New energy power generation equipment health management platform based on large model

Abstract

The present application relates to the technical field of data analysis, and more particularly to a new energy power generation equipment health management platform based on a large model, comprising a data acquisition and perception layer, an edge computing layer, a cloud processing layer and an application service layer; the present application uses static historical data or a single equipment parameter as a health detection benchmark, which has the shortcoming of being unable to reflect the dynamic changes of the environment and the influence of equipment aging, compared with the prior art; the present application uses multi-dimensional simulation modeling technology, collects equipment parameters, weather parameters and other real-time working condition data to construct a digital twin model, simulates the equipment operating state under different scenes (such as photovoltaic efficiency decay under extreme temperature, fan aerodynamic load under salt spray environment), and generates dynamic health benchmark values including power generation, instantaneous current / voltage, equipment temperature, etc. This method can accurately capture the interaction effect of the environment and the equipment, dynamically adjust the health detection threshold with the working condition, and the abnormal identification accuracy is improved by more than 35% compared with the traditional method.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, and particularly relates to a new energy power generation equipment health management platform based on a large model. BACKGROUND

[0002] New energy power generation equipment faces severe challenges in long-term stable operation under complex natural environment and dynamic grid operating conditions. The existing health management technology system has significant defects in environmental adaptability, real-time performance and diagnostic accuracy.

[0003] Taking an offshore wind turbine as an example, the traditional health detection method relies on fixed threshold or historical experience data to set abnormal alarm line, but the coupling of equipment aging, salt spray corrosion and sea wave impact leads to prominent problems of sensor drift and performance degradation. Industry measurement data shows that in the area where the salt spray concentration is higher than 5 milligrams per cubic meter, the monthly average drift error of the current sensor can reach 1.5 percent, and the existing calibration technology only supports periodic manual maintenance or simple primary and standby switching mechanism, which cannot realize dynamic error compensation, resulting in that the equipment false alarm rate is maintained at more than 25 percent for a long time.

[0004] In the aspect of data processing, the traditional cloud centralized architecture is limited by network bandwidth and transmission delay. For key features such as high-frequency vibration signal and instantaneous current fluctuation, the uploading of full data leads to more than 40 percent of effective information loss rate. A case of a certain northern European wind farm shows that the twelve kilohertz high-frequency vibration signal caused by early wear of the gearbox cannot be timely identified due to data compression distortion, which eventually leads to the breakage of the transmission chain, with indirect loss of millions of euros.

[0005] In the aspect of fault diagnosis model, the current mainstream technology is based on a single physical model or statistical learning algorithm, such as an expert system relying on vibration spectrum feature matching. Its knowledge base update is seriously lagging behind the device technology iteration speed, which cannot effectively identify new composite fault modes. The International Renewable Energy Agency report points out that 35 percent of the newly added faults in the past three years are not covered by the traditional model, and the difference rate between manual repair results and algorithm diagnosis conclusions is as high as 30 percent. However, due to the lack of closed-loop feedback channel, the model iteration period is as long as six months, which seriously restricts the operation and maintenance efficiency.

[0006] In addition, in extreme environmental scenarios such as sandstorm invasion of desert photovoltaic power stations or biological attachment corrosion of tidal power generators, the existing sensor calibration technology cannot dynamically adapt to the dramatic changes of environmental parameters, leading to distortion of key monitoring data such as temperature and pressure, and further causing misjudgment. Industry data shows that the unplanned downtime caused by sensor drift accounts for 45 percent of the total equipment failure time, which seriously weakens the economic benefits of new energy power generation.

[0007] In this context, it is urgent to build a new health management technology system that integrates multi-physical field dynamic simulation, edge intelligent computing, adaptive calibration and closed-loop optimization feedback, to systematically solve the core problems of environmental interference suppression, real-time data processing, composite fault identification and model self-evolution. SUMMARY

[0008] In order to overcome the problems presented in the above background art, the present application proposes a new energy power generation equipment health management platform based on large model.

[0009] The technical solution of the present application is: a new energy power generation equipment health management platform based on large model, comprising:

[0010] The data acquisition and perception layer is used to acquire real-time operation data through multi-source sensors and Internet of Things devices, and combines external data including meteorological parameters and power grid state; real-time operation data is acquired through multi-source sensors and Internet of Things devices, and external data such as meteorological parameters and power grid state are fused, which can obtain comprehensive and multi-angle operation information of new energy power generation equipment, provide rich and accurate data basis for subsequent analysis and diagnosis, and help to more comprehensively understand the equipment operation condition;

[0011] The edge computing layer deploys a lightweight anomaly detection model to filter the data collected by the data acquisition and perception layer and extract key features; the lightweight anomaly detection model filters the collected data and extracts key features, which can perform preliminary processing at the data source, effectively reducing the amount of data transmitted to the cloud, reducing network transmission burden, improving data transmission efficiency, and at the same time filtering and extracting features at the edge, which can quickly respond to equipment anomalies, discover potential problems in time, avoid delay in fault handling due to data transmission delay, and improve the real-time performance and reliability of the system;

[0012] The cloud processing layer is used to build a digital twin simulation model, predict the working state of the new energy power generation equipment in combination with external data, generate dynamic health benchmark values, and perform health diagnosis and management through large models; the digital twin simulation model is built, the working state of the new energy power generation equipment is predicted in combination with external data, dynamic health benchmark values are generated, which can more accurately evaluate the health condition of the equipment, discover potential health hazards of the equipment in advance, realize preventive maintenance of the equipment, reduce the equipment failure rate, and prolong the service life of the equipment;

[0013] The application service layer is used to provide equipment health assessment, operation and maintenance work order management, visualization services and decision support based on the data results of the cloud processing layer; the equipment health assessment service is provided based on the data results of the cloud processing layer, which can present complex equipment health information to users in an intuitive way, facilitate users to quickly understand the equipment health condition, and facilitate equipment management and maintenance decision-making.

[0014] As preferred, the data acquisition and perception layer further comprises an intelligent calibration module for intelligently calibrating the sensors in the data acquisition and perception layer. The intelligent calibration module intelligently calibrates the sensors, can adjust the working state of the sensors in real time, reduces the data deviation caused by factors such as sensor self-error, environmental interference, etc., thereby ensuring that the collected data is more accurate and reliable, and providing a high-quality data basis for subsequent equipment health management and decision-making. The intelligent calibration module adopts a master-backup redundant architecture for core parameter sensors, specifically including:

[0015] S11: data synchronization acquisition, the master sensor acquires data in real time, and the backup sensor synchronously samples at a fixed period in a dormant state, and data synchronization of the master sensor and the backup sensor is realized through a time stamp; the master sensor acquires data in real time, and the backup sensor synchronously samples at a fixed period in a dormant state, and data synchronization is realized through a time stamp. This master-backup redundant architecture ensures that when one sensor fails or data is abnormal, the other sensor can take over the data acquisition task in time, avoids interruption of data acquisition, and guarantees the continuity and integrity of data;

[0016] S12: D-S evidence theory decision, first, calculate the data support of the master sensor and the backup sensor, and generate a joint probability distribution based on a confidence matrix to determine whether the data of the master sensor and the backup sensor meet a consistency threshold, wherein the principle formula for calculating the data support of the master sensor and the backup sensor is:

[0017] ;

[0018] wherein, represents the data support of the master sensor and the backup sensor, is the measurement value of the master sensor, is the measurement value of the backup sensor, and are the range of the sensor, i.e. the maximum and minimum values that can be measured by the sensor; D-S evidence theory decision is adopted to calculate the data support of the master sensor and the backup sensor, and generate a joint probability distribution based on a confidence matrix to determine whether the data of the master sensor and the backup sensor meet a consistency threshold. Through comprehensive analysis and consistency judgment of the data of the two sensors, abnormal conditions of the sensors can be found in time, the detection capability of the system for sensor failure is improved, and the reliability of the system is enhanced;

[0019] S13: Offset judgment and switching strategy, if the offset of the main sensor and the backup sensor exceeds the threshold value for 3 times in a row, and the automatic switching logic is triggered, the backup sensor takes over the main control right and sends a fault warning; if the offset of the main sensor and the backup sensor exceeds the threshold value for 3 times in a row, and the automatic switching logic is triggered, the backup sensor takes over the main control right and sends a fault warning. This automatic switching strategy can quickly respond when the sensor fails, reduce the impact of failure on the normal operation of the system, and improve the stability and availability of the system;

[0020] S14: After switching, the original main sensor enters calibration mode, verifies its drift characteristics through standard source input, and if it cannot be restored, it is marked as a fault to be replaced, wherein the standard source input is the reference value data in the healthy reference library; After switching, the original main sensor enters calibration mode, verifies its drift characteristics through standard source input. This process helps to accurately determine whether the original main sensor has a fault and the type and degree of the fault, providing clear basis for troubleshooting and maintenance, improving the efficiency and accuracy of equipment maintenance.

[0021] As a preferred, the intelligent calibration module adopts dynamic reference value weight for error compensation when calibrating the sensors in the data acquisition and perception layer for non-critical parameter sensors, specifically including;

[0022] S21: Environment parameter correlation matrix construction, determine the non-critical parameters and their sensitive environmental variables, establish the compensation coefficient matrix through historical data analysis or physical model derivation; adopt dynamic reference value adjustment strategy, discretize the environmental variables into state space, and calculate the adjustment vector group according to the deviation of the current environmental parameters and the threshold value. This dynamic adjustment mechanism can respond to environmental changes in real time, compensate for sensor data in a timely manner, avoid the accumulation of data deviation caused by environmental changes, and ensure the accuracy of data in different environments;

[0023] S22: Dynamic adjustment of reference value, discretize the environmental variables into state space, the action space is the reference value adjustment amplitude, and adopt the decay learning rate strategy, then input the parameter correlation matrix, the deviation of the current environmental parameters and the threshold value, calculate the adjustment vector group; In the process of dynamic adjustment of reference value, the decay learning rate strategy is adopted, so that the algorithm can quickly learn and adapt to the environment in the initial stage, and gradually reduce the learning rate as the learning process progresses, to avoid excessive adjustment and oscillation. This strategy can improve the stability and convergence of the algorithm, so that the system can better adapt to complex and variable environments;

[0024] S23: Calibration Verification and Update. A sliding window mechanism is introduced to calculate the error variance after the N most recent adjustments. If the variance continues to increase, the exploration rate is reset, forcing the algorithm to re-explore the environment. The optimized reference values ​​are then written into the digital twin model, and a new health benchmark library is generated. This mechanism can promptly detect situations where the algorithm may be trapped in a local optimum or where significant environmental changes occur. By re-exploring the environment, a more suitable reference value adjustment strategy is found, enhancing the system's adaptability and robustness. Furthermore, the optimized reference values ​​are written into the digital twin model, and a new health benchmark library is generated. This allows the health benchmark library to more accurately reflect the normal operating status of the equipment under different environments, providing a more reliable basis for equipment health diagnosis and management. Based on the updated health benchmark library, the health status of the equipment can be more accurately assessed, and potential faults can be detected in advance.

[0025] Preferably, the edge computing layer specifically includes:

[0026] A11: The fault handling module employs a benchmark statistical model to filter out abnormal segments and extract statistical features. Fault detection is then performed based on these features, and real-time fault handling is conducted according to a built-in fault strategy library. The benchmark statistical model utilizes statistical methods including the 3σ rule and moving average analysis to filter out abnormal segments and extract statistical features, which are then used for fault detection. The 3σ rule effectively identifies outliers that deviate significantly from the normal data distribution, while moving average analysis smooths data fluctuations and captures data trend changes. The combination of these two methods enables more accurate fault detection during equipment operation, reducing the probability of false alarms and missed alarms. Furthermore, the built-in fault strategy library allows for rapid real-time fault handling once a fault is detected. This rapid response mechanism can promptly control the scope of fault impact, prevent further deterioration, reduce equipment downtime and production losses, and improve the operational stability and reliability of new energy power generation equipment.

[0027] A12: The data filtering module is used to filter the data collected by the data acquisition and perception layer based on the data results of the fault identification model and preset rules, retaining abnormal data and key feature data for uploading to the cloud processing layer; This effectively reduces unnecessary data transmission, lowers network bandwidth usage, and improves data transmission efficiency, especially suitable for situations where new energy power generation equipment is typically deployed in remote areas with limited network bandwidth.

[0028] A13: Scheduling optimization module, used to establish a health assessment model for compute nodes and dynamically allocate tasks based on CPU and memory load. This dynamic scheduling method can rationally allocate computing tasks according to the actual load of each compute node, avoiding situations where some compute nodes are overloaded while others are idle, improving the utilization of computing resources, and ensuring the efficient operation of the edge computing layer.

[0029] As a preferred option, the cloud processing layer specifically includes:

[0030] A21: The simulation module is used to build a digital twin simulation model. Based on real-time environmental data and external data, it simulates the new energy power generation equipment to obtain predicted operating status data. The digital twin simulation model comprehensively considers real-time environmental data (such as wind speed and light intensity) and external data (such as grid dispatch instructions and weather forecasts) to simulate the new energy power generation equipment. This comprehensive simulation method can more accurately simulate the equipment's operation under different conditions, obtaining more realistic predicted operating status data, providing a reliable basis for subsequent anomaly identification and fault diagnosis.

[0031] A22: Anomaly detection module, used to identify anomalies based on real-time data and predicted operating status data; by comparing and analyzing real-time data with predicted operating status data obtained from the simulation module, it can quickly identify abnormal situations during equipment operation. This data comparison-based anomaly detection method has high sensitivity and accuracy, and can promptly detect subtle changes in equipment performance, saving valuable time for subsequent fault diagnosis;

[0032] A23: The fault diagnosis module is used to diagnose faults in new energy power generation equipment based on the data results from the anomaly identification module. Based on the data provided by the anomaly identification module, the fault diagnosis module can deeply analyze the equipment's operating data and status information to accurately pinpoint the specific location and cause of the fault. This helps maintenance personnel quickly take effective repair measures, shorten equipment downtime, and improve equipment availability.

[0033] Preferably, the cloud processing layer, when in operation, specifically includes:

[0034] S31: Multi-source data fusion and processing. First, it receives filtered abnormal data, key feature data, and external environmental parameters from the edge computing layer. This data is received via a distributed message queue, and then cleaned and aligned. Finally, the heterogeneous data is converted into a standardized format that can be parsed by the digital twin model. Data received from the edge computing layer may contain noise, missing values, or inconsistent timestamps. Through data cleaning and alignment, invalid data can be removed, missing values ​​filled, and the time reference of the data unified, thereby improving data quality and usability and providing a reliable foundation for subsequent analysis and modeling.

[0035] S32: The digital twin simulation model is dynamically updated. A digital twin simulation model is built based on equipment design parameters through the simulation module. Cleaned data is then injected into the model to simulate equipment operation and obtain equipment health baseline values. Building a digital twin simulation model based on equipment design parameters accurately reflects the physical characteristics and operating patterns of the equipment. Injecting cleaned data into the model for equipment operation simulation yields more realistic equipment health baseline values, providing accurate reference for anomaly identification and fault diagnosis.

[0036] S33: Anomaly Detection. This module calculates the deviation between real-time data and simulated prediction data, uses moving average analysis to identify short-term fluctuations, and employs wavelet transform to detect periodic anomalies. By calculating the deviation between real-time data and simulated prediction data, and combining moving average analysis to identify short-term fluctuations and wavelet transform to detect periodic anomalies, it enables comprehensive data analysis from different perspectives, improving the accuracy and sensitivity of anomaly detection. Moving average analysis effectively smooths data and captures long-term trends, while wavelet transform excels at detecting periodic anomalies and local abrupt changes in the data.

[0037] S34: Fault Diagnosis. The fault diagnosis module extracts statistical and time-frequency features of abnormal segments to form a high-dimensional feature vector. This vector is then used for fault location and root cause analysis, generating an interpretable report. Extracting statistical and time-frequency features of abnormal segments to form a high-dimensional feature vector comprehensively reflects the characteristic information of the fault. Through the analysis and processing of this high-dimensional feature vector, the location and root cause of the fault can be more accurately pinpointed, providing a clear direction for fault repair.

[0038] Preferably, when the cloud processing layer dynamically updates the digital twin simulation model, it specifically includes:

[0039] S41: Digital twin simulation model construction. Based on equipment design parameters, a three-dimensional geometric model of the new energy power generation equipment is constructed using parametric modeling tools, and a finite element mesh is generated through a mesh generation algorithm. This accurate geometric foundation provides a basis for subsequent analysis and simulation, making the simulation results closer to the actual operation of the equipment. Furthermore, the finite element mesh generated through the mesh generation algorithm discretizes the continuous physical model into a finite number of elements, facilitating numerical calculation and analysis. A reasonable mesh generation can improve computational accuracy and efficiency, ensuring the reliability of the simulation results.

[0040] S42: Multiphysics embedding, including mechanical fields, fluid fields, and thermoelectric coupling. The mechanical field involves solving the stress-strain distribution through finite element analysis; the fluid field uses computational fluid dynamics to simulate the aerodynamic loads of wind speed and irradiance on new energy power generation equipment; and the thermoelectric coupling establishes a correlation model between temperature and power generation efficiency for photovoltaic modules. Multiphysics embedding considers various physical phenomena and interactions during equipment operation, making the simulation results closer to the actual operating state of the equipment. Compared to single-physics field simulations, multiphysics coupled simulations provide more comprehensive and accurate information, offering stronger support for equipment health management and performance optimization.

[0041] S43: Real-time data-driven and dynamic updates inject data transmitted from the edge computing layer into the model, dynamically updating boundary conditions and load status. Models based on real-time data-driven analysis can more accurately predict the future operating status of equipment, providing timely warnings and decision support for maintenance personnel. For example, when the model detects abnormal changes in certain parameters of the equipment, it can issue an early warning, allowing maintenance personnel to take timely measures based on the warning information to prevent equipment failure.

[0042] S44: Co-simulation and prediction utilizes digital twin simulation models to simulate and predict the operating status of new energy power generation equipment. This helps maintenance personnel understand the equipment's performance under different operating conditions. By comparing simulations of various operating strategies, the optimal strategy can be selected to improve the equipment's power generation efficiency and energy utilization rate.

[0043] S45: Dynamic health baseline generation. This involves generating a probability distribution of equipment health status through Monte Carlo simulation, and combining this with confidence level assessments to output dynamic baseline values. These output dynamic baseline values ​​include vibration amplitude thresholds, temperature drift tolerances, and electrical output parameters. These dynamic baseline values ​​provide a scientific basis for equipment health assessment, enabling a more accurate determination of whether the equipment is operating normally.

[0044] Preferably, when the fault diagnosis module extracts the statistical and time-frequency features of abnormal segments to form a high-dimensional feature vector, performs fault location and root cause analysis, and generates an interpretable report, it specifically includes:

[0045] S51: Feature vector construction involves extracting statistical and time-frequency features from abnormal segments to form a high-dimensional feature vector. Extracting these features allows for analysis of fault signals from different perspectives. Statistical features such as mean, variance, and kurtosis reflect the overall distribution and fluctuations of the signal; time-frequency features such as wavelet coefficients and short-time Fourier transform coefficients capture the signal's changes in time and frequency. Combining these features into a high-dimensional feature vector provides a more comprehensive characterization of the fault, offering rich information for subsequent diagnosis.

[0046] S52: Embedding physical knowledge, combined with equipment mechanism models, constrains the diagnostic scope; by embedding physical knowledge into the fault diagnosis process, the scope of diagnosis can be constrained. Equipment mechanism models reflect the physical characteristics and operating laws of equipment. Through matching analysis between fault characteristics and mechanism models, some fault assumptions that do not conform to physical laws can be eliminated, thereby narrowing the scope of fault diagnosis and improving diagnostic efficiency.

[0047] S53: Large-scale model semantic analysis utilizes pre-trained models to parse operation and maintenance logs and historical work order texts, associating fault modes. This pre-trained model enables the extraction of valuable information from these unstructured text data. Operation and maintenance logs and historical work orders record past equipment operation and fault handling experience; semantic analysis can extract key information such as fault modes, fault causes, and handling methods, providing a reference for current fault diagnosis.

[0048] S54: Multimodal fusion reasoning employs an attention mechanism to dynamically weight features from different modalities and uses a graph neural network for cross-modal relationship reasoning. The attention mechanism dynamically adjusts the weighting based on the importance of different modal features in fault diagnosis. Fault diagnosis of new energy power generation equipment involves data from multiple modalities, such as vibration signals, temperature signals, and image data. The attention mechanism can automatically learn the importance of different modal features, allowing important features to play a greater role in diagnosis, thereby comprehensively utilizing multi-source information and improving diagnostic accuracy.

[0049] S55: Fault Location and Root Cause Analysis. Based on Bayesian networks, it calculates the probability distribution of faults and generates interpretable reports. This Bayesian network-based calculation comprehensively considers the causal relationships and uncertainties among various fault factors, accurately calculating the probability distribution of different fault locations and causes. By selecting the fault locations and causes with the highest probabilities, precise fault location and root cause analysis can be achieved, providing maintenance personnel with a clear direction for repairs.

[0050] Preferably, when calculating the fault distribution probability based on a Bayesian network and generating an interpretable report, the specific steps include:

[0051] S61: Probability Output Distribution. This method inputs feature vectors into a Bayesian network to output the failure probability of each component. The failure warning threshold is adjusted based on the equipment aging coefficient. By inputting feature vectors into a Bayesian network and outputting the failure probability of each component, it can quantitatively assess the likelihood of failure for each component. The Bayesian network, based on probabilistic reasoning, comprehensively considers the causal relationships and uncertainties between various factors, making the calculation of failure probabilities more accurate and reliable. Maintenance personnel can prioritize different components of the equipment based on their failure probabilities, addressing components with higher failure probabilities first to improve maintenance efficiency. Furthermore, adjusting the failure warning threshold based on the equipment aging coefficient makes the warning mechanism more aligned with the actual operating conditions of the equipment. As equipment age, the probability of failure increases. By introducing an equipment aging coefficient, the warning threshold can be dynamically adjusted, making warnings more timely and accurate. When equipment is severely aged, the warning threshold can be appropriately lowered to detect potential failures earlier; when the equipment is newer, the warning threshold can be appropriately raised to avoid false alarms.

[0052] S62: Fault propagation path visualization, generating multi-level fault chains and labeling the confidence levels of key nodes; This visualization visually presents the propagation path of a fault within the device. Maintenance personnel can clearly understand how a fault propagates from one component to another, and the probability of failure at each key node, through visual charts. This helps maintenance personnel gain a deeper understanding of the fault occurrence mechanism, providing more comprehensive information for fault diagnosis and repair.

[0053] S63: Interpretable Report Generation. This feature generates natural language reports based on a large model. The generated reports include fault cause probability analysis, recommended fault inspection solutions, and recommended fault repair solutions. This comprehensive fault information provides maintenance personnel with a complete understanding of the faults. Fault cause probability analysis helps maintenance personnel understand the various possible factors leading to faults and their probabilities, providing a reference for fault localization. Recommended fault inspection solutions guide maintenance personnel to conduct targeted equipment inspections, improving inspection efficiency. Recommended fault repair solutions provide maintenance personnel with specific repair methods and steps, ensuring the smooth progress of maintenance work.

[0054] As a preferred option, the fault diagnosis module is also used for iterative optimization based on actual inspection work orders and actual repair work orders. The specific process is as follows:

[0055] S71: Abnormal event triggering. After manual maintenance is completed, the predicted fault type is compared with the actual fault type. When a difference is found, an abnormal event is triggered. This method can promptly detect deviations between system diagnostic results and actual conditions, providing clear triggering conditions for subsequent optimization. By promptly detecting errors, problems such as maintenance delays and resource waste caused by incorrect diagnoses can be avoided, thus improving operation and maintenance efficiency.

[0056] S72: Discrepancy Type Classification. If the manually confirmed actual fault type does not match the system diagnosis, it is marked as a diagnostic deviation, and the reason for the deviation is recorded. If a fault mode not in the Bayesian network is found, it is marked as a newly added fault type, triggering the knowledge base expansion process. Cases where the manually confirmed actual fault type does not match the system diagnosis are marked as either diagnostic deviations or newly added fault types. This classification method can accurately locate the type of problem, providing targeted guidance for subsequent optimization work. For diagnostic deviations, the reasons for the deviation can be analyzed in depth to optimize the diagnostic algorithm; for newly added fault types, the knowledge base can be expanded in a timely manner to improve the system's adaptability and coverage.

[0057] S73: Feature vector reconstruction involves extracting multi-dimensional features from deviation cases and establishing a diagnostic deviation feature matrix that compares the original and corrected features of the deviation cases. This method allows for in-depth analysis of the characteristic factors leading to diagnostic deviations, enabling the reconstruction and optimization of feature vectors. The optimized feature vectors more accurately reflect the essential characteristics of the fault, improving the accuracy of fault diagnosis.

[0058] S74: Dynamic expansion of the knowledge graph. For newly added fault types, a digital twin simulation model is used to simulate the state data of the corresponding fault type and construct new entity relationships, including defining fault node attributes and associated environmental variables. This approach can quickly incorporate relevant knowledge of new fault types into the knowledge graph, enhancing the system's knowledge reserves. The dynamic expansion of the knowledge graph allows the system to continuously adapt to new fault situations, improving the comprehensiveness and accuracy of fault diagnosis. Furthermore, constructing new entity relationships and defining fault node attributes and associated environmental variables helps support more complex fault analyses. By establishing correlations between faults and various environmental variables and equipment parameters, a deeper understanding of the fault occurrence mechanism and influencing factors can be achieved, providing more comprehensive information for fault diagnosis and maintenance.

[0059] S75: Model iterative optimization employs Gibbs sampling to dynamically adjust the Bayesian network node parameters and evaluates the rationality of the network topology through scoring row structure learning. This approach dynamically optimizes the Bayesian network model based on actual data and feedback information, improving the model's diagnostic accuracy. By continuously adjusting node parameters and optimizing the network topology, the model can better fit actual fault data, reducing diagnostic errors.

[0060] The beneficial effects of this invention are:

[0061] 1. Compared to existing technologies that use static historical data or single equipment parameters as health monitoring benchmarks, which fail to reflect dynamic environmental changes and the impact of equipment aging, this solution employs multi-dimensional simulation modeling technology. It integrates equipment parameters, weather parameters, and other real-time operating data to construct a digital twin model. By simulating equipment operating states under different scenarios (such as photovoltaic efficiency degradation under extreme temperatures and wind turbine aerodynamic loads in salt spray environments), it generates dynamic health benchmark values ​​including power generation, instantaneous current / voltage, and equipment temperature. This method can accurately capture the interaction between the environment and the equipment, allowing health monitoring thresholds to dynamically adjust with operating conditions. The accuracy of anomaly identification is improved by more than 35% compared to traditional methods.

[0062] 2. Compared to existing technologies that rely on a single algorithm model for fault diagnosis, which suffers from high false alarm rates and delayed identification of novel faults, this solution introduces a manual maintenance work order feedback mechanism. Upon discovering abnormal data, the fault knowledge base is dynamically optimized by comparing the results of manual on-site verification with the model's diagnostic results. If the actual fault type is inconsistent with the model's diagnosis, the fault label is corrected and distinguishing features are added. If a novel fault is discovered, samples are collected and injected into the model for incremental training. Through this closed-loop iteration, the model can adapt to equipment aging and environmental changes, reducing the false alarm rate by 40% and shortening the novel fault identification cycle from an average of 6 months to 2 weeks.

[0063] 3. Compared to existing technologies that upload all data to the cloud for processing, which suffers from high bandwidth consumption and poor real-time performance, this solution deploys lightweight statistical models (such as the 3σ rule and moving average analysis) at edge nodes, extracting features (such as vibration spectrum kurtosis and current fluctuation variance) only from abnormal segments and compressing them before uploading. This strategy reduces the amount of data uploaded by 70%, while significantly improving fault response speed through local real-time processing (such as triggering overcurrent protection commands within 10ms). Under extreme conditions, system latency is reduced from 5 seconds in traditional solutions to less than 0.5 seconds.

[0064] 4. Compared to existing technologies that employ fixed-period calibration or direct use of single sensor data, which suffer from high false alarm rates due to sensor drift and poor environmental adaptability, this solution adopts a primary-backup redundancy architecture for core parameters (such as current and voltage), and achieves automatic switching of faulty sensors through DS evidence theory decision-making. For non-critical parameters (such as temperature and pressure), an environmental correlation matrix is ​​constructed, and the reference value weights are dynamically updated using reinforcement learning (e.g., the compensation coefficient increases by 0.15 for every 10% increase in humidity). This mechanism reduces the sensor data error rate by 60% and maintains a data confidence level of over 95% even in harsh environments such as salt spray and sandstorms. Attached Figure Description

[0065] Fig. 1 The diagram shown is a structural schematic of the new energy power generation equipment health management platform based on a large model according to the present invention.

[0066] Fig. 2 The diagram illustrates the workflow of the cloud processing layer in the new energy power generation equipment health management platform based on a large model, as presented in this invention. Detailed Implementation

[0067] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0068] Please see Figs. 1-2 This invention provides an embodiment: a health management platform for new energy power generation equipment based on a large model, comprising:

[0069] Data Acquisition and Sensing Layer:

[0070] It is used to collect operational data in real time through multi-source sensors and IoT devices, and combine it with external data including meteorological parameters and power grid status;

[0071] The data acquisition and sensing layer also includes an intelligent calibration module. This module performs intelligent calibration on the sensors within the data acquisition and sensing layer. For core parameter sensors, the intelligent calibration module employs a primary-backup redundancy architecture. Specifically, this includes: synchronous data acquisition, where the primary sensor acquires data in real time, and the backup sensor synchronously samples at fixed intervals while in sleep mode, achieving data synchronization between the primary and backup sensors through timestamps; and DS evidence theory decision-making, which first calculates the data support of the primary and backup sensors, and generates a joint probability distribution based on the confidence matrix to determine whether the data from the primary and backup sensors meet the consistency threshold. The formula for calculating the data support of the primary and backup sensors is as follows: ;in, This indicates the data support between the main sensor and the backup sensor. The measured value of the main sensor, The measurement value is from the backup sensor. and The range of the sensor is the maximum and minimum values ​​that the sensor can measure; the offset judgment and switching strategy is as follows: if the offset of the main sensor and the backup sensor exceeds the threshold for three consecutive times and triggers the automatic switching logic, the backup sensor takes over the main control and sends a fault warning; after switching, the original main sensor enters the calibration mode and verifies its drift characteristics through the standard source input. If it cannot be recovered, it is marked as faulty and needs to be replaced. The standard source input is the reference value data in the health benchmark library.

[0072] For non-critical parameter sensors, dynamic reference value weighting is used for error compensation. This includes: constructing an environmental parameter correlation matrix to identify non-critical parameters and their sensitive environmental variables, and establishing a compensation coefficient matrix through historical data analysis or physical model derivation; dynamically adjusting reference values ​​by discretizing environmental variables into a state space and the action space into the reference value adjustment range, employing a decaying learning rate strategy, and then calculating the adjustment vector set by inputting the parameter correlation matrix and the deviation between the current environmental parameters and the threshold; and calibration verification and updating by introducing a sliding window mechanism to calculate the error variance after the most recent N adjustments. If the variance continues to increase, the exploration rate is reset, forcing the algorithm to re-explore the environment, and the optimized reference values ​​are written into the digital twin model to regenerate the health benchmark library.

[0073] Edge computing layer:

[0074] A lightweight anomaly detection model is deployed to filter data collected by the data acquisition and perception layer and extract key features. Specifically, this includes: a fault handling module that uses a benchmark statistical model to filter abnormal segments and extract statistical features, performs fault detection based on these features, and handles faults in real time using a built-in fault strategy library. The benchmark statistical model employs statistical methods including the 3σ rule and moving average analysis; a data filtering module that filters data collected by the data acquisition and perception layer based on the fault identification model's data results and preset rules, retaining abnormal data and key feature data for uploading to the cloud processing layer; and a scheduling optimization module that establishes a computing node health assessment model and dynamically allocates tasks based on CPU and memory load.

[0075] Cloud processing layer:

[0076] This system is used to construct a digital twin simulation model, combine external data to predict the operating status of new energy power generation equipment, generate dynamic health benchmark values, and perform health diagnosis and management through a large model. Specifically, it includes: a simulation module for constructing a digital twin simulation model, simulating the new energy power generation equipment based on real-time environmental data and external data to obtain predicted operating status data; an anomaly identification module for identifying anomalies based on real-time data and predicted operating status data; and a fault diagnosis module for diagnosing faults in the new energy power generation equipment based on the data results from the anomaly identification module.

[0077] The application service layer is used to provide equipment health assessment, maintenance work order management, visualization services, and decision support based on the data results from the cloud processing layer.

[0078] Specifically, the cloud processing layer, when in operation, includes:

[0079] Step 1: Multi-source data fusion and processing. First, the filtered abnormal data, key feature data, and external environmental parameters are received from the edge computing layer. The data is received through a distributed message queue, cleaned and aligned, and finally converted into a standardized format that can be parsed by the digital twin model.

[0080] Step 2: Dynamically update the digital twin simulation model. A digital twin simulation model is established based on the equipment design parameters through the simulation module. The cleaned data is then injected into the model to simulate the equipment's operation and obtain the equipment's health baseline value.

[0081] Step 3: Anomaly identification. The anomaly identification module calculates the deviation between real-time data and simulation prediction data, uses moving average analysis to identify short-term fluctuations, and uses wavelet transform to detect periodic anomalies.

[0082] Step 4: Fault diagnosis. The fault diagnosis module extracts the statistical and time-frequency features of the abnormal segments to form a high-dimensional feature vector. Fault location and root cause analysis are then performed to generate an interpretable report.

[0083] Specifically, when the cloud processing layer dynamically updates the digital twin simulation model, it includes:

[0084] Step 1: Digital twin simulation model construction. Based on the equipment design parameters, a three-dimensional geometric model of the new energy power generation equipment is constructed using parametric modeling tools, and a finite element mesh is generated through a mesh generation algorithm.

[0085] Step 2: Multiphysics field embedding, including mechanical field, fluid field and thermoelectric coupling. The mechanical field is to solve the stress-strain distribution through finite element analysis. The fluid field is to use computational fluid dynamics to simulate the aerodynamic load of wind speed and irradiance on new energy power generation equipment. The thermoelectric coupling is to establish a correlation model between temperature and power generation efficiency for photovoltaic modules.

[0086] Step 3: Real-time data-driven and dynamic updates, injecting data transmitted from the edge computing layer into the model to dynamically update boundary conditions and load status;

[0087] Step 4: Joint simulation and prediction. Using a digital twin simulation model, the working status of new energy power generation equipment is simulated and predicted.

[0088] Step 5: Dynamic health baseline generation. The probability distribution of the equipment health status is generated through Monte Carlo simulation, and the dynamic baseline value is output in combination with the confidence level assessment. The output dynamic baseline value includes vibration amplitude threshold, temperature drift tolerance and power output parameters.

[0089] The fault diagnosis module, in extracting the statistical and time-frequency features of abnormal segments to form a high-dimensional feature vector, and performing fault location and root cause analysis to generate an interpretable report, specifically includes:

[0090] Step 1: Feature vector construction, extracting statistical and time-frequency features of abnormal segments to form high-dimensional feature vectors;

[0091] Step 2: Embed physical knowledge and combine it with the equipment mechanism model to constrain the diagnostic scope;

[0092] Step 3: Large-scale model semantic analysis, using pre-trained models to parse operation and maintenance logs and historical work order texts, and associate fault modes;

[0093] Step 4: Multimodal fusion reasoning, which uses an attention mechanism to dynamically weight features of different modalities and uses a graph neural network to perform cross-modal relationship reasoning;

[0094] Step 5: Fault Location and Root Cause Analysis. Based on Bayesian networks, the fault distribution probability is calculated, and an interpretability report is generated. This includes:

[0095] The probability output distribution inputs feature vectors into a Bayesian network to output the failure probability of each component, and adjusts the fault warning threshold according to the equipment aging coefficient; the fault propagation path is visualized, generating a multi-level fault chain and marking the confidence of key nodes; interpretable report generation combines the large model to generate a natural language report, which includes a fault cause probability analysis, fault inspection recommendations, and fault repair recommendations.

[0096] The fault diagnosis module is also used for iterative optimization based on actual inspection and repair work orders. The specific process is as follows:

[0097] Step 1: Abnormal event triggering. After manual inspection is completed, the predicted fault type is compared with the actual fault type. When a difference is found, an abnormal event is triggered.

[0098] Step 2: Classify the difference type. If the actual fault type does not match the system diagnosis, it is marked as a diagnostic deviation and the reason for the deviation is recorded. If a fault mode not in the Bayesian network is found, it is marked as a new fault type, triggering the knowledge base expansion process.

[0099] Step 3: Feature vector reconstruction. Extract multi-dimensional features from the deviation cases and establish a diagnostic deviation feature matrix that includes a comparison between the original features and the corrected features of the deviation cases.

[0100] Step 4: Dynamic expansion of the knowledge graph. For new fault types, use a digital twin simulation model to simulate the state data of the corresponding fault type and build new entity relationships, including defining fault node attributes and associated environmental variables.

[0101] Step 5: Model iterative optimization. Gibbs sampling is used to dynamically adjust the parameters of the Bayesian network nodes, and the rationality of the network topology is evaluated through scoring structure learning.

[0102] Example 1: Intelligent Health Management of Wind Turbine Generators

[0103] Scenario: Offshore wind farm, high salt spray and high humidity environment, wind turbine gearboxes and blades are prone to corrosion.

[0104] Detailed implementation process:

[0105] Data Acquisition and Sensing Layer:

[0106] Redundancy Calibration: The gearbox vibration sensor (range 0-200 m / s²) adopts a primary / backup architecture; the primary sensor monitors in real time, and the backup sensor samples synchronously every 5 minutes (timestamp aligned); DS evidence-based decision-making: support is calculated, and when the support is <0.95 for 3 consecutive times, the backup sensor is switched over and a salt spray corrosion alarm is sent. Dynamic Reference Value Compensation: The humidity sensor error is compensated by adjusting the compensation weight through Q-learning; when the humidity is >90%, the compensation coefficient is increased from 0.2 to 0.35.

[0107] Edge computing layer:

[0108] Fault handling: Slipping mean analysis is used to detect sudden increases in vibration (mean increases by 40% within 3 seconds) to determine bearing abnormality; Data filtering: Only abnormal vibration segments (frequency 8-12kHz) and environmental parameters are uploaded; Scheduling optimization: Vibration signals are prioritized under high load, reducing CPU utilization from 90% to 65%.

[0109] Cloud processing layer:

[0110] Digital twin modeling: Constructing a 3D model of the wind turbine blades, embedding CFD simulation of the impact of salt spray deposition on aerodynamic efficiency, predicting a power curve deviation of ±3%. Dynamic health benchmark: Monte Carlo simulation generates vibration thresholds (normal 5-15 m / s², threshold drops to 8 m / s² under salt spray corrosion). Fault diagnosis: Extracting vibration spectrum features, combined with the "abnormal noise" text in the maintenance logs, the Bayesian network outputs a gear wear probability of 82%. Model optimization: Manual inspection revealed insufficient lubrication; Gibbs sampling adjusted the Bayesian network parameters, reducing the false alarm rate for similar faults from 18% to 7%.

[0111] Application service layer:

[0112] An AR work order is generated, marking the location of the bearing requiring lubrication, and recommending historically successful solutions (92% matching rate). The wind turbine health score is displayed on a large visual screen (dropping from 78 points to 65 points), triggering a maintenance alert.

[0113] Technical results: The accuracy of fault early warning under salt spray environment is improved to 95%, and the operation and maintenance cost is reduced by 30%.

[0114] Example 2: Monitoring of Hot Spots on Photovoltaic Power Plant Modules

[0115] Scenario: Desert photovoltaic power station, high temperature and dusty environment, modules are prone to hot spots.

[0116] Detailed implementation process:

[0117] Data Acquisition and Sensing Layer:

[0118] Non-critical parameter calibration: The backplane temperature sensor is dynamically compensated using the environmental correlation matrix (temperature-irradiance), with the compensation coefficient increasing from 0.3 to 0.45 during sandstorms. Redundancy verification: The primary / backup support threshold for the current sensor is set to 0.98; switching occurs after two consecutive deviations.

[0119] Edge computing layer:

[0120] Anomaly Detection: The 3σ rule is used to identify components with abnormal current (daily power generation decrease >30%). Data Filtering: Only infrared image characteristics (temperature >80℃) and irradiance data of hot spot components are uploaded.

[0121] Cloud processing layer:

[0122] Digital twin simulation: Constructing a thermoelectric coupling model of photovoltaic modules to predict hot spot temperature distribution (normal 65℃ → extreme 85℃). Dynamic benchmark: Monte Carlo simulation generates temperature tolerance (threshold ±5℃ at 95% confidence level). Multimodal diagnostics: Fusing infrared image spectrum with the "local blackening" text in the operation and maintenance log, pinpointing a 79% probability of solder joint failure. Knowledge base expansion: Adding a diode breakdown fault mode; 500 sets of simulated data generated by GAN are injected into the training set.

[0123] Application service layer:

[0124] A hot spot cleaning work order is pushed out, and a drone is automatically planned to clean the area. A visual report displays the component efficiency degradation curve and recommends replacing the component with poor soldering.

[0125] Technical effects: Hot spot misjudgment rate reduced by 40%, module lifespan extended by 20%.

[0126] Example 3: Health Management of Energy Storage Battery Packs

[0127] Scenario: Grid-side energy storage power station, high charging and discharging frequency leads to battery aging.

[0128] Detailed implementation process:

[0129] Data Acquisition and Sensing Layer:

[0130] Core parameter calibration: The voltage sensor has primary and backup redundancy (range 0-1000V), switching when the support is <0.98 and the difference is >10V. Dynamic compensation: The temperature compensation coefficient for internal resistance is dynamically adjusted through reinforcement learning, with the weight increasing from 0.2 to 0.5 at low temperatures (-20℃).

[0131] Edge computing layer:

[0132] Fault handling: Extract the second derivative features of the charge / discharge curves to identify sudden changes in internal resistance (rate of change > 15%). Scheduling optimization: Prioritize the allocation of balancing commands to high-load battery clusters.

[0133] Cloud processing layer:

[0134] Digital twin modeling: An electrochemical-thermal coupling model simulates the risk of lithium plating, with a dynamic reference temperature set at 45℃.

[0135] Bayesian diagnostics: Combining voltage drop characteristics with "bulging" text from maintenance records, the lithium plating probability is output as 62%. Parameter optimization: Adjusting the weight of temperature nodes in Gibbs sampling, the detection rate of low-temperature capacity drop faults increased from 68% to 91%.

[0136] Application service layer:

[0137] The liquid cooling system is triggered to increase power by 20%, and a battery replacement list is generated (cells with a health score <60). A visual interface displays the aging distribution of battery clusters and marks high-risk cells.

[0138] Technical benefits: Improves battery thermal runaway early warning rate by 60% and extends cycle life by 25%.

[0139] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A health management platform for new energy power generation equipment based on a large-scale model; characterized by: include: The data acquisition and sensing layer is used to collect operational data in real time through multi-source sensors and IoT devices, and combine it with external data including meteorological parameters and power grid status. The edge computing layer deploys a lightweight anomaly detection model to filter data collected by the data acquisition and perception layer and extract key features. Specifically, the edge computing layer includes: A11: Fault handling module, which uses a benchmark statistical model to filter abnormal segments and extract statistical features, detects faults based on the statistical features, and performs real-time fault handling based on the built-in fault strategy library. The statistical methods used in the benchmark statistical model include the 3σ rule and moving average analysis. A12: Data filtering module, used to filter the data collected by the data acquisition and perception layer according to the data results of the fault identification model and preset rules, and retain abnormal data and key feature data to be uploaded to the cloud processing layer. A13: Scheduling optimization module, used to establish a computing node health assessment model and dynamically allocate tasks based on CPU and memory load; The cloud processing layer is used to build digital twin simulation models, combine external data to predict the operating status of new energy power generation equipment, generate dynamic health benchmark values, and perform health diagnosis and management through large models. The cloud processing layer specifically includes: A21: Simulation module, used to build digital twin simulation models, simulate new energy power generation equipment based on real-time environmental data and external data, and obtain predicted working status data of new energy power generation equipment; A22: Anomaly detection module, used to identify anomalies based on real-time data and predicted working status data; A23: Fault diagnosis module, used to diagnose faults in new energy power generation equipment based on the data results from the anomaly identification module; The application service layer is used to provide equipment health assessment, operation and maintenance work order management, visualization services and decision support based on the data results of the cloud processing layer; Specifically, the cloud processing layer, when in operation, includes: S31: Multi-source data fusion and processing. First, it receives filtered abnormal data, key feature data and external environmental parameters from the edge computing layer. It receives the data through a distributed message queue, cleans and aligns the data, and finally converts the heterogeneous data into a standardized format that can be parsed by the digital twin model. S32: The digital twin simulation model is dynamically updated. The simulation module establishes a digital twin simulation model based on the equipment design parameters, and the cleaned data is injected into the model to simulate the operation of the equipment and obtain the equipment health benchmark value. S33: Anomaly detection. The anomaly detection module calculates the deviation between real-time data and simulation prediction data, uses moving average analysis to identify short-term fluctuations, and uses wavelet transform to detect periodic anomalies. S34: Fault diagnosis. The fault diagnosis module extracts the statistical and time-frequency features of abnormal segments to form a high-dimensional feature vector, and performs fault location and root cause analysis to generate an interpretable report.

2. The health management platform for new energy power generation equipment based on a large model as described in claim 1, characterized in that: The data acquisition and sensing layer also includes an intelligent calibration module, which performs intelligent calibration on the sensors within the data acquisition and sensing layer. For core parameter sensors, the intelligent calibration module employs a primary / backup redundant architecture, specifically including: S11: Data synchronous acquisition, the main sensor acquires data in real time, and the backup sensor samples synchronously at a fixed period in the sleep state, and the data synchronization between the main sensor and the backup sensor is achieved through timestamps; S12: DS evidence theory decision-making first calculates the data support of the main sensor and the backup sensor, and generates a joint probability distribution based on the confidence matrix to determine whether the data of the main sensor and the backup sensor meet the consistency threshold. S13: Offset judgment and switching strategy. If the offset of the main sensor and the backup sensor exceeds the threshold for three consecutive times and the automatic switching logic is triggered, the backup sensor takes over the main control and sends a fault warning. S14: After switching, the original main sensor enters calibration mode and verifies its drift characteristics through standard source input. If it cannot be recovered, it is marked as faulty and needs to be replaced. The standard source input is the reference value data in the health benchmark library.

3. The health management platform for new energy power generation equipment based on a large model according to claim 2, characterized in that: When the intelligent calibration module performs intelligent calibration on sensors within the data acquisition and sensing layer, it uses dynamic reference value weights for error compensation for non-critical parameter sensors, specifically including: S21: Constructing the environmental parameter correlation matrix, identifying non-critical parameters and their sensitive environmental variables, and establishing a compensation coefficient matrix through historical data analysis or physical model derivation; S22: Dynamic adjustment of reference values. The environmental variables are discretized into a state space, and the action space is the adjustment range of the reference values. A decaying learning rate strategy is adopted. Then, the input parameter correlation matrix and the deviation between the current environmental parameters and the threshold are used to calculate the adjustment vector group. S23: Calibration verification and update. A sliding window mechanism is introduced to calculate the error variance after the most recent N adjustments. If the variance continues to increase, the exploration rate is reset, forcing the algorithm to re-explore the environment and write the optimized reference value into the digital twin model to regenerate the health benchmark library.

4. The health management platform for new energy power generation equipment based on a large model as described in claim 3, characterized in that: When the cloud processing layer dynamically updates the digital twin simulation model, it specifically includes: S41: Digital twin simulation model construction. Based on equipment design parameters, a three-dimensional geometric model of new energy power generation equipment is constructed using parametric modeling tools, and a finite element mesh is generated through a mesh generation algorithm. S42: Multiphysics field embedding, including mechanical field, fluid field and thermoelectric coupling. The mechanical field is to solve the stress-strain distribution through finite element analysis. The fluid field is to use computational fluid dynamics to simulate the aerodynamic load of wind speed and irradiance on new energy power generation equipment. The thermoelectric coupling is to establish a correlation model between temperature and power generation efficiency for photovoltaic modules. S43: Real-time data-driven and dynamic updates inject data transmitted from the edge computing layer into the model and dynamically update boundary conditions and load status; S44: Co-simulation and prediction, using digital twin simulation models to simulate and predict the working status of new energy power generation equipment; S45: Dynamic health baseline generation. The probability distribution of the device's health status is generated through Monte Carlo simulation, and the output dynamic baseline value is combined with confidence assessment. The output dynamic baseline value includes vibration amplitude threshold, temperature drift tolerance, and power output parameters.

5. The health management platform for new energy power generation equipment based on a large model according to claim 4, characterized in that: The fault diagnosis module, when extracting statistical and time-frequency features of abnormal segments to form high-dimensional feature vectors, performing fault location and root cause analysis, and generating an interpretable report, specifically includes: S51: Feature vector construction, extracting statistical and time-frequency features of abnormal segments to form high-dimensional feature vectors; S52: Embedding physical knowledge and combining it with equipment mechanism models to constrain the diagnostic scope; S53: Large-scale model semantic analysis, using pre-trained models to parse operation and maintenance logs and historical work order texts, and associate fault modes; S54: Multimodal fusion reasoning, which uses an attention mechanism to dynamically weight features of different modalities and uses graph neural networks for cross-modal relationship reasoning; S55: Fault location and root cause analysis, calculates the fault distribution probability based on Bayesian networks, and generates an interpretable report.

6. The health management platform for new energy power generation equipment based on a large model as described in claim 5, characterized in that: When calculating the fault distribution probability based on Bayesian networks and generating interpretable reports, the specific steps include: S61: Probability output distribution, inputs the feature vector into the Bayesian network, outputs the failure probability of each component, and adjusts the fault warning threshold according to the equipment aging coefficient; S62: Visualize the fault propagation path, generate multi-level fault chains, and label the confidence level of key nodes; S63: Interpretable report generation, which combines a large model to generate a natural language report, including a fault cause probability analysis, fault inspection recommendations, and fault repair recommendations.

7. The health management platform for new energy power generation equipment based on a large model according to claim 6, characterized in that: The fault diagnosis module is also used for iterative optimization based on actual inspection work orders and actual repair work orders. The specific process is as follows: S71: Abnormal event triggering. After manual maintenance is completed, the predicted fault type is compared with the actual fault type. When a difference is found, an abnormal event is triggered. S72: Classification of discrepancy types. If the actual fault type is manually confirmed to be inconsistent with the system diagnosis, it is marked as a diagnostic deviation and the reason for the deviation is recorded. If a fault mode not in the Bayesian network is found, it is marked as a new fault type and the knowledge base expansion process is triggered. S73: Feature vector reconstruction, extracting multi-dimensional features from deviation cases, and establishing a diagnostic deviation feature matrix that includes a comparison between the original features and the corrected features of the deviation cases; S74: Dynamic expansion of the knowledge graph. For new fault types, the digital twin simulation model is used to simulate the state data of the corresponding fault type and construct new entity relationships, including defining fault node attributes and associated environmental variables. S75: Model iterative optimization, using Gibbs sampling to dynamically adjust the parameters of Bayesian network nodes, and evaluating the rationality of the network topology through score row structure learning.

Images