A wind turbine generator system monitoring data alignment method

Abstract

In the wind turbine monitoring data alignment method, device and storage medium provided in the application, a monitoring data set of at least one operating state of a wind turbine is constructed, a target operating state to be analyzed and a monitoring variable to be aligned in the target operating state are determined, whether data of the monitoring variable to be aligned in the monitoring data set meets an alignment condition is determined, and in response to the data in the monitoring data set meeting the alignment condition, alignment data of the monitoring variable to be aligned is generated. Thus, the application provides a wind turbine monitoring data alignment method, generates "new" data capable of reflecting the operating state mode of the wind turbine, thereby expanding the sample size of the state monitoring data, achieving alignment of the monitoring data, and providing a data basis for subsequent accurate management and control of the state of the wind turbine.

Description

Technical Field

[0001] This application relates to the field of wind turbine generator data analysis technology, and in particular to a method, device and storage medium for aligning monitoring data of wind turbine generators. Background Technology

[0002] The causes of abnormal state modes in wind turbine generators are complex. An abnormal state is the result of the combined effects of multiple coupling factors within the system itself, as well as the external environment in which the system operates. Furthermore, because the probability of abnormal operation varies among different types of subsystems within a wind turbine generator, the probability of different types of abnormal state modes occurring also varies, resulting in different amounts of data monitoring data at different times during each abnormal state mode.

[0003] Meanwhile, wind turbine generators are complex electromechanical-hydraulic coupled systems, with continuous exchange of matter, information, and energy between their subsystems. The combination of these subsystems under different state modes results in numerous abnormal state modes for wind turbine generators. Consequently, some rare wind turbine generator operating state modes may emerge, with limited monitoring data available for these modes. Therefore, a method for aligning wind turbine generator monitoring data is urgently needed to "align" the limited monitoring data with the data from normal operating state modes, enabling better analysis of the characteristics of these operating state modes based on the aligned data. Summary of the Invention

[0004] This application provides a method, device, and storage medium for aligning monitoring data of wind turbine generator sets to solve the technical problems mentioned above.

[0005] The first aspect of this application proposes a method for aligning monitoring data of wind turbine generator sets, including:

[0006] Construct a monitoring dataset for at least one operating state of a wind turbine;

[0007] Identify the target operating state to be analyzed and the monitoring variables that need to be aligned within the target operating state;

[0008] Determine whether the data of the monitoring variables that need to be aligned in the monitoring dataset meet the alignment conditions;

[0009] In response to the data in the monitoring dataset satisfying the alignment conditions, aligned data for the monitoring variables that need to be aligned is generated.

[0010] A second aspect of this application provides a wind turbine generator monitoring data alignment device, comprising:

[0011] A building module for constructing monitoring datasets for at least one operating state of a wind turbine;

[0012] The first determining module is used to determine the target operating state to be analyzed and the monitoring variables that need to be aligned in the target operating state;

[0013] The second determining module is used to determine whether the data of the monitoring variables that need to be aligned in the monitoring dataset meet the alignment conditions.

[0014] The generation module is used to generate aligned data for the monitoring variables that need to be aligned in response to the data in the monitoring dataset meeting the alignment conditions.

[0015] The computer storage medium proposed in the third aspect of this application stores computer-executable instructions; after being executed by a processor, the computer-executable instructions can implement the method described in the first aspect above.

[0016] The computer device proposed in the fourth aspect of this application includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it is able to implement the method described in the first aspect above.

[0017] The technical solutions provided by the embodiments of this application bring at least the following beneficial effects:

[0018] The wind turbine generator monitoring data alignment method, device, and storage medium proposed in this application construct a monitoring dataset for at least one operating state of the wind turbine generator, determine the target operating state to be analyzed and the monitoring variables that need to be aligned in the target operating state, determine whether the data of the monitoring variables that need to be aligned in the monitoring dataset meets the alignment conditions, and generate aligned data for the monitoring variables that need to be aligned in response to the data in the monitoring dataset meeting the alignment conditions. Therefore, in this application, when the data in the monitoring dataset meets the alignment conditions, aligned data for the monitoring variables that need to be aligned is generated, thereby improving the effectiveness of data alignment. Simultaneously, this application proposes a wind turbine generator monitoring data alignment method that generates "new" data that can reflect the operating state mode of the wind turbine generator, thereby expanding the sample size of the state monitoring data, realizing the alignment of monitoring data, and providing a data foundation for subsequent precise control of the wind turbine generator state.

[0019] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0020] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0021] Figure 1 This is a flowchart illustrating a method for aligning monitoring data of a wind turbine generator set according to an embodiment of this application.

[0022] Figure 2 This is a schematic diagram of the structure of a wind turbine generator monitoring data alignment device according to an embodiment of this application. Detailed Implementation

[0023] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0024] The following describes a method and apparatus for aligning monitoring data of a wind turbine generator set according to embodiments of this application, with reference to the accompanying drawings.

[0025] Example 1

[0026] Figure 1 This is a flowchart illustrating a method for aligning monitoring data of a wind turbine generator set according to an embodiment of this application. Figure 1 As shown, it may include:

[0027] Step 101: Construct a monitoring dataset for at least one operating state of the wind turbine.

[0028] In this embodiment of the disclosure, monitoring data of the wind turbine generator set can be collected based on the hardware and software deployment of the wind turbine generator set, and a monitoring data set that can reflect the operating status of the wind turbine generator set can be constructed.

[0029] Specifically, in this embodiment of the disclosure, monitoring data of the wind turbine generator can be collected through at least one of the following systems: Supervisory Control and Data Acquisition (SCADA), Condition Monitoring System (CMS), and Structural Health Monitoring System (SHM).

[0030] In this embodiment of the disclosure, the data acquisition and monitoring system monitors the wind turbine generator set at low resolution and provides data and alarm channels for the generator set; the status monitoring system monitors the key subsystems (important components) of the wind turbine generator set at high resolution and performs fault diagnosis or pre-diagnosis; and the structural health monitoring system monitors the key structural (tower, etc.) components of the wind turbine generator set at lower resolution.

[0031] Furthermore, in this embodiment of the disclosure, the above-mentioned different types of monitoring data are organized according to different operating states such as collection time, wind turbine generator number, and current operating status of the wind turbine generator, to construct a wind turbine generator status monitoring dataset.

[0032] Step 102: Determine the target operating state to be analyzed and the monitoring variables that need to be aligned in the target operating state.

[0033] In this embodiment of the disclosure, operating states that occur less than a first threshold number of times can be identified as target operating states to be analyzed. In this embodiment of the disclosure, monitoring variables in the target states whose data volume is less than a second threshold can be identified as monitoring variables in the target states that need to be aligned.

[0034] Furthermore, in this embodiment of the disclosure, the target operating state to be analyzed and the monitoring variables to be aligned in the target operating state can also be manually set.

[0035] Furthermore, in this embodiment of the disclosure, after determining the aligned monitoring variables as described above, other variables that are coupled with the monitoring variables can be obtained through analysis. These variables can fully characterize the target's operating state.

[0036] Specifically, in this embodiment, based on the physical mechanism of wind turbine operation: a wind turbine is a complex electromechanical-hydraulic coupled system, and there is a continuous flow of material, energy and information in each key subsystem (important component). The relationship between several monitoring data corresponding to the key subsystem is regarded as a coupling relationship. The K monitoring variables with the strongest coupling relationship in the final target operating state are determined by the correlation mining algorithm between monitoring variables, and the monitoring variables that need to be corresponding are determined as the monitoring variables that need to be aligned.

[0037] In this embodiment of the disclosure, the aforementioned association mining algorithm may include clustering algorithms (e.g., density peak clustering algorithm, or DBSCAN clustering algorithm, or K-means clustering algorithm), association rule mining algorithms (e.g., Apriori algorithm, or FP-Growth algorithm), and similarity measurement methods (e.g., Manhattan distance, or Euclidean distance, or cosine similarity).

[0038] Step 103: Determine whether the data of the monitoring variables that need to be aligned in the monitoring dataset meet the alignment conditions.

[0039] In this embodiment of the disclosure, the method for determining whether the data of the monitoring variables that need to be aligned in the monitoring dataset meet the alignment conditions may include the following steps:

[0040] Step a: Determine whether the data of the monitoring variables that need to be aligned have alignment features in the monitoring dataset.

[0041] In this embodiment of the disclosure, the alignment feature may include at least one of the following:

[0042] Chaos;

[0043] Non-stationarity;

[0044] Long-range correlation.

[0045] Furthermore, in this embodiment of the disclosure, the method for determining whether the data of the monitoring variables that need to be aligned in the monitoring dataset is chaotic may include: determining the chaos of the data of the monitoring variables that need to be aligned in the monitoring dataset based on the maximum Lyapunov exponent.

[0046] In this embodiment of the disclosure, the maximum Lyapunov exponent λ based on the Wolf method is... max The calculation method may include the following steps:

[0047] 1) Choose an appropriate method to calculate the time series {x} of length N. n Given the delay time τ and embedding dimension m, the reconstructed phase space is obtained as follows:

[0048] X = [x n ,x n+τ ,...,x n+(m-1)τ ]∈R M n=1,2,...,N-(m-1)τ

[0049] 2) Set the initial point x1 as the reference point, and find the phase point (excluding phase point x1) that is closest to the reference point in the reconstructed phase space. The distance between two points, as endpoints.

[0050] 3) The distance between the baseline and the endpoint changes from L1 to L′1 after T1 evolution steps, resulting in the formula for calculating the exponential growth rate:

[0051]

[0052] 4) In the reconstructed phase space, except Choose one of the phase points other than the one with The nearest phase point As the new endpoint, and this endpoint should make the vector and The included angle θ1 between them is the smallest. The distance between the new endpoint and the new endpoint is denoted as

[0053] 5) Select a new evolution step size as T2, and a baseline. With endpoints They evolved into and Let the distance between the two points be L′2. The formula for calculating the exponential growth rate is:

[0054]

[0055] 6) Continue selecting new evolution step sizes until the last phase point in the phase space. Let M be the total number of evolution steps, and let λ be the average of the exponential growth rate. max The estimated value, λ max The calculation formula is:

[0056]

[0057] Furthermore, in this embodiment of the disclosure, the method for determining whether the data of the monitoring variable to be aligned in the monitoring dataset is non-stationary may include: determining the non-stationarity of the data of the monitoring variable to be aligned in the monitoring dataset based on the ADF (Argumented Dickey-Fuller test).

[0058] In this embodiment of the disclosure, the method for determining the nonstationarity of the monitoring variables to be aligned in the monitoring dataset based on the ADF test may include the following steps:

[0059] In this embodiment of the disclosure, if all elements in a time series X are randomly sampled from the same probability distribution and simultaneously satisfy the following three conditions, then according to statistical analysis theory, the time series is considered stationary. Furthermore, the time series consists of monitoring data of the same monitoring variable at different times.

[0060] ① The mean of a time series X is a constant that is independent of time;

[0061] ②The variance of the time series X is a constant independent of time;

[0062] ③ The autocovariance of a time series X is a constant that depends only on the time interval and is independent of time.

[0063] The stationarity of a time series is determined using the ADF (Advanced Derivative Function) test. The three models for the ADF test are as follows:

[0064] Model 1:

[0065] Model 2:

[0066] Model 3:

[0067] The alternative hypothesis tests H1: H < 0, indicating no unit root; and the null hypothesis H0: H = 0, indicating the existence of a unit root. The stationarity of a time series can be determined using the following two conditions:

[0068] (i) If the hypothesis test results of at least one model reject the null hypothesis, then the model is stationary;

[0069] (ii) If the hypothesis test results of all models cannot reject the null hypothesis, then the model is non-stationary.

[0070] Furthermore, in this embodiment of the disclosure, the method for determining whether the data of the monitoring variables that need to be aligned have long-range correlation in the monitoring dataset may include: determining the long-range correlation of the data of the monitoring variables that need to be aligned in the monitoring dataset based on DFA (Detrended Fluctuation Analysis).

[0071] In this embodiment of the disclosure, the method for determining the long-range correlation of the monitoring variables to be aligned in the monitoring dataset based on DFA may include the following steps:

[0072] Step 1: From the original time series {x} n The new time series {y} is calculated. n},in:

[0073]

[0074] Step 2: Decompose the new time series y(n) into lengths s. If the length N of y(n) is not an integer multiple of s, the decomposition will produce a segment with a length less than s. In order not to lose this part of the information, the decomposition is repeated from the other end of y(n), resulting in a total of 2N segments. s A segment;

[0075] Step 3: Obtain y(n) and its least squares fit p v The differences between them are as follows:

[0076] Y(i)=y(i)-p v (i)

[0077] Step 4: The variance of the v-th segment is obtained as follows:

[0078]

[0079] Step 5: Calculate the mean of the variances of all segments, and obtain the formula for calculating the detrended volatility function as follows:

[0080]

[0081] Step 6: If the original time series {x} n If a variable exhibits long-range correlation, then its detrended fluctuation function will grow exponentially, i.e.:

[0082] F(s)∝s α

[0083] Step 7: Optimize the above algorithm to obtain the correction function and the optimized detrended fluctuation function, as shown below:

[0084]

[0085]

[0086] In the above formula, s′ represents the length of each part obtained by shuffling and then dividing the original time series. Typically, length s′ ≈ Length(x) n ) / 20.

[0087] Step b: If the data of the monitoring variables that need to be aligned have alignment characteristics in the monitoring dataset, then it is determined that the data of the monitoring variables that need to be aligned in the monitoring dataset meet the alignment conditions.

[0088] In this embodiment of the disclosure, if the data of the monitoring variable that needs to be aligned have alignment characteristics in the monitoring dataset, it indicates that the alignment effect of the data of the monitoring variable that needs to be aligned in the monitoring dataset is good, and then it is determined that the data of the corresponding monitoring variable in the monitoring dataset meets the alignment condition.

[0089] Step 104: In response to the data in the monitoring dataset meeting the alignment conditions, generate aligned data for the monitoring variables that need to be aligned.

[0090] In embodiments of this disclosure, the method for generating aligned data for monitoring variables that need to be aligned may include the following steps:

[0091] Step 1041: Based on the reconstruction power, determine the width of the time window and the number of sampling points.

[0092] Among them, in the embodiments of the present disclosure, the above reconstruction success rate is as follows: A number of data are randomly selected from a time series to reconstruct the system equation. If the reconstructed system equation meets the prerequisite conditions (if all the reconstructed time series data are within the specified upper and lower limits), it is considered a successful reconstruction. The percentage of the number of successful reconstructions in several random data samplings to the total number of times is the reconstruction success rate. The reconstruction success rate reflects the matching degree of the monitored variable to the dynamic relationship model and can also be used to measure the reasonable degree of the current wind turbine generator set segmented in time and space.

[0093] Moreover, in the embodiments of the present disclosure, the method for determining the width of the time window and the number of sampling points may include: Select the range of the width D of the time window as [1h / sampling interval time, 24h / sampling interval time], select the range of the number of sampling points M as [2, 4h / sampling interval time], calculate the reconstruction success rates under different combinations of the width D and the number of sampling points M, and select the values of D and M in the combination with the maximum success rate.

[0094] For example, in the embodiments of the present disclosure, for the monitoring data of the wind turbine generator set with a recording point every 10 minutes, select the range of the width D of the time window as [6, 144], and select the range of the number of sampling points M as [2, 24]. Moreover, both the above time window width D and the number of sampling points M are taken as integers, and the number of combinations of success rates to be calculated is maintained at 3197. The calculation formula is as follows:

[0095] (144 - 6 + 1) × (24 - 2 + 1) = 3197.

[0096] Step 1042: Based on the number of sampling points and the width of the time window, perform multivariate dynamic relationship modeling on the target operating state by a method based on the compressed sensing theory to obtain the target system equation of the target operating state.

[0097] Among them, in the embodiments of the present disclosure, the method for performing multivariate dynamic relationship modeling on the target operating state by a method based on the compressed sensing theory may include:

[0098] For a signal x ∈ R N , the compressed sensing problem can be described as, on the basis of knowing a certain measurement matrix Φ ∈ R M×N (M << N) and the linear measurement value y ∈ R of the signal x under this matrix M , solve the system of equations:

[0099] y = Φx

[0100] The process of obtaining the original signal x is described above. Since the dimension of y is much lower than the dimension of x, the above formula is an underdetermined system of equations with infinitely many solutions, making it impossible to reconstruct the original signal x. However, if the original signal x is sparse, and y and Φ in the system of equations satisfy certain conditions, it can be proven that the reconstruction of the original signal x can be achieved by solving a problem that minimizes the L0 norm:

[0101]

[0102] In the above formula: is the reconstructed value of the original signal x; ||x||0 is the L0 norm of vector x, that is, the number of non-zero elements in vector x.

[0103] Furthermore, in this embodiment, compressed sensing theory states that to achieve accurate reconstruction of a K-sparse signal x, the number of measurements M (i.e., the dimension of y) and the matrix Φ must satisfy M = O[K·lg(N)] and RIP (Restricted Isometry Property) conditions, respectively. However, since most natural signals in the time domain are not sparsity, the reconstruction of natural signals cannot simply follow the above process. Instead, a sparse representation of the natural signal x can be achieved first through a transformation Ψ based on signal sparse representation theory.

[0104] x=Ψα

[0105] In the above formula, α represents the sparse representation of the natural signal x in the Ψ transform domain.

[0106] And according to the measurement formula y=Φx, we have:

[0107] y=Φx=ΦΨα=Aα

[0108] In the above formula: A is the sensing matrix, A = ΦΨ; y is the measurement value of the sparse signal α with respect to the sensing matrix A.

[0109] Furthermore, in this embodiment of the disclosure, if the perception matrix A satisfies the RIP condition, the sparse signal α can be reconstructed by solving the problem of minimizing the L0 norm as shown in the following expression:

[0110]

[0111] Among them, the above α is the reconstructed value of the sparse representation of the original signal x.

[0112] In this embodiment of the disclosure, a mathematical model of a nonlinear system equation is constructed in the following form, wherein it is assumed that the system contains only three monitoring variables s1, s2, and s3, and the lowest power of the monitoring variables is set to 0 and the highest power to 2. The expression of the system equation is as follows:

[0113]

[0114] In the above formula: s1, s2, and s3 represent the signals generated by each monitoring point in the wind turbine generator system, s1 * s2 * s3 * The calculated monitoring data values ​​obtained from the system equations are represented by s′1, s′2, and s′3, which are the intensity values ​​(differentials) of the signal changes at the monitored variables s1, s2, and s3, respectively.

[0115] Specifically, in this embodiment of the disclosure, the target's operating state is modeled using the aforementioned nonlinear system equations and based on compressed sensing theory, as follows:

[0116] Based on the characteristics of wind turbine generator sets, the multivariate dynamic relationship model is set in the following form:

[0117] s * =F(s)

[0118] In the above formula, s represents a set of monitoring variables in a complex electromechanical system, s∈R M F is a function reflecting the dynamic relationship between the monitored variables; s * The monitoring data values ​​of each monitoring point calculated by the model.

[0119] In this embodiment of the disclosure, the lowest power of each variable is set to 0, and the highest power is n. The k-th component s of s is... k The equation for (k = 1, 2, ..., m) can be expanded into the following form:

[0120]

[0121] In the above formula, 'a' is the coefficient vector of the equation, which is sparse. Since F(s) k The expansion of ) does not contain s k Therefore, the coefficient vector a contains only (1+n) elements. m-1 +m-1 elements.

[0122] Furthermore, in this embodiment of the disclosure, the monitoring variable s is not adjusted at any time t. k Instead of conducting measurements, a number of monitoring data points are randomly selected within a certain time interval for reconstructing the system equations. This reduces the impact of noise in the monitoring data of the wind turbine generator, thereby effectively reducing the influence of some noise-sensitive data on the reconstructed system equations. From the time interval [t]... i ,t j In each step, w monitoring variables s are randomly selected. kThe monitoring data is used as the measured value, and the following measurement vector is obtained:

[0123]

[0124] Among them, in the embodiments of the present disclosure, different from the mathematical model for reconstructing the non-linear system equation, in the multi-variable dynamic relationship modeling of a wind turbine generator set, the reconstructed system equation may not necessarily reflect the dynamic relationship between the monitoring variables due to reasons such as the influence of noise on the monitoring data, and it needs to be tested to prove its ability to reflect the dynamic relationship between the monitoring variables. Also, in the embodiments of the present disclosure, the testing method is for all the monitoring data s i , t j (i, j ∈ N, and i < j) within the time interval t ∈ [t k (t), there is:

[0125]

[0126] Among them, in the above formula: σ is the threshold of the reconstruction error of the monitoring data s k (t). When σ is small enough, it can be proved that the system equation satisfying the above formula can reflect the dynamic relationship between the monitoring variables.

[0127] Step 1043: Based on the target system equation, construct a training dataset for the adversarial network.

[0128] Among them, in the embodiments of the present disclosure, the method for constructing a training dataset for the adversarial network based on the target system equation may include the following steps:

[0129] Step a: According to the width of the time window and the target system equation, obtain the analytical data of the monitoring variables corresponding to the next moment;

[0130] Step b: If the analytical data of the monitoring variables to be aligned at the next moment is greater than the preset deviation threshold from the real data, store the analytical data at the next moment into the training dataset of the adversarial network; otherwise, slide to the next moment according to the time window and repeat the above process.

[0131] Among them, in the embodiments of the present disclosure, the above preset deviation threshold can be set as needed, and different preset deviation thresholds correspond to different accuracy requirements. Also, according to different accuracy requirements, the sliding step size can be determined and adjusted. In the embodiments of the present disclosure, STEP is set to 1, that is, at an interval of 1 moment.

[0132] Step 1044: Based on the training dataset, perform adversarial training on the Im-TimeGAN model to obtain a trained Im-TimeGAN model.

[0133] In this embodiment of the disclosure, the Im-TimeGAN model may include: an embedding network, a recovery network, a generation network, and a discriminator.

[0134] Step 1045: Use the trained Im-TimeGAN model to generate aligned data for the monitoring variables that need to be aligned.

[0135] In this embodiment of the disclosure, the method for generating aligned data for the monitoring variables that need to be aligned using a trained Im-TimeGAN model may include:

[0136] (1) Embedded networks via e s :S→H s e x :H s ×H x ×X→H x Map the data from the original feature space to a higher-dimensional vector space.

[0137] h s =e s (s)

[0138] h t =e x (h s ,h t-1 ,x t )

[0139] Among them, e s It is a static feature embedding network, e x It is an embedding network for dynamic features, where S and X represent the vector spaces of static and dynamic features, respectively, and H... s H x Let S and X represent the potential vector spaces corresponding to S and X, respectively.

[0140] (2) Restore network access via r s :H s →S,r x :H x →X restores the higher-dimensional vector space to the original static and dynamic features.

[0141]

[0142]

[0143] Where, r s It is a static feature recovery network, r x It is a recovery network for dynamic features. The embedding network and the recovery network can be parameterized using any chosen method; the only stipulation is that they are autoregressive and follow a causal order.

[0144] (3) Generate network via g s :Z S →ss,g x :H S ×H X ×Z X →xx t The synthesized output is generated directly in the feature space.

[0145]

[0146]

[0147] Among them, g s It is a static feature generation network, g x It is a dynamic feature generation network. Im-TimeGAN directly generates the synthetic output in the feature space. During discrimination, the data generated in the feature space needs to be embedded into the latent vector space through an embedding function. s and z x It is a known distribution subspace.

[0148] (4) The discriminator passes through d S d X The output layer classification function is used for discrimination.

[0149]

[0150]

[0151] in, and These represent the forward and reverse hidden state sequences, respectively. and It is a recursive function. As an invertible mapping between the feature space and the latent space, the embedding and recovery functions should be able to extract values ​​from s, X. 1:T The latent high-dimensional embedding space vector representation h S ,h 1:T Accurately recovering the representation to the original low-dimensional feature space Therefore, the reconstruction loss function is expressed as:

[0152]

[0153] Furthermore, in this embodiment of the disclosure, the generator network primarily minimizes two types of losses. Specifically, in pure open-loop mode, the autoregressive generator receives synthetic embeddings. (i.e., its own previous output) to generate the next composite vector and h tThe difference between them; maximizing the accuracy of the discriminator (discrimination) It is synthesized, h S ,h 1:T (It is the original) and the probability of the minimized generator being recognized ( The probability of being identified as synthetic is given by the loss function, which is:

[0154]

[0155] Furthermore, in this embodiment of the disclosure, the distribution of generated data is reduced through closed-loop and alternating training. Distribution of the original data (p(h)) t |h S ,h 1:t-1 The difference between h and h is used to address the issue that the binary adversarial feedback of the discriminator may not be sufficient to motivate the generator to capture the stepwise conditional distribution in the data. t-1 It is the actual data x t-1 The loss function is calculated through the embedded network and is expressed as:

[0156]

[0157] In this embodiment of the disclosure, if the trained Im-TimeGAN model generates data corresponding to new variables other than the monitoring variables that need to be aligned, then the process of generating aligned data in steps 1042 to 1045 is repeated using the data of the new variables to obtain the aligned data of the new variables, until the trained Im-TimeGAN model no longer generates data of the new variables, and the generation of aligned data ends.

[0158] Furthermore, in this embodiment of the disclosure, the method for determining whether there is a new variable may include: determining whether the relative error between the actual value of the monitoring data at the next moment and the analytical value of the monitoring data obtained through the target system equation at the previous moment is less than a third threshold; if it is greater than the third threshold for Y consecutive times, it indicates that a new variable has appeared, the new variable needs to be stored, and an alignment process is carried out on the new variable.

[0159] It should be noted that, in this embodiment, the trained Im-TimeGAN model first embeds monitoring data points that satisfy the system model into a high-dimensional vector space. Then, it extracts data points from a known distribution and generates them into a feature space through a generative network. This feature space is then mapped to the embedding space, and a discriminator determines the type of these two types of data in the embedding space. By employing the generative adversarial approach, "new" data that reflects the target operating state pattern of the wind turbine generator is generated, thereby expanding the sample size of the operating state monitoring data and solving the problem of aligning typical monitoring data of wind turbine generators under multiple influencing factors.

[0160] Furthermore, in this embodiment of the disclosure, the above method further includes: in response to the data in the monitoring dataset not meeting the alignment conditions, processing the data in the monitoring dataset, and not generating aligned data for the monitoring variables that need to be aligned.

[0161] In summary, the wind turbine generator monitoring data alignment method proposed in this application constructs a monitoring dataset for at least one operating state of the wind turbine generator, determines the target operating state to be analyzed and the monitoring variables that need to be aligned within the target operating state, and determines whether the data of the monitoring variables to be aligned in the monitoring dataset meets the alignment conditions. In response to the data in the monitoring dataset meeting the alignment conditions, aligned data for the monitoring variables to be aligned is generated. Therefore, in this application, when the data in the monitoring dataset meets the alignment conditions, aligned data for the monitoring variables to be aligned is generated, thereby improving the effectiveness of data alignment. Simultaneously, this application proposes a wind turbine generator monitoring data alignment method that generates "new" data that reflects the operating state pattern of the wind turbine generator, thereby expanding the sample size of the state monitoring data, achieving monitoring data alignment, and providing a data foundation for subsequent precise control of the wind turbine generator state.

[0162] Example 2

[0163] Figure 2 This is a schematic diagram of the structure of a wind turbine generator monitoring data alignment device according to an embodiment of this application, as shown below. Figure 2 As shown, the device may include:

[0164] Module 201 is used to construct a monitoring dataset for at least one operating state of a wind turbine.

[0165] The first determining module 202 is used to determine the target operating state to be analyzed and the monitoring variables that need to be aligned in the target operating state;

[0166] The second determining module 203 is used to determine whether the data of the monitoring variables that need to be aligned in the monitoring dataset meet the alignment conditions.

[0167] The generation module 204 is used to generate aligned data for the monitoring variables that need to be aligned in response to the data in the monitoring dataset meeting the alignment conditions.

[0168] In summary, the wind turbine generator monitoring data alignment device proposed in this application constructs a monitoring dataset for at least one operating state of the wind turbine generator, determines the target operating state to be analyzed and the monitoring variables that need to be aligned within the target operating state, and determines whether the data of the monitoring variables to be aligned in the monitoring dataset meets the alignment conditions. In response to the data in the monitoring dataset meeting the alignment conditions, aligned data for the monitoring variables to be aligned is generated. Therefore, in this application, when the data in the monitoring dataset meets the alignment conditions, aligned data for the monitoring variables to be aligned is generated, thereby improving the effectiveness of data alignment. Simultaneously, this application proposes a wind turbine generator monitoring data alignment method, generating "new" data that reflects the operating state mode of the wind turbine generator, thereby expanding the sample size of the state monitoring data, achieving monitoring data alignment, and providing a data foundation for subsequent precise control of the wind turbine generator state.

[0169] To implement the above embodiments, this disclosure also proposes a computer storage medium.

[0170] The computer storage medium provided in this embodiment stores an executable program; after the executable program is executed by a processor, it can achieve the following: Figure 1 The method shown.

[0171] To implement the above embodiments, this disclosure also proposes a computer device.

[0172] The computer device provided in this disclosure includes a memory, a processor, and a computer program stored in the memory and executable on the processor; when the processor executes the program, it can achieve the following: Figure 1 Any of the methods shown.

[0173] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0174] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0175] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for aligning monitoring data of a wind turbine generator set, characterized in that, The method includes: Construct a monitoring dataset for at least one operating state of a wind turbine; Identify the target operating state to be analyzed and the monitoring variables that need to be aligned within the target operating state; Determine whether the data of the monitoring variables that need to be aligned in the monitoring dataset meet the alignment conditions; In response to the data in the monitoring dataset satisfying the alignment conditions, aligned data for the monitoring variables that need to be aligned is generated; The step of determining whether the data of the monitoring variable requiring alignment in the monitoring dataset meets the alignment conditions includes: Determine whether the data of the monitoring variables that need alignment exist in the monitoring dataset, wherein the alignment features include at least one of the following: Chaos; Non-stationarity; Long-range correlation; If the data of the monitoring variable that needs to be aligned in the monitoring dataset have alignment features, then it is determined that the data of the monitoring variable that needs to be aligned in the monitoring dataset meet the alignment conditions. The process of generating alignment data for the monitoring variables that need alignment includes: Based on the reconstruction success rate, determine the width of the time window and the number of sampling points; Based on the number of sampling points and the width of the time window, a multivariate dynamic relationship model is performed on the target operating state using a method based on compressed sensing theory, thereby obtaining the target system equation for the target operating state. Based on the target system equations, a training dataset for the adversarial network is constructed; Based on the training dataset, the Im-TimeGAN model is subjected to adversarial training to obtain a trained Im-TimeGAN model. The trained Im-TimeGAN model is used to generate aligned data for the monitoring variables that need to be aligned.

2. The method according to claim 1, characterized in that, The training dataset for constructing the adversarial network based on the target system equation includes: Based on the width of the time window and the target system equation, the analytical data of the monitoring variables that need to be aligned at the next moment are obtained; If the parsed data of the monitoring variable that needs to be aligned at the next time step is greater than the actual data by a preset deviation threshold, then the parsed data at the next time step is stored in the training dataset of the adversarial network; otherwise, the process is repeated by sliding to the next time step according to the time window.

3. The method according to claim 1, characterized in that, The Im-TimeGAN model includes: an embedding network, a recovery network, a generative network, and a discriminator.

4. The method according to claim 3, characterized in that, The step of generating aligned data for the monitoring variables that need to be aligned using the trained Im-TimeGAN model includes: Embedded networks through , Map the data from the original feature space to a higher-dimensional vector space; , ,in It is a static feature embedding network. It is an embedding network for dynamic features, where S and X represent the vector spaces of static and dynamic features, respectively. , Let S and X represent the latent vector spaces corresponding to S and X, respectively. Restore network access , This restores the higher-dimensional vector space to the original static and dynamic features. It is a static feature recovery network. It is a dynamic feature recovery network; Generative networks through , The synthesized output is generated directly in the feature space. It is a static feature generation network. It is a dynamic feature generation network. and It is a known distribution subspace; Discriminator through The output layer classification function is used for discrimination.

5. The method according to claim 4, characterized in that, The method further includes: If the trained Im-TimeGAN model generates new data aligned to variables other than the monitoring variables that need to be aligned, the process of generating aligned data is repeated using the data of the new variables to obtain the aligned data of the new variables, until the trained Im-TimeGAN model no longer generates data of the new variables, and the generation of aligned data ends.

6. The method according to claim 1, characterized in that, The method further includes: In response to the data in the monitoring dataset not meeting the alignment conditions, the data in the monitoring dataset is processed, and alignment data for the monitoring variables that need to be aligned is not generated.

7. A wind turbine generator monitoring data alignment device, characterized in that, The device is used to implement the wind turbine generator monitoring data alignment method as described in claim 1, and the device includes: A building module for constructing monitoring datasets for at least one operating state of a wind turbine; The first determining module is used to determine the target operating state to be analyzed and the monitoring variables that need to be aligned in the target operating state; The second determining module is used to determine whether the data of the monitoring variables that need to be aligned in the monitoring dataset meet the alignment conditions. The generation module is used to generate aligned data for the monitoring variables that need to be aligned in response to the data in the monitoring dataset meeting the alignment conditions.

8. A computer storage medium, wherein, The computer storage medium stores computer-executable instructions; when executed by a processor, the computer-executable instructions can implement the method described in any one of claims 1-6.

Images