Health assessment methods and systems for equipment groups applicable to integrated wind, solar, hydro, and storage operation

Abstract

This invention belongs to the field of operation and maintenance technology for wind-solar-hydro-storage integrated commissioning systems, specifically relating to a health assessment method and system for equipment groups suitable for wind-solar-hydro-storage integrated commissioning. It includes the following steps: S1, panoramic perception data acquisition and management; S2, multimodal large-scale model construction; S3, cross-energy equipment group health standard construction; S4, dynamic early warning; S5, model verification and iterative optimization. It aims to solve the technical problems of fragmented equipment perception, inconsistent equipment group health standards, and poor coordination between operation and maintenance and commissioning in wind-solar-hydro-storage integrated commissioning scenarios.

Description

Technical Field

[0001] This invention belongs to the field of operation and maintenance technology of wind, solar, hydro and storage integrated commissioning system equipment groups, specifically involving a health assessment method and system for equipment groups applicable to wind, solar, hydro and storage integrated commissioning. Background Technology

[0002] With the rapid development of the new energy industry, technologies related to the condition assessment of new energy power plant equipment have also developed rapidly. For example, Chinese patent document CN120832503B discloses a method for condition assessment and early warning of new energy power plant equipment. This method includes: S1: Real-time collection of operational, environmental, and inspection image data through multi-source sensing devices, followed by standardized processing to generate a dataset; S2: Selection of a large AI model supporting multimodal input, incorporating fault knowledge graphs and industry standards, pre-training with general power data, and fine-tuning with power plant-specific data to generate an adapted model; S3: Extracting time-domain / frequency-domain features and image damage information, dynamically weighting and constructing a unified health index; S4: Predicting the HI value for the next 24 hours based on LSTM, dynamically adjusting thresholds according to operating conditions, and issuing early warnings in three levels; S5: Updating the model quarterly, verifying indicator requirements, and optimizing rules through false alarm analysis.

[0003] Currently, the integrated commissioning model of wind power, photovoltaic, hydropower, and energy storage systems (hereinafter referred to as: wind-solar-hydro-storage) has become a core direction for energy structure transformation. Through the complementary characteristics of wind power, photovoltaic, and hydropower, and the peak-shaving and valley-filling capabilities of energy storage systems, it can significantly improve the stability and efficiency of energy supply. The integrated commissioning scenario of wind-solar-hydro-storage is characterized by a complex array of equipment types (including wind turbines, photovoltaic inverters, hydro-generators, energy storage converters, etc.), variable operating conditions, significant environmental differences, and dispersed data sources. The aforementioned patents only focus on a single new energy power station (wind power or photovoltaic power station), lacking system-level integrated commissioning and assessment, having a single data dimension, insufficient depth of multi-source heterogeneous integration, failing to cover hydropower and energy storage systems, and being unable to adapt to heterogeneous equipment clusters that complement wind, solar, hydro, and storage, resulting in insufficient dynamic early warning capabilities.

[0004] Therefore, existing technologies cannot perform health assessments on integrated wind, solar, hydro, and storage system equipment clusters. Specifically: First, existing technologies are mostly designed for single energy devices or single-dimensional data collection, lacking panoramic perception capabilities that cover the entire scenario of wind, solar, hydro, and storage. Multi-source heterogeneous data (operational data, environmental data, and inspection image data) are difficult to integrate effectively, resulting in a fragmented perception system that cannot provide comprehensive data support for determining the health of equipment groups in joint commissioning scenarios.

[0005] Second, due to the lack of coordination and adaptability in the health evaluation indicators, weight settings, and fault classifications of different energy equipment, there is a problem of inconsistent health standards for equipment groups, which in turn makes it difficult to compare the health status of equipment groups across different equipment and energy types. At the same time, existing health standards are mostly applicable to single equipment and cannot meet the overall operation, maintenance, and joint commissioning needs of equipment groups.

[0006] Third, current methods for assessing equipment health rely heavily on fixed thresholds and expert experience, with weights failing to dynamically adapt to different equipment operating conditions and the collaborative characteristics of equipment clusters. Existing general-purpose technologies such as multimodal large-scale models and AHP are not optimized for wind-solar-hydro-storage integrated commissioning scenarios, resulting in issues such as feature extraction bias, low judgment accuracy, and insufficient early warning lead time, which can easily lead to unplanned downtime. Directly applying existing general-purpose technologies such as multimodal large-scale models and AHP to the equipment clusters of wind-solar-hydro-storage integrated commissioning systems can easily affect the stability of these systems.

[0007] Fourth, the application results of health standards are disconnected from the joint commissioning strategy of wind, solar, hydro and storage, making it impossible to optimize the joint commissioning and scheduling scheme through the health status of equipment groups. Existing technologies are unable to achieve deep synergy between health assessment and joint commissioning. Summary of the Invention

[0008] The technical problem to be solved by this invention is to provide a method and system for assessing the health of equipment groups suitable for integrated wind, solar, hydro, and storage commissioning. It aims to address the technical issues of fragmented equipment sensing, inconsistent health standards for equipment groups, and poor coordination between operation and maintenance and commissioning in integrated wind, solar, hydro, and storage commissioning scenarios.

[0009] The technical solution adopted by this invention to solve its technical problem is: a health assessment method for equipment groups applicable to the joint commissioning of wind, solar, hydro, and storage systems, comprising the following steps: S1, Panoramic Perception Data Acquisition and Governance: Construct a multi-source perception network covering the entire scenario of wind, solar, hydro, and energy storage, collect panoramic perception data, including the operation characteristic data, environmental adaptation characteristic data, and inspection image characteristic data of various equipment in wind power, photovoltaic, hydro power, and energy storage systems. Based on the data source characteristics of different equipment, the panoramic perception data is standardized and governed to generate a standardized panoramic perception dataset. S2, Multimodal Large Model Construction: Select an AI large model that supports input of time series data, image data and standardized panoramic perception datasets, inject knowledge graphs of fault mechanisms of wind, solar, hydro and storage equipment and industry operation and maintenance procedures, first use general power datasets and public datasets of wind, solar, hydro and storage for pre-training, then inject dedicated data for wind, solar, hydro and storage joint commissioning for a second training, and use LoRA technology to adjust the weights of operation features, environmental adaptation features and inspection image features to generate a multimodal large model adapted to the joint commissioning scenario; S3, Construction of health standards across energy equipment groups: Based on the multimodal large model in step S2, extract the time domain features, frequency domain features and image features of each device, and combine the operating characteristics of different devices in the wind, solar, hydro and storage system with the collaborative requirements of the equipment group to construct health standards across energy equipment groups; S4, Dynamic Early Warning: Based on the health standards of cross-energy equipment groups, combined with the time-series reasoning capabilities of the LSTM model and the multimodal large model in step S2, the unified health index of single equipment and equipment groups is predicted within a certain period of time in the future, and fault early warning information is generated. Combined with the equipment degradation characteristics and joint commissioning adaptation strategies of wind, solar, hydro and storage, the fault early warning threshold and early warning classification strategy are dynamically optimized to achieve dynamic early warning. S5, Model Validation and Iterative Optimization: Through historical data backtracking and field verification, verify the effectiveness of fault early warning and the rationality of cross-energy equipment group health standards. Regularly update the multimodal large model with new data. Based on false alarm analysis, update the equipment degradation characteristics in step S4 and the cross-energy equipment group health standards in step S3.

[0010] Furthermore, in step S1, operational characteristic data is obtained by connecting to the SCADA system via the DL / T634.5104 protocol; environmental adaptation characteristic data is collected through the environmental sensing system; inspection image characteristic data is integrated through the API interface, with a data delay of ≤10 seconds; missing data is filled by linear interpolation, with a unified time granularity of 5 minutes average; and the collected data is labeled with device ID, energy type, collection time, and operating condition label.

[0011] Furthermore, in step S2, the knowledge graph includes causal relationship chains of faults in various equipment of wind power, photovoltaic, hydropower and energy storage systems; the industry operation and maintenance procedures include GB / T38335, NB / T11346 and NB / T11570 standards.

[0012] Furthermore, in steps S3 and S4, the formula for calculating the Unified Health Index (HI) is as follows: HI = A × Operational Feature Score + B × Environment Adaptation Feature Score + C × Inspection Image Feature Score; Where A is the weight coefficient of the running feature, B is the weight coefficient of the environment adaptation feature, and C is the weight coefficient of the inspection image feature, satisfying A+B+C=1 and A>0, B>0, C>0; A, B, and C are dynamically adjusted according to equipment type using the Analytic Hierarchy Process (AHP), and their final weights are determined after a consistency test (CR < 0.1).

[0013] Furthermore, the formula for calculating the running feature score is as follows: Running feature score = (Actual feature parameter compliance rate) × 100; Actual feature parameter compliance rate = Number of feature parameters that meet the rated threshold range / Total number of feature parameters.

[0014] Furthermore, the formula for calculating the environment adaptation feature score is as follows: Environmental adaptability feature score = Σ (adaptability coefficient of single environmental factor × factor weight) × 100; Single environmental factor adaptation coefficient = 1 - |actual environmental parameter - optimal adaptation parameter| / (tolerance upper limit - tolerance lower limit); When the calculated result of the single environmental factor fit coefficient is less than 0, the single environmental factor fit coefficient is forced to be 0.

[0015] Furthermore, the formula for calculating the feature score of the inspection image is as follows: Inspection image feature score = 100 - Σ (fault level weight × number of faults); The fault levels are divided into minor faults, general faults, and serious faults, with weights of 0.1 for minor faults, 0.3 for general faults, and 0.6 for serious faults.

[0016] Furthermore, the method also includes the step of: the unified health index HI is normalized using min-max. The normalization formula is: HI_norm=(HI-HI_min) / (HI_max-HI_min)×100; in, HI_min represents the lowest historical health index of the corresponding energy type of equipment; HI_max is the highest historical health index of the corresponding energy type of equipment; HI_norm is the normalized standard health index.

[0017] Furthermore, the early warning classification strategy in step S4 is as follows: When HI_norm≤30, it is a Level 1 warning, which is pushed to the operation and maintenance terminal and triggers an audible and visual alarm, and the linkage and debugging system adjusts the equipment load. When 30 < HI_norm ≤ 60, it is a level 2 warning, and a work order to be processed is generated; When 60 < HI_norm ≤ 80, it is a Level 3 warning and should be included in the periodic inspection plan; No warning is given when 80 < HI_norm.

[0018] A health assessment system for equipment clusters suitable for wind, solar, hydro, and storage integrated commissioning adopts a health assessment method for equipment clusters suitable for wind, solar, hydro, and storage integrated commissioning. It includes a panoramic perception data acquisition and management module, a multimodal large model construction module, a cross-energy equipment cluster health standard construction module, a dynamic early warning module, and a model verification and iterative optimization module. The panoramic perception data acquisition and management module is configured to build a multi-source perception network covering the entire scenario of wind, solar, hydro and energy storage, and to collect panoramic perception data. The panoramic perception data includes the operation characteristic data, environmental adaptation characteristic data and inspection image characteristic data of each device in the wind power, photovoltaic, hydro power and energy storage system. The panoramic perception data is standardized and managed according to the data source characteristics of different devices to generate a standardized panoramic perception dataset. The multimodal large model construction module is configured to select AI large models that support input of time series data, image data and standardized panoramic perception datasets, and inject knowledge graphs of fault mechanisms of wind, solar and hydro storage equipment and industry operation and maintenance procedures. It first uses general power datasets and public datasets of wind, solar and hydro storage for pre-training, and then injects dedicated data for wind, solar and hydro storage joint commissioning for a second training. It uses LoRA technology to adjust the weights of operating features, environmental adaptation features and inspection image features to generate a multimodal large model adapted to the joint commissioning scenario. The cross-energy equipment group health standard construction module is configured based on the multimodal large model in step S2, extracts the time domain features, frequency domain features and image features of each device, and combines the operating characteristics of different devices in the wind, solar, water and storage system with the collaborative requirements of the equipment group to construct the cross-energy equipment group health standard. The dynamic early warning module is configured to predict the unified health index of a single device and a group of devices within a certain period of time based on the health standard of cross-energy equipment groups, combined with the temporal reasoning capabilities of the LSTM model and the multimodal large model in step S2, and generate fault early warning information. Combining the equipment degradation characteristics and joint commissioning adaptation strategies of wind, solar, hydro and storage, it dynamically optimizes the fault early warning threshold and early warning classification strategy to achieve dynamic early warning. The model verification and iterative optimization module is configured to verify the effectiveness of fault warning and the rationality of cross-energy equipment group health standards through historical data backtracking and field verification, periodically update the multimodal large model with new data, and update the equipment degradation characteristics in step S4 and the cross-energy equipment group health standards in step S3 based on false alarm analysis.

[0019] Compared with existing technologies, the beneficial effects of this invention are: This invention provides a method and system for assessing the health of equipment groups suitable for integrated wind, solar, hydro, and storage commissioning. It aims to solve the technical problems of fragmented equipment perception, inconsistent health standards for equipment groups, and poor coordination between operation and maintenance and commissioning in integrated wind, solar, hydro, and storage commissioning scenarios. It achieves panoramic fusion of multi-source data, unified health assessment across energy equipment groups, and dynamic adaptation of early warning and commissioning strategies, thereby improving the operation and maintenance efficiency and stability of the commissioning system and reducing power generation losses and operation and maintenance costs.

[0020] Specifically: First, the panoramic perception capability adapts to the equipment group commissioning needs: it constructs a multi-source perception network covering the entire scenario of wind, solar, hydro and storage, integrates multiple types of data and standardizes governance, solves the problem of fragmentation in traditional perception, provides comprehensive data support for cross-energy equipment group assessment, and adapts to multi-energy collaborative operation and equipment group management scenarios.

[0021] Second, the level of intelligence and assessment accuracy are improved: by combining multimodal large models and Fourier transform technology, the ability to identify weak fault characteristics is enhanced. The scores of each dimension are calculated using quantitative formulas to avoid reliance on expert experience. Combined with dynamic early warning strategies, the accuracy and lead time of early warning meet the needs of joint commissioning and maintenance.

[0022] Third, the unified health standard for equipment groups breaks through the limitations of cross-energy systems: a unified health index for cross-energy equipment groups is established. Through dynamic weight adjustment and normalization, the pain points of chaotic cross-energy equipment health judgment standards and the inability to make overall comparisons are solved. This enables the comparability of health status of different types of equipment such as wind turbines, hydropower equipment, and energy storage equipment, as well as the overall judgment of equipment groups. It fills the gap in the unified health standard for equipment groups in the scenario of wind, solar, hydro and storage joint commissioning, and has significant industry innovation.

[0023] Fourth, the equipment group has strong synergy with the joint commissioning: the early warning strategy is dynamically adapted to the joint commissioning conditions of wind, solar, hydro and storage, and the equipment group assessment results directly serve the optimization of joint commissioning and scheduling. The load distribution can be adjusted in a coordinated manner through the health status of the equipment group, reducing the impact of single equipment failure on the equipment group and joint commissioning system, improving energy consumption efficiency and system stability, and promoting the transformation of the station to unmanned operation and intelligent joint commissioning. Attached Figure Description

[0024] Figure 1 This is a flowchart of the present invention; Detailed Implementation

[0025] The following is in conjunction with the appendix Figure 1 The invention is further illustrated by the embodiments.

[0026] A health assessment method for equipment groups applicable to integrated wind, solar, hydro, and storage operation includes the following steps: S1. Collection and Management of Panoramic Perception Data: Construct a multi-source perception network covering the entire scenario of wind, solar, hydro, and energy storage, and collect panoramic perception data. The panoramic perception data includes the operation characteristic data, environmental adaptation characteristic data, and inspection image characteristic data of various equipment in wind power, photovoltaic, hydro power, and energy storage systems. Based on the data source characteristics of different equipment, the panoramic perception data is standardized and managed to generate a standardized panoramic perception dataset.

[0027] Specifically, in step S1, operational characteristic data is obtained by connecting to the SCADA system via the DL / T634.5104 protocol; environmental adaptation characteristic data is collected through the environmental sensing system; inspection image characteristic data is integrated through the API interface, with a data latency of ≤10 seconds; missing data is filled using linear interpolation, with a unified time granularity of 5 minutes average; and the collected data is labeled with device ID, energy type, collection time, and operating condition label. The environmental sensing system is an existing technology.

[0028] The data source characteristics of a device include data source protocol, format, timing, and quality characteristics.

[0029] Specifically, based on the data source protocols, formats, timing and quality characteristics of different devices, standardized governance of panoramic perception data is achieved through unified access, model mapping, cleaning and verification, standard encoding and metadata management.

[0030] S2, Constructing a Multimodal Large Model: Select an AI large model that supports input of time-series data, image data, and standardized panoramic perception datasets. Inject knowledge graphs of fault mechanisms of wind, solar, hydro, and storage equipment and industry operation and maintenance procedures. First, pre-train using general power datasets and publicly available wind, solar, hydro, and storage datasets. Then, inject dedicated data for wind, solar, hydro, and storage joint commissioning for a second training. Use LoRA technology to adjust the weights of operating features, environmental adaptation features, and inspection image features to generate a multimodal large model adapted to joint commissioning scenarios.

[0031] Specifically, in step S2, the knowledge graph includes causal relationship chains of faults in various equipment of wind power, photovoltaic, hydropower and energy storage systems; the industry operation and maintenance procedures include GB / T38335, NB / T11346 and NB / T11570 standards.

[0032] S3, Constructing a cross-energy equipment group health standard: Based on the multimodal large model in step S2, extract the time-domain features, frequency-domain features, and image features of each device. Combine the operating characteristics of different devices in the wind, solar, hydro, and storage system with the collaborative requirements of the equipment group to construct a cross-energy equipment group health standard. This enables the extension from single-device health assessment to overall equipment group health assessment, while solving the technical pain point that existing general methods cannot adapt to cross-energy equipment group collaborative scenarios.

[0033] The health standard for cross-energy equipment clusters includes a unified health index, operational characteristic score, environmental adaptability characteristic score, inspection image characteristic score, and equipment cluster health standard weighting coefficient.

[0034] The weighting coefficients for the health standards of the equipment group include weighting coefficients for operational characteristics, weighting coefficients for environmental adaptation characteristics, and weighting coefficients for inspection image characteristics.

[0035] The time-domain features include the mean, variance, and peak factor, while the frequency-domain features are the characteristic frequencies of the vibration signal obtained through Fourier transform.

[0036] S4, Dynamic Early Warning: Based on the health standards of cross-energy equipment groups, combined with the temporal reasoning capabilities of the LSTM model and the multimodal large model in step S2, predict the unified health index of individual equipment and equipment groups within a certain period of time in the future (such as within 24 hours, 48 ​​hours, 72 hours, etc.), and generate fault early warning information. Combining the equipment degradation characteristics and joint commissioning adaptation strategies of wind, solar, hydro and storage systems, dynamically optimize the fault early warning threshold and early warning classification strategy to achieve dynamic early warning.

[0037] Specifically, in steps S3 and S4, the formula for calculating the Unified Health Index (HI) is as follows: HI = A × Operational Feature Score + B × Environment Adaptation Feature Score + C × Inspection Image Feature Score; Where A is the weight coefficient of the running feature, B is the weight coefficient of the environment adaptation feature, and C is the weight coefficient of the inspection image feature, satisfying A+B+C=1 and A>0, B>0, C>0; A, B, and C are dynamically adjusted according to equipment type using the Analytic Hierarchy Process (AHP), and their final weights are determined after a consistency test (CR < 0.1).

[0038] Specifically, addressing the shortcomings of existing AHP feature weight coefficient applications, which are fixed and unable to adapt to differences across energy equipment groups, the feature weight coefficients are dynamically adjusted using the Analytic Hierarchy Process (AHP) based on equipment type (e.g., wind turbines, photovoltaic inverters, hydro turbine generators, energy storage converters, etc.). The adjustment is based on the weight ratio of the impact of equipment failures in various dimensions on the joint operation of wind, solar, hydro, and energy storage. The final feature weight value is determined by constructing a judgment matrix and performing a consistency test (CR < 0.1 is considered acceptable). For example, the operating feature weight coefficient for wind turbines is set at 0.5–0.6, and the environmental adaptability weight coefficient for hydropower equipment is set at 0.3–0.4. This accurately adapts to the core risk points of different equipment and the collaborative needs of equipment groups, solving the problem of unbalanced weights in health assessment across energy equipment.

[0039] Specifically, the operational feature score is obtained by comparing the time-domain and frequency-domain features extracted from the multimodal large model with the corresponding rated operating parameter thresholds of the equipment (determined according to the equipment's factory standards and industry standards) and then using normalization calculation. The formula for calculating the running feature score is as follows: Running feature score = (Actual feature parameter compliance rate) × 100; Actual feature parameter compliance rate = Number of feature parameters that meet the rated threshold range / Total number of feature parameters; The total characteristic parameters are the sum of features further increased based on the selected time-domain and frequency-domain features, differentiated according to equipment type. For example, wind turbines add speed and pitch angle features, hydropower equipment adds head and flow rate features, and energy storage equipment adds charge / discharge efficiency and SOC features.

[0040] Specifically, the environmental adaptability score is calculated based on the degree of matching between the collected actual environmental parameters (including temperature, humidity, wind speed, dust concentration, water level and water quality, collected differently according to energy type) and the equipment's tolerance threshold. The formula for calculating the environment adaptation feature score is as follows: Environmental adaptability feature score = Σ (adaptability coefficient of single environmental factor × factor weight) × 100; Single environmental factor adaptation coefficient = 1 - |actual environmental parameter - optimal adaptation parameter| / (tolerance upper limit - tolerance lower limit); When the calculated result of the single environmental factor fit coefficient is less than 0, the single environmental factor fit coefficient is forced to be 0.

[0041] Among them, the optimal adaptation parameters, the upper limit of tolerance, and the lower limit of tolerance are determined according to the equipment technical manual and the NB / T11346 standard; the factor weights are determined with the help of AHP based on the degree of influence of the environment on equipment failure and the stability of the joint commissioning system. For example, in the wind power scenario, the wind speed factor weight is higher than the humidity factor weight, and in the hydropower scenario, the water level factor weight is higher than the temperature factor weight.

[0042] Specifically, the inspection image score is calculated based on the fault identification results of the inspection image using a multimodal large model. The formula for calculating the feature score of the inspection image is as follows: Inspection image feature score = 100 - Σ (fault level weight × number of faults); The fault levels are classified according to a unified standard (i.e., based on the proportion, characteristics and scope of the defect), and are applicable to various wind, solar, hydro and storage equipment. They are divided into minor faults, general faults and serious faults, with weights of 0.1 for minor faults, 0.3 for general faults and 0.6 for serious faults.

[0043] Minor faults: The defect accounts for less than 5% of the image, with blurred edges, small grayscale differences, no obvious structural abnormalities, and only manifests as minor surface flaws. It does not involve the core functional areas of the equipment and does not affect the joint commissioning and operation. General faults: Defects account for 5%-30% of the image, with medium contrast. Defects can be clearly distinguished from normal areas. They involve core components of the equipment (such as connectors and sealing surfaces) and may affect the local performance of the equipment. The compatibility of the joint debugging needs to be monitored. Serious fault: The defect accounts for more than 30% of the image, with sharp edges and huge differences in grayscale / color. Core functional areas (such as main components, key structures, and safety protection areas) show breakage, missing parts, or large-area anomalies, which can directly determine that the core components have failed. This will seriously affect the stability of wind, solar, hydro, and storage joint commissioning and requires immediate handling.

[0044] Specifically, the early warning classification strategy in step S4 is as follows: When HI_norm≤30, it is a Level 1 warning, which is pushed to the operation and maintenance terminal and triggers an audible and visual alarm, and the linkage and debugging system adjusts the equipment load. When 30 < HI_norm ≤ 60, it is a level 2 warning, and a work order to be processed is generated; When 60 < HI_norm ≤ 80, it is a Level 3 warning and should be included in the periodic inspection plan; No warning is given when 80 < HI_norm.

[0045] Specifically, equipment degradation characteristics mainly include cumulative operating time and annual failure frequency.

[0046] Specifically, the integration and adaptation strategy is shown in the table below:

[0047] S5, Model Validation and Iterative Optimization: Through historical data backtracking and field verification, verify the effectiveness of fault early warning and the rationality of cross-energy equipment group health standards. Regularly update the multimodal large model with new data. Based on false alarm analysis, update the equipment degradation characteristics in step S4 and the cross-energy equipment group health standards in step S3.

[0048] Specifically, it involves updating the weighting coefficients of the equipment group health standard in the cross-energy equipment group health standard, including the weighting coefficients of operating characteristics, environmental adaptation characteristics, and inspection image characteristics.

[0049] Specifically, the Unified Health Index (HI) eliminates assessment biases among different energy devices through min-max normalization; it enables ranking and overall assessment of the health status of energy device groups, providing a global basis for joint commissioning and operation scheduling.

[0050] The normalization formula is: HI_norm=(HI-HI_min) / (HI_max-HI_min)×100; in, HI_min represents the lowest historical health index of the corresponding energy type of equipment; HI_max is the highest historical health index of the corresponding energy type of equipment; HI_norm is a normalized standard health index used to rank the health status of equipment across wind, solar, hydro, and storage scenarios and determine maintenance priorities.

[0051] Set verification indicators: early warning accuracy ≥90%, recall rate ≥95%, and lead time ≥24 hours; retrain the multimodal large model in step S2 every quarter using newly added joint debugging data and fault data, update the equipment degradation characteristics in step S4 and the cross-energy equipment group health standards in step S3 based on false alarm analysis, and form a closed-loop system of "perception-judgment-early warning-joint debugging-optimization" to continuously improve the accuracy of equipment group health judgment and joint debugging adaptability.

[0052] The equipment group health assessment system applicable to wind, solar, hydro, and storage integrated commissioning adopts equipment group health assessment methods applicable to wind, solar, hydro, and storage integrated commissioning, including a panoramic perception data acquisition and management module, a multimodal large model construction module, a cross-energy equipment group health standard construction module, a dynamic early warning module, and a model verification and iterative optimization module.

[0053] The panoramic perception data acquisition and management module is configured to build a multi-source perception network covering the entire scenario of wind, solar, hydro and energy storage, and to collect panoramic perception data. The panoramic perception data includes the operation characteristic data, environmental adaptation characteristic data and inspection image characteristic data of each device in the wind power, photovoltaic, hydro power and energy storage system. The panoramic perception data is standardized and managed according to the data source characteristics of different devices to generate a standardized panoramic perception dataset. The multimodal large model construction module is configured to select AI large models that support input of time series data, image data and standardized panoramic perception datasets, and inject knowledge graphs of fault mechanisms of wind, solar and hydro storage equipment and industry operation and maintenance procedures. It first uses general power datasets and public datasets of wind, solar and hydro storage for pre-training, and then injects dedicated data for wind, solar and hydro storage joint commissioning for a second training. It uses LoRA technology to adjust the weights of operating features, environmental adaptation features and inspection image features to generate a multimodal large model adapted to the joint commissioning scenario. The cross-energy equipment group health standard construction module is configured based on the multimodal large model in step S2, extracts the time domain features, frequency domain features and image features of each device, and combines the operating characteristics of different devices in the wind, solar, water and storage system with the collaborative requirements of the equipment group to construct the cross-energy equipment group health standard. The dynamic early warning module is configured to predict the unified health index of a single device and a group of devices within a certain period of time based on the health standard of cross-energy equipment groups, combined with the temporal reasoning capabilities of the LSTM model and the multimodal large model in step S2, and generate fault early warning information. Combining the equipment degradation characteristics and joint commissioning adaptation strategies of wind, solar, hydro and storage, it dynamically optimizes the fault early warning threshold and early warning classification strategy to achieve dynamic early warning. The model verification and iterative optimization module is configured to verify the effectiveness of fault warning and the rationality of cross-energy equipment group health standards through historical data backtracking and field verification, periodically update the multimodal large model with new data, and update the equipment degradation characteristics in step S4 and the cross-energy equipment group health standards in step S3 based on false alarm analysis.

[0054] Example

[0055] This embodiment takes a wind-solar-hydro-storage integrated power station (including one 20MW wind power plant, one 10MW photovoltaic power plant, one 20MW hydropower plant, and one 5MW energy storage power station) as an example. The specific implementation steps are as follows: S1, Panoramic Perception Data Acquisition and Governance Multi-source sensing devices are deployed: vibration sensors and speed sensors are deployed on all wind turbines in the 20MW wind farm; string current / voltage sensors are deployed on the modules in the 10MW photovoltaic power station; water level sensors and water flow velocity sensors are deployed in the 20MW hydropower station; and charge / discharge status sensors are deployed in the 5MW energy storage power station. Operational characteristic data of each device is collected through these sensors. Environmental data collection covers wind speed, temperature, humidity, dust concentration (wind / photovoltaic), water level, and water quality (hydropower station) for each site. Inspection image feature data is collected through drone inspections and handheld devices, covering the appearance, core components, and sealing surfaces of equipment at each site. The system connects to the power station's SCADA system via the DL / T634.5104 protocol, and the API interface integrates the inspection systems of each site, with data latency controlled within 8 seconds. Missing data is filled using linear interpolation, with a unified time granularity of 5-minute average values. Equipment ID, site type (wind / solar / hydro / storage), collection time, and operating condition label are provided (e.g., "wind speed 8m / s" for wind turbines, "water level 15m" for hydropower stations).

[0056] S2, Multimodal Large Model Construction

[0057] The ViT-L / 14 model, which supports both time-series and image inputs, was selected. Fault mechanism knowledge graphs for equipment at various wind, solar, hydro, and storage power plants (including fault chains for wind turbine gearboxes, hydro turbine cavitation, energy storage converters, and photovoltaic module hot spot fault chains) and industry operation and maintenance regulations such as GB / T38335 were injected. The model was pre-trained for 100 rounds using the IEEE PES power dataset and publicly available fault datasets for wind, solar, hydro, and storage. A second training round was then conducted using nearly three years of proprietary data from each station of the integrated commissioning power station (including data from 500+ typical faults and cross-station collaborative operation). LoRA technology was used to adjust the weights of operational features, environmental adaptation features, and inspection image features, generating a multimodal large-scale model adapted to the integrated commissioning scenario. After fine-tuning, the model's feature extraction accuracy reached 92%.

[0058] S3, Construction of Health Standards Across Energy Equipment Clusters

[0059] The weights of each device were determined using AHP: wind turbine (A1=0.55, B1=0.25, C1=0.2), photovoltaic inverter (A2=0.5, B2=0.3, C2=0.2), hydro turbine generator set (A3=0.45, B3=0.35, C3=0.2), and energy storage converter (A4=0.6, B4=0.2, C4=0.2). The consistency check of the judgment matrix CR=0.08<0.1, indicating that the weights are valid.

[0060] Taking a single wind turbine in a wind farm as an example, the operating characteristics include 8 items such as vibration frequency, speed, and pitch angle. If 7 items meet the rated threshold, the compliance rate = 7 / 8 = 0.875, and the operating characteristic score = 87.5. The environmental factors include wind speed, temperature, and humidity, with weights of 0.6, 0.3, and 0.1, respectively. The actual wind speed is 6 m / s (optimal 5 m / s, tolerance upper limit 12 m / s, lower limit 0 m / s). The single environmental factor adaptation coefficient = 1 - |6 - 5| / (12 - 0) = 0.917. The temperature and humidity adaptation coefficients are 0.95 and 0.98, respectively. The environmental adaptation characteristic score = (0.917 × 0.6 + 0.95 × 0.3 + 0. 98×0.1)×100≈93.3; the inspection image shows no faults, with a score of 100; then the unified health index HI of the wind turbine is 0.55×87.5+0.25×93.3+0.2×100≈91.45. Combining the historical unified health index HI range (20~98) of the wind turbine group in the wind farm, we calculate HI_norm=(91.45-20) / (98-20)×100≈91.6. According to the health standard of this invention, the wind turbine is judged to be in good health and there is no warning. At the same time, the overall health status of the entire equipment group of the entire station is formed by summarizing the unified health index HI values ​​of individual equipment in each station, which provides a basis for cross-station joint commissioning and scheduling.

[0061] S4, Dynamic Warning

[0062] Based on deep learning prediction algorithms, the system can predict and analyze the health index of individual equipment and equipment groups, and automatically generate fault warning information. Combining the equipment degradation characteristics and joint commissioning adaptation strategies in the wind, solar, water and storage joint commissioning scenario, the system can dynamically optimize the fault warning threshold and warning classification strategy to achieve dynamic warning.

[0063] Based on the LSTM model, the predicted Unified Health Index (HI) value for the wind turbine over the next 24 hours is 88.2, with HI_norm ≈ 87.4, indicating no warning. The inspection strategy is adjusted according to equipment degradation characteristics and joint commissioning adaptation strategies. Simultaneously, the predicted Unified Health Index (HI) value for the entire wind turbine group is 89.1, indicating stable equipment status at each site. If the predicted Unified Health Index (HI) value for a certain energy storage converter in the energy storage power station is 28, with HI_norm ≈ 26.5, a Level 1 warning will be triggered. Based on equipment degradation characteristics and joint commissioning adaptation strategies, the joint commissioning system will immediately reduce the charging and discharging load of this energy storage device, simultaneously assess its impact on the energy storage power station equipment group and the overall joint commissioning operation of the entire site, and push warning information to the operation and maintenance terminal.

[0064] S5, Model Validation and Iterative Optimization

[0065] Through retrospective verification of historical fault data from various stations of the joint commissioning power station, the early warning accuracy rate was 93%, the recall rate was 96%, and the average lead time was 26 hours, meeting the verification indicators. Every quarter, new fault data from each station (such as hot spot faults of photovoltaic modules, wear faults of hydropower turbines, and abnormal noise faults of wind turbine gearboxes) are included to retrain the model, adjust the environmental adaptation factor weight of the converter of the energy storage power station, optimize the early warning threshold and health standard parameters, and continuously improve the health judgment and early warning effect of cross-station equipment groups.

[0066] The specific embodiments described are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. All equivalent changes made to the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A health assessment method for equipment groups applicable to integrated wind, solar, hydro, and storage commissioning, characterized in that, Including the following steps: S1, Panoramic Perception Data Acquisition and Governance: Construct a multi-source perception network covering the entire scenario of wind, solar, hydro, and energy storage, collect panoramic perception data, including the operation characteristic data, environmental adaptation characteristic data, and inspection image characteristic data of various equipment in wind power, photovoltaic, hydro power, and energy storage systems. Based on the data source characteristics of different equipment, the panoramic perception data is standardized and governed to generate a standardized panoramic perception dataset. S2, Multimodal Large Model Construction: Select an AI large model that supports input of time series data, image data and standardized panoramic perception datasets, inject knowledge graphs of fault mechanisms of wind, solar, hydro and storage equipment and industry operation and maintenance procedures, first use general power datasets and public datasets of wind, solar, hydro and storage for pre-training, then inject dedicated data for wind, solar, hydro and storage joint commissioning for a second training, and use LoRA technology to adjust the weights of operation features, environmental adaptation features and inspection image features to generate a multimodal large model adapted to the joint commissioning scenario; S3, Construction of health standards across energy equipment groups: Based on the multimodal large model in step S2, extract the time domain features, frequency domain features and image features of each device, and combine the operating characteristics of different devices in the wind, solar, hydro and storage system with the collaborative requirements of the equipment group to construct health standards across energy equipment groups; S4, Dynamic Early Warning: Based on the health standards of cross-energy equipment groups, combined with the time-series reasoning capabilities of the LSTM model and the multimodal large model in step S2, the unified health index of single equipment and equipment groups is predicted within a certain period of time in the future, and fault early warning information is generated. Combined with the equipment degradation characteristics and joint commissioning adaptation strategies of wind, solar, hydro and storage, the fault early warning threshold and early warning classification strategy are dynamically optimized to achieve dynamic early warning. S5, Model Validation and Iterative Optimization: Through historical data backtracking and field verification, verify the effectiveness of fault early warning and the rationality of cross-energy equipment group health standards. Regularly update the multimodal large model with new data. Based on false alarm analysis, update the equipment degradation characteristics in step S4 and the cross-energy equipment group health standards in step S3.

2. The health assessment method for equipment groups applicable to integrated wind, solar, hydro, and storage operation as described in claim 1, characterized in that, In step S1, the system connects to the SCADA system via the DL / T634.5104 protocol to obtain operational characteristic data; it collects environmental adaptation characteristic data through the environmental sensing system; it integrates inspection image characteristic data through the API interface, with a data delay of ≤10 seconds; it fills in missing data using linear interpolation, with a unified time granularity of 5 minutes average; and it labels the collected data with device ID, energy type, collection time, and operating condition tag.

3. The health assessment method for equipment groups applicable to integrated wind, solar, hydro, and storage operation as described in claim 1, characterized in that, In step S2, the knowledge graph includes causal relationship chains of faults in various equipment of wind power, photovoltaic, hydropower and energy storage systems; the industry operation and maintenance procedures include GB / T38335, NB / T11346 and NB / T11570 standards.

4. The health assessment method for equipment groups applicable to integrated wind, solar, hydro, and storage operation as described in claim 1, characterized in that, In steps S3 and S4, the formula for calculating the Unified Health Index (HI) is as follows: HI = A × Operational Feature Score + B × Environment Adaptation Feature Score + C × Inspection Image Feature Score; Where A is the weight coefficient of the running feature, B is the weight coefficient of the environment adaptation feature, and C is the weight coefficient of the inspection image feature, satisfying A+B+C=1 and A>0, B>0, C>0; A, B, and C are dynamically adjusted according to equipment type using the analytic hierarchy process (AHP), and their final weights are determined after a consistency check.

5. The health assessment method for equipment groups applicable to integrated wind, solar, hydro, and storage operation as described in claim 4, characterized in that, The formula for calculating the running feature score is as follows: Running feature score = (Actual feature parameter compliance rate) × 100; Actual feature parameter compliance rate = Number of feature parameters that meet the rated threshold range / Total number of feature parameters.

6. The health assessment method for equipment groups applicable to integrated wind, solar, hydro, and storage operation as described in claim 4, characterized in that, The formula for calculating the environment adaptation feature score is as follows: Environmental adaptability feature score = Σ (adaptability coefficient of single environmental factor × factor weight) × 100; Single environmental factor adaptation coefficient = 1 - |actual environmental parameter - optimal adaptation parameter| / (tolerance upper limit - tolerance lower limit); When the calculated result of the single environmental factor fit coefficient is less than 0, the single environmental factor fit coefficient is forced to be 0.

7. The health assessment method for equipment groups applicable to integrated wind, solar, hydro, and storage operation as described in claim 4, characterized in that, The formula for calculating the feature score of the inspection image is as follows: Inspection image feature score = 100 - Σ (fault level weight × number of faults); The fault levels are divided into minor faults, general faults, and serious faults, with weights of 0.1 for minor faults, 0.3 for general faults, and 0.6 for serious faults.

8. The health assessment method for equipment groups applicable to integrated wind, solar, hydro, and storage operation as described in claim 1, characterized in that, It also includes the step of: the unified health index HI is normalized by min-max; The normalization formula is: HI_norm=(HI-HI_min) / (HI_max-HI_min)×100; in, HI_min represents the lowest historical health index of the corresponding energy type of equipment; HI_max is the highest historical health index of the corresponding energy type of equipment; HI_norm is the normalized standard health index.

9. The health assessment method for equipment groups applicable to integrated wind, solar, hydro, and storage operation as described in claim 8, characterized in that, The early warning classification strategy in step S4 is as follows: When HI_norm≤30, it is a Level 1 warning, which is pushed to the operation and maintenance terminal and triggers an audible and visual alarm, and the linkage and debugging system adjusts the equipment load. When 30 < HI_norm ≤ 60, it is a level 2 warning, and a work order to be processed is generated; When 60 < HI_norm ≤ 80, it is a Level 3 warning and should be included in the periodic inspection plan; No warning is given when 80 < HI_norm.

10. A health assessment system for equipment groups suitable for integrated wind, solar, hydro, and storage operation, characterized in that: The method for assessing the health of equipment clusters applicable to the joint commissioning of wind, solar, hydro, and storage as described in any one of claims 1-9 includes a panoramic perception data acquisition and management module, a multimodal large model construction module, a cross-energy equipment cluster health standard construction module, a dynamic early warning module, and a model verification and iterative optimization module. The panoramic perception data acquisition and management module is configured to build a multi-source perception network covering the entire scenario of wind, solar, hydro and energy storage, and to collect panoramic perception data. The panoramic perception data includes the operation characteristic data, environmental adaptation characteristic data and inspection image characteristic data of each device in the wind power, photovoltaic, hydro power and energy storage system. The panoramic perception data is standardized and managed according to the data source characteristics of different devices to generate a standardized panoramic perception dataset. The multimodal large model construction module is configured to select AI large models that support input of time series data, image data and standardized panoramic perception datasets, and inject knowledge graphs of fault mechanisms of wind, solar and hydro storage equipment and industry operation and maintenance procedures. It first uses general power datasets and public datasets of wind, solar and hydro storage for pre-training, and then injects dedicated data for wind, solar and hydro storage joint commissioning for a second training. It uses LoRA technology to adjust the weights of operating features, environmental adaptation features and inspection image features to generate a multimodal large model adapted to the joint commissioning scenario. The cross-energy equipment group health standard construction module is configured based on the multimodal large model in step S2, extracts the time domain features, frequency domain features and image features of each device, and combines the operating characteristics of different devices in the wind, solar, water and storage system with the collaborative requirements of the equipment group to construct the cross-energy equipment group health standard. The dynamic early warning module is configured to predict the unified health index of a single device and a group of devices within a certain period of time based on the health standard of cross-energy equipment groups, combined with the temporal reasoning capabilities of the LSTM model and the multimodal large model in step S2, and generate fault early warning information. Combining the equipment degradation characteristics and joint commissioning adaptation strategies of wind, solar, hydro and storage, it dynamically optimizes the fault early warning threshold and early warning classification strategy to achieve dynamic early warning. The model verification and iterative optimization module is configured to verify the effectiveness of fault warning and the rationality of cross-energy equipment group health standards through historical data backtracking and field verification, periodically update the multimodal large model with new data, and update the equipment degradation characteristics in step S4 and the cross-energy equipment group health standards in step S3 based on false alarm analysis.

Images