A data-independent collaborative optimization control method for wind turbine bearings
By constructing a collaborative control architecture combining multiphysics simulation and safety reinforcement learning, the problems of fault warning and life management of wind turbine bearings in data-scarce scenarios were solved, achieving efficient and safe bearing control.
Patent Information
- Application Number
- CN202611435781.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-16
- Publication Date
- 2026-11-17
AI Technical Summary
Existing wind turbine bearing control technologies fail to achieve synergistic optimization of multi-physics coupling effects, cannot provide early warning of faults, have poor applicability in scenarios with scarce data, and lack safety and lifespan management.
An integrated architecture is constructed, which combines multi-physics simulation, safety reinforcement learning, multi-timescale collaboration, and closed-loop feedback. It integrates mechanical dynamics, heat conduction, lubrication, electromagnetic interference, and fault evolution models, and combines physical information security layers and dynamic safety assessment rules to achieve data-independent collaborative optimization control.
It enables safe control, early warning of faults, and life management of wind turbine bearings, adapts to complex operating conditions, improves control accuracy and operation and maintenance efficiency, and reduces downtime losses and labor costs.
Smart Images

Figure CN123113367A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wind turbine operation and maintenance and control technology, and more specifically to a data-independent collaborative optimization control method for wind turbine bearings based on physical information security reinforcement learning. It is applicable to multi-physics field collaborative control, early fault warning, full life cycle life management and intelligent maintenance decision-making of key bearing components such as wind turbine main shaft bearings and generator bearings. It is especially suitable for intelligent control scenarios of newly commissioned wind farms or new types of wind turbine bearings with scarce data. Background Technology
[0002] Current wind turbine bearing control and maintenance technologies mainly revolve around condition monitoring, fault diagnosis, and regular maintenance. Existing technical solutions mainly fall into the following categories, but they have obvious limitations and lack innovation: 1. Condition monitoring and fault diagnosis technology: Existing technologies mostly focus on signal acquisition and analysis of a single physical field. They identify abnormal data quality of vibration signals through time-domain and frequency-domain analysis, thereby achieving fault diagnosis. However, they do not consider the coupling effect of multiple physical fields, and can only achieve passive diagnosis. They cannot provide early warning of potential faults, nor can they simultaneously achieve control optimization.
[0003] 2. Maintenance strategy technology: Existing technologies mainly adopt periodic maintenance, post-event repair, or predictive maintenance based on a single parameter, which do not achieve multi-objective collaborative optimization and lack dynamic consideration of safety constraints. Some improvement schemes focus on material optimization and lubrication technology upgrades to extend bearing life, but do not combine intelligent control algorithms to achieve integrated monitoring-control-maintenance. For example, existing technologies improve bearing performance through nano-lubricants, laser remelting repair, etc., but do not involve reinforcement learning and multi-physics field collaborative control logic, nor are they adapted to the complex operating conditions of wind turbine bearings.
[0004] 3. Control optimization technology: Existing bearing control is mostly focused on a single dimension such as lubrication and cooling, and adopts fixed parameter control strategies. It does not consider the multi-physics field coupling effects of mechanical load, thermal effect, vibration excitation and electromagnetic interference. A few schemes combine reinforcement learning, but do not introduce a physical information security layer, which cannot guarantee the safety of control actions. Moreover, they rely on a large amount of historical operating data, and have poor applicability in scenarios with scarce data.
[0005] 4. Industry technical pain points: Currently, patents in the field of wind power transmission system bearings are mainly concentrated in the technical branches of cages, materials, processes, conventional monitoring and testing, sealing, and lubrication. Existing technologies have not achieved the collaborative integration of physical information security reinforcement learning, multi-timescale coordination, and multi-physics field simulation, and cannot take into account safety constraints, fault warning, performance optimization, and life management. Furthermore, no dedicated control architecture has been designed specifically for the operating characteristics of wind turbine bearings. Summary of the Invention To overcome the shortcomings of the existing technologies, this invention discloses a data-independent collaborative optimization control method for wind turbine bearings. This invention abandons the isolated mode of single monitoring / diagnosis / maintenance in existing technologies and constructs an integrated architecture of multi-physics simulation, security reinforcement learning, multi-timescale collaboration, and closed-loop feedback. It does not rely on a large amount of historical operating data. Through the deep integration of the physical information security layer and the multi-physics coupling model, it adapts to the complex operating characteristics of wind turbine bearings and realizes the coordinated advancement of bearing safety control, early fault warning, performance optimization, and life management.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A data-independent collaborative optimization control method for wind turbine bearings includes: constructing an integrated architecture adapted to wind turbine bearings, comprising multi-physics simulation, security reinforcement learning, multi-timescale collaboration, and closed-loop feedback; and utilizing the constructed integrated architecture for collaborative optimization control of wind turbine bearings; wherein the integrated architecture includes: Multiphysics coupling simulation module: It integrates five sub-models, namely mechanical dynamics, heat conduction, lubrication, electromagnetic interference and fault evolution, to simulate the multiphysics coupling effect during the operation of wind turbine bearings and generate operation state simulation data that closely matches the actual operation scenario of wind turbine bearings; Physical information security reinforcement learning agent: Based on simulated data, model training is performed to generate collaborative control actions. On the basis of traditional reinforcement learning, a physical information security layer and dynamic security assessment rules are added. Combined with the safety operation requirements of wind turbine bearings, the action generation, security verification and correction optimization are integrated. Multi-timescale collaborative control module: Dividing into four time levels: ultra-short-term, short-term, medium-term, and long-term, and combining the operating conditions of wind turbine bearings, an inter-level information exchange and conflict resolution mechanism is established to achieve the synergy between short-term safety control and long-term life management, and to generate collaborative optimized control actions based on collaborative control actions; Multi-source fusion closed-loop feedback module: executes collaborative optimization control actions and integrates real-time data collected by multiple types of sensors and multi-physics simulation data to adapt to the monitoring needs of wind turbine bearings, realize bidirectional data calibration, and feed back the state after control execution to the intelligent agent to complete online learning and strategy iteration.
[0007] 1. Multiphysics Coupled Simulation Module Preferably, the construction of the multiphysics coupling simulation module includes: A five-dimensional multiphysics coupling model is constructed, including a mechanical dynamics sub-model, a heat conduction sub-model, a lubrication sub-model, an electromagnetic interference sub-model, and a fault evolution sub-model. Each sub-model is coupled and linked through a data interface to simulate the real operating environment of wind turbine bearings. Input the basic parameters of the wind turbine bearing and initialize the parameters of the coupled model; Various operating conditions of the wind turbine bearing are input into the coupled model for simulation, and the real-time operating status parameters of the bearing are output, including vibration characteristics, temperature information, load information, lubrication status, electromagnetic interference parameters and fault characteristics. Environmental noise and sensor error simulations are added to make the simulated state closely resemble the real operating scenario of wind turbine bearings. At the same time, an adjustable fault injection interface is set up to simulate the dynamic evolution process of common failure modes of wind turbine bearings, which can be used to train the agent's fault response capabilities without relying on real fault data.
[0008] Preferably, the mechanical dynamics sub-model is used to simulate the load and vibration coupling effect of the wind turbine bearing, the heat conduction sub-model is used to simulate the influence of bearing frictional heat generation and nacelle ambient temperature distribution, the lubrication sub-model is used to simulate bearing oil film formation and failure, the electromagnetic interference sub-model is used to simulate the influence of generator shaft voltage and shaft current on the bearing, and the fault evolution sub-model is used to simulate the dynamic development of common failure modes of wind turbine bearings.
[0009] 2. Physical Information Security Reinforcement Learning Intelligent Agent Preferably, the construction of the physical information security reinforcement learning agent includes: The design incorporates a multi-dimensional coupled state space, encompassing five dimensions: mechanical, thermal, lubrication, electromagnetic, and fault. Specifically, it includes: vibration time-domain / frequency-domain characteristics, temperature of various bearing components, radial / axial loads, oil film parameters, electromagnetic interference intensity, fault severity, and health status, forming a complete state vector adapted to the operating characteristics of wind turbine bearings. The design incorporates a multi-dimensional collaborative action space, including five dimensions: dynamic load distribution, adaptive adjustment of lubrication parameters, intelligent control of the cooling system, fine-tuning of rotation speed, and dynamic generation of maintenance decisions. The action range is dynamically adjusted based on the physical characteristics of the wind turbine bearings, rather than being controlled by fixed parameters. An adaptive multi-objective reward function is constructed, which comprehensively considers four dimensions: safety, economy, operation performance and lifespan. The weight coefficients are dynamically adjusted according to the real-time operating status of the wind turbine bearings, rather than fixed weights, to guide the agent to learn the optimal collaborative strategy that adapts to the operating conditions of the wind turbine bearings. A dynamic safety assessment rule is constructed, and a deep neural network is used to train the assessment model. The real-time state vector of the wind turbine bearing is input, and the safety probability of each safety dimension is output. This is different from the existing fixed threshold safety judgment and realizes dynamic safety verification that is adapted to the operating characteristics of wind turbine bearings. A physical information security layer is constructed, and a quadratic programming algorithm is used to perform security correction on the original control actions generated by the intelligent agent. The core is to satisfy all physical safety constraints of the wind turbine bearings while ensuring that the difference between the control actions and the original actions is minimized. The agent is trained in a hierarchical manner. Based on the simulation environment provided by the multiphysics coupling simulation module, which is adapted to the wind turbine bearing, the basic control strategy is first trained under simplified working conditions. The complexity of multiphysics coupling, fault scenarios and electromagnetic interference are gradually increased. After training, the agent is directly deployed to the actual control system of the wind turbine bearing.
[0010] 3. Multi-timescale collaborative control module Preferably, the construction of the multi-timescale collaborative control module includes: The system divides the time into four independent and coordinated time levels, clarifies the core tasks and decision-making frequency of each level, distinguishes itself from the existing single time scale control, and adapts to the operating characteristics of wind turbine bearings. Establish an inter-layer collaboration mechanism to achieve two-way information transmission, time scale matching, and conflict resolution; Layered learning and optimization: each layer of intelligent agent independently learns the control strategy corresponding to the time scale, while achieving overall collaborative optimization through inter-layer information interaction. This ensures that short-term control does not affect long-term lifespan, and long-term decision-making does not ignore short-term safety, thus adapting to the operational needs of wind turbine bearings throughout their entire life cycle.
[0011] Preferably, the division into four independent and coordinated time levels includes: Millisecond-level ultra-short-term layer: responsible for real-time vibration suppression and electromagnetic interference buffering, with high decision frequency and rapid response to safety risks caused by instantaneous gusts and sudden electromagnetic interference from wind turbines; The short-term layer, measured in seconds, is responsible for dynamic load distribution, real-time adjustment of lubrication and cooling parameters, and capturing dynamic changes in the operating status of wind turbine bearings. The minute-level mid-term layer is responsible for early fault warning, operating parameter optimization, identifying early fault characteristics of wind turbine bearings, and adjusting control strategies. Long-term layer (hours and above): Responsible for dynamic generation of maintenance strategies and dynamic lifespan management, and for developing optimal maintenance plans adapted to wind power operation and maintenance scenarios based on the cumulative damage of wind turbine bearings.
[0012] Preferably, the establishment of the inter-layer collaboration mechanism to achieve bidirectional information transmission, time scale matching, and conflict resolution includes: Information transmission: High-level decisions serve as constraints for low-level controls, while the real-time status of low-level controls serves as the basis for high-level decisions. Time scale matching: Through a buffering mechanism, the control actions at different time levels are ensured to be coordinated and consistent, avoiding decision conflicts and adapting to the dynamic operating conditions of wind turbine bearings. Conflict resolution: Following the priorities of safety first, performance second, lifespan best, and cost lowest, resolve decision conflicts at different levels to meet the operation and maintenance needs of wind turbines.
[0013] 4. Multi-source fusion closed-loop feedback module Preferably, the construction of the multi-source fusion closed-loop feedback module includes: A multi-source sensor acquisition system was built, deploying triaxial vibration sensors, multi-point temperature sensors, load sensors, speed sensors, lubrication sensors, and electromagnetic interference sensors to collect real-time operating status data of wind turbine bearings, adapting to the monitoring environment of wind power sites. The real data collected by the sensor is bidirectionally calibrated with the simulation data output by the multiphysics coupling simulation module that is adapted to the wind turbine bearing. This corrects the simulation model parameters, improves the simulation accuracy, and makes up for the limitations of single real data or simulation data. A smart actuator is built to receive control actions output by a safety reinforcement learning agent, and to perform load adjustment, lubrication control, and cooling control operations. The execution process is fed back to the feedback module in real time, adapting to the control requirements of wind turbine bearings. Collect wind turbine bearing status data after executing control actions, calculate reward function values, and feed them back to the safety reinforcement learning agent for online learning and policy iteration. At the same time, feed the status data back to the multiphysics coupling simulation module to achieve dynamic model optimization and adapt to the dynamic changes in wind power conditions. A tiered early warning and emergency response mechanism is set up. When the security probability output by the security assessment rule of the intelligent agent is lower than the preset threshold, the corresponding level of early warning signal is immediately triggered, and emergency control actions are initiated to prevent the fault from escalating. Maintenance suggestions adapted to the wind turbine operation and maintenance scenario are generated simultaneously.
[0014] 5. Model Training Preferably, the method further includes training a model for the physical information security reinforcement learning agent, wherein the model training includes: Start the multiphysics coupling simulation module, input the basic parameters of the wind turbine bearing, initialize the five-dimensional multiphysics coupling model, set various operating conditions, fault scenarios and electromagnetic interference scenarios to adapt to wind power scenarios, and generate simulation state data; Initialize the physical information security reinforcement learning agent, set up a multi-dimensional coupled state space, cooperative action space, adaptive multi-objective reward function and dynamic security constraint parameters adapted to wind turbine bearings, and train dynamic security assessment and security layer optimization algorithms; A hierarchical training strategy is adopted to train the agent in a simulated environment, gradually increasing the complexity of multi-physics coupling, fault scenarios and electromagnetic interference intensity, and optimizing the control strategy until the agent can stably generate cooperative control actions that meet safety constraints, take into account multi-objective optimization and adapt to the operating conditions of wind turbine bearings. Through the multi-timescale collaborative control module, the agent strategies at each time level are collaboratively optimized to ensure that the control tasks at different time scales are coordinated and consistent, forming a complete collaborative control strategy adapted to wind turbine bearings. After training is completed, save the parameters of the intelligent agent model and the multiphysics coupling simulation module model to complete the training phase.
[0015] 6. Actual Operation Preferably, the collaborative optimization control of wind turbine bearings using the constructed integrated architecture includes: Step 1: Deploy the integrated architecture to the wind turbine bearing control system, start the sensor acquisition system and actuator of the multi-source fusion closed-loop feedback module, and initialize the wind turbine bearing parameters and multi-physics coupling simulation module; Step 2: The multiphysics coupling simulation module generates simulation data without data dependence. The sensor acquisition system collects the operating status data of the wind turbine bearing in real time and transmits it to the multi-source fusion closed-loop feedback module. The simulation data of the multiphysics coupling simulation module is bidirectionally calibrated and outputs an accurate real-time bearing status vector. Step 3: The physical information security reinforcement learning agent receives the state vector and judges whether the wind turbine bearing is safe through dynamic security assessment rules. If it is safe, it generates the original cooperative control action and outputs it. If it is not safe, it corrects the cooperative control action through the physical information security layer before outputting it. Step 4: The multi-timescale collaborative control module coordinates the control actions output by the intelligent agent in a hierarchical manner, and sends the control actions of each time scale to the actuator synchronously to achieve collaborative control and adapt to the dynamic operating conditions of the wind turbine. Step 5: The actuator performs control actions, while the sensor acquisition system collects the state data after execution, calculates the reward function value, and feeds it back to the agent and the multiphysics coupling simulation module to realize online learning, strategy iteration and simulation model optimization of the agent, and adapt to the dynamic changes of wind power conditions; Step 6: If an abnormal operation of the wind turbine bearing is detected, immediately trigger the corresponding level of early warning signal, initiate emergency control actions, and generate dynamic maintenance suggestions adapted to the wind power operation and maintenance scenario to guide operation and maintenance personnel to perform precise maintenance. Step 7: Repeat steps 2-6 to achieve coordinated optimization control of the wind turbine bearing throughout its entire life cycle, until the bearing reaches its design life or is replaced.
[0016] The beneficial effects of this invention are: 1. No reliance on historical data, wider applicability, and complete avoidance of the data dependency defects of existing technologies: Existing technologies all rely on a large amount of historical operating data or fault data for model training, which cannot be applied in scenarios with scarce data, such as newly commissioned wind farms and new wind turbine bearings; This application uses a five-dimensional multiphysics coupling simulation module to simulate various operating conditions, faults and electromagnetic interference scenarios of wind turbine bearings without historical data, and completes the training of intelligent agents. The scope of application covers various types of wind turbine bearings, especially solving the control problem in scenarios with scarce data, which is completely different from the technical route of existing technologies.
[0017] 2. Achieving multi-physics field coordinated control with higher control precision, avoiding the limitations of existing single-physics field control: Existing technologies mostly focus on the monitoring or control of a single physical field, without considering the coupling effects of mechanical, thermal, lubrication, and electromagnetic factors, and are not adapted to the complex operating characteristics of wind turbine bearings, resulting in low control precision. This application constructs a five-dimensional multi-physics field coupling model to achieve multi-physics field coordinated control. At the same time, through multi-source data fusion calibration, the control precision is significantly improved. Furthermore, it covers electromagnetic interference, a key factor affecting the operation of wind turbine bearings that is easily overlooked in existing technologies, making the technical approach innovative.
[0018] 3. Enhanced safety and optimization through synergy, avoiding the shortcomings of existing reinforcement learning technologies that prioritize optimization over safety: Existing bearing control technologies that combine reinforcement learning lack a dedicated safety layer, failing to guarantee the safety of control actions and easily leading to safety risks. Furthermore, they are not adapted to the safe operation requirements of wind turbine bearings. This application adds a physical information security layer and dynamic safety assessment rules, achieving integrated action generation, safety verification, and correction optimization. This ensures that control actions meet both optimization objectives and the physical safety constraints of wind turbine bearings, fundamentally improving bearing operation safety. This design differs from the core design of existing technologies.
[0019] 4. Multi-timescale coordination, taking into account both short-term safety and long-term lifespan, avoiding the shortcomings of existing single-timescale control: Existing technologies only focus on short-term control (such as lubrication and cooling) or long-term maintenance, and cannot achieve coordination between the two. This can easily lead to short-term optimization damaging long-term lifespan or long-term maintenance neglecting short-term safety, and it is not adapted to the operating conditions of wind turbine bearings. This application divides four time levels and establishes a coordination mechanism to achieve coordination between short-term vibration suppression, electromagnetic interference buffering and long-term lifespan management and maintenance decision-making, resulting in better overall benefits.
[0020] 5. Integrated collaborative control for higher operation and maintenance efficiency, avoiding the limitations of isolated modules in existing technologies: The monitoring, diagnosis, control, and maintenance modules of existing technologies are independent of each other, unable to achieve collaborative linkage, resulting in low operation and maintenance efficiency and not adapted to the operation and maintenance scenarios of wind turbines; This application constructs an integrated architecture of multi-physics simulation, security reinforcement learning, multi-timescale collaboration, and closed-loop feedback, realizing the collaboration of the entire process of monitoring, control, early warning, and maintenance. Dynamic optimization and fault response can be completed without manual intervention, significantly improving operation and maintenance efficiency, reducing labor costs, and adapting to the operation and maintenance needs of wind turbines.
[0021] 6. Enhanced dynamic adaptability, adapting to complex operating conditions and avoiding the shortcomings of existing fixed parameter control: Existing technologies mostly adopt fixed parameter control or fixed weight optimization, which cannot adapt to the complex operating conditions of wind turbines such as wind speed fluctuations, load changes, and electromagnetic interference; The adaptive multi-objective reward function, dynamic safety assessment rules, and closed-loop feedback mechanism of this application can dynamically adjust the control strategy according to the real-time operating status of the wind turbine bearing, adapting to various complex operating conditions, and can be adapted to wind turbine bearings of different models and operating environments, making it more versatile.
[0022] 7. Early warning of faults, lower losses, and avoidance of the shortcomings of passive diagnosis in existing technologies: Existing technologies can only achieve passive diagnosis after a fault occurs, and cannot provide early warning of potential faults, which can easily lead to the expansion of faults and is not adapted to the fault evolution characteristics of wind turbine bearings. This application can achieve accurate early warning of faults through dynamic safety assessment rules and multi-dimensional status monitoring, and at the same time initiate emergency control actions to eliminate faults in their infancy, significantly reducing downtime losses and maintenance costs, which is different from the core objectives of existing fault diagnosis technologies. Attached Figure Description
[0023] Figure 1 This is a flowchart of the actual operation phase of the present invention. Detailed Implementation
[0024] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention.
[0025] Example 1 A data-independent collaborative optimization control method for wind turbine bearings, the overall architecture of which is designed as follows: The overall architecture of this invention comprises four core modules, each operating independently yet working in tandem to form a complete closed-loop control system. This differs from the isolated module design of existing technologies, and each module is specifically designed to adapt to the operating characteristics of wind turbine bearings. 1. Multiphysics Coupled Simulation Module: Unlike existing single-physics simulations, this module integrates five sub-models: mechanical dynamics, heat conduction, lubrication, electromagnetic interference, and fault evolution. It accurately simulates the multiphysics coupling effect during the operation of wind turbine bearings and can generate operating status data that closely matches the actual operating scenario of wind turbine bearings without relying on historical data. 2. Physical Information Security Reinforcement Learning Agent Module: The core improvement lies in adding a physical information security layer and dynamic security assessment rules to the traditional reinforcement learning, and combining the safety operation requirements of wind turbine bearings to achieve the integration of action generation, security verification and correction optimization, thus avoiding the defects of existing reinforcement learning without security constraints. 3. Multi-timescale collaborative control module: Dividing the time into four time levels: ultra-short-term, short-term, medium-term, and long-term, and combining the operating conditions of wind turbine bearings, an inter-level information exchange and conflict resolution mechanism is established to achieve the synergy between short-term safety control and long-term life management, which is different from the single-timescale control of existing technologies. 4. Multi-source fusion closed-loop feedback module: Integrates real-time data collected by multiple types of sensors and multi-physics simulation data, adapts to the monitoring needs of wind turbine bearings, realizes bidirectional data calibration, and feeds back the state after control execution to the intelligent agent to complete online learning and strategy iteration, avoiding the existing single data acquisition or single simulation technical route.
[0026] Example 2 This embodiment further elaborates on the above embodiments, and details the specific implementation steps of each module.
[0027] ① Implementation steps of the multiphysics coupling simulation module The core of this module is to address the shortcomings of existing technologies that do not fully consider multi-physics coupling and rely on historical data. It is designed specifically for the operating characteristics of wind turbine bearings. The specific steps are as follows: Step 1: Construct a five-dimensional multiphysics coupling model, including a mechanical dynamics model (simulating the load and vibration coupling effect of wind turbine bearings), a heat conduction model (simulating the influence of bearing friction heat generation and nacelle ambient temperature distribution), a lubrication model (simulating bearing oil film formation and failure), an electromagnetic interference model (simulating the influence of generator shaft voltage and shaft current on bearings), and a fault evolution model (simulating the dynamic development of common failure modes of wind turbine bearings). Each sub-model is coupled and linked through a data interface to accurately simulate the real operating environment of wind turbine bearings. Step 2: Input the basic parameters of the wind turbine bearing and initialize the model parameters. There is no need to import historical operating data or fault data. Step 3: Simulate various operating conditions of the wind turbine bearing (wind speed fluctuations, sudden load changes, etc.), input them into the coupled model, and output the real-time operating status parameters of the bearing, including vibration characteristics, temperature information, load information, lubrication status, electromagnetic interference parameters and fault characteristics; Step 4: Add environmental noise and sensor error simulation to make the simulation state fit the real operating scenario of wind turbine bearings. At the same time, set an adjustable fault injection interface to simulate the dynamic evolution process of common failure modes of wind turbine bearings such as wear failure and fatigue spalling. This can be used to train the agent's fault response capability without relying on real fault data.
[0028] ② Implementation steps of the reinforcement learning intelligent agent module for physical information security This module is the core innovation of this invention. Unlike existing applications of reinforcement learning in bearing control, it focuses on adding a physical information security layer and dynamic security assessment rules, and is adapted to the operational requirements of wind turbine bearings. The specific steps are as follows: Step 1: Design a multi-dimensional coupled state space that covers five dimensions: mechanical, thermal, lubrication, electromagnetic, and fault. Unlike the existing single-dimensional state space, it specifically includes: vibration time-domain / frequency-domain characteristics, temperature of various parts of the bearing, radial / axial load, oil film parameters, electromagnetic interference intensity, fault severity, and health status, forming a complete state vector that adapts to the operating characteristics of wind turbine bearings. Step 2: Design a multi-dimensional collaborative action space, including five dimensions: dynamic load distribution, adaptive adjustment of lubrication parameters, intelligent control of the cooling system, fine-tuning of speed, and dynamic generation of maintenance decisions. The action range is dynamically adjusted based on the physical characteristics of the wind turbine bearing, rather than controlled by fixed parameters. Step 3: Construct an adaptive multi-objective reward function that comprehensively considers four dimensions: safety, economy, operation performance and lifespan. The weight coefficients are dynamically adjusted according to the real-time operating status of the wind turbine bearings, rather than fixed weights, to guide the agent to learn the optimal collaborative strategy that adapts to the operating conditions of the wind turbine bearings. Step 4: Construct dynamic safety assessment rules, use deep neural network to train the assessment model, input the real-time state vector of the wind turbine bearing, and output the safety probability of each safety dimension (vibration, temperature, load, lubrication, electromagnetic). This is different from the existing fixed threshold safety judgment and realizes dynamic safety verification adapted to the operating characteristics of wind turbine bearings. Step 5: Construct a physical information security layer and use a quadratic programming algorithm to correct actions. The core is to ensure that all physical safety constraints of the wind turbine bearings are met, such as vibration amplitude, temperature, and load not exceeding limits, while minimizing the difference between the control action and the original action. This avoids the shortcomings of existing reinforcement learning that emphasizes optimization but neglects safety. Step 6: Layered training of the agent. Based on the simulation environment provided by the multiphysics coupling simulation module and adapted to the wind turbine bearing, the basic control strategy is first trained under simplified operating conditions. The complexity of multiphysics coupling, fault scenarios and electromagnetic interference are gradually increased. After training, the agent can be directly deployed to the actual control system of the wind turbine bearing without relying on historical operating data.
[0029] ③ Implementation steps of the multi-timescale collaborative control module This module addresses the shortcomings of existing technologies that rely on a single time-scale control, failing to balance short-term safety and long-term lifespan. Furthermore, it is designed specifically to address the unique operating conditions of wind turbine bearings. The specific steps are as follows: Step 1: Divide the time into four independent yet coordinated time levels, clarify the core tasks and decision frequencies of each level, differentiate it from existing single-time-scale control, and adapt it to the operating characteristics of wind turbine bearings: (1) Ultra-short-term layer (millisecond level): responsible for real-time vibration suppression and electromagnetic interference buffering, with high decision frequency and rapid response to safety risks caused by instantaneous gusts and sudden electromagnetic interference of wind turbines; (2) Short-term layer (second level): responsible for dynamic load distribution, real-time adjustment of lubrication and cooling parameters, and capturing dynamic changes in the operating status of wind turbine bearings, such as load changes caused by wind speed fluctuations; (3) Mid-term layer (minute level): responsible for early warning of faults, optimization of operating parameters, identification of early fault characteristics of wind turbine bearings, and adjustment of control strategies; (4) Long-term layer (hour level and above): responsible for dynamic generation of maintenance strategies and dynamic life management, and formulate the optimal maintenance plan that is suitable for wind power operation and maintenance scenarios based on the cumulative damage of wind turbine bearings. Step 2: Establish an inter-layer collaboration mechanism to achieve two-way information transmission, time scale matching, and conflict resolution. (1) Information transmission: High-level decisions serve as constraints for low-level controls, and the real-time status of low-level controls serves as the basis for high-level decisions. (2) Time scale matching: Through the buffering mechanism, the control actions at different time levels are coordinated and consistent to avoid decision conflicts and adapt to the dynamic operating conditions of wind turbine bearings. (3) Conflict resolution: Following the priority of safety first, performance second, lifespan best, and cost lowest, resolve decision conflicts at different levels to meet the operation and maintenance needs of wind turbines; Step 3: Layered learning and optimization. Each layer of the intelligent agent independently learns the control strategy corresponding to the time scale. At the same time, through inter-layer information interaction, overall collaborative optimization is achieved to ensure that short-term control does not affect long-term lifespan and long-term decision-making does not ignore short-term safety, thus adapting to the operational needs of the wind turbine bearing throughout its entire life cycle.
[0030] ④ Implementation steps of the multi-source fusion closed-loop feedback module This module addresses the shortcomings of existing technologies, such as limited data and delayed feedback, enabling bidirectional calibration between simulated and real data. It is also adapted to the monitoring and control requirements of wind turbine bearings. The specific steps are as follows: Step 1: Build a multi-source sensor acquisition system, deploying triaxial vibration sensors, multi-point temperature sensors, load sensors, speed sensors, lubrication sensors, and electromagnetic interference sensors to collect real-time operating status data of wind turbine bearings and adapt to the monitoring environment of wind power sites. Step 2: Perform bidirectional calibration between the real data collected by the sensor and the simulation data output by the multiphysics coupling simulation module that is adapted to the wind turbine bearing, correct the simulation model parameters, improve the simulation accuracy, and make up for the limitations of single real data or simulation data. Step 3: Build an intelligent actuator to receive corrective actions output by the safety reinforcement learning agent, and perform operations such as load adjustment, lubrication control, and cooling control. The action execution process is fed back to the feedback module in real time to adapt to the control requirements of wind turbine bearings. Step 4: Collect the bearing status data of the wind turbine after executing the control action, calculate the reward function value, and feed it back to the safety reinforcement learning agent for online learning and policy iteration. At the same time, feed the status data back to the multiphysics coupling simulation module to realize dynamic model optimization and adapt to the dynamic changes of wind power conditions. Step 5: Set up a tiered early warning and emergency response mechanism. When the safety probability output by the safety assessment rules is lower than the preset threshold, the corresponding level of early warning signal will be triggered immediately, and emergency control actions will be initiated at the same time to prevent the fault from escalating. Maintenance suggestions adapted to the wind turbine operation and maintenance scenario will be generated simultaneously.
[0031] Example 3 This embodiment further elaborates on the above embodiments, and illustrates the usage process of the method of the present invention.
[0032] The method of this invention is divided into a model training phase and an actual operation phase. Its core advantage lies in the fact that the training phase does not rely on historical data, thus avoiding the limitations of existing technologies that depend on data. Furthermore, it is fully adaptable to the application scenarios of wind turbine bearings, as detailed below: ① Model training phase Step 1: Start the multiphysics coupling simulation module, input the basic parameters of the wind turbine bearing, initialize the five-dimensional multiphysics coupling model, set various operating conditions, fault scenarios and electromagnetic interference scenarios to adapt to wind power scenarios, and generate simulation state data; Step 2: Initialize the physical information security reinforcement learning agent, set the multi-dimensional coupled state space, cooperative action space, adaptive multi-objective reward function and dynamic security constraint parameters adapted to the wind turbine bearing, and train the dynamic security assessment model and security layer optimization algorithm; Step 3: Using a hierarchical training strategy, train the agent in a simulated environment, gradually increasing the complexity of multi-physics coupling, fault scenarios and electromagnetic interference intensity, and optimizing the control strategy until the agent can stably generate cooperative control actions that meet safety constraints, take into account multi-objective optimization and adapt to the operating conditions of wind turbine bearings. Step 4: Through the multi-timescale collaborative control module, the agent strategies at each time level are collaboratively optimized to ensure that the control tasks at different time scales are coordinated and consistent, forming a complete collaborative control strategy adapted to the wind turbine bearing. Step 5: After training is complete, save the parameters of the intelligent agent model and the parameters of the multiphysics coupling simulation model to complete the training phase. No historical running data or fault data needs to be imported.
[0033] ② In the actual operation phase, such as Figure 1 As shown: Step 1: Deploy the trained intelligent agent model, multi-timescale collaborative control module, and multi-source fusion closed-loop feedback module to the wind turbine bearing control system, start the sensor acquisition system and actuators, complete system initialization, and adapt to the operating environment of the wind power site; Step 2: The sensor acquisition system collects real-time operating status data of the wind turbine bearing and transmits it to the multi-source fusion closed-loop feedback module. The data is then bidirectionally calibrated with the output data of the multi-physics coupling simulation module to output an accurate real-time bearing status vector. Step 3: The physical information security reinforcement learning agent receives the state vector and judges whether the wind turbine bearing is safe through dynamic security assessment rules. If it is safe, it generates the original cooperative control action. If it is not safe, it corrects the action through the physical information security layer to ensure that the action meets all security constraints. Step 4: The multi-timescale collaborative control module coordinates the control actions output by the intelligent agent in a hierarchical manner, and sends the control actions of each time scale to the actuator synchronously to achieve collaborative control and adapt to the dynamic operating conditions of the wind turbine. Step 5: The actuator performs control actions, while the sensors collect the state data after execution, calculate the reward function value, and feed it back to the agent and the multiphysics coupling simulation module to realize online learning, strategy iteration and simulation model optimization of the agent, and adapt to the dynamic changes of wind power conditions; Step 6: If an abnormal operation of the wind turbine bearing is detected, the system will immediately trigger the corresponding level of early warning signal, initiate emergency control actions, and generate dynamic maintenance suggestions adapted to the wind power operation and maintenance scenario to guide operation and maintenance personnel to perform precise maintenance. Step 7: Repeat steps 2-6 to achieve coordinated optimization control of the wind turbine bearing throughout its entire life cycle, until the bearing reaches its design life or is replaced.
[0034] The key points of this invention are as follows: The key points of this invention focus on collaborative integration and no data dependency, completely avoiding the single technical route of existing patents, and are designed entirely around the application scenario of wind turbine bearings. The core key points are as follows: 1. Multiphysics coupling simulation without historical data dependence: Abandoning the dependence of existing technologies on historical operating data and fault data, a five-dimensional multiphysics coupling model is constructed to simulate various operating conditions and fault scenarios of wind turbine bearings, realize full-scenario training of intelligent agents, and solve the problem of wind turbine bearing control in data-scarce scenarios. 2. Deep integration of physical information security layer and reinforcement learning: Unlike the simple application of reinforcement learning in bearing control, a physical information security layer and dynamic safety assessment rules are added. Combined with the safety operation requirements of wind turbine bearings, the integration of action generation, safety verification and correction optimization is realized. This ensures that the control actions meet both the optimization objectives and physical safety constraints, avoiding the shortcomings of existing methods that emphasize optimization but neglect safety. 3. Multi-timescale collaborative control mechanism: Divide the time into four time levels, and combine the operating conditions of wind turbine bearings to establish an inter-level information exchange and conflict resolution mechanism to achieve synergy between short-term safety control and long-term life management, which is different from the existing single-timescale control. 4. Multi-source data fusion closed-loop feedback: Integrates real sensor data and multi-physics simulation data to meet the monitoring needs of wind turbine bearings, achieve bidirectional calibration, and build a closed-loop system of acquisition-control-feedback-optimization to improve control accuracy and adaptability, avoiding the existing single data acquisition or single simulation technical route. 5. Multi-objective adaptive collaborative optimization: Construct a multi-objective reward function with adaptive weights, and combine it with the operating characteristics of wind turbine bearings to achieve collaborative optimization of safety, economy, operating performance and lifespan, which is different from the existing single-objective optimization or fixed-weight optimization technical solutions.
[0035] The embodiments of the present invention have been described in detail above, but the present invention is not limited to the described embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and these equivalents or substitutions are all included within the scope defined by the claims of the present invention.
Claims
1. A data-independent collaborative optimization control method for wind turbine bearings, characterized in that, include: An integrated architecture adaptable to wind turbine bearings, comprising multiphysics simulation, safety reinforcement learning, multi-timescale collaboration, and closed-loop feedback, is constructed. This integrated architecture is then used for collaborative optimization control of the wind turbine bearings. The integrated architecture includes: Multiphysics coupling simulation module: It integrates five sub-models, namely mechanical dynamics, heat conduction, lubrication, electromagnetic interference and fault evolution, to simulate the multiphysics coupling effect during the operation of wind turbine bearings and generate operation state simulation data that closely matches the actual operation scenario of wind turbine bearings; Physical information security reinforcement learning agent: Based on simulated data, model training is performed to generate collaborative control actions. On the basis of traditional reinforcement learning, a physical information security layer and dynamic security assessment rules are added. Combined with the safety operation requirements of wind turbine bearings, the action generation, security verification and correction optimization are integrated. Multi-timescale collaborative control module: Dividing into four time levels: ultra-short-term, short-term, medium-term, and long-term, and combining the operating conditions of wind turbine bearings, an inter-level information exchange and conflict resolution mechanism is established to achieve the synergy between short-term safety control and long-term life management, and to generate collaborative optimized control actions based on collaborative control actions; Multi-source fusion closed-loop feedback module: executes collaborative optimization control actions and integrates real-time data collected by multiple types of sensors and multi-physics simulation data to adapt to the monitoring needs of wind turbine bearings, realize bidirectional data calibration, and feed back the state after control execution to the intelligent agent to complete online learning and strategy iteration.
2. The data-independent collaborative optimization control method for wind turbine bearings as described in claim 1, characterized in that, The construction of the multiphysics coupling simulation module includes: A five-dimensional multiphysics coupling model is constructed, including a mechanical dynamics sub-model, a heat conduction sub-model, a lubrication sub-model, an electromagnetic interference sub-model, and a fault evolution sub-model. Each sub-model is coupled and linked through a data interface to simulate the real operating environment of wind turbine bearings. Input the basic parameters of the wind turbine bearing and initialize the parameters of the coupled model; Various operating conditions of the wind turbine bearing are input into the coupled model for simulation, and the real-time operating status parameters of the bearing are output, including vibration characteristics, temperature information, load information, lubrication status, electromagnetic interference parameters and fault characteristics. Environmental noise and sensor error simulations are added to make the simulated state closely resemble the real operating scenario of wind turbine bearings. At the same time, an adjustable fault injection interface is set up to simulate the dynamic evolution process of common failure modes of wind turbine bearings, which can be used to train the agent's fault response capabilities without relying on real fault data.
3. The data-independent collaborative optimization control method for wind turbine bearings as described in claim 2, characterized in that, The mechanical dynamics sub-model is used to simulate the load and vibration coupling effect of wind turbine bearings; the heat conduction sub-model is used to simulate the influence of bearing frictional heat generation and nacelle ambient temperature distribution; the lubrication sub-model is used to simulate bearing oil film formation and failure; the electromagnetic interference sub-model is used to simulate the influence of generator shaft voltage and shaft current on bearings; and the fault evolution sub-model is used to simulate the dynamic development of common failure modes of wind turbine bearings.
4. The data-independent collaborative optimization control method for wind turbine bearings as described in claim 1, characterized in that, The construction of the physical information security reinforcement learning agent includes: The design incorporates a multi-dimensional coupled state space, encompassing five dimensions: mechanical, thermal, lubrication, electromagnetic, and fault. Specifically, it includes: vibration time-domain / frequency-domain characteristics, temperature of various bearing components, radial / axial loads, oil film parameters, electromagnetic interference intensity, fault severity, and health status, forming a complete state vector adapted to the operating characteristics of wind turbine bearings. The design incorporates a multi-dimensional collaborative action space, including five dimensions: dynamic load distribution, adaptive adjustment of lubrication parameters, intelligent control of the cooling system, fine-tuning of rotation speed, and dynamic generation of maintenance decisions. The action range is dynamically adjusted based on the physical characteristics of the wind turbine bearings, rather than being controlled by fixed parameters. An adaptive multi-objective reward function is constructed, which comprehensively considers four dimensions: safety, economy, operation performance and lifespan. The weight coefficients are dynamically adjusted according to the real-time operating status of the wind turbine bearings, rather than fixed weights, to guide the agent to learn the optimal collaborative strategy that adapts to the operating conditions of the wind turbine bearings. A dynamic safety assessment rule is constructed, and a deep neural network is used to train the assessment model. The real-time state vector of the wind turbine bearing is input, and the safety probability of each safety dimension is output. This is different from the existing fixed threshold safety judgment and realizes dynamic safety verification that is adapted to the operating characteristics of wind turbine bearings. A physical information security layer is constructed, and a quadratic programming algorithm is used to perform security correction on the original control actions generated by the intelligent agent. The core is to satisfy all physical safety constraints of the wind turbine bearings while ensuring that the difference between the control actions and the original actions is minimized. The agent is trained in a hierarchical manner. Based on the simulation environment provided by the multiphysics coupling simulation module, which is adapted to the wind turbine bearing, the basic control strategy is first trained under simplified working conditions. The complexity of multiphysics coupling, fault scenarios and electromagnetic interference are gradually increased. After training, the agent is directly deployed to the actual control system of the wind turbine bearing.
5. The data-independent collaborative optimization control method for wind turbine bearings as described in claim 1, characterized in that, The construction of the multi-timescale collaborative control module includes: The system divides the time into four independent and coordinated time levels, clarifies the core tasks and decision-making frequency of each level, distinguishes itself from the existing single time scale control, and adapts to the operating characteristics of wind turbine bearings. Establish an inter-layer collaboration mechanism to achieve two-way information transmission, time scale matching, and conflict resolution; Layered learning and optimization: each layer of intelligent agent independently learns the control strategy corresponding to the time scale, while achieving overall collaborative optimization through inter-layer information interaction. This ensures that short-term control does not affect long-term lifespan, and long-term decision-making does not ignore short-term safety, thus adapting to the operational needs of wind turbine bearings throughout their entire life cycle.
6. The data-independent collaborative optimization control method for wind turbine bearings as described in claim 5, characterized in that, The division into four independent and coordinated time levels includes: Millisecond-level ultra-short-term layer: responsible for real-time vibration suppression and electromagnetic interference buffering, with high decision frequency and rapid response to safety risks caused by instantaneous gusts and sudden electromagnetic interference from wind turbines; The short-term layer, measured in seconds, is responsible for dynamic load distribution, real-time adjustment of lubrication and cooling parameters, and capturing dynamic changes in the operating status of wind turbine bearings. The minute-level mid-term layer is responsible for early fault warning, operating parameter optimization, identifying early fault characteristics of wind turbine bearings, and adjusting control strategies. Long-term layer (hours and above): Responsible for dynamic generation of maintenance strategies and dynamic lifespan management, and for developing optimal maintenance plans adapted to wind power operation and maintenance scenarios based on the cumulative damage of wind turbine bearings.
7. The data-independent collaborative optimization control method for wind turbine bearings as described in claim 5, characterized in that, The establishment of an inter-layer collaboration mechanism to achieve two-way information transmission, time scale matching, and conflict resolution includes: Information transmission: High-level decisions serve as constraints for low-level controls, while the real-time status of low-level controls serves as the basis for high-level decisions. Time scale matching: Through a buffering mechanism, the control actions at different time levels are ensured to be coordinated and consistent, avoiding decision conflicts and adapting to the dynamic operating conditions of wind turbine bearings. Conflict resolution: Following the priorities of safety first, performance second, lifespan best, and cost lowest, resolve decision conflicts at different levels to meet the operation and maintenance needs of wind turbines.
8. The data-independent collaborative optimization control method for wind turbine bearings as described in claim 1, characterized in that, The construction of the multi-source fusion closed-loop feedback module includes: A multi-source sensor acquisition system was built, deploying triaxial vibration sensors, multi-point temperature sensors, load sensors, speed sensors, lubrication sensors, and electromagnetic interference sensors to collect real-time operating status data of wind turbine bearings, adapting to the monitoring environment of wind power sites. The real data collected by the sensor is bidirectionally calibrated with the simulation data output by the multiphysics coupling simulation module that is adapted to the wind turbine bearing. This corrects the simulation model parameters, improves the simulation accuracy, and makes up for the limitations of single real data or simulation data. A smart actuator is built to receive control actions output by a safety reinforcement learning agent, and to perform load adjustment, lubrication control, and cooling control operations. The execution process is fed back to the feedback module in real time, adapting to the control requirements of wind turbine bearings. Collect wind turbine bearing status data after executing control actions, calculate reward function values, and feed them back to the safety reinforcement learning agent for online learning and policy iteration. At the same time, feed the status data back to the multiphysics coupling simulation module to achieve dynamic model optimization and adapt to the dynamic changes in wind power conditions. A tiered early warning and emergency response mechanism is set up. When the security probability output by the security assessment rule of the intelligent agent is lower than the preset threshold, the corresponding level of early warning signal is immediately triggered, and emergency control actions are initiated to prevent the fault from escalating. Maintenance suggestions adapted to the wind turbine operation and maintenance scenario are generated simultaneously.
9. The data-independent collaborative optimization control method for wind turbine bearings as described in claim 1, characterized in that, It also includes training a model for the physical information security reinforcement learning agent, wherein the model training includes: Start the multiphysics coupling simulation module, input the basic parameters of the wind turbine bearing, initialize the five-dimensional multiphysics coupling model, set various operating conditions, fault scenarios and electromagnetic interference scenarios to adapt to wind power scenarios, and generate simulation state data; Initialize the physical information security reinforcement learning agent, set up a multi-dimensional coupled state space, cooperative action space, adaptive multi-objective reward function and dynamic security constraint parameters adapted to wind turbine bearings, and train dynamic security assessment and security layer optimization algorithms; A hierarchical training strategy is adopted to train the agent in a simulated environment, gradually increasing the complexity of multi-physics coupling, fault scenarios and electromagnetic interference intensity, and optimizing the control strategy until the agent can stably generate cooperative control actions that meet safety constraints, take into account multi-objective optimization and adapt to the operating conditions of wind turbine bearings. Through the multi-timescale collaborative control module, the agent strategies at each time level are collaboratively optimized to ensure that the control tasks at different time scales are coordinated and consistent, forming a complete collaborative control strategy adapted to wind turbine bearings. After training is completed, save the parameters of the intelligent agent model and the multiphysics coupling simulation module model to complete the training phase.
10. The data-independent collaborative optimization control method for wind turbine bearings as described in claim 1, characterized in that, The aforementioned collaborative optimization control of wind turbine bearings using the constructed integrated architecture includes: Step 1: Deploy the integrated architecture to the wind turbine bearing control system, start the sensor acquisition system and actuator of the multi-source fusion closed-loop feedback module, and initialize the wind turbine bearing parameters and multi-physics coupling simulation module; Step 2: The multiphysics coupling simulation module generates simulation data without data dependence. The sensor acquisition system collects the operating status data of the wind turbine bearing in real time and transmits it to the multi-source fusion closed-loop feedback module. The simulation data of the multiphysics coupling simulation module is bidirectionally calibrated and outputs an accurate real-time bearing status vector. Step 3: The physical information security reinforcement learning agent receives the state vector and judges whether the wind turbine bearing is safe through dynamic security assessment rules. If it is safe, it generates the original cooperative control action and outputs it. If it is not safe, it corrects the cooperative control action through the physical information security layer before outputting it. Step 4: The multi-timescale collaborative control module coordinates the control actions output by the intelligent agent in a hierarchical manner, and sends the control actions of each time scale to the actuator synchronously to achieve collaborative control and adapt to the dynamic operating conditions of the wind turbine. Step 5: The actuator performs control actions, while the sensor acquisition system collects the state data after execution, calculates the reward function value, and feeds it back to the agent and the multiphysics coupling simulation module to realize online learning, strategy iteration and simulation model optimization of the agent, and adapt to the dynamic changes of wind power conditions; Step 6: If an abnormal operation of the wind turbine bearing is detected, immediately trigger the corresponding level of early warning signal, initiate emergency control actions, and generate dynamic maintenance suggestions adapted to the wind power operation and maintenance scenario to guide operation and maintenance personnel to perform precise maintenance. Step 7: Repeat steps 2-6 to achieve coordinated optimization control of the wind turbine bearing throughout its entire life cycle, until the bearing reaches its design life or is replaced.
