Wind turbine generator performance evaluation method and system based on multi-source data
By collecting and standardizing data from multiple dimensions, and combining evaluation models and feedback mechanisms, the problems of incomplete data and rigid models in wind turbine performance evaluation have been solved, enabling dynamic and accurate performance evaluation and optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-11
- Publication Date
- 2026-04-10
AI Technical Summary
Existing wind turbine performance evaluation methods rely on limited data, lack cross-scenario correlation and unstructured data, resulting in one-sided evaluation results and rigid models that cannot be dynamically adjusted, leading to a decline in evaluation accuracy.
By collecting multi-dimensional feature data, performing standardization and correlation mapping, a standardized evaluation dataset is generated. The evaluation model is then used to output performance evaluation results, and the model is updated based on feedback data to achieve continuous optimization.
It improves the accuracy and robustness of performance evaluation, provides health scores and component risk identification, supports real-time operational adjustments, reduces operation and maintenance costs, and increases power generation revenue.
Smart Images
Figure CN121834278A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of wind power generation, in particular to a wind turbine performance evaluation method and system based on multi-source data. The method integrates multiple data sources of wind turbines to achieve comprehensive and dynamic performance evaluation and operation optimization, suitable for scenarios such as wind farm operation and management, grid dispatching coordination, etc. BACKGROUND
[0002] With the rapid development of wind power generation, the reliability and operation efficiency of wind turbines have become the focus of attention. Traditional performance evaluation methods usually rely on limited operation data, which is difficult to fully reflect the state of the unit, especially in complex operating environments. The existing technology has the following shortcomings: only using unit operation data, lacking cross-scene correlation data and unstructured data, resulting in one-sided evaluation results; the evaluation model is usually static and cannot be dynamically adjusted according to real-time scenarios, and does not consider individual differences and life cycle changes of the unit; the evaluation results are disconnected from operation adjustment, lacking model updating based on feedback, resulting in a decline in evaluation accuracy over time.
[0003] Therefore, there is an urgent need in the art for a wind turbine performance evaluation method and system that can integrate multi-source data, dynamically adapt to changes in scenarios, and have self-learning capabilities. SUMMARY
[0004] The purpose of the present application is to provide a wind turbine performance evaluation method and system based on multi-source data to solve the problems of incomplete evaluation and rigid models in the prior art. The present application collects multi-dimensional feature data, performs standardization and correlation mapping processing, uses an evaluation model to output performance evaluation results, and generates operation adjustment recommendations based on the results, while updating the model through feedback data to achieve continuous optimization.
[0005] In some embodiments of the present application, a wind turbine performance evaluation method based on multi-source data is provided, comprising:
[0006] Collecting multi-dimensional feature data of the wind turbine to generate an original data set;
[0007] Standardizing and correlating the mapping processing of the original data set to generate a standardized evaluation data set;
[0008] Generating performance evaluation results of the standardized evaluation data set based on an evaluation model;
[0009] Generating operation adjustment recommendations according to the performance evaluation results and generating control instructions based on the operation adjustment recommendations, and using feedback data after execution to update the evaluation model.
[0010] In some embodiments of the present application, when collecting multi-dimensional feature data of the wind turbine to generate an original data set, it comprises:
[0011] The multi-dimensional feature data includes unit operation-related data, cross-scenario related data, unstructured data, and full lifecycle data;
[0012] Real-time power, rotor speed, vibration parameters of key components, temperature parameters, and blade pitch angle are obtained from sensors and the unit control system to generate relevant data on unit operation.
[0013] Cross-scenario related data are generated by obtaining power grid peak-shaving commands, frequency regulation commands, and power load change data through the power grid interface, as well as wind speed, ambient temperature, air density, and extreme weather warning data obtained through meteorological monitoring equipment.
[0014] Unstructured data is generated from manual maintenance records recorded by mobile terminals and blade surface images and temperature field images of key components acquired by image acquisition devices. The manual maintenance records include equipment anomaly descriptions, maintenance operations, and inspection results.
[0015] The unit design parameters, rated performance indicators and life parameters obtained through the archive system, as well as historical fault records, component replacement records and performance change trend data obtained through the operation and maintenance management system, are used to generate full life cycle data.
[0016] The unit operation-related data, cross-scenario associated data, unstructured data, and full lifecycle data are associated and stored according to the unit's unique identifier and timestamp to form the original dataset.
[0017] In some embodiments of this application, when generating a standardized evaluation dataset by performing standardization and association mapping processing on the original dataset, the following steps are included:
[0018] The original dataset is cleaned and standardized to generate standardized data.
[0019] Based on the standardized data, cross-scenario correlation analysis is performed to generate data correlation mapping relationships;
[0020] Using the aforementioned data association mapping relationship, unstructured data is parsed and quantified to generate structured state parameters;
[0021] The structured state parameters are corrected by combining full lifecycle data to generate performance correction coefficients;
[0022] The standardized data, structured state parameters, and performance correction coefficients are integrated to generate a standardized evaluation dataset.
[0023] In some embodiments of this application, when outputting performance evaluation results using an evaluation model based on the standardized evaluation dataset, the following are included:
[0024] based on the cross-scene correlation data and the unit operation related data in the standardized evaluation data set, scene judgment is performed to obtain a scene classification result;
[0025] a pre-stored model library is dynamically selected to select an evaluation model corresponding to the scene classification result;
[0026] the selected evaluation model is called to process the standardized evaluation data set to obtain a unit health score and a component risk identification result;
[0027] based on the component risk identification result, a unit control parameter adjustment instruction is generated in real time and is sent to a unit actuator;
[0028] when in a cross-scene collaborative evaluation mode, encrypted model parameters from a federated learning aggregation node are received, the encrypted model parameters are adapted in combination with individual characteristics of the unit, and then the local evaluation model is updated;
[0029] based on the model output result, real-time feedback data, and a performance degradation physical law of a key component, a health degree benchmark parameter of the evaluation model is adaptively corrected.
[0030] In some embodiments of the present application, when the evaluation model corresponding to the scene classification result is dynamically selected from the pre-stored model library, it includes:
[0031] based on the scene classification result, a core performance dimension to be evaluated under the current scene is determined;
[0032] from the pre-stored model library, an evaluation model optimized for the core performance dimension is selected, and the evaluation model includes a first type of evaluation model and a second type of evaluation model;
[0033] The first type of evaluation model is used to generate a control parameter adjustment instruction for maintaining stable power generation performance, and the second type of evaluation model is used to generate a protective control parameter adjustment instruction for ensuring structural load safety.
[0034] In some embodiments of the present application, when in the cross-scene collaborative evaluation mode, the encrypted model parameters from the federated learning aggregation node are received, the encrypted model parameters are adapted in combination with individual characteristics of the unit, and then the local evaluation model is updated, which includes:
[0035] adaptability evaluation related to individual characteristics of the unit is performed on the received encrypted model parameters;
[0036] The adaptability evaluation is used to dynamically adjust a weight coefficient of the encrypted model parameters according to historical operation and maintenance data of the unit and a component aging coefficient;
[0037] When the encrypted model parameters pass the adaptability evaluation, they are used to update the local evaluation model.
[0038] In some embodiments of the present application, when the health degree benchmark parameter of the evaluation model is adaptively corrected based on the model output result, real-time feedback data, and the performance degradation physical law of the key component, the correction includes:
[0039] The real-time feedback data is compared with the expected state data predicted based on the performance degradation physical law to generate a residual sequence;
[0040] The evaluation model is incrementally updated based on the residual sequence, historical performance degradation trend, and online learning mechanism;
[0041] The correction amount of the health degree benchmark parameter is associated with the material fatigue characteristics of the key component.
[0042] In some embodiments of the present application, when the performance evaluation result of the standardized evaluation data set generated based on the evaluation model is generated, the generation includes:
[0043] A model update strategy is generated, which is used to adaptively adjust the strength of model update according to the dynamic characteristics of the residual sequence and the historical performance degradation trend;
[0044] In the model update process, a retrospective constraint on the historical knowledge distribution is synchronously introduced;
[0045] The evaluation model parameters are cooperatively optimized based on the model update strategy and the retrospective constraint.
[0046] In some embodiments of the present application, according to the performance evaluation result, a running adjustment suggestion is generated and a control instruction is generated based on the running adjustment suggestion, and the feedback data after execution is used to update the evaluation model, including:
[0047] According to the unit health degree score and the component risk identification result output by the evaluation model, a performance evaluation conclusion including the overall health state, component abnormal type and level, and potential operation risk is generated;
[0048] Based on the performance evaluation conclusion and the preset risk control mapping rule, a unit operation adjustment suggestion is generated;
[0049] The unit operation adjustment suggestion is converted into a control instruction and is issued to a unit execution mechanism to realize real-time adjustment of unit operation parameters;
[0050] The running state data after the unit executes the control instruction is collected as feedback data, and the feedback data is fed back to the evaluation model to correct the model parameters.
[0051] In some embodiments of the present application, the wind turbine performance evaluation system based on multi-source data comprises:
[0052] A data acquisition module is configured to acquire multi-dimensional feature data of the wind turbine and generate an original data set;
[0053] A data processing module is configured to perform standardization and correlation mapping processing on the original data set and generate a standardized evaluation data set;
[0054] A model evaluation module is configured to call an evaluation model based on the standardized evaluation data set and dynamically adjust the evaluation logic according to a real-time operation scene of the unit;
[0055] A decision feedback module is configured to generate an operation adjustment suggestion according to the performance evaluation result, generate a control instruction based on the operation adjustment suggestion, and use the executed feedback data to update the evaluation model.
[0056] Compared with the prior art, the wind turbine performance evaluation method based on multi-source data in the preferred embodiments of the present application has the following advantages:
[0057] The present application fundamentally solves the problem of incomplete state perception caused by single data source in the prior art by systematically collecting and deeply integrating unit operation related data, cross-scene correlation data, unstructured data and full life cycle data, and constructing a multi-dimensional original data set. Through standardized data cleaning, correlation mapping and analysis and quantization of unstructured data, a standardized evaluation data set is generated, laying a reliable data foundation for accurate model evaluation. The method can dynamically select the optimal evaluation model according to the real-time scene, and use the federated learning and the self-adaptive correction mechanism based on the physical law to realize the continuous evolution and adaptation of the evaluation model under the premise of protecting data privacy, thereby greatly improving the accuracy, robustness and forward-looking warning ability of the performance evaluation result.
[0058] At the same time, the present application can output a comprehensive evaluation conclusion containing health score, component risk identification and specific control instruction, and automatically convert the operation adjustment suggestion into executable control instruction and issue it to the unit execution mechanism, forming a complete evaluation-decision-execution-feedback-optimization closed loop control circuit. In addition, the evaluation conclusion and health status information are synchronized to the power grid dispatching system, enhancing the controllability and schedulability of the wind turbine as a distributed power source, realizing the collaborative optimization of the station and the power grid, and ultimately achieving the comprehensive economic benefits of reducing the full life cycle operation and maintenance cost and improving the power generation income. BRIEF DESCRIPTION OF DRAWINGS
[0059] Figure 1 is a flowchart of the wind turbine performance evaluation method based on multi-source data in the preferred embodiments of the present application. DETAILED DESCRIPTION
[0060] The specific embodiments of the present application will be further described in details below with reference to the drawings and examples. The following examples are used to illustrate the present application, but not to limit the scope of the present application.
[0061] In the description of the present application, it needs to be understood that the terms "center", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.
[0062] The terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.
[0063] In the description of the present application, it needs to be explained that, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connection" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium, or internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.
[0064] In some embodiments of the present application, a wind turbine performance evaluation method based on multi-source data is provided, comprising:
[0065] Collecting multi-dimensional feature data of the wind turbine to generate an original data set;
[0066] Standardizing and correlating the original data set to generate a standardized evaluation data set;
[0067] Generating performance evaluation results of the standardized evaluation data set based on the evaluation model;
[0068] According to the performance evaluation results, generating operation adjustment suggestions and generating control instructions based on the operation adjustment suggestions, and using the feedback data after execution to update the evaluation model.
[0069] In this embodiment, the multi-dimensional feature data refers to various data reflecting the state of the wind turbine from different sources and angles, including unit operation related data, cross-scenario correlation data, unstructured data, and full life cycle data. These data are obtained through various collection devices and stored in association with the unit unique identifier and timestamp to form an original data set. This multi-dimensional integration ensures the comprehensiveness and traceability of the data.
[0070] In some embodiments of the present application, when generating an original data set according to the multi-dimensional feature data of the wind turbine, the following steps are included:
[0071] The multi-dimensional feature data includes unit operation related data, cross-scenario correlation data, unstructured data, and full life cycle data.
[0072] Real-time power, rotor speed, key component vibration parameters, temperature parameters, and pitch angle obtained through sensors and unit control systems generate unit operation related data.
[0073] Power grid peak shaving instructions, frequency modulation instructions, and power load change data obtained through grid interfaces, as well as wind farm wind speed, environmental temperature, air density, and extreme weather warning data obtained through weather monitoring devices generate cross-scenario correlation data.
[0074] Artificial operation records recorded through mobile terminals, as well as blade surface images and key component temperature field images obtained through image collection devices generate unstructured data, wherein the artificial operation records include device anomaly description, maintenance operation, and inspection results.
[0075] Unit design parameters, rated performance indicators, and life parameters obtained through archive systems, as well as historical fault records, component replacement records, and performance change trend data obtained through operation and maintenance management systems generate full life cycle data.
[0076] The unit operation related data, cross-scenario correlation data, unstructured data, and full life cycle data are stored in association with the unit unique identifier and timestamp to form an original data set.
[0077] In this embodiment, the unit operation related data is real-time data obtained through sensors and unit control systems, such as real-time power, rotor speed, key component vibration parameters, temperature parameters, and pitch angle. These data directly reflect the real-time operation state of the unit.
[0078] In this embodiment, the cross-scenario correlation data is power grid peak shaving instructions, frequency modulation instructions, and power load change data obtained through grid interfaces, as well as wind farm wind speed, environmental temperature, air density, and extreme weather warning data obtained through weather monitoring devices. These data are used to associate the influence of external scenario changes on unit performance.
[0079] In this embodiment, the unstructured data includes manual operation and maintenance records recorded by a mobile terminal, and blade surface images and key component temperature field images obtained by an image acquisition device. These data need to be subsequently analyzed and quantified to extract structured information.
[0080] In this embodiment, the full life cycle data includes unit design parameters, rated performance indicators and life parameters obtained by an archive system, and historical failure records, component replacement records and performance change trend data obtained by an operation and maintenance management system. These data are used to evaluate the unit health status from a long-term perspective.
[0081] In some embodiments of the present application, when generating a standardized evaluation data set according to the standardization and correlation mapping processing of the original data set, the following steps are included:
[0082] Cleaning and standardizing the original data set to generate standardized data;
[0083] Based on the standardized data, cross-scene correlation analysis is performed to generate a data correlation mapping relationship;
[0084] Using the data correlation mapping relationship, the unstructured data is analyzed and quantified to generate structured state parameters;
[0085] Combining the full life cycle data, the structured state parameters are corrected to generate performance correction coefficients;
[0086] Integrating the standardized data, structured state parameters and performance correction coefficients to generate a standardized evaluation data set.
[0087] In this embodiment, the standardization and correlation mapping processing refers to cleaning, format unification, correlation analysis and quantitative conversion of the original data to eliminate data heterogeneity and establish the internal relationship between the data. The standardized evaluation data set is a structured data set after processing, which is suitable for model input, including standardized data, structured state parameters and performance correction coefficients.
[0088] In this embodiment, the cleaning and standardization processing refers to denoising, missing value filling and format unification of the original data set to generate standardized data.
[0089] In this embodiment, the cross-scene correlation analysis is based on the standardized data to analyze the correlation between different data sources and generate a data correlation mapping relationship.
[0090] In this embodiment, the unstructured data analysis and quantification refers to using natural language processing technology to analyze operation and maintenance records and extract key events, using image processing technology to analyze blade images and quantify the surface defect degree, thereby generating structured state parameters.
[0091] In the embodiment, the full life cycle data correction refers to correcting the structured state parameters to generate performance correction coefficients in combination with the full life cycle data.
[0092] In the embodiment, the integration of the standardized evaluation dataset is to integrate the standardized data, the structured state parameters and the performance correction coefficients into a unified dataset for model evaluation.
[0093] In some embodiments of the present application, when the evaluation model outputs the performance evaluation result based on the standardized evaluation dataset, the following steps are included:
[0094] Based on the cross-scenario correlation data and the unit operation related data in the standardized evaluation dataset, scene judgment is performed to obtain a scene classification result;
[0095] An evaluation model corresponding to the scene classification result is dynamically selected from a preset model library;
[0096] The selected evaluation model is called to process the standardized evaluation dataset to obtain a unit health score and a component risk identification result;
[0097] Based on the component risk identification result, a unit control parameter adjustment instruction is generated in real time and is issued to a unit actuator;
[0098] When in the cross-scene collaborative evaluation mode, encrypted model parameters from a federated learning aggregation node are received, the encrypted model parameters are adapted in combination with individual characteristics of the unit, and then the local evaluation model is updated;
[0099] Based on the model output result, real-time feedback data and the performance degradation physical law of the key components, the health degree benchmark parameter of the evaluation model is adaptively corrected.
[0100] In the embodiment, the evaluation model is a model set for analyzing data and outputting evaluation results, which can be dynamically selected according to the scene and supports the fusion of machine learning and physical law. The performance evaluation result includes a unit health score, a component risk identification result and the like, which are used to guide operation adjustment.
[0101] In the embodiment, scene judgment is performed based on the cross-scenario correlation data and the unit operation related data in the standardized evaluation dataset to generate a scene classification result. According to the wind speed and the grid instruction, the scene is classified into normal power generation and peak regulation mode.
[0102] In the embodiment, the model processing and output call the selected evaluation model to process the standardized evaluation dataset to generate a unit health score and a component risk identification result. Meanwhile, based on the component risk identification result, a unit control parameter adjustment instruction is generated in real time and is issued to a unit actuator.
[0103] In some embodiments of the present application, when the evaluation model corresponding to the scene classification result is dynamically selected from the preset model library, it includes:
[0104] According to the scene classification result, the core performance dimension to be evaluated in the current scene is determined;
[0105] An evaluation model optimized for the core performance dimension is selected from the preset model library, and the evaluation model includes a first type of evaluation model and a second type of evaluation model;
[0106] The first type of evaluation model is used to generate a control parameter adjustment instruction for maintaining stable power generation performance, and the second type of evaluation model is used to generate a protective control parameter adjustment instruction for ensuring structural load safety.
[0107] In the present embodiment, the corresponding evaluation model is dynamically selected according to the scene classification result from the preset model library. The model library includes a plurality of models, the first type of model is used for power generation performance stability, and the second type of model is used for structural load safety. When selecting, the core performance dimension to be evaluated in the current scene is first determined, and then the model optimized for the dimension is selected.
[0108] In some embodiments of the present application, when in the cross-scene collaborative evaluation mode, the encrypted model parameters from the federated learning aggregation node are received, the encrypted model parameters are adapted in combination with the individual characteristics of the local group, and then the local evaluation model is updated, including:
[0109] The received encrypted model parameters are subjected to adaptability evaluation related to the individual characteristics of the local group;
[0110] The adaptability evaluation is used to dynamically adjust the weight coefficient of the encrypted model parameters according to the historical operation and maintenance data of the local group and the component aging coefficient;
[0111] When the encrypted model parameters pass the adaptability evaluation, they are used to update the local evaluation model.
[0112] In the present embodiment, the cross-scene collaborative evaluation is when in the cross-scene collaborative evaluation mode, the encrypted model parameters from the federated learning aggregation node are received. Federated learning allows the model to be updated without sharing the original data. Then, the encrypted model parameters are adapted in combination with the individual characteristics of the local group, such as historical operation and maintenance data, component aging coefficient, and geographical location microclimate characteristics, and the weight coefficient is dynamically adjusted. Only when the adaptability evaluation is passed, the local evaluation model is updated.
[0113] In some embodiments of the present application, when the health degree benchmark parameter of the evaluation model is adaptively corrected based on the model output result, the real-time feedback data, and the performance degradation physical law of the key component, the method comprises:
[0114] The real-time feedback data is compared with the expected state data predicted based on the performance degradation physical law to generate a residual sequence;
[0115] The evaluation model is incrementally updated based on the residual sequence, the historical performance degradation trend, and the online learning mechanism;
[0116] The correction amount of the health degree benchmark parameter is associated with the material fatigue characteristics of the key component.
[0117] In the present embodiment, the health degree benchmark parameter of the evaluation model is adaptively corrected based on the model output result, the real-time feedback data, and the performance degradation physical law of the key component. Specifically, the real-time feedback data is compared with the expected state data predicted based on the physical law to generate a residual sequence; the evaluation model is incrementally updated based on the residual sequence and the historical performance degradation trend through an online learning mechanism. The correction amount is associated with the material fatigue characteristics of the key component.
[0118] In some embodiments of the present application, when the performance evaluation result of the standardized evaluation data set generated based on the evaluation model is generated, the method comprises:
[0119] A model update strategy is generated, which is used to adaptively adjust the strength of model updating according to the dynamic characteristics of the residual sequence and the historical performance degradation trend;
[0120] In the model updating process, a retrospective constraint on the historical knowledge distribution is synchronously introduced;
[0121] The evaluation model parameters are cooperatively optimized based on the model update strategy and the retrospective constraint.
[0122] In the present embodiment, the model updating comprises: constructing a model update strategy, adaptively adjusting the strength of model updating according to the dynamic characteristics of the residual sequence and the historical performance degradation trend; in the model updating process, a retrospective constraint on the historical knowledge distribution is synchronously introduced to prevent the model from forgetting the knowledge learned in the past; the evaluation model parameters are cooperatively optimized based on the update strategy and the retrospective constraint to ensure the stability and accuracy of the model.
[0123] In some embodiments of the present application, according to the performance evaluation result, a running adjustment suggestion is generated, a control instruction is generated based on the running adjustment suggestion, and the feedback data after execution is used to update the evaluation model, which comprises:
[0124] According to the unit health degree score and the component risk identification result output by the evaluation model, a performance evaluation conclusion including an overall health state, an abnormal type and level of a component, and a potential operation risk is generated.
[0125] Based on the performance evaluation conclusion and a preset risk control mapping rule, a unit operation adjustment suggestion is generated.
[0126] The unit operation adjustment suggestion is converted into a control instruction and is sent to a unit execution mechanism to realize real-time adjustment of a unit operation parameter.
[0127] Operation state data after the unit executes the control instruction are collected as feedback data, and the feedback data are fed back to the evaluation model to correct model parameters.
[0128] In the embodiment, the operation adjustment suggestion refers to an optimization measure such as a parameter adjustment and a maintenance plan generated based on an evaluation result; the control instruction is a directly executable operation command; and the feedback data refer to unit state data after the instruction is executed, which are used for model iteration optimization.
[0129] In the embodiment, the performance evaluation conclusion is generated according to the unit health degree score and the component risk identification result output by the evaluation model.
[0130] In the embodiment, the operation adjustment suggestion is generated based on the performance evaluation conclusion and a preset risk control mapping rule.
[0131] In the embodiment, the control instruction is generated by converting the operation adjustment suggestion into a control instruction such as an adjustment of a pitch angle and a setting of a power upper limit, and is sent to a unit execution mechanism to realize real-time adjustment of a unit operation parameter.
[0132] In the embodiment, the feedback and model updating are performed by collecting operation state data after the unit executes the control instruction as feedback data, feeding back the feedback data to the evaluation model, and correcting model parameters.
[0133] In the embodiment, the power grid cooperation is performed by synchronizing the performance evaluation conclusion and the health state information to a power grid dispatching system to realize cooperation of unit operation states and power grid dispatching.
[0134] In some embodiments of the present application, a wind turbine performance evaluation system based on multi-source data includes:
[0135] A data acquisition module is configured to acquire multi-dimensional feature data of a wind turbine to generate an original data set.
[0136] The data processing module is configured to perform normalization and correlation mapping on the original data set to generate a normalized evaluation data set.
[0137] The model evaluation module is configured to call an evaluation model based on the normalized evaluation data set, and dynamically adjust evaluation logic according to a real-time operation scene of the unit.
[0138] The decision feedback module is configured to generate an operation adjustment suggestion according to the performance evaluation result, generate a control instruction based on the operation adjustment suggestion, and use feedback data after execution to update the evaluation model.
[0139] In the embodiment, the data acquisition module is configured to acquire multi-dimensional feature data of the wind turbine to generate an original data set. The module integrates sensors, power grid interfaces, meteorological equipment, image acquisition equipment, and the like, and supports real-time data flow.
[0140] In the embodiment, the data processing module is configured to perform normalization and correlation mapping on the original data set to generate a normalized evaluation data set. The module includes a data cleaning unit, a correlation analysis unit, and an unstructured data analysis unit.
[0141] In the embodiment, the model evaluation module is configured to call an evaluation model based on the normalized evaluation data set, and dynamically adjust evaluation logic according to a real-time operation scene of the unit. The module includes a scene classification unit, a model selection unit, and a model execution unit, and supports federated learning and online learning.
[0142] In the embodiment, the decision feedback module is configured to generate an operation adjustment suggestion according to the performance evaluation result, generate a control instruction based on the operation adjustment suggestion, and use feedback data after execution to update the evaluation model. The module includes a suggestion generation unit, an instruction issuing unit, and a feedback processing unit.
[0143] The above only describes the preferred embodiments of the present application. It should be noted that, for those skilled in the art, without departing from the technical principles of the present application, several improvements and replacements can be made, which should also be considered as the protection scope of the present application.
Claims
1. A wind turbine performance evaluation method based on multi-source data, characterized in that, include: Collect multi-dimensional feature data of wind turbine units to generate raw datasets; The original dataset is standardized and associated with a mapping process to generate a standardized evaluation dataset. Performance evaluation results for the standardized evaluation dataset are generated based on the evaluation model. Based on the performance evaluation results, operational adjustment suggestions are generated, and control instructions are generated based on the operational adjustment suggestions. The feedback data after execution is used to update the evaluation model.
2. The wind turbine performance evaluation method based on multi-source data according to claim 1, characterized in that, When generating the original dataset based on the multi-dimensional feature data of the collected wind turbine units, it includes: The multi-dimensional feature data includes unit operation-related data, cross-scenario related data, unstructured data, and full lifecycle data; Real-time power, rotor speed, vibration parameters of key components, temperature parameters, and blade pitch angle are obtained from sensors and the unit control system to generate relevant data on unit operation. Cross-scenario related data are generated by obtaining power grid peak-shaving commands, frequency regulation commands, and power load change data through the power grid interface, as well as wind speed, ambient temperature, air density, and extreme weather warning data obtained through meteorological monitoring equipment. Unstructured data is generated from manual maintenance records recorded by mobile terminals and blade surface images and temperature field images of key components acquired by image acquisition devices. The manual maintenance records include equipment anomaly descriptions, maintenance operations, and inspection results. The unit design parameters, rated performance indicators and life parameters obtained through the archive system, as well as historical fault records, component replacement records and performance change trend data obtained through the operation and maintenance management system, are used to generate full life cycle data. The unit operation-related data, cross-scenario associated data, unstructured data, and full lifecycle data are associated and stored according to the unit's unique identifier and timestamp to form the original dataset.
3. The wind turbine performance evaluation method based on multi-source data according to claim 1, characterized in that, When generating a standardized evaluation dataset by performing standardization and association mapping on the original dataset, the following steps are included: The original dataset is cleaned and standardized to generate standardized data. Based on the standardized data, cross-scenario correlation analysis is performed to generate data correlation mapping relationships; Using the aforementioned data association mapping relationship, unstructured data is parsed and quantified to generate structured state parameters; The structured state parameters are corrected by combining full lifecycle data to generate performance correction coefficients; The standardized data, structured state parameters, and performance correction coefficients are integrated to generate a standardized evaluation dataset.
4. The wind turbine performance evaluation method based on multi-source data according to claim 1, characterized in that, When outputting performance evaluation results using the evaluation model based on the standardized evaluation dataset, the following are included: Based on the cross-scenario related data and unit operation-related data in the standardized evaluation dataset, scenario determination is performed to obtain scenario classification results; Dynamically select the evaluation model corresponding to the scene classification result from the pre-set model library; The selected assessment model is used to process the standardized assessment dataset to obtain the unit health score and component risk identification results; Based on the risk identification results of the components, the unit control parameter adjustment instructions are generated in real time and sent to the unit actuators. When in cross-field collaborative evaluation mode, the encrypted model parameters are received from the federated learning aggregation node. The encrypted model parameters are then adapted to the individual characteristics of the local group before the local evaluation model is updated. Based on the model output results, real-time feedback data, and the physical laws governing the performance degradation of key components, the health benchmark parameters of the evaluation model are adaptively corrected.
5. The wind turbine performance evaluation method based on multi-source data according to claim 4, characterized in that, When dynamically selecting an evaluation model corresponding to the scene classification result from a pre-set model library, the process includes: Based on the scenario classification results, determine the core performance dimensions to be evaluated in the current scenario; Select an evaluation model optimized for the core performance dimension from a pre-built model library. The evaluation model includes a first type of evaluation model and a second type of evaluation model. The first type of evaluation model is used to generate control parameter adjustment instructions to maintain the stability of power generation performance, while the second type of evaluation model is used to generate protective control parameter adjustment instructions to ensure the safety of structural loads.
6. The wind turbine performance evaluation method based on multi-source data according to claim 4, characterized in that, When in cross-field collaborative evaluation mode, after receiving encrypted model parameters from federated learning aggregation nodes, adapting the encrypted model parameters to the individual characteristics of the local group, and then updating the local evaluation model, the process includes: The received encryption model parameters are evaluated for their compatibility with the individual characteristics of the local group. The adaptability assessment is used to dynamically adjust the weighting coefficients of the encryption model parameters based on the historical operation and maintenance data of the local unit and the component aging coefficient. When the encryption model parameters pass the adaptability evaluation, they are used to update the local evaluation model.
7. The wind turbine performance evaluation method based on multi-source data according to claim 4, characterized in that, When adaptively adjusting the health baseline parameters of the evaluation model based on model output results, real-time feedback data, and the physical laws governing performance degradation of key components, the following steps are taken: The real-time feedback data is compared with the expected state data predicted based on the physical laws of performance degradation to generate a residual sequence; The evaluation model is incrementally updated based on the residual sequence, historical performance degradation trend, and online learning mechanism. The correction amount of the health baseline parameter is related to the material fatigue characteristics of the key components.
8. The wind turbine performance evaluation method based on multi-source data according to claim 7, characterized in that, When generating performance evaluation results for the standardized evaluation dataset based on the evaluation model, the following are included: A model update strategy is generated, which is used to adaptively adjust the intensity of model updates based on the dynamic characteristics and historical performance degradation trend of the residual sequence. During the model update process, a retrospective constraint on the distribution of historical knowledge is introduced simultaneously. The model update strategy and the retrospective constraints are used to evaluate the synergistic optimization of the model parameters.
9. The wind turbine performance evaluation method based on multi-source data according to claim 1, characterized in that, Based on the performance evaluation results, generating operational adjustment suggestions and generating control instructions based on the operational adjustment suggestions, and using the feedback data after execution to update the evaluation model, includes: Based on the unit health score and component risk identification results output by the evaluation model, a performance evaluation conclusion is generated that includes the overall health status, component anomaly type and level, and potential operational risks. Based on the performance evaluation conclusions and the pre-set risk control mapping rules, recommendations for unit operation adjustment are generated. The unit operation adjustment suggestions are converted into control commands and sent to the unit actuators to realize real-time adjustment of unit operating parameters; The system collects the operating status data of the unit after executing the control command and uses it as feedback data. The feedback data is then sent back to the evaluation model to correct the model parameters.
10. A wind turbine performance evaluation system based on multi-source data, applied to the wind turbine performance evaluation method based on multi-source data as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to collect multi-dimensional feature data of wind turbines and generate raw datasets; The data processing module is used to perform standardization and association mapping on the original dataset to generate a standardized evaluation dataset; The model evaluation module is used to call the evaluation model based on the standardized evaluation dataset and dynamically adjust the evaluation logic according to the real-time operation scenario of the unit. The decision feedback module is used to generate operation adjustment suggestions based on the performance evaluation results and generate control instructions based on the operation adjustment suggestions, and use the feedback data after execution to update the evaluation model.