A method for locating abnormal periods of load distribution in wind power converters

By establishing normal ranges for load increase and decrease in wind power converters and distinguishing abnormal cooling temperature rise, the false alarm and missed alarm problems in the location of abnormal periods of converters in the existing technology have been solved, and accurate location of abnormal periods has been achieved.

CN123121981APending Publication Date: 2026-11-17HUANENG HUILI WIND POWER GENERATION CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202611628005.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-21
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively distinguish between normal thermal hysteresis and genuine anomalies during the frequent load changes of wind power converters, leading to false alarms or missed alarms and making it impossible to accurately pinpoint abnormal periods.

Method used

By acquiring data on the output power, cooling inlet temperature, and cooling outlet temperature of the liquid-cooled converter of the wind turbine generator set, normal ranges for load increase and load decrease are established, forming a common normal area. Only data that are outside the two normal ranges at the same time as the cooling temperature rise are identified as abnormal, and abnormal periods are formed according to the sampling time.

Benefits of technology

It enables accurate differentiation between normal thermal hysteresis and true anomalies without adding new measuring points, providing direct conclusions on the load range and time at which the converter experienced an anomaly, thus reducing false alarms and missed alarms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123121981A_ABST
    Figure CN123121981A_ABST
Patent Text Reader

Abstract

This invention relates to the field of wind power converter monitoring technology, and discloses a method for locating abnormal periods of load distribution in wind power converters. The method includes: acquiring the output power, cooling inlet temperature, and cooling outlet temperature of the liquid-cooled converter of a wind turbine generator set in relation to the sampling time; determining the cooling temperature rise based on the difference between the cooling outlet temperature and the cooling inlet temperature; distinguishing between load increase data and load decrease data from normal historical operating data, establishing normal ranges for load increase and load decrease according to load location, and forming a common normal area from the two types of normal ranges at the same load location; confirming abnormal data when the cooling temperature rise of the operating data to be located is on the same outer side of the two types of normal ranges; grouping consecutive abnormal data into abnormal periods according to the sampling time sequence, determining the abnormal start time, abnormal end time, minimum power of the period, and maximum power of the period, thus forming the result of locating abnormal periods of load distribution.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of wind power converter monitoring technology, and more specifically, to a method for locating abnormal periods of load distribution in wind power converters. Background Technology

[0002] Megawatt-class wind turbine converters commonly employ liquid cooling. The cooling medium's temperature rises after flowing through the power devices, and the difference between the cooling outlet temperature and the cooling inlet temperature (i.e., cooling temperature rise) is a commonly used existing measurement point in wind farm SCADA systems for judging the converter's heating status. However, as the turbines continuously follow wind speed changes, the converters repeatedly experience load increases and decreases during grid-connected operation. Power devices, cold plates, and the cooling medium possess heat capacity, and their temperature changes lag behind load changes. Therefore, under the same output power, the cooling temperature rise during load increases and decreases does not coincide. This round-trip difference is normal for the turbines, but it presents practical difficulties for judging anomalies based on cooling temperature rise.

[0003] Several Chinese patents have addressed the assessment and protection of converter cooling status. CN108235656A (Control Method and Control System for Cooling System of Wind Turbine Converter) utilizes the inlet temperature, outlet temperature, and temperature difference of the cooling medium to regulate and protect the cooling system. Its assessment focuses on the current temperature and temperature difference themselves, without considering the temperature rise difference in different load directions under the same load. CN112084661A (Cooling Status Assessment and Early Warning Method for Water-Cooled System of Wind Turbine Converter) assesses the cooling status based on measurement point data and issues an early warning. The assessment conclusion is for the entire cooling system, without specifying the load range and start / end time of the anomaly. CN115633489A (Cooling System Anomaly Identification Method, Device, Computer Equipment, and Storage Medium) groups the cooling temperature difference by hub speed, performs bicentric clustering within each group, and uses the average distance between the two cluster centers to determine whether the entire historical data segment is abnormal. Its conclusion granularity is the entire data segment, and only a single benchmark is formed within each speed group; the normal temperature rise difference between load increases and decreases is incorporated into the same benchmark. CN117992887A (A Multi-Constraint Anomaly Detection Method for Wind Turbine Units under Different Operating Conditions) establishes detection models for each operating condition, but still uses a single detection benchmark within each condition, and similarly fails to distinguish between load direction differences under the same operating condition. CN115045807A (A Method for Detecting Anomaly in the Generator Speed ​​of Wind Turbine Units) can connect consecutively exceeding limits into an abnormal time period, but its exceedance criterion assumes that the detected quantity has only one normal level under each operating condition. CN115578084A (A Fault Early Warning Method for Wind Turbine Unit Inverters Based on Deep Convolutional Autoencoders) relies on a large number of historical samples for training, outputs the warning level, and does not directly give the abnormal start time and abnormal end time of each load.

[0004] When applying the above approach to converters operating under frequent load changes, two common problems arise: First, establishing only one normal level for each load range or operating condition makes it easy for the normal cooling temperature rise difference between load increases and decreases to be reported as an anomaly, resulting in false alarms. Second, relaxing the judgment boundaries to suppress these false alarms shifts the detection location of the actual anomaly, leading to anomaly periods that start earlier, end later, or are interrupted in the middle, making it difficult to provide directly usable time period conclusions for on-site inspections. Therefore, without adding new measurement points and using existing SCADA records, distinguishing between normal thermal hysteresis and actual anomalies, and directly providing the load range and time period from which an anomaly occurred for a specific converter, is a practical problem that urgently needs to be solved in converter condition monitoring. Summary of the Invention

[0005] This invention provides a method for locating abnormal periods of load distribution in wind power converters, which solves the technical problems mentioned in the background art.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solution.

[0007] A method for locating abnormal load periods in wind power converters, comprising:

[0008] The output power, cooling inlet temperature, and cooling outlet temperature of the liquid-cooled converter of the wind turbine generator set are obtained in relation to the sampling time. The obtained data includes normal historical operating data and operating data to be located.

[0009] The cooling temperature rise is determined based on the difference between the cooling outlet temperature and the cooling inlet temperature at the same sampling time.

[0010] Based on the change in output power over sampling time, load increase data and load decrease data are distinguished from normal historical operating data;

[0011] Based on the load location represented by the output power, establish the normal range of load increase data and the normal range of load decrease data respectively. The two normal ranges of the same load location form a common normal area, so that the same load location retains two types of normal thermal response.

[0012] The output power of the operating data to be located is mapped to the common normal area. Only when the cooling temperature rise at the same sampling time is located on the same outer side of the normal range of load increase and the normal range of load decrease, is the data at that sampling time confirmed as abnormal data.

[0013] Based on the chronological order of sampling time, consecutive abnormal data are grouped into abnormal time periods, and the start and end times of the abnormal time periods are determined.

[0014] Determine the minimum and maximum power during the abnormal period, and combine them with the target converter number, the start time of the abnormality, and the end time of the abnormality to form the result of load distribution abnormal period location.

[0015] Compared with the prior art, the present invention has the following substantial features and significant progress:

[0016] The monitoring processor only reads the existing output power, cooling inlet temperature, and cooling outlet temperature of the wind turbine generator's liquid-cooled converter. It establishes normal ranges for load increase and load decrease based on the load location represented by the output power, and forms a common normal area from the two types of normal ranges at the same load location. This ensures that the normal thermal hysteresis caused by the heat capacity of the device and cooling medium during the load increase and load decrease processes is retained within the judgment criteria. Only the operating data to be located that is simultaneously located on the same outer side of both types of normal ranges is confirmed as abnormal data. This avoids the problems of judging normal thermal hysteresis as abnormal when monitoring based on a single criterion, or the problem of missing judgments due to relaxing the criteria to take into account both types of normal thermal responses when monitoring. The confirmed abnormal data are arranged into abnormal time periods with abnormal start and end times according to the chronological relationship of the sampling time, and are accompanied by the minimum power and maximum power of the time period. Based on this, maintenance personnel can directly know in which load range and at what time the target converter experienced cooling abnormality. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the system modules of the present invention;

[0018] Figure 2 This is a schematic diagram of the method flow of the present invention;

[0019] Figure 3 This is the power-temperature rise data graph of the present invention. Detailed Implementation

[0020] The following description is provided in conjunction with the accompanying drawings. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the invention; those skilled in the art can make equivalent substitutions or combinations for the specific implementations.

[0021] This embodiment relates to cooling monitoring of liquid-cooled converters in wind turbine generators during load-changing grid connection. During grid-connected operation, wind turbine converters frequently increase and decrease loads according to wind speed and grid dispatch. When output power increases, the losses of power devices increase accordingly, causing device temperatures to rise initially. Heat is conducted to the cooling medium via the cold plate, and the cooling medium carries away the heat, causing its temperature to rise, and the cooling outlet temperature changes accordingly. When output power decreases, the above process reverses. Power devices, cold plates, and cooling media all have heat capacity, and the change in cooling outlet temperature lags behind the change in output power. Therefore, under the same output power, the normal load increase process and the normal load decrease process have different cooling temperature rises. During load increase, the power device temperature has not yet fully risen, resulting in a relatively low cooling temperature rise; during load decrease, the power device temperature has not yet dropped, resulting in a relatively high cooling temperature rise. This difference is a normal thermal lag that changes with load direction. During one load cycle, the same output power corresponds to the normal cooling temperature rise during the load increase phase and the normal cooling temperature rise during the load decrease phase, and the values ​​between the two temperature rises are all within the normal range. The specific technical problem to be solved by this embodiment is: when the load of the wind power converter changes back and forth, causing the same load to correspond to two normal cooling temperature rises of increasing load and decreasing load, distinguishing between the normal thermal hysteresis and the abnormality that deviates from both normal thermal responses at the same time, and determining the start and end time of the abnormality. The confirmation of the abnormality and the location of the start and end boundaries are completed using data from existing measuring points.

[0022] To address the aforementioned issues, this implementation retains two types of normal responses caused by thermal hysteresis under the same load: Based on the change in output power over sampling time, load increase data and load decrease data are distinguished from normal historical operating data. Normal ranges for load increase and load decrease are established separately, and the two types of normal ranges at the same load location form a common normal region, thus retaining two types of normal thermal responses at the same load location. Only when the cooling temperature rise of the operating data to be located is simultaneously located on the same outer side of both normal ranges, i.e., it cannot be explained by either normal load direction, is it identified as abnormal data. Subsequently, consecutive abnormal data are grouped into abnormal time periods according to the chronological order of sampling time, determining the start and end times of the abnormality, and providing the output power range within the time period, forming the load-specific abnormal time period location results.

[0023] See Figure 1The wind turbine generator set 111 is equipped with a liquid-cooled converter 112 and a cooling circuit 113. The cooling medium circulates between the cooling circuit 113 and the liquid-cooled converter 112. The side where the cooling medium flows into the liquid-cooled converter 112 is the cooling inlet, and the side where it flows out of the liquid-cooled converter 112 is the cooling outlet. A cooling inlet temperature sensor 114 is installed at the cooling inlet, and the measured cooling inlet temperature reflects the thermal state of the cooling medium before it enters the liquid-cooled converter 112. A cooling outlet temperature sensor 115 is installed at the cooling outlet, and the measured cooling outlet temperature reflects the thermal state of the cooling medium after it has flowed through the liquid-cooled converter 112. The converter power measurement channel 116 is connected to the liquid-cooled converter 112 to obtain the output power of the liquid-cooled converter 112. The monitoring processor 117 receives three data streams from the cooling inlet temperature sensor 114, the cooling outlet temperature sensor 115, and the converter power measurement channel 116. These three data streams are merged into the same data bus, aligned according to the sampling time, and correlated with the sampling time. The acquired data includes normal historical operating data and operating data to be located. The monitoring processor 117 can be any of the following: a wind turbine controller, a site edge computer, or a SCADA server. The load distribution anomaly location results output by the monitoring processor 117 are written to the SCADA event log and displayed on the monitoring interface. The SCADA event log and the monitoring interface are integrated. Figure 1 The SCADA event recording and monitoring interface 118 is used. The above three physical quantities are all taken from the existing measurement points of the liquid-cooled converter 112, and the construction of the converter power circuit, cooling pipeline, circulating pump and wind turbine remains in its existing state.

[0024] See Figure 2 The wind power converter load abnormality period location method of this embodiment includes steps S211 to S225. Steps S211 and S212 complete data acquisition and recording. Steps S213 to S218 establish two types of normal ranges for each load location based on normal historical operating data and form a common normal area. Steps S219 to S221 judge the operating data to be located window by window and confirm abnormal data. Steps S222 to S225 group the abnormal data into abnormal time periods and output the location results.

[0025] See Figure 1 and Figure 2 In step S211, the monitoring processor 117 acquires the output power, cooling inlet temperature, and cooling outlet temperature of the liquid-cooled converter 112 of the wind turbine generator set in relation to the sampling time. The acquired data includes normal historical operating data and operating data awaiting location. Normal historical operating data is taken from records of the liquid-cooled converter 112 during normal operation and is used to establish normal load increase and decrease ranges. Operating data awaiting location is the record of periods awaiting location of abnormalities. Each record includes three physical quantities at the same sampling time. The sampling time is denoted as... , indicating the first The recording time is recorded by the SCADA clock; the output power is recorded as... , indicating the first The active power on the AC side of the converter is recorded in this record, by the converter controller, in MW; the cooling inlet temperature is recorded as... , indicating the first The recorded cooling medium inlet temperature is measured by cooling inlet temperature sensor 114, in degrees Celsius; the cooling outlet temperature is recorded as... , indicating the first The cooling medium outlet temperature recorded in the first record was measured by cooling outlet temperature sensor 115, and the unit is degrees Celsius. Three physical quantities are related to the sampling time, i.e., the [number]th [item / item / etc.]. records , and Corresponding to the same sampling time Each record thus has both a numerical basis for comparing load and temperature rise, and a basis for chronological arrangement. The monitored liquid-cooled converter 112 has a unique target converter number in the SCADA system, which marks the liquid-cooled converter to which the abnormal period belongs in the location results. The cooling status of liquid-cooled converter 112 is normal within the time period covered by the normal historical operation data, and the load increase and decrease processes within this time period are both sources of normal thermal response; the operation data to be located is a record continuously generated during operation, updated one by one with the sampling cycle.

[0026] See Figure 2 In step S212, the monitoring processor 117 organizes the acquired records. Normal historical operating data and operating data to be located come from records of the same measuring point, the same unit, and the same sampling period. Both types of data have consistent dimensions and time intervals in the comparison of load and temperature rise. Each record is arranged in ascending order of sampling time. After sorting, the temporal relationship between adjacent records is the chronological relationship in the data sequence. The determination of load direction and the determination of time period boundaries are based on this order. When multiple records exist at the same sampling time, the last complete record is retained. A complete record is one where output power, cooling inlet temperature, and cooling outlet temperature are all complete. Records missing any of these three values ​​are considered blank records and are not included in subsequent calculations. After organizing the records, records of grid-connected loads with output power between the lowest and highest monitored power are retained. The lowest monitored power is denoted as... This indicates the minimum output power required for monitoring; the maximum monitoring power is denoted as... This indicates the highest output power participating in the monitoring; both are monitoring configuration values ​​set during the site monitoring configuration. Standby and grid-connected transition records with output power lower than the minimum monitoring power are removed from monitoring. The grid-connected load range is defined by the output power itself, and this monitoring power range shares the minimum and maximum monitoring power boundary values ​​with the load interval division.

[0027] See Figure 2 In step S213, the monitoring processor 117 determines the cooling temperature rise based on the difference between the cooling outlet temperature and the cooling inlet temperature at the same sampling time. For the first... A complete record, with cooling temperature rise recorded as :

[0028]

[0029] in, Indicates the first The temperature rise is the difference between the cooling outlet temperature and the cooling inlet temperature, expressed in degrees Celsius. The cooling inlet temperature represents the thermal state of the cooling medium before it enters the liquid-cooled converter 112. This thermal state varies with the ambient temperature and the unit's operating conditions. After constructing the cooling temperature rise using the difference, this variation is subtracted. The cooling temperature rise directly reflects the heat loss carried away by the cooling medium as it flows through the liquid-cooled converter 112, constituting the longitudinal physical quantity in the relationship between load and temperature rise. The subsequent two normal ranges are both based on the cooling temperature rise as the longitudinal value. The cooling temperature rise increases with increasing output power, and its value is determined by the losses of the liquid-cooled converter 112 and the heat dissipation capacity of the cooling medium. Under the same load, the cooling temperature rise takes different normal values ​​depending on the load direction.

[0030] See Figure 2 In step S214, the monitoring processor 117 organizes the normal historical running data and the running data to be located into a sliding window according to a preset window duration and window step size. The window start time is recorded as... , indicating the first The start time of each sliding window; the window duration is denoted as... , representing the duration of each sliding window's coverage; the window step size is denoted as . , representing the difference in start time between two adjacent sliding windows. Both window duration and window step size are monitoring configuration values. Normal historical operating data and operating data to be located are organized into sliding windows with the same window duration and window step size. The window load and window temperature rise of both types of data have a consistent time scale. Starting from the first valid moment, the monitoring processor 117 establishes sliding windows by moving forward window by window with the window step size. The sliding window coverage sampling time is at The record within the sliding window shows that the end time is the sum of the window's start time and its duration. When the window step size is less than the window duration, adjacent sliding windows share a portion of the records, with each complete record serving as an input to all sliding windows that cover it. When the number of complete records within a sliding window reaches the minimum window point count, that sliding window participates in distinguishing between load increase and load decrease data, establishing normal ranges for load increase and load decrease, and confirming abnormal data; the minimum window point count is denoted as... This is a monitoring configuration value. If the number of complete records is less than the minimum number of window points for a sliding window, the data is insufficient and only participates in record saving.

[0031] For the sliding window involved in the above process, the median output power within the sliding window is taken as the window load, and the median cooling temperature rise within the sliding window is taken as the window temperature rise. The window load is denoted as... The window temperature rise is recorded as :

[0032]

[0033]

[0034] in, Indicates the first The median output power within each sliding window, in MW; Indicates the first The median of cooling temperature rise within a sliding window, expressed in degrees Celsius. When the number of records in the sliding window is even, the median is the average of the two middle values ​​after sorting by numerical value. When individual records in the window are affected by transient disturbances, the median keeps the window load and temperature rise at the level of most records in the window. When a single record in the window deviates significantly from the rest of the records in the same window, the median value is determined by the rest of the records, and the value is updated window by window as the window slides.

[0035] See Figure 2 In step S215, the monitoring processor 117 operates at the lowest monitoring power. With the highest monitoring power The load intervals are divided into continuous, equally wide sections according to a preset load interval. The load interval is denoted as... This represents the width of each load interval, a monitoring configuration value. Each sliding window is assigned a load interval number based on the load interval in which the window's load falls; the load interval number is denoted as [value]. :

[0036]

[0037] Each load interval is characterized by its center, which is denoted as . , indicating the first Central power of each load zone:

[0038]

[0039] in, The difference between the window load and the minimum monitored power is divided by the load interval and then rounded down. The resulting load interval number ensures that the window load falls within the corresponding load interval, and there is a one-to-one correspondence between the window load and the load interval number. The load range in question is the load location corresponding to the sliding window, and the load location is indicated by the load range number. Mark and center the load interval Characterization. Continuous, equally wide load intervals seamlessly cover the range from the lowest to the highest monitored power. Both normal ranges are established according to the load intervals and values ​​are taken at the center of the load intervals.

[0040] See Figure 2 In step S216, the monitoring processor 117 distinguishes between load increase data and load decrease data from a sliding window of normal historical operating data based on the change in output power over sampling time. The sliding windows are arranged according to their start times, and the load change is determined based on the difference in window load between two adjacent sliding windows. The load change is denoted as... :

[0041]

[0042] When the load change is positive and its absolute value reaches a preset directional threshold, the data in the corresponding sliding window represents load increase data; when the load change is negative and its absolute value reaches a directional threshold, the data in the corresponding sliding window represents load decrease data. The directional threshold is denoted as... This is the minimum change value used to distinguish between load increases, load decreases, and load stabilization, and is a monitoring configuration value. The load direction of the sliding window is denoted as... :

[0043]

[0044] in, Indicates the first The load change category for each sliding window is calculated based on the difference between the sliding window with a later start time and the sliding window with a earlier start time. The load direction is marked in the later sliding window. The load direction is... The data in the sliding window is the load increase data, and the load direction is... The data in the sliding window is the load reduction data, and the load direction is... The sliding window is the stable load window. If the load change in the stable load window does not reach the directional threshold, its data is neither included in the load increase data nor the load decrease data, and it is excluded from the calculation of the two broken lines. The directional distinction is performed within the normal historical operating data, so that normal temperature rise samples in the load increase direction and normal temperature rise samples in the load decrease direction are accumulated simultaneously within the same load range; the sliding window of the operating data to be located is uniformly compared with the two normal ranges in step S219, and its own load direction is irrelevant to the anomaly judgment.

[0045] See Figure 2 In step S217, the monitoring processor 117 statistically analyzes the temperature rise benchmarks in two directions within each load interval. The median of the window temperature rise for the load increase data is determined as the load increase center temperature rise, and the median of the window temperature rise for the load decrease data is determined as the load decrease center temperature rise. The load increase temperature rise fluctuation and the load decrease temperature rise fluctuation, respectively characterizing the dispersion of the window temperature rise, are then determined. The load increase center temperature rise is denoted as... The temperature rise at the load reduction center is recorded as The temperature rise fluctuation due to increased load is denoted as The temperature rise fluctuation due to load reduction is recorded as .by Standardize the marking direction. Pick or :

[0046]

[0047]

[0048] in, Number the load interval. Load interval number in normal historical operation data equal And the direction of the load equal The sliding window number. Temperature rise fluctuation is measured by taking the median of the absolute values ​​of the deviations of the window temperature rise from the center temperature rise in the same direction, multiplied by a conversion factor. This coefficient ensures that the temperature rise fluctuation is on the same order of magnitude as the standard deviation when the window temperature rise is normally discrete; the larger the temperature rise fluctuation, the greater the dispersion of the normal window temperature rise in that direction. Only when the number of load increase and load decrease data points within the same load range both reach the minimum number of directional points are the load increase center temperature rise, load decrease center temperature rise, load increase temperature rise fluctuation, and load decrease temperature rise fluctuation of that load range determined. The minimum number of directional points is denoted as... The minimum number of historical sliding windows required for each load interval in each direction is the monitoring configuration value. The minimum number of points in each direction is counted separately for each load interval and for each direction. The number of sliding windows for increasing load and decreasing load are counted independently. If the number of sliding windows in a certain direction is less than the minimum number of points in that direction, the sliding windows for the operating data to be located in that load interval will be excluded from the confirmation of abnormal data, and the temperature rise benchmark will be determined after the samples in that direction have accumulated to the minimum number of points in that direction.

[0049] See Figure 2 In step S218, the monitoring processor 117 connects the temperature rise of each load increase center in ascending order of load interval center to obtain a load increase curve, and connects the temperature rise of each load decrease center in the same order to obtain a load decrease curve. The load increase curve represents the relationship between normal cooling temperature rise and load during the load increase process, and the load decrease curve represents the relationship between normal cooling temperature rise and load during the load decrease process. The longitudinal difference between the two curves at the same load position is the normal thermal hysteresis under that load. The ascending order of load interval centers is the order of load positions from low to high. The horizontal coordinates of the two curves are the same, and the longitudinal values ​​of the two curves at the same load position are directly comparable. During one round trip of the load, the same load position experiences the values ​​on both the load increase curve and the load decrease curve, and the two curves together cover the normal temperature rise range of that load position.

[0050] Window load of sliding window Located at the center of two adjacent load zones and When the load is between adjacent load intervals, linear interpolation is performed based on the position of the window load between the centers of adjacent load intervals. Interpolation is performed separately for the load increase and load decrease line curves to obtain the current center temperature rise in the current direction, denoted as . Linear interpolation is performed on the temperature rise fluctuations during load increase and load decrease at this location to obtain the temperature rise fluctuation in the current direction, denoted as . :

[0051]

[0052]

[0053] in, Pick hour, and For the direction of load increase in the window load The central temperature rise and temperature rise fluctuation at the location; Pick hour, and To reduce load in the direction of The center temperature rise and temperature rise fluctuation at the location. When the window load coincides with the center of a certain load range, the center temperature rise and temperature rise fluctuation in the current direction are taken as the corresponding values ​​of that load range.

[0054] The temperature rise fluctuation in the current direction is then amplified using the corridor magnification factor. The corridor magnification factor is denoted as... , where is the amplification factor for the current temperature rise fluctuation, and is the monitoring configuration value; when the current temperature rise fluctuation is lower than the minimum temperature rise fluctuation, the minimum temperature rise fluctuation is used for amplification, and the minimum temperature rise fluctuation is denoted as . This represents the lower limit of the temperature rise fluctuation in the current direction, and is the monitoring configuration value. The upper limit of the corresponding direction is determined by the sum of the current direction center temperature rise and the amplified current direction temperature rise fluctuation, and the lower limit of the corresponding direction is determined by the difference between the current direction center temperature rise and the amplified current direction temperature rise fluctuation. The normal range for increasing load is defined by the upper and lower limits of the increasing load direction, the normal range for decreasing load is defined by the upper and lower limits of the decreasing load direction, and the common normal area is defined by the larger of the upper limits of the increasing load direction and the smaller of the lower limits of the decreasing load direction. The upper edge of the common normal area is the upper edge of the corridor. The lower side is the lower edge of the corridor. :

[0055]

[0056]

[0057] in, This indicates the maximum normal temperature rise allowed by the current sliding window. This indicates the minimum allowable normal temperature rise for the current sliding window, in degrees Celsius. At each load location, the common normal zone covers both the normal range for increasing and decreasing loads. Sliding windows with cooling temperature rises within this zone are considered normal regardless of whether measured by the normal path in the increasing or decreasing load direction. Thus, two types of normal thermal responses are retained at the same load location. The width of the common normal zone is determined by the temperature rise fluctuations in both directions, with the zone being wider at load locations with larger temperature rise fluctuations.

[0058] See Figure 2 In step S219, the monitoring processor 117 compares the window temperature rise of each sliding window of the operational data to be located with the normal range of load increase and load decrease at the load position corresponding to the window load of that sliding window. The output power of the operational data to be located is mapped to the load position of the common normal area based on the window load, and the upper limit of the load increase direction is obtained at that load position. Upper limit of the load reduction direction and the lower bounds in both directions , Window temperature rise Simultaneously, comparisons are made with upper and lower bounds in both directions. This comparison is performed for each sliding window of the operational data to be located, regardless of whether the load direction is increasing, decreasing, or stable. Through linear interpolation, continuous upper and lower bounds are obtained for the load in any window between the centers of two adjacent load intervals, and the comparison is performed window-by-window as the window load changes.

[0059] In step S220, it is determined whether the window temperature rise is located on the same outer side of the normal range for both increasing and decreasing loads. The same outer side means that the window temperature rise is simultaneously higher than the upper bound of both the increasing and decreasing load directions, or simultaneously lower than the lower bound of both the increasing and decreasing load directions. The window temperature rise is the median of the cooling temperature rise within the sliding window. If the window temperature rise is higher than the upper bound of one direction but not higher than the upper bound of the other direction, or if the window temperature rise is between the upper and lower bounds of either direction, it is considered a state that can still be explained by a certain normal load direction. If the determination is negative, return to step S219 and process the next sliding window according to the window step size; if the determination is positive, proceed to step S221. The determination is performed window by window in the order of the window start time. The determination of each sliding window is independent of each other, and the association between windows is established in step S222 based on the adjacency relationship.

[0060] In step S221, the window anomaly flag of the sliding window is set to one, and the data within the sliding window is confirmed as abnormal data. The window anomaly flag is denoted as... :

[0061]

[0062]

[0063] in, Indicates the first A window exception flag for each sliding window, with a value of zero or one; This means that it holds true in both directions simultaneously. This indicates that one of two simultaneous boundary crossing scenarios is met. The window anomaly flag is set to one if the window temperature rise simultaneously crosses the upper bound of two directions or the lower bound of two directions; otherwise, the window anomaly flag remains zero. Data for each sampling time covered by the sliding window is confirmed according to the window anomaly flag result. Data for each sampling time within a sliding window with a window anomaly flag set to one is confirmed as abnormal data. Since there are normal longitudinal differences between the load increase and decrease curves at the same load position, a window temperature rise that crosses the boundary in only one direction is still within the normal range of the other direction. If only the temperature rise relationship of a single load is considered, this normal longitudinal difference and the true anomaly are compressed into the same allowable range. Temperature rises that simultaneously cross two normal ranges cannot be distinguished from normal thermal hysteresis. This implementation uses simultaneous boundary crossing on both sides of the two curves as the criterion, thus distinguishing between normal thermal hysteresis and the true anomaly. The confirmation of abnormal data utilizes information from three aspects: load position, load direction, and temperature rise level.

[0064] See Figure 2 In step S222, the monitoring processor 117 assigns the sliding window containing the abnormal data to the abnormal window set. The abnormal window set is denoted as... :

[0065]

[0066] Within the abnormal window set, window adjacency relationships are established between two sliding windows whose window start time difference does not exceed the maximum adjacent interval and whose load interval number difference does not exceed the maximum interval hop count. All window adjacency relationships constitute the window adjacency set, denoted as […]. :

[0067]

[0068] in, The maximum adjacent interval represents the maximum allowed time interval between abnormal sliding windows. The time difference is calculated as the difference between the window start time of the subsequent sliding window and the window start time of the previous sliding window. `maximum interval jump count` represents the number of load intervals that abnormal sliding windows are allowed to cross; both are monitoring configuration values. When the maximum adjacent interval is set to the same order of magnitude as the window step size, abnormal sliding windows that move forward continuously according to the window step size establish window adjacency relationships one by one, and the windows that continuously cross the boundary are thus connected into a whole segment. The sliding windows connected along the window adjacency set constitute an abnormal window group, denoted as . , indicating the first A group of consecutive abnormal sliding windows belongs to the same abnormal window group. Two abnormal sliding windows are considered to be in the same abnormal window group if they are directly adjacent through window adjacency or reachable via other abnormal sliding windows. When the coverage duration of an abnormal window group reaches the shortest abnormal duration, the abnormal window group forms an abnormal time period. The coverage duration is the difference between the start time of the earliest window and the end time of the latest sliding window within the group; the shortest abnormal duration is denoted as... This is the minimum duration that an abnormal period must reach, and it is a monitoring configuration value. Two abnormal sliding windows whose start time difference exceeds the maximum adjacent interval, or whose load interval number difference exceeds the maximum interval jump number, belong to different abnormal window groups. The out-of-bounds window caused by a sudden load location crossing is thus separated from the consecutive out-of-bounds windows near the same load and formed into segments.

[0069] See Figure 2 In step S223, the monitoring processor 117 determines the start and end boundaries of each abnormal time period. The abnormal start time is the earliest window start time within the abnormal window group, and the abnormal end time is the latest sliding window end time within the abnormal window group, denoted as […]. and :

[0070]

[0071]

[0072] In step S222, the window adjacency set has established all window adjacency relationships. The start and end boundaries of each abnormal window group are directly obtained by the above formula. The abnormal time period is formed as the group closes. The boundary is the earliest window start time and the latest sliding window end time in the group. Both boundaries are directly derived from the window start time and window duration of each sliding window in the group.

[0073] See Figure 2 In step S224, the monitoring processor 117 determines the load range for each abnormal period. The lowest power of a period is the minimum load within the window during the abnormal period, and the highest power of a period is the maximum load within the window during the abnormal period, denoted as […]. and :

[0074]

[0075]

[0076] The minimum power and maximum power of the time period give the output power range covered by the abnormal period. Both are taken from the window load in the abnormal window group, reflecting the actual load range of the liquid-cooled converter 112 during the abnormal period. The load range in which the abnormality occurred can be directly read from this output power range.

[0077] See Figure 2 In step S225, the monitoring processor 117 generates a load-sharing anomaly period location result, including the target converter number, anomaly start time, anomaly end time, minimum power of the period, and maximum power of the period. The location result is given in the form that the load-sharing anomaly period for the target converter is from a certain start time to a certain end time, and the corresponding output power is from a certain minimum value to a certain maximum value: the target converter number indicates the liquid-cooled converter to which the anomaly period belongs, the anomaly start time and anomaly end time give the period boundary, and the minimum power of the period and the maximum power of the period give the load range of that period. The load-sharing anomaly period location result is written into the SCADA event log and displayed on the monitoring interface. Figure 1 The SCADA event recording and monitoring interface 118 is used. The circulating pump frequency, converter power command, and fan start / stop status still operate according to the existing control mode of the liquid-cooled converter 112. The operator locates the equipment according to the target converter number, traces back the operating data of the corresponding time period according to the abnormal start time and abnormal end time, and determines the load range covered by the abnormality according to the minimum power and maximum power of the time period.

[0078] Example 1:

[0079] For example, a direct-drive wind turbine generator set with a rated output of 5MW and a liquid-cooled full-power converter is used. The liquid-cooled converter 112 is supplied with circulating cooling medium by the cooling circuit 113. The cooling inlet temperature and cooling outlet temperature are measured by two temperature sensors originally installed in the equipment, each with a resolution of 0.1 degrees Celsius. The converter controller writes a record to the SCADA system every 10 seconds. Each record includes the output power, cooling inlet temperature and cooling outlet temperature. The three physical quantities share the same controller timestamp, and the sampling period is 10 seconds.

[0080] The period from 00:00:00 on June 1, 2026 to 23:59:50 on June 30, 2026 was selected as the normal historical period. A total of 259,200 records were sampled at 10-second intervals. Of these, 612 records were considered missing due to any of the three physical quantities being missing, and 46 records with duplicate timestamps retained only the last complete record. After processing, 258,542 complete records were obtained, which were used as the normal historical operational data. The operational data to be located was taken from the records from 14:00:00 to 14:10:00 on July 18, 2026.

[0081] Minimum monitoring power Take 0.50MW as the maximum monitoring power. Records of standby and grid-connected transitions with an output power below 0.50MW (5.00MW) are removed from monitoring. Window duration. Use 60 seconds, window step size Given a 10-second time limit, with each complete window containing 6 complete records, what is the minimum number of window points? If the total number of records in a sliding window is less than 5, it will be used for recording and saving only.

[0082] In step S213, the cooling temperature rise is determined for each complete record. In the record from 14:03:00 on July 18, 2026, the output power was 3.12MW, the cooling inlet temperature was 31.2 degrees Celsius, the cooling outlet temperature was 36.7 degrees Celsius, and the cooling temperature rise was... Degrees Celsius. In step S214, a sliding window with a starting time of 14:02:30 covers the period from 14:02:30 to 14:03:30, containing 6 complete records, and their output powers are as follows: MW, cooling temperature rise in sequence is The window load is the average of the two middle values ​​of output power, 3.12MW and 3.13MW, which is 3.125MW. The window temperature rise is the average of the two middle values ​​of cooling temperature rise, 5.47 degrees Celsius and 5.50 degrees Celsius, which is 5.485 degrees Celsius. The window load and window temperature rise cover the information of all 6 complete records within the window. A single record is only used as an input in each sliding window that covers it.

[0083] In step S215, the load interval Starting with 0.25MW, the power is sequentially formed from the lowest monitored power of 0.50MW. MW MW until There are 18 consecutive load zones of equal width, with the center of each load zone increasing sequentially in 0.25 MW increments starting from 0.625 MW. In step S216, the directional threshold... Taking 0.03MW as an example, an increase in window load of 0.03MW or more between adjacent sliding windows is recorded as a load increase, and a decrease in window load of 0.03MW or more is recorded as a load decrease. Window with an absolute load change of less than 0.03MW is considered a stable load window. In step S217, the minimum number of direction points... Taking 100, each load interval under monitoring in the normal historical period has more than 100 load increase sliding windows and load decrease sliding windows; a certain high load interval has only 83 load decrease sliding windows. The sliding windows for the operation data to be located in this load interval are excluded from the confirmation of abnormal data, and the temperature rise benchmark will be determined after the historical data is added.

[0084] In step S217, taking three adjacent load intervals with centers of 2.875MW, 3.125MW, and 3.375MW as examples, the temperature rise at the load center is as follows: The temperature rise at the load reduction center is as follows: (degrees Celsius) The temperature rise fluctuations under increased load are as follows: (Celsius, temperature fluctuations under increased load are...) The temperature rise fluctuations during load reduction are as follows: (Celsius, load reduction temperature fluctuations are...) Temperature in degrees Celsius. At the center of the same load range, the temperature rise at the center of decreasing load is greater than that at the center of increasing load. to The difference in degrees Celsius represents the normal thermal hysteresis; a normal temperature rise between the two broken lines does not constitute abnormal data.

[0085] In step S218, the minimum temperature rise fluctuation Take 0.05 degrees Celsius as the magnification factor for the corridor. Let's take 3. Taking a sliding window with a load of 3.06MW as an example, this window load is located between the center of the load range of 2.875MW and 3.125MW, with a positional ratio of... Linear interpolation yields the center temperature rise in the current load-increasing direction. Temperature rise at the center of the current load reduction direction: Celsius Temperature in degrees Celsius; temperature rise fluctuation in the current load increase direction Temperature fluctuation in Celsius, current load reduction direction The temperature rise fluctuations are both 0.05 degrees Celsius higher than the minimum temperature rise fluctuation, which is not included in the substitution. The upper limit of the load increase direction is... The upper limit of the load reduction direction is Celsius. Celsius, along the upper edge of the corridor Take the larger of the two values, 5.0250 degrees Celsius; the lower limit for the load increase direction is... The lower bound of the load reduction direction is Celsius. Temperature in Celsius, along the lower edge of the corridor Take the smaller of the two values, 3.9106 degrees Celsius.

[0086] In steps S219 to S221, for the operational data to be located from 14:00:00 to 14:10:00 on July 18, 2026, a sliding window is organized with the same window duration of 60 seconds and window step size of 10 seconds, and the window load and window temperature rise are calculated. The same comparison as with normal historical operational data is performed. For the sliding window starting at 14:01:00, the window load is 3.02MW and the window temperature rise is 4.78 degrees Celsius, which is within the common normal area of ​​the corresponding load position, and the window anomaly is marked as zero. For the sliding window starting at 14:01:10, the window load is 3.04MW and the window temperature rise is 4.91 degrees Celsius, which is higher than the upper limit of the load increase direction but not higher than the upper limit of the load decrease direction, and does not simultaneously cross the upper limits of both directions, so the window anomaly is marked as zero. The sliding window that started at 14:01:20 had a load of 3.06MW and a temperature rise of 5.08 degrees Celsius, which was 5.0250 degrees Celsius higher than the upper edge of the corridor. That is, it was higher than both the upper limit of 4.6150 degrees Celsius in the load increase direction and the upper limit of 5.0250 degrees Celsius in the load decrease direction. The window anomaly flag was set to 1 for the first time, and the data in this sliding window was confirmed as abnormal data.

[0087] The sliding window continues to move forward by the window step size. The window loads for the sliding windows starting at 14:01:30, 14:01:40, 14:01:50, 14:02:00, and 14:02:10 are as follows: MW, window temperature rise in sequence is After interpolation, the temperature rise of each window simultaneously exceeded the upper limit in both directions, and the abnormal window markers were all set to 1. From 14:02:20 to 14:06:50, all sliding windows maintained the same side high temperature exceeding the limit, with window loads varying between 3.15MW and 3.29MW, and window temperature rises varying between 5.36°C and 5.67°C. During this period, a cooling outlet temperature record was missing at 14:04:00. The sliding window at that time still had 5 complete records, reaching the minimum window point count of 5. The window load and window temperature rise were determined by the median of these 5 complete records, and the missing cooling outlet temperature was not interpolated.

[0088] The sliding window that started at 14:07:00 had a load of 3.20MW and a temperature rise of 5.24 degrees Celsius. The sliding window that started at 14:07:10 had a load of 3.18MW and a temperature rise of 5.19 degrees Celsius. This load was located between the center of the load range of 3.125MW and 3.375MW, with a positional ratio of [missing information]. Interpolation yields the center temperature rise in the current load direction. Temperature rise at the center of the current load reduction direction: Celsius The temperature rise fluctuation in the current load reduction direction is [degrees Celsius]. The upper limit of the load reduction direction is Celsius. The window temperature rise of 5.19 degrees Celsius was higher than the upper limit of 5.1428 degrees Celsius in the load reduction direction and also higher than the upper limit in the load increase direction, so the window anomaly marker remained at 1. At 14:07:20, the sliding window load was 3.17MW, and the window temperature rise was 5.12 degrees Celsius, which was lower than the upper edge of the corridor at the corresponding load position (5.13 degrees Celsius), so the window anomaly marker returned to zero. From then until 14:10:00, all sliding windows were within the common normal range for their corresponding load positions.

[0089] In step S222, the maximum adjacent interval Maximum number of jumps in a 20-second interval Set to 1, shortest anomaly duration A 60-second interval is used. From 14:01:20 to 14:07:10, each abnormal sliding window has a 10-second difference in start time between adjacent windows. The load interval numbers are the same or differ by 1. All windows connected along their adjacent sets form the same abnormal window group. This group consists of 36 abnormal sliding windows, from the window starting at 14:01:20 to the window starting at 14:07:10, with a 10-second window step. The latest sliding window in this group starts at 14:07:10 and ends at 14:08:10, while the earliest window starts at 14:01:20, with a coverage duration of 410 seconds, reaching the shortest abnormal duration of 60 seconds. This abnormal window group constitutes an abnormal period. When a single sliding window in a group has insufficient data due to missing complete records, if the difference in the window start time between the abnormal sliding windows before and after it does not exceed the maximum adjacent interval of 20 seconds and the difference in the load interval number does not exceed the maximum interval jump number of 1, they are still connected to the same abnormal window group by the window adjacency relationship; if the difference in the window start time reaches 30 seconds on both sides of the gap, the window adjacency relationship is broken, the abnormal sliding windows on both sides belong to two different abnormal window groups and the coverage duration is determined separately.

[0090] In step S223, the earliest window start time within the abnormal window group is 14:01:20 on July 18, 2026, which is the start time of the abnormality; the latest sliding window end time within the group is 14:08:10 on July 18, 2026, which is the end time of the abnormality. In step S224, the minimum load within the group is 3.06MW, and the maximum load is 3.29MW, which are the lowest and highest power during the time period. In step S225, the load distribution abnormality period location result is generated and written into the SCADA event record. The record content is: the load distribution abnormality period for converter C01 is from 14:01:20 to 14:08:10 on July 18, 2026, with a corresponding output power of 3.06MW to 3.29MW, where C01 is the target converter number of the liquid-cooled converter 112. The location result is also displayed on the monitoring interface.

[0091] See Figure 1 The above method is executed by the monitoring processor 117. The monitoring processor 117 receives three data streams from the cooling inlet temperature sensor 114, the cooling outlet temperature sensor 115, and the converter power measurement channel 116. Internally, it processes data in the following order: acquiring, determining the cooling temperature rise, distinguishing between load increase and load decrease data, establishing normal ranges for load increase and load decrease and forming a common normal area, confirming abnormal data, forming abnormal time periods, and outputting the results. This process corresponds to the method flow of steps S211 to S225. The output load-specific abnormal time period location results are written to the SCADA event log and monitoring interface 118. As normal historical operating data accumulates, the temperature rise benchmark and common normal area for each load interval are updated according to steps S217 and S218. The determination of the operating data to be located is executed according to steps S219 to S225 as the sliding window scrolls. The method runs continuously within the SCADA system. New location results are written to the SCADA event log one by one as abnormal time periods are formed. The monitoring interface displays the load-specific abnormal time periods of each liquid-cooled converter in the order of writing.

[0092] Example 2:

[0093] See Figure 3 The horizontal axis represents the output power of the liquid-cooled converter, and the vertical axis represents the cooling temperature rise of the corresponding sliding window. Figure 3 This is used to represent the relative relationship between two types of normal thermal responses and the operational data to be located under different load positions. The positional relationship between each curve and data point is used to illustrate the anomaly determination principle of this embodiment. The actual anomaly determination is still performed according to the method described in steps S211 to S225.

[0094] This embodiment uses the same wind turbine generator set and monitoring parameters as Embodiment 1. The rated output power of the wind turbine generator set is 5MW. The converter controller writes the output power, cooling inlet temperature, and cooling outlet temperature to the SCADA system at a sampling period of 10 seconds. The window duration is 60 seconds, the window step size is 10 seconds, the minimum monitored power is 0.50MW, the maximum monitored power is 5.00MW, the load interval is 0.25MW, the directional threshold is 0.03MW, the minimum number of directional points is 100, the minimum temperature rise fluctuation is 0.05 degrees Celsius, and the corridor magnification factor is 3.

[0095] First, based on the direction of load change between adjacent sliding windows in the normal historical operating data, the window data is divided into load increase data and load decrease data. Within the same load range, the window temperature rise corresponding to the load increase data and the load decrease data are statistically analyzed separately. The normal range of load increase is determined by the temperature rise at the center of load increase and the fluctuation of the load increase temperature rise, and the normal range of load decrease is determined by the temperature rise at the center of load decrease and the fluctuation of the load decrease temperature rise.

[0096] See Figure 3 At the same output power, there is a longitudinal difference in the normal cooling temperature rise corresponding to the load increase and load decrease processes. This longitudinal difference arises from the normal thermal hysteresis caused by the heat capacity of the power devices, cold plates, and cooling media. Therefore, the normal thermal response in the load increase and load decrease directions is retained separately in the data graph, rather than merging the data from the two directions into a single normal reference.

[0097] For each output power location, a common normal region is formed by the normal ranges corresponding to the load increase direction and the load decrease direction. Specifically, the larger of the two directional upper bounds is taken as the upper boundary of the common normal region, and the smaller of the two directional lower bounds is taken as the lower boundary of the common normal region. Figure 3 The common normal region extends continuously along the output power direction, so that the temperature rise during load increase and the temperature rise during load decrease caused by normal thermal hysteresis at the same load location are both included within the allowable normal range.

[0098] When the data point corresponding to the operational data to be located is within the common normal area, or only exceeds the normal range of load increase and load decrease, but can still be explained by the normal range of the other load direction, the data point is not confirmed as abnormal data.

[0099] For example, taking the operation data to be located on July 18, 2026 in Example 1, the window load of the sliding window starting at 14:01:00 is 3.02MW and the window temperature rise is 4.78 degrees Celsius. This data point is located in the common normal area of ​​the corresponding load position, so the window anomaly mark remains at 0.

[0100] The sliding window load that began at 14:01:10 was 3.04 MW, and the window temperature rise was 4.91 degrees Celsius. Although this temperature rise was higher than the upper limit for the load increase direction, it was not higher than the upper limit for the load decrease direction. Therefore, it could still be explained by the normal thermal response in the load decrease direction, and the window anomaly flag remained at 0. This data point... Figure 3 The data corresponds to data that is close to the common normal area boundary but has not yet constituted a crossover in both directions on the same side.

[0101] At 14:01:20, the sliding window load was 3.06 MW, and the window temperature rise was 5.08 degrees Celsius. At this load position, the upper boundary of the common normal zone determined by the normal range of increasing and decreasing loads is 5.0250 degrees Celsius. The window temperature rise of 5.08 degrees Celsius is higher than both the upper limit of the increasing load direction and the upper limit of the decreasing load direction, therefore it cannot be explained by either normal load direction. The window anomaly flag for this sliding window is set to 1.

[0102] Therefore, in Figure 3In this context, the data point corresponding to this window is located on the same outer side of the two normal ranges, that is, above the common normal area. Figure 3 The continuous abnormal data points located above the common normal region are used to represent the above-mentioned high temperature overshooting state on the same side in both directions.

[0103] As the sliding windows continued to advance in 10-second increments, multiple sliding windows, starting at 14:01:30, remained simultaneously above the normal upper limits in both directions. From 14:02:20 to 14:06:50, the sliding windows maintained a state of high temperature exceeding the limit on the same side, with window loads between 3.15MW and 3.29MW and window temperature rises between 5.36°C and 5.67°C. Therefore, these sliding windows... Figure 3 This forms a set of abnormal data points that are continuously distributed along the direction of output power on the upper side of the common normal region.

[0104] The sliding window load that started at 14:07:10 was 3.18MW, and the window temperature rise was 5.19 degrees Celsius. The upper limit of the load reduction direction obtained by interpolation at the corresponding load position is 5.1428 degrees Celsius. The window temperature rise is higher than the upper limits of both the load increase and load decrease directions, so the window anomaly flag is still 1.

[0105] The sliding window that started at 14:07:20 had a load of 3.17 MW and a temperature rise of 5.12 degrees Celsius, corresponding to an upper boundary of the common normal zone at that load location of 5.13 degrees Celsius. Since the temperature rise had fallen back below the common normal zone, the window anomaly flag was reset to 0. Subsequent sliding windows all remained within the common normal zone at their corresponding load locations.

[0106] As can be seen from the above continuous judgment results, abnormal data is manifested as a set of data points that appear continuously at adjacent time positions and adjacent load positions, and are all located on the same outer side of the normal range of load increase and load decrease.

[0107] Furthermore, sliding windows marked as abnormal (1) are categorized into an abnormal window set, and window adjacency relationships are established based on the maximum adjacent interval and the maximum interval jump count. In this embodiment, the maximum adjacent interval is 20 seconds, the maximum interval jump count is 1, and the shortest abnormal duration is 60 seconds.

[0108] The abnormal sliding windows that started between 14:01:20 and 14:07:10 all had a 10-second difference in the start time of their adjacent windows, and the load interval numbers of the adjacent windows were the same or differed by 1. Therefore, they were connected along the window adjacency relationship to form the same abnormal window group.

[0109] The earliest sliding window in the abnormal window group started at 14:01:20 on July 18, 2026, and the latest sliding window started at 14:07:10 and ended at 14:08:10. Therefore, the abnormal start time was determined to be 14:01:20 and the abnormal end time was determined to be 14:08:10.

[0110] Figure 3 In this context, T1 can be used to represent the start time of the anomaly, and T2 can be used to represent the end time of the anomaly. The data points between T1 and T2 represent continuous abnormal operation data that are classified into the same abnormal window group, so that the spatial boundary of the cooling temperature rise relative to the two types of normal thermal response can be further mapped to the abnormal period in the time dimension.

[0111] Statistics were compiled on all window loads within the abnormal window group. The minimum window load was 3.06MW and the maximum window load was 3.29MW. Therefore, the minimum power during this abnormal period was determined to be 3.06MW and the maximum power during this period was determined to be 3.29MW.

[0112] The final result of the load distribution anomaly location is as follows: the anomaly start time of the target converter C01 is 14:01:20 on July 18, 2026, and the anomaly end time is 14:08:10, with a corresponding output power range of 3.06MW to 3.29MW.

[0113] Depend on Figure 3 As shown in the data distribution, under the same output power, different normal cooling temperature rises are allowed in the load increase direction and the load decrease direction. Data that is between the two types of normal thermal responses or can still be explained by the normal range of one of the directions is not identified as abnormal. Abnormal data is only formed when the data to be located crosses the upper limit of both directions or crosses the lower limit of both directions at the same time.

[0114] In this way, normal thermal hysteresis is retained in the normal judgment criteria at different load locations, while continuous out-of-bounds data on the same side are further grouped into anomaly window groups according to sampling time and load location, so that the anomaly judgment result can simultaneously give the time range and load range of the anomaly occurrence.

[0115] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A method for locating abnormal load periods in wind power converters, characterized in that, include: The output power, cooling inlet temperature, and cooling outlet temperature of the liquid-cooled converter of the wind turbine generator set are obtained in relation to the sampling time. The obtained data includes normal historical operating data and operating data to be located. The cooling temperature rise is determined based on the difference between the cooling outlet temperature and the cooling inlet temperature at the same sampling time. Based on the change in output power over sampling time, load increase data and load decrease data are distinguished from normal historical operating data; Based on the load location represented by the output power, establish the normal range of load increase data and the normal range of load decrease data respectively. The two normal ranges of the same load location form a common normal area, so that the same load location retains two types of normal thermal response. The output power of the operating data to be located is mapped to the common normal area. Only when the cooling temperature rise at the same sampling time is located on the same outer side of the normal range of load increase and the normal range of load decrease, is the data at that sampling time confirmed as abnormal data. Based on the chronological order of sampling time, consecutive abnormal data are grouped into abnormal time periods, and the start and end times of the abnormal time periods are determined. Determine the minimum and maximum power during the abnormal period, and combine them with the target converter number, the start time of the abnormality, and the end time of the abnormality to form the result of load distribution abnormal period location.

2. The method for locating abnormal load periods in a wind power converter according to claim 1, characterized in that, According to the preset window duration and window step size, the normal historical operation data and the operation data to be located are organized into a sliding window; the median output power within the sliding window is taken as the window load, and the median cooling temperature rise within the sliding window is taken as the window temperature rise; a continuous load interval of equal width is divided between the minimum monitoring power and the maximum monitoring power according to the preset load interval, and the load interval number is determined according to the load interval in which the window load is located. Each load interval is represented by the center of the load interval, and the load position is the load interval in which the window load is located.

3. The method for locating abnormal load periods in a wind power converter according to claim 2, characterized in that, The load change is determined based on the difference in window load between two adjacent sliding windows; when the load change is positive and its absolute value reaches a preset directional threshold, the data for the corresponding sliding window is the load increase data; when the load change is negative and its absolute value reaches the directional threshold, the data for the corresponding sliding window is the load decrease data; the normal range for load increase is established based on the load increase data, and the normal range for load decrease is established based on the load decrease data.

4. The method for locating abnormal load periods in a wind power converter according to claim 2, characterized in that, Within each load interval, the median of the window temperature rise of the load increase data is determined as the load increase center temperature rise, and the median of the window temperature rise of the load decrease data is determined as the load decrease center temperature rise. The load increase temperature rise fluctuation and the load decrease temperature rise fluctuation, characterizing the dispersion of the window temperature rise, are determined respectively. Only when the number of load increase data and load decrease data within the same load interval reaches the minimum number of directional points are the load increase center temperature rise, load decrease center temperature rise, load increase temperature rise fluctuation, and load decrease temperature rise fluctuation of that load interval determined. Connecting the load increase center temperature rises in ascending order of the load interval center yields a load increase line, and connecting the load decrease center temperature rises yields a load decrease line. For both the load increase and load decrease lines, linear interpolation is performed according to the position of the window load between the centers of adjacent load intervals to obtain the current directional center temperature rise. The load increase temperature rise fluctuation is then analyzed according to this position. Momentum and temperature rise fluctuation under load reduction are linearly interpolated to obtain the temperature rise fluctuation in the current direction. The temperature rise fluctuation in the current direction is amplified by the corridor amplification factor. When the temperature rise fluctuation in the current direction is lower than the minimum temperature rise fluctuation, the minimum temperature rise fluctuation is used for amplification. The upper boundary of the corresponding direction is determined by the sum of the temperature rise at the center of the current direction and the amplified temperature rise fluctuation in the current direction, and the lower boundary of the corresponding direction is determined by the difference between the temperature rise at the center of the current direction and the amplified temperature rise fluctuation in the current direction. The normal range of load increase is defined by the upper and lower boundaries of the load increase direction, the normal range of load reduction is defined by the upper and lower boundaries of the load reduction direction, and the common normal region is defined by the larger of the upper boundary of the load increase direction and the smaller of the lower boundary of the lower boundary of the load reduction direction. The same outer side is when the cooling temperature rise is simultaneously higher than the upper boundary of the load increase direction and the lower boundary of the lower boundary of the load reduction direction.

5. The method for locating abnormal load periods in a wind power converter according to claim 2, characterized in that, For each sliding window of the operation data to be located, when the window temperature rise of the sliding window is located on the same outer side of the normal range of load increase and the normal range of load decrease at the load position corresponding to the window load of the sliding window, the window abnormality mark of the sliding window is set to one; the abnormal data is the data in the sliding window where the window abnormality mark is set to one.

6. The method for locating abnormal load periods in a wind power converter according to claim 2, characterized in that, The sliding window containing the abnormal data is categorized into an abnormal window set. Within the abnormal window set, two sliding windows whose window start times differ by no more than the maximum adjacent interval and whose load interval numbers differ by no more than the maximum interval jump number are established as window adjacency relationships. All window adjacency relationships constitute a window adjacency set. Sliding windows connected along the window adjacency set constitute an abnormal window group. The abnormal window group whose coverage duration reaches the shortest abnormal duration forms the abnormal period. The abnormal start time is the earliest window start time in the abnormal window group, and the abnormal end time is the latest sliding window end time in the abnormal window group.

7. The method for locating abnormal load periods in a wind power converter according to claim 2, characterized in that, The minimum power during the time period is the minimum of the window loads within the abnormal time period, and the maximum power during the time period is the maximum of the window loads within the abnormal time period; The results of locating the abnormal load periods are written into the SCADA event log and displayed on the monitoring interface.

8. The method for locating abnormal load periods in a wind power converter according to claim 2, characterized in that, The normal historical operating data and the operating data to be located are records from the same measuring point, the same unit, and the same sampling period. When there are multiple records at the same sampling time, the last written complete record is retained. The complete record is a record with complete output power, cooling inlet temperature, and cooling outlet temperature. When the number of complete records in the sliding window reaches the minimum number of window points, the sliding window participates in distinguishing the load increase data and the load decrease data, establishing the normal range of the load increase and the normal range of the load decrease, and confirming the abnormal data.

Citation Information

Patent Citations

  • Control method and control system of wind generating set converter cooling system

    CN108235656A

  • Wind turbine generator converter water cooling system cooling state evaluation and early warning method

    CN112084661A

  • Method for detecting abnormal rotating speed of generator of wind generating set

    CN115045807A

  • Wind turbine generator frequency converter fault early warning method based on deep convolution auto-encoder

    CN115578084A

  • Cooling system abnormity identification method and device, computer equipment and storage medium

    CN115633489A