Method and device for cleaning fan operation data, electronic equipment and storage medium
Through iterative operations and data density analysis, the operating data of the blower is cleaned, which solves the problem of poor data quality in existing technologies and achieves efficient data cleaning and model training.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ANTELOPE IND INTERNET CO LTD
- Filing Date
- 2026-02-05
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies cannot effectively clean wind turbine operating data, resulting in poor data quality and affecting subsequent model training and tasks such as anomaly warning or fault diagnosis.
Through iterative operations, based on data density and preset attenuation coefficient sorting, wind turbine operation data with lower density rankings are gradually removed. Data density is calculated using wind speed and power grids and rectangular regions. Combined with iteration termination conditions and data distance judgment, effective cleaning of wind turbine operation data is achieved.
It effectively removed poor-quality and stacked wind turbine operation data, improved data sparsity and quality, and ensured the accuracy of model training and the reliability of anomaly warning.
Smart Images
Figure CN121658796B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, electronic device, and storage medium for cleaning wind turbine operating data. Background Technology
[0002] Supervisory Control and Data Acquisition (SCADA) is the main system currently used in wind turbine control. The wind turbine operation data collected by SCADA is used for model training and participates in the early warning of anomalies and operation and maintenance decisions of onshore and offshore wind turbine equipment.
[0003] Most wind turbines operate in harsh outdoor environments such as low temperature, high humidity, and strong wind for extended periods. SCADA data is often affected by various factors such as component failure, sensor error, transmission anomalies, and human-induced power throttling, resulting in poor data quality. This makes it difficult to depict the normal operating state of the wind turbine and cannot be directly used to build a Normal Behavior Model (NBM) to complete anomaly warning or fault diagnosis tasks.
[0004] Therefore, it is necessary to clean the wind turbine operation data and remove the wind turbine operation data with poor data quality. However, at present, there is a situation where there is a pile of wind turbine operation data with poor data quality. Existing cleaning methods cannot effectively clean it, which seriously affects the training of the model in the later stage and affects the wind turbine abnormality warning or fault diagnosis tasks. Summary of the Invention
[0005] This invention provides a method, apparatus, electronic device, and storage medium for cleaning wind turbine operation data, which addresses the shortcomings of existing cleaning methods that cannot effectively clean stacked wind turbine operation data with poor data quality. This invention achieves effective cleaning of wind turbine operation data with poor data quality.
[0006] This invention provides a method for cleaning wind turbine operating data, comprising the following steps.
[0007] Receives operating data from multiple wind turbines;
[0008] Repeat the following iterative operation until the iteration termination condition is met to determine the target wind turbine operating data among the multiple wind turbine operating data:
[0009] Determine the data density corresponding to each wind turbine operating data point, whereby the data density describes the density of the data points determined based on the wind turbine operating data within the wind speed-power grid.
[0010] Based on the order of all data density from high to low, the wind turbine operating data prior to this iteration is sorted to obtain the sorting result;
[0011] The top N wind turbine operating data in the sorting results will be used as the wind turbine operating data for the next iteration.
[0012] Wherein, N is determined based on the amount of wind turbine operating data mentioned before this iteration and the preset attenuation coefficient.
[0013] According to a method for cleaning fan operating data provided by the present invention, each fan operating data includes a wind speed value and a power value;
[0014] Determining the data density corresponding to each wind turbine's operating data includes:
[0015] Based on the wind speed value and the power value, the target grid corresponding to each data point determined by the wind turbine operation data is determined in the wind speed and power grid, which is constructed based on a preset wind speed step size and a preset power step size;
[0016] The number of data points located within a preset area of the target grid is taken as the data density of the wind turbine operation data belonging to the target grid.
[0017] According to a method for cleaning wind turbine operation data provided by the present invention, the preset area is a rectangular area consisting of the target grid, a first grid adjacent to the target grid on the left and right, and a second grid adjacent to the target grid on the top and bottom.
[0018] According to a method for cleaning wind turbine operating data provided by the present invention, the iteration termination condition includes:
[0019] Maximum number of iterations; or
[0020] The data distance between the operating data of the first fan and the operating data of the second fan is less than the corresponding preset distance;
[0021] Wherein, the first wind turbine operating data is the wind turbine operating data that will be retained in the next iteration operation, and the second wind turbine operating data is the wind turbine operating data that will be removed in the next iteration operation.
[0022] According to a method for cleaning wind turbine operating data provided by the present invention, the data distance includes the average difference in wind speed and the average difference in power.
[0023] The data distance between the first wind turbine operating data and the second wind turbine operating data was calculated using the following method:
[0024] Based on the first wind turbine operating data, the average power corresponding to each wind speed range and the average wind speed corresponding to each power range are determined.
[0025] Based on the average power corresponding to each wind speed range and the average wind speed corresponding to each power range, the wind speed difference and power difference corresponding to each second wind turbine operating data are determined.
[0026] The average wind speed difference is determined based on the wind speed difference corresponding to the operating data of each of the second wind turbines, and the average power difference is determined based on the power difference corresponding to the operating data of each of the second wind turbines.
[0027] According to a method for cleaning wind turbine operating data provided by the present invention, the method further includes:
[0028] Based on the power value in the operating data of each target wind turbine, the operating data of the target wind turbines in the multiple wind turbine operating data are grouped to obtain multiple power groups, which include multiple first power groups with the largest power and multiple second power groups with the smallest power.
[0029] A first reference wind speed value is determined based on multiple first power groups, and a second reference wind speed value is determined based on multiple second power groups;
[0030] The wind turbine operating data that have wind speed values greater than the first reference wind speed value and less than the cut-off wind speed value, and whose power values are in the first power group and have been removed, are added to the target wind turbine operating data; and / or
[0031] The wind turbine operating data that have wind speed values less than the second reference wind speed value but greater than zero, power values in the second power group, and have been removed, are added to the target wind turbine operating data.
[0032] According to a method for cleaning wind turbine operating data provided by the present invention, the first reference wind speed value is the median wind speed when the target wind turbine operating data in the first power group is sorted according to the wind speed value; the second reference wind speed value is the median wind speed when the target wind turbine operating data in the second power group is sorted according to the wind speed value.
[0033] The present invention also provides a cleaning device for fan operation data, comprising the following modules:
[0034] The receiving module is used to receive operating data from multiple wind turbines;
[0035] The processing module is used to repeatedly perform the following iterative operation until the iteration termination condition is met, in order to determine the target wind turbine operating data among the multiple wind turbine operating data:
[0036] Determine the data density corresponding to each wind turbine operating data point, whereby the data density describes the density of the data points determined based on the wind turbine operating data within the wind speed-power grid.
[0037] Based on the order of all data density from high to low, the wind turbine operating data prior to this iteration is sorted to obtain the sorting result;
[0038] The top N wind turbine operating data in the sorting results will be used as the wind turbine operating data for the next iteration.
[0039] Wherein, N is determined based on the amount of wind turbine operating data mentioned before this iteration and the preset attenuation coefficient.
[0040] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for cleaning fan operating data as described above.
[0041] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for cleaning wind turbine operating data as described above.
[0042] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the method for cleaning wind turbine operating data as described above.
[0043] This invention provides a method, apparatus, electronic device, and storage medium for cleaning wind turbine operation data. By repeatedly performing iterative operations until the iteration termination condition is met, target wind turbine operation data among multiple wind turbine operation data can be identified. During the iterative operation, wind turbine operation data with lower density ranking can be removed. Wind turbine operation data with poor data quality and stacking gradually become sparse and are eventually identified and filtered. In this process, effective cleaning of wind turbine operation data with poor data quality can be achieved. Attached Figure Description
[0044] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0045] Figure 1 It is a schematic diagram of the theoretical wind speed-power curve in the relevant technical solution.
[0046] Figure 2 It is a wind speed-power scatter plot drawn based on SCADA data reported during the actual operation of the wind turbine in the relevant technical solution.
[0047] Figure 3 This is one of the flowcharts illustrating the method for cleaning wind turbine operating data provided by the present invention.
[0048] Figure 4 This is a flowchart illustrating the process of determining the data density corresponding to the operating data of each wind turbine before the current iteration operation, provided by the present invention.
[0049] Figure 5 This is a flowchart illustrating the process of calculating the data distance between the operating data of the first wind turbine and the operating data of the second wind turbine, as provided by the present invention.
[0050] Figure 6 This is the second flowchart illustrating the method for cleaning wind turbine operating data provided by the present invention.
[0051] Figure 7 This is one of the schematic diagrams illustrating the cleaning effect of the fan operation data provided by the present invention.
[0052] Figure 8 This is the second schematic diagram of the wind turbine operation data cleaning effect provided by the present invention.
[0053] Figure 9 This is the third schematic diagram of the wind turbine operation data cleaning effect provided by the present invention.
[0054] Figure 10 This is a schematic diagram of the structure of the fan operation data cleaning device provided by the present invention.
[0055] Figure 11 This is a schematic diagram of the structure of the electronic device provided by the present invention.
[0056] Figure label:
[0057] 1001: Receiver module; 1002: Processing module; 1110: Processor; 1120: Communication interface; 1130: Memory; 1140: Communication bus. Detailed Implementation
[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0059] The following is combined with Figures 1 to 11 The present invention describes a method, apparatus, electronic device, and storage medium for cleaning wind turbine operating data.
[0060] In related technical solutions, the cleaning of wind turbine operating data is mostly based on wind speed-power curves (referred to as power curves). The power curve of a wind turbine depicts the active power value corresponding to different wind speeds, and is further divided into two types: theoretical power curves and actual power curves. For example... Figure 1 As shown, the theoretical wind speed-power curve includes five zones: A, B, C, D, and E. Among them, A is the start-up zone during wind turbine operation, B is the maximum wind energy capture zone during wind turbine operation, C is the constant speed zone during wind turbine operation, D is the constant power zone during wind turbine operation, and E is the shutdown zone during wind turbine operation.
[0061] like Figure 2 As shown, the theoretical power curve ignores the non-uniqueness of wind speed and output power. In fact, according to the principle of wind power generation, the output power is related to variables such as wind speed, rotational speed, pitch angle, blade length, and atmospheric density. Therefore, the output power at the same wind speed is not unique. The actual power curve differs significantly from the theoretical power curve. The wind speed-power scatter plot, based on SCADA data reported during actual wind turbine operation, shows that although most wind turbine operating data are distributed near the power curve (high density points), there are also many wind turbine operating data that deviate from the curve, which are the wind turbine operating data that need to be deleted.
[0062] The present invention provides a method for cleaning wind turbine operation data. The executing entity can be an electronic device with data processing capabilities, such as a server, a local computer, or a cloud computing platform, or it can be a functional module embedded in the wind turbine main control system or the wind farm monitoring system.
[0063] Figure 3 This is one of the flowcharts illustrating the method for cleaning fan operating data provided by the present invention, such as... Figure 3 As shown, the method includes the following:
[0064] Step 301: Receive operating data from multiple wind turbines;
[0065] Step 302: Repeat the following iterative operation until the iteration termination condition is met to determine the target wind turbine operating data among multiple wind turbine operating data; determine the data density corresponding to each wind turbine operating data, which is used to describe the density of data points determined based on the wind turbine operating data in the wind speed power grid; sort the wind turbine operating data before this iteration operation based on the order of all data densities from high to low, and obtain the sorting result; take the top N wind turbine operating data in the sorting result as the wind turbine operating data for the next iteration operation; where N is determined based on the number of wind turbine operating data before this iteration operation and the preset attenuation coefficient.
[0066] In some embodiments, the wind turbine operating data is part or all of the operating parameter data collected and reported in real time by the wind turbine SCADA during operation.
[0067] For example, operating parameter data can be obtained in real time from the wind farm's database, historical data files (such as CSV, TXT, etc.), or via a network interface. Each operating parameter data can logically be regarded as a data point, which may contain one or more dimensions of information such as wind speed, generator power, main shaft speed, and pitch angle, to describe the operating status of the wind turbine at a certain moment.
[0068] In some embodiments, data density is an indicator used to quantify the density of neighboring wind turbine operating data around a given wind turbine operating data point. A neighborhood-based approach can be used, for example, counting how many other wind turbine operating data points are contained within a preset radius centered on a given wind turbine operating data point, and using this number as the data density of that data point; alternatively, the average distance between the wind turbine operating data point and its K nearest neighbors can be calculated, and the reciprocal of this average distance can be used as the data density.
[0069] The value of N is dynamically determined, based on the amount of wind turbine operating data prior to the current iteration and a preset attenuation coefficient. The attenuation coefficient is a preset proportionality factor, typically close to but less than 1, such as 0.99, 0.995, or 0.98. The formula for calculating N is: N = floor(total number of current wind turbine operating data × attenuation coefficient), where floor is a floor function.
[0070] For example, if the current iteration processes 10,000 wind turbine operation data points, and the attenuation coefficient is set to 0.99, then N is 9,900. In this case, the method will retain the top 9,900 wind turbine operation data points by density in the sorting results as the data for the next iteration; while the bottom 100 wind turbine operation data points by density will be considered outliers to be removed in this iteration. This proportional removal method ensures that the cleaning intensity of each iteration is gentle and controllable.
[0071] By repeatedly performing iterative operations until the iteration termination condition is met, the target wind turbine operation data among multiple wind turbine operation data can be determined. During the iterative operation, wind turbine operation data with lower density ranking can be removed. For wind turbine operation data with poor data quality and stacking, the data gradually becomes sparse and is eventually identified and filtered. In this process, effective cleaning of wind turbine operation data with poor data quality can be achieved.
[0072] This invention introduces an iterative filtering method based on attenuation coefficients to remove low-density outliers from raw wind turbine operating data layer by layer in a gentle manner. Compared to one-time threshold filtering methods, this approach has better processing performance for complex conditions such as stacked outliers, is less sensitive to parameter settings, and is more robust. Therefore, it can more reliably and accurately extract high-quality target wind turbine operating data from massive amounts of low-quality data.
[0073] In some embodiments, the operating data for each wind turbine includes wind speed and power values;
[0074] like Figure 4 As shown, the data density corresponding to each wind turbine operating data point is determined, including:
[0075] Step 401: Based on the wind speed and power values, determine the target grid corresponding to the data point determined by each wind turbine operation data in the wind speed and power grid.
[0076] The wind speed and power grid is constructed based on preset wind speed and power step sizes. Specifically, the wind speed and power grid is a virtual grid system obtained by discretizing a continuous two-dimensional wind speed-power coordinate space. When constructing this grid system, the granularity of the wind speed axis and power axis needs to be preset, i.e., the preset wind speed step size and the preset power step size.
[0077] For example, the preset wind speed step size can be set to 0.01 m / s, 0.1 m / s, or 0.2 m / s, etc., and the preset power step size can be set to 0.5%, 1%, or 2% of the rated power of the wind turbine, etc., or it can be a fixed power value, such as 10 kW or 50 kW. Choosing an appropriate step size can achieve a balance between calculation efficiency and representation accuracy.
[0078] After the wind speed and power grid is constructed, each data point determined by the wind turbine's operating data can be uniquely mapped to a specific target grid within that grid, based on its wind speed and power values. For example, a data point with a wind speed of 7.53 m / s and a power of 1520 kW can be assigned to the target grid with a wind speed range of [7.5 m / s, 7.6 m / s] and a power range of [1500 kW, 1550 kW] in a grid system with a wind speed step size of 0.1 m / s and a power step size of 50 kW. This process of assigning continuous data points to discrete grids greatly simplifies the complexity of subsequent processing.
[0079] Step 402: The number of data points located within the preset area of the target grid is taken as the data density of the wind turbine operation data belonging to the target grid.
[0080] The preset region is a neighborhood defined around a target grid. Data density calculation no longer relies solely on the number of data points within the target grid itself, but rather on the sum of all data points within this broader preset region. This approach effectively smooths the data distribution, avoiding significant deviations in density calculations caused by occasional fluctuations in the number of data points in a single grid, resulting in more stable and reliable density estimation results.
[0081] The preset region can be defined in various ways. One simple approach is that the preset region is the target grid itself. In this case, the data density of the target grid is equal to the number of data points determined by the wind turbine operation data contained within that grid. Another approach is that the preset region can be a circular area with radius R centered on the target grid, and the total number of data points determined by the wind turbine operation data within all grids covered by this circular area is counted. Yet another approach is that the preset region can be a rectangular area centered on the target grid.
[0082] After calculating the density value corresponding to each grid, all data points determined by the wind turbine operation data assigned to the same target grid are assigned the same density value, i.e., the density value of that target grid. This completes the determination of the data density for each wind turbine operation data point prior to this iteration.
[0083] By adopting the above technical solution, this invention discretizes the continuous data space into a grid and approximates the density by counting the number of data points within a preset area. This is a computationally extremely efficient method. It avoids the complex process of calculating a large number of point-to-point distances required in traditional density estimation algorithms, significantly improving the overall operational efficiency of the data cleaning method. This makes it possible to process massive amounts of wind turbine SCADA data quickly, while ensuring the stability and effectiveness of density estimation.
[0084] In some embodiments, the preset area is a rectangular area consisting of a target grid, a first grid adjacent to the target grid on the left and right, and a second grid adjacent to the target grid on the top and bottom.
[0085] The first and second grids describe the size of the neighborhood. More specifically, the first grid adjacent to the target grid on the left and right refers to extending a predetermined number of grids to the left and right, centered on the target grid, along the wind speed axis (usually the horizontal axis) of the wind speed power grid. For example, a wind speed neighborhood parameter `adj_ws` can be set to indicate extending `adj_ws` grids to the left and right. Thus, along the wind speed axis, the total width of this rectangular area covers 2... adj_ws+1 grids (including the target grid itself).
[0086] The second grid adjacent to the target grid vertically refers to a predetermined number of grids extending upwards and downwards from the target grid along the power axis (usually the vertical axis) of the wind speed power grid. For example, a power neighborhood parameter `adj_p` can be set to indicate an extension of `adj_p` grids upwards and downwards. Thus, along the power axis, the total height of this rectangular region covers 2... adj_p+1 grids (including the target grid itself).
[0087] Based on the above definition, the size of this rectangular region is jointly determined by the wind speed neighborhood parameter adj_ws and the power neighborhood parameter adj_p. For example, when adj_ws=2 and adj_p=3, for any target grid, its corresponding rectangular region is a grid with a width of 5 (2 2+1 ）、 The height is 7 grids (2 A 3×1 rectangle. When calculating the density of this target grid, it is necessary to sum the data points determined by the operating data of all wind turbines within all 35 grids covered by this 5×7 rectangular area.
[0088] To give a more concrete example, suppose we set adj_ws=1 and adj_p=1. Then, for a target grid, its corresponding preset region is a 3×3 rectangle, containing the target grid itself and its eight adjacent grids (top, bottom, left, right, upper left, upper right, lower left, and lower right). The density value of this target grid is the sum of the number of data points determined by the wind turbine operation data within these nine grids.
[0089] Using rectangular regions as preset regions offers significant technical advantages. First, it is extremely efficient in computation. By employing algorithms such as integral image or 2D prefix sum, the entire wind speed-power grid can be preprocessed once, and then the sum of all values within any rectangular region can be calculated, reducing the time complexity to O(1), i.e., independent of the size of the rectangular region. This allows for rapid density calculation even with a large neighborhood. Second, by independently setting the wind speed neighborhood parameter adj_ws and the power neighborhood parameter adj_p, rectangular regions with different aspect ratios can be flexibly defined to adapt to different wind turbine power curve shapes. For example, in regions with relatively flat power curves, a larger adj_ws and a smaller adj_p can be set; in regions with relatively steep power curves, a smaller adj_ws and a larger adj_p can be set, thus more accurately capturing the local clustering characteristics of the data.
[0090] By adopting the above technical solution, this embodiment of the invention concretizes the preset region into a flexibly configurable rectangular region and utilizes its structural advantages in computation to achieve highly efficient and accurate estimation of data density. This not only further improves the computational performance of the entire data cleaning process, but also enhances the adaptability of the method to different data distribution patterns through parameterized neighborhood definition, thereby improving the robustness and accuracy of density calculation.
[0091] In some embodiments, the iteration termination condition includes:
[0092] Maximum number of iterations; or
[0093] The data distance between the operating data of the first fan and the operating data of the second fan is less than the corresponding preset distance;
[0094] The first wind turbine operating data is the wind turbine operating data that will be retained in the next iteration operation, and the second wind turbine operating data is the wind turbine operating data that will be removed in the next iteration operation.
[0095] In this embodiment, the iteration termination condition provides two specific and complementary implementation schemes, namely, the iteration terminates when either condition is met, which aims to ensure that the iteration process terminates at the appropriate time, effectively removing abnormal data while avoiding excessive cleaning of normal data.
[0096] Specifically, the first iteration termination condition is setting a maximum number of iterations. This is a simple and direct control method. Before starting the iteration process, a positive integer is preset as the upper limit for the number of iterations, such as 50, 100, or 200. After each iteration operation, the system checks whether the number of iterations already executed has reached this upper limit. Once it is reached, the iteration process will be forcibly terminated regardless of the data cleaning effect. The main purpose of setting a maximum number of iterations is to prevent the iteration process from falling into an infinite loop due to the inability to meet other termination conditions in certain special cases, thus ensuring that the algorithm always terminates within a finite amount of time.
[0097] The second iteration termination condition is a more intelligent and adaptive control mechanism. It determines the termination based on the data distance between the first and second wind turbine operating data being less than a corresponding preset distance. Here: the first wind turbine operating data refers to the set of wind turbine operating data determined to be retained in the next iteration. In the current iteration, this data is the portion with higher density that is retained after being sorted by data density. It is considered to be closer to the normal state at this stage. The second wind turbine operating data refers to the set of wind turbine operating data determined to be removed in the next iteration. In the current iteration, this data is the portion with lower density that is filtered out.
[0098] Data distance is a measure used to quantify the degree of separation between two sets of data. This distance can be defined from multiple dimensions. For example, it can be a geometric distance, calculating the average Euclidean or Mahalanobis distance between points in the second set of wind turbine operation data and the data cluster formed by the first set of wind turbine operation data. It can also be a statistical distance, such as comparing the distribution differences of the two sets on certain key features (e.g., wind speed, power), using KL divergence or Wasserstein distance.
[0099] The preset distance is a threshold corresponding to the distance in the data mentioned above. This threshold needs to be preset based on experience or through experimentation.
[0100] The principle behind this termination condition is as follows: In the early stages of iteration, the eliminated second-wind turbine operating data are typically obvious outliers far from the core data cluster, resulting in a large data distance between them and the retained first-wind turbine operating data. As the iteration progresses, outliers are gradually cleared, and only marginal points slightly deviating from the core data cluster are removed. At this point, the second-wind turbine operating data set becomes increasingly closer to the first-wind turbine operating data set, and the calculated data distance decreases accordingly. When this data distance decreases to less than a preset distance (i.e., a threshold), it can be considered that the algorithm has begun to erode the boundary region of normal data, and continuing the iteration may mistakenly delete useful data. Therefore, the iteration should be stopped at this point to retain the maximum amount of valid data.
[0101] By adopting the above technical solution, this embodiment of the invention achieves intelligent control of the iteration process by setting a maximum number of iterations as a safety guarantee and introducing an adaptive termination condition based on data distance. This dual guarantee mechanism not only ensures the finiteness and efficiency of the algorithm, but more importantly, it can dynamically determine the optimal stopping time based on the distribution characteristics of the data itself, effectively preventing excessive cleaning of normal data, thereby maximizing the preservation of data integrity and validity while ensuring the cleaning effect.
[0102] In some embodiments, the data distance includes the average difference in wind speed and the average difference in power.
[0103] like Figure 5 As shown, the data distance between the operating data of the first wind turbine and the operating data of the second wind turbine is calculated in the following way:
[0104] Step 501: Based on the operating data of the first wind turbine, determine the average power corresponding to each wind speed range and the average wind speed corresponding to each power range.
[0105] This step aims to use the first wind turbine operating data, which is currently considered normal, to construct a benchmark or reference that can represent the normal behavior of the wind turbine, i.e., a baseline.
[0106] To establish this baseline, the wind speed axis and power axis first need to be divided into a series of continuous, non-overlapping intervals. For example, the wind speed axis can be divided into multiple wind speed intervals with a step size of 0.1 m / s, and the power axis can be divided into multiple power intervals with a step size of 100 kW.
[0107] Then, iterating through all the first wind turbine operating data, for each wind speed interval, the arithmetic mean of the power values of all first wind turbine operating data falling within that interval is calculated to obtain the average power corresponding to that wind speed interval. Similarly, for each power interval, the arithmetic mean of the wind speed values of all first wind turbine operating data falling within that interval is calculated to obtain the average wind speed corresponding to that power interval. In this way, we obtain two mapping relationships: one from wind speed intervals to average power, and the other from power intervals to average wind speed. These two mappings together constitute the baseline representing normal behavior. To avoid unstable results due to data sparsity when calculating the average value, neighborhood smoothing can be introduced, that is, data from adjacent intervals are also included in the calculation.
[0108] In some embodiments, the average power corresponding to each wind speed range is the average power including adjacent wind speed ranges, and similarly, the average wind speed corresponding to each power range is the average wind speed including adjacent power ranges.
[0109] For example, the average power corresponding to the i-th wind speed interval is based on the average power corresponding to the (i-1)-th wind speed interval, the average power corresponding to the i-th wind speed interval, and the average power corresponding to the (i+1)-th wind speed interval.
[0110] Similarly, the average wind speed corresponding to the i-th power range is based on the average of the average wind speeds corresponding to the (i-1)-th power range, the i-th power range, and the (i+1)-th power range.
[0111] Step 502: Based on the average power corresponding to each wind speed range and the average wind speed corresponding to each power range, determine the wind speed difference and power difference corresponding to the operating data of each second fan.
[0112] In this step, all the second wind turbine operating data that were removed in this iteration are traversed. For any data point, assume its coordinates are (v_d, p_d), where v_d is the wind speed value at that point and p_d is the power value at that point.
[0113] First, calculate the power difference. Based on the wind speed value v_d at this point, find the wind speed range to which it belongs, and retrieve the average power p_baseline corresponding to this wind speed range from the baseline constructed in the first step. The power difference at this point is |p_d - p_baseline| (taking the absolute value).
[0114] Next, calculate the wind speed difference. Based on the power value p_d at this point, find the power range to which it belongs, and look up the average wind speed v_baseline corresponding to this power range from the baseline. The wind speed difference at this point is |v_d - v_baseline| (take the absolute value).
[0115] Through this step, each eliminated second wind turbine operating data point will receive a power difference value and a wind speed difference value, which represent its distance from the normal behavior baseline in the vertical and horizontal directions, respectively.
[0116] Step 503: Determine the average wind speed difference based on the wind speed difference corresponding to the operating data of each second wind turbine, and determine the average power difference based on the power difference corresponding to the operating data of each second wind turbine.
[0117] In this step, the deviation of all data points removed in this iteration is summarized to obtain the final distance metric.
[0118] Specifically, the wind speed differences calculated from all the second wind turbine operating data are summed, and then divided by the total number of second wind turbine operating data. The result is the average wind speed difference for this iteration. Similarly, the average power difference is obtained by summing all the power differences.
[0119] In this embodiment, the average difference in wind speed and the average difference in power are used to quantify the average deviation of the removed data points (i.e., the second wind turbine operating data) from the retained normal data (i.e., the first wind turbine operating data) in the dimensions of wind speed and power, respectively.
[0120] Specifically, when determining the termination of the iteration, these two values can be compared with their respective preset distance thresholds (e.g., preset wind speed distance threshold and power distance threshold). The iteration termination condition is met when the average difference in wind speed and the average difference in power are both less than or either one is less than its corresponding preset distance threshold.
[0121] By adopting the above technical solution, this invention provides a clear and physically meaningful method for calculating data distance. This method constructs a normal data baseline and calculates the average deviation of the removed data points in the two key dimensions of wind speed and power, accurately quantifying the "cleaning margin effect" of each iteration. This makes the determination of iteration termination no longer a vague geometric distance, but an interpretable physical quantity deviation directly related to the wind turbine's operating characteristics, thus making the adaptive stopping control of the iteration process more precise, reliable, and robust.
[0122] In some embodiments, such as Figure 6 As shown, the methods for cleaning wind turbine operating data also include:
[0123] Step 601: Based on the power value in the operating data of each target wind turbine, group the target wind turbine operating data in multiple wind turbine operating data to obtain multiple power groups.
[0124] Among them, the multiple power groups include multiple first power groups with the highest power and multiple second power groups with the lowest power.
[0125] The purpose of this step is to analyze the distribution characteristics of the normal data obtained after cleaning along the power dimension. Specifically, the power axis can be divided into several continuous intervals from the minimum to the maximum value, with each interval being a power group. For example, the interval width can be divided into intervals of 100kW or 1% of the rated power. Then, all target wind turbine operating data are assigned to the corresponding power groups based on their own power values.
[0126] After grouping, it is necessary to identify the first power groups with the highest power and the second power groups with the lowest power from the multiple power groups. "Multiple" can be a preset number; for example, the three power groups with the highest power values can be selected as the first power groups, and the three power groups with the lowest power values can be selected as the second power groups. These groups represent normal data samples of the wind turbine under near-full-capacity and near-start-stop conditions, respectively.
[0127] Step 602: Determine a first reference wind speed value based on multiple first power groups, and determine a second reference wind speed value based on multiple second power groups.
[0128] The purpose of this step is to learn and determine the typical wind speed range boundaries corresponding to the high-power and low-power ranges from known normal data. The first and second reference wind speed values are representative wind speed statistics extracted from the target wind turbine operating data contained in the first and second power groups, respectively. For example, these reference wind speed values can be obtained by calculating the arithmetic mean, median, or a predetermined quantile of the wind speed values from these target wind turbine operating data. These two reference wind speed values provide crucial judgment criteria for subsequent data supplementation steps.
[0129] Step 603: Add the wind turbine operating data of the target wind turbine to the wind turbine operating data if the wind speed value is greater than the first reference wind speed value and less than the cut-out wind speed value, the power value is in the first power group and has been removed; and / or add the wind turbine operating data of the target wind turbine to the wind turbine operating data if the wind speed value is less than the second reference wind speed value and greater than zero, the power value is in the second power group and has been removed.
[0130] This step may include one or two of the following parallel sub-steps:
[0131] On one hand, wind turbine operating data that were previously excluded but whose wind speed values are greater than the first reference wind speed value, less than the cut-off wind speed value, and whose power values are within the first power group are added to the target wind turbine operating data. The cut-off wind speed value is an inherent safety parameter of the wind turbine, referring to the upper limit of wind speed at which the wind turbine must stop operating to protect its structural safety. The principle behind this step is that in high wind speed ranges, the wind turbine should be stable near its rated power. Data points in this area may be relatively sparse and easily deleted during iterations due to their relatively low density. Therefore, identifying and recalling data points that were excluded during iterations but whose wind speed and power values conform to the physical laws of high-power operation (i.e., wind speed between the first reference wind speed and the cut-off wind speed, and power within the high-power range) can effectively restore the integrity of the power curve at high wind speeds.
[0132] On the other hand, and / or, wind turbine operating data that were previously excluded but whose wind speed values are less than the second reference wind speed value, are greater than zero, and whose power values are in the second power group are added to the target wind turbine operating data. The principle of this step is similar to the above; it targets data from the wind turbine's start-up and shutdown phase at low wind speeds. Data points in this phase may also be mistakenly deleted due to sparsity. Therefore, identifying and adding back data points that were excluded during iterations but whose wind speed and power values conform to the physical laws of low wind speed operation (i.e., wind speed between 0 and the second reference wind speed value, and power within the low power range) ensures that the cleaned data can fully cover the transition phase from wind turbine standstill to power generation.
[0133] In this embodiment, valid data points that may be mistakenly deleted during the iteration process and are located at both ends of the wind turbine's normal operating envelope (i.e., near zero power at low wind speed and near rated power at high wind speed) are identified and supplemented, thereby improving the integrity of the target wind turbine operating data obtained after cleaning.
[0134] In some embodiments, the removed wind turbine operating data is wind turbine operating data that is removed during the execution of an iterative operation.
[0135] In some embodiments, the first reference wind speed value is the median wind speed when the target wind turbine operating data in the first power group is sorted according to the wind speed value; the second reference wind speed value is the median wind speed when the target wind turbine operating data in the second power group is sorted according to the wind speed value.
[0136] In this embodiment, after identifying the first power group with the highest power values (e.g., the three highest power groups), we extract the operating data of all target wind turbines falling into these groups. Then, we sort the wind speed values of these target wind turbine operating data from low to high. After sorting, we take the wind speed value located in the middle of the sorted list as the first reference wind speed value. If the total number of data points is even, we can take the arithmetic mean of the two middle wind speed values. A significant advantage of choosing the median as the statistic is its insensitivity to outliers. Even if a few data points with abnormal wind speeds remain in these high-power groups (e.g., extremely low or extremely high wind speed values), the median will not be significantly affected by these extreme values as much as the average, thus reflecting the typical wind speed level under high-power operating conditions more stably and reliably.
[0137] Similarly, after identifying the second power group with the lowest power values (e.g., the three lowest power groups), we extract the operating data of all target wind turbines falling into these groups. Then, we sort the wind speed values of these target wind turbine operating data from low to high and take the median wind speed value as the second reference wind speed value. This allows for a robust estimation of the upper limit of the typical wind speed range for wind turbines in low-power states near shutdown or just after startup.
[0138] For example, suppose we extract 101 target wind turbine operating data points when analyzing data from the first power group (high-power zone). After sorting the wind speed values of these 101 data points in ascending order, the wind speed value ranked 51st (i.e., the median) will be selected as the first reference wind speed value. When supplementing data later, high-power data points that were removed during the iteration and whose wind speed values are greater than this first reference wind speed value will be recalled.
[0139] By adopting the above technical solution, this embodiment of the invention effectively improves the robustness of boundary delineation by using the median wind speed as a statistical measure to determine the reference wind speed value. Since the median can resist the influence of a few extreme outliers, the calculated first and second reference wind speed values can more accurately represent the central trend of wind speed distribution under high-power and low-power operating conditions. This provides a more accurate and reliable judgment benchmark for subsequent data supplementation steps, further ensuring the stability and effectiveness of the entire data cleaning method.
[0140] In some embodiments, such as Figure 7 As shown, the wind turbine operation data of wind turbine No. 1 is cleaned. After 6 iterations and post-processing, the wind turbine operation data with the lowest density can be removed. The wind turbine operation data with poor data quality and stacking gradually becomes sparse and is eventually identified and filtered. In this process, the wind turbine operation data with poor data quality can be effectively cleaned.
[0141] Specifically, such as Figure 8 As shown, after 67 iterations and post-processing, the operating data of wind turbine No. 2 can remove the operating data of wind turbines with lower density rankings. The operating data of wind turbines with poor data quality and stacking gradually become sparse and are eventually identified and filtered. In this process, effective cleaning of wind turbine operating data with poor data quality can be achieved.
[0142] Similarly, such as Figure 9 As shown, after 41 iterations and post-processing, the operating data of wind turbine No. 3 can remove the operating data of wind turbines with lower density rankings. The operating data of wind turbines with poor data quality and stacking gradually become sparse and are eventually identified and filtered. In this process, effective cleaning of wind turbine operating data with poor data quality can be achieved.
[0143] in, Figure 7 , Figure 8 and Figure 9 The vertical axis of each diagram represents power. Red dots in the diagram indicate normal wind turbine operating data, while blue dots indicate abnormal wind turbine operating data.
[0144] like Figure 8 and Figure 9 As shown, the method for cleaning fan operation data proposed in this embodiment has strong consistency for different fans. The same parameters can be applied to fans of different specifications and operating conditions, and fine-tuning of parameters will not cause significant changes in the results, greatly reducing the difficulty of parameter tuning.
[0145] The following describes the wind turbine operation data cleaning device provided by the present invention. The wind turbine operation data cleaning device described below and the wind turbine operation data cleaning method described above can be referred to in correspondence.
[0146] In some embodiments, a cleaning device for fan operation data is proposed, such as... Figure 10 As shown, it includes the following modules:
[0147] Receiver module 1001 is used to receive operating data from multiple wind turbines;
[0148] Processing module 1002 is used to repeatedly perform the following iterative operation until the iteration termination condition is met, in order to determine the target wind turbine operating data among multiple wind turbine operating data:
[0149] Determine the data density corresponding to each wind turbine operation data. The data density is used to describe the density of data points determined based on wind turbine operation data in the wind speed and power grid.
[0150] Based on the order of all data density from high to low, the wind turbine operation data before this iteration is sorted to obtain the sorting result;
[0151] Use the top N wind turbine operating data from the sorting results as the wind turbine operating data for the next iteration.
[0152] Where N is determined based on the amount of wind turbine operating data before this iteration and the preset attenuation coefficient.
[0153] Figure 11 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 11 As shown, the electronic device may include: a processor 1110, a communication interface 1120, a memory 1130, and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1140. The processor 1110 can call logical instructions in the memory 1130 to execute a method for cleaning wind turbine operating data. The method includes: receiving multiple wind turbine operating data; repeatedly executing the following iterative operations until the iteration termination condition is met to determine the target wind turbine operating data among the multiple wind turbine operating data: determining the data density corresponding to each wind turbine operating data before the current iteration operation, where the data density is used to describe the density of data points determined based on the wind turbine operating data in the wind speed power grid; sorting the wind turbine operating data before the current iteration operation based on the descending order of all data densities to obtain a sorting result; and using the top N wind turbine operating data in the sorting result as the wind turbine operating data for the next iteration operation; wherein N is determined based on the number of wind turbine operating data before the current iteration operation and a preset attenuation coefficient.
[0154] Furthermore, the logical instructions in the aforementioned memory 1130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0155] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the wind turbine operation data cleaning method provided by the above methods. The method includes: receiving multiple wind turbine operation data; repeatedly performing the following iterative operations until the iteration termination condition is met to determine the target wind turbine operation data among the multiple wind turbine operation data: determining the data density corresponding to each wind turbine operation data, the data density being used to describe the density of the data points determined based on the wind turbine operation data in the wind speed power grid; sorting the wind turbine operation data before the current iteration operation based on the descending order of all data densities to obtain a sorting result; and using the top N wind turbine operation data in the sorting result as the wind turbine operation data for the next iteration operation; wherein, N is determined based on the number of wind turbine operation data before the current iteration operation and a preset attenuation coefficient.
[0156] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a method for cleaning wind turbine operating data provided by the methods described above. This method includes: receiving multiple wind turbine operating data sets; repeatedly performing the following iterative operations until an iteration termination condition is met to determine target wind turbine operating data among the multiple wind turbine operating data sets; determining the data density corresponding to each wind turbine operating data set; sorting the wind turbine operating data before the current iteration operation based on the descending order of all data densities to obtain a sorting result; and using the top N wind turbine operating data sets in the sorting result as the wind turbine operating data for the next iteration operation; wherein N is determined based on the number of wind turbine operating data sets before the current iteration operation and a preset attenuation coefficient.
[0157] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0158] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.
[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for cleaning fan operating data, characterized in that, include: Receives operating data from multiple wind turbines; Repeat the following iterative operation until the iteration termination condition is met to determine the target wind turbine operating data among the multiple wind turbine operating data: Determine the data density corresponding to each wind turbine operating data point, whereby the data density describes the density of the data points determined based on the wind turbine operating data within the wind speed-power grid. Based on the order of all data density from high to low, the wind turbine operating data prior to this iteration is sorted to obtain the sorting result; The top N wind turbine operating data in the sorting results will be used as the wind turbine operating data for the next iteration. Where N is determined based on the amount of wind turbine operating data mentioned before this iteration and the preset attenuation coefficient; The iteration termination conditions include: Maximum number of iterations; or The data distance between the operating data of the first fan and the operating data of the second fan is less than the corresponding preset distance; Wherein, the first wind turbine operating data is the wind turbine operating data that will be retained in the next iteration operation, and the second wind turbine operating data is the wind turbine operating data that will be removed in the next iteration operation.
2. The method for cleaning fan operation data according to claim 1, characterized in that, Each of the aforementioned wind turbine operating data includes wind speed and power values; Determining the data density corresponding to each wind turbine's operating data includes: Based on the wind speed value and the power value, the target grid corresponding to each data point determined by the wind turbine operation data is determined in the wind speed and power grid, which is constructed based on a preset wind speed step size and a preset power step size; The number of data points located within a preset area of the target grid is taken as the data density of the wind turbine operation data belonging to the target grid.
3. The method for cleaning fan operation data according to claim 2, characterized in that, The preset area is a rectangular area consisting of the target grid, a first grid adjacent to the target grid on the left and right, and a second grid adjacent to the target grid on the top and bottom.
4. The method for cleaning fan operating data according to claim 1, characterized in that, The data distance includes the average difference in wind speed and the average difference in power. The data distance between the first wind turbine operating data and the second wind turbine operating data was calculated using the following method: Based on the first wind turbine operating data, the average power corresponding to each wind speed range and the average wind speed corresponding to each power range are determined. Based on the average power corresponding to each wind speed range and the average wind speed corresponding to each power range, the wind speed difference and power difference corresponding to each second wind turbine operating data are determined. The average wind speed difference is determined based on the wind speed difference corresponding to the operating data of each of the second wind turbines, and the average power difference is determined based on the power difference corresponding to the operating data of each of the second wind turbines.
5. The method for cleaning wind turbine operating data according to any one of claims 1 to 4, characterized in that, The method for cleaning the fan operating data also includes: Based on the power value in the operating data of each target wind turbine, the operating data of the target wind turbines in the multiple wind turbine operating data are grouped to obtain multiple power groups, which include multiple first power groups with the largest power and multiple second power groups with the smallest power. A first reference wind speed value is determined based on multiple first power groups, and a second reference wind speed value is determined based on multiple second power groups; The wind turbine operating data that have wind speed values greater than the first reference wind speed value and less than the cut-off wind speed value, and whose power values are in the first power group and have been removed, are added to the target wind turbine operating data; and / or The wind turbine operating data that have wind speed values less than the second reference wind speed value but greater than zero, power values in the second power group, and have been removed, are added to the target wind turbine operating data.
6. The method for cleaning fan operating data according to claim 5, characterized in that, The first reference wind speed value is the median wind speed when the target wind turbine operating data in the first power group is sorted according to the wind speed value; the second reference wind speed value is the median wind speed when the target wind turbine operating data in the second power group is sorted according to the wind speed value.
7. A device for cleaning fan operating data, characterized in that, include: The receiving module is used to receive operating data from multiple wind turbines; The processing module is used to repeatedly perform the following iterative operation until the iteration termination condition is met, in order to determine the target wind turbine operating data among the multiple wind turbine operating data: Determine the data density corresponding to each wind turbine operating data point, whereby the data density describes the density of the data points determined based on the wind turbine operating data within the wind speed-power grid. Based on the order of all data density from high to low, the wind turbine operating data prior to this iteration is sorted to obtain the sorting result; The top N wind turbine operating data in the sorting results will be used as the wind turbine operating data for the next iteration. Where N is determined based on the amount of wind turbine operating data mentioned before this iteration and the preset attenuation coefficient; The iteration termination conditions include: Maximum number of iterations; or The data distance between the operating data of the first fan and the operating data of the second fan is less than the corresponding preset distance; Wherein, the first wind turbine operating data is the wind turbine operating data that will be retained in the next iteration operation, and the second wind turbine operating data is the wind turbine operating data that will be removed in the next iteration operation.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for cleaning wind turbine operating data as described in any one of claims 1 to 6.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for cleaning wind turbine operating data as described in any one of claims 1 to 6.