A wind power monitoring data fusion preprocessing method and system based on timestamp and running state matching

Abstract

The application discloses a wind power monitoring data fusion preprocessing method and system based on timestamp and running state matching, and comprises the following steps: determining the wind turbine running state switching position by extracting the state transition boundary in the pitch angle, rotating speed, cabin azimuth and active power; then correcting the time position of the vibration, bearing temperature, yaw and variable pitch monitoring data according to the state transition boundary sequence, weakening the influence of the multi-channel sampling delay difference on the fusion result; further combining the time credible sequence and the state credible sequence to generate a sample weight sequence, eliminating the mismatched sample section and keeping the continuous sample section, improving the boundary consistency and continuity of the fusion sample from the source, and finally improving the accuracy, stability and subsequent analysis reliability of the wind power monitoring data fusion preprocessing.

Description

Technical Field

[0001] This invention relates to the technical field of wind turbine operation monitoring data processing and fusion preprocessing, and in particular to a wind power monitoring data fusion preprocessing method and system based on timestamp and operation status matching. Background Technology

[0002] With the continuous expansion of wind power installed capacity and the increasing complexity of single-unit control strategies, wind turbine operation monitoring has gradually evolved from single-parameter acquisition to a multi-monitoring channel collaborative analysis mode. Existing wind power monitoring chains typically simultaneously access multiple monitoring parameters such as pitch angle, speed, nacelle azimuth, active power, vibration, temperature, yaw, and pitch, using them as the basic inputs for fault diagnosis, health assessment, performance analysis, and operation adjustment. In actual wind farm deployment environments, the sampling cycle, buffering method, transmission delay, and edge-side processing order of different acquisition channels are not entirely consistent. Even if each monitoring channel is accompanied by a timestamp, the response position of the same operating event in different monitoring channels may still be offset. When such offsets enter subsequent analysis stages, they can easily cause feature splicing distortion, state boundary misalignment, continuous sample segment breakage, and anomaly judgment bias. Therefore, conducting fusion preprocessing to ensure the consistency of time position and state boundary of multi-source monitoring data during the operation state switching process has become one of the fundamental issues in wind turbine operation monitoring technology.

[0003] CN112459970A discloses a wind turbine anomaly detection method using data adversarial learning. This scheme preprocesses the wind turbine monitoring data, extracts normal data samples, and constructs a generative adversarial network to complete the anomaly detection. Its technical focus is on improving the anomaly recognition capability through adversarial learning. However, its processing focus is still on the anomaly detection model itself. The preprocessing steps mainly serve the subsequent model training and do not establish a dedicated correction mechanism for the time position mismatch caused by differences in sampling delay, buffer delay, and response order among multiple monitoring channels. Nor does it constrain the correspondence between the operating state transition position and the response position of each monitoring channel. Therefore, in operating state switching scenarios such as wind turbine start-up and shutdown, yaw adjustment, pitch adjustment, and power fluctuation, this type of technology may still have problems such as asynchronous responses of each monitoring channel to the same state transition, disordered boundary sequence, and insufficient continuity of fused sample segments.

[0004] Furthermore, existing wind power monitoring and analysis technologies also include schemes that focus on constructing operational status datasets, dynamic analysis, or performance anomaly identification. For example, by establishing operational status datasets and analyzing the unit's operational status, the accuracy and reliability of performance anomaly detection can be improved. Although such schemes can identify abnormal states from the perspective of unit operational characteristics, they mainly focus on anomaly result judgment or state trend analysis, rather than specifically processing the time axis offset, state transition boundary preservation, and selection of retainable sample segments of multi-source monitoring data before the monitoring data enters the identification stage. In other words, most existing technologies assume that multi-channel data can be directly merged, resampled, or uniformly processed based on the original timestamps. They lack a preprocessing mechanism that uses reverse constraint time correction based on operational status transition boundaries and combines boundary preservation to retain and remove sample segments. Therefore, it is difficult to reduce the cumulative error of misalignment in the fusion results from the data source.

[0005] In summary, existing wind power monitoring data processing technologies still suffer from problems such as time-location mismatch of monitoring channels, difficulty in unifying state transition boundaries, and insufficient continuity of fused sample segments. This invention proposes a wind power monitoring data fusion preprocessing method based on timestamp and operating state matching. This method sequentially arranges pitch angle monitoring data, speed monitoring data, nacelle azimuth monitoring data, and active power monitoring data and extracts the on-chain state transition boundary sequence. Then, based on the on-chain state transition boundary sequence, it extracts response positions from vibration monitoring data, bearing temperature monitoring data, yaw monitoring data, and pitch monitoring data, generating a channel offset sequence and a corrected time axis. Furthermore, it combines a reliable time sequence, a reliable state sequence, and a sample weight sequence to complete the removal of boundary mismatched sample segments and the retention of continuous boundary sample segments. This effectively addresses the problems of inaccurate alignment and stable fusion of multiple monitoring channels of wind turbines before and after operating state switching. Summary of the Invention

[0006] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0007] In view of the aforementioned existing problems, the present invention is proposed.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a wind power monitoring data fusion preprocessing method based on timestamp and operating status matching, including: arranging the pitch angle monitoring data, speed monitoring data, nacelle orientation monitoring data and active power monitoring data of the same wind turbine in the same monitoring batch according to the original timestamp, filtering out the corresponding direction reversal positions and incremental jump positions between adjacent monitoring points, and generating a chain state transition boundary sequence in chronological order; Based on the chain state transition boundary sequence, the response positions located in the intervals before and after each boundary are extracted from the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data and pitch monitoring data of the same wind turbine. A channel offset sequence is generated according to the difference between the response positions and the corresponding boundary positions. The original timestamps of each monitoring data involved in the correction are shifted and corrected according to the channel offset sequence to generate a corrected time axis. A reliable time sequence is generated based on the deviation of adjacent correction time intervals in the correction time axis. A reliable state sequence is generated based on the change in the order of the boundary transitions of the on-chain state transition boundary sequence before and after correction and the offset of the interval between adjacent boundaries. A sample weight sequence is generated based on the reliable time sequence and the reliable state sequence. According to the correction time axis, all monitoring data involved in the correction are mapped to the same correction position. Based on the sample weight sequence, sample segments with inconsistent boundary order or excessive adjacent boundary spacing are removed, while continuous boundary sample segments are retained to generate a state-matched fusion preprocessing result.

[0009] Secondly, the present invention provides a wind power monitoring data fusion preprocessing system based on timestamp and operating status matching, including: a boundary sequence generation module, used to arrange the pitch angle monitoring data, speed monitoring data, nacelle orientation monitoring data and active power monitoring data of the same wind turbine in the same monitoring batch according to the original timestamp, filter out the direction reversal position and incremental jump position that appear between adjacent monitoring points, and generate the on-chain state transition boundary sequence in chronological order; The channel correction module is used to extract the response positions located in the intervals before and after each boundary from the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data and pitch monitoring data of the same wind turbine according to the state transition boundary sequence on the chain, generate a channel offset sequence according to the difference between the response positions and the corresponding boundary positions, and perform translation correction on the original timestamps of each monitoring data involved in the correction according to the channel offset sequence to generate a corrected time axis. The weight sequence generation module is used to generate a reliable time sequence based on the deviation of adjacent correction time intervals in the correction time axis, generate a reliable state sequence based on the change in the order of the boundary transitions of the on-chain state transition boundary sequence before and after correction and the offset of the interval between adjacent boundaries, and generate a sample weight sequence based on the reliable time sequence and the reliable state sequence. The fusion preprocessing module is used to map each monitoring data participating in the correction to the same correction position according to the correction time axis, remove sample segments with inconsistent boundary order or excessive adjacent boundary spacing according to the sample weight sequence, and retain continuous boundary sample segments to generate a state-matched fusion preprocessing result.

[0010] Thirdly, the present invention provides a computer device, comprising: One or more processors; The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the wind power monitoring data fusion preprocessing method based on timestamp and operating status matching as described above.

[0011] Fourthly, the present invention provides a computer-readable medium for storing software, the software including instructions executable by one or more computers, the instructions causing the one or more computers to perform operations, the operations including the flow of the wind power monitoring data fusion preprocessing method based on timestamp and operating status matching as described above.

[0012] The beneficial effects of this invention are as follows: This invention determines the switching position of wind turbine operating state by extracting state transition boundaries from pitch angle, rotational speed, nacelle azimuth, and active power; then, it corrects the time position of vibration, bearing temperature, yaw, and pitch monitoring data based on the state transition boundary sequence, reducing the impact of multi-channel sampling delay differences on the fusion results; further, it generates a sample weight sequence by combining reliable time and state sequences, removes mismatched sample segments and retains continuous sample segments, improving the boundary consistency and continuity of fused samples from the source, and ultimately improving the accuracy, stability, and reliability of wind power monitoring data fusion preprocessing and subsequent analysis. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1This is a flowchart illustrating the wind power monitoring data fusion and preprocessing method based on timestamp and operating status matching as shown in this invention. Figure 2 This is a schematic diagram of the modules of the wind power monitoring data fusion and preprocessing system based on timestamp and operating status matching as shown in this invention. Detailed Implementation

[0014] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0015] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0017] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates a wind power monitoring data fusion and preprocessing method based on timestamp and operational status matching, which specifically includes the following steps: S1. Arrange the pitch angle monitoring data, speed monitoring data, nacelle azimuth monitoring data, and active power monitoring data of the same wind turbine unit and the same monitoring batch in sequence according to the original timestamps. Filter out the corresponding direction reversal positions and incremental jump positions between adjacent monitoring points, and generate a chain state transition boundary sequence in chronological order. Note that the following should be noted in this step: In a preferred embodiment, this embodiment applies to a multi-source monitoring data fusion preprocessing scenario of the same wind turbine during a continuous operation. The same monitoring batch can correspond to continuous data collection during a start-up and shutdown process, a yaw correction process, or a pitch adjustment process. For example, in a wind turbine yaw correction scenario, the data collection period is from 10:15:00 on January 27, 2025 to 10:18:00 on January 27, 2025. The sampling interval for pitch angle monitoring data is 200 ms, the sampling interval for speed monitoring data is 200 ms, the sampling interval for nacelle azimuth monitoring data is 200 ms, the sampling interval for active power monitoring data is 200 ms, the sampling interval for vibration monitoring data is 100 ms, the sampling interval for bearing temperature monitoring data is 1 s, the sampling interval for yaw monitoring data is 100 ms, and the sampling interval for pitch monitoring data is 100 ms.

[0018] It should be noted that the pitch angle monitoring data is collected by the angle feedback sensor of the blade pitch drive unit, and the collected content is the blade pitch angle value at each sampling time; the speed monitoring data is collected by the main shaft encoder or the unit speed measurement unit, and the collected content is the main shaft speed or generator-side converted speed value at each sampling time; the nacelle azimuth monitoring data is collected by the yaw angle sensor, and the collected content is the nacelle angle value relative to the reference azimuth at each sampling time; the active power monitoring data is collected by the unit electrical measurement unit, and the collected content is the unit active power output value at each sampling time. The reason for selecting the above four types of monitoring data is that the pitch angle monitoring data, speed monitoring data, nacelle azimuth monitoring data, and active power monitoring data correspond to the four operating state dimensions of wind turbine aerodynamic regulation, mechanical rotation, yaw attitude, and electrical power output, respectively. When the wind turbine starts or stops, yaws are corrected, or pitch is adjusted, at least two of the above four types of monitoring data will show a reversal of direction or a sudden jump in increment at adjacent time positions, so they can be used as the data source for boundary position extraction.

[0019] S1.1 Arrange the pitch angle monitoring data, speed monitoring data, nacelle azimuth monitoring data and active power monitoring data of the same wind turbine unit and the same monitoring batch in sequence according to the original timestamp, and arrange them side by side according to the same time position to form a status monitoring sequence.

[0020] Specifically, the original timestamps and their corresponding monitoring values ​​are first extracted from the same monitoring batch of pitch angle monitoring data, speed monitoring data, nacelle azimuth monitoring data, and active power monitoring data of the same wind turbine unit, and then sorted in ascending order of the original timestamps. After sorting, the earliest original timestamp among the four types of monitoring data is used as the starting time position, and the latest original timestamp among the four types of monitoring data is used as the ending time position. Then, a unified time position sequence is constructed according to the minimum sampling interval common to the four types of monitoring data. Subsequently, the monitoring values ​​corresponding to the unified time position sequence in each type of monitoring data are arranged one by one in chronological order at the same time position to form a status monitoring sequence composed of time position, pitch angle monitoring value, speed monitoring value, nacelle azimuth monitoring value, and active power monitoring value.

[0021] In this embodiment, the parallel arrangement refers to placing the pitch angle monitoring value, speed monitoring value, nacelle azimuth monitoring value, and active power monitoring value side by side in the same sequence unit at the same time position, so that any time position corresponds to one and only one set of pitch angle monitoring value, speed monitoring value, nacelle azimuth monitoring value, and active power monitoring value; when there is no original sampled value for a certain type of monitoring data at that time position, the smaller time interval between the most recent sampled value before that time position and the most recent sampled value after that time position is selected as the corresponding monitoring value; when the time intervals on both sides are the same, the earlier sampled value is selected as the corresponding monitoring value.

[0022] For example, a wind turbine underwent yaw correction between 10:15:20 and 10:15:22 on May 27, 2025. The pitch angle monitoring data changed from 2.1° to 2.2° and then to 2.3° at adjacent time points; the speed monitoring data changed from 14.8 r / min to 14.6 r / min and then to 14.2 r / min; the nacelle azimuth monitoring data changed from 183.4° to 184.2° and then to 186.0°; and the active power monitoring data changed from 1840 kW to 1810 kW and then to 1765 kW. By juxtaposing these monitoring values ​​at the same time points, the combined changes in pitch angle, speed, nacelle azimuth, and active power can be observed simultaneously at the same time point.

[0023] S1.2 In the condition monitoring sequence, calculate the pitch angle change, speed change, nacelle orientation change and active power change for each adjacent monitoring point, and compare the changes of the current adjacent monitoring point with the same changes of the previous adjacent monitoring point.

[0024] Specifically, in the status monitoring sequence, any two adjacent time positions are selected chronologically. The monitoring value corresponding to the previous time position is subtracted from the monitoring value corresponding to the later time position to obtain the pitch angle change, speed change, nacelle azimuth change, and active power change of the adjacent monitoring point. Then, the pitch angle change of the current adjacent monitoring point is compared with that of the previous adjacent monitoring point; the speed change of the current adjacent monitoring point is compared with that of the previous adjacent monitoring point; the nacelle azimuth change of the current adjacent monitoring point is compared with that of the previous adjacent monitoring point; and the active power change of the current adjacent monitoring point is compared with that of the previous adjacent monitoring point. The comparison between the two adjacent points includes two types: directional comparison and amplitude comparison. The comparison order is to first compare the directions of the same type of change, and then compare the absolute values ​​of the same type of change. The comparison is only performed when the current adjacent monitoring point and the previous adjacent monitoring point belong to the same parameter category; cross-parameter category comparisons are not performed.

[0025] In a preferred embodiment, the changes in pitch angle, rotational speed, nacelle azimuth, and active power are calculated as follows: in, For the first The change in pitch angle corresponding to each adjacent monitoring point; The first in the condition monitoring sequence The propeller pitch angle monitoring value corresponding to each time position; The first in the condition monitoring sequence The propeller pitch angle monitoring value corresponding to each time position; in, For the first The change in rotational speed corresponding to each adjacent monitoring point; The first in the condition monitoring sequence Rotational speed monitoring values ​​corresponding to each time position; The first in the condition monitoring sequence Rotational speed monitoring values ​​corresponding to each time position; in, For the first The change in cabin orientation corresponding to each adjacent monitoring point; The first in the condition monitoring sequence The cabin orientation monitoring value corresponding to each time location; The first in the condition monitoring sequence The cabin orientation monitoring value corresponding to each time location; in, For the first The change in active power corresponding to each adjacent monitoring point; The first in the condition monitoring sequence Active power monitoring values ​​corresponding to each time location; The first in the condition monitoring sequence The active power monitoring value corresponding to each time location.

[0026] Furthermore, if the change in a parameter corresponding to the current adjacent monitoring point is positive and the change in the same parameter corresponding to the previous adjacent monitoring point is negative, then the parameter changes from decreasing to increasing at the current time position; if the change in a parameter corresponding to the current adjacent monitoring point is negative and the change in the same parameter corresponding to the previous adjacent monitoring point is positive, then the parameter changes from increasing to decreasing at the current time position; if the absolute value of the change in a parameter corresponding to the current adjacent monitoring point is greater than the absolute value of the change in the same parameter corresponding to the previous adjacent monitoring point, then the parameter experiences a sudden jump in amplitude at the current time position.

[0027] For example, if the cabin azimuth monitoring values ​​at three consecutive time points are 183.4∘, 183.1∘, and 184.0∘, then the cabin azimuth change at the previous adjacent monitoring point is -0.3∘, and the cabin azimuth change at the current adjacent monitoring point is +0.9∘. This time point is recorded as a single-parameter reversal position. If the active power monitoring values ​​are 1840 kW, 1825 kW, and 1765 kW, then the active power change at the previous adjacent monitoring point is -15 kW, and the active power change at the current adjacent monitoring point is -60 kW. Since the absolute value of the current change is greater than the absolute value of the previous change, this time point is recorded as a single-parameter jump position.

[0028] S1.3. Based on the comparison results, the time position where the current change of the same parameter is greater than zero and the previous change is less than zero, or the current change is less than zero and the previous change is greater than zero, is recorded as the single parameter reversal position. The time position where at least two parameters appear at the single parameter reversal position at the same time is recorded as the direction reversal position. S1.4 In the comparison results, the time position where the absolute value of the current change of the same parameter is greater than the absolute value of the previous change is recorded as the single parameter jump position, and the time position where at least two parameters appear at the single parameter jump position at the same time is recorded as the incremental jump position. S1.5 Merge and sort the direction reversal positions and incremental jump positions according to their time sequence to generate the state transition boundary sequence on the chain.

[0029] In a preferred embodiment, within the same monitoring batch, the direction reversal positions are 10:15:20.400, 10:15:38.600, and 10:16:02.000, and the incremental jump positions are 10:15:20.600, 10:15:39.000, and 10:16:01.800, respectively. First, the direction reversal positions and incremental jump positions are merged into a single time position set, then sorted from earliest to latest time, resulting in the boundary sequence 10:15:20.400, 10:15:20.600, 10:15:38.600, 10:15:39.000, 10:16:01.800, and 10:16:02.000. If the time difference between two positions is less than or equal to 200... If the value is ms, then the earlier one is retained as the boundary position, and the later one is merged into the boundary position, resulting in the final on-chain state transition boundary sequence of 10:15:20.400, 10:15:38.600 and 10:16:01.800.

[0030] Preferably, this method can prevent the same state transition from being repeatedly recorded as multiple boundary positions in a very short time.

[0031] S2. Based on the state transition boundary sequence on the chain, extract the response positions located within the intervals before and after each boundary from the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data, and pitch monitoring data of the same wind turbine. Generate a channel offset sequence according to the difference between the response position and the corresponding boundary position. Then, perform translation correction on the original timestamps of each monitoring data involved in the correction based on the channel offset sequence to generate a corrected time axis. Note that the following should be noted in this step: S2.1. Based on the state transition boundary sequence on the chain, extract continuous monitoring intervals before and after each boundary position from the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data and pitch monitoring data of the same wind turbine, respectively, to form the boundary response interval.

[0032] In a preferred embodiment, vibration monitoring data is collected by vibration sensors installed in the main bearing housing, gearbox, or nacelle structure; bearing temperature monitoring data is collected by temperature sensors near the main bearing, gearbox bearing, or generator bearing; yaw monitoring data is collected by angular displacement sensors, current sampling units, or yaw motion status monitoring units of the yaw drive mechanism; and pitch monitoring data is collected by angle feedback units, current sampling units, or motion status monitoring units of the pitch drive mechanism.

[0033] Specifically, for each boundary position in the state transition boundary sequence on the chain, a continuous monitoring interval is extracted forward and backward from the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data, and pitch monitoring data, with the time corresponding to the boundary position as the center. The forward continuous monitoring interval and the backward continuous monitoring interval are then merged into the boundary response interval corresponding to the boundary position.

[0034] In this embodiment, extracting continuous monitoring intervals before and after each boundary position specifically means: taking the time corresponding to the boundary position as the center time point, selecting continuous monitoring data of a fixed duration before the center time point, and selecting continuous monitoring data of the same duration after the center time point.

[0035] As an example, for vibration monitoring data, yaw monitoring data, and pitch monitoring data, a continuous monitoring interval of 2 seconds before and 2 seconds after the boundary position can be selected; for bearing temperature monitoring data, a continuous monitoring interval of 10 seconds before and 10 seconds after the boundary position can be selected.

[0036] Preferably, a longer interval is selected for the bearing temperature monitoring data because the change in bearing temperature has a more significant lag compared to the change in mechanical action.

[0037] S2.2 Within the boundary response interval, calculate the change of the previous adjacent monitoring point, the change of the current adjacent monitoring point, and the change of the next adjacent monitoring point for the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data, and pitch monitoring data, respectively. The time position where the absolute value of the change of the current adjacent monitoring point is greater than the absolute value of the change of the previous adjacent monitoring point and the absolute value of the change of the next adjacent monitoring point is recorded as the candidate response position.

[0038] Specifically, for any intermediate time position within a certain boundary response interval, three adjacent differences are constructed based on the monitoring values ​​corresponding to the previous time position, the current time position, and the next time position. These three adjacent differences correspond to the changes in the previous adjacent monitoring point, the current adjacent monitoring point, and the next adjacent monitoring point, respectively. Then, the absolute value of the change in the current adjacent monitoring point is compared with the absolute values ​​of the changes in the previous and next adjacent monitoring points. When the absolute value of the change in the current adjacent monitoring point is greater than both the absolute values ​​of the changes in the previous and next adjacent monitoring points, this time position is recorded as a candidate response position. The candidate response position corresponds to the time position where the local change is most significant within the boundary response interval.

[0039] In a preferred embodiment, the changes in the previous adjacent monitoring point, the current adjacent monitoring point, and the next adjacent monitoring point within the boundary response interval for any monitoring channel can be calculated as follows: in, This represents the change at the previous adjacent monitoring point; This represents the change in the current adjacent monitoring points; This represents the change at the next adjacent monitoring point; For the boundary response interval, the first Monitoring values ​​corresponding to each time location; For the boundary response interval, the first Monitoring values ​​corresponding to each time location; For the boundary response interval, the first Monitoring values ​​corresponding to each time location; For the boundary response interval, the first Monitoring values ​​corresponding to each time location; These can represent vibration monitoring values, bearing temperature monitoring values, yaw monitoring values, or pitch monitoring values, respectively; the criteria for determining the candidate response location are: and For example, within the boundary response interval of a certain yaw monitoring data, the yaw monitoring values ​​at four consecutive time positions are 0.12∘, 0.18∘, 0.44∘, and 0.51∘, respectively. Then the change of the previous adjacent monitoring point is 0.06∘, the change of the current adjacent monitoring point is 0.26∘, and the change of the next adjacent monitoring point is 0.07∘. Therefore, the current time position is recorded as the candidate response position.

[0040] S2.3. Among the candidate response locations, select the candidate response location that is located after the corresponding boundary location and has the smallest absolute time difference with the corresponding boundary location, and use it as the response location of the monitoring data corresponding to the boundary location; when there is no candidate response location after the corresponding boundary location, select the candidate response location that is located before the corresponding boundary location and has the smallest absolute time difference with the corresponding boundary location, and use it as the response location. S2.4 Calculate the time difference between the response position and the corresponding boundary position, and form a channel offset sequence in the order of vibration monitoring data, bearing temperature monitoring data, yaw monitoring data and pitch monitoring data.

[0041] Specifically, for each monitoring channel corresponding to each boundary position, the time corresponding to the response position is subtracted from the time corresponding to the boundary position to obtain the time difference value of the monitoring channel corresponding to the boundary position. If the time difference value is positive, it indicates that the response position of the monitoring channel is after the boundary position; if the time difference value is negative, it indicates that the response position of the monitoring channel is before the boundary position. The time difference values ​​obtained from each monitoring channel are arranged in the order of vibration monitoring data, bearing temperature monitoring data, yaw monitoring data, and pitch monitoring data to form the channel offset sequence corresponding to the boundary position.

[0042] In a preferred embodiment, the time difference is calculated as follows: in, For a certain monitoring channel at the 1st Time difference values ​​at each boundary position; For monitoring channels in the first The response time corresponding to each boundary position; For the first The boundary position time corresponding to each boundary position.

[0043] As an example, if the time at a certain boundary position is 10:15:20.400, and the response time of the vibration monitoring data is 10:15:20.520, then the time difference between the vibration monitoring data and the response time is 120 ms; if the response time of the bearing temperature monitoring data is 10:15:21.400, then the time difference between the bearing temperature monitoring data and the response time is 1.0 s; if the response time of the yaw monitoring data is 10:15:20.480, then the time difference between the yaw monitoring data and the response time is 80 ms; if the response time of the pitch monitoring data is 10:15:20.600, then the time difference between the pitch monitoring data and the response time is 200 ms; therefore, the channel offset sequence corresponding to this boundary position can be recorded as a sequence containing 120 ms, 1.0 s, 80 ms, and 200 ms in sequence.

[0044] S2.5. Based on the channel offset sequence, the original timestamps of each monitoring data involved in the correction are shifted and corrected, and the corrected time positions are arranged in chronological order to generate a correction time axis.

[0045] In a preferred embodiment, the concentrated distribution position of the time difference corresponding to each boundary position of the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data, and pitch monitoring data is statistically analyzed, and the concentrated distribution position is used as the translation correction amount of the monitoring channel; preferably, the median value of the time difference of the same monitoring channel at all boundary positions can be taken as the translation correction amount; then, the translation correction amount is uniformly subtracted from all the original timestamps of the monitoring channel to obtain the corrected time position of the monitoring channel; then the corrected time positions of each monitoring channel are uniformly arranged in chronological order to form a correction time axis.

[0046] It should be noted that using the intermediate value as the translation correction amount can reduce the impact of abnormal time differences at individual boundary positions on the overall correction result.

[0047] During the same operation of a wind turbine, although multi-channel monitoring data may point to the same state change event, the change locations in each channel do not coincide due to sampling delay, actuator response delay, and channel processing delay. This leads to time misalignment of the same event in different monitoring channels during subsequent fusion. By performing the aforementioned boundary response interval interception, response location screening, time difference calculation, and original timestamp shift correction processes, the multi-channel monitoring data can achieve higher consistency near the boundary positions, thereby reducing time mismatch, lowering the probability of mis-splicing, and improving the accuracy of subsequent sample segment retention.

[0048] S3. Generate a reliable time sequence based on the deviation of adjacent correction time intervals in the correction time axis; generate a reliable state sequence based on the changes in the order of boundary states before and after correction and the offset of adjacent boundary intervals in the on-chain state transition boundary sequence; and generate a sample weight sequence based on the reliable time sequence and the reliable state sequence. Note that the following should be noted in this step: S3.1 In the correction time axis, calculate the correction time interval between every two adjacent correction time positions in turn, and take the difference between the current correction time interval and the previous correction time interval as the time interval deviation value.

[0049] Specifically, according to the chronological order of time in the correction timeline, any two adjacent correction time positions are selected, and the previous correction time position is subtracted from the subsequent correction time position to obtain the corresponding correction time interval. The current correction time interval is then compared with the previous correction time interval, and the difference between the two is used as the time interval deviation value. If the adjacent correction time positions in the correction timeline are relatively evenly distributed, the time interval deviation value is small. If there is local compression or local stretching in the correction timeline, the time interval deviation value increases.

[0050] In a preferred embodiment, the correction time interval and the time interval deviation value can be calculated as follows: in, For the first One correction time interval; To correct the first in the timeline One correction time position; To correct the first in the timeline One correction time position; For the first Deviation value for each time interval; For the first One correction time interval.

[0051] S3.2. Record the time position where the time interval deviation value is less than the larger of the two adjacent correction time intervals as the time stable position, and record the time position where the time interval deviation value is greater than or equal to the larger of the two adjacent correction time intervals as the time abnormal position, and generate a reliable time sequence according to the time sequence. S3.3. Compare the boundary positions in the state transition boundary sequence on the chain with the corresponding corrected boundary positions in sequence. Record the time positions where the order of the boundaries has not changed and the change in the interval between adjacent boundaries is less than the interval between adjacent boundaries before correction as stable state positions. Record the other time positions as abnormal state positions and generate a reliable state sequence according to the time sequence. S3.4. Compare the time-trusted sequence and the state-trusted sequence at the same time position. Record the time position where the time-stable position and the state-stable position coincide as the high-weight position, the time position where the time-abnormal position and the state-abnormal position coincide as the low-weight position, and the remaining time positions as the middle-weight positions. S3.5 Arrange the high-weight positions, middle-weight positions, and low-weight positions in chronological order to generate a sample weight sequence.

[0052] For example, if during a startup process, the judgment results corresponding to ten consecutive time positions on the correction timeline are, in order, high weight position, high weight position, middle weight position, high weight position, low weight position, low weight position, middle weight position, high weight position, high weight position, and high weight position, then these are arranged in chronological order to form a sample weight sequence. This sample weight sequence can be directly used as the basis for checking the subsequent correction sample sequence: consecutive high weight positions correspond to the stable rising phase during the startup process, low weight positions correspond to positions where sampling is misaligned or the state order is disrupted, and middle weight positions correspond to positions where there is local time jitter but the state order is not significantly disrupted.

[0053] S4. Map all monitoring data participating in the correction to the same correction position according to the correction time axis. Based on the sample weight sequence, remove sample segments with inconsistent boundary order or excessive adjacent boundary spacing, and retain continuous boundary sample segments to generate a state-matched fusion preprocessing result. Note that the following should be noted in this step: S4.1. According to the correction time axis, align and arrange the monitoring data involved in the correction at the same correction time position to form a correction sample sequence.

[0054] Specifically, each correction time position in the correction time axis is used as a unified reference time position. At each correction time position, the correction monitoring value corresponding to that correction time position is extracted from the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data, and pitch monitoring data, and arranged in a fixed order under the same correction time position to form a corresponding correction sample unit. All correction sample units are arranged in chronological order to form a correction sample sequence. When there is no completely identical correction time position in a certain monitoring channel for a certain correction time position, the correction monitoring value with the smallest time difference from that correction time position is selected as the corresponding monitoring value for that correction time position. If the time difference between two consecutive correction monitoring values ​​and the correction time position is the same, the earlier one is selected.

[0055] S4.2 In the calibration sample sequence, the continuous time positions are checked sequentially according to the sample weight sequence. The continuous sample intervals that contain low weight positions, or whose corresponding boundary order has changed, or whose adjacent boundary interval offset is greater than or equal to the adjacent boundary interval before calibration, and which have no missing calibration positions between two adjacent time positions are recorded as boundary mismatch sample segments. S4.3 In the calibration sample sequence, the continuous time positions are checked sequentially according to the sample weight sequence. The continuous sample interval that does not contain low weight positions, whose corresponding boundary order has not changed, whose adjacent boundary interval offset is less than the adjacent boundary interval before calibration, and whose two adjacent time positions have no missing calibration positions is recorded as the boundary continuous sample segment. S4.4 Remove the boundary mismatched sample segments from the corrected sample sequence and retain the boundary continuous sample segments; S4.5. The retained boundary continuous sample segments are sequentially spliced ​​according to the correction time position to generate the state-matched fusion preprocessing result.

[0056] In this embodiment, sequential splicing means preserving the original time order within each boundary continuous sample segment and connecting them only according to the starting time position of each boundary continuous sample segment without changing the sample arrangement within the boundary continuous sample segment. The resulting state matching fusion preprocessing result not only has a consistent multi-channel data correspondence at the same correction time position, but also maintains a consistent boundary evolution order between adjacent sample positions.

[0057] As an example, if a wind turbine begins yaw correction around 10:15:20 and pitch angle increase occurs around 10:16:02, step S1 extracts the chain state transition boundary sequence from the pitch angle monitoring data, speed monitoring data, nacelle azimuth monitoring data, and active power monitoring data, with boundary positions of 10:15:20.400, 10:15:38.600, and 10:16:01.800, respectively. Combined with step S2, the corresponding response positions are extracted from the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data, and pitch monitoring data, respectively, and the translation correction amounts for each monitoring channel are obtained as 120 ms, 1.0 s, 80 ms, and 200 ms, respectively. According to step S3, three low-weight positions, two intermediate-weight positions, and nine high-weight positions are determined on the corrected time axis. The low-weight positions are mainly concentrated around 10:15:38.400 to 10:15:38.800, corresponding to the short-term jitter interval of yaw drive. In step S4, this short-term jitter interval is recorded as a boundary mismatch sample segment in the corrected sample sequence and removed. Two boundary continuous sample segments, 10:15:20.400 to 10:15:38.200 and 10:15:39.000 to 10:16:02.200, are retained and spliced ​​in chronological order to form the final fusion preprocessing result.

[0058] It should be noted that even after time-shift correction, multi-source monitoring data may still have issues such as disordered boundary order, distorted boundary intervals, or broken continuous samples in local intervals. If all samples are retained directly, subsequent analysis will result in data segments that are inconsistent with the actual state evolution being mixed in. By constructing corrected sample sequences, identifying boundary mismatched sample segments, retaining continuous boundary sample segments, and sequentially splicing them, a fusion preprocessing result with consistent state matching, continuous time location, and stable boundary order can be obtained. This improves the accuracy of subsequent operational state identification, reduces interference from abnormal samples, and enhances the consistency of multi-source data.

[0059] In the application of the above embodiments, refer to Figure 2 Other aspects disclosed in the embodiments of the present invention also propose a wind power monitoring data fusion preprocessing system based on timestamp and operating status matching, including: The boundary sequence generation module is used to arrange the pitch angle monitoring data, speed monitoring data, nacelle orientation monitoring data and active power monitoring data of the same wind turbine in the same monitoring batch according to the original timestamp, filter out the corresponding direction reversal positions and incremental jump positions between adjacent monitoring points, and generate the on-chain state transition boundary sequence in chronological order. The channel correction module is used to extract the response positions located in the intervals before and after each boundary from the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data and pitch monitoring data of the same wind turbine based on the state transition boundary sequence on the chain. It generates a channel offset sequence according to the difference between the response position and the corresponding boundary position, and performs translation correction on the original timestamps of each monitoring data involved in the correction based on the channel offset sequence to generate a corrected time axis. The weight sequence generation module is used to generate a reliable time sequence based on the deviation of adjacent correction time intervals in the correction time axis, generate a reliable state sequence based on the change in the order of the state transition boundary sequences before and after correction and the offset of the adjacent boundary intervals, and generate a sample weight sequence based on the reliable time sequence and the reliable state sequence. The fusion preprocessing module is used to map the monitoring data involved in the correction to the same correction position according to the correction time axis, remove sample segments with inconsistent boundary order or excessive adjacent boundary intervals based on the sample weight sequence, and retain continuous boundary sample segments to generate a state-matched fusion preprocessing result.

[0060] Other aspects disclosed in the embodiments of the present invention also provide a computer device including one or more processors and a memory.

[0061] The memory is used to store operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the wind power monitoring data fusion preprocessing method based on timestamp and operating status matching in the foregoing embodiments, especially... Figure 1 The flowchart of the method is shown.

[0062] Other aspects disclosed in the embodiments of the present invention also propose a computer-readable medium for storing software including instructions executable by one or more computers, which, upon execution, cause the one or more computers to perform operations including the flow of the wind power monitoring data fusion preprocessing method based on timestamp and operating status matching of the foregoing embodiments, particularly... Figure 1 The flowchart of the method is shown.

[0063] It should be recognized that embodiments of the present invention may be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium.

[0064] The method can be implemented using standard programming techniques, including a non-transitory computer-readable storage medium configured with a computer program in the computer program, wherein the storage medium is configured such that the computer operates in a specific and predefined manner.

[0065] Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system; however, if required, the program can be implemented in assembly or machine language.

[0066] In any case, the language can be either compiled or interpreted.

[0067] Furthermore, for this purpose, the program can run on programmed application-specific integrated circuits.

[0068] The processes described herein (or variations and / or combinations thereof) can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program includes a plurality of instructions executable by one or more processors.

[0069] Furthermore, the method can be implemented in any suitable computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices.

[0070] Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether portable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein.

[0071] Furthermore, machine-readable code, or parts thereof, can be transmitted via wired or wireless networks.

[0072] When such media includes instructions or programs that combine with a microprocessor or other data processor to implement the steps described above, the invention described herein includes these and other different types of non-transitory computer-readable storage media.

[0073] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A wind power monitoring data fusion and preprocessing method based on timestamp and operational status matching, characterized in that, include: The pitch angle monitoring data, speed monitoring data, nacelle orientation monitoring data and active power monitoring data of the same wind turbine unit and the same monitoring batch are arranged in order according to the original timestamp. The corresponding direction reversal positions and incremental jump positions between adjacent monitoring points are screened out, and the on-chain state transition boundary sequence is generated in chronological order. Based on the chain state transition boundary sequence, the response positions located in the intervals before and after each boundary are extracted from the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data and pitch monitoring data of the same wind turbine. A channel offset sequence is generated according to the difference between the response positions and the corresponding boundary positions. The original timestamps of each monitoring data involved in the correction are shifted and corrected according to the channel offset sequence to generate a corrected time axis. A reliable time sequence is generated based on the deviation of adjacent correction time intervals in the correction time axis. A reliable state sequence is generated based on the change in the order of the boundary transitions of the on-chain state transition boundary sequence before and after correction and the offset of the interval between adjacent boundaries. A sample weight sequence is generated based on the reliable time sequence and the reliable state sequence. According to the correction time axis, all monitoring data involved in the correction are mapped to the same correction position. Based on the sample weight sequence, sample segments with inconsistent boundary order or excessive adjacent boundary spacing are removed, while continuous boundary sample segments are retained to generate a state-matched fusion preprocessing result.

2. The wind power monitoring data fusion and preprocessing method based on timestamp and operating status matching according to claim 1, characterized in that, The generation of the on-chain state transition boundary sequence includes: The pitch angle monitoring data, speed monitoring data, nacelle azimuth monitoring data and active power monitoring data of the same wind turbine unit and the same monitoring batch are arranged in order according to the original timestamps, and then arranged side by side according to the same time position to form a status monitoring sequence. In the state monitoring sequence, for each adjacent monitoring point, the pitch angle change, speed change, nacelle orientation change and active power change are calculated, and the changes of the current adjacent monitoring point are compared with the same changes of the previous adjacent monitoring point. Based on the comparison results, the time position where the current change of the same parameter is greater than zero and the previous change is less than zero, or the current change is less than zero and the previous change is greater than zero, is recorded as the single parameter reversal position. The time position where at least two parameters appear at the single parameter reversal position at the same time is recorded as the direction reversal position. In the comparison results, the time position where the absolute value of the current change of the same parameter is greater than the absolute value of the previous change is recorded as the single parameter jump position, and the time position where at least two parameters appear at the single parameter jump position at the same time is recorded as the incremental jump position. The direction reversal positions and the incremental jump positions are merged and sorted according to time sequence to generate the on-chain state transition boundary sequence.

3. The wind power monitoring data fusion and preprocessing method based on timestamp and operating status matching according to claim 2, characterized in that, The extraction of the response location includes: Based on the chain state transition boundary sequence, continuous monitoring intervals before and after each boundary position are extracted from the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data and pitch monitoring data of the same wind turbine, respectively, to form boundary response intervals. Within the boundary response interval, the change in the previous adjacent monitoring point, the change in the current adjacent monitoring point, and the change in the next adjacent monitoring point are calculated for the vibration monitoring data, the bearing temperature monitoring data, the yaw monitoring data, and the pitch monitoring data, respectively. The time position where the absolute value of the change in the current adjacent monitoring point is simultaneously greater than the absolute value of the change in the previous adjacent monitoring point and the absolute value of the change in the next adjacent monitoring point is recorded as the candidate response position. Among the candidate response locations, the candidate response location that is located after the corresponding boundary location and has the smallest absolute time difference with the corresponding boundary location is selected as the response location of the monitoring data corresponding to the boundary location; when there is no candidate response location after the corresponding boundary location, the candidate response location that is located before the corresponding boundary location and has the smallest absolute time difference with the corresponding boundary location is selected as the response location.

4. The wind power monitoring data fusion and preprocessing method based on timestamp and operating status matching according to claim 3, characterized in that, The generation of the corrected timeline includes: Calculate the time difference between the response position and the corresponding boundary position, and form a channel offset sequence according to the order of the vibration monitoring data, the bearing temperature monitoring data, the yaw monitoring data, and the pitch monitoring data; Based on the channel offset sequence, the original timestamps of each monitoring data involved in the correction are shifted and corrected, and the corrected time positions are arranged in chronological order to generate the corrected time axis.

5. The wind power monitoring data fusion and preprocessing method based on timestamp and operating status matching according to claim 1, characterized in that, The generation of the time-credible sequence and the state-credible sequence includes: In the correction time axis, the correction time interval between every two adjacent correction time positions is calculated sequentially, and the difference between the current correction time interval and the previous correction time interval is used as the time interval deviation value. The time position where the time interval deviation is less than the larger of the two adjacent correction time intervals is recorded as the time stable position, and the time position where the time interval deviation is greater than or equal to the larger of the two adjacent correction time intervals is recorded as the time abnormal position, and a reliable time sequence is generated according to the time sequence. The boundary positions in the state transition boundary sequence on the chain are compared sequentially with the corresponding corrected boundary positions. The time positions where the order of the boundaries has not changed and the change in the interval between adjacent boundaries is less than the interval between adjacent boundaries before correction are recorded as stable state positions, and the remaining time positions are recorded as abnormal state positions. A reliable state sequence is generated according to the time sequence.

6. The wind power monitoring data fusion and preprocessing method based on timestamp and operating status matching according to claim 5, characterized in that, The generation of the sample weight sequence includes: The time-credible sequence and the state-credible sequence are compared at the same time position. The time position where the time-stable position and the state-stable position coincide is recorded as the high-weight position, the time position where the time-abnormal position and the state-abnormal position coincide is recorded as the low-weight position, and the remaining time positions are recorded as the middle-weight positions. The high-weight positions, the middle-weight positions, and the low-weight positions are arranged in chronological order to generate the sample weight sequence.

7. The wind power monitoring data fusion and preprocessing method based on timestamp and operating status matching according to claim 1, characterized in that, The generation of the fusion preprocessing result includes: According to the correction timeline, the monitoring data involved in the correction are aligned and arranged at the same correction time position to form a correction sample sequence; In the corrected sample sequence, the continuous time positions are checked sequentially according to the sample weight sequence. The continuous sample intervals that contain low weight positions, or whose corresponding boundary order has changed, or whose adjacent boundary interval offset is greater than or equal to the adjacent boundary interval before correction, and which have no missing correction positions between two adjacent time positions are recorded as boundary mismatch sample segments. In the corrected sample sequence, the continuous time positions are checked sequentially according to the sample weight sequence. The continuous sample interval that does not contain low weight positions, whose corresponding boundary order has not changed, whose adjacent boundary interval offset is less than the adjacent boundary interval before correction, and whose two adjacent time positions have no missing correction positions is recorded as the boundary continuous sample segment. Remove the boundary mismatched sample segments from the corrected sample sequence, and retain the boundary continuous sample segments; The retained boundary continuous sample segments are sequentially spliced ​​according to the correction time position to generate a state-matched fusion preprocessing result.

8. A wind power monitoring data fusion preprocessing system based on timestamp and operating status matching, based on the wind power monitoring data fusion preprocessing method based on timestamp and operating status matching as described in any one of claims 1 to 7, characterized in that, include: The boundary sequence generation module is used to arrange the pitch angle monitoring data, speed monitoring data, nacelle orientation monitoring data and active power monitoring data of the same wind turbine in the same monitoring batch according to the original timestamp, filter out the corresponding direction reversal positions and incremental jump positions between adjacent monitoring points, and generate the on-chain state transition boundary sequence in chronological order. The channel correction module is used to extract the response positions located in the intervals before and after each boundary from the vibration monitoring data, bearing temperature monitoring data, yaw monitoring data and pitch monitoring data of the same wind turbine according to the state transition boundary sequence on the chain, generate a channel offset sequence according to the difference between the response positions and the corresponding boundary positions, and perform translation correction on the original timestamps of each monitoring data involved in the correction according to the channel offset sequence to generate a corrected time axis. The weight sequence generation module is used to generate a reliable time sequence based on the deviation of adjacent correction time intervals in the correction time axis, generate a reliable state sequence based on the change in the order of the boundary transitions of the on-chain state transition boundary sequence before and after correction and the offset of the interval between adjacent boundaries, and generate a sample weight sequence based on the reliable time sequence and the reliable state sequence. The fusion preprocessing module is used to map each monitoring data participating in the correction to the same correction position according to the correction time axis, remove sample segments with inconsistent boundary order or excessive adjacent boundary spacing according to the sample weight sequence, and retain continuous boundary sample segments to generate a state-matched fusion preprocessing result.

9. A computer device, characterized in that, include: One or more processors; The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the wind power monitoring data fusion preprocessing method based on timestamp and operating status matching as described in any one of claims 1 to 7.

10. A computer-readable medium for storing software, characterized in that: The software includes instructions executable by one or more computers, which cause the one or more computers to perform operations, including the flow of the wind power monitoring data fusion preprocessing method based on timestamp and operating status matching as described in any one of claims 1 to 7.

Images