A coordinated control system for wind turbine load reduction and opportunistic maintenance under maintenance window constraints
By using the state perception and physical motion mask generation module to determine the feasibility of wind turbine operation and maintenance actions, and the collaborative decision-making module to select joint actions, the problem of operation control and maintenance under limited maintenance windows is solved. This enables load reduction risk buffering and multi-component simultaneous maintenance, forming a closed-loop feedback and improving the operational reliability and maintenance efficiency of wind turbines.
Patent Information
- Application Number
- CN202611620875.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
Under conditions of limited maintenance windows, existing technologies struggle to accurately determine the feasibility of operational control actions and maintenance actions, lack a risk buffering mechanism for load reduction during maintenance waiting periods, and are unable to dynamically form maintenance combinations within the same window based on the real-time status of multiple components, and lack feedback updates on execution results.
A unified state vector is obtained by a state awareness module, the feasibility of actions is judged by a physical action mask generation module, the final joint action is selected by a collaborative decision-making module, and control commands are output by an instruction execution module. Collaborative control is performed by combining component health status and maintenance window status.
This enables the avoidance of unexecutable operation and maintenance actions under limited maintenance windows, slows down component degradation, reduces repeated downtime, and forms a closed loop of operation control, maintenance execution, and status feedback, thereby improving the operational reliability and maintenance efficiency of wind turbine units.
Smart Images

Figure CN123109939A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wind turbine generator operation control and intelligent operation and maintenance technology, and more specifically to a wind turbine generator load reduction and opportunistic maintenance collaborative control system under maintenance window constraints. Background Technology
[0002] With the continuous expansion of installed wind power capacity and the increasing capacity of individual wind turbine units, the operational reliability, power generation efficiency, and maintenance economy of wind turbine units have gradually become important factors affecting the long-term stable operation of wind farms. Wind turbine units operate under the combined influence of random wind speeds, alternating loads, and complex weather conditions, leading to varying degrees of degradation in critical components such as generators, gearboxes, main bearings, and pitch systems. Failures in these critical components can result in unplanned downtime, power generation losses, and increased maintenance costs. Therefore, it is necessary to coordinate wind turbine operation control and maintenance based on wind conditions, operating loads, and component health status.
[0003] Existing wind turbine maintenance methods mainly include corrective maintenance, preventive maintenance, condition-based maintenance, and opportunistic maintenance. Corrective maintenance is usually implemented after a component fails, which can easily lead to prolonged unplanned downtime. Preventive maintenance is mainly performed according to a preset cycle, which may result in premature or delayed maintenance. Condition-based maintenance determines the timing of maintenance based on component health indicators or reliability, but its decision-making effectiveness is affected by the accuracy of condition assessment and threshold settings.
[0004] Opportunistic maintenance typically utilizes periods of low wind speed, existing downtime, or other available work windows to schedule maintenance tasks for multiple components within the same window, reducing redundant downtime and work. However, wind turbine maintenance is constrained by factors such as average wind speed, maximum gusts, turbine shutdown status, maintenance resources, and work plans. When critical components require maintenance but current conditions do not meet the requirements, maintenance tasks must be postponed. During the maintenance waiting period, if the wind turbine continues to operate at its original power level, critical components will still be subjected to continuous loads and abnormal wind conditions, potentially leading to a further decline in component health.
[0005] Power derating can reduce the output power of wind turbines and their equivalent load by adjusting power setpoints, generator torque, or pitch control parameters. However, in some existing systems, power derating control and maintenance decisions are independent. Derating actions are mainly determined by wind conditions or power demand, while maintenance actions are mainly determined by health thresholds or maintenance plans. There is a lack of coordination mechanism between the two for the maintenance waiting process. Therefore, the system struggles to dynamically determine whether to drate and the level of derating based on component risk and maintenance window status.
[0006] Furthermore, existing opportunistic maintenance solutions typically use fixed thresholds, preset priorities, or manual rules to determine maintenance targets. When there are differences in the degradation rate, maintenance intensity, and maintenance duration of multiple components, fixed rules may struggle to simultaneously consider the components with primary maintenance needs, potential opportunistic maintenance components, and the task combinations within the same downtime window, resulting in scattered maintenance tasks or repeated downtime.
[0007] Some existing intelligent operation and maintenance systems use reinforcement learning or other optimization models to generate operation and maintenance decisions. However, in the joint action space formed by wind turbine operation control and multi-component maintenance, some actions are limited by wind conditions, maintenance occupancy status, remaining downtime, and maintenance window status. If the decision model does not explicitly handle these physical constraints, it may generate operation control actions or maintenance actions that cannot be executed under the current conditions. At the same time, some solutions lack a closed-loop process from joint decision-making and command issuance to feedback of actual execution results and status updates.
[0008] Therefore, existing technologies still have the following problems: when the maintenance window is limited, it is difficult to accurately determine the feasibility of operation control actions and maintenance actions; when maintenance needs have arisen but maintenance is temporarily not feasible, there is a lack of a mechanism to implement load reduction risk buffering based on component health status and load level; after maintenance conditions are met, it is difficult to dynamically form a maintenance combination for the same window based on the real-time status of multiple components and continuously update the status of wind turbine units based on actual execution results.
[0009] Therefore, how to achieve maintenance feasibility judgment, graded load reduction risk buffering during maintenance waiting period, multi-component simultaneous maintenance, and execution result feedback and update are problems that urgently need to be solved by those skilled in the art. Summary of the Invention
[0010] In view of the above problems, the present invention proposes a wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints, which aims to overcome or at least partially solve the above problems.
[0011] To achieve the above objectives, the present invention adopts the following technical solution:
[0012] This invention provides a coordinated control system for wind turbine load shedding and opportunistic maintenance under maintenance window constraints, comprising: The state awareness module is used to acquire a unified state vector and identify components with major maintenance needs and components with potential maintenance opportunities. The unified state vector includes the wind turbine operating state vector for the current decision cycle, the health state vectors of each key component, and the maintenance window availability state vector. The physical action mask generation module is used to encode the current physical operation constraints into a physical action mask based on the wind turbine operating state vector and the maintenance window available state vector, combined with the main maintenance requirement components and opportunity maintenance candidate components; the physical action mask is used to mark whether each joint action in the joint action space, including operation control actions and maintenance actions, can be executed at the current moment. The collaborative decision-making module is used to filter out the final joint action from the candidate joint actions marked as executable by the physical action mask according to the unified state vector and the physical action mask. The final joint action includes the final operation control action, the final maintenance intensity, and the final repair component combination. The instruction execution module is used to output corresponding control instructions based on the final combined action.
[0013] Furthermore, the key components include a generator, gearbox, main bearing, and pitch system.
[0014] Furthermore, the state-aware module includes: The data acquisition unit is used to acquire at least the operating mode, start / stop status, average wind speed, maximum gust, maintenance occupancy status, remaining downtime and maintenance window status of the wind turbine during operation. The load status assessment unit is used to determine the average load factor and abnormal fluctuation driving quantity of the wind turbine based on the average wind speed and maximum gust during the current decision cycle, combined with the average actual output power and rated power. The component health assessment unit is used to determine the equivalent aging increment of each key component in the current decision cycle based on the average load factor and abnormal fluctuation driving amount; update the equivalent service time of each key component based on the equivalent aging increment; and determine the reliability of each key component based on the updated equivalent service time. The overall machine health assessment unit is used to determine the overall reliability of the wind turbine based on the reliability of each key component, determine the minimum value among the reliability of multiple key components as the bottleneck component reliability, and determine the overall machine health level based on the overall machine reliability. The maintenance requirement identification unit is used to identify the main maintenance requirement components and the opportunity maintenance candidate components based on the reliability of each key component and the preset main maintenance reference boundary and opportunity maintenance candidate boundary. The state vector generation unit is used to normalize and fuse the operating mode, start / stop status, average wind speed, maximum gust, average load factor, abnormal fluctuation driving quantity, reliability vector of multiple key components, overall machine reliability, bottleneck component reliability, overall machine health level, maintenance occupancy status, remaining downtime, maintenance window status, set of main maintenance requirement components and set of opportunity maintenance candidate components to generate a unified state vector for the current decision cycle.
[0015] Furthermore, the physical action mask generation module specifically includes: The maintenance execution condition judgment unit is used to determine whether the wind turbine meets the maintenance operation conditions based on the wind turbine operating state vector and the maintenance window available state vector, and to generate a maintenance execution identifier. The operation control action filtering unit is used to determine the feasibility of operation control actions based on the wind turbine's operating state vector. The operation control actions include rated operation actions, slight load reduction actions, and deep load reduction actions. The maintenance action screening unit is used to determine the feasibility of maintenance actions based on maintenance executability identifiers, a set of main maintenance requirement components, and a set of opportunity maintenance candidate components; the maintenance action is determined by the maintenance intensity and the combination of repair components, and the maintenance intensity includes no maintenance, minor repair, and major repair; The risk buffer action constraint unit is used to determine the operational risk during the maintenance waiting period based on the reliability of bottleneck components, average load factor, and abnormal fluctuation driving quantity when the set of components requiring major maintenance is not empty and maintenance actions are not executable. The joint action mask encoding unit is used to set an executable or non-executable flag for each joint action in the joint action space composed of operation control actions, maintenance intensity and maintenance components, and generate a physical action mask corresponding to the current decision cycle.
[0016] Furthermore, the operation control action filtering unit is also used to: shield the rated operation action, the slight load reduction action, and the deep load reduction action when the wind turbine is in a maintenance occupancy state or the remaining downtime is greater than zero.
[0017] Furthermore, the maintenance action filtering unit is also used to: when the maintenance executable identifier is zero, or when the maintenance executable identifier is 1 and the main maintenance requirement component set is empty, to block minor repairs, major repairs and all repair component combinations; and when the maintenance executable identifier is 1 and the main maintenance requirement component set is not empty, to determine at least one repairable component combination from the main maintenance requirement component set and the opportunity maintenance candidate component set.
[0018] Furthermore, the risk buffer action constraint unit is also used to: when the operational risk reaches the preset risk buffer condition, shield the rated operational action, and retain at least one of the mild load reduction action and the deep load reduction action.
[0019] Furthermore, the collaborative decision-making module specifically includes: The state feature encoding unit is used to extract and fuse features from the unified state vector of the current decision cycle to generate shared state features; The operation control decision unit is used to determine the operation control candidate scores corresponding to rated operation, light load reduction and deep load reduction respectively based on the shared state characteristics; The maintenance decision unit is used to determine the maintenance intensity candidate score and the repair component combination candidate score based on the shared state characteristics, the set of main maintenance requirement components, and the set of opportunity maintenance candidate components, respectively. The mask strategy generation unit is used to combine the candidate scores of operation control, maintenance intensity, and repair component combination into a joint action score, and to filter out executable candidate joint actions from the joint action space according to the physical action mask, and to calculate the probability distribution of the candidate joint actions based on the joint action score. The joint action determination unit is used to determine the final operation control action, the final maintenance intensity, and the final repair component combination based on the probability distribution of the candidate joint actions, and combine them into the final joint action for the current decision cycle; when the maintenance intensity is minor repair or major repair, the operation control action is determined to be maintenance shutdown.
[0020] Furthermore, the instruction execution module specifically includes: The joint action parsing unit is used to receive the final joint action and parse the final operation control action, final maintenance intensity, and final repair component combination from the final joint action; An operation control instruction generation unit is used to generate corresponding power control instructions based on the final operation control action; A maintenance instruction generation unit is used to generate maintenance instructions based on the final maintenance intensity and the final repair component combination; The multi-component simultaneous maintenance task arrangement unit is used to arrange multiple maintenance tasks in the final maintenance component combination within the same maintenance window, and generate a shutdown command and multiple component maintenance tasks. The safety interlock and scheduling unit is used to perform consistency verification on the instruction before it is issued, based on the current physical action mask, maintenance executable identifier, wind turbine start / stop status, and remaining downtime; when the maintenance instruction conflicts with the power control instruction, the power control instruction is masked and the maintenance instruction is executed first.
[0021] Furthermore, it also includes a status feedback update module, which is used to update the wind turbine operating status vector, the health status vector of each key component, and the maintenance window availability status vector according to the execution feedback data of the control command, and use the updated status as the input for the next decision cycle.
[0022] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints, which has the following beneficial effects: 1. This invention organically combines critical component health status assessment, maintenance feasibility constraints, load reduction risk buffering during maintenance waiting periods, and multi-component simultaneous maintenance. It avoids intelligent decision-making models outputting operation and maintenance actions that are impossible to execute under current physical conditions, slows component degradation during maintenance waiting periods, reduces repeated downtime caused by asynchronous maintenance of multiple components, and ensures that subsequent decisions reflect actual load reduction effects and maintenance recovery results. This system can be deployed on wind turbine edge controllers, wind farm monitoring servers, or operation and maintenance scheduling platforms, providing technical support for long-term operation control and intelligent maintenance of wind turbines.
[0023] 2. This invention uses physical action masks to constrain the executability of operation control actions, maintenance intensity, and repair component combinations. It can shield unexecutable actions based on wind conditions, shutdown status, maintenance occupancy status, remaining downtime, and maintenance window status, reducing the inconsistency between decision results and actual operation and maintenance conditions.
[0024] 3. When critical components have generated maintenance needs but maintenance actions are temporarily not feasible, this invention implements graded load reduction risk buffering based on the reliability of bottleneck components, average load factor, and abnormal fluctuation driving quantity, thereby reducing the output power and the equivalent aging increment of components during the maintenance waiting period and slowing down the further accumulation of component risks.
[0025] 4. When maintenance conditions are met, the present invention arranges at least one major maintenance-required component and zero or at least one candidate component for maintenance within the same downtime window for repair, and updates component reliability, downtime status and maintenance window status based on the actual execution results of operation control and maintenance tasks, thereby reducing repeated downtime and forming a closed loop of operation control, maintenance execution and status feedback updates. Attached Figure Description
[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0027] Figure 1 This is a schematic diagram of the overall framework of the wind turbine load reduction and opportunistic maintenance collaborative control system under the condition of limited maintenance window provided in this embodiment of the invention; Figure 2This is a schematic diagram illustrating the changes in the overall reliability of the wind turbine and the reliability of key components over service time, as provided in this embodiment of the invention. Figure 3 This is a schematic diagram of the collaborative control process for wind turbine load reduction and multi-component opportunistic maintenance under the condition of limited maintenance window provided in the embodiments of the present invention; Figure 4 This is a schematic diagram of the process for implementing a graded load reduction risk buffer when the physical action mask generation and maintenance actions are not executable, as provided in this embodiment of the invention. Figure 5 This is a schematic diagram of the structure of the FAM-DC-PPO collaborative decision-making model provided in this embodiment of the invention; Figure 6 This is a schematic diagram illustrating the coordinated changes in load control, maintenance execution, and overall machine health status under the limited conditions of the maintenance window provided in this embodiment of the invention. Detailed Implementation
[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0029] This invention discloses a coordinated control system for load shedding and opportunistic maintenance of wind turbines under maintenance window constraints, such as... Figure 1 As shown, it includes the following modules: The state awareness module is used to acquire a unified state vector and identify components with major maintenance needs and components with opportunities for maintenance. The unified state vector includes the wind turbine operating state vector for the current decision cycle, the health state vectors of each key component, and the state vector of available maintenance windows. The physical action mask generation module is used to encode the current physical operation constraints into a physical action mask based on the wind turbine's operating state vector and the maintenance window's available state vector, combined with the main maintenance requirement components and opportunity maintenance candidate components. The physical action mask is used to mark whether each joint action in the joint action space, including operation control actions and maintenance actions, can be executed at the current moment. The collaborative decision-making module is used to select the final joint action from the candidate joint actions marked as executable by the physical action mask based on the unified state vector and the physical action mask. The final joint action includes the final operation control action, the final maintenance intensity, and the final repair component combination. The instruction execution module is used to output corresponding control instructions based on the final combined action.
[0030] Next, each of the above modules will be explained in detail.
[0031] 1. State Awareness Module: See Figure 1 The status awareness module is used to acquire the current operating status of the wind turbine, the health status of its key components, and the availability of maintenance windows. This status awareness module includes a data acquisition unit, a load status assessment unit, a component health assessment unit, a whole-machine health assessment unit, a maintenance requirement identification unit, and a status vector generation unit; specifically: (1) Data acquisition unit: The data acquisition unit is used to acquire wind condition data, operational status data, key component status data, and maintenance status data during the operation of the wind turbine. It also performs time synchronization and periodic processing on the acquired data to form the input data required by the status awareness module. This data acquisition unit includes the following sub-units: 1) Wind condition data acquisition subunit: The wind condition data acquisition subunit is used to acquire instantaneous wind speed, gust wind speed and environmental condition data at the location of the wind turbine through wind speed measurement devices, meteorological monitoring devices or wind farm monitoring and data acquisition systems.
[0032] Instantaneous wind speed is used to characterize the basic wind conditions input received by the wind turbine at the corresponding sampling time; gust wind speed is used to characterize the short-term peak wind speed and serves as input for judging the safety conditions of maintenance operations and assessing abnormal fluctuations; environmental status data includes at least one of ambient temperature, air humidity, and air pressure.
[0033] 2) Run the data acquisition subunit: The operation data acquisition subunit is used to obtain actual output power, target power setpoint, rotor speed, start / stop status, current operating mode, fault alarm information and control command execution status through the wind turbine monitoring and data acquisition system.
[0034] Among them, the actual output power is used to characterize the real power output under the combined effect of wind conditions and control actions; the target power setpoint is used to characterize the power level required by the wind turbine controller; the rotor speed is used to assist in judging the wind turbine's operating status and power response process; the current operating mode includes at least the rated operating mode, the light load reduction mode, and the deep load reduction mode; and the control command execution status is used to determine whether the operating control commands generated in the previous decision cycle have been executed.
[0035] 3) Component status acquisition subunit: The component status acquisition subunit is used to acquire status monitoring data for multiple key components. Key components include at least the generator, gearbox, main bearing, and pitch system; the status monitoring data includes at least one of the following: vibration data, temperature data, current data, lubrication status data, cumulative operating time, component alarm information, and historical maintenance information.
[0036] The component status acquisition subunit associates the status monitoring data of different key components according to the component identification and acquisition time, so that each key component has a status record corresponding to the current sampling time, providing a data foundation for subsequent updates of the equivalent service time of key components, reliability calculation and maintenance requirement identification.
[0037] 4) Maintenance Status Acquisition Subunit: The maintenance status acquisition subunit obtains maintenance window status, maintenance occupancy status, remaining downtime, actual maintenance parts, actual maintenance intensity, maintenance task progress, and maintenance completion feedback information through the operation and maintenance management terminal or maintenance scheduling platform.
[0038] Among them, the maintenance window status is used to indicate whether there is a time window that meets the requirements of maintenance work; the maintenance occupancy status is used to indicate whether the wind turbine is performing maintenance tasks; the remaining downtime is used to limit the feasibility of operation control actions and new maintenance actions within the current decision cycle; and the maintenance completion feedback information is used to determine whether the corresponding key components have entered the maintenance recovery and update process.
[0039] 5) Data preprocessing subunit: The data preprocessing subunit is connected to the wind condition data acquisition subunit, the operation data acquisition subunit, the component status acquisition subunit, and the maintenance status acquisition subunit, respectively, and is used to sequentially perform timestamp alignment, physical validity verification, outlier handling, missing value handling, decision cycle aggregation, and unified data object generation.
[0040] During the physical validity verification process: when the data to be verified is within the corresponding physical allowable range, the data is retained; when the data to be verified exceeds the corresponding physical allowable range, it is marked as invalid data. For continuous data that is missing for a short period of time, it is supplemented by interpolation of adjacent valid data; for discrete data such as operating mode, start / stop status, maintenance occupancy status, and maintenance window status, it is supplemented by maintaining the most recent valid status. When the proportion of valid data in a decision cycle is lower than the preset integrity threshold, the decision cycle is marked as a data abnormal cycle, and the generation of new operation and maintenance control instructions is stopped, or the safe operating status corresponding to the previous valid decision cycle is maintained. After the validity verification is completed, the data preprocessing subunit aggregates the sampled data into decision cycle-level data.
[0041] In actual data usage, the operating mode adopts the most frequently occurring effective operating mode within the current decision-making cycle; the start / stop status, maintenance occupancy status, remaining downtime, and maintenance window status adopt the latest effective status at the end of the current decision-making cycle; and the status data of key components are aggregated according to the data type using the average value, maximum value, change range, or latest effective value.
[0042] Through the above preprocessing, wind condition data, operational status data, key component status data, and maintenance status data from different sources and with different sampling frequencies are converted into input data with a unified time base and data format, providing a data foundation for subsequent load status calculations, key component reliability updates, overall machine health assessments, and maintenance requirement identification.
[0043] (2) Load condition assessment unit: The load condition assessment unit is used to determine the average load factor and abnormal fluctuation driving quantity of the wind turbine based on the average wind speed, maximum gust, average actual output power and rated power in the current decision cycle. The average loading factor for the current decision cycle is expressed as:
[0044] in, This represents the average loading factor of the current decision-making cycle. This represents the average actual output power. This indicates the rated power of the wind turbine generator; the average load factor It is a dimensionless parameter used to characterize the average power load level of the wind turbine relative to its rated operating state within the current decision-making period.
[0045] The abnormal fluctuation driver in the current decision-making cycle is represented as:
[0046] in, This indicates the driving force behind abnormal fluctuations in the current decision-making cycle. This indicates the maximum gust of wind in the current decision-making cycle. This represents the average wind speed during the current decision-making cycle. This indicates the permissible deviation of gusts from the threshold. This represents the normalized scale of abnormal fluctuations, and Greater than 0, where 0 is the minimum output value when the abnormal fluctuation is not activated; this abnormal fluctuation driving quantity Used to characterize maximum gust relative to average wind speed The additional impact of abnormal deviations on the operating status of critical components; when the maximum gust of wind... With average wind speed When the difference does not exceed the allowable gust deviation threshold, the abnormal fluctuation driving force The value is set to zero; when the difference exceeds the allowable gust deviation threshold, the abnormal fluctuation driving force is normalized according to the degree of excess. This allows normal wind speed fluctuations to be primarily characterized by the average load factor, while significant gust fluctuations serve as an additional risk input in updating component health status.
[0047] (3) Component health assessment unit: The component health assessment unit is connected to the load condition assessment unit and is used to determine the equivalent aging increment of each key component in the current decision cycle based on the average load factor and abnormal fluctuation driving amount; update the equivalent service time of each key component based on the equivalent aging increment, and determine the reliability of each key component based on the updated equivalent service time. In this embodiment, key components are vectorized according to a fixed order: generator, gearbox, main bearing, and pitch system. The equivalent aging increment vector of each key component is represented as follows:
[0048] in, This represents the equivalent aging increment vector of multiple key components within the current decision-making cycle; This represents a vector of basic aging coefficients for multiple key components; This represents a unit vector where all elements are one. This represents a vector of load acceleration coefficients for multiple key components. This represents a vector of acceleration coefficients for abnormal fluctuations in multiple key components. Indicates the average loading factor. This represents element-wise multiplication between vector elements at the same position.
[0049] The larger the average load factor, the greater the equivalent aging increment of each key component; when the abnormal fluctuation driving force is activated, the equivalent aging increment of the corresponding key component further increases. Different basic aging coefficients, load acceleration coefficients, and abnormal fluctuation acceleration coefficients are used for different key components to characterize the heterogeneous degradation characteristics of multiple components. The updated equivalent service time vector of each key component is expressed as follows:
[0050] in, This represents the updated equivalent service time vector of each key component at the end of the current decision-making cycle; This represents the equivalent service time vector of each key component at the start of the current decision-making cycle; This represents the equivalent aging increment vector for the current decision-making cycle. Based on the updated equivalent service time, the reliability vectors of each key component are determined using the Weibull reliability model:
[0051] in, This represents the reliability vector of multiple key components at the end of the current decision-making cycle; This represents a Weibull scale parameter vector for multiple key components, used to characterize the lifespan of each key component; The Weibull shape parameter vector represents multiple key components and is used to characterize the variation of the failure rate of each key component over service time. Represents an exponential function with the natural constant as its base; This represents element-wise division between vector elements at the same position. This means that element-wise exponentiation is performed on each key component using the element at the corresponding position in the shape parameter vector as the exponent.
[0052] The values of each element in the reliability vector range from 0 to 1, with larger values indicating better health of the corresponding critical component. The reliability vector is then sent to the overall system health assessment unit and the maintenance requirement identification unit.
[0053] (4) Overall machine health assessment unit: The overall turbine health assessment unit is used to determine the overall turbine reliability and bottleneck component reliability based on the reliability of multiple critical components. Critical components of the wind turbine are treated as a series system; failure of any critical component will affect the normal operation of the wind turbine. See also... Figure 2 In the entire life cycle operation of this embodiment, the reliability of each key component changes with the equivalent service time and recovers accordingly after maintenance is implemented; the overall reliability evolves with the changes in the reliability of each key component, thereby characterizing the impact of key component degradation and maintenance recovery on the overall health status of the machine.
[0054] Overall machine reliability is expressed as:
[0055] in, This indicates the overall system reliability during the current decision-making cycle; Indicates the first Reliability of key components Assign part numbers; This represents the number of critical components. Bottleneck component reliability characterizes the critical component with the worst current health status, and is expressed as:
[0056] in, Indicates the reliability of bottleneck components in the current decision-making cycle; This indicates selecting the minimum value from all elements of the reliability vector. Bottleneck component reliability characterizes the critical component with the worst current health status, while overall system reliability characterizes the system-level operational risk under the combined effect of multiple critical components.
[0057] Furthermore, the overall health level of the machine is expressed as follows:
[0058] in, Indicates the overall health level of the machine; This indicates the maintenance of early warning boundaries; Indicates a high-risk boundary; Denotes the failure boundary, and satisfies Health level Indicates normal condition, health level Indicates a warning status and health level. Indicates a high-risk status, health level This indicates a failure status.
[0059] Health levels are used for status description and risk feedback, and do not directly generate maintenance instructions according to fixed rules.
[0060] (5) Maintenance requirement identification unit: The maintenance requirement identification unit is connected to the component health assessment unit and the whole machine health assessment unit respectively, and is used to identify the components with major maintenance requirements and the components with potential for opportunity maintenance based on the reliability of each key component, the preset major maintenance reference boundary and the preset opportunity maintenance candidate boundary.
[0061] The set of components requiring primary maintenance is represented as follows:
[0062] The set of candidate parts for opportunity maintenance is represented as:
[0063] in, This represents the set of components requiring primary maintenance. This indicates the opportunity to maintain a set of candidate components; This refers to the collection of key components consisting of a generator, gearbox, main bearing, and pitch system. Represents the set of key components Any key component in it; Indicates key components Corresponding reliability; This indicates that the primary maintenance reference boundary is being maintained; This indicates that the opportunity maintains the candidate boundary and satisfies .
[0064] Main maintenance requirement component set Used to record critical components whose reliability has entered the major maintenance zone; a set of candidate components for opportunity maintenance. This is used to record critical components that have not yet entered the main maintenance area, but whose health status is close to that of the maintenance area and have the value for early collaborative repair. The two sets mentioned above are only used to limit the scope of repair candidates and do not directly determine whether to perform repairs or the number of components actually repaired.
[0065] (6) State vector generation unit: The state vector generation unit is connected to the load state assessment unit, component health assessment unit, overall machine health assessment unit, and maintenance requirement identification unit, respectively; the state vector generation unit will generate the operating mode within the current decision cycle. Start-stop status Average wind speed Maximum gust Average loading factor and abnormal fluctuation driving force This data serves as the corresponding operating status of the wind turbine; it also represents the reliability vectors of multiple key components. Overall machine reliability Reliability of bottleneck components And overall machine health level Data corresponding to the health status of key components; maintenance occupancy status. Remaining downtime Maintain window status Main maintenance requirement components and the opportunity to maintain the candidate component set As the data corresponding to the maintenance state, the above-mentioned data types are normalized and combined to generate a unified state vector corresponding to the current decision cycle.
[0066] The unified state vector is represented as follows:
[0067] in, A unified state vector representing the current decision-making cycle; This represents a normalization operation, used to map continuous data of different dimensions to a preset numerical range; This means combining the state variables into a state vector according to a predetermined order; This indicates a set-indicator encoding operation used to convert the component affiliation relationships in a component set into a binary vector. This represents the set of components requiring primary maintenance. The corresponding binary code; This indicates the opportunity to maintain the candidate component set. The corresponding binary code.
[0068] For any critical component, when it belongs to the corresponding set, the corresponding position in the set indicator code is set to 1; when it does not belong to the corresponding set, the corresponding position is set to 0. The state vector generation unit will unify the state vector. Main maintenance requirement components and opportunity maintenance candidate component set Send to the physical action mask generation module and the collaborative decision-making module.
[0069] Through the above-mentioned status assessment process, the original wind conditions, power, component monitoring and maintenance status data are converted into unified status information that can reflect continuous load, abnormal fluctuations, heterogeneous degradation of components, overall operation risks and maintenance needs of multiple components, providing a status basis for subsequent maintenance feasibility judgment, physical action mask generation and joint operation and maintenance decision-making.
[0070] 2. Physical Action Mask Generation Module: See Figure 1 , Figure 3 and Figure 4 The physical action mask generation module is connected to both the state perception module and the collaborative decision-making module, and is used to generate a unified state vector based on the current decision cycle. The system determines whether each operational control action, maintenance intensity, and repair component combination meets the actual execution conditions and masks out unexecutable combined actions. The physical action mask generation module includes a maintenance execution condition judgment unit, an operational control action filtering unit, a maintenance action filtering unit, a risk buffer action constraint unit, and a combined action mask encoding unit.
[0071] (1) Maintain execution condition judgment unit: The maintenance execution condition judgment unit is used to unify the state vector according to the current decision cycle. Average wind speed in Maximum gust Maintenance Occupancy Status Remaining downtime and maintain window status Determine whether the wind turbine meets the conditions for maintenance work and generate a maintenance execution flag. :
[0072] in, Indicates maintenance of executable identifiers; This indicates that the current decision-making cycle allows for the initiation of a new maintenance task. This indicates that starting a new maintenance task is not allowed in the current decision-making cycle; This represents the average wind speed during the current decision-making cycle; This indicates the maximum gust of wind in the current decision-making cycle; This indicates the upper limit of the average wind speed at which maintenance work is permitted; This indicates the maximum allowable gust of wind for maintenance work. Indicates the maintenance occupancy status, where This indicates that the wind turbine is undergoing maintenance. This indicates that the wind turbine unit is not occupied by maintenance tasks; Indicates the remaining downtime; Indicates the maintenance window status, where This indicates that the maintenance window is open. This indicates that the maintenance window is closed.
[0073] When the average wind speed or maximum gust When the corresponding maintenance operation safety threshold is exceeded, or when the wind turbine is under maintenance, the remaining downtime is greater than zero, or the maintenance window is closed, the maintenance can be performed. Set to 0; the maintenance can be performed only if all of the above maintenance conditions are met. Take 1.
[0074] (2) Operation control action screening unit: The operation control action filtering unit is used to unify the state vector according to the current decision cycle. Start-stop status Maintenance Occupancy Status and remaining downtime The feasibility of the operation control actions is determined. These operation control actions include rated operation actions, minor load reduction actions, and deep load reduction actions.
[0075] When the wind turbine is under maintenance and occupied, that is or remaining downtime When the wind turbine is in operation, not under maintenance and with zero remaining downtime, the rated operation, slight load reduction and deep load reduction actions are shielded.
[0076] The rated operation is used to ensure the wind turbine operates at its normal power setting; the slight load reduction operation is used to reduce the power setting according to a first load reduction ratio; and the deep load reduction operation is used to reduce the power setting according to a second load reduction ratio, where the output power corresponding to the second load reduction ratio is lower than the output power corresponding to the first load reduction ratio. In one embodiment, rated operation, slight load reduction, and deep load reduction correspond to 100%, 90%, and 80% of the rated power, respectively; the ratios can also be adjusted according to the wind turbine model, control capabilities, and load characteristics of key components.
[0077] (3) Maintenance action screening unit: The maintenance action filtering unit is connected to the maintenance execution condition judgment unit, and is used to determine the maintenance executability based on the maintenance action identifier. Main maintenance requirement components and opportunity maintenance candidate component set The feasibility of maintenance actions is determined. These actions are determined by the maintenance intensity and the combination of repair components, whereby maintenance intensity includes no maintenance, minor repairs, and major repairs.
[0078] Main maintenance requirement component set With opportunity to maintain the candidate component set Together, they form a pool of candidate repair parts:
[0079] in, This represents the pool of candidate repair parts for the current decision-making cycle; This represents the set of components requiring primary maintenance. This indicates the opportunity to maintain a set of candidate components; This represents the union operation of sets.
[0080] When maintaining executable identifiers At that time, minor repairs, major repairs, and their corresponding repair component combinations are disabled, leaving only maintenance-free operations enabled; when And the main maintenance requirement components set When empty, only maintenance-free actions are retained; no maintenance component combinations are generated to avoid creating maintenance tasks when there is no explicit maintenance requirement. And the main maintenance requirement components set When not in use, maintenance-free, minor, and major repairs are allowed to enter the candidate pool, and repair candidate parts are allowed to be selected from the maintenance candidate part pool. One or more maintenance component combinations are formed. Any executable maintenance component combination includes at least one primary maintenance requirement component and may include zero or at least one opportunity maintenance candidate component; maintenance component combinations consisting only of opportunity maintenance candidate components and not containing a primary maintenance requirement component are masked so that opportunity maintenance is attached to the established primary maintenance requirement and available maintenance window.
[0081] For the same combination of repairable components, minor repairs and major repairs are coded as different maintenance intensities. If the health status or structural type of a critical component does not allow for the execution of a corresponding maintenance intensity, then the repair action consisting of that critical component and the maintenance intensity is disabled.
[0082] (4) Risk buffer action constraint unit: The risk buffer action constraint unit is connected to the maintenance execution condition judgment unit, and is used to determine the main maintenance requirement components. When the system is not empty and maintenance actions are not feasible, the operational risk during the maintenance waiting period is determined based on the reliability of the bottleneck component, the mean load factor, and the abnormal fluctuation driving force.
[0083] The operational risk value during the maintenance waiting period is expressed as follows:
[0084] And satisfy:
[0085] in, This indicates the operational risk value during the maintenance waiting period; Indicates the reliability of the bottleneck component; Indicates the average loading factor; Indicates the driving force behind abnormal fluctuations; This indicates the weight corresponding to the health risk of the bottleneck component; This indicates the weight corresponding to the sustained load risk; This indicates the weight corresponding to the risk of abnormal fluctuations.
[0086] The permissible indicator for the rated operating action is determined based on the aforementioned operating risk value:
[0087] in, A permissible indicator indicating the rated operating action; This indicates that the rated operating action can be retained. This indicates that the rated operating action is disabled; Indicates maintenance of executable identifiers; This represents the set of components requiring primary maintenance. Represents the empty set; This indicates the preset risk buffer trigger threshold.
[0088] When maintaining executable identifiers Main maintenance requirement components Non-empty and operating risk value Reaching or exceeding the preset risk buffer trigger threshold At that time, the permitted operation indicator will be displayed. Setting it to 0 disables rated operating actions and retains at least one of light load reduction actions and deep load reduction actions. The collaborative decision-making module determines specific operating control actions based on the remaining executable load reduction actions to reduce output power and equivalent aging increments of critical components during maintenance waiting periods.
[0089] When running risk values The preset risk buffer trigger threshold was not reached. At this time, rated operation, light load reduction and deep load reduction actions can all be retained, and the collaborative decision-making module determines the final operation control action by combining the current state and physical action mask.
[0090] (5) Joint action mask coding unit: The joint action mask encoding unit is connected to the operation control action filtering unit, the maintenance action filtering unit, and the risk buffer action constraint unit, respectively. It is used to combine operation control actions, maintenance intensity, and repair component combinations, and set executable or non-executable identifiers for each candidate joint action in the joint action space according to the aforementioned physical operation constraints. It generates a physical action mask corresponding to the current decision cycle and sends the physical action mask to the collaborative decision module, so that the collaborative decision module determines the final joint action only from the candidate joint actions marked as executable.
[0091] Any candidate joint action is represented as:
[0092] in, Indicates joint action by candidates; This indicates the code for the operation control action, corresponding to one of the following: rated operation, light load reduction, deep load reduction, and maintenance shutdown. This indicates the maintenance intensity code, corresponding to one of no maintenance, minor repair, and major repair. This indicates the combination code for the repair parts. The three components listed in parentheses in a predetermined order form an ordered combination.
[0093] The physical operational constraints of the joint action include at least the following: when the wind turbine is under maintenance or the remaining downtime is greater than zero, new operation control actions or new maintenance tasks are not allowed to be executed; when the maintenance is available... When maintenance intensity is minor or major, and their corresponding maintenance component combinations are not allowed, the operation control action is determined to be maintenance shutdown. When maintenance intensity is no maintenance, no maintenance component combination is generated. Any executable maintenance component combination must include at least one major maintenance requirement component and may include zero or at least one opportunity maintenance candidate component. Maintenance component combinations consisting only of opportunity maintenance candidate components and not containing a major maintenance requirement component are masked. When maintenance action is not executable and the major maintenance requirement component set is not executed, the operation control action is determined to be maintenance shutdown. Non-empty and maintenance waiting period operation risk value Reaching or exceeding the preset risk buffer trigger threshold At that time, the rated operating action of the shield is activated.
[0094] The joint action mask encoding unit sets mask values for each candidate joint action in a preset order of joint actions. The mask value corresponding to the executable candidate joint action is set to 1, and the mask value corresponding to the non-executable candidate joint action is set to 0. The mask values corresponding to each candidate joint action form the physical action mask for the current decision period.
[0095] The physical action mask is sent to the collaborative decision-making module to filter out candidate joint actions that do not meet the current physical operation conditions, so that the final determined operation control actions, maintenance intensity and maintenance component combinations meet the physical constraints formed by the current wind conditions, shutdown status, maintenance window status and health status of key components.
[0096] 3. Collaborative Decision-Making Module: See Figure 1 , Figure 3 and Figure 5 The collaborative decision-making module is connected to the state perception module, the physical action mask generation module, and the instruction execution module, respectively. It is used to determine the operation control actions, maintenance intensity, and repair component combinations based on the current state vector and the physical action mask. The collaborative decision-making module includes a state feature encoding unit, an operation control decision-making unit, a maintenance decision-making unit, a mask strategy generation unit, a joint action determination unit, and a decision result output unit.
[0097] (1) State feature coding unit: The state feature encoding unit is connected to both the state perception module and the physical action mask generation module, and is used to receive the unified state vector output by the state perception module. It also extracts the correlation features between wind conditions, load, component reliability, overall machine reliability, operating mode, shutdown status, and maintenance window status to generate shared state features for the current decision-making cycle.
[0098] in, A unified state vector representing the current decision-making cycle; This represents the encoded shared state characteristics; This represents a state feature encoding function. Preferably, the state feature encoding function consists of a fully connected layer, a normalization layer, and a nonlinear activation layer.
[0099] (2) Operation control decision unit: The operation control decision unit is based on shared state characteristics The candidate scores for operation control were determined for rated operation, light load reduction, and deep load reduction, respectively:
[0100] in, This represents the set of candidate scores for operational control. This indicates the candidate score corresponding to the rated operating action; This represents the candidate score corresponding to a slight load reduction action; This represents the candidate score corresponding to the deep load reduction action; This represents the runtime control decision function; This indicates a shared state characteristic.
[0101] The three candidate scores correspond one-to-one with the three operation control actions of rated operation, slight load reduction, and deep load reduction, respectively, and are used to characterize the relative tendency to select the corresponding operation control action under the current shared state characteristics.
[0102] When there is a maintenance requirement but the maintenance action is not feasible, if the physical action mask marks the rated operation action as not feasible based on the reliability of the bottleneck component, the average load factor and the abnormal fluctuation driving quantity, then the candidate score corresponding to the rated operation action will not participate in the subsequent joint action screening, and only the candidate scores corresponding to the slight load reduction and deep load reduction will be retained to participate in the subsequent joint decision.
[0103] (3) Maintaining the decision-making unit: The maintenance decision unit is connected to the state feature encoding unit and includes a maintenance intensity decision branch and a repair component combination decision branch. The maintenance intensity decision branch is used to determine the shared state features. Candidate scores for maintenance intensity are determined for no maintenance, minor repair, and major repair, respectively:
[0104] in, This represents the set of candidate scores for maintenance strength. This indicates the candidate score corresponding to no maintenance. This indicates the candidate score corresponding to the minor repair; This indicates the candidate score corresponding to the major repair. Represents the maintenance strength decision function; This indicates a shared state characteristic.
[0105] Repair component combination decision branch is used based on shared state characteristics Main maintenance requirement components and opportunity maintenance candidate component set The candidate scores for different repair component combinations are determined, wherein each candidate repair component combination includes at least one component with a primary maintenance requirement and may include zero or at least one candidate component with a chance of maintenance. The candidate scores for repair component combinations are expressed as follows:
[0106] in, This represents the set of candidate scores for repair component combinations; Represents the decision function for the combination of repair parts; This represents the set of components requiring primary maintenance. This indicates the opportunity to maintain the set of candidate components.
[0107] The candidate scores for maintenance intensity and repair component combinations, together with the candidate scores for operation control, are used to determine the scores for subsequent joint actions. These are then combined with a physical action mask to filter out candidate joint actions that are not feasible under the current physical conditions.
[0108] (4) Mask strategy generation unit: The mask strategy generation unit is connected to the operation control decision unit, the maintenance decision unit, and the physical action mask generation module, respectively. It is used to combine the candidate scores of operation control, the candidate scores of maintenance intensity, and the candidate scores of maintenance component combination, and to mask the candidate joint actions that are not executable under the current physical conditions according to the physical action mask, so as to form the probability distribution of executable candidate joint actions.
[0109] Let the set of operation control actions, the set of maintenance intensity, and the set of repair parts combinations be respectively: , and .
[0110] The candidate joint action space is then represented as:
[0111] in, Represents the Cartesian product; This represents the set of candidate joint actions. Any candidate joint action... Represented as:
[0112] in, Indicates the candidate operation control action, Indicates the candidate maintenance strength. This indicates a combination of candidate repair parts.
[0113] For candidate joint actions The combined action score is expressed as:
[0114] in, Represents the set of candidate scores for operation control. Zhongyu The corresponding score; Represents the set of candidate scores for maintenance strength Zhongyu The corresponding score; Represents the set of candidate scores for repair component combinations. Zhongyu The corresponding score.
[0115] set up For candidate joint action The corresponding physical action mask value, where This indicates that the candidate joint action is executable. If the candidate combined action is not executable, then the candidate combined action... The selection probability is expressed as:
[0116] in, Indicates joint action by candidates The probability of choosing; Represents the set of candidate joint actions Any candidate joint action used for summation; Indicates exponentiation; Indicates joint action by candidates The corresponding physical action mask value; Indicates joint action by candidates The corresponding joint action score. Therefore, the selection probability of unexecutable candidate joint actions is set to zero, and the remaining executable candidate joint actions are normalized to form the feasible strategy distribution for the current decision cycle.
[0117] (5) Joint Action Determination Unit: The joint action determination unit determines the final joint action for the current decision cycle from the candidate joint actions marked as executable by the physical action mask, based on the feasible strategy distribution:
[0118] in, This indicates the final joint action determined in the current decision-making cycle; This indicates selecting the candidate joint action with the highest probability from the set of candidate joint actions; Indicates joint action by candidates Belongs to the candidate set of joint actions That is, in the candidate set of joint actions The process iterates through all candidate joint actions.
[0119] Due to candidate joint action The coding is based on three components: operational control actions, maintenance intensity, and repair component combinations. Therefore, it is determined that... Then, it is parsed according to the same predetermined encoding order to obtain:
[0120] in, Indicates the final execution control action; Indicates the final maintenance intensity; This represents the final repair component combination. The three components are not selected independently, but rather are three parts of the same candidate joint action, determined jointly based on the joint action score, physical action mask, and feasible strategy distribution.
[0121] When final maintenance intensity When maintenance is not required, the final operation control action will be used. Perform rated operation, slight de-loading, or deep de-loading; when final maintenance intensity For minor or major repairs, the operating control status is set to maintenance shutdown, and the final repair component assembly is followed. Perform the corresponding maintenance tasks.
[0122] Preferably, the collaborative decision-making module adopts the FAM-DC-PPO model, which reduces the difficulty of solving joint actions through three decomposed policy branches: operation control, maintenance intensity, and repair component combination, and excludes unexecutable actions through physical action masks. During training, a dual pruning mechanism is used to limit the policy update magnitude caused by abnormal negative rewards, thereby improving the stability of long-term collaborative decision-making. See also Figure 5 The FAM-DC-PPO collaborative decision-making model first extracts the current state features through a state feature encoding unit, and then generates candidate decision results through operation control decision branches, maintenance intensity decision branches, and maintenance component combination decision branches, respectively. A physical action mask generation unit is used to mask currently unexecutable actions, and the joint action determination unit determines the operation control, maintenance intensity, and maintenance component combination accordingly. The state, action, and reward information generated by model interaction is stored in a sampling trajectory cache, and parameter optimization is completed by combining GAE advantage estimation, dual pruning strategy updates, and value network updates, thereby achieving collaborative decision-making for operation control and multi-component maintenance under maintenance window constraints.
[0123] 4. Instruction Execution Module: See Figure 1 and Figure 3 The instruction execution module is connected to both the collaborative decision-making module and the status feedback update module. It is used to parse the final joint action, generate operation control instructions or maintenance instructions, and complete safety verification, multi-component maintenance task scheduling, and instruction issuance. The instruction execution module includes a joint action parsing unit, an operation control instruction generation unit, a maintenance instruction generation unit, a multi-component multi-component maintenance task scheduling unit, and a safety interlock and scheduling unit.
[0124] (1) Combined Action Analysis Unit: The joint action parsing unit is used to receive the final joint action output by the collaborative decision-making module. The final operation control action, final maintenance intensity, and final repair component combination are obtained by parsing according to the preset coding order.
[0125] in, Indicates the final joint action of the current decision-making cycle; Indicates the final execution control action; Indicates the final maintenance intensity; This represents the final repair component combination. The three components are jointly determined by the collaborative decision-making module based on the joint action score, physical action mask, and candidate joint action probability distribution. When maintenance is not required, the operation control command generation unit generates commands according to... Generate corresponding operation control commands; when For minor or major repairs, the operating control status is set to maintenance shutdown, and the maintenance command generation unit determines the shutdown status based on the specified parameters. and Generate the corresponding maintenance instructions.
[0126] (2) Operation control instruction generation unit: The operation control instruction generation unit is connected to the joint action parsing unit, and is used to generate operation control instructions based on the final operation control actions. Determine the power control command for the wind turbine. The power control command... Represented as:
[0127] in, This indicates the power control command issued to the wind turbine controller; Indicates the rated power of the wind turbine unit; Indicates the power factor under slight load reduction; Represents the power factor for deep load reduction and satisfies The operation control command generation unit encapsulates the power control command and the corresponding operation mode into an operation control command, and sends it to at least one of the wind turbine main controller, pitch controller, or generator torque controller.
[0128] (3) Maintenance instruction generation unit: The maintenance instruction generation unit is connected to the joint action parsing unit, and is used to determine the final maintenance intensity. and final repair parts combination Generate maintenance tasks. The final repair component assembly... Represented using binary encoding:
[0129] in, Indicates the number of key components involved in the condition assessment; Indicates the first The maintenance selection identifier for each key component involved in the condition assessment; will be the first The maintenance selection identifier for each key component involved in the condition assessment is recorded as follows: ,and ;when When, it indicates the first Several key components have been included in the current maintenance tasks; when When, it indicates the first Several key components have not been included in the current maintenance tasks.
[0130] Based on final maintenance intensity and repair selection indicator , No. The estimated repair time for each critical component is expressed as follows:
[0131] in, Indicates the first Estimated repair time for key components; Indicates the first Estimated repair time for minor repairs of key components; Indicates the first Estimated repair time for major overhauls of key components. When final maintenance intensity... No maintenance task is generated when there is no maintenance required; when the final maintenance intensity is reached... For minor or major repairs, the final assembly of repaired components. The maintenance task is generated for the critical component with the identifier 1 selected in the maintenance selection. The maintenance instruction includes at least the wind turbine identifier, the final maintenance intensity, the component to be maintained identifier, the maintenance window start time, the estimated maintenance time, and the task number, and is sent to the maintenance scheduling platform or operation and maintenance management terminal.
[0132] (4) Multi-component simultaneous repair task arrangement unit: The multi-component simultaneous repair task scheduling unit is connected to the maintenance instruction generation unit, and is used to schedule repair tasks based on the final combination of repaired components. The estimated maintenance time for each selected key component is used to schedule multiple maintenance tasks to be performed in the same maintenance window and during the same downtime, generating corresponding downtime arrangements and multi-component maintenance tasks.
[0133] When multiple maintenance tasks in the final maintenance component assembly meet the personnel, tool, and operational safety requirements and can be performed in parallel, the estimated downtime for this collaborative maintenance is expressed as follows:
[0134] in, This indicates the estimated downtime for this classmate's repair. This indicates the time required for shutdown, personnel organization, and maintenance preparation. Indicates the first Estimated repair time for key components; Indicates the first Repair option labels for key components; Indicates the first Several key components were included in the final repair parts assembly; This indicates that the maximum estimated repair time is taken from all critical components included in the current maintenance task.
[0135] When some maintenance tasks cannot be carried out in parallel due to limitations in personnel, tools, workspace, or maintenance procedures, the multi-component concurrent maintenance task scheduling unit sorts the maintenance tasks according to the corresponding constraints and completes the maintenance tasks sequentially in the same maintenance window and during the same downtime, so as to reduce repeated downtime and preparation work.
[0136] (5) Safety interlock and scheduling unit: The safety interlock and scheduling unit is connected to the operation control command generation unit, the maintenance command generation unit, and the multi-component simultaneous maintenance task arrangement unit, respectively. It is used to perform consistency verification on the operation control command and maintenance command before the command is issued, based on the current physical action mask, maintenance executable identifier, wind turbine start-stop status, remaining downtime, and communication and controller status.
[0137] When a conflict arises between operation control commands and maintenance commands, the operation control commands are blocked and the maintenance commands are executed first. When the verification result meets the command issuance conditions, the operation control commands are sent to at least one of the wind turbine main controller, pitch controller, or generator torque controller, and the maintenance commands are sent to the maintenance dispatch platform or operation and maintenance management terminal.
[0138] The instruction can be issued flag is indicated as follows:
[0139] in, This indicates that the instruction is authorized to be issued; Indicates wind condition and safety signs; Indicates a normal communication link; This indicates that the controller is available. Each of the above identifiers has a value of either 0 or 1. When... hour, This indicates that the corresponding operation control command or maintenance command can be issued; when any flag is 0, Suspend issuing new instructions and maintain the current safe operation status or maintenance shutdown status.
[0140] The execution result of the instruction includes at least the instruction number, actual output power, actual repaired parts, actual maintenance intensity, estimated downtime, and instruction issuance result, and is sent to the status feedback update module for subsequent confirmation of execution results and equipment status updates.
[0141] 5. Status Feedback Update Module: See Figure 1 and Figure 3 The status feedback update module is connected to the instruction execution module, the status perception module, and the physical action mask generation module, respectively. It is used to obtain the execution feedback data of power control instructions and maintenance instructions, update the wind turbine operating status vector, the health status vector of each key component, and the maintenance window available status vector, and generate the input data for the next decision cycle.
[0142] The status feedback update module includes an execution feedback acquisition unit, a running status update unit, a component degradation update unit, a maintenance recovery update unit, a shutdown and window update unit, and a next cycle status generation unit.
[0143] (1) Execution feedback collection unit: The execution feedback acquisition unit is used to acquire the actual output power, actual operation control actions, actual maintenance intensity, actual repair component combinations, maintenance completion status, and actual downtime returned by the instruction execution module, and to form the execution feedback data for the current decision cycle:
[0144] in, Indicates the first Execution feedback data for each decision-making cycle; Indicates the decision-making cycle number; Indicates the actual output power; Indicates the actual execution of operational control actions; Indicates the actual maintenance intensity performed; Indicates the actual combination of repair parts performed; Indicates that the repair is complete; This indicates the actual downtime.
[0145] When the maintenance task is completed, When the maintenance task is not completed, The execution feedback acquisition unit combines the actual execution results with the final joint action determined by the collaborative decision-making module.
[0146] This process is used to determine whether operational control actions and maintenance actions have actually been completed. If the controller or maintenance platform does not return a valid execution result, the corresponding instruction is marked as incomplete, the corresponding maintenance recovery update is not triggered, and the actual equipment state is retained in the next decision cycle.
[0147] (2) Running status update unit: The operation status update unit updates the operation status for the current decision cycle based on actual output power, actual operation control actions, and the start-stop status of the wind turbine. The actual power ratio for the current decision cycle is expressed as:
[0148] in, Indicates the first The actual power ratio of each decision cycle; Indicates the first The actual output power of each decision cycle; This indicates the rated power of the wind turbine. Under normal grid-connected operation conditions, when... When it is close to 1, it means that the actual output power is close to the rated power; when When, it indicates that the actual output power is lower than the rated power; when When the wind turbine is in a certain state, it indicates that the wind turbine has no power output. The actual operating state is determined by combining the actual operation control actions and the start-stop status of the wind turbine. The actual power ratio is... As input for the load state calculation in the next decision cycle.
[0149] (3) Component degradation and renewal unit: The component degradation update unit is used to update the equivalent service time of critical components that have not undergone maintenance and restoration within the current decision-making cycle. For the [number missing]th [period missing]... A key component, which is in the first The equivalent service time at the end of each decision cycle is expressed as:
[0150] in, Indicates the first The key component is in the first The updated equivalent service time at the end of each decision cycle; This indicates the equivalent service life of the critical component at the start of the current decision-making cycle; Indicates the first The equivalent aging increment of key components within the current decision-making cycle.
[0151] The equivalent aging increment Based on the actual power ratio of the current decision cycle, the abnormal fluctuation driving force, and the first The degradation parameters of key components are determined. When the actual power ratio decreases, the equivalent load on the key components decreases accordingly, thereby reducing the corresponding equivalent aging increment; when the abnormal fluctuation driving force increases, the corresponding equivalent aging increment increases accordingly.
[0152] For critical components that have been repaired and meet the maintenance and recovery conditions within the current decision-making cycle, the above-mentioned cumulative degradation results are not used as the final state. Instead, the maintenance and recovery update unit updates their equivalent service time and reliability based on the actual maintenance intensity.
[0153] (4) Maintenance, recovery, and update unit: The maintenance and recovery update unit is used to restore and update the equivalent service time of corresponding key components based on the actual maintenance intensity after the maintenance task is completed. For the first... A key component, which is in the first The equivalent service time at the end of each decision cycle is updated as follows:
[0154] in, Indicates the first The key component is in the first The updated equivalent service time at the end of each decision cycle; This indicates the equivalent service life of the critical component at the start of the current decision-making cycle; Indicates the first The equivalent aging increment of key components within the current decision-making cycle; Indicates the first The minor repair recovery coefficient of each key component must meet the following requirements:
[0155] in, This indicates the proportion of equivalent service time retained after a minor repair. Minor repairs are used to partially restore critical components, while major repairs are used to restore the equivalent service time of critical components to a preset initial state; in this embodiment, the preset initial state is zero. The corresponding maintenance recovery update is performed only when the repair completion status is 1.
[0156] Based on the updated equivalent service time, the Weibull reliability model is used to update the [previous] [reliability]. Reliability of key components:
[0157] in, Indicates the first Reliability of key components after updates; Indicates the first Weibull scale parameters of key components; Indicates the first Weibull shape parameters of key components.
[0158] (5) Shutdown and Window Update Unit: The shutdown and window update unit is used to update the maintenance occupancy status, remaining downtime, and maintenance window status based on the maintenance execution results of the current decision cycle. For the first... The remaining downtime for each decision cycle is updated as follows:
[0159] in, Indicates the first Remaining downtime at the start of each decision cycle; Indicates the remaining downtime before entering the next decision cycle; This indicates the length of time corresponding to a decision-making cycle. This indicates that the maximum value within the parentheses is taken to ensure that the remaining downtime is not less than zero. The maintenance occupancy status is updated based on the remaining downtime and the completion status of the maintenance task. When Maintenance occupancy status ;when And when the current maintenance task has been completed, the maintenance occupancy status is... Maintain window state The conditions, operational resources, and maintenance plans for the next decision cycle will be redefined and used as input for the feasibility assessment of maintenance in the next decision cycle.
[0160] (6) Next cycle state generation unit: The next cycle state generation unit is used to summarize the update results output by the operation state update unit, component degradation update unit, maintenance recovery update unit, and shutdown and window update unit, and to form a unified state vector for the next decision cycle together with the wind conditions and operation data collected in the next decision cycle.
[0161] In the After the decision cycle begins, the state perception module uses the newly collected average wind speed... Maximum gust Average loading factor and abnormal fluctuation driving force Combined with the key component reliability vector obtained from the previous decision cycle update Overall machine reliability Reliability of bottleneck components Overall machine health level Operating Mode Start-stop status Maintenance Occupancy Status Remaining downtime Maintain window status Main maintenance requirement components and the opportunity to maintain the candidate component set Generate a unified state vector for the next decision cycle:
[0162]
[0163] in, Indicates the first A unified state vector for each decision cycle; This represents a normalization operation, used to map continuous data of different dimensions to a preset numerical range; This means combining the state variables into a state vector according to a predetermined order; This indicates a set-indicator encoding operation; the meanings of the other variables are the same as those defined above.
[0164] The unified state vector As input for the generation of physical action masks and collaborative decision-making in the next decision cycle, this forms a closed-loop control process of "state perception - action constraint - collaborative decision-making - instruction execution - state feedback update". (See also...) Figure 6 When critical components require maintenance but the current maintenance window is not yet available, the system implements mild or deep load reduction to reduce operating load and slow down health deterioration. When the maintenance window meets the execution conditions, the system executes the corresponding maintenance task and updates the reliability of critical components and the overall health status based on the maintenance results.
[0165] In the wind turbine load shedding and opportunistic maintenance collaborative control system proposed in this invention under limited maintenance windows, each module is connected through a standardized interface and can be independently configured and deployed according to the wind turbine model, key component composition, and site conditions. In one embodiment, the raw operating data acquisition interval is set to 10 minutes, and the operation and maintenance decision cycle is set to 1 day, which can also be adjusted according to control response speed and maintenance management requirements.
[0166] In the specific software implementation, Python, C++, or other programming languages can be used for development. Numerical computing libraries can be used to complete data preprocessing, state assessment, and matrix operations, and machine learning frameworks can be used to build collaborative decision-making models. Parameters in the system, such as the maintenance wind speed limit, gust limit, reliability boundary, load reduction ratio, maintenance recovery coefficient, and maintenance duration, can be adjusted through configuration files or the operation and maintenance management interface. Wind condition data, component reliability, physical action masks, joint actions, and maintenance feedback information can be stored in relational databases, time-series databases, or file storage systems.
[0167] This invention can be applied to the full lifecycle operation and maintenance management of onshore and offshore wind turbines. When critical components require maintenance but current conditions do not permit repair, the system uses physical action masks to block inoperable maintenance actions and implements slight or deep load reduction. When a maintenance window is open, components requiring primary maintenance and candidate components for opportunistic maintenance are scheduled for repair within the same downtime window. After maintenance is completed, component reliability, downtime status, and maintenance window status are updated based on the actual execution results, thus forming a closed loop between operation control, maintenance scheduling, and status feedback.
[0168] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0169] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints, characterized in that, include: The state awareness module is used to acquire a unified state vector and identify components with major maintenance needs and components with potential maintenance opportunities. The unified state vector includes the wind turbine operating state vector for the current decision cycle, the health state vectors of each key component, and the maintenance window availability state vector. The physical action mask generation module is used to encode the current physical operation constraints into a physical action mask based on the wind turbine operating state vector and the maintenance window available state vector, combined with the main maintenance requirement components and opportunity maintenance candidate components; the physical action mask is used to mark whether each joint action in the joint action space, including operation control actions and maintenance actions, can be executed at the current moment. The collaborative decision-making module is used to filter out the final joint action from the candidate joint actions marked as executable by the physical action mask according to the unified state vector and the physical action mask. The final joint action includes the final operation control action, the final maintenance intensity, and the final repair component combination. The instruction execution module is used to output corresponding control instructions based on the final combined action.
2. The wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints as described in claim 1, characterized in that, The key components include the generator, gearbox, main bearing, and pitch system.
3. The wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints as described in claim 1, characterized in that, The state awareness module includes: The data acquisition unit is used to acquire at least the operating mode, start / stop status, average wind speed, maximum gust, maintenance occupancy status, remaining downtime and maintenance window status of the wind turbine during operation. The load status assessment unit is used to determine the average load factor and abnormal fluctuation driving quantity of the wind turbine based on the average wind speed and maximum gust during the current decision cycle, combined with the average actual output power and rated power. The component health assessment unit is used to determine the equivalent aging increment of each key component in the current decision cycle based on the average load factor and abnormal fluctuation driving amount; update the equivalent service time of each key component based on the equivalent aging increment; and determine the reliability of each key component based on the updated equivalent service time. The overall machine health assessment unit is used to determine the overall reliability of the wind turbine based on the reliability of each key component, determine the minimum value among the reliability of multiple key components as the bottleneck component reliability, and determine the overall machine health level based on the overall machine reliability. The maintenance requirement identification unit is used to identify the main maintenance requirement components and the opportunity maintenance candidate components based on the reliability of each key component and the preset main maintenance reference boundary and opportunity maintenance candidate boundary. The state vector generation unit is used to normalize and fuse the operating mode, start / stop status, average wind speed, maximum gust, average load factor, abnormal fluctuation driving quantity, reliability vector of multiple key components, overall machine reliability, bottleneck component reliability, overall machine health level, maintenance occupancy status, remaining downtime, maintenance window status, set of main maintenance requirement components and set of opportunity maintenance candidate components to generate a unified state vector for the current decision cycle.
4. The wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints as described in claim 3, characterized in that, The physical action mask generation module specifically includes: The maintenance execution condition judgment unit is used to determine whether the wind turbine meets the maintenance operation conditions based on the wind turbine operating state vector and the maintenance window available state vector, and to generate a maintenance execution identifier. The operation control action filtering unit is used to determine the feasibility of operation control actions based on the wind turbine's operating state vector. The operation control actions include rated operation actions, slight load reduction actions, and deep load reduction actions. The maintenance action screening unit is used to determine the feasibility of maintenance actions based on maintenance executability identifiers, a set of main maintenance requirement components, and a set of opportunity maintenance candidate components; the maintenance action is determined by the maintenance intensity and the combination of repair components, and the maintenance intensity includes no maintenance, minor repair, and major repair; The risk buffer action constraint unit is used to determine the operational risk during the maintenance waiting period based on the reliability of bottleneck components, average load factor, and abnormal fluctuation driving quantity when the set of components requiring major maintenance is not empty and maintenance actions are not executable. The joint action mask encoding unit is used to set an executable or non-executable flag for each joint action in the joint action space composed of operation control actions, maintenance intensity and maintenance components, and generate a physical action mask corresponding to the current decision cycle.
5. The wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints as described in claim 4, characterized in that, The operation control action filtering unit is also used to: shield the rated operation action, light load reduction action and deep load reduction action when the wind turbine is under maintenance or the remaining downtime is greater than zero.
6. The wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints as described in claim 4, characterized in that, The maintenance action filtering unit is further configured to: when the maintenance executable identifier is zero, or when the maintenance executable identifier is 1 and the main maintenance requirement component set is empty, to block minor repairs, major repairs, and all repair component combinations; and when the maintenance executable identifier is 1 and the main maintenance requirement component set is not empty, to determine at least one repairable component combination from the main maintenance requirement component set and the opportunity maintenance candidate component set.
7. The wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints as described in claim 4, characterized in that, The risk buffer action constraint unit is also used to: when the operational risk reaches the preset risk buffer condition, shield the rated operational action and retain at least one of the mild load reduction action and the deep load reduction action.
8. The wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints as described in claim 4, characterized in that, The collaborative decision-making module specifically includes: The state feature encoding unit is used to extract and fuse features from the unified state vector of the current decision cycle to generate shared state features; The operation control decision unit is used to determine the operation control candidate scores corresponding to rated operation, light load reduction and deep load reduction respectively based on the shared state characteristics; The maintenance decision unit is used to determine the maintenance intensity candidate score and the repair component combination candidate score based on the shared state characteristics, the set of main maintenance requirement components, and the set of opportunity maintenance candidate components, respectively. The mask strategy generation unit is used to combine the candidate scores of operation control, maintenance intensity, and repair component combination into a joint action score, and to filter out executable candidate joint actions from the joint action space according to the physical action mask, and to calculate the probability distribution of the candidate joint actions based on the joint action score. The joint action determination unit is used to determine the final operation control action, the final maintenance intensity, and the final repair component combination based on the probability distribution of the candidate joint actions, and combine them into the final joint action for the current decision cycle; when the final maintenance intensity is a minor repair or a major repair, the final operation control action is determined to be a maintenance shutdown.
9. The wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints as described in claim 4, characterized in that, The instruction execution module specifically includes: The joint action parsing unit is used to receive the final joint action and parse the final operation control action, final maintenance intensity, and final repair component combination from the final joint action; An operation control instruction generation unit is used to generate corresponding power control instructions based on the final operation control action; A maintenance instruction generation unit is used to generate maintenance instructions based on the final maintenance intensity and the final repair component combination; The multi-component simultaneous maintenance task arrangement unit is used to arrange multiple maintenance tasks in the final maintenance component combination within the same maintenance window, and generate a shutdown command and multiple component maintenance tasks. The safety interlock and scheduling unit is used to perform consistency verification on the instruction before it is issued, based on the current physical action mask, maintenance executable identifier, wind turbine start / stop status, and remaining downtime; when the maintenance instruction conflicts with the power control instruction, the power control instruction is masked and the maintenance instruction is executed first.
10. The wind turbine load reduction and opportunistic maintenance coordinated control system under maintenance window constraints as described in claim 1, characterized in that, It also includes a status feedback update module, which is used to update the wind turbine operating status vector, the health status vector of each key component, and the maintenance window availability status vector according to the execution feedback data of the control command, and use the updated status as the input for the next decision cycle.
