Method and device for identifying and correcting power limiting data of wind power plant

Abstract

The invention relates to the technical field of water, wind and light power prediction, and discloses a wind power plant power limit data identification and correction method and device, and the method comprises the steps: employing a CNN-LSTM model to learn normal wind speed-power curve features, and through the combination of one-dimensional convolution and a pooling layer, effectively capturing the correlation features of wind speed and power, and carrying out the recognition and correction of the power limit data of a wind power plant. The identification accuracy and stability of the abnormal section of the wind power plant are improved; by adopting multi-dimensional wind regime feature analysis and similarity calculation, the complex wind power plant operation state can be effectively identified and processed, the limitation of only depending on a single feature in the prior art is overcome, and the comprehensiveness and accuracy of wind power plant abnormal data identification are improved; the weight of each similar section is determined based on the similarity of the fan states, and the reconstruction power value of the power limiting section is obtained through a weighted summation mode, so that the reliability and the accuracy of the operation data of the wind power plant are improved.

Description

Technical Field

[0001] This invention relates to the field of hydropower, wind power, and solar power prediction technology, specifically to a method and apparatus for identifying and correcting wind farm power curtailment data. Background Technology

[0002] With the rapid development of wind power technology, the installed capacity of wind power has continued to expand and grow rapidly, becoming a key development direction for new power systems in the modern energy system. Currently, during wind farm operation, power curtailment occurs frequently due to multiple constraints such as grid dispatch mechanisms and power absorption capacity. The existence of power curtailment data not only disrupts the continuity and regularity of normal power data from wind farms but also brings many challenges to subsequent data analysis and processing.

[0003] Currently, the mainstream methods for processing wind farm power curtailment data fall into two categories: statistical analysis based on historical data and simple mathematical models. Statistical analysis based on historical data typically involves simply filling in the mean or matching similar patterns based on the distribution characteristics of power data from the same historical period. However, this method fails to fully consider the dynamic impact of key meteorological factors such as wind speed, wind direction, and turbulence intensity on power, and is also ill-suited to the complex and ever-changing operating environment of wind farms, leading to significant discrepancies between the reconstructed data and the actual situation. Simple mathematical models (such as linear regression models), due to the highly nonlinear and uncertain nature of wind farm power changes, struggle to accurately capture the complex relationships behind the curtailment data, and similarly cannot achieve accurate reconstruction of the data. Summary of the Invention

[0004] In view of this, the present invention provides a method and apparatus for identifying and correcting wind farm power curtailment data, in order to solve the problems of low identification accuracy and poor reconstruction precision of power curtailment data.

[0005] In a first aspect, the present invention provides a method for identifying and correcting wind farm power curtailment data, comprising: learning the characteristics of a normal wind speed-power curve using a CNN-LSTM model based on preprocessed wind speed-power time series data; identifying abnormal segments of the wind speed-power time series data based on the characteristics of the normal wind speed-power curve learned by the CNN-LSTM model, and verifying and correcting the abnormal segments; extracting wind condition characteristics within the power curtailment segment and performing similarity analysis, and obtaining the reconstructed power value of the power curtailment segment based on the similarity of the wind turbine status.

[0006] This invention employs a CNN-LSTM model to learn the characteristics of normal wind speed-power curves. Through a combination of one-dimensional convolution and pooling layers, it effectively captures the correlation between wind speed and power, improving the accuracy and stability of identifying abnormal wind farm sections. By using multi-dimensional wind condition feature analysis and similarity calculation, it effectively identifies and processes complex wind farm operating states, overcoming the limitations of existing technologies that rely solely on a single feature, and improving the comprehensiveness and accuracy of identifying abnormal wind farm data. Based on the similarity of wind turbine states, the weights of each similar section are determined, and the reconstructed power value of the curtailment section is obtained through weighted summation, improving the reliability and accuracy of wind farm operating data.

[0007] In one optional implementation, the preprocessing of wind speed-power time series data includes: normalizing the wind speed-power time series data using quantile normalization; and filtering the wind speed and power data using a Savitzky-Golay filter.

[0008] In one optional implementation, the process of learning the features of a normal wind speed-power curve using a CNN-LSTM model includes: using preprocessed wind speed-power time series data as a training set, dividing the continuous wind speed-power time series data into multiple samples according to a time window, and performing preprocessing operations on the sample data; using one-dimensional convolution to capture the correlation features between wind speed and power at the same time point; adding MaxPooling1D to the pooling layer to compress the dimension of the correlation features and retain key local features; inputting the feature vector output by the CNN into the LSTM layer to capture the change pattern of features over time; connecting the output of the LSTM layer to three fully connected layers to further fuse spatiotemporal features; and using a linear activation function in the output layer to output reconstructed wind speed-power data, with the output data having the same dimension as the input data.

[0009] In one optional implementation, the process of identifying abnormal segments in wind speed-power time series data includes: calculating the error value between the reconstructed power and the input power; when the reconstructed error exceeds a preset threshold, the window is considered to contain a suspected power curtailment segment; based on the detected reconstructed error window, further analyzing the steep drop characteristics of the power series; based on the steep drop characteristics, calculating the power acceleration of the abnormal segment, and marking segments with acceleration less than a preset threshold as the first segment; verifying the first segment through SCADA status information, filtering out power steep drops caused by extreme weather, and obtaining the power curtailment segments.

[0010] This invention, through SCADA status information verification and wind condition feature analysis, can effectively filter out power drops caused by extreme weather, ensuring accurate identification of power curtailment sections and overcoming the shortcomings of existing technologies in excluding wind turbine shutdown event data.

[0011] In one optional implementation, the process of extracting wind condition features within the power rationing area and performing similarity analysis includes: extracting wind condition features within the power rationing area; searching for multiple normal similar segments with similar wind conditions to the power rationing area within the same seasonal time range as the power rationing area, and using them as candidate segments; extracting wind speed features, wind direction features, and turbulence intensity features of the power rationing area and the candidate segments in the same season as the power rationing area; calculating the Euclidean distance between the wind speed features, wind direction features, and turbulence intensity of the power rationing area and the candidate segments to obtain the initial similarity of each candidate segment; selecting a preset number of target candidate segments based on the initial similarity; and using DTW to calculate the similarity of wind conditions between the power rationing area and the target candidate segments to obtain multiple similar segments.

[0012] In one optional implementation, the process of obtaining the reconfigurable power value of the power-canceling section based on the similarity of wind turbine status includes: determining the weight of each similar section based on the similarity of wind turbine status; and performing a weighted summation of the power data of the similar sections based on the weight values ​​to obtain the reconfigurable power value of the power-canceling section.

[0013] In one optional implementation, when a power rationing period exists simultaneously in multiple overlapping base time windows, the process of obtaining multiple reconstructed powers includes: extracting the reconstructed power values ​​of power rationing periods in multiple segments; for each segment containing power rationing periods, first locating the time points corresponding to the power rationing period and extracting the reconstructed power of these time points; fusing candidate values ​​according to segment credibility weights, since each segment containing power rationing periods has different credibility in representing the power rationing period, and using weighted averaging after assigning weights, the final reconstructed power is obtained.

[0014] Secondly, the present invention provides a wind farm power curtailment data identification and correction device, the device comprising: a model training module, used to learn the normal wind speed-power curve features using a CNN-LSTM model based on preprocessed wind speed-power time series data; an identification and correction module, used to identify abnormal segments of the wind speed-power time series data based on the normal wind speed-power curve features learned by the CNN-LSTM model, and to verify and correct the abnormal segments; and a reconstruction module, used to extract wind condition features within the power curtailment segment and perform similarity analysis, and to obtain the reconstructed power value of the power curtailment segment based on the similarity of the wind turbine status.

[0015] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the wind farm power curtailment data identification and correction method of the first aspect or any corresponding embodiment described above.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the wind farm power curtailment data identification and correction method of the first aspect or any corresponding embodiment described above.

[0017] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the wind farm power curtailment data identification and correction method described in the first aspect or any corresponding embodiment. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the wind farm power curtailment data identification and correction method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] According to an embodiment of the present invention, a method for identifying and correcting wind farm power curtailment data is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0022] This embodiment provides a method for identifying and correcting wind farm power curtailment data, which can be used in the aforementioned mobile terminal. Figure 1 This is a flowchart of a wind farm power curtailment data identification and correction method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S1: Based on the preprocessed wind speed-power time series data, the CNN-LSTM model is used to learn the characteristics of the normal wind speed-power curve.

[0023] Optionally, the preprocessing of wind speed-power time series data includes: normalizing the wind speed-power time series data using quantile normalization; and filtering the wind speed and power data using a Savitzky-Golay filter to retain trend characteristics while suppressing high-frequency noise.

[0024] Specifically, the wind speed-power time series data are normalized using quantile normalization. This method differs from traditional min-max normalization or Z-score standardization; its core principle is to map the original data to a preset quantile distribution interval (usually [0,1]) to achieve a uniform correction of the data distribution pattern. Secondly, the wind speed and power data are filtered using a Savitzky-Golay filter. This filter is a sliding window filtering method based on local polynomial fitting, particularly suitable for processing time series data containing high-frequency noise.

[0025] In practice, it is necessary to set appropriate window sizes and polynomial orders based on the sampling frequency of wind speed and power data. The window size determines the time scale of local fitting; too small a window will fail to effectively filter noise, while too large a window may smooth out key short-term fluctuations. The polynomial order affects the flexibility of the fitted curve, requiring a balance between noise suppression and feature preservation. For wind speed data, filtering can eliminate glitches caused by high-frequency sensor vibrations (such as instantaneous wind speed jumps), highlighting its true gradual trend. For power data, it can filter high-frequency noise caused by grid load fluctuations, restoring the power output pattern of the unit under stable wind speeds. After filtering, the wind speed-power sequence has smoother data curves and more prominent key features, significantly improving the learning accuracy of subsequent models for normal operating modes.

[0026] Step S2: Learn the characteristics of normal wind speed-power curve based on CNN-LSTM model, identify abnormal segments in wind speed-power time series data, and verify and correct abnormal segments.

[0027] Specifically, to avoid misjudgment, abnormal sections need to be verified twice: firstly, by combining external data such as power grid dispatch records and meteorological warnings, non-power curtailment anomalies caused by extreme weather (such as unit shutdowns due to typhoons) or equipment failures can be ruled out; secondly, by using sliding window analysis, the continuity and fluctuation characteristics of the abnormal section can be verified to ensure that they conform to the typical performance of power curtailment phenomena (such as maintaining a stable low value after a sudden drop in power). For the verified and confirmed power curtailment anomaly sections, further interpolation correction is used to eliminate noise interference and ensure the accuracy of the anomaly boundary.

[0028] Step S3: Extract wind condition features within the power-restricted section and perform similarity analysis. Based on the similarity of wind turbine status, obtain the reconfigurable power value of the power-restricted section.

[0029] Optionally, the process of obtaining the reconfigurable power value of the power-restricted section based on the similarity of the wind turbine status includes: (1) determining the weight of each similar section based on the similarity of the wind turbine status; and (2) performing a weighted summation of the power data of the similar sections based on the weight values ​​to obtain the reconfigurable power value of the power-restricted section.

[0030] In some alternative implementations, the process of learning the characteristics of the normal wind speed-power curve using a CNN-LSTM model includes: (1) The preprocessed wind speed-power time series data is used as the training set, and the continuous wind speed-power time series data is divided into multiple samples according to the time window, and the sample data is preprocessed.

[0031] Optionally, the sample window length should cover the longest power curtailment interval in the sample. That is, when the longest power curtailment time in the historical data of a wind farm is 1 day, the sample window length should be no less than 1 day.

[0032] For example, to adapt to the batch training requirements of the model, the continuous data needs to be divided into time windows. Taking the common sampling frequency of wind farms as an example, if the data is collected at fixed intervals of 15 minutes or 30 minutes, each record contains the wind speed (unit: m / s) and power (unit: MW) data at the corresponding time, that is, the format of a single data record is (v t p t ), where v t p represents the wind speed value at time t. t This represents the power value at time t. As time progresses, these discrete records form a continuous time series: with a sampling interval of 15 minutes, 4 data points are generated per hour, and 96 data points per day; with a sampling interval of 30 minutes, 48 ​​data points are generated per day. If a longer time span is needed to capture intraday or cross-day variations, the total length of the time series can be set to T (e.g., T=96, corresponding to 1 day of data at 15-minute intervals; or T=48, corresponding to 1 day of data at 30-minute intervals). In this case, the complete time series can be represented as [(v1,p1),(v2,p2),...,(v...]. T ,p T Each element is a wind speed-power data pair at a specific moment, and the elements are arranged in chronological order.

[0033] Based on the aforementioned continuous time series, a sliding window method is used for sample partitioning: the window length for each sample is set to 60 time steps (i.e., containing 60 pairs of wind speed-power data). The window slides across the time series with a fixed step size (e.g., 10 time steps) until it covers the entire series. Since each time step contains two features (wind speed and power), each sample will ultimately form a 60×2 matrix, where the rows represent time steps and the columns correspond to the wind speed and power features, respectively.

[0034] (2) One-dimensional convolution is used to capture the correlation features between wind speed and power at the same time point.

[0035] Optionally, assuming 64 convolutional kernels are used, each with a size of 3 (i.e., covering 3 consecutive time points), and all are non-linearly transformed using the ReLU activation function. Each convolutional kernel can be viewed as a 3×2 weight matrix (3 rows corresponding to 3 time points, and 2 columns corresponding to wind speed and power features respectively). The 64 convolutional kernels will generate 64 feature maps of size 58×1 (the length becomes 60-3+1=58 after 3×1 convolution of 60 time points), and each feature map corresponds to a spatial distribution of an association pattern. For example, in a certain feature map, the feature values ​​from time point t=10 to t=20 are generally high, indicating that the wind speed and power association in this interval conforms to the typical pattern learned by this convolutional kernel (such as linear power output under stable wind speed); while the region with a feature value of 0 may correspond to abnormal fluctuations or irrelevant patterns.

[0036] (3) MaxPooling1D is added to the pooling layer to compress the dimension of the associated features and retain the key local features.

[0037] Specifically, after processing by convolutional layers, the feature maps contain a large amount of correlation information, but they are high-dimensional and redundant. A MaxPooling1D pooling layer is introduced, with a pooling window size of 2, to sample the feature maps output by the convolution: the maximum value within each window is selected as the representative value for that region. This not only compresses the feature dimension to half of its original value, reducing the computational load of the model, but also strengthens the salient features within each local region (such as the power jump signal during sudden wind speed changes) through the "maximum" operation, filtering out secondary information so that subsequent processing can focus on key correlation patterns.

[0038] (4) Input the feature vector output by CNN into the LSTM layer to capture the change pattern of features over time.

[0039] Specifically, a bidirectional LSTM is employed, consisting of two LSTM layers, each with 128-256 neurons. The stacking of the two LSTM layers enables layer-by-layer feature abstraction: the first LSTM layer (128 neurons) processes pooled local features, extracting mid-level patterns such as "wind speed fluctuation cycles within one hour"; the second LSTM layer (256 neurons), based on the output of the first layer, further mines cross-time-period correlations (such as "similar wind speed-power patterns at the same time of day for three consecutive days"). Setting the number of neurons to 128-256 is the optimal choice after comprehensively considering the complexity of wind farm time-series data (requiring sufficient parameters to store long-range dependencies) and the computational efficiency of the model (avoiding overfitting due to over-parameterization). Ultimately, this allows the LSTM layers to accurately capture multi-scale time-series patterns from minute to day, providing rich dynamic information for feature fusion in the fully connected layers.

[0040] (5) Connect the output of the LSTM layer to three fully connected layers to further integrate spatiotemporal features.

[0041] Specifically, although the feature vectors output by the LSTM layer already contain the changing patterns over time, the interactive relationships between the features still need to be strengthened. Through layer-by-layer mapping of three fully connected layers (with 128, 64, and 32 neurons respectively), and by using a combination of linear transformations and non-linear activation functions (such as ReLU), cross-dimensional feature fusion is achieved.

[0042] (6) The output layer uses a linear activation function to output the reconstructed wind speed-power data. The output data is in the same dimension as the input data.

[0043] Specifically, the feature vectors fused by the fully connected layer are transformed by the output layer using a linear transformation (e.g., setting two output neurons, corresponding to the reconstructed values ​​of wind speed and power respectively) to generate reconstructed data of the same dimension as the input samples. The choice of the linear activation function ensures that the numerical range of the output data remains consistent with the original input (avoiding numerical distortion caused by nonlinear transformation), allowing the reconstructed data to be directly compared with the original input data for error calculation (e.g., mean squared error), providing accurate loss feedback for model training, thereby continuously optimizing model parameters and improving the learning accuracy of normal wind speed-power curve features.

[0044] In some alternative implementations, the process of identifying outlier segments in wind speed-power time series data includes: (1) Calculate the error between the reconstructed power and the input power. When the reconstructed error exceeds the preset threshold, the window is considered to contain a suspected power-limited section.

[0045] (2) Based on the detection of the reconstruction error window, further analyze the steep drop characteristics of the power sequence.

[0046] For suspected windows with excessive reconstruction errors, the power time series within them is extracted for steep drop feature analysis to distinguish power curtailment from other types of anomalies (such as short-term equipment failure). Simultaneously, the start time of the steep drop, its duration (e.g., 5 time steps from t=10 to t=15), and the maximum drop (e.g., a single-step drop of 3MW) are recorded to form a steep drop feature vector.

[0047] (3) Based on the steep drop characteristics, calculate the power acceleration of the abnormal section and mark the section with acceleration less than the preset threshold as the first section.

[0048] Within the steep power drop zone, the acceleration value at each time point is calculated, and the minimum value is taken as the characteristic acceleration of that zone. That is, the power drop caused by power curtailment usually exhibits the characteristic of "rapid decline followed by stabilization", so the acceleration value is relatively small; while equipment failure may cause drastic fluctuations in acceleration. The preset threshold can be set according to the acceleration distribution of historical power curtailment events. When the characteristic acceleration of a zone is less than the threshold, it is marked as the first zone (i.e., the high-probability power curtailment zone).

[0049] (4) Verify the first section through SCADA status information, filter out the power drop caused by extreme weather, and obtain the power restriction section.

[0050] SCADA (Supervisory Control and Data Acquisition) records the real-time operating parameters of the wind farm, including wind speed, wind direction, ambient temperature, and turbine status codes. After verification, the remaining first segment is the final identified power curtailment segment, ensuring the accuracy and reliability of the identification results.

[0051] In some optional implementations, when a power-limiting period exists simultaneously within multiple overlapping base time windows, and is thus identified as multiple overlapping power-limiting segments, resulting in multiple reconfiguration power, i.e.: For example, if the power rationing period is from 16:15 to 18:00 on a certain day of a certain month of a certain year (if each sample point is 15 minutes, then there are 8 power rationing periods), and if the sample window length is 20 (i.e., 5 hours), the fixed step size is 4 (i.e., 1 hour), and the start time of the sample selection is the hour, then the power rationing interval exists within 5 sample intervals: 12:00-16:45, 13:00-17:45, 14:00-18:45, 15:00-19:45, and 16:00-20:45.

[0052] The principle for obtaining multiple reconstructed powers in the above situation is "based on segment overlap weight, fusion of multi-window reconstructed results". That is, each window is determined to be a power-restricted segment after screening, and each obtains a set of reconstructed powers (the power value at each time point within the corresponding window) through "weighting of similar segments". The specific implementation method is as follows: (1) Extract the reconfigured power values ​​of the power restriction time in multiple sections. For each section containing the power restriction time, first locate the time point corresponding to the power restriction period and extract the reconfigured power of these time points.

[0053] (2) After extraction, there are multiple candidate values ​​for reconfigurable power at each time point during the power restriction period. For example, in the above example, there are 5 reconfigurable power values ​​at the power restriction time of 16:30.

[0054] (3) The candidate values ​​are fused according to the segment credibility weight. Each segment containing the power curtailment time has a different “representation credibility” for the power curtailment time. It needs to be assigned a weight and then weighted to obtain the final reconstructed power. The weight is set according to the overlap ratio between the segment and the power curtailment time.

[0055] The higher the percentage of time a power outage occurs within a given section, the higher its weight. The weight is calculated using the following formula:

[0056] in, L overlap It refers to the overlap duration between the power outage section and the power rationing time. L window It is the total duration of the segment.

[0057] (4) For the multiple candidate reconstruction powers at each time point, the final value is calculated by weighted average, and the calculation formula is as follows:

[0058] Where k is the number of power rationing sections that include that time point; w i It is the first i The weight of each segment (can be calculated using a single dimension or a weighted sum of multiple dimensions); P i ( t ) is the first i Each segment in time t Reconstruction power.

[0059] If the power outage occurs only in a single section, the reconfiguration power of that section will be used directly as the final value, without the need for merging.

[0060] Therefore, based on this, when a power curtailment occurs in multiple segments, instead of selecting a single reconfigurable power, candidate values ​​for the corresponding time points in multiple segments are extracted and weighted according to segment reliability. This approach maximizes the accuracy and robustness of the final reconfigurable power.

[0061] In some optional implementations, the process of extracting wind condition characteristics within the power rationing zone and performing similarity analysis includes: (1) Extract wind characteristics in the power rationing area.

[0062] For example, a "power restriction zone" has been identified: denoted as (Time range), and extreme weather / fault interference has been excluded; a "candidate section pool" has been prepared: all normally operating wind speed-power sections within the same season, denoted as... The duration of the power rationing section and all candidate sections is the same (e.g., 24 hours, with a sampling frequency of 15 minutes, i.e., 96 15-minute sampling points).

[0063] The core wind condition features, such as wind speed, wind direction, and turbulence intensity, are uniformly extracted. The specific process for extracting wind condition features within the power rationing area is as follows: For power rationing sections and each candidate segment The following features were extracted (all are time-series features, i.e., each feature is a sequence of length T, where T is the number of sampling points in the segment), and then normalized: ① Wind speed characteristics (core wind conditions that directly determine power generation):

[0064] in, This represents the wind speed values ​​at various sampling times within the section, such as the wind speed values ​​sampled at 15 / 30 minutes, in m / s. t =1,2,…,T.

[0065] ② Wind direction characteristics (wind turbines are sensitive to wind direction; at the same wind speed, the power generation efficiency differs depending on the wind direction):

[0066] in, This refers to wind direction angles, such as 0° = northerly wind, 90° = easterly wind. t =1,2,…,T.

[0067] ③ Turbulence intensity characteristics (reflecting wind speed stability; high turbulence intensity indicates large power fluctuations and is a key indicator of wind condition similarity):

[0068] in, The turbulence intensity at each sampling time within the section. t =1,2,…,T.

[0069] (2) Within the same season as the power rationing area, search for multiple normal similar areas with similar wind conditions to the power rationing area and use them as candidate areas.

[0070] Specifically, wind conditions at wind farms exhibit significant seasonal characteristics (e.g., lower wind speeds in summer, higher wind speeds in winter, or seasonal reversals of wind direction in monsoon regions). Therefore, the search scope needs to be limited to the same season as the curtailment area to reduce the interference of seasonal differences on wind condition similarity. Within this timeframe, all normal operating periods without curtailment are selected, and these periods are divided into consecutive segments of equal length based on the duration of the curtailment area. Through preliminary matching, a batch of segments with roughly matching wind condition characteristics are selected from a large number of normal segments as candidate segments to avoid excessive computational load in subsequent steps.

[0071] (3) Extract wind speed characteristics, wind direction characteristics, and turbulence intensity characteristics of the power rationing section and the candidate section in the same season as the power rationing section.

[0072] (4) Calculate the Euclidean distance between the wind speed characteristics, wind direction characteristics and turbulence intensity characteristics of the power-restricted section and the candidate section to obtain the initial similarity of each candidate section.

[0073] Specifically, the Euclidean distances between the power-restricted section and each candidate section are calculated separately in three dimensions: wind speed, wind direction, and turbulence intensity. The combined distance is then obtained by weighted summation. The initial similarity is inversely proportional to the combined distance; that is, the smaller the combined distance, the greater the initial similarity between the candidate section and the power-restricted section, indicating that their basic wind characteristics are closer.

[0074] Optionally, Euclidean distance is the most basic "numerical similarity" used for quick coarse screening of candidate segments. The calculation logic is as follows: ① Calculation of Euclidean distance for a single feature For each type of characteristic, calculate the power rationing section separately. and candidate segments Euclidean distance: Wind speed characteristic distance ( The calculation formula for ) is as follows:

[0075] in, Power rationing area and candidate segments The wind speed characteristics of the Euclidean distance. Power rationing area Inside t Wind speed value at the time of sampling. Candidate segment Inside t Wind speed value at the time of sampling.

[0076] Wind direction characteristic distance ( The calculation formula for ) is as follows:

[0077] in, Power rationing area and candidate segments The wind direction characteristics of the Euclidean distance. Power rationing area Inside t The wind direction angle value at the sampling time. Candidate segment Inside t The wind direction angle value at the time of sampling.

[0078] Turbulence intensity characteristic distance ( The calculation formula for ) is as follows:

[0079] in, Power rationing area and candidate segments The Euclidean distance of the turbulence intensity characteristics Power rationing area Inside t The turbulence intensity value at the sampling time. Candidate segment Inside t The turbulence intensity value at the sampling time.

[0080] ② Initial similarity The weighted sum of the distances of the three types of features (the weights can be adjusted according to the actual scenario) yields the comprehensive distance;

[0081] in, The weight of the distance characteristic of wind speed; The weight of the distance characteristic of wind direction; This represents the weight of the turbulence intensity characteristic distance. The initial similarity is inversely proportional to the comprehensive distance; that is, the smaller the comprehensive distance, the greater the initial similarity between the candidate segment and the power-restricted segment, meaning the closer the basic wind condition characteristics are.

[0082] (5) Based on the initial similarity, select a preset number of target candidate segments.

[0083] Specifically, a predetermined number of segments with the highest similarity are selected as target candidate segments. The purpose of this step is to eliminate obviously dissimilar segments through preliminary screening, thereby reducing the load on subsequent high-precision calculations. The predetermined number needs to be determined based on the amount of historical data from the wind farm, ensuring sufficient candidate samples to avoid missing the optimal segment, while also keeping it within the limits of computing resources.

[0084] For example, the comprehensive distance for all candidate segments Sort the segments in ascending order (closer distances indicate greater similarity), and select the top 100 segments as "target candidate segments" (denoted as...). Through the above steps, most dissimilar segments can be quickly eliminated, reducing the computational cost of subsequent DTW calculations (DTW computation is much greater than Euclidean distance).

[0085] (6) DTW was used to calculate the similarity of wind conditions between the power-restricted section and the target candidate section, and multiple similar sections were obtained.

[0086] Specifically, the DTW (Dynamic Time Warping) algorithm effectively solves the problems of inconsistent time series data lengths or phase shifts, and is particularly suitable for sequence alignment with dynamic characteristics such as wind conditions. For the target candidate segment and the power-restricted segment, continuous wind speed and wind direction time series are extracted respectively, and the time series distance between them is calculated using DTW. The smaller the DTW distance, the more similar the wind condition time series change patterns of the two segments are. Combining the DTW distance with the initial similarity, the segments with the highest overall similarity are finally selected as similar segments, providing accurate reference samples for subsequent power reconfiguration.

[0087] For example, the drawback of Euclidean distance is that it requires two time series to be strictly aligned (comparing values ​​at the same point in time), but in reality, wind condition series may have "phase shifts" (for example, the peak wind speed in the power-restricted section is at 10:00, while in the candidate section it is at 11:00, but the overall trend is the same) - DTW (Dynamic Time Warping) can solve this problem and allow the series to be "flexibly aligned".

[0088] ①The core logic of DTW DTW finds the optimal alignment path between two sequences by constructing a "cost matrix" and calculates the total cost of the path as the distance (the smaller the cost, the higher the similarity).

[0089] ②The DTW calculation steps are as follows, where wind speed characteristics are taken as an example, and wind direction / turbulence intensity are calculated similarly.

[0090] Assuming wind speed sequence in the power rationing area The wind speed sequence for the target candidate section is : A. Constructing the cost matrix M (size T × T ): (Each element is the absolute error between corresponding points in two sequences) i =1,2,…,T, j =1,2,…,T; B. Constructing the cumulative cost matrix D The specific process is as follows: initialization: ; First line: (You can only move to the right); First column: (You can only go down); Other locations: (You can move to the right, down, or diagonally; choose the path with the least cost.) C.DTW distance: (Divided by length normalization). ③ Overall DTW similarity Similar to the Euclidean distance, the DTW distance is weighted and summed for wind speed, wind direction, and turbulence intensity (the weights can be adjusted according to the actual scenario and may differ from the weights in the initial similarity formula) to obtain the final DTW distance:

[0091] After the initial screening, precise screening is needed, such as selecting 5 similar segments, as follows: For 100 target candidate segments Sort in ascending order; Take the first 5 segments as the "final similar segments" (denoted as...). ); These five sections are the normal operating sections with the most similar wind conditions and power rationing within the same season.

[0092] This embodiment also provides a wind farm power curtailment data identification and correction device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0093] This embodiment provides a wind farm power curtailment data identification and correction device, the device comprising: The model training module is used to learn the features of normal wind speed-power curves based on preprocessed wind speed-power time series data using a CNN-LSTM model. The identification and correction module is used to learn the characteristics of normal wind speed-power curves based on the CNN-LSTM model, identify abnormal segments in the wind speed-power time series data, and verify and correct the abnormal segments. The reconfiguration module is used to extract wind condition characteristics within the power curtailment zone and perform similarity analysis. Based on the similarity of the wind turbine status, the reconfiguration power value of the power curtailment zone is obtained.

[0094] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0095] In this embodiment, the wind farm power curtailment data identification and correction device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0096] This invention also provides a computer device having the above-described features. Figure 2 The device shown is for identifying and correcting power curtailment data in wind farms.

[0097] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 2 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 2 Take a processor 10 as an example.

[0098] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0099] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0100] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0101] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0102] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 2 Taking the example of a connection between China and Israel via a bus.

[0103] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0104] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0105] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0106] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A method for identifying and correcting wind farm power curtailment data, characterized in that, include: Based on the preprocessed wind speed-power time series data, the CNN-LSTM model is used to learn the characteristics of the normal wind speed-power curve. Based on the CNN-LSTM model, the normal wind speed-power curve features are learned, abnormal segments of the wind speed-power time series data are identified, and the abnormal segments are verified and corrected. Wind condition characteristics within the power curtailment zone are extracted and similarity analysis is performed. Based on the similarity of wind turbine status, the reconfigurable power value of the power curtailment zone is obtained.

2. The wind farm power curtailment data identification and correction method according to claim 1, characterized in that, The preprocessing of wind speed-power time series data includes: The wind speed-power time series data were normalized using the quantile normalization method. The wind speed and power data were filtered using a Savitzky-Golay filter.

3. The wind farm power curtailment data identification and correction method according to claim 1, characterized in that, The process of learning the characteristics of the normal wind speed-power curve using a CNN-LSTM model includes: The preprocessed wind speed-power time series data is used as the training set, and the continuous wind speed-power time series data is divided into multiple samples according to the time window, and the sample data is preprocessed. One-dimensional convolution is used to capture the correlation features between wind speed and power at the same time point; By incorporating MaxPooling1D into a pooling layer, the dimension of associated features is compressed, while key local features are preserved. The feature vector output by the CNN is input into the LSTM layer to capture the changes in features over time. The output of the LSTM layer is connected to three fully connected layers to further fuse spatiotemporal features. The output layer uses a linear activation function to output reconstructed wind speed-power data, and the output data has the same dimension as the input data.

4. The wind farm power curtailment data identification and correction method according to claim 3, characterized in that, The process of identifying outlier segments in wind speed-power time series data includes: Calculate the error between the reconstructed power and the input power. When the reconstructed error exceeds a preset threshold, the window is considered to contain a suspected power-limited section. Based on the detection of the reconstruction error window, the steep drop characteristics of the power sequence are further analyzed; Based on the steep drop characteristics, the power acceleration of the abnormal section is calculated, and the section with acceleration less than a preset threshold is marked as the first section; The first section is verified by SCADA status information, and power-restricted sections are filtered out due to sharp power drops caused by extreme weather.

5. The method for identifying and correcting wind farm power curtailment data according to claim 1, characterized in that, The process of extracting wind condition characteristics within power rationing areas and performing similarity analysis includes: Extract wind condition characteristics within the power rationing area; Within the same seasonal timeframe as the power rationing area, search for multiple normal similar areas with similar wind conditions to the power rationing area and use them as candidate areas. Extract wind speed, wind direction, and turbulence intensity characteristics from the power rationing area and candidate areas in the same season as the power rationing area; Calculate the Euclidean distance between the power-restricted section and the candidate section in terms of wind speed characteristics, wind direction characteristics, and turbulence intensity to obtain the initial similarity of each candidate section; Based on the initial similarity, a preset number of target candidate segments are selected; DTW was used to calculate the similarity of wind conditions between the power-restricted sections and the target candidate sections, resulting in multiple similar sections.

6. The method for identifying and correcting wind farm power curtailment data according to claim 1, characterized in that, The process of obtaining the reconfigurable power value of the power-canceling section based on the similarity of wind turbine conditions includes: The weights of each similar segment are determined based on the similarity of the wind turbine's condition. The power data of similar sections are weighted and summed based on the weight values ​​to obtain the reconfigured power value of the power-restricted section.

7. The wind farm power curtailment data identification and correction method according to claim 6, characterized in that, When a power curtailment period exists simultaneously within multiple overlapping base time windows, the process of obtaining multiple reconfigured power includes: Extract the reconfigurable power values ​​at the power rationing time in multiple segments. For each segment containing the power rationing time, first locate the time point corresponding to the power rationing period and extract the reconfigurable power at these time points. Candidate values ​​are fused according to the segment credibility weight. Each segment containing the power rationing time has a different credibility in representing the power rationing period. The final reconstructed power is obtained by weighting the values ​​after assigning weights.

8. A wind farm power curtailment data identification and correction device, characterized in that, The device includes: The model training module is used to learn the features of normal wind speed-power curves based on preprocessed wind speed-power time series data using a CNN-LSTM model. The identification and correction module is used to learn the characteristics of the normal wind speed-power curve based on the CNN-LSTM model, identify abnormal segments in the wind speed-power time series data, and verify and correct the abnormal segments. The reconfiguration module is used to extract wind condition characteristics within the power curtailment zone and perform similarity analysis. Based on the similarity of the wind turbine status, the reconfiguration power value of the power curtailment zone is obtained.

9. A computer device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the wind farm curtailment data identification and correction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the wind farm power curtailment data identification and correction method according to any one of claims 1 to 7.

11. A computer program product, characterized in that, It includes computer instructions for causing a computer to execute the wind farm power curtailment data identification and correction method according to any one of claims 1 to 7.

Images