Predictive maintenance method and system for alarm and maintenance linkage of wind power plant

Abstract

The invention discloses a predictive maintenance method and system for wind power plant alarm maintenance linkage, and relates to the technical field of wind power operation and maintenance scheduling. The predictive maintenance method and system for the alarm maintenance linkage of the wind power plant comprises the following steps: S1, periodically collecting operation data of the wind power plant, and executing preprocessing; s2, evaluating the operation risk of each fan, judging whether a fault risk exists or not, triggering an alarm, and marking the fan to be overhauled; s3, judging whether to enter a maintenance scheduling process or not by combining the thermal mechanical shock item and the output deviation item, evaluating the maintenance urgency degree of each to-be-maintained fan, determining a maintenance distribution sequence, and distributing a maintenance channel; s4, after on-site maintenance is completed, maintenance fact data are recorded, maintenance closed-loop entries are constructed, and the upper limit and the lower limit of the disposal level coefficient are dynamically adjusted. The problem of resource abuse caused by full-coverage inspection due to the fact that an existing wind power plant is still dispatched according to fixed inspection and artificial experience is solved.

Description

Technical Field

[0001] This invention relates to the field of wind power operation and maintenance scheduling technology, specifically a predictive maintenance method and system for wind farm alarm and maintenance linkage. Background Technology

[0002] As the installed capacity of wind power continues to increase, the operation and maintenance of wind turbines is gradually evolving from reactive emergency repairs to planned maintenance and predictive maintenance. Existing technologies have proposed various maintenance decision-making schemes for wind turbine reliability and life-cycle cost control, taking into account factors such as the operating status of wind turbines, component degradation trends, and maintenance resource investment. These schemes are gaining attention in high-value scenarios such as offshore wind farms.

[0003] For example, the invention with announcement number CN109376872B relates to an offshore wind turbine maintenance system, including a data input module for inputting real-time status data of each component of the offshore wind turbine; a maintenance judgment module for obtaining the risk level of each component based on the real-time status data using the Weibull proportional hazards model, and determining whether there are any components whose risk level exceeds the corresponding preventive maintenance threshold; and a maintenance scope acquisition module for obtaining the combined risk level of all permutations and combinations of components using a multi-component risk level model based on the Copula function, in real-time response to the judgment result of the maintenance judgment module, and using the component combination with the lowest cost-effectiveness ratio as the maintenance scope of the wind turbine.

[0004] For example, the invention disclosed in CN119740771A discloses a method for determining the preventive maintenance interval of a single component in offshore wind power, including the following steps: based on the analysis of historical failure data of the component, the failure rate distribution function of the single component of the wind turbine is fitted using a Weibull distribution; by introducing a service life reduction factor and a failure rate increase factor, the relationship between the failure rate corresponding to the i-th preventive maintenance and the failure rate corresponding to the (i+1)-th preventive maintenance of the single component is calculated; based on the maintenance threshold, the interval period for each preventive maintenance is calculated; the maintenance cost per unit time of the component is calculated, and the optimal maintenance plan of the component within its life cycle is calculated through an optimization algorithm, thereby determining the optimal number of preventive maintenance operations for the component; after determining the optimal number of preventive maintenance operations, the maintenance process is simulated to obtain the maintenance interval and maintenance cost, and it is evaluated whether it meets the requirements of the wind farm. If it does, the decision parameters are determined; if it does not, the preventive maintenance strategy is adjusted, and the parameters are re-optimized.

[0005] However, existing technologies mostly focus on estimating the timing and scope of maintenance based on component failure rates, lifespan distribution, and maintenance costs. In the actual operation of wind farms, the following needs are not fully addressed: First, a large number of real-time status variables generated during operation are often only used for alarm triggering, rather than for finely differentiating the degree of risk, resulting in different wind turbines not being prioritized at the same time. Second, the actual results formed after on-site maintenance are usually only recorded in the maintenance logbook and are not structured and written back to the alarm decision-making side. The system cannot identify alarm types that recur but have been proven multiple times to be without substantial faults, and may continue to send such alarms out for maintenance as high-priority signals.

[0006] Therefore, in response to the above problems, there is an urgent need for predictive maintenance methods and systems that link alarms and maintenance in wind farms. Summary of the Invention

[0007] Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides a predictive maintenance method and system for alarm and maintenance linkage in wind farms. This solves the problem of resource abuse caused by the current practice of wind farms relying on fixed inspections and manual experience-based work order assignments, which leads to full-coverage inspections.

[0009] Technical solution

[0010] To achieve the above objectives, the present invention provides the following technical solution: a predictive maintenance method for wind farm alarm and maintenance linkage, comprising: S1, periodically collecting wind farm operation data and performing time alignment, outlier removal, missing data interpolation, and standardization on the wind farm operation data to obtain pre-processed wind farm operation data; S2, assessing the operational risk of each wind turbine based on the pre-processed data, determining whether there is a fault risk based on the assessment results, triggering alarms, and marking the wind turbines to be maintained; S3, receiving the sequence of wind turbines to be maintained, determining whether to enter the maintenance scheduling process based on thermomechanical shock and output deviation items, assessing the maintenance urgency of each wind turbine to be maintained, determining the maintenance dispatch order, and allocating maintenance channels based on downtime and number of over-threshold occurrences; S4, after completing on-site maintenance, recording maintenance fact data, constructing maintenance closed-loop entries, and dynamically adjusting the upper and lower limits of the disposal level coefficient based on the maintenance results to achieve maintenance closed-loop control.

[0011] Furthermore, the specific steps for periodically collecting wind farm operation data and performing time alignment, outlier removal, missing data interpolation, and standardization on the wind farm operation data to obtain preprocessed wind farm operation data are as follows: A fixed-width sliding time window is set as one sampling period. Wind farm operation data is periodically collected, including turbine number, wind speed, rotor speed, generator speed, power generation, generator bearing temperature, gearbox oil temperature, gearbox vibration acceleration, and main shaft vibration acceleration. For the collected wind farm operation data, a fixed... The system employs a long-time grid and a nearest-neighbor resampling algorithm to align and resample the timestamps of different signals. It calculates the mean and standard deviation of the time series data for each wind farm based on the three-sigma criterion, identifying and removing outliers exceeding ±3 times the standard deviation of the mean. The system uses a Kalman filter to filter the wind farm operation data, suppressing high-frequency measurement noise. It supplements missing data caused by communication interruptions and packet loss using a piecewise linear interpolation algorithm. Finally, it performs Z-Score standardization on all wind farm operation data to unify the numerical scale and eliminate dimensional differences.

[0012] Furthermore, based on the preprocessed data, the operational risks of each wind turbine are assessed, and the specific steps for determining whether there is a fault risk based on the assessment results are as follows: Extract the preprocessed wind farm operation data, set N sampling periods as one assessment period, and aggregate them according to the wind turbine number to construct a wind turbine operation dataset with assessment periods as the unit; for each assessment period, calculate the average values ​​of gearbox oil temperature, generator bearing temperature, and power generation; add the square of the real-time gearbox vibration acceleration to the square of the real-time main shaft vibration acceleration, add one, and take the natural logarithm to obtain the vibration impact intensity term; average the real-time gearbox oil temperature and the real-time generator bearing temperature to obtain the real-time average temperature; average the average gearbox oil temperature and the average generator bearing temperature within the assessment period to obtain the periodic reference temperature average; and use the real-time temperature... The thermal suppression term is obtained by subtracting the mean of the periodic reference temperature from the mean value, dividing the difference by the average gearbox oil temperature within the evaluation period, and taking the exponential function value of the negative of the ratio with the natural constant e as the base. The thermomechanical shock term is obtained by dividing the vibration and shock intensity term by the sum of the thermal suppression term and the constant one. The output deviation term is obtained by taking the square root of the ratio of the absolute value of the difference between the average power generation within the evaluation period and the real-time power generation, divided by the average power generation within the evaluation period. The output deviation term is obtained by adding the squares of the real-time wind speed, the real-time impeller speed, and the real-time generator speed, adding one, taking the natural logarithm, dividing the logarithm by the average of the real-time impeller speed and the real-time generator speed, and taking the square root of the ratio. The operating stress term is obtained by adding the thermomechanical shock term, the output deviation term, and the operating stress term. The wind turbine operation risk assessment value is obtained by adding the thermomechanical shock term, the output deviation term, and the operating stress term.

[0013] Furthermore, the specific steps for triggering alarms and marking wind turbines to be inspected are as follows: compare the wind turbine operation risk assessment value and risk threshold in real time. When the wind turbine operation risk assessment value is less than or equal to the risk threshold, the corresponding wind turbine is determined to be operating stably and no action is taken. When the wind turbine operation risk assessment value is greater than the risk threshold, the corresponding wind turbine is determined to have a fault risk, triggering a risk alarm, marking the corresponding wind turbine as a wind turbine to be inspected, and extracting all wind turbines to be inspected to construct a wind turbine sequence to be inspected.

[0014] Further, the specific steps for receiving the sequence of wind turbines to be inspected and determining whether to enter the maintenance scheduling process based on the thermomechanical shock item and output deviation item are as follows: Receive the sequence of wind turbines to be inspected; for each turbine, read the thermomechanical shock item, output deviation item, and wind turbine operation risk assessment value at the end of the assessment cycle; and count the number of times the wind turbine operation risk assessment value of the corresponding wind turbine to be inspected exceeds the risk threshold and the cumulative duration of unplanned shutdowns within this assessment cycle, recording them as the number of times exceeding the threshold and the shutdown duration, respectively; compare the thermomechanical shock item and output deviation item with the corresponding multi-level shock threshold and deviation threshold in real time; when both the thermomechanical shock item and output deviation item are not less than the corresponding second-level shock threshold and second-level deviation threshold, [the process is initiated]. Wind turbines awaiting maintenance are marked as requiring immediate maintenance and enter the maintenance scheduling process, with a handling level coefficient of 2. When both the thermomechanical shock item and the output deviation item are less than the corresponding first-level shock threshold and first-level deviation threshold, the wind turbine awaiting maintenance is marked as requiring routine tracking, the sampling frequency is increased, and it is not included in the maintenance scheduling. Otherwise, it is marked as requiring urgent review and enters the review scheduling process: real-time wind farm operation data is extracted to assess the wind turbine operation risk assessment value, and the deviation of the wind turbine operation risk assessment value is calculated. When the recalculated wind turbine operation risk assessment value is less than the risk threshold and the deviation is less than the deviation threshold, it is not included in the maintenance scheduling. Otherwise, it enters the maintenance scheduling process and is configured with a handling level coefficient of 1.

[0015] Further, the specific steps for assessing the urgency of maintenance for each turbine to be inspected and determining the maintenance dispatch order are as follows: For all turbines entering the maintenance dispatch process, the square of the thermomechanical shock term and the square of the output deviation term are added together and the average value is taken to obtain the comprehensive anomaly intensity term; the number of times the threshold is exceeded is increased by one and the natural logarithm is taken, then increased by one again to obtain the continuous anomaly amplification term; the handling level coefficient, the comprehensive anomaly intensity term, and the continuous anomaly amplification term are multiplied sequentially to obtain the dispatch priority term; the downtime is increased by one and the square root is taken, then multiplied by the turbine operation risk assessment value to obtain the downtime impact term; the dispatch priority term and the downtime impact term are added together to obtain the maintenance urgency assessment value; all turbines entering the maintenance dispatch process are arranged in descending order of maintenance urgency assessment value to obtain the maintenance dispatch sequence, and the ranking position of the maintenance dispatch sequence is used as the basis for the maintenance dispatch order.

[0016] Furthermore, the specific steps for allocating maintenance channels based on downtime and number of over-threshold events are as follows: Extract thermomechanical shock items, output deviation items, wind turbine operation risk assessment values, number of over-threshold events, and downtime. Based on comprehensive calculation, obtain the maintenance channel determination value for each wind turbine to be dispatched. When the maintenance channel determination value is negative, use the ranking position of the corresponding wind turbine's maintenance urgency assessment value, the handling level coefficient, downtime, and number of over-threshold events as the dispatch parameters for the corresponding wind turbine. Generate a maintenance linkage record containing the wind turbine number and the end timestamp of the assessment cycle, and push it to the power generation output maintenance channel. Otherwise, generate a maintenance linkage record and push it to the transmission chain maintenance channel.

[0017] Further, the specific steps for extracting the thermomechanical shock term, output deviation term, wind turbine operation risk assessment value, number of over-threshold events, and downtime, and obtaining the maintenance channel determination value for each wind turbine to be dispatched based on comprehensive calculation are as follows: For each wind turbine in the maintenance dispatch sequence, add one to the number of over-threshold events, take the natural logarithm, add one again, and multiply by the thermomechanical shock term to obtain the thermomechanical shock intensity term; divide the thermomechanical shock intensity term by the sum of the square root of the output deviation term and a constant one to obtain the output deviation impact term; add one to the downtime, take the square root, and multiply by the output deviation term to obtain the downtime impact term; divide the downtime impact term by the sum of the square root of the thermomechanical shock term and a constant one to obtain the output deviation correction term; take the natural logarithm of the ratio of the added thermomechanical shock term to the added output deviation term, and multiply by the wind turbine operation risk assessment value to obtain the risk correction term; subtract the output deviation correction term from the output deviation impact term, and add the risk correction term to obtain the maintenance channel determination value.

[0018] Furthermore, after completing on-site maintenance, the maintenance fact data is recorded, maintenance closed-loop entries are constructed, and the upper and lower limits of the handling level coefficient are dynamically adjusted based on the maintenance results to achieve maintenance closed-loop control. The specific steps are as follows: After completing on-site maintenance, maintenance fact data is collected and recorded. The maintenance fact data includes: fault confirmation flag, turbine number, maintenance duration, maintenance end timestamp, number of participants in the maintenance, corresponding downtime, replacement part code, and number of replacement parts. Among them, a fault confirmation flag of 1 indicates that a physical fault is confirmed, and 0 indicates that no physical fault is found. The maintenance fact data is bound to the maintenance linkage record of the corresponding turbine according to the turbine number and the end timestamp of the evaluation cycle to form a maintenance closed-loop entry. Based on the maintenance closed-loop entries accumulated for the same turbine in the same maintenance channel, the number of times the fault confirmation flag is 1 and 0 are counted respectively, and the upper and lower limits of the handling level coefficient are dynamically adjusted.

[0019] The second aspect of this invention provides a predictive maintenance system for wind farm alarm and maintenance linkage, comprising: a data acquisition and preprocessing module, a risk assessment and alarm decision module, a maintenance linkage dispatching module, and a maintenance result write-back closed-loop module. The data acquisition and preprocessing module periodically acquires wind farm operation data and performs time alignment, outlier removal, missing data imputation, and standardization on the wind farm operation data to obtain preprocessed wind farm operation data. The risk assessment and alarm decision module assesses the operational risk of each wind turbine based on the preprocessed data and, based on the assessment results... If a fault risk is determined, an alarm is triggered and the fan to be repaired is marked; the maintenance linkage dispatch module is used to receive the sequence of fans to be repaired, determine whether to enter the maintenance dispatch process based on thermomechanical shock and output deviation, assess the urgency of each fan to be repaired, determine the maintenance dispatch order, and then allocate maintenance channels based on downtime and number of over-threshold events; the maintenance result write-back closed-loop module is used to record maintenance fact data after the on-site maintenance is completed, construct maintenance closed-loop entries, and dynamically adjust the upper and lower limits of the disposal level coefficient based on the maintenance results to achieve maintenance closed-loop control.

[0020] Beneficial effects

[0021] The present invention has the following beneficial effects:

[0022] (1) Predictive maintenance method and system for alarm and maintenance linkage in wind farms. By processing operating data such as vibration, temperature rise, power deviation, and speed conditions into quantitative indicators such as thermomechanical shock items, output deviation items, and operating stress items, and further combining downtime and over-threshold times, maintenance urgency assessment value and maintenance channel judgment value are formed, so that the severity of the units has a clear and comparable order, supporting the scheduling and hierarchical response of batch wind turbines, rather than manual judgment of each unit.

[0023] (2) Predictive maintenance method and system for alarm and maintenance linkage in wind farms. By not only indicating whether maintenance is needed, but also dividing the wind turbines to be inspected into different maintenance channels during the dispatching stage, such as channels that tend to have problems with power generation output or channels that tend to have problems with transmission chain, the dispatching results directly correspond to the maintenance side, reducing unprepared maintenance that has to be judged on-site, and improving the pertinence of maintenance upon arrival.

[0024] (3) A predictive maintenance method and system for alarm-based maintenance linkage in wind farms. This system sets multi-level thresholds and review processes before entering maintenance scheduling: wind turbines with slightly abnormal conditions will not be directly dispatched for maintenance, but the sampling frequency will be increased first; wind turbines in the middle range will enter expedited review, and only wind turbines that still show an abnormal trend after review will be included in maintenance scheduling. This mechanism screens out situations where turbines are forcibly dispatched due to slight fluctuations at the process level, reducing unnecessary shutdowns and unnecessary tower climbing.

[0025] (4) A predictive maintenance method and system for alarm and maintenance linkage in wind farms. After on-site maintenance is completed, maintenance data is collected and recorded, and the results are bound to the corresponding dispatch records, which are then used to dynamically adjust the upper and lower limits of the handling level coefficient. In this way, the system's response level to similar alarms is no longer a static rule, but will gradually converge with the actual maintenance results, realizing iterative updates and continuous correction of the operation and maintenance strategy. Attached Figure Description

[0026] Figure 1 A flowchart of a predictive maintenance method for alarm-based maintenance linkage in wind farms;

[0027] Figure 2 A structural diagram of a predictive maintenance system for alarm and maintenance linkage in wind farms;

[0028] Figure 3 A schematic diagram of the dispatching process for joint maintenance;

[0029] Figure 4 This is a schematic diagram showing the results of the inspection channel assessment. Detailed Implementation

[0030] 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.

[0031] Please see Figures 1-4 This invention provides a technical solution: a predictive maintenance method for wind farm alarm and maintenance linkage, comprising: S1, periodically collecting wind farm operation data, and performing time alignment, outlier removal, missing data interpolation, and standardization on the wind farm operation data to obtain pre-processed wind farm operation data; S2, assessing the operational risk of each wind turbine based on the pre-processed data, determining whether there is a fault risk based on the assessment results, triggering alarms, and marking the wind turbines to be maintained; S3, receiving the sequence of wind turbines to be maintained, determining whether to enter the maintenance scheduling process based on thermomechanical shock and output deviation items, assessing the maintenance urgency of each wind turbine to be maintained, determining the maintenance dispatch order, and then allocating maintenance channels based on downtime and number of over-threshold occurrences; S4, after completing on-site maintenance, recording maintenance fact data, constructing maintenance closed-loop entries, and dynamically adjusting the upper and lower limits of the disposal level coefficient based on the maintenance results to achieve maintenance closed-loop control.

[0032] Specifically, the following steps are taken to periodically collect wind farm operation data and perform time alignment, outlier removal, missing data imputation, and standardization on the data to obtain preprocessed wind farm operation data: A fixed-width sliding time window is set as one sampling period. Wind farm operation data is collected periodically. Real-time measurements of the corresponding wind turbines are recorded within each sampling period to form a serialized raw measurement record indexed by the sampling timestamp. The wind farm operation data includes turbine number, wind speed, rotor speed, generator speed, power generation, generator bearing temperature, gearbox oil temperature, gearbox vibration acceleration, and main shaft vibration acceleration. The turbine number is obtained by reading the turbine's... Unique identifiers are used for data collection. Wind speed is collected via a wind speed sensor in the wind turbine nacelle; impeller speed is collected via a speed measurement device on the impeller shaft; generator speed is collected via a speed measurement device on the generator shaft; power generation is collected via a power measurement device on the generator output side; generator bearing temperature is collected via a temperature sensor at the bearing location; gearbox oil temperature is collected via a temperature sensor in the gearbox lubrication oil circuit; gearbox vibration acceleration is collected via a vibration acceleration measurement device at the gearbox housing; and main shaft vibration acceleration is collected via a vibration acceleration measurement device at the main shaft support location. The collected wind farm operating data is then used to establish a unified sampling period. A fixed-step time grid is used, and the timestamps of different signals are aligned and resampled using a nearest-neighbor resampling algorithm. This ensures that wind speed, rotor speed, generator speed, power generation, generator bearing temperature, gearbox oil temperature, gearbox vibration acceleration, and main shaft vibration acceleration form a time-by-time correspondence on the same time grid, with the turbine number remaining unchanged as the index key. The mean and standard deviation of the time series data for each wind farm are calculated point-by-point based on the three-sigma criterion. Outliers exceeding ±3 times the standard deviation are identified and removed. During outlier removal, the sampling timestamp and turbine number of the removed data points are simultaneously retained, while unremoved data points continue to be processed. A Kalman filter algorithm is then applied. The wind farm operation data is filtered to suppress high-frequency measurement noise. During the filtering process, recursive state estimation is performed on wind speed, rotor speed, generator speed, power generation, generator bearing temperature, gearbox oil temperature, gearbox vibration acceleration, and main shaft vibration acceleration to generate smoothed time series. The missing data caused by communication interruption and packet loss is supplemented by a piecewise linear interpolation algorithm. When supplementing the missing data, the original sampling period and fixed step time grid remain unchanged. The missing data supplementation is performed independently for each channel of wind speed, rotor speed, generator speed, power generation, generator bearing temperature, gearbox oil temperature, gearbox vibration acceleration, and main shaft vibration acceleration.Z-Score standardization was performed on all wind farm operation data to unify the numerical scale and eliminate dimensional differences. Z-Score standardization used the time series of the same wind turbine number within the same sampling period as the statistical object, calculating the standardization results for wind speed, rotor speed, generator speed, power generation, generator bearing temperature, gearbox oil temperature, gearbox vibration acceleration, and main shaft vibration acceleration, resulting in preprocessed wind farm operation data.

[0033] In this implementation plan, by comprehensively collecting, aligning, filtering, smoothing, completing, and standardizing wind turbine numbers, wind speeds, impeller speeds, generator speeds, power generation, generator bearing temperatures, gearbox oil temperatures, gearbox vibration accelerations, and main shaft vibration accelerations under a unified sampling period, consistent, continuous, and comparable wind farm operation data inputs are formed for each wind turbine in each sampling period. This avoids instability in status judgments caused by collection noise, timestamp drift, and missing data interference in the subsequent risk assessment stage. It ensures that the calculations of thermomechanical shock items, output deviation items, and operating stress items are based on the same time grid and the same numerical scale, so that the maintenance urgency assessment value and maintenance channel judgment value can accurately reflect the differences in wind turbine operating status, supporting maintenance dispatch based on operating status rather than a fixed inspection sequence.

[0034] Specifically, the steps for assessing the operational risk of each wind turbine based on the preprocessed data and determining the existence of fault risks based on the assessment results are as follows: Extract the preprocessed wind farm operation data, set N sampling periods as one assessment period, and aggregate the data by wind turbine number. For the same wind turbine number, the wind speed, impeller speed, generator speed, power generation, generator bearing temperature, gearbox oil temperature, gearbox vibration acceleration, and main shaft vibration acceleration within N consecutive sampling periods are merged in chronological order to construct a wind turbine operation dataset with the assessment period as the unit; where N is a positive integer greater than one; for each assessment period, calculate the average values ​​of gearbox oil temperature, generator bearing temperature, and power generation. The average gearbox oil temperature is used to characterize the thermal level of the gearbox lubrication circuit during the evaluation period; the average generator bearing temperature is used to characterize the thermal level of the generator bearings during the evaluation period; and the average power generation is used to characterize the active power output level during the evaluation period. The vibration impact intensity term is obtained by adding the square of the real-time gearbox vibration acceleration to the square of the real-time spindle vibration acceleration, then adding one, and taking the natural logarithm. This term measures the concentration of the current transmission chain vibration amplitude in an instantaneous state. The average real-time gearbox oil temperature and the real-time generator bearing temperature are averaged to obtain the real-time temperature mean. The average gearbox oil temperature and the average generator bearing temperature during the evaluation period are then calculated... The average temperature of the period is calculated by subtracting the average temperature of the periodic reference temperature from the average temperature of the real-time temperature. The difference is then divided by the average gearbox oil temperature within the evaluation period. The negative of this ratio is then used as an exponential function with the natural constant e as the base to obtain the thermal suppression term. This term reflects the degree of suppression of the real-time temperature rise relative to the reference thermal level of the evaluation period. The vibration shock intensity term is then divided by the sum of the thermal suppression term and the constant e to obtain the thermomechanical shock term. This term characterizes the comprehensive shock intensity of the instantaneous vibration response under the current temperature rise background. Finally, the absolute value of the difference between the average power generation within the evaluation period and the real-time power generation is divided by the average power generation within the evaluation period. The square root of the resulting ratio is taken to obtain the power output. The deviation term, specifically the output deviation term, measures the degree of deviation of real-time power generation from the average output level of the assessment period. The natural logarithm is obtained by adding the squares of the real-time wind speed, the real-time impeller speed, and the real-time generator speed, then dividing this logarithm by the average of the real-time impeller speed and the real-time generator speed, and taking the square root of the resulting ratio. This yields the operating stress term, which characterizes the coupling strength between the real-time aerodynamic drive state and the transmission chain rotation state. The wind turbine operating risk assessment value is obtained by adding the thermomechanical shock term, the output deviation term, and the operating stress term. This value is used to quantify the operating risk level of the wind turbine at the end of the assessment period, serving as an input for subsequent maintenance decisions.

[0035] The specific formula for calculating the risk assessment value of wind turbine operation is as follows:

[0036] ;

[0037] In the formula, This indicates the risk assessment value for wind turbine operation. This indicates the average gearbox oil temperature. This represents the average temperature of the generator bearings. Indicates the real-time gearbox oil temperature. This indicates the real-time generator bearing temperature. This represents the real-time gearbox vibration acceleration. This represents the real-time spindle vibration acceleration. This represents the average power generation during the assessment period. Indicates real-time power generation. Indicates real-time wind speed. Indicates the real-time impeller speed. This indicates the real-time generator speed.

[0038] In this implementation plan, wind speed, impeller speed, generator speed, power generation, generator bearing temperature, gearbox oil temperature, gearbox vibration acceleration, and main shaft vibration acceleration are aggregated during the assessment period. At the end of the assessment period, vibration impact intensity, thermal suppression, thermomechanical shock, output deviation, and operating stress are merged into a single wind turbine operation risk assessment value. This transforms risk judgment from a single alarm signal to a joint quantitative indicator based on multiple physical quantities, avoiding the need to trigger maintenance based on a single point of sensing, preventing minor fluctuations from being treated as serious faults, and establishing directly comparable risk levels between different wind turbines, so that maintenance triggering no longer relies on manual experience judgment and fixed inspection sequences.

[0039] Specifically, the steps for triggering an alarm and marking a wind turbine for maintenance are as follows: Real-time comparison of the wind turbine's operational risk assessment value and risk threshold, where the risk threshold is the numerical boundary used to distinguish between a stable state and a fault risk state; When the wind turbine's operational risk assessment value is less than or equal to the risk threshold, the corresponding wind turbine is determined to be operating stably, no action is taken, and the wind turbine is not allowed to enter the subsequent maintenance process. Simultaneously, the determination result of the wind turbine in this assessment cycle is recorded as a stable state, for subsequent screening of wind turbines already determined to be stable; When the wind turbine's operational risk assessment value is greater than the risk threshold, the corresponding wind turbine is determined to have a fault risk, triggering a risk alarm, marking the corresponding wind turbine as a wind turbine for maintenance, recording the wind turbine's number and the corresponding assessment cycle end timestamp as maintenance identification information, extracting all wind turbines for maintenance to construct a wind turbine sequence for maintenance, and using this wind turbine sequence as input for subsequent maintenance dispatch, ensuring that subsequent maintenance dispatch is based solely on the wind turbine sequence for maintenance to perform downstream processing.

[0040] In this implementation plan, the wind turbine operation risk assessment value is compared with the risk threshold for each turbine. The quantitative value is used to determine whether a turbine is included in the maintenance sequence. This ensures that the maintenance sequence only includes turbines whose operation risk assessment value is higher than the risk threshold. This avoids the practice of manually selecting maintenance targets one by one based on experience and avoids the situation where low-risk turbines are passively dispatched for maintenance under a fixed inspection sequence. At the same time, it ensures that the turbine number and the timestamp at the end of the assessment cycle are recorded in pairs. Subsequent maintenance dispatch is based on the same maintenance sequence and is carried out in a targeted manner, rather than being arranged indiscriminately for the entire site.

[0041] Specifically, the steps for receiving the sequence of wind turbines to be inspected and determining whether to enter the maintenance scheduling process based on thermomechanical shock and output deviation are as follows: Receive the sequence of wind turbines to be inspected; for each turbine, read the thermomechanical shock, output deviation, and wind turbine operation risk assessment value at the end of the assessment period; and count the number of times the wind turbine operation risk assessment value of the corresponding wind turbine to be inspected exceeds the risk threshold and the cumulative unplanned downtime within this assessment period, recorded as the number of times the threshold is exceeded and the downtime, respectively. The number of times the threshold is exceeded is used to characterize the frequency of high-risk states of the corresponding wind turbine to be inspected within this assessment period, and the downtime is used to characterize the scope of grid interruption impact caused by the corresponding wind turbine to be inspected within this assessment period; compare the thermomechanical shock and output deviation with the corresponding multi-level shock threshold and deviation threshold in real time to determine the emergency level of the corresponding wind turbine maintenance. The specific process is as follows: Figure 3 As shown; where the multi-level impact threshold includes a first-level impact threshold and a second-level impact threshold, and the multi-level deviation threshold includes a first-level deviation threshold and a second-level deviation threshold; when both the thermomechanical impact item and the current output deviation item are not less than the corresponding second-level impact threshold and second-level deviation threshold, the fan to be repaired is marked as an immediate repair level, enters the repair scheduling process, and is configured with a handling level coefficient of 2. The handling level coefficient is used to indicate the urgency of issuing a repair instruction directly; when both the thermomechanical impact item and the output deviation item are less than the corresponding first-level impact threshold and first-level deviation threshold, the fan to be repaired is marked as a routine tracking level, the sampling frequency is increased, it is not included in the repair scheduling, and in this assessment week During the period, the wind turbine will be registered as a continuously monitored object to maintain data continuity without occupying maintenance scheduling resources; otherwise, it will be marked as an urgent review level and enter the review scheduling process: extract real-time wind farm operation data to evaluate the wind turbine operation risk assessment value, and calculate the deviation of the wind turbine operation risk assessment value. The deviation is used to characterize the change before and after the review. When the recalculated wind turbine operation risk assessment value is less than the risk threshold and the deviation is less than the deviation threshold, it will not be included in the maintenance scheduling and the state of increasing the sampling frequency will be maintained; otherwise, it will enter the maintenance scheduling process and be configured with a handling level coefficient of 1. When the handling level coefficient is 1, it means that maintenance needs to be arranged but the priority is lower than the immediate maintenance level.

[0042] In this implementation plan, before entering the maintenance scheduling process, a tiered screening is conducted based on thermomechanical shock items, output deviation items, wind turbine operation risk assessment values, number of threshold exceedances, and downtime. Only wind turbines that reach the immediate maintenance level or maintain high-risk performance after reviewing the scheduling process are included in the maintenance scheduling process. Furthermore, wind turbines that do not reach the first-level threshold are registered as continuous monitoring targets, and only the sampling frequency is increased, without directly issuing maintenance instructions. This avoids treating minor fluctuations as emergency faults and scheduling them for operation, and suppresses the resource abuse caused by experience-based full-coverage inspections. At the same time, a disposal level coefficient is assigned to each wind turbine entering the maintenance scheduling process, forming the priority basis for the subsequent maintenance dispatch order, so that maintenance resources are concentrated on wind turbines that are truly high-risk.

[0043] Specifically, the steps for assessing the urgency of maintenance for each turbine to be inspected and determining the maintenance dispatch order are as follows: For all turbines entering the maintenance dispatch process, the square of the thermomechanical shock term and the square of the output deviation term are added together and averaged to obtain the comprehensive abnormality intensity term. The comprehensive abnormality intensity term is used to characterize the combined magnitude of the vibration shock level and output deviation level of the turbine at the end of the current assessment period; the number of times the threshold is exceeded is incremented by one, the natural logarithm is taken, and then incremented by one again to obtain the continuous abnormality amplification term. The continuous abnormality amplification term is used to characterize the cumulative degree of high-risk status of the turbine during the current assessment period; the handling level coefficient, the comprehensive abnormality intensity term, and the continuous abnormality amplification term are multiplied sequentially to obtain the dispatch priority term, which is used to reflect the turbine's priority in dispatch. The urgency of intervention in the dispatching process should be prioritized; the downtime duration is added by one, the square root is taken, and then multiplied by the wind turbine operation risk assessment value to obtain the downtime impact item. The downtime impact item is used to quantify the cumulative exposure level of the wind turbine's unplanned downtime impact and operation risk intensity within the current assessment period; the dispatching priority item and the downtime impact item are added to obtain the maintenance urgency assessment value, which serves as a quantitative indicator of the necessity of current intervention for the wind turbine; all wind turbines entering the maintenance dispatching process are arranged from largest to smallest according to their maintenance urgency assessment value to obtain the maintenance dispatch sequence, and the order of maintenance dispatch is based on the ranking position in the maintenance dispatch sequence. Wind turbines with higher ranking positions in the maintenance dispatch sequence are given priority for issuing maintenance instructions.

[0044] The specific formula for calculating the maintenance urgency assessment value is as follows:

[0045] ;

[0046] In the formula, This indicates the assessment value for the urgency of maintenance. Indicates the level of treatment coefficient. Indicates the thermomechanical shock term. Indicates the deviation of output. This indicates the risk assessment value for wind turbine operation. Indicates the number of times the threshold is exceeded. Indicates the downtime.

[0047] In this implementation plan, thermomechanical shock items, output deviation items, number of times exceeding the threshold, downtime, handling level coefficient, and wind turbine operation risk assessment value are uniformly converted into maintenance urgency assessment value. The wind turbines entering the maintenance scheduling process are then ranked according to the maintenance urgency assessment value to form a maintenance dispatch sequence. This makes the maintenance dispatch order no longer dependent on fixed inspection routes, but rather on centralized decision-making based on quantified risk exposure intensity, abnormality persistence, and downtime impact. It can clearly determine which wind turbine to maintain first and which to maintain later within the same assessment cycle, avoiding the average distribution of maintenance resources to low-risk wind turbines and reducing the situation of unplanned on-site inspections of each turbine.

[0048] Specifically, the steps for allocating maintenance channels based on downtime and the number of times exceeding the threshold are as follows: Extract the thermomechanical shock item, output deviation item, wind turbine operation risk assessment value, number of times exceeding the threshold, and downtime. Based on comprehensive calculation, obtain the maintenance channel determination value for each wind turbine to be dispatched. The maintenance channel determination value is used to distinguish the main anomalies of the wind turbine to be dispatched at the end of the current assessment cycle. The thermomechanical shock item reflects the vibration and impact intensity of the transmission chain; the output deviation item reflects the deviation intensity of power generation; the wind turbine operation risk assessment value reflects the overall operational risk exposure level; the number of times exceeding the threshold reflects the frequency of high-risk manifestations; and the downtime reflects the operational interruption exposure caused by unplanned downtime. When the maintenance channel determination value is negative, the ranking position of the corresponding wind turbine's maintenance urgency assessment value, the handling level coefficient, downtime, and number of times exceeding the threshold are used as the dispatch parameters for the corresponding wind turbine. All parameters are aligned with the end timestamp of the same assessment cycle, and a maintenance linkage record containing the wind turbine number and the end timestamp of the assessment cycle is generated and pushed to the power generation output maintenance channel. Among these, the power generation output maintenance... The maintenance channel is used for on-site maintenance and handling related to power generation output capacity. The targets of these maintenance and handling include the power output chain, unit power decline performance, and grid connection output capacity reduction. Maintenance linkage records issued through this channel are used on-site to investigate power output capacity recovery issues and restore power output from the normal range. Otherwise, a maintenance linkage record is generated and pushed to the drivetrain maintenance channel. The drivetrain maintenance channel is used for on-site maintenance and handling related to drivetrain impact behavior. The targets of these maintenance and handling include main shaft vibration behavior, gearbox vibration behavior, and gearbox thermal load performance. Maintenance linkage records issued through this channel are used on-site to investigate drivetrain structural impact risks and suppress continuous high-impact operation of the drivetrain. During the above-mentioned push process, each maintenance linkage record uses the turbine number and the end time stamp of the assessment cycle as the retrieval index, allowing the dispatched maintenance target to directly locate the corresponding turbine and the source of the anomaly during the execution phase, using this source as the starting point for maintenance, without needing to conduct initial attribution confirmation upon arrival.

[0049] In this implementation plan, the maintenance linkage record corresponding to the wind turbine to be dispatched is pushed to the power generation output maintenance channel or the transmission chain maintenance channel based on the maintenance channel judgment value during the dispatching process. The maintenance linkage record also carries the ranking position of the maintenance urgency assessment value, the handling level coefficient, the downtime, and the number of times the threshold is exceeded. This ensures that each wind turbine to be dispatched is bound with a clear handling side, handling priority, and handling background indicators before it is dispatched. This allows the on-site maintenance personnel to directly use the maintenance linkage record as the execution entry point and conduct targeted investigations according to the problem side corresponding to the assigned maintenance channel. This reduces the initial confirmation steps after arrival, shortens the time span between arrival and substantive intervention, and avoids the same wind turbine being repeatedly dispatched as a new problem in different stages.

[0050] Specifically, the steps for extracting the thermomechanical shock term, output deviation term, wind turbine operation risk assessment value, number of threshold exceedances, and downtime, and obtaining the maintenance channel judgment value for each wind turbine to be dispatched based on comprehensive calculation are as follows: For each wind turbine in the maintenance dispatch sequence, add one to the number of threshold exceedances, take the natural logarithm, add one again, and multiply by the thermomechanical shock term to obtain the thermomechanical shock intensity term, which is used to characterize the amplification effect of the frequency of high-risk states of the wind turbine on the degree of transmission chain impact during the current assessment period; divide the thermomechanical shock intensity term by the sum of the square root of the output deviation term and a constant one to obtain the output deviation impact term, which is used to characterize the relative impact intensity of the degree of power generation deviation under the current transmission chain impact background; add one to the downtime, take the square root, and multiply by the output deviation term to obtain the downtime impact term, which is used to characterize the cumulative exposure of unplanned downtime to the wind turbine. The impact of power generation deviation is assessed. The shutdown impact term is divided by the sum of the square root of the thermomechanical shock term and a constant, yielding a power deviation correction term. This correction term is used to normalize the shutdown impact term in the presence of high transmission chain impact, ensuring it maintains the same scale as the transmission chain impact background in subsequent calculations. The ratio of the thermomechanical shock term plus one to the power deviation term plus one is taken as its natural logarithm and multiplied by the wind turbine's operational risk assessment value, resulting in a risk correction term. This risk correction term is used to adjust the relative dominance of transmission chain impact behavior and power generation deviation behavior by incorporating the overall operational risk exposure level. Subtracting the power deviation correction term from the power deviation impact term and then adding the risk correction term yields a maintenance path determination value. This value quantifies the wind turbine's abnormal dominance at the end of the current assessment period and serves as the basis for subsequent allocation of power generation maintenance paths or transmission chain maintenance paths.

[0051] The specific calculation formula for the inspection passage judgment value is as follows:

[0052] ;

[0053] In the formula, This indicates the inspection channel judgment value. Indicates the thermomechanical shock term. Indicates the deviation of output. This indicates the risk assessment value for wind turbine operation. Indicates the number of times the threshold is exceeded. Indicates the downtime.

[0054] In this embodiment, Table 1 is a data table of maintenance access judgment values, listing the quantitative parameters of the operating characteristics of five fans to be inspected and the corresponding maintenance access judgment values. The selected quantitative parameters of operating characteristics include thermomechanical shock, output deviation, fan operation risk assessment value, number of times exceeding the threshold, and downtime. Specifically: the thermomechanical shock of fan 1 to be inspected is 2.5, the output deviation is 0.4, the fan operation risk assessment value is 1.8, the number of times exceeding the threshold is 3, the downtime is 1.5, and the corresponding maintenance access judgment value is 5.059; the thermomechanical shock of fan 2 to be inspected is 1.2, the output deviation is 0.9, the fan operation risk assessment value is 1.1, the number of times exceeding the threshold is 1, the downtime is 0.2, and the corresponding maintenance access judgment value is 0.733; the thermomechanical shock of fan 3 to be inspected is 3.8, the output deviation is 0.3, the fan operation risk assessment value is 1.8, the number of times exceeding the threshold is 3, the downtime is 1.5, and the corresponding maintenance access judgment value is 5.059; The estimated value is 2.4, the number of times the threshold is exceeded is 5, the downtime is 4.0, and the corresponding maintenance channel judgment value is 9.762; the thermomechanical shock item of the fan to be inspected is 0.7, the output deviation item is 1.5, the fan operation risk assessment value is 0.9, the number of times the threshold is exceeded is 2, the downtime is 1.0, and the corresponding maintenance channel judgment value is -0.842; the thermomechanical shock item of the fan to be inspected is 2.9, the output deviation item is 2.2, the fan operation risk assessment value is 1.6, the number of times the threshold is exceeded is 4, the downtime is 6.5, and the corresponding maintenance channel judgment value is 1.135.

[0055] Table 1. Inspection Passage Judgment Value Data Table

[0056]

[0057] like Figure 4The figure shows the maintenance channel determination values ​​and corresponding maintenance channel allocation results for five wind turbines awaiting maintenance. Different colored bars distinguish the handling aspects of each turbine, visually demonstrating the identification of the dominant source of anomalies for different turbines within the same assessment period. When the maintenance channel determination value is non-negative, corresponding to the orange bar, it is automatically assigned to the drivetrain maintenance channel, prioritizing the inspection of parts related to drivetrain impact behavior. When the maintenance channel determination value is negative, corresponding to the blue bar, it is automatically assigned to the power generation output maintenance channel, prioritizing the handling of reduced power generation output capacity. As can be seen from the figure, the maintenance channel determination values ​​for turbines 1, 2, 3, and 5 are all non-negative, indicating they are assigned to the drivetrain maintenance channel. This suggests that the main anomaly of these turbines is dominated by thermomechanical impact, manifested as high drivetrain impact levels and exposure of drivetrain-related operating stresses. The maintenance channel determination value for turbine 4 is below zero, indicating it is assigned to the power generation output maintenance channel. This suggests that the main anomaly of this turbine is dominated by output deviation, manifested as a relatively prominent decline in power generation. Figure 4 This intuitively demonstrates the ability to classify wind turbines based on the maintenance channel judgment value, so that each wind turbine to be maintained is clearly diverted to the transmission chain maintenance channel or the power generation output maintenance channel at the dispatch stage, realizing the maintenance direction in advance, avoiding manual attribution judgment after arrival, and reducing the ineffective dispatch of vehicles for repeated tower inspections.

[0058] In this implementation plan, thermomechanical shock items, output deviation items, wind turbine operation risk assessment values, over-threshold times, and downtime duration are converted into maintenance channel judgment values ​​through a unified calculation process. These maintenance channel judgment values ​​are then used to distinguish between power generation output problems and transmission chain problems. This ensures that the division of maintenance channels is not determined by a single instantaneous signal, but rather by simultaneously considering the intensity of transmission chain impact, the degree of power generation deviation, the frequency of high-risk conditions, the exposure of operational interruptions caused by unplanned shutdowns, and the overall level of operational risk exposure. This transforms the judgment of the source of influence into a reproducible numerical judgment process, ensuring that the fault indications of the same wind turbine remain consistent across different assessment cycles, different maintenance teams, and different dispatch batches, reducing misallocation of the fault due to differences in human interpretation.

[0059] Specifically, after completing on-site maintenance, the maintenance fact data is recorded, maintenance closed-loop entries are constructed, and the upper and lower limits of the handling level coefficient are dynamically adjusted based on the maintenance results to achieve maintenance closed-loop control. The specific steps are as follows: After completing on-site maintenance, maintenance fact data is collected and recorded. The maintenance fact data includes: fault confirmation flag, turbine number, maintenance duration, maintenance end timestamp, number of participants in the maintenance, corresponding downtime, replacement part code, and number of replacement parts; where, a fault confirmation flag of 1 indicates that a physical fault is confirmed, and 0 indicates that no physical fault is found; the maintenance duration is used to characterize the duration of the maintenance of the turbine from the start to the end of the maintenance process. The duration of the maintenance and the maintenance end timestamp are used to identify the end time of the maintenance loop. The number of participants in the maintenance represents the intensity of manpower deployment on-site. The corresponding downtime of the maintenance represents the duration of grid interruption caused by the maintenance. The replacement component code and the number of replaced components represent the component replacement behavior involved in this maintenance. The maintenance fact data and the corresponding wind turbine maintenance linkage record are bound by wind turbine number and evaluation cycle end timestamp to form a maintenance loop entry. The maintenance loop entry is used to simultaneously retain the maintenance linkage record on the dispatch side and the maintenance fact data on the on-site side in the same data structure, so that the handling level coefficient at the time of dispatch is consistent with the actual result after execution. Establish a corresponding relationship; based on the accumulated maintenance closed-loop entries of the same wind turbine under the same maintenance channel, count the number of maintenance closed-loop entries with a fault confirmation flag of 1, and record it as the fault count; count the number of maintenance closed-loop entries with a fault confirmation flag of 0, and record it as the no-fault count. The specific steps for dynamically adjusting the upper and lower limits of the handling level coefficient are as follows: Record the handling level coefficients corresponding to the immediate maintenance level and the expedited review level as the upper limit and lower limit of the handling level coefficient, respectively; perform the following update rule on the upper limit of the handling level coefficient: divide the upper limit of the handling level coefficient by one and add the no-fault count to obtain the updated upper limit of the handling level coefficient; perform the following update rule on the lower limit of the handling level coefficient. New rule: Multiply the lower limit of the handling level coefficient by one plus the natural logarithm of the fault count, and then add one to obtain the updated lower limit of the handling level coefficient; use the updated upper limit and lower limit of the handling level coefficient as the upper and lower limits of the handling level coefficient for the wind turbine in the maintenance channel, respectively, for the handling level coefficient constraint when the wind turbine triggers the wind turbine to be maintained again in the subsequent evaluation cycle. This ensures that the larger the number of no faults, the lower the corresponding upper limit of the handling level coefficient, and the larger the number of faults, the higher the corresponding lower limit of the handling level coefficient. In this way, the handling level coefficient is automatically updated in the subsequent evaluation cycle according to the actual maintenance effectiveness of the wind turbine in the maintenance channel.

[0060] In this implementation plan, maintenance data is linked to corresponding maintenance linkage records by turbine number and assessment cycle end timestamp, forming a closed-loop maintenance entry that can be traced back to a specific turbine, specific time period, and specific handling level coefficient. This ensures that each dispatch, each arrival, and each handling result is recorded in pairs, and the updated results of the handling level coefficient upper and lower limits are solidified in subsequent assessment cycles. This establishes a continuous and verifiable handling history for the same turbine, so that different maintenance teams use the same set of quantitative records when facing the same type of alarm in subsequent assessment cycles, instead of making independent judgments. This avoids the situation where the same turbine location is given contradictory conclusions repeatedly in different batches of maintenance.

[0061] like Figure 2 As shown, the second aspect of this invention provides a predictive maintenance system for wind farm alarm and maintenance linkage, including: a data acquisition and preprocessing module, a risk assessment and alarm decision module, a maintenance linkage dispatching module, and a maintenance result write-back closed-loop module. The data acquisition and preprocessing module is used to periodically acquire wind farm operation data and perform time alignment, outlier removal, missing data imputation, and standardization on the wind farm operation data to obtain preprocessed wind farm operation data. The risk assessment and alarm decision module is used to assess the operational risk of each wind turbine based on the preprocessed data and, based on the assessment... The system assesses the risk of failure, triggers alarms, and marks the turbines to be inspected. The maintenance linkage dispatch module receives the sequence of turbines to be inspected, determines whether to enter the maintenance dispatch process based on thermomechanical shock and output deviation, assesses the urgency of each turbine's maintenance, determines the maintenance dispatch order, and allocates maintenance channels based on downtime and number of over-threshold events. The maintenance result write-back closed-loop module records maintenance data after on-site maintenance, constructs maintenance closed-loop entries, and dynamically adjusts the upper and lower limits of the handling level coefficient based on the maintenance results to achieve closed-loop maintenance control.

[0062] In this implementation plan, by introducing a sequential connection of four stages within the same technical framework—data acquisition and preprocessing, risk assessment and alarm decision-making, maintenance linkage dispatch and scheduling, and maintenance result write-back closed loop—the operating status data of the wind turbine is limited to a unified data format that can be used for dispatch decision-making from the beginning of sampling. This ensures that the data format has not been manually rewritten before entering the maintenance linkage dispatch and scheduling. At the same time, the maintenance fact data collected at the end of the on-site maintenance is forcibly bound to the previous dispatch basis and fed back to update the upper and lower limits of the disposal level coefficient, forming a corrective capability for the dispatch decision itself. Thus, within the same technical framework, it simultaneously possesses risk identification capability, maintenance dispatch capability, and self-correction capability, making the dispatch logic no longer a one-way execution, but one that can be constrained and corrected by subsequent actual maintenance results.

[0063] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0064] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A predictive maintenance method of wind farm alarm repair linkage, characterized in that, The method comprises the following steps: S1, periodically collecting wind farm operation data, and performing time alignment, abnormal point elimination, missing data interpolation and standardization processing on the wind farm operation data to obtain preprocessed wind farm operation data; S2, evaluating the operation risk of each wind turbine based on the preprocessed data, determining whether there is a fault risk based on the evaluation result, triggering an alarm and marking the wind turbine to be repaired; S3, receiving the sequence of wind turbines to be repaired, determining whether to enter the repair scheduling process in combination with the thermal mechanical impact item and the output deviation item, evaluating the repair urgency of each wind turbine to be repaired, determining the repair dispatching order, and then distributing the repair channels in combination with the shutdown time and the number of times exceeding the threshold; S4, after completing the on-site repair, recording the repair factual data, constructing a repair closed loop item, and dynamically adjusting the upper and lower limits of the disposal level coefficient based on the repair result to realize closed loop control of repair.

2. The predictive maintenance method of claim 1, wherein: The specific steps of periodically collecting wind farm operation data and performing time alignment, abnormal point elimination, missing data interpolation and standardization processing on the wind farm operation data to obtain preprocessed wind farm operation data are as follows: A fixed-width sliding time window is set as one sampling period, and wind farm operation data is periodically collected. The wind farm operation data includes wind turbine number, wind speed, impeller speed, generator speed, power generation, generator bearing temperature, gearbox oil temperature, gearbox vibration acceleration and main shaft vibration acceleration; For the collected wind farm operation data, a fixed step time grid is established based on a unified sampling period, and the timestamps of different signals are aligned and resampled by a nearest neighbor resampling algorithm; the mean value and standard deviation of each wind farm operation data time series are calculated point by point based on the three sigma criterion, and abnormal points exceeding ±3 times the standard deviation of the mean value are identified and eliminated; the wind farm operation data is filtered by a Kalman filter algorithm to suppress high-frequency measurement noise; missing data caused by communication interruption and packet loss is supplemented by a piecewise linear interpolation algorithm; and all wind farm operation data is subjected to Z-Score standardization processing to unify the numerical scale and eliminate the dimensional difference.

3. The wind farm alarm-based predictive maintenance method of claim 1, wherein: The specific steps of evaluating the operation risk of each wind turbine based on the preprocessed data and determining whether there is a fault risk based on the evaluation result are as follows: The preprocessed wind farm operation data is extracted, N sampling periods are set as one evaluation period, and the wind turbine number is aggregated to construct a wind turbine operation data set with an evaluation period as a unit; for each evaluation period, the average values of the gearbox oil temperature, the generator bearing temperature and the power generation are calculated; The vibration impact intensity term is obtained by adding the square of the real-time gearbox vibration acceleration and the square of the real-time main shaft vibration acceleration, taking the natural logarithm of the sum of 1; the real-time temperature average is obtained by averaging the real-time gearbox oil temperature and the real-time generator bearing temperature; the period reference temperature average is obtained by averaging the average of the gearbox oil temperature in the evaluation period and the average of the generator bearing temperature in the evaluation period; the thermal inhibition term is obtained by subtracting the period reference temperature average from the real-time temperature average, dividing the difference by the average of the gearbox oil temperature in the evaluation period, taking the exponential function value of the inverse of the obtained ratio with the natural constant e as the base; the thermal mechanical impact term is obtained by dividing the vibration impact intensity term by the sum of the thermal inhibition term and a constant 1; the power deviation term is obtained by taking the square root of the ratio of the absolute value of the difference between the average of the generated power in the evaluation period and the real-time generated power to the average of the generated power in the evaluation period; the working condition stress term is obtained by adding the square of the real-time wind speed, the square of the real-time impeller speed and the square of the real-time generator speed, taking the natural logarithm of the sum of 1, dividing the logarithmic value by the average of the real-time impeller speed and the real-time generator speed, and taking the square root of the obtained ratio; the fan operation risk assessment value is obtained by adding the thermal mechanical impact term, the power deviation term and the working condition stress term.

4. The wind farm alarm-based predictive maintenance method of claim 1, wherein: The specific steps of triggering an alarm and marking the fan to be repaired are as follows: The real-time fan operation risk assessment value is compared with the risk threshold value, and when the fan operation risk assessment value is less than or equal to the risk threshold value, it is determined that the corresponding fan is running stably and no processing is performed; When the fan operation risk assessment value is greater than the risk threshold value, it is determined that the corresponding fan has a fault risk, a risk alarm is triggered, the corresponding fan is marked as a fan to be repaired, and all fans to be repaired are extracted to construct a sequence of fans to be repaired.

5. The wind farm alarm-based predictive maintenance method of claim 1, wherein: The specific steps of receiving the sequence of fans to be repaired and determining whether to enter the maintenance scheduling process in combination with the thermal mechanical impact term and the power deviation term are as follows: The sequence of fans to be repaired is received, the thermal mechanical impact term, the power deviation term and the fan operation risk assessment value at the end of the evaluation period are read for each fan, and the number of times that the fan operation risk assessment value of the corresponding fan to be repaired exceeds the risk threshold value and the cumulative length of unplanned downtime in the evaluation period are counted, which are denoted as the threshold exceeding number and the downtime length, respectively; The thermal mechanical shock item and the output deviation item are compared with corresponding multi-stage shock threshold and deviation threshold in real time, when the thermal mechanical shock item and the output deviation item are not less than corresponding secondary shock threshold and secondary deviation threshold, the wind turbine to be overhauled is marked as an immediate overhaul level, enters an overhaul scheduling process, and a disposal level coefficient is configured as 2; when the thermal mechanical shock item and the output deviation item are less than corresponding first-stage shock threshold and first-stage deviation threshold, the wind turbine to be overhauled is marked as a routine tracking level, a sampling frequency is improved, and is not included in the overhaul scheduling; otherwise, the wind turbine to be overhauled is marked as an urgent review level, enters a review scheduling process, real-time wind farm operation data is extracted to evaluate a wind turbine operation risk evaluation value, and a deviation amount of the wind turbine operation risk evaluation value is calculated, when the recalculated wind turbine operation risk evaluation value is less than a risk threshold and the deviation amount is less than a deviation threshold, the wind turbine to be overhauled is not included in the overhaul scheduling; otherwise, the wind turbine to be overhauled enters the overhaul scheduling process, and the disposal level coefficient is configured as 1.

6. The wind farm alarm-based predictive maintenance method of claim 1, wherein: The specific steps of evaluating the overhaul urgency of each wind turbine to be overhauled and determining an overhaul dispatch sequence are as follows: For all wind turbines entering the overhaul scheduling process, the square of the thermal mechanical shock item and the square of the output deviation item are added and averaged to obtain a comprehensive abnormal intensity item; the natural logarithm of the super-threshold number plus one is obtained, and then one is added to obtain a sustained abnormal amplification item; the disposal level coefficient, the comprehensive abnormal intensity item and the sustained abnormal amplification item are multiplied in turn to obtain a scheduling priority item; the square root of the shutdown time plus one is obtained, and then multiplied by the wind turbine operation risk evaluation value to obtain a shutdown impact item; the scheduling priority item and the shutdown impact item are added to obtain an overhaul urgency evaluation value; all wind turbines entering the overhaul scheduling process are arranged in descending order of the overhaul urgency evaluation value to obtain an overhaul dispatch sequence, and the ordering position of the overhaul dispatch sequence is taken as the basis of the overhaul dispatch sequence.

7. The wind farm alarm-based predictive maintenance method of claim 1, wherein: The specific steps of distributing the overhaul channels in combination with the shutdown time and the super-threshold number are as follows: The thermal mechanical shock item, the output deviation item, the wind turbine operation risk evaluation value, the super-threshold number and the shutdown time are extracted, and the overhaul channel judgment value of each wind turbine to be dispatched is obtained based on comprehensive calculation; when the overhaul channel judgment value is a negative value, the ordering position of the overhaul urgency evaluation value of the corresponding wind turbine, the disposal level coefficient, the shutdown time and the super-threshold number are taken as the dispatch parameters of the corresponding wind turbine, an overhaul linkage record containing the wind turbine number and the evaluation period end timestamp is generated, and is pushed to the power output overhaul channel; Otherwise, the overhaul linkage record is generated and is pushed to the transmission chain overhaul channel.

8. The prognostic maintenance method of wind farm alarm corrective maintenance linkage according to claim 7, characterized in that: The specific steps of extracting the thermal mechanical shock item, the output deviation item, the wind turbine operation risk evaluation value, the super-threshold number and the shutdown time, and obtaining the overhaul channel judgment value of each wind turbine to be dispatched based on comprehensive calculation are as follows: For each fan in the maintenance dispatch sequence, the threshold number of times is added one, then the natural logarithm is taken, and then the thermal mechanical impact term is multiplied by one, to obtain the thermal mechanical impact intensity term; the thermal mechanical impact intensity term is divided by the square root of the sum of the output deviation term and the constant one, to obtain the output deviation influence term; the shutdown duration is added one, then the square root is taken, and then the output deviation term is multiplied, to obtain the shutdown influence term; the shutdown influence term is divided by the square root of the sum of the thermal mechanical impact term and the constant one, to obtain the output deviation correction term; the ratio of the thermal mechanical impact term plus one to the output deviation term plus one is taken the natural logarithm, and then the fan operation risk assessment value is multiplied, to obtain the risk correction term; The output deviation influence term is subtracted from the output deviation correction term, and then the risk correction term is added, to obtain the maintenance access decision value.

9. The wind farm alarm-based predictive maintenance method of claim 1, wherein: After the on-site maintenance is completed, the maintenance fact data is recorded, the maintenance closed loop item is constructed, and the upper and lower limits of the disposal level coefficient are dynamically adjusted based on the maintenance result, to realize the specific steps of the maintenance closed loop control as follows: After the on-site maintenance is completed, the maintenance fact data is collected and recorded, and the maintenance fact data includes: fault confirmation flag, fan number, maintenance duration, maintenance end time stamp, number of participants in maintenance, maintenance corresponding shutdown duration, replacement component code and replacement component quantity; wherein the fault confirmation flag takes 1 to indicate that there is an entity fault, and takes 0 to indicate that no entity fault is found; the maintenance fact data and the maintenance linkage record of the corresponding fan are bound according to the fan number and the evaluation period end time stamp, to form the maintenance closed loop item; based on the cumulative maintenance closed loop items of the same fan under the same maintenance access, the number of times of the fault confirmation flag being 1 and 0 is counted respectively, and the upper and lower limits of the disposal level coefficient are dynamically adjusted.

10. A predictive maintenance system for wind farm alarm repair linkage, characterized in that: It comprises: a data acquisition preprocessing module, a risk assessment alarm decision module, a maintenance linkage dispatch scheduling module and a maintenance result write-back closed loop module, wherein: The data acquisition preprocessing module is used for periodically collecting wind farm operation data, and performing time alignment, outlier rejection, missing interpolation and standardization processing on the wind farm operation data, to obtain preprocessed wind farm operation data; The risk assessment alarm decision module is used for assessing the operation risk of each fan based on the preprocessed data, determining whether there is a fault risk based on the assessment result, triggering an alarm and marking the fan to be maintained; The maintenance linkage dispatch scheduling module is used for receiving the sequence of the fan to be maintained, determining whether to enter the maintenance scheduling process in combination with the thermal mechanical impact term and the output deviation term, evaluating the maintenance urgency of each fan to be maintained, determining the maintenance dispatch sequence, and then distributing the maintenance access in combination with the shutdown duration and the threshold number of times; The maintenance result write-back closed loop module is used for recording the maintenance fact data after completing the on-site maintenance, constructing the maintenance closed loop item, and dynamically adjusting the upper and lower limits of the disposal level coefficient based on the maintenance result, to realize the maintenance closed loop control.

Images