Sentinel-machine-based panoramic monitoring and health assessment system and method for a wind farm

Abstract

The application provides a kind of wind farm panoramic monitoring and health assessment system and method based on sentinel machine group.The system includes: sentinel machine group selection module, representative wind turbine is identified as sentinel machine group based on global simulation technology;Multi-source sensing network, key component structure state and operating environment data are collected on sentinel machine group;Data acquisition and processing platform, fusion processing sentinel machine group sensing data, full field SCADA and CMS data;Health assessment and early warning center, comprehensive assessment and early warning of full field wind turbine based on fusion data.The application selects sentinel machine group and installs special sensor, combines multi-source data fusion and deep learning algorithm, realizes the energy efficiency evaluation of global wind turbine of wind farm, fault early warning and health quantification evaluation, effectively improves operation and maintenance efficiency, reduces unplanned downtime risk.

Description

Technical Field

[0001] This document relates to the field of wind farm monitoring technology, and in particular to a panoramic monitoring and health assessment system and method for wind farms based on sentinel turbines. Background Technology

[0002] With the continuous expansion of wind power installed capacity, wind farm operation and maintenance faces prominent problems such as a large number of equipment, remote distribution, low efficiency of manual inspections, and delayed fault response. Traditional operation and maintenance models rely on regular inspections and post-incident repairs, which make it difficult to achieve real-time perception and early warning of equipment status, resulting in serious losses from unplanned downtime and high operation and maintenance costs.

[0003] To improve the intelligent operation and maintenance level of wind farms, existing technologies have proposed wind turbine condition monitoring and fault diagnosis systems. These systems collect equipment operating data by deploying vibration sensors, temperature sensors, etc., and combine this data with SCADA system data for fault analysis. However, these solutions typically have the following shortcomings: First, sensor deployment lacks specificity, with a uniform monitoring configuration used for all turbines in the farm, resulting in insufficient monitoring accuracy for critical high-risk units and wasted monitoring resources for ordinary units. Second, the degree of multi-source data fusion is low, with vibration, temperature, and electrical parameter data analyzed in isolation, making it difficult to comprehensively reflect the health status of the equipment. Third, fault early warning models rely on a single data source, resulting in limited accuracy and an inability to transfer early warning capabilities from a few benchmark units to all units in the farm.

[0004] Therefore, how to achieve efficient and accurate monitoring and health assessment of wind turbines across the entire wind farm has become a pressing technical problem to be solved in this field. Summary of the Invention

[0005] This invention provides a panoramic monitoring and health assessment system and method for wind farms based on sentinel turbines, aiming to solve the above-mentioned problems.

[0006] According to an embodiment of the present invention, a panoramic monitoring and health assessment system for wind farms based on sentinel turbines is provided, comprising: The sentinel turbine selection module is used to perform load and wind condition analysis on all wind farm locations based on wind farm full-domain simulation technology, and to identify and select representative wind turbines as sentinel turbines using a preset selection method. A multi-source sensor network, deployed on the sentinel unit, includes dedicated sensors for collecting structural status data and operating environment data of key components of the wind turbine. The data acquisition and processing platform communicates with the multi-source sensor network and the existing SCADA and CMS systems of the wind farm, and is used to collect and process sensor data from sentinel units, SCADA operation data of all units in the wind farm and CMS vibration data in real time. The health assessment and early warning center is connected to the data acquisition and processing platform and is used to conduct comprehensive assessment and early warning of all wind turbines based on the fused multi-source data.

[0007] According to an embodiment of the present invention, a method for panoramic monitoring and health assessment of wind farms based on sentinel turbines is provided, comprising: By using the sentinel turbine selection module, load and wind conditions are analyzed for all wind farm turbine locations based on wind farm full-domain simulation technology, and representative wind turbines are identified and selected as sentinel turbines. The structural status data and operating environment data of key wind turbine components are collected through a multi-source sensor network deployed on the sentinel unit. Through the data acquisition and processing platform, sensor data from the sentinel unit, SCADA operation data of all units, and CMS vibration data are collected and processed in real time. The health assessment and early warning center conducts a comprehensive assessment and early warning of all wind turbines based on the fused multi-source data.

[0008] By employing the embodiments of this invention, high-precision sensing resources are precisely deployed to high-risk, highly representative wind turbines through the sentinel turbine selection module, achieving a balance between monitoring accuracy and cost. The data acquisition and processing platform integrates data from dedicated sensors on the sentinel turbines with SCADA and CMS data from the entire wind farm, forming a comprehensive, three-dimensional monitoring network. The fault early warning module uses an attention mechanism deep neural network combined with transfer learning and physical information fusion to significantly reduce false alarm and missed alarm rates. The health assessment module quantifies and visualizes the health status of wind turbines, providing a scientific basis for operation and maintenance decisions. The overhead line monitoring subsystem accurately locates potential discharge hazards in the collection lines, forming a comprehensive monitoring system for the core assets of the wind farm, thus comprehensively improving the safety and reliability of wind farm operation. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of a wind farm panoramic monitoring and health assessment system based on sentinel turbines, according to an embodiment of the present invention. Figure 2 This is a flowchart of the wind farm panoramic monitoring and health assessment method based on sentinel generators, according to an embodiment of the present invention. Detailed Implementation

[0011] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0012] System Implementation Examples According to embodiments of the present invention, a panoramic monitoring and health assessment system for wind farms based on sentinel turbines is provided. Figure 1 This is a schematic diagram of a wind farm panoramic monitoring and health assessment system based on sentinel turbines, according to an embodiment of the present invention. Figure 1 As shown, the wind farm panoramic monitoring and health assessment system based on sentinel turbines according to an embodiment of the present invention specifically includes: The sentinel turbine selection module is used to analyze the load and wind conditions of all wind farm turbine locations based on wind farm full-domain simulation technology. It identifies and selects representative wind turbines as sentinel turbines using a preset selection method. The sentinel turbine selection module is the core of the point selection in this embodiment of the invention. Its core function is to conduct a refined analysis of the load and wind conditions of all wind farm turbine locations based on wind farm full-domain simulation technology and combined with multi-dimensional information such as wind farm geography, meteorology, and equipment layout. Through a scientific selection method, it selects wind turbines with high load representativeness, typical wind conditions, and wake influence radiation as sentinel turbines, thus solving the problems of resource waste and insufficient monitoring of key points caused by the indiscriminate deployment of sensors across the entire domain in traditional monitoring.

[0013] The sentry unit selection module is specifically used for: A holographic digital twin model of a wind farm is established. The geographical information of the wind farm, the layout of wind turbines, the topography, and historical meteorological data are input into the holographic digital twin model. Computational fluid dynamics (CFD) simulation algorithm and wake evolution algorithm are coupled to simulate the wind resource distribution characteristics of each wind turbine throughout its entire life cycle under different wind conditions, seasons, and terrain conditions. At the same time, the dynamic load spectrum of core components such as wind turbine blades, towers, and transmission chains is calculated, including key load data such as fatigue load, ultimate load, and instantaneous impact load, so as to achieve a quantitative characterization of the load state of each wind turbine.

[0014] Based on the dynamic load spectrum obtained from simulation, a load damage evolution matrix is ​​constructed. Combining material fatigue damage theory, the cumulative damage rate and remaining life decay curve of key components of the computer group, the probability distribution of the cumulative damage rate of all units in the wind farm is fitted, and the inflection point or abrupt change point on the cumulative damage rate distribution curve is identified. Units located after the inflection point or abrupt change point are identified as initial candidate sentinel units. These units are the core monitoring objects with the highest load damage risk and the highest failure probability in the wind farm. Representative propagation coefficients are calculated for initial candidate sentinel units. Combining the two core indicators of cumulative damage rate and representative propagation coefficient, a multi-objective optimization algorithm (NSGA-Ⅲ algorithm) is used to solve the problem. Under the premise of comprehensive monitoring coverage and optimal resource investment, the final number and distribution of sentinel units are determined. In this system, the number of sentinel units is preferably 10% to 15% of the total number of units in the wind farm, and full coverage of different wind condition areas, different load characteristic areas, and different wake influence areas is achieved. The representative propagation coefficient is used to characterize the degree to which the load characteristics of the unit can affect the surrounding units through wake propagation. The higher the propagation coefficient, the more its operating status reflects the typical operating conditions of the surrounding units. This module outputs information such as the location number, distribution coordinates, and key monitoring points of the sentinel units, providing accurate location data for the subsequent deployment of multi-source sensor networks. At the same time, it synchronizes the simulated load and wind condition data to the data acquisition and processing platform as the basis for subsequent health assessments.

[0015] A multi-source sensor network, deployed on the sentinel turbine units, includes dedicated sensors for collecting structural status data and operating environment data of key wind turbine components. The network deploys multiple types and locations of sensors targeting core vulnerable components such as wind turbine blades, towers, drive trains, and foundation rings, as well as key operating environment factors such as incoming wind. All sensors are industrial-grade high-precision equipment, meeting the operational requirements of harsh outdoor environments in wind farms (high and low temperatures, high humidity, strong winds, electromagnetic interference), and supporting real-time data acquisition and wireless transmission. The sensor deployment and functions of this network are as follows: Tower foundation ring monitoring: A fiber optic strain sensor array is deployed at the interface between the tower foundation ring and the concrete. The array layout enables full circumferential monitoring of the foundation ring, and real-time acquisition of strain data of the foundation ring is used to monitor uneven settlement of the tower, cracking of the concrete around the foundation ring, and stress deformation of the foundation ring itself, so as to detect early structural damage to the tower foundation in a timely manner.

[0016] Tower dynamic attitude monitoring: Dual-axis tilt sensors are deployed in the middle and top of the tower to collect the roll and pitch angle data of the tower in real time. Combined with time series analysis, the dynamic tilt and sway trajectory of the tower is generated, and the sway displacement of the tower is quantitatively characterized. This enables real-time monitoring of changes in the verticality and stiffness of the tower, and provides early warning of risks such as tower tilt and structural fatigue.

[0017] Blade root load monitoring: Fiber optic strain sensors are deployed at the flange position at the blade root, arranged at multiple points along the blade spanwise and chordwise, to monitor the changes in bending moment and torque at the blade root in real time, accurately capture the load characteristics of the blade under different wind conditions, and provide early warning of problems such as blade load exceeding limits and load imbalance, providing data support for the assessment of the blade structural health status.

[0018] Operating environment wind condition monitoring: A lidar wind measurement device is deployed on the top of the nacelle to measure the wind speed, wind direction, turbulence intensity, wind shear and other wind characteristics of the wind flowing in front of the nacelle in real time, providing high-precision environmental wind input data for the sentinel unit. At the same time, this data can be used to correct the global simulation model of the wind farm and improve the accuracy of the simulation results.

[0019] Early damage monitoring of the transmission chain: Multi-channel acoustic emission sensors are deployed on the gearbox housing and main bearing housing. The sensors are closely attached to the equipment surface to capture acoustic emission signals generated by the initiation and development of early micro-cracks in the gearbox and main bearing during operation in real time. This enables accurate identification of early hidden damage in the transmission chain and solves the problem that traditional vibration monitoring is difficult to detect micro-damage.

[0020] Tower structure stiffness monitoring: A vibration sensor array is deployed around the tower portal, which is a weak area of ​​the tower structure and is prone to stress concentration and stiffness degradation. By collecting vibration signals around the tower portal, the local modal changes of the tower are monitored, and this data is fused and analyzed with the strain data collected by fiber optic strain sensors to accurately identify the location and degree of stiffness degradation of the tower structure, thereby realizing the location and quantification of tower structure damage.

[0021] The data acquisition and processing platform communicates with the multi-source sensor network and the existing SCADA and CMS systems of the wind farm, and is used to collect and process sensor data from sentinel units, SCADA operation data of all units in the wind farm and CMS vibration data in real time. The data acquisition and processing platform is specifically used for: Perform protocol parsing and format unification processing on the accessed multi-source heterogeneous data; The unified data is time-series aligned and missing values ​​are imputed to form a standardized multi-source fusion dataset. The multi-source fusion dataset is pushed to the health assessment and early warning center and the external centralized control system respectively according to the preset data interface specifications.

[0022] The health assessment and early warning center is connected to the data acquisition and processing platform and is used to conduct comprehensive assessment and early warning of all wind turbines based on the fused multi-source data.

[0023] The health assessment and early warning center includes: The energy efficiency assessment module analyzes the power generation, availability, and power curves of all wind turbines in the wind farm based on SCADA data to evaluate turbine performance. It also performs multi-dimensional and quantitative energy efficiency assessments of each turbine's power generation performance based on SCADA operational data from the fused data set, combined with wind farm wind resource data. This identifies turbines with low energy efficiency and key factors affecting power generation efficiency, providing a basis for wind farm power optimization and operation and maintenance scheduling. The module's assessment index system covers three main categories: power generation indicators, equipment utilization indicators, and power characteristic indicators, specifically including: Power generation indicators: actual power generation, theoretical power generation, wind curtailment rate, and breakdown of power loss; Equipment utilization metrics: time availability, energy availability, mean time between failures (MTBF), mean time to repair (MTTR). Power characteristic indicators: power curve conformity, wind energy utilization coefficient, and power characteristic consistency coefficient.

[0024] The fault early warning module is used to combine fused multi-source data and run a preset diagnostic model to provide fault early warnings for key components of the entire wind turbine field. This module, in conjunction with the fused multi-source data, runs a preset intelligent diagnostic model to provide early fault warnings, fault location, and fault level classification for key components such as blades, towers, drive trains, and generators of the entire wind turbine field. This addresses the problems of traditional fault diagnosis relying on a single data source, low early warning accuracy, and strong lag. The fault early warning module includes: The multimodal feature extraction unit is used to extract time-domain features, frequency-domain features, and time-frequency-domain features from the multi-source data fused by the data acquisition and processing platform, and to construct a high-dimensional feature vector characterizing the operating status of the equipment. The deep neural network model based on the attention mechanism takes the high-dimensional feature vector as input, automatically identifies the contribution of different sensor data to fault identification through attention weight allocation, and outputs the health status and fault probability of each key component. The transfer learning unit is used to transfer the parameters of the deep neural network model trained on the sentinel units to all non-sentinel units in the field through the domain adaptation method. The SCADA and CMS data of the non-sentinel units are used to fine-tune the transferred model to achieve fault early warning for all wind turbines. The physical information fusion unit is used to fuse the fault probability output by the deep neural network model with the load damage evolution result based on the physical model, and triggers a confirmation alarm when both exceed a preset threshold.

[0025] The health assessment module integrates energy efficiency evaluation results, fault statistical analysis, and fault early warning information to quantitatively evaluate and visually display the health status of each wind turbine in the entire site. The health assessment module uses the analytic hierarchy process (AHP) to determine the weight of each evaluation indicator, constructing a multi-level evaluation indicator system. The primary indicators include energy efficiency status (weight 30%), fault history status (weight 20%), and real-time operating status (weight 50%). Each primary indicator is further subdivided into several secondary indicators. The health score of a single wind turbine is calculated through a weighted summation method, and based on the score, the wind turbine's health status is divided into four levels: healthy (90-100 points), sub-healthy (70-89 points), concerning (50-69 points), and severe (0-49 points).

[0026] This module provides a panoramic view of the health status of all wind turbines in the wind farm through a visual interface. It supports filtering and querying by health status, turbine location, components, and other dimensions. It also generates a trend curve of wind turbine health status changes, providing a scientific basis for predictive maintenance, operation and maintenance resource scheduling, and equipment life assessment of wind farms, and realizing the transformation of operation and maintenance mode from post-maintenance to predictive maintenance.

[0027] Furthermore, the wind farm panoramic monitoring and health assessment system based on sentinel turbine units also includes an overhead line monitoring subsystem, comprising: Traveling wave monitoring terminals deployed on collector lines are used to collect fault traveling wave current and power frequency fault current at the time of line faults. The diagnostic analysis module communicates with the traveling wave monitoring terminal and is used to accurately locate and warn of potential discharge hazards in overhead lines based on the received fault information.

[0028] By employing the embodiments of the present invention, the following beneficial effects are achieved: The sentinel turbine selection module precisely deploys high-precision sensing resources to high-risk, highly representative turbines, achieving a balance between monitoring accuracy and cost. The data acquisition and processing platform integrates data from dedicated sensors on the sentinel turbines with data from the entire SCADA and CMS systems, forming a comprehensive, three-dimensional monitoring network. The fault early warning module employs an attention mechanism deep neural network combined with transfer learning and physical information fusion to significantly reduce false alarm and missed alarm rates. The health assessment module quantifies and visualizes the health status of wind turbines, providing a scientific basis for operation and maintenance decisions. The overhead line monitoring subsystem accurately locates potential discharge hazards in the collection lines, forming a comprehensive monitoring system for the core assets of the wind farm, thus comprehensively improving the safety and reliability of wind farm operation.

[0029] Method Implementation Examples According to embodiments of the present invention, a method for panoramic monitoring and health assessment of wind farms based on sentinel turbines is provided. Figure 2 This is a flowchart of a wind farm panoramic monitoring and health assessment method based on sentinel turbines, according to an embodiment of the present invention. Figure 2 As shown, the wind farm panoramic monitoring and health assessment method based on sentinel turbines according to an embodiment of the present invention specifically includes: S1. Using the sentinel turbine selection module, based on the wind farm full-domain simulation technology, load and wind conditions are analyzed for all wind farm turbine locations, and representative wind turbines are identified and selected as sentinel turbines. S2. Collect structural status data and operating environment data of key components of the wind turbine through a multi-source sensor network deployed on the sentinel unit; S3. Through the data acquisition and processing platform, sensor data from the sentinel unit, SCADA operation data of all units, and CMS vibration data are collected and processed in real time. S4. Through the health assessment and early warning center, a comprehensive assessment and early warning of all wind turbines in the field is conducted based on the fused multi-source data.

[0030] The comprehensive assessment and early warning of all wind turbines in the field through the health assessment and early warning center, based on fused multi-source data, includes: The energy efficiency assessment module analyzes the power generation, availability, and power curves of all wind turbines in the field based on SCADA data to generate energy efficiency assessment results. Based on the multi-source fusion dataset, the fault early warning module runs a preset diagnostic model to provide fault early warnings for key components of the entire wind turbine and generate fault early warning information. The health evaluation module integrates the energy efficiency assessment results, fault statistical analysis, and fault early warning information to conduct a quantitative health evaluation of each wind turbine in the entire site, generating a health evaluation score.

[0031] The step of using the fault early warning module to run a preset diagnostic model based on the multi-source fusion dataset to provide fault early warning for key components of the entire wind turbine specifically includes: The multimodal feature extraction unit extracts time-domain features, frequency-domain features, and time-frequency-domain features from the multi-source fusion dataset to construct a high-dimensional feature vector characterizing the device's operating status. Using a deep neural network model based on an attention mechanism, with the high-dimensional feature vector as input, the contribution of different sensor data to fault identification is automatically identified through attention weight allocation, and the health status and fault probability of each key component are output. Through the transfer learning unit, the parameters of the deep neural network model trained on the sentinel unit are transferred to all non-sentinel units in the field through the domain adaptation method. The SCADA and CMS data of the non-sentinel units are used to fine-tune the transferred model to achieve fault early warning of wind turbines in the entire field. The physical information fusion unit fuses the fault probability output by the deep neural network model with the load damage evolution result based on the physical model. When both exceed a preset threshold, a confirmation alarm is triggered.

[0032] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A panoramic monitoring and health assessment system for wind farms based on sentinel turbines, characterized in that... include: The sentinel turbine selection module is used to perform load and wind condition analysis on all wind farm locations based on wind farm full-domain simulation technology, and to identify and select representative wind turbines as sentinel turbines using a preset selection method. A multi-source sensor network, deployed on the sentinel unit, includes dedicated sensors for collecting structural status data and operating environment data of key components of the wind turbine; The data acquisition and processing platform communicates with the multi-source sensor network and the existing SCADA and CMS systems of the wind farm, and is used to collect and process sensor data from sentinel units, SCADA operation data of all units in the wind farm and CMS vibration data in real time. The health assessment and early warning center is connected to the data acquisition and processing platform and is used to conduct comprehensive assessment and early warning of all wind turbines based on the fused multi-source data.

2. The system according to claim 1, characterized in that, The sentry unit selection module is specifically used for: A holographic digital twin model of a wind farm is established. The geographical information of the wind farm, the layout of wind turbines, the topography and geomorphology, and historical meteorological data are input into the holographic digital twin model. By coupling CFD simulation and wake evolution algorithm, the dynamic load spectrum of each wind turbine during its entire life cycle is simulated. Based on the dynamic load spectrum, a load damage evolution matrix is ​​constructed, and the cumulative damage rate and remaining life decay curve of the key components of the computer group are obtained. The cumulative damage rate of the entire field unit is fitted to the distribution, and the inflection point or abrupt change point on the cumulative damage rate distribution curve is identified. The unit located after the inflection point or abrupt change point is determined as the initial candidate sentinel unit. Representative propagation coefficients are calculated for the initial candidate sentry units. Combining the cumulative damage rate and the representative propagation coefficients, a multi-objective optimization algorithm is used to determine the final number and distribution of sentry units. The representative propagation coefficients are used to characterize the extent to which the load characteristics of the unit can affect the surrounding units through wake propagation.

3. The system according to claim 1, characterized in that, The multi-source sensor network includes: A fiber optic strain sensor array deployed at the junction of the tower foundation ring and the concrete is used to monitor uneven settlement of the tower and cracking of the concrete around the foundation ring. Dual-axis tilt sensors deployed in the middle and top of the tower are used to monitor the dynamic tilt and sway trajectory of the tower. A fiber optic strain sensor deployed at the blade root is used to monitor changes in bending moment and torque at the blade root in real time. The lidar deployed on the top of the cabin is used to measure the wind speed, wind direction and turbulence characteristics of the wind flowing in front of the cabin in real time, providing the sentry crew with environmental wind input data; Multichannel acoustic emission sensors deployed in gearboxes and main bearing housings are used to capture acoustic emission signals of early microcrack initiation and development. An array of vibration sensors deployed around the tower's portal openings is used to monitor local modal changes in the tower and fuse them with data from the fiber optic strain sensor to identify the location of stiffness degradation and damage in the tower structure.

4. The system according to claim 1, characterized in that, The data acquisition and processing platform is specifically used for: Perform protocol parsing and format unification processing on the accessed multi-source heterogeneous data; The unified data is time-series aligned and missing values ​​are imputed to form a standardized multi-source fusion dataset. The multi-source fusion dataset is pushed to the health assessment and early warning center and the external centralized control system respectively according to the preset data interface specifications.

5. The system according to claim 1, characterized in that, The health assessment and early warning center includes: The energy efficiency assessment module is used to analyze the power generation, availability, and power curves of all wind turbines in the field based on SCADA data to evaluate the performance of the wind turbines. The fault early warning module is used to combine the fused multi-source data, run the preset diagnostic model, and provide fault early warning for key components of the wind turbines throughout the field. The health assessment module is used to comprehensively analyze energy efficiency assessment results, fault statistics and early warning information, and to quantitatively assess and visually display the health of each wind turbine in the entire site.

6. The system according to claim 5, characterized in that, The fault early warning module includes: The multimodal feature extraction unit is used to extract time-domain features, frequency-domain features, and time-frequency-domain features from the multi-source data fused by the data acquisition and processing platform, and to construct a high-dimensional feature vector characterizing the operating status of the equipment. The deep neural network model based on the attention mechanism takes the high-dimensional feature vector as input, automatically identifies the contribution of different sensor data to fault identification through attention weight allocation, and outputs the health status and fault probability of each key component. The transfer learning unit is used to transfer the parameters of the deep neural network model trained on the sentinel units to all non-sentinel units in the field through the domain adaptation method. The SCADA and CMS data of the non-sentinel units are used to fine-tune the transferred model to achieve fault early warning for all wind turbines. The physical information fusion unit is used to fuse the fault probability output by the deep neural network model with the load damage evolution result based on the physical model, and triggers a confirmation alarm when both exceed a preset threshold.

7. The system according to claim 1, characterized in that, The system also includes an overhead line monitoring subsystem, comprising: Traveling wave monitoring terminals deployed on collector lines are used to collect fault traveling wave current and power frequency fault current at the time of line faults. The diagnostic analysis module communicates with the traveling wave monitoring terminal and is used to accurately locate and warn of potential discharge hazards in overhead lines based on the received fault information.

8. An evaluation method based on the wind farm panoramic monitoring and health assessment system based on sentinel turbines as described in any one of claims 1-7, characterized in that, include: By using the sentinel turbine selection module, load and wind conditions are analyzed for all wind farm turbine locations based on wind farm full-domain simulation technology, and representative wind turbines are identified and selected as sentinel turbines. The structural status data and operating environment data of key wind turbine components are collected through a multi-source sensor network deployed on the sentinel unit. Through the data acquisition and processing platform, sensor data from the sentinel unit, SCADA operation data of all units, and CMS vibration data are collected and processed in real time. The health assessment and early warning center conducts a comprehensive assessment and early warning of all wind turbines based on the fused multi-source data.

9. The method according to claim 8, characterized in that, The comprehensive assessment and early warning of all wind turbines in the field through the health assessment and early warning center, based on fused multi-source data, includes: The energy efficiency assessment module analyzes the power generation, availability, and power curves of all wind turbines in the field based on SCADA data to generate energy efficiency assessment results. Based on the multi-source fusion dataset, the fault early warning module runs a preset diagnostic model to provide fault early warnings for key components of the entire wind turbine and generate fault early warning information. The health evaluation module integrates the energy efficiency assessment results, fault statistical analysis, and fault early warning information to conduct a quantitative health evaluation of each wind turbine in the entire site, generating a health evaluation score.

10. The method according to claim 9, characterized in that, The step of using the fault early warning module to run a preset diagnostic model based on the multi-source fusion dataset to provide fault early warning for key components of the entire wind turbine specifically includes: The multimodal feature extraction unit extracts time-domain features, frequency-domain features, and time-frequency-domain features from the multi-source fusion dataset to construct a high-dimensional feature vector characterizing the device's operating status. Using a deep neural network model based on an attention mechanism, with the high-dimensional feature vector as input, the contribution of different sensor data to fault identification is automatically identified through attention weight allocation, and the health status and fault probability of each key component are output. Through the transfer learning unit, the parameters of the deep neural network model trained on the sentinel unit are transferred to all non-sentinel units in the field through the domain adaptation method. The SCADA and CMS data of the non-sentinel units are used to fine-tune the transferred model to achieve fault early warning of wind turbines in the entire field. The physical information fusion unit fuses the fault probability output by the deep neural network model with the load damage evolution result based on the physical model. When both exceed a preset threshold, a confirmation alarm is triggered.

Images