Fault early warning method and device for anemometer based on adaptive sliding window division

Abstract

The application discloses a wind speed instrument fault early warning method and device based on adaptive sliding window division, and the method comprises the following steps: taking the average level of the wind speed of the wind fan in the normal state of the wind field as a reference, and combining a 3sigma abnormal data checking criterion to determine an alarm threshold; an adaptive size sliding window is used to slide detection on an online monitoring data stream; the time sequence of the average level of the wind speed of the wind speed instrument to be monitored in the sliding window and the wind speed of other wind fans is monitored, and when the average level exceeds the alarm threshold, an alarm is triggered; the sliding window division method based on combined information entropy can dynamically fuse the trend information of the online monitoring data stream of the equipment, adaptively adjust the size of the sliding window, focus on monitoring the abnormal mutation of the data distribution in real time, can fully capture and monitor the abnormal state of the wind speed instrument, and timely early warning.

Description

Technical Field

[0001] This invention belongs to the field of mechanical equipment condition monitoring, and particularly relates to a fault early warning method for anemometers, specifically a fault early warning method and device for anemometers based on adaptive sliding window partitioning. Background Technology

[0002] The wind speed parameters measured by an anemometer are crucial references for wind turbine start-up, shutdown, and pitch control. Their measurement accuracy directly affects the overall control and operation of the wind turbine, thus significantly impacting input power and electricity generation. Therefore, timely and accurate fault warnings for anemometers are of great practical significance for maintaining the safe and stable operation of wind turbines.

[0003] Based on the comparative analysis of differences between adjacent wind turbines in the same wind farm, the fault status of the anemometer of the wind turbine under test can be detected in a timely manner. When the anemometer is in an abnormal state or malfunctions, the residual distribution of the wind speed parameter will be significantly deviated from the residual distribution under the normal state benchmark level. Therefore, the wind speed index change trend of the wind turbine anemometer can be continuously monitored based on the mean and standard deviation of the residual sequence within the sliding window.

[0004] The commonly used sliding window method is characterized by a fixed window size and sliding step size, which moves sequentially and delineates equal-length sequence slices. However, the time-series monitoring of this method is easily affected by the window size: if the window is too large, the data base within a unit sequence slice is too large, resulting in poor dynamic real-time performance and making it difficult to respond to abnormal changes in the distribution of monitoring data caused by environmental interference or changes in operating conditions; if the window is too small, the data base within a unit sequence slice is insufficient. Although the real-time detection performance is improved, it is prone to local overfitting and may cause false detections for normal fluctuations in monitoring data or isolated monitoring caused by random or objective factors.

[0005] Therefore, selecting a sliding window of appropriate size can quickly and continuously reflect the changing trend of the wind speed residual sequence of the wind turbine, while improving the sensitivity of the anemometer's fault warning and reducing false alarms caused by an excessively small window. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a method and device for early warning of anemometer faults based on adaptive sliding window partitioning. This method can adaptively set the sliding window for the equipment monitoring data stream, taking into account the trend change characteristics of the time series, thereby accurately grasping the dynamic evolution law of the time series with the operating conditions. At the same time, it simplifies the difficulty of data processing by using symbolization, which can effectively provide early warning of anemometer faults and improve the timeliness and sensitivity of the warning.

[0007] To achieve the above objectives, the technical solution adopted by this invention is: a method for early warning of anemometer faults based on adaptive sliding window partitioning, comprising the following steps:

[0008] The average wind speed of the wind turbines under normal conditions in the same wind field is used as the benchmark, and the alarm threshold is determined in combination with the 3σ abnormal data verification criterion.

[0009] Based on the symbolic sliding window partitioning method of combined information entropy, an adaptive sliding window is designed;

[0010] Based on the adaptive-size sliding window, sliding detection is performed on the online monitoring data stream to monitor the time series of the average wind speed of the anemometer under monitoring and other wind turbines within the sliding window. If the average wind speed exceeds the alarm threshold, an alarm is triggered.

[0011] As a further optimization, the average wind speed of wind turbines under normal conditions in the same wind farm is used as a benchmark, and alarm thresholds are determined in conjunction with abnormal data verification criteria, including:

[0012] Using the wind speed meter of the wind turbine under test as a reference, the radius of the area where the adjacent wind turbines are located is determined by taking the reference as the center and the area of ​​the region where the airflow operation law in the wind field is similar. At the same time, all wind turbine units in this area are considered as adjacent wind turbines.

[0013] Using the anemometer of the unit under test as a reference point, the wind speed measured by each anemometer of the nearby wind turbine is obtained and mapped to the wind speed at the reference point position based on the wind speed profile model.

[0014] A baseline curve is plotted by mapping the hourly wind speed of all nearby wind turbines to the reference point and taking the average value. Its probability distribution can be approximately described by a normal distribution. The position parameter of the baseline curve is calculated as the mean μ and the scale parameter is σ. A 3σ criterion model is established.

[0015] The 3σ criterion is used to determine the residual between the measured anemometer data and the fitted data. If the residual is not within the interval (μ-3σ, μ+3σ), it is determined to be an outlier.

[0016] As a further optimization, based on the symbolic sliding window partitioning method of combined information entropy, an adaptive-size sliding window is designed, including:

[0017] Identify the extreme points in a time series where the fluctuation amplitude reaches a set level, and the data points with large fluctuations within a set time period;

[0018] The data stream is statically encoded and dynamically encoded separately, and the static and dynamic sequences are jointly encoded. The joint encoded data is then symbolized to obtain a symbolic data stream.

[0019] Calculate the entropy of the symbolic combination information of the symbolic data stream;

[0020] The sliding window is divided based on the symbolic combination information entropy.

[0021] As a further optimization, the extreme point where the fluctuation amplitude reaches a set level is determined according to the following principles:

[0022] Given a constant R, the points that satisfy the following formula (excluding the left and right endpoints) are the extreme points where the fluctuation amplitude reaches the set level.

[0023]

[0024] x m >x m-1 x m / x m-1 >R or x m <x m-1 x m-1 / x m >R

[0025] x m Points in a time series;

[0026] When given a data point with significant fluctuations within a given time period, and a constant K, if two directly adjacent points satisfy the following formula, then x... m For data points that fluctuate significantly in a short period of time,

[0027]

[0028] x m These are points in a time series.

[0029] As a further optimization, the data stream is statically encoded and dynamically encoded separately, and the static and dynamic sequences are jointly encoded. The symbolization process of the joint encoding includes:

[0030] Static encoding reflects the amplitude of a time series. A threshold is set, and values ​​above the threshold are set to state "1", while values ​​below the threshold are set to state "0".

[0031] Dynamic encoding reflects the trend of the sequence. It compares two adjacent trend turning points, setting the upward trend as state "1" and the downward trend as state "0".

[0032] The static and dynamic codes are combined to encode four types: 00, 01, 10, and 11, and are assigned the symbols a, b, c, and d respectively.

[0033] As a further optimization, the information entropy of discrete random sequences is improved, and the formula for calculating the symbolic combination information entropy is as follows:

[0034] H(X J)=-p(x a log2(p(x) a ))-p(x b log2(p(x) b ))

[0035] -p(x c log2(p(x) c ))-p(x d log2(p(x) d ))

[0036] Where p(x) a p(x) represents the proportion of the symbol "a" in the symbol sequence. a p(x) represents the proportion of the symbol "b" in the symbol sequence. a p(x) represents the proportion of the symbol "c" in the symbol sequence. a ) represents the proportion of the symbol "d" in the symbol sequence.

[0037] As a further optimization, the sliding window is divided based on the symbolic combination information entropy, including:

[0038] Acquire the online monitoring data stream of the device under normal conditions for a set time period, with the time slice width L from the sliding window width adjustment point up to the current inflow data stream; calculate the mean μ of the information entropy stream, and set thresholds σ and σ'. max As a reference value, the mean μ of the current sliding window sequence is calculated in real time. i Calculate Δ i =|μ i -μ|, sliding window width w new as follows:

[0039]

[0040] When the mean value of information entropy stream μ i The changes are not significant, that is When the sliding window maintains its initial width w, the mean of the information entropy stream μ... i Significant changes, namely When the sliding window width decreases and the current time slice is re-probing, it is possible to better monitor and analyze the local trends of the online monitoring stream time series; when the mean μ is within a continuous time slice of width 3w... i If all return to normal, the sliding window will return to its initial width w.

[0041] As a further optimization, based on the adaptive-sized sliding window, sliding detection is performed on the online monitoring data stream to monitor the time series of the average wind speed of the anemometer under monitoring and other wind turbines within the sliding window. If the average wind speed exceeds the alarm threshold, an alarm is triggered, including:

[0042] The wind speed measured by the anemometer of the target unit is compared with the baseline curve obtained by averaging the wind speeds of nearby units. The mean and standard deviation of the residuals are calculated as monitoring statistics. The monitoring statistics stream is slidable using the sliding window, and the width of the sliding window is adaptively adjusted according to the trend of the time series. If the mean or standard deviation of the residuals exceeds the range of (μ-3σ, μ+3σ), the data is judged to be abnormal, and the position of the sliding window is recorded. When there are data abnormalities in three consecutive sliding windows, it is confirmed that the anemometer of the target wind turbine is faulty, and an alarm message is issued.

[0043] In addition, the present invention also provides a computer device, including a processor and a memory, wherein the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and the processor can implement the anemometer fault early warning method based on adaptive sliding window partitioning described in the present invention when executing the computer executable program.

[0044] Simultaneously, a computer-readable storage medium can be provided, which stores a computer program. When the computer program is executed by a processor, it can implement the anemometer fault early warning method based on adaptive sliding window partitioning described in this invention.

[0045] Compared with existing technologies, the present invention has at least the following beneficial effects: The present invention proposes a method for early warning of anemometer faults based on an adaptive sliding window. This method constructs a baseline curve model of wind turbine speedometers in the same wind field and introduces a symbolic sliding window partitioning method based on combined information entropy to perform fault discrimination based on the 3σ criterion alarm threshold for continuous sliding time windows, thereby improving the sensitivity and accuracy of anemometer abnormal state monitoring and fault warning; The above-mentioned sliding window early warning method based on combined information entropy, compared with the method of using a fixed sliding time window for anemometer fault warning, can dynamically integrate the trend information of the online monitoring data stream of the equipment, adaptively adjust the size of the sliding window, and focus on abnormal changes in the distribution of monitoring data in real time, so as to fully capture and monitor the abnormal state of the anemometer and provide timely warnings.

[0046] Furthermore, based on the wind speed profile model mapped to the wind speed at the reference point, the wind speed obtained by each wind turbine within the target area is down-standardized to eliminate the influence of different heights; thus providing basic data for drawing the baseline curve for adjacent wind turbines.

[0047] Furthermore, this invention proposes a symbolic sliding window partitioning method based on combined information entropy. This method realizes the pre-partitioning of the online monitoring data stream of the device, extracts the trend turning points of the time series and carries out dynamic and static symbolic hybrid encoding, then calculates the combined information entropy of the symbolic sequence, and adaptively changes the sliding window width based on this. Attached Figure Description

[0048] Figure 1 The flowchart of the symbolic online device monitoring data stream sliding window partitioning method based on combined information entropy provided by the present invention is shown below.

[0049] Figure 2 The flowchart of the anemometer fault early warning method based on adaptive sliding window provided by the present invention is shown. Detailed Implementation

[0050] The present invention will now be described in further detail through specific implementation steps and in conjunction with the accompanying drawings.

[0051] Determination of alarm threshold level for wind turbines in the same wind farm based on the 3σ criterion (using the average level of wind turbines in the same wind farm as a benchmark, and combining it with 3σ to determine the alarm threshold).

[0052] (1) Fan selection and fixed area radius

[0053] In wind farms, wind turbine sites are typically selected in open environments, and anemometers are usually installed at a certain height. Therefore, near-surface atmospheric motion within a certain range generally follows common patterns. However, airflow caused by surrounding buildings, terrain, and obstacles can also affect wind speed measurements. Therefore, it is necessary to conduct a site survey of the wind farm's geographical environment to determine the radius (in meters). Using the anemometer being measured as a reference, it is assumed that the airflow patterns within a circular area centered on the reference anemometer and defined by the determined radius are approximately consistent. All wind turbines within this area are considered neighboring turbines.

[0054] (2) Time interval determination

[0055] To express wind speed, the average wind speed over a certain time interval is usually used. In the "Ground Meteorological Observation Specification", the average wind speed measured by different anemometers refers to the average value over a specified time (3s, 2min, 10min). Based on the research of existing observation structures, the average wind speed over different time intervals does not decrease as the time interval increases. Considering the convenience of data processing, the average time interval is set to one hour. The average data of 30 minutes before and after each hour of each wind turbine speed meter is selected as the wind speed of that speed meter for that hour.

[0056] (3) Wind speed mapping based on wind speed profile model

[0057] Wind speed profile models can be used to reflect the relationship between wind speed and altitude. The anemometer of the unit under test is selected as the reference location point. The wind speed measured by each anemometer of the neighboring wind turbines in the specified area is substituted into the following formula to map the wind speed at the reference point location. The influence of wind speed caused by geographical location differences should be eliminated.

[0058]

[0059] Among them, v r The wind speed at the reference location is v0, and the wind speed at the original location is h. r h0 is the height at the reference position, and h0 is the height at the original position.

[0060] The alarm threshold is based on the 3σ criterion. The 3σ criterion first assumes that a set of detection data contains only random errors, calculates and processes it to obtain σ and μ, and determines an interval (μ-3σ, μ+3σ). The probability of the set of data being in the (μ-3σ, μ+3σ) interval is 99%, so values ​​exceeding this interval can be judged as outliers.

[0061] The hourly wind speed data v of all nearby units mapped by the wind speed profile. r The average value is used to plot the baseline curve, whose probability distribution can be approximately described by a normal distribution. The position parameter of the baseline curve is calculated as μ (mean) and the scale parameter is calculated as σ. A 3σ criterion model is established.

[0062] The 3σ criterion is used to determine the residual between the measured anemometer data and the fitted data. If the residual is within the interval (μ-3σ, μ+3σ), it is considered a normal value; otherwise, it is considered an outlier.

[0063] Design a symbolic sliding window based on combined information entropy.

[0064] The sliding window partitioning method designed in this invention is as follows: Figure 1 As shown, the trend points of the time series are first extracted and encoded using a hybrid dynamic and static symbolization method. Then, the combined information entropy of the selected data trend points is used to divide the sliding window. Detailed design content is as follows:

[0065] 1. Identify the trend turning points of a time series.

[0066] For a time series {(x1, t1), ... (x m-1 , t m-1 ), (x m , t m ), ...(x n , t n Extract the following two types of trend turning points in time series:

[0067] (1) The extreme point where the fluctuation amplitude reaches the set level, that is, given a constant R, the point that satisfies the following formula except for the two endpoint values.

[0068]

[0069] x m >x m-1 x m / xm-1 >R or x m <x m-1 x m-1 / x m >R

[0070] (2) For data points that fluctuate significantly in a short period of time, i.e., given a constant K, if two directly adjacent points satisfy the following formula, then x m for.

[0071]

[0072] 2. Mixed encoding of static and dynamic symbols

[0073] The data streams are jointly encoded and symbolized, and the specific steps are as follows:

[0074] (1) Static coding reflects the amplitude of the time series. A threshold line is set. If the value is higher than the threshold, it is set to state "1" and if the value is lower than the threshold, it is set to state "0".

[0075] (2) Dynamic encoding reflects the trend of the sequence. It compares two adjacent trend turning points, setting the upward trend as state "1" and the downward trend as state "0".

[0076] (3) Coding and symbolization

[0077] The static and dynamic sequences are jointly encoded and then symbolically represented according to the following rules, thereby converting the numerical sequence into a symbolic sequence, as shown in the table below:

[0078]

[0079] 3. Calculate the combined information entropy

[0080] For the four symbols “a”, “b”, “c”, and “d” obtained above, the amplitude and trend information covered by the online monitoring data stream are fused from the perspective of information theory using combined information entropy, so that the time series contains the dynamic characteristics of the four symbols simultaneously.

[0081] Information entropy is used to measure the degree of orderliness of a discrete time series; the magnitude of the time series increases with the increasing degree of discretization. This invention improves upon the information entropy of discrete random sequences, proposing the following formula for symbolic combination information entropy:

[0082] H(X J )=-p(x a log2(p(x) a ))-p(x b log2(p(x) b ))-p(x clog2(p(x) c ))-p(x d log2(p(x) d ))

[0083] Where p(x) a p(x) represents the proportion of the symbol "a" in the symbol sequence. a p(x) represents the proportion of the symbol "b" in the symbol sequence. a p(x) represents the proportion of the symbol "c" in the symbol sequence. a ) represents the proportion of the symbol "d" in the symbol sequence.

[0084] 4. Divide the sliding window based on information entropy

[0085] Acquire the online monitoring data stream of the device under normal conditions over a period of time. Let the time slice width of the incoming data stream be L, from the point where the sliding window width is adjusted to the current data slice width. Calculate the mean value μ of the information entropy stream, and set thresholds σ and σ0. max As a reference value, the mean μ of the current sliding window sequence is calculated in real time. i .

[0086] Calculate Δ i =|μ i -μ|, the sliding window width is defined as:

[0087]

[0088] When the mean value of the information entropy stream μi does not change much (i.e.) When the information entropy stream mean μ is reached, the sliding window maintains its initial width w; when the information entropy stream mean μ is reached, the sliding window maintains its initial width w. i Significant changes (i.e.) When the sliding window width decreases and the current time slice is re-probing, it is used to better monitor and analyze the local trends of the online monitoring stream time series; when the mean μ is within a continuous time slice of width 3w... i If all return to normal, the sliding window will return to its initial width w.

[0089] Speedometer status monitoring and fault early warning based on adaptive sliding window.

[0090] The wind speed measured by the anemometer of the target unit is compared with the baseline curve obtained by averaging the wind speeds of nearby units. The mean and standard deviation of the residuals are calculated as monitoring statistics. A designed sliding window is used to slide the monitoring data stream, and the width of the sliding window is adaptively adjusted according to the trend characteristics of the time series.

[0091] The residuals between the anemometer data and the baseline data of the time slice within the sliding window are judged according to the 3σ criterion. If the mean or standard deviation of the residuals exceeds (μ-3σ, μ+3σ), the data is judged to be abnormal, and the position of the sliding window is recorded.

[0092] Because the sliding window's adaptive width design can accurately track the trend information of data distribution, when the data distribution tends to be stable and the fluctuation range is small, the monitoring sequence data base covered by the sliding window is large. However, when the data distribution trend changes significantly and the fluctuation range is large, the monitoring information data base covered by the sliding window is small but contains rich trend change information. When there are abnormal data in three consecutive sliding windows, it is determined that the anemometer of the target wind turbine is faulty.

[0093] An alarm will be triggered if three consecutive sliding windows malfunction or if the first sliding time window malfunctions. The fault information can be pushed to the control center and terminal equipment. The operator can also monitor the status display on the page to make a detailed judgment on the fault information.

[0094] In summary, this invention proposes a method for anemometer fault early warning based on adaptive sliding window partitioning. This method constructs a baseline curve model of wind turbine speedometers in adjacent wind fields and introduces a symbolic sliding window partitioning method based on combined information entropy to perform fault discrimination based on the 3σ criterion alarm threshold for continuous sliding time windows. This improves the sensitivity and accuracy of anemometer abnormal state monitoring and fault early warning. Compared with methods that use fixed sliding time windows for anemometer fault early warning, the aforementioned sliding window early warning method based on combined information entropy can dynamically integrate the trend information of the online monitoring data stream of the equipment, adaptively adjust the size of the sliding window, and focus on abnormal changes in the distribution of monitoring data in real time. It can fully capture and monitor the abnormal state of the anemometer and provide timely early warning.

[0095] In addition, the present invention can also provide a computer device, including a processor and a memory, wherein the memory is used to store a computer executable program, the processor reads part or all of the computer executable program from the memory and executes it, and when the processor executes part or all of the computer executable program, it can realize the anemometer fault early warning method based on adaptive sliding window partitioning described in the present invention.

[0096] On the other hand, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the anemometer fault early warning method based on adaptive sliding window partitioning described in the present invention.

[0097] The computer device may be a laptop, a desktop computer, or a workstation.

[0098] The processor can be a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or an off-the-shelf programmable gate array (FPGA).

[0099] The memory described in this invention can be an internal storage unit of a laptop, desktop computer, or workstation, such as memory or hard disk; or it can be an external storage unit, such as a portable hard disk or flash memory card.

[0100] Computer-readable storage media can include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer-readable storage media can include: read-only memory.

[0101] (ROM, Read Only Memory), Random Access Memory (RAM), Solid State Drives (SSD), or Optical Discs, etc. Among them, Random Access Memory can include Resistive Random Access Memory (ReRAM) and Dynamic Random Access Memory (DRAM).

Claims

1. A method for early warning of anemometer faults based on adaptive sliding window partitioning, characterized in that, Includes the following steps: The average wind speed of the wind turbines under normal conditions in the same wind field is used as the benchmark, and the alarm threshold is determined in combination with the 3σ abnormal data verification criterion. A symbolic sliding window partitioning method based on combined information entropy is used to design an adaptive sliding window. Specifically, this includes: identifying extreme points in the time series where the fluctuation amplitude reaches a set level and data points with large fluctuations within a set time period; performing static and dynamic encoding on the data stream respectively, jointly encoding the static and dynamic sequences, and symbolizing the joint encoding to obtain a symbolic data stream; calculating the symbolic combined information entropy of the symbolic data stream; and partitioning the sliding window based on the symbolic combined information entropy. The partitioning of the sliding window based on the symbolic combined information entropy includes: Acquire the online monitoring data stream of the device under normal conditions over a period of time, with the time slice width from the point where the sliding window width is adjusted up to the current inflow data stream being L; calculate the mean of the information entropy stream. Set threshold and As a reference value, the mean of the current sliding window sequence slices is calculated in real time. ,calculate Sliding window width as follows: When the information entropy flow mean The changes are not significant, that is At that time, the sliding window maintains its initial width. When the information entropy stream mean Significant changes, namely When the sliding window width decreases and the current time slice is re-probing, it is used to better monitor and analyze the local trends of the online monitoring stream time series; when the width is 3... If the mean value within a continuous time slice If all values ​​return to normal, the sliding window will return to its initial width. ; Based on the adaptive-size sliding window, sliding detection is performed on the online monitoring data stream to monitor the time series of the average wind speed of the anemometer under monitoring and other wind turbines within the sliding window. If the average wind speed exceeds the alarm threshold, an alarm is triggered.

2. The anemometer fault early warning method based on adaptive sliding window partitioning according to claim 1, characterized in that, The alarm threshold is determined by using the average wind speed of the wind turbines under normal conditions in the same wind farm as a benchmark, and combining it with the abnormal data verification criteria. Using the wind speed meter of the wind turbine under test as a reference, the radius of the area where the adjacent wind turbines are located is determined by taking the reference as the center and the area of ​​the region where the airflow operation law in the wind field is similar. At the same time, all wind turbine units in this area are considered as adjacent wind turbines. Using the anemometer of the unit under test as a reference point, the wind speed measured by each anemometer of the nearby wind turbine is obtained and mapped to the wind speed at the reference point position based on the wind speed profile model. A baseline curve is plotted by mapping the hourly wind speed of all nearby wind turbines to the reference point location and taking the average value. Its probability distribution can be approximately described by a normal distribution. The position parameter of the baseline curve is calculated as the mean μ and the scale parameter is σ. A 3σ criterion model is established. The 3σ criterion is used to determine the residual between the measured anemometer data and the fitted data. If the residual is not within the interval (μ -3σ, μ+3σ), it is determined to be an outlier.

3. The anemometer fault early warning method based on adaptive sliding window partitioning according to claim 1, characterized in that, The extreme point where the fluctuation amplitude reaches the set level is determined according to the following principles: Given constant Except for the left and right endpoints, the points that satisfy the following formula are the extreme points where the fluctuation amplitude reaches the set level. Points in a time series; When setting data points with large fluctuations within a given time period, a constant is given. If two directly adjacent points satisfy the following formula, then For data points that fluctuate significantly in a short period of time, These are points in a time series.

4. The anemometer fault early warning method based on adaptive sliding window partitioning according to claim 1, characterized in that, The data stream is statically and dynamically encoded separately, and the static and dynamic sequences are jointly encoded. The symbolization of the joint encoding includes: Static encoding reflects the amplitude of a time series. A threshold is set, and values ​​above the threshold are set to state "1", while values ​​below the threshold are set to state "0". Dynamic encoding reflects the trend of the sequence. It compares two adjacent trend turning points, setting the upward trend as state "1" and the downward trend as state "0". The static and dynamic codes are combined to encode four types: 00, 01, 10, and 11, and are assigned the symbols a, b, c, and d respectively.

5. The anemometer fault early warning method based on adaptive sliding window partitioning according to claim 1, characterized in that, An improved formula for calculating the information entropy of discrete random sequences is as follows: ( ) ( ) ( ) ( ) in, It represents the proportion of the symbol "a" in the symbol sequence. It represents the proportion of the symbol "b" in the symbol sequence. It represents the proportion of the symbol "c" in the symbol sequence. It represents the proportion of the symbol "d" in the symbol sequence.

6. The anemometer fault early warning method based on adaptive sliding window partitioning according to claim 1, characterized in that, Based on the adaptive-size sliding window, sliding detection is performed on the online monitoring data stream to monitor the time series of the average wind speed of the anemometer under test and other wind turbines within the sliding window. If the average wind speed exceeds the alarm threshold, an alarm is triggered, including: The wind speed measured by the anemometer of the target unit is compared with the baseline curve obtained by averaging the wind speeds of nearby units. The mean and standard deviation of the residuals are calculated as monitoring statistics. The monitoring statistics stream is slidable using the sliding window, and the width of the sliding window is adaptively adjusted according to the trend of the time series. If the mean or standard deviation of the residuals exceeds the range of (μ -3σ, μ+3σ), the data is judged to be abnormal, and the position of the sliding window is recorded. When there are data abnormalities in three consecutive sliding windows, it is confirmed that the anemometer of the target wind turbine is faulty, and an alarm message is issued.

7. A computer device, characterized in that, It includes a processor and a memory, the memory being used to store a computer-executable program, the processor reading the computer-executable program from the memory and executing it, and the processor executing the computer-executable program being able to implement the anemometer fault early warning method based on adaptive sliding window partitioning as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the anemometer fault early warning method based on adaptive sliding window partitioning as described in any one of claims 1 to 5.

Images