sensors. At the same time, different inverters, combiner boxes, and meteorological monitoring equipment belong to different manufacturers, and their hardware interfaces, data transmission protocols, and data storage formats are not standardized. Existing acquisition solutions only perform simple data translation and lack a unified protocol gateway and multi-source data fusion and calibration mechanism, which easily generates a large amount of abnormal, missing, and asynchronous invalid data, making it difficult to guarantee the quality of the underlying data source.
[0004] In terms of power plant data analysis and intelligent diagnosis, existing technologies fall into two main categories: one relies on simple thresholds, linear regression, and basic neural networks to establish a power generation prediction model, which only adopts a single data-driven modeling approach and does not combine the physical mechanism of photovoltaic modules. The model cannot adapt to complex operating conditions such as dust accumulation, aging, and partial shading, and has poor generalization ability. The other type of fault diagnosis method is mostly based on training machine learning models with independent time-series data of a single device. It can only judge anomalies based on data correlation and cannot characterize the electrical coupling relationship between strings, combiner boxes, and inverters. It is difficult to locate complex faults caused by the coupling of multiple factors such as hot spots, PID attenuation, and branch mismatch, and it cannot output an explanation of the underlying physical cause of the fault. The fault identification accuracy is low and false alarms and missed alarms occur frequently.
[0005] At the system architecture and operation and maintenance management level, most monitoring, fault diagnosis, operation and maintenance scheduling, and material management systems on the market are independent and separate platforms. Data from each system is not interconnected, forming typical information silos, which cannot achieve closed-loop collaboration of the entire process of "data collection - performance evaluation - fault diagnosis - operation and maintenance dispatch - resource scheduling". Operation and maintenance task scheduling relies solely on manual experience for scheduling, without intelligent optimization based on multiple constraints such as fault urgency, weather conditions, personnel qualifications, and spare parts inventory. As a result, the utilization rate of operation and maintenance manpower and material resources is low, and operation and maintenance costs remain high.
[0006] In terms of security and communication support, conventional operation and maintenance platforms often use single-line networks for communication links, and single-point failures can easily cause data interruptions across the entire domain. Data transmission often uses general encryption algorithms, which are not adapted to domestic industrial information security compliance requirements. They lack in-depth protection measures such as layered industrial firewalls, intrusion detection, fine-grained role and permission control, and tamper-proof operation logs. The massive amounts of power plant operation monitoring data and operation and maintenance business data are at risk of leakage, tampering, and unauthorized access. Summary of the Invention
[0007] The present invention aims to at least partially solve one of the technical problems in the related art.
[0008] Therefore, the first objective of this invention is to propose an intelligent integrated operation and maintenance method for photovoltaic power plants, comprising the following steps:
[0009] S1, Collect the operating status data of photovoltaic power station equipment, preprocess and fuse the data to generate a data stream with unified time alignment; S2, based on a hybrid model combining a physical mechanism model and a data-driven model, analyzes the data stream to diagnose the fault type, fault location, and fault severity of the equipment; S3, based on the fault diagnosis results, generates alarm information according to the preset classification rules, and executes self-healing control operations corresponding to the fault level; S4 generates maintenance work orders based on fault diagnosis results and self-healing control operations, and coordinates materials, personnel, and plans to execute closed-loop maintenance management.
[0010] In one embodiment of the present invention, S1 includes: Data from different manufacturers' devices in different formats is converted into a unified timestamp alignment format; outliers are identified and removed using a method based on the 3σ criterion and the isolated forest algorithm. For short-term data gaps, linear interpolation or historical data interpolation based on similar days can be used to fill in the gaps. The irradiance and power signals are smoothed by median filtering or wavelet denoising based on sliding window to obtain preprocessed data with quality labels. The unified data access gateway supports access to Modbus RTU / TCP, IEC 61850, MQTT and OPC-UA protocols, and automatically identifies device types and performs protocol parsing. The Kalman filter algorithm is used to dynamically estimate and eliminate noise in the time series data of irradiance, temperature, and power. For multi-source measurements of the same physical quantity, a confidence-based weighted fusion method is adopted. The weights are dynamically allocated according to the historical accuracy of each sensor and the current environmental state to generate a high-quality data stream with unified time alignment.
[0011] In one embodiment of the present invention, the method for performing dynamic estimation and noise cancellation is as follows: A state-space model of the photovoltaic system is established. The state at the next moment is predicted by the time update equation. The prediction result is corrected by the actual observation value through the measurement update equation to obtain the optimal estimate.
[0012] In one embodiment of the present invention, S2 includes: The five-parameter model of a single diode in a photovoltaic module is used as the physical mechanism model. The current-voltage relationship of the physical mechanism model is given by the implicit equation:
[0013] in, For photocurrent, This is the reverse saturation current of the diode. For series resistance, For parallel resistors, For diode ideality factor, Thermoelectric voltage; A Long Short-Term Memory (LSTM) network is used as the data-driven error compensation model. Environmental humidity, wind speed, estimated dust accumulation levels, irradiance, temperature, and the theoretical maximum power output from the physical model are used as input features to predict the error. The final theoretical power generation prediction is as follows:
[0014] in, This represents the theoretical maximum power output by the physical model. This is the error compensation value predicted by the LSTM model.
[0015] In one embodiment of the present invention, the fault type, fault location, and fault severity of the diagnostic device include: Each string, combiner box, and inverter in the power station is treated as a graph node. Electrical connections are represented as graph edges. Constructing a graph structure ; By using the message passing mechanism of a graph neural network to learn the mutual influence between devices, for the first... Layers, nodes The hidden state update process is as follows: , in, Employ the attention aggregation mechanism in graph attention networks; The real-time operating data of each node is used as the input node feature matrix. After being encoded by a multi-layer graph neural network, the fault type, fault location, fault severity, and cause explanation are output through graph-level readout or node-level classifier.
[0016] In one embodiment of the present invention, S3 includes: The fault levels are divided into three levels: Level 1, Level 2, and Level 3. For Level 1 faults, the fault is automatically recorded and an attempt is made to perform a software reset or switch to a backup sensor. For level 2 faults, the system will automatically adjust the operating strategy and generate a maintenance work order; For level 3 faults, alarms will be immediately pushed through multiple channels such as APP, SMS and email, and complete data before and after the fault will be recorded for analysis.
[0017] In one embodiment of the present invention, S4 includes: Based on equipment failure prediction results, equipment maintenance cycle, weather forecast, personnel availability and spare parts inventory, a genetic algorithm is used to optimize the solution of task priority, time window and resource constraints, and output maintenance work orders including task type, task content, suggested time, required personnel skills and required spare parts. Based on the skill requirements and urgency of the maintenance work order, automatically match personnel with the corresponding qualifications and currently light workload; Based on the failure rate statistics and remaining service life prediction of each piece of equipment, calculate the safety stock level and replenishment point of spare parts, and automatically generate a purchase order when the inventory is below the threshold.
[0018] In one embodiment of the present invention, it further includes: A redundant communication network is used to ensure data transmission between modules, and the data is encrypted using the national cryptographic SM2 / SM4 algorithm. At the same time, access control is based on roles to manage permissions and all operation logs are recorded.
[0019] To achieve the above objectives, a second aspect of the present invention provides an intelligent integrated operation and maintenance system for photovoltaic power plants, comprising: The data fusion module is used to collect operating status data of photovoltaic power station equipment, preprocess and fuse the data, and generate a data stream with unified time alignment. The hybrid diagnostic module is used to analyze the data stream based on a hybrid model that combines a physical mechanism model and a data-driven model, and to diagnose the fault type, fault location and fault severity of the equipment. The alarm self-healing module is used to generate alarm information according to preset classification rules based on the fault diagnosis results, and to execute self-healing control operations corresponding to the fault level. The closed-loop operation and maintenance module is used to generate operation and maintenance work orders based on fault diagnosis results and self-healing control operations, and coordinate materials, personnel and plans to execute closed-loop operation and maintenance management.
[0020] To achieve the above objectives, a third aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.
[0021] Compared with the prior art, the methods, systems, and storage media of this invention have the following advantages: (1) Data quality improvement: Through proactive data quality management and environmental adaptive dynamic calibration, the problem of inaccurate sensor data under extreme environments has been solved, and the reliability of data sources has been improved.
[0022] (2) Improved diagnostic accuracy: By combining physical models with data-driven hybrid modeling methods (single diode five-parameter model + LSTM / GNN), the accuracy and generalization ability of fault diagnosis are improved, and the accurate identification and cause explanation of complex faults such as hot spots and PID attenuation are realized.
[0023] (3) Closed-loop intelligent control: Through model-driven real-time control and hierarchical fault self-healing, closed-loop management from string-level monitoring to inverter control is realized, reducing manual intervention and improving response speed.
[0024] (4) System integration and collaboration: Through the integrated operation and maintenance decision-making platform, the management links such as material management, personnel scheduling, financial analysis, and knowledge base are connected, solving the problem of system silos and realizing global optimization.
[0025] (5) Comprehensive security protection: A three-dimensional security protection system has been built through redundant communication networks, national cryptographic algorithm encryption, multi-factor authentication and fine-grained permission management.
[0026] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0027] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of an intelligent integrated operation and maintenance method for a photovoltaic power plant according to an embodiment of the present invention; Figure 2 This is a structural diagram of an intelligent integrated operation and maintenance system for a photovoltaic power plant according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the overall system structure according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the data acquisition and processing fusion module structure according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the data storage and model training module structure according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the control optimization and fault diagnosis module according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the operation and maintenance decision-making and auxiliary management module according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the communication protection and security module according to an embodiment of the present invention. Detailed Implementation
[0028] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0030] The following description, with reference to the accompanying drawings, describes an intelligent integrated operation and maintenance method and system for photovoltaic power plants according to embodiments of the present invention.
[0031] Example 1 Figure 1 This is a flowchart of a data processing method according to an embodiment of the present invention.
[0032] like Figure 1 As shown, the data processing method includes the following steps: S1, collect the operating status data of photovoltaic power station equipment, preprocess and fuse the data to generate a high-quality data stream with unified time alignment.
[0033] The process of collecting operational status data from photovoltaic power plant equipment begins with the deployment of a sensor network at key equipment locations to acquire various electrical and environmental parameters reflecting the equipment's operational status in real time. The collected data then enters a preprocessing stage, designed to eliminate data quality issues caused by sensor errors, communication interference, or sudden environmental changes. This includes format standardization and timestamp alignment for data from different sources and formats, identifying and removing outliers exceeding physical limits or exhibiting abnormal rates of change, imputing missing data due to momentary interruptions, and smoothing high-frequency noise in the signal. Preprocessed data is assigned a quality label reflecting its reliability, allowing subsequent stages to differentiate processing based on data quality. After preprocessing, the data enters a fusion stage. This stage integrates heterogeneous data sources by using a unified protocol access mechanism compatible with multiple industrial communication protocols. Based on this, a time-series data fusion method is employed to dynamically estimate key parameters and suppress noise. Simultaneously, for multiple-source measurements of the same physical quantity, weighted fusion is performed based on the historical accuracy of each measurement source and the current environmental conditions to obtain a more accurate and stable estimate. Through the collaborative processing of data collection, preprocessing, and fusion described above, a high-quality data stream with unified time alignment and controllable quality is finally generated, providing a reliable data foundation for subsequent analysis.
[0034] As a specific implementation method, sensor networks can be deployed in key locations such as photovoltaic arrays, combiner boxes, inverters, and weather stations. Preprocessing can employ methods such as anomaly identification based on statistical criteria and isolated forest algorithms, handling missing values through linear interpolation or historical data imputation, and noise filtering methods such as sliding window filtering or wavelet denoising. The fusion stage can achieve dynamic estimation of time-series data through the Kalman filtering algorithm and use a confidence-based weighted fusion method to process multi-source measurement data.
[0035] This step, through systematic data acquisition, preprocessing, and fusion, effectively improves the accuracy, completeness, and consistency of the raw data, providing reliable data support for subsequent fault diagnosis and closed-loop control based on high-quality data streams, thereby significantly enhancing the analytical accuracy and decision-making reliability of the entire operation and maintenance system.
[0036] S2, based on a hybrid model combining a physical mechanism model and a data-driven model, analyzes the high-quality data stream to diagnose the fault type, fault location, and fault severity of the equipment.
[0037] In this step, a hybrid model integrating physical mechanisms and data-driven approaches is used to perform in-depth analysis of the high-quality data stream generated after preprocessing and fusion, in order to diagnose faults in photovoltaic power plant equipment. The core of this hybrid model lies in its approach: it does not rely solely on statistical data or a single physical principle, but rather organically combines a mechanistic model based on the equipment's physical characteristics with a data-driven model capable of learning complex nonlinear patterns from historical data. This allows for a comprehensive characterization and analysis of the equipment's operating status. Specifically, the hybrid model first uses a physical mechanism model to calculate the theoretical operating baseline of the equipment under ideal or standard conditions based on real-time environmental parameters (such as irradiance and temperature). This baseline reflects the inherent physical response characteristics of the equipment. Subsequently, the data-driven model is used to capture deviations between actual operating data and the theoretical baseline. These deviations often stem from complex factors that are difficult for the physical model to accurately describe, such as equipment aging, partial shading, uneven dust accumulation, or multi-factor coupling effects. By analyzing the patterns and characteristics of these deviations, the hybrid model can identify whether the equipment has deviated from its normal state and further infer the specific fault type, the precise location of the fault, and the severity of the fault.
[0038] As one implementation method, this hybrid model can specifically combine a single-diode five-parameter physical model of a photovoltaic module with a data-driven model based on a graph neural network. The physical model provides a theoretical power generation benchmark, and the current-voltage relationship of this physical mechanism model is given by the implicit equation:
[0039] in, For photocurrent, This is the reverse saturation current of the diode. For series resistance, For parallel resistors, For diode ideality factor, Thermoelectric voltage Graph neural networks, based on the topological connections of power plant equipment, learn the mutual influence between devices through a message passing mechanism, thereby achieving accurate diagnosis and location of complex faults such as hot spot effect, PID attenuation, string mismatch, and inverter IGBT faults.
[0040] This technical step significantly improves the accuracy and robustness of fault diagnosis by introducing a hybrid model that combines physical mechanisms with data-driven approaches. It can not only identify complex fault modes that are difficult to capture by traditional single models, but also provide more physically interpretable diagnostic results, thus providing a reliable basis for subsequent precise operation and maintenance decisions.
[0041] S3 generates alarm information according to preset classification rules based on the fault diagnosis results, and executes self-healing control operations corresponding to the fault level. Based on the fault diagnosis results, the system automatically assesses and classifies the severity of the fault according to preset grading rules, generates corresponding alarm information for different levels, and triggers self-healing control operations matching the fault level. These grading rules are pre-set based on the impact of the fault on equipment safety, power generation efficiency, and system stability, covering multiple levels from minor anomalies to severe faults. Alarm information is generated in formats adapted to different receiving terminals, including but not limited to system interface prompts, mobile application push notifications, SMS or email notifications, ensuring that maintenance personnel can be promptly informed of the fault status. The self-healing control operations automatically execute a series of preset countermeasures according to the fault level, aiming to quickly mitigate or eliminate the impact of the fault and reduce the need for manual intervention.
[0042] In one preferred implementation, for minor faults, the system can automatically perform operations such as data logging, sensor reset, or switching to a backup channel; for general faults, the system can automatically adjust the operating parameters or strategies of relevant equipment to maintain stable operation in degraded mode and generate maintenance work orders for subsequent processing; for severe faults, the system immediately executes emergency protection actions, such as cutting off the power supply to the faulty equipment, isolating the faulty area, or activating backup equipment, while simultaneously pushing high-priority alarms through multiple channels and recording complete data before and after the fault for analysis. Through this hierarchical self-healing mechanism, the system achieves closed-loop control from fault detection to automatic response, significantly improving the automation level and response speed of photovoltaic power plant operation and maintenance.
[0043] This step, through a combination of tiered alarms and self-healing control, automates and differentiates fault handling, effectively shortens fault response time, reduces manual inspection and intervention costs, and avoids secondary damage caused by improper handling, thereby improving the reliability and safety of photovoltaic power plant operation.
[0044] S4 generates maintenance work orders based on fault diagnosis results and self-healing control operations, and coordinates materials, personnel, and plans to execute closed-loop maintenance management. Based on fault diagnosis results and self-healing control operations, the system automatically generates maintenance work orders and coordinates materials, personnel, and maintenance plans based on these work orders to execute closed-loop maintenance management. Specifically, this step aims to transform the fault information and handling requirements output from the diagnosis and self-healing stages into executable maintenance tasks. By integrating multi-dimensional constraints such as material inventory status, personnel skills and availability, equipment maintenance cycles, and environmental conditions, it achieves optimized allocation of maintenance resources and task scheduling. The maintenance work order covers key elements such as task type, task content, suggested execution time, required personnel skills, and required spare parts, thus providing a unified information carrier for subsequent material allocation, personnel assignment, and scheduling. Regarding material coordination, the system automatically triggers spare parts demand assessment and procurement processes based on the fault diagnosis results and the type and severity of equipment involved in the self-healing operation, combined with spare parts inventory levels and supplier supply capabilities, ensuring the timely arrival of materials needed for repair. Regarding personnel coordination, the system automatically matches qualified personnel with relatively light workloads from the maintenance personnel database based on the skills and qualifications required by the work order and the urgency level, and can dynamically adjust shifts to handle urgent tasks. In terms of planning and coordination, the system comprehensively considers multiple factors such as equipment failure prediction results, equipment maintenance cycles, weather forecasts, personnel availability, and spare parts inventory. It employs optimization algorithms to solve for task priorities, time windows, and resource constraints, generating an optimized operation and maintenance plan. Through the collaborative management of materials, personnel, and plans, the system achieves a closed-loop process from fault diagnosis results to operation and maintenance execution, reducing manual intervention and improving operation and maintenance efficiency and resource utilization. For example, the system can generate work orders based on fault diagnosis results, including task type, required spare parts, and personnel skills. It then automatically checks spare parts inventory and triggers purchase orders, simultaneously assigning qualified personnel to perform the tasks and using genetic algorithms to optimize task scheduling.
[0045] This step, by automatically generating maintenance work orders and coordinating materials, personnel, and plans, achieves efficient transformation of fault diagnosis results into maintenance execution, significantly improving the automation level of maintenance management and the accuracy of resource allocation, reducing manual decision-making costs and response delays, thereby enhancing the closed-loop control capability and overall operational efficiency of photovoltaic power plant maintenance.
[0046] Example 2 like Figure 2 As shown, this invention proposes an intelligent integrated operation and maintenance system 10 for photovoltaic power plants, comprising: The data fusion module 100 is used to collect the operating status data of photovoltaic power station equipment, preprocess and fuse the data, and generate a data stream with unified time alignment. The hybrid diagnostic module 200 is used to analyze the data stream based on a hybrid model that combines a physical mechanism model and a data-driven model, and to diagnose the fault type, fault location and fault degree of the equipment. The alarm self-healing module 300 is used to generate alarm information according to preset classification rules based on the fault diagnosis results, and to execute self-healing control operations corresponding to the fault level. The closed-loop operation and maintenance module 400 is used to generate operation and maintenance work orders based on fault diagnosis results and self-healing control operations, and coordinate materials, personnel and plans to execute closed-loop operation and maintenance management.
[0047] Example 3 like Figure 3 As shown, this invention proposes an intelligent integrated operation and maintenance system for photovoltaic power plants, consisting of five core modules arranged in order of data flow and functional hierarchy: 1. Data acquisition and processing fusion module; 2. Data storage and model training module; 3. Control optimization and fault diagnosis module; 4. Operation and maintenance decision-making and auxiliary management module; and 5. Communication assurance and security protection module. Each module works collaboratively through standardized data interfaces, with the communication assurance and security protection module providing network communication and security guarantees for the other modules.
[0048] The system comprises five modules: Data Acquisition and Processing Fusion Module 1, which serves as the system's input, collecting and fusing operational status data from equipment in real time; Data Storage and Model Training Module 2, the system's foundation, storing and managing historical operating data of the power plant and training relevant models; Control Optimization and Fault Diagnosis Module 3, the core decision-making component, used for performance evaluation, optimized control, and fault diagnosis of equipment; Operation and Maintenance Decision and Auxiliary Management Module 4, the system's long-term decision-making component, used for intelligent management of operation and maintenance plans, materials, personnel, and finances; and Communication Assurance and Security Protection Module 5, providing network communication and security services to other modules. All modules interact through standardized data interfaces to ensure the overall coordinated operation of the system.
[0049] The internal structure and technical implementation details of each module will be explained below.
[0050] Module 1: Data Acquisition and Processing Fusion Module like Figure 4 As shown, this module is the system's input, responsible for real-time acquisition and fusion processing of the device's operating status data. The data acquisition and processing fusion module 1 includes a sensor network unit 11, a data preprocessing unit 12, and a data fusion unit 13.
[0051] (1) Sensor network unit The sensor network deployed for key equipment in photovoltaic power plants is arranged as follows: ① Photovoltaic array area: Deploy high-precision irradiance meter (accuracy ±2%), module backsheet temperature sensor (accuracy ±0.5℃), string current sensor (accuracy ±0.5%), and string voltage sensor.
[0052] ② Combiner box layer: Deploy combiner box monitoring units to collect current, voltage and surge protector status of each branch.
[0053] ③ Inverter layer: Deploy inverter status monitoring units to collect DC side input power, AC side output power, inverter efficiency, module temperature and IGBT status.
[0054] ④ Weather station: Deploy ambient temperature sensors, wind speed sensors, humidity sensors and dust concentration sensors.
[0055] The sensors are arranged according to the structural characteristics of the equipment and the parts prone to failure. For example, irradiance meters are deployed at the edge and center of the array to capture the effects of local shading, and temperature sensors are deployed near the inverter power module to monitor the heat dissipation status.
[0056] (2) Data preprocessing unit The collected data undergoes preliminary processing, including the following steps: ① Format unification conversion: Convert data of different formats from different manufacturers' devices into a unified timestamp aligned format.
[0057] ② Outlier identification: A method based on the 3σ criterion and the isolated forest algorithm is used to identify and remove outliers caused by sensor failure or communication interference.
[0058] ③ Missing value imputation: For short-term missing data, linear interpolation or imputation methods based on historical data of similar days are used to fill the missing data.
[0059] ④ Noise filtering: Median filtering or wavelet denoising based on sliding window is used to smooth the irradiance and power signals.
[0060] The preprocessed data is labeled with quality tags ("normal" / "suspicious" / "invalid") for easy use by subsequent modules.
[0061] (3) Data fusion unit Responsible for the fusion and processing of multi-source heterogeneous data: ① Protocol Access: Through a unified data access gateway, it supports access to multiple industrial protocols such as Modbus RTU / TCP, IEC 61850, MQTT, and OPC-UA, and automatically identifies device types and performs protocol parsing.
[0062] ② Time-series data fusion: The Kalman filter algorithm is used to dynamically estimate and eliminate noise in time-series data such as irradiance, temperature, and power. Specific steps include: establishing a state-space model of the photovoltaic system; predicting the state at the next time step using a time update equation; and correcting the prediction results using actual observations through a measurement update equation to obtain the optimal estimate.
[0063] ③ Multi-source measurement fusion: For multi-source measurements of the same physical quantity (such as data from multiple irradiance meters), a confidence-based weighted fusion method is adopted, which dynamically allocates weights according to the historical accuracy of each sensor and the current environmental state.
[0064] The merged data forms a unified, time-aligned, high-quality data stream, providing data support for subsequent modules.
[0065] Module 2: Data Storage and Model Training Module like Figure 5 As shown, this module is the foundation of the system, responsible for storing historical data and training various models. The data storage and model training module 2 includes a data storage unit 21, a data quality management unit 22, and a model training unit 23.
[0066] (1) Data storage unit Employing a distributed storage architecture (such as HDFS) and off-site backup strategies, storage is categorized as follows: ① Raw data: Raw values collected by the sensor without processing.
[0067] ② High-quality data after cleaning: Data that has undergone preprocessing and fusion.
[0068] ③ Fault case data: Includes complete data and fault labels for the time periods before and after the fault.
[0069] ④ Maintenance records: Operation records such as equipment repair, replacement, and cleaning.
[0070] A data version management mechanism is introduced, which generates a new version for each data cleaning or calibration operation and retains historical versions, facilitating data traceability and retrospective analysis.
[0071] (2) Data Quality Management Unit Establish a sensor performance database to record the model, installation location, historical accuracy, fault records, and performance under different environmental conditions for each sensor. The workflow of this unit is as follows: ① Combining environmental sensor data (temperature, humidity, dust concentration, etc.), the reliability of each sensor's current data is dynamically evaluated through multiple quality check rules, outputting a confidence score between 0 and 1. Specific check rules include: physical range check, determining whether the data exceeds the sensor's measurement range (e.g., if the irradiance meter reading exceeds the local astronomical theoretical upper limit, it is deemed invalid); rate of change check, determining whether the data change at adjacent times exceeds the maximum allowable rate of change (e.g., if the temperature changes by more than 10°C within 1 second, it is deemed suspicious); and persistence check, determining whether the data remains unchanged for a long time, suggesting sensor malfunction (e.g., if irradiance remains unchanged for several hours after sunrise, it is deemed invalid). For multiple sensors measuring the same physical quantity in the same area (e.g., multiple irradiance meters in the same area), their readings are further compared. If a sensor's reading consistently and significantly deviates from the population median (e.g., exceeding 3 times the standard deviation), its confidence score is reduced accordingly. Based on the combined check results of all rules, points are deducted for data that fail individual checks, ultimately outputting a comprehensive confidence score between 0 and 1.
[0072] ② For data with a confidence level below the threshold, mark it and trigger the sensor self-test process.
[0073] ③ For calibrable sensors (such as radiometers), real-time compensation is performed at the software level using historical calibration data and current environmental conditions. For example, when dust is detected on the surface of the radiometer, the measured value is corrected based on the degree of dust accumulation and the cumulative duration.
[0074] (3) Model training unit (completed) Responsible for training two types of core models: The first category is theoretical power generation prediction models for photovoltaic arrays. A hybrid modeling approach is used, combining physical mechanisms with data-driven methods: Underlying Physical Mechanism Model: Adopting a five-parameter model of a single diode in a photovoltaic module, the current-voltage (IV) relationship is given by the following implicit equation:
[0075] The meanings of each parameter are as follows: The component output current (A). The component output voltage (V); Photocurrent (A) is proportional to the current irradiance and reflects the number of charge carriers generated by the light. This is the reverse saturation current (A) of the diode, which is temperature-dependent and reflects the dark current characteristics. The series resistance (Ω) reflects losses such as electrode contact and semiconductor bulk resistance. The parallel resistance (Ω) reflects the power loss caused by the leakage current path. The diode ideality factor reflects how close the PN junction is to an ideal PN junction; For thermal voltage, where Boltzmann constant (1.38 × 10⁻⁶) - ²³ J / K), The component temperature (K) is the component temperature. The electron charge (1.6 × 10⁻⁶) - ¹ 9 C).
[0076] The model uses irradiance and component temperature As input, the theoretical IV curve is obtained by solving the above implicit equations, and then the theoretical maximum power point is extracted. That is, the theoretical maximum power generation capacity.
[0077] Among them, photocurrent The relationship with irradiance is as follows:
[0078] In the formula, This represents the current irradiance (W / m²). Irradiance under standard test conditions (1000 W / m²). Photocurrent under standard test conditions The temperature coefficient of short-circuit current. The component temperature is 25°C under standard test conditions.
[0079] Upper Layer – Data-Driven Error Compensation Model: Since the physical model cannot fully encompass the complex factors in actual operation (such as component aging, shading, uneven dust accumulation, etc.), a Long Short-Term Memory (LSTM) network is used to learn and compensate for the prediction errors of the physical model. Definition of Error The LSTM model uses estimates of ambient humidity, wind speed, and dust accumulation, as well as the physical model's inputs (irradiance, temperature) and outputs (…). As input features, for error A forecast was made. The final theoretical power generation forecast is as follows:
[0080] in, This is the error compensation value predicted by the LSTM model.
[0081] Training: Using normal operating condition data from historical data, with mean squared error (MSE) as the loss function:
[0082] The network parameters of the LSTM part are optimized using the backpropagation algorithm, while the calibration parameters of the physical model part remain unchanged.
[0083] The second type is the deep causal fault diagnosis model. It employs a graph neural network (GNN) architecture. Each string, combiner box, and inverter in the power station is treated as a graph node. Electrical connections are represented as graph edges. Constructing a graph structure It learns the mutual influence between devices through message passing mechanisms.
[0084] The core idea of graph neural networks is to update the representation of each node by iteratively aggregating information from its neighbors. For the th... Layers, nodes The hidden state update process is as follows:
[0085] in: For nodes In the The hidden state vector of the layer; For nodes The set of neighboring nodes; Aggregation functions (such as summation pooling, mean pooling, or attention mechanisms) are used to summarize information from neighboring nodes. This unit adopts the attention aggregation mechanism in Graph Attention Network (GAT) to assign different weights to different neighbors. Its advantage is that it can automatically learn the strength of mutual influence between devices. The update function is typically a multilayer perceptron (MLP), which combines the node's own state with aggregated neighbor information to generate a new node representation.
[0086] The model input consists of real-time operating data from each node (device), including features such as voltage, current, power, and temperature. The input is a node feature matrix. After multi-layer GNN encoding, the fault diagnosis results are output through graph-level readout or node-level classifier. Specifically, the fault type (hot spot effect, PID attenuation, string mismatch, inverter IGBT fault), fault location (specific to a certain string or combiner box branch), fault degree (such as power loss percentage) and cause explanation.
[0087] The advantage of GNN models over traditional independent diagnostic methods is that when a node fails, its impact propagates to its neighboring nodes through the graph edges. Through this message passing process, the model can capture the cascading effects and multi-factor coupling relationships between devices, thereby providing more globally consistent diagnostic results and explanations of fault causes.
[0088] Training: Cross-validation is used, dividing historical failure cases into training and validation sets in an 8:2 ratio. The Adam optimizer is employed, with the cross-entropy loss function:
[0089] Perform parameter optimization, among which Number of fault categories For the sample In category The real labels on it This represents the corresponding probability predicted by the model.
[0090] Module 3: Control Optimization and Fault Diagnosis Module like Figure 6 As shown, this module is the core decision-making part of the system, responsible for real-time performance evaluation, optimized control, and fault diagnosis. The control optimization and fault diagnosis module 3 includes a performance evaluation unit 31, an optimization control unit 32, a fault diagnosis unit 33, and an alarm and self-healing unit 34.
[0091] (1) Performance evaluation unit Performance evaluation is conducted by combining real-time environmental data and theoretical power generation prediction models. ① Input the real-time irradiance and component temperature into the theoretical power generation prediction model to calculate the theoretical power generation under the current conditions.
[0092] ② Collect the actual output power of the inverter and calculate the performance ratio PR = actual power generation / theoretical power generation × 100%.
[0093] ③ Analyze the PR differences of each combiner box branch and each string to identify units with low performance (e.g., if the PR of a certain string is significantly lower than that of the adjacent strings, there may be problems such as shading, contamination or component attenuation).
[0094] ④ Output system efficiency (PR), power generation loss, performance degradation rate and other indicators.
[0095] (2) Optimize the control unit Dynamically adjust power plant operating parameters based on model predictions and equipment operating constraints: ① Inverter active power control: Adjust the output power according to the grid dispatch instructions and the real-time power generation capacity of the power station to participate in grid peak shaving and frequency regulation.
[0096] ②Reactive power regulation: Adjust reactive power output according to the voltage conditions at the grid connection point to support grid voltage stability.
[0097] ③ Component tilt tracking strategy: For a flat single-axis tracking system, the tracking angle is optimized according to the sun's position and weather conditions (cloudy, windy) to achieve a balance between power generation revenue and mechanical losses.
[0098] ④ Cleaning plan trigger: When the power generation loss caused by dust accumulation exceeds the cleaning cost threshold, a cleaning task is automatically generated.
[0099] (3) Fault diagnosis unit The diagnostic process, which utilizes a deep causal fault diagnosis model for inference and analysis, is as follows: ① The model performs forward calculations on the data from all monitoring points at the current moment and outputs the probability of anomalies for each device and string.
[0100] ② Locate and analyze units with an abnormal probability exceeding the threshold, for example, by comparing the current of each branch under the same combiner box to identify abnormal branches.
[0101] ③ Assess the severity of the fault after it has been located, for example, by analyzing the characteristics of the IV curve (maximum power point shift, fill factor change) to evaluate the degree of component degradation.
[0102] ④ Utilize the model's interpretability module to output explanations of the causes of failures.
[0103] For example, the diagnosis result can be expressed as: "The current of string A3 is 60% of the normal value, suspected hot spot effect, confidence level 92%, possible cause is that there are leaves blocking the bottom of the string or the diode inside the component is broken down. (4) Alarm and self-healing unit The faults are classified into three levels based on their severity: Level 1 (Minor): If a single sensor reading is abnormal but the main device is normal → automatically record and attempt a software reset or switch to a backup sensor.
[0104] Level 2 (General): If the power generation of a certain string decreases by less than 20% due to obstruction, the operating strategy will be automatically adjusted (such as reducing the voltage of the MPPT branch where the string is located) and a maintenance work order will be generated.
[0105] Level 3 (Severe): If an inverter failure causes a shutdown or an arc fault is detected, an alarm will be immediately pushed through multiple channels such as APP, SMS, and email, and complete data before and after the fault will be recorded for analysis.
[0106] Module 4: Operation and Maintenance Decision-Making and Auxiliary Management Module like Figure 7 As shown, this module is the long-term decision-making part of the system, responsible for the intelligent management of operation and maintenance plans, materials, personnel, and finances. The operation and maintenance decision-making and auxiliary management module 4 includes a material management unit 41, an operation and maintenance planning unit 42, a personnel management unit 43, a financial management unit 44, and a knowledge base unit 45.
[0107] (1) Materials Management Unit Spare parts management based on fault diagnosis and prediction results: Based on the failure rate statistics and remaining service life prediction of each device, the safety stock level and replenishment point of spare parts are calculated. For example, if the failure probability of a certain type of inverter IGBT module is predicted to be 15% in the next month, the inventory warning threshold of that spare part will be automatically increased. It integrates with the supplier's system to automatically generate purchase orders and track logistics status when inventory falls below a threshold.
[0108] (2) Operation and maintenance planning unit The optimized operation and maintenance plan is generated by comprehensively considering multiple factors, including: ① Equipment fault prediction results (e.g., a certain string needs cleaning, a certain inverter needs maintenance).
[0109] ② Equipment maintenance cycle (e.g., the junction box should be inspected every six months).
[0110] ③ Weather forecast (e.g., sunny skies for the next three days, suitable for outdoor work).
[0111] ④ Staff availability (e.g., 3 maintenance personnel are on duty this week).
[0112] ⑤ Spare parts inventory status.
[0113] Genetic algorithms are used to optimize and solve task priorities, time windows, and resource constraints, and output maintenance work orders (including task type, task content, suggested time, required personnel skills, and required spare parts).
[0114] (3) Personnel Management Unit Maintain a database of maintenance personnel (basic information, skills and qualifications such as high-voltage electrician certificate / high-altitude work certificate, historical performance, and current workload). Based on the skill requirements and urgency of maintenance work orders, automatically match qualified personnel with a light workload. For urgent faults, the schedule can be dynamically adjusted to assign nearby personnel for priority handling.
[0115] (4) Financial Management Unit Link maintenance records and financial data: Record the material costs, labor costs, and power generation losses (downtime losses) for each maintenance activity. Calculate the levelized cost of electricity (LCOE) of the equipment and evaluate the return on investment of different operation and maintenance strategies. For example, compare the annual costs and power generation gains of the two strategies of "periodic cleaning" and "on-demand cleaning" to provide data support for subsequent decision-making.
[0116] (5) Knowledge base unit Building an operations and maintenance knowledge base based on knowledge graphs: The physical components include: equipment model, fault symptoms, diagnostic methods, repair procedures, spare parts models, etc.
[0117] Relationships include: "caused by", "the solution is", "related spare parts are", etc.
[0118] Maintenance personnel can use natural language queries (such as "how to troubleshoot low string current") to obtain historical similar cases and handling suggestions.
[0119] After each new fault is resolved, the system automatically adds the case to the knowledge base for continuous updates.
[0120] Module 5: Communication Assurance and Security Protection Module like Figure 8 As shown, this module provides network communication and security guarantees for all modules. The communication guarantee and security protection module 5 includes a communication network unit 51, a security protection unit 52, and an access control unit 53.
[0121] (1) Communication network unit Construct a power plant-level redundant communication network: Backbone network: Fiber optic ring network, the ring topology ensures uninterrupted communication in the event of a single point of failure.
[0122] Array area communication: RS485 bus or PLC power line carrier communication connects the combiner box and string monitoring equipment.
[0123] Supplemental coverage: Large ground-mounted power stations supplement LoRa or 4G wireless networks.
[0124] Critical command channels: Inverter start-up and shutdown, power regulation, etc. are transmitted through independent high-priority channels, and priority queuing and traffic shaping technology are used to ensure low latency.
[0125] (2) Safety protection unit Implement a defense-in-depth strategy: Network boundary: Deploy industrial firewalls and configure access control lists to allow only authorized IPs and protocols to pass through.
[0126] Intrusion Detection: Deploy IDS at critical nodes to analyze network traffic in real time and identify and block port scanning, abnormal data packets, and other behaviors.
[0127] Data encryption: The national cryptographic algorithms SM2 / SM4 are used. SM2 is used for key negotiation and digital signature, and SM4 is used for symmetric encryption of large amounts of data.
[0128] Data backup: Daily incremental backups and weekly full backups, with backup data stored in an off-site disaster recovery center.
[0129] (3) Access Control Unit Implement role-based access control: Predefined roles: Operations Engineer (view real-time data, receive alarms, and perform remote operations), Data Analyst (access historical data, train models, and generate reports), and Power Plant Administrator (all permissions, configure parameters, and manage users).
[0130] Multi-factor authentication: password + dynamic verification code.
[0131] Operation Log: All operation records (login time, operation content, operation object) are written to an immutable operation log and stored for no less than one year.
[0132] Collaborative workflow between modules The following example of a typical hot spot fault handling illustrates the collaborative workflow of each module: In the first step, the sensor network unit 11 in the data acquisition and processing fusion module 1 acquires data of an abnormal drop in the current of a certain string. The data preprocessing unit 12 cleans the data and identifies the anomalies. The data fusion unit 13 then merges the data and sends it to the control optimization and fault diagnosis module 3 and the data storage and model training module 2.
[0133] In the second step, the fault diagnosis unit 33 in the control optimization and fault diagnosis module 3 calls the fault diagnosis model provided by the data storage and model training module 2 to determine that the string has a hot spot effect with a confidence level of 92%, and outputs the fault location to the specific component. The alarm and self-healing unit 34 generates a secondary alarm.
[0134] The third step involves the data storage and model training module 2 storing the fault data into the fault case library, while the model training unit 23 uses the new data to incrementally update the model in the background.
[0135] In the fourth step, after the operation and maintenance decision and auxiliary management module 4 receives the fault diagnosis result, the operation and maintenance planning unit 42 generates a maintenance work order to replace the component, the material management unit 41 checks whether there is a compatible component in the inventory, and the personnel management unit 43 assigns personnel with replacement skills to perform the task.
[0136] Fifth, the communication assurance and security protection module 5 ensures the security and reliability of data transmission between the above modules, and the access control unit 53 records all operation logs.
[0137] Through the above collaborative work, this system has achieved closed-loop management of the entire process from fault perception, diagnosis, decision-making to handling.
[0138] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned intelligent integrated operation and maintenance method for photovoltaic power plants.
[0139] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0140] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.
Claims
1. A method for intelligent integrated operation and maintenance of photovoltaic power plants, characterized in that, Includes the following steps: S1, Collect the operating status data of photovoltaic power station equipment, preprocess and fuse the data to generate a data stream with unified time alignment; S2, based on a hybrid model combining a physical mechanism model and a data-driven model, analyzes the data stream to diagnose the fault type, fault location, and fault severity of the equipment; S3, based on the fault diagnosis results, generates alarm information according to the preset classification rules, and executes self-healing control operations corresponding to the fault level; S4 generates maintenance work orders based on fault diagnosis results and self-healing control operations, and coordinates materials, personnel, and plans to execute closed-loop maintenance management.
2. The method as described in claim 1, characterized in that, S1 includes: Data from different manufacturers' devices in different formats is converted into a unified timestamp alignment format; outliers are identified and removed using a method based on the 3σ criterion and the isolated forest algorithm. For short-term data gaps, linear interpolation or historical data interpolation based on similar days can be used to fill in the gaps. The irradiance and power signals are smoothed by median filtering or wavelet denoising based on sliding window to obtain preprocessed data with quality labels. The unified data access gateway supports access to Modbus RTU / TCP, IEC 61850, MQTT and OPC-UA protocols, and automatically identifies device types and performs protocol parsing. The Kalman filter algorithm is used to dynamically estimate and eliminate noise in the time series data of irradiance, temperature, and power. For multi-source measurements of the same physical quantity, a confidence-based weighted fusion method is adopted. The weights are dynamically allocated according to the historical accuracy of each sensor and the current environmental state to generate a high-quality data stream with unified time alignment.
3. The method as described in claim 2, characterized in that, The method for dynamic estimation and noise cancellation is as follows: A state-space model of the photovoltaic system is established. The state at the next moment is predicted by the time update equation. The prediction result is corrected by the actual observation value through the measurement update equation to obtain the optimal estimate.
4. The method as described in claim 1, characterized in that, S2 includes: The five-parameter model of a single diode in a photovoltaic module is used as the physical mechanism model. The current-voltage relationship of the physical mechanism model is given by the implicit equation: in, For photocurrent, This is the reverse saturation current of the diode. For series resistance, For parallel resistors, For diode ideality factor, Thermoelectric voltage; A Long Short-Term Memory (LSTM) network is used as the data-driven error compensation model. Environmental humidity, wind speed, estimated dust accumulation levels, irradiance, temperature, and the theoretical maximum power output from the physical model are used as input features to predict the error. The final theoretical power generation prediction is as follows: in, The theoretical maximum power output by the physical model. This is the error compensation value predicted by the LSTM model.
5. The method as described in claim 4, characterized in that, The fault type, fault location, and fault severity of the diagnostic equipment include: Each string, combiner box, and inverter in the power station is treated as a graph node. Electrical connections are represented as graph edges. Constructing a graph structure ; By using the message passing mechanism of a graph neural network to learn the mutual influence between devices, for the first... Layers, nodes The hidden state update process is as follows: , in, Employ the attention aggregation mechanism in graph attention networks; The real-time operating data of each node is used as the input node feature matrix. After being encoded by a multi-layer graph neural network, the fault type, fault location, fault severity, and cause explanation are output through graph-level readout or node-level classifier.
6. The method as described in claim 1, characterized in that, S3 includes: The fault levels are divided into three levels: Level 1, Level 2, and Level 3. For Level 1 faults, the fault is automatically recorded and an attempt is made to perform a software reset or switch to a backup sensor. For level 2 faults, the system will automatically adjust the operating strategy and generate a maintenance work order; For level 3 faults, alarms will be immediately pushed through multiple channels such as APP, SMS and email, and complete data before and after the fault will be recorded for analysis.
7. The method as described in claim 1, characterized in that, S4 includes: Based on equipment failure prediction results, equipment maintenance cycle, weather forecast, personnel availability and spare parts inventory, a genetic algorithm is used to optimize the solution of task priority, time window and resource constraints, and output maintenance work orders including task type, task content, suggested time, required personnel skills and required spare parts. Based on the skill requirements and urgency of the maintenance work order, automatically match personnel with the corresponding qualifications and currently light workload; Based on the failure rate statistics and remaining service life prediction of each piece of equipment, calculate the safety stock level and replenishment point of spare parts, and automatically generate a purchase order when the inventory is below the threshold.
8. The method as described in claim 1, characterized in that, Also includes: A redundant communication network is used to ensure data transmission between modules, and the data is encrypted using the national cryptographic SM2 / SM4 algorithm. At the same time, access control is based on roles to manage permissions and all operation logs are recorded.
9. A photovoltaic power plant intelligent integrated operation and maintenance system, characterized in that, include: The data fusion module is used to collect operating status data of photovoltaic power station equipment, preprocess and fuse the data, and generate a data stream with unified time alignment. The hybrid diagnostic module is used to analyze the data stream based on a hybrid model that combines a physical mechanism model and a data-driven model, and to diagnose the fault type, fault location and fault severity of the equipment. The alarm self-healing module is used to generate alarm information according to preset classification rules based on the fault diagnosis results, and to execute self-healing control operations corresponding to the fault level. The closed-loop operation and maintenance module is used to generate operation and maintenance work orders based on fault diagnosis results and self-healing control operations, and coordinate materials, personnel and plans to execute closed-loop operation and maintenance management.
10. A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method of any one of claims 1-8.
