Wind power equipment service quality unstable operation state identification method and system

Abstract

The invention provides a wind power equipment service quality unstable operation state identification method and system, and the method comprises the following sub-steps: obtaining a historical data set of the service of wind power equipment, the historical data set comprising time sequence data of the operation of the wind power equipment in a steady state and an unstable state; training the data in the historical data set based on the generative adversarial network CAN and MMD evaluation indexes, and determining the minimum sample size during model training; according to the minimum sample size and the historical data set, performing data amplification training on the unsteady data in the historical data set based on a WGAN model to generate unsteady similar sample data; according to the method, data amplification can be carried out on the unstable operation state sample data, quantitative identification of the unstable operation degree of the wind power equipment and positioning of reason variables are realized in combination with the LSTM model and the residual spectrum, and adaptive adjustment can be carried out on the state identification threshold range, so that the accuracy of service quality evaluation of the wind power equipment is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power equipment operating state detection, and in particular to a method and system for identifying an unstable operating state of a wind power equipment service quality. Background Art

[0002] Wind power equipment is a complex electromechanical system with highly coupled functions and structures, coordinated operation of multiple components, and powerful operation control capabilities. Any abnormality in any component may affect the operating status of other components through interactions between components, thus causing system failure and even huge economic losses and safety issues.

[0003] A method, device, equipment and storage medium for detecting the operating status of a wind turbine set with publication number CN117633439A discloses the method comprising: obtaining real-time data of key parts in the wind turbine set; performing smoothing preprocessing on the real-time data to obtain preprocessed data; reconstructing the preprocessed data through a preset NLC-TCN model to obtain reconstructed data, wherein the preset NLC-TCN model comprises a plurality of TCNConvNeXt modules and a Linear residual module, wherein the Linear residual module is used to preliminarily extract time information features and spatial information features of any time length in the preprocessed data, and output the time information features and spatial information features of any time length to each TCNConvNeXt module, wherein each TCNConvNeXt module is used to further sequentially extract the time information features and spatial information features of a target time length in any time length; determining an operating status score through the reconstructed data and the real-time data, and determining whether the operating status of the wind turbine set is abnormal according to the operating status score and the real-time abnormality threshold.

[0004] The existing wind power equipment operating status monitoring methods have the following problems: insufficient sample size of unstable operating states, difficulty in measuring the degree of operating state instability, and poor model adaptability to time-varying system states, which makes it difficult to evaluate the quality of wind power equipment and reduces the accuracy of wind power equipment operating status evaluation. Summary of the invention

[0005] In view of this, the present invention proposes a method and system for identifying the unstable operation state of the service quality of wind power equipment, which can perform data amplification on the unstable operation state sample data, and combine the WGAN-LSTM model and the residual spectrum to realize the quantitative identification of the unstable operation degree of wind power equipment and the location of the cause variables, and can adaptively adjust the state identification threshold range, thereby improving the accuracy of the service quality assessment of wind power equipment.

[0006] The technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides a method for identifying the unstable operation state of the service quality of a wind power device, the method comprising the following sub-steps:

[0007] S1, obtaining a historical data set of wind power equipment in service, wherein the historical data set includes time series data of the wind power equipment operating in a steady state and an unsteady state;

[0008] S2, training the data in the historical data set based on the CAN and MMD evaluation indicators of the generative adversarial network to determine the minimum sample size for model training;

[0009] S3, based on the minimum sample size and historical data set, perform data augmentation training on the non-steady-state data in the historical data set based on the WGAN model to generate non-steady-state similar sample data;

[0010] S4, according to the time series data running under non-steady state in the historical data set, the generated non-steady-state similar sample data is reordered based on the Reshuffling time series learning to obtain non-steady-state similar time series data;

[0011] S5, construct a wind power equipment state identification model based on LSTM long short-term memory network, merge the non-steady-state similar time series data with the historical data set as training data and input it into the wind power equipment state identification model for training, and output the wind power equipment operation state measurement index;

[0012] S6, preset a steady-state range threshold, and determine the operating state of the wind power equipment according to the wind power equipment operating state measurement index.

[0013] On the basis of the above technical solution, preferably, step S2 includes the following sub-steps:

[0014] S21, constructing a generative adversarial network (CAN) model, and initializing the parameters of the generator and discriminator of the generative adversarial network (CAN) model;

[0015] S22, presetting a plurality of learning rate parameters within the learning rate range, and presetting a plurality of groups of different alternation times;

[0016] S23, training using the historical data set of the wind power equipment in service according to each learning rate parameter, and performing an iterative training process for each group of alternation times to obtain a generated data set;

[0017] S24, after each iterative training, the MMD indicator is used to calculate the distance between the distribution of the generated sample data set and the historical data set, and the sample size at which the MMD indicator value begins to stabilize and approaches the minimum is selected as the minimum sample size for model training. The calculation expression is:

[0018]

[0019] In the formula, φ(·) represents the mapping function of the original feature space X to the Hilbert space H, X(s) represents the distribution of historical data set samples, X (t) represents the distribution of the generated dataset samples, represents samples drawn from historical data sets, represents the samples drawn from the generated dataset; n s Represents the sample size of the historical data set, n t represents the sample size of the generated dataset, ||·|| H represents the norm in Hilbert space.

[0020] Based on the above technical solution, preferably, step S3 includes the following sub-steps:

[0021] S31, preprocessing the data in the historical data set of wind power equipment service based on the data normalization method, the formula for data normalization is:

[0022]

[0023] In the formula, x ij ′ is the historical sample data of the i-th variable after normalization at the j-th moment, x ij is the historical sample data of the ith variable at the jth moment, max(X j ),min(X j ) are the maximum and minimum values ​​of the jth variable in all historical sample data respectively;

[0024] S32, constructing a WGAN model, and randomly initializing the parameters of the generator G and the discriminator D of the WGAN model;

[0025] S33, define the loss function and optimizer, and set the learning rate parameters and decay rate. The loss function expression is:

[0026]

[0027]

[0028] Where, L D represents the loss function of the discriminator, L G represents the loss function of the generator, D(x) represents the score of the discriminator on the historical sample data x, and D(G(z)) represents the score of the discriminator on the generated sample data G(z);

[0029] S34, input the normalized time series data of wind power equipment operating in non-steady state in the historical data set into the WGAN model for iterative training, adjust the learning rate parameters and decay rate of the WGAN model based on the optimizer until the WGAN model reaches a convergence state, and the WGAN model outputs the generated non-steady-state similar sample data.

[0030] On the basis of the above technical solution, preferably, step S4 includes the following sub-steps:

[0031] S41, constructing a first time series and a second time series, wherein the length of the first time series X is T, wherein each monitoring point of the first time series X has q variables, and the first time series X corresponds to the time series data running under non-steady state in the historical data set, and the first time series expression is:

[0032]

[0033] In the formula, x Tq is the monitoring value of the monitoring point at the Tth moment of the qth variable of the first time series X;

[0034] The length of the second time series Y is T, where each monitoring of the second time series Y has q variables, and the expression of the first time series is:

[0035]

[0036] In the formula, y Tq is the monitoring value of the monitoring point at the Tth moment of the qth variable of the second time series Y;

[0037] S42, respectively obtaining a sequential sequence of the first time series and the second time series;

[0038] The order sequence calculation expression of the first time series is:

[0039]

[0040] In the formula, For X j The observed value at the i-th moment, For X j 'The i′th moment observation value;

[0041] The order sequence calculation expression of the second time series is:

[0042]

[0043] In the formula, Y j The observed value at the kth moment, Y j 'The k'th moment observation value;

[0044] S43, according to X j 'The index i of each observation' for the first time series X jThe corresponding observation value i is marked to obtain the temporal trend characteristics of the first time series, which is expressed as:

[0045]

[0046] S44, the second time series learns the temporal trend characteristics of the first time series, reorders the generated non-steady-state similar sample data, and obtains non-steady-state similar time series data, which is expressed as:

[0047]

[0048] In the formula, Y j The value of "at the kth moment", is the value of the temporal trend feature of the first time series at the k'th moment.

[0049] Based on the above technical solution, preferably, the steady-state range threshold in step S6 is:

[0050]

[0051] In the formula, is the maximum value of the steady-state range threshold at the i-th moment, is the minimum value of the steady-state range threshold at the i-th moment, μ j is the average value of the j-th variable metric, σ j is the variance of the j-th variable metric;

[0052] When the wind power equipment operating state measurement indicators output by each variable are all within the corresponding steady-state range threshold, the wind power equipment is determined to be operating in a stable state, otherwise the wind power equipment is determined to be operating in an unstable state.

[0053] On the basis of the above technical solution, preferably, step S7 is further included, locating the unstable operation state variable based on the residual spectrum to determine the cause of the unstable operation state of the wind power equipment, including the following sub-steps:

[0054] S71, obtaining real-time data of the wind power equipment based on real-time data sliding window sampling, performing variable similarity distance matrix calculation, and obtaining a real-time distance matrix;

[0055] S72, calculating a variable similarity distance matrix based on the time series data of the wind power equipment operating in a steady state to obtain an original distance matrix;

[0056] S73, based on the trained WGAN model, inputting the time series data of the wind power equipment operating in a steady state to obtain steady-state generated sample data, performing variable similarity distance matrix calculation based on the steady-state generated sample data, and obtaining a generated distance matrix;

[0057] S74, performing difference calculation on corresponding elements according to the original distance matrix and the standard distance matrix to obtain a standard residual matrix;

[0058] S75, converting the standard residual matrix and the real-time distance matrix into a standard residual spectrum and a real-time residual spectrum respectively, and identifying the key variables causing the unstable operation state of the wind power equipment according to the difference between the two, and determining the location of the cause of the unstable operation state of the wind power equipment.

[0059] On the basis of the above technical solution, preferably, the variable similarity distance matrix calculation expression is:

[0060]

[0061] In the formula, x t1 and x t2 are two time step variables, and n is the total number of samples;

[0062] The distance matrix expression is:

[0063]

[0064] On the basis of the above technical solution, preferably, it further includes step S8, adaptively updating the steady-state range threshold based on unsupervised clustering and incremental learning, which includes the following sub-steps:

[0065] S81, normalizing the real-time monitoring data of the wind power equipment and inputting it into the wind power equipment state identification model in time sequence to obtain output measurement indicators corresponding to each data;

[0066] S82, obtaining a state metric index threshold library, in which the steady-state range threshold corresponding to each variable and each moment is stored, and comparing each data output metric index with the corresponding steady-state range threshold;

[0067] S83, if the number of output metrics corresponding to a state category in the real-time monitoring data within a time window that is within the corresponding steady-state range threshold is greater than half, then the steady-state range threshold in the state metric threshold library is updated, and the expression is:

[0068]

[0069] Otherwise, the real-time monitoring data at that moment will be marked as an outlier and stored in the outlier database;

[0070] S83, preset a response threshold, use a counter to count the number of outliers in the outlier library, when the number of outliers in the outlier library reaches the preset response threshold, use a clustering algorithm to cluster the outlier library to obtain outlier data sets of different category states, if the threshold is not reached, the currently detected outlier is regarded as a noise point, and the system operation status continues to be analyzed;

[0071] S84, determining whether the sample size in the outlier data set of each category state reaches the update threshold of the wind power equipment state identification model. If the sample size of a certain category reaches the update threshold, the wind power equipment state identification model is trained based on the outlier data set of the category state until convergence.

[0072] S85, based on the trained wind power equipment state identification model, calculating the steady-state threshold corresponding to the state, and adding it to the state measurement indicator threshold library;

[0073] The steady-state threshold expression corresponding to this state is:

[0074]

[0075] In a second aspect, the present invention further provides a system for identifying the unstable operation state of the service quality of wind power equipment, which is implemented by using the method for identifying the unstable operation state of the service quality of wind power equipment according to any one of the above claims, and the system comprises:

[0076] A data acquisition module, used to obtain a historical data set of wind power equipment in service, wherein the historical data set includes time series data of the wind power equipment operating in a steady state and an unsteady state;

[0077] The first generation module is used to train the data in the historical data set based on the CAN and MMD evaluation indicators of the generative adversarial network to determine the minimum sample size during model training;

[0078] The second generation module is used to perform data augmentation training on the non-steady-state data in the historical data set based on the WGAN model according to the minimum sample size and the historical data set, and generate non-steady-state similar sample data;

[0079] The sample reorganization module is used to reorder the generated non-steady-state similar sample data based on the time series data running in the non-steady state in the historical data set based on the Reshuffling time series learning to obtain non-steady-state similar time series data;

[0080] The model building module is used to build a wind power equipment state identification model based on the LSTM long short-term memory network, merge the non-steady-state similar time series data with the historical data set as training data and input it into the wind power equipment state identification model for training, and output the wind power equipment operation state measurement indicators;

[0081] The state determination module is used to preset a steady-state range threshold and determine the operating state of the wind power equipment according to the wind power equipment operating state measurement index.

[0082] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above-mentioned methods for identifying the unstable operating state of the service quality of wind power equipment when executing the computer program.

[0083] The method and system for identifying the unstable operation state of the service quality of wind power equipment of the present invention have the following beneficial effects compared with the prior art:

[0084] (1) The WGAN-LSTM model can be used to amplify the sample data of the non-steady state operation, generate more non-steady-state similar sample data, and improve the generalization ability of the model under non-steady-state conditions. The wind power equipment state identification model constructed by the LSTM long short-term memory network is trained by combining the historical data set and the amplified non-steady-state similar time series data, which significantly improves the accuracy and robustness of the model in identifying the non-steady state of wind power equipment.

[0085] (2) By combining the residual spectrum, the unstable operation degree of wind power equipment can be quantitatively identified and the cause variables can be located, thereby further improving the accuracy of wind power equipment service quality assessment;

[0086] (3) The dynamic update of the steady-state range threshold can reflect the changes in the actual operating status of the wind power equipment, thereby improving the accuracy of state identification, the adaptability and robustness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0088] Figure 1 It is a flow chart of the method for identifying the unstable operation state of the service quality of wind power equipment of the present invention;

[0089] Figure 2 It is a schematic diagram of wind power equipment state identification accuracy under various unbalanced degrees of the wind power equipment service quality unstable operation state identification method of the present invention;

[0090] Figure 3A diagram showing the training steps of WGAN-Reshuffling of the method for identifying the unstable operation state of the service quality of wind power equipment of the present invention;

[0091] Figure 4 A wind power equipment state identification model identification flow chart of the wind power equipment service quality unstable operation state identification method of the present invention;

[0092] Figure 5 This is a flow chart of locating the cause of the unstable operation state of the method for identifying the unstable operation state of the service quality of wind power equipment of the present invention. DETAILED DESCRIPTION

[0093] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0094] In order to solve the problems that the quality of wind power equipment is difficult to evaluate due to insufficient sample size of unstable operating states of wind power equipment, difficulty in measuring the degree of unstable operating state and poor adaptability of the model to time-varying system states, the present invention provides a method for identifying the unstable operating state of the service quality of wind power equipment, which comprehensively predicts and evaluates the quality risks during the operation of wind power equipment.

[0095] like Figure 2 As shown, the influence of unbalanced characteristics on the state identification of wind power equipment is quantitatively analyzed based on the sample size difference between steady-state and unsteady-state data of wind power equipment, which includes the following steps:

[0096] Select evaluation indicators of wind power equipment state identification model under unbalanced data according to task characteristics;

[0097] Based on evaluation indicators such as F1_score, G_mean, AUC and Kappa coefficient, the quantitative impact of data imbalance on the accuracy of equipment state identification is obtained.

[0098] F1_score is the harmonic average of recall and precision, and the calculation formula is:

[0099]

[0100] AUC is the area between the curve with FPR as the horizontal axis and Recall as the vertical axis, where FPR is the proportion of all samples that are actually class 2 that are incorrectly predicted;

[0101] G_mean represents the performance evaluation standard of the state recognition model that unifies the false alarm rate and recall rate. The calculation formula is:

[0102]

[0103] The Kappa coefficient is often used for consistency testing and is also suitable for measuring comprehensive classification accuracy. Its value ranges from 0 to 1. The closer it is to 1, the better the recognition learning effect. The calculation formula is:

[0104]

[0105] In the formula, p o represents the accuracy of the test set;

[0106]

[0107] In the formula, TP is the positive sample predicted by the model as the positive class, TN is the negative sample predicted by the model as the negative class, FP is the negative sample predicted by the model as the positive class, and FN is the positive sample predicted by the model as the negative class.

[0108] Taking the imbalance factor into consideration, this embodiment can explore the quantitative influence of different imbalance degrees on the equipment state identification accuracy through multiple specific indicators. The analysis shows that the original sample size of the unstable operation state of the wind power equipment is insufficient, resulting in data imbalance between the steady-state data and the non-steady-state data, which in turn affects the identification accuracy of the identification model. Based on this problem, the present invention performs data amplification on the unstable operation state sample data, and combines the WGAN-LSTM model and the residual spectrum to realize the quantitative identification of the unstable operation degree of the wind power equipment and the location of the causal variables, and can adaptively adjust the state identification threshold range, thereby improving the accuracy of the service quality assessment of the wind power equipment.

[0109] like Figure 1 As shown, a method for identifying the unstable operation state of the service quality of a wind power device according to the present invention comprises the following sub-steps:

[0110] S1, obtaining a historical data set of the service of the wind power equipment, wherein the historical data set includes time series data of the wind power equipment operating in a steady state and an unsteady state.

[0111] It is understandable that the data sources may include the wind farm's monitoring system, sensor network, maintenance records and any database related to the operating status of the wind power equipment. The collected data parameters comprehensively cover the operating conditions of the wind power equipment, including time series data under steady state operation and time series data under non-steady state operation.

[0112] S2, based on the generative adversarial network CAN and MMD evaluation indicators, train the data in the historical data set to determine the minimum sample size for model training.

[0113] Specifically, step S2 in this embodiment includes the following sub-steps:

[0114] S21, constructing a generative adversarial network (CAN) model, and initializing the parameters of the generator and discriminator of the generative adversarial network (CAN) model;

[0115] S22, presetting a plurality of learning rate parameters within the learning rate range, and presetting a plurality of groups of different alternation times;

[0116] S23, training using the historical data set of the wind power equipment in service according to each learning rate parameter, and performing an iterative training process for each group of alternation times to obtain a generated data set;

[0117] S24, after each iterative training, the MMD indicator is used to calculate the distance between the distribution of the generated sample data set and the historical data set, and the sample size at which the MMD indicator value begins to stabilize and approaches the minimum is selected as the minimum sample size for model training. The calculation expression is:

[0118]

[0119] In the formula, φ(·) represents the mapping function of the original feature space X to the Hilbert space H, X (s) represents the distribution of historical data set samples, X (t) represents the distribution of the generated dataset samples, represents samples drawn from historical data sets, represents the samples drawn from the generated dataset; n s Represents the sample size of the historical data set, n t represents the sample size of the generated dataset, ||·|| H represents the norm in Hilbert space.

[0120] It should be noted that by generating adversarial network CAN and MMD evaluation indicators, the minimum sample size required for model training can be accurately determined, which avoids data waste, ensures the effectiveness of model training, and improves the efficiency and accuracy of data processing.

[0121] like Figure 3 As shown, S3, according to the minimum sample size and the historical data set, performs data augmentation training on the non-steady-state data in the historical data set based on the WGAN model to generate non-steady-state similar sample data.

[0122] Specifically, step S3 in this embodiment includes the following sub-steps:

[0123] S31, preprocessing the data in the historical data set of wind power equipment service based on the data normalization method, the formula for data normalization is:

[0124]

[0125] In the formula, x ij ′ is the historical sample data of the i-th variable after normalization at the j-th moment, x ij is the historical sample data of the ith variable at the jth moment, max(X j ),min(X j ) are the maximum and minimum values ​​of the jth variable in all historical sample data respectively;

[0126] S32, constructing a WGAN model, and randomly initializing the parameters of the generator G and the discriminator D of the WGAN model;

[0127] S33, define the loss function and optimizer, and set the learning rate parameters and decay rate. The loss function expression is:

[0128]

[0129] Where, L D represents the loss function of the discriminator, L G represents the loss function of the generator, D(x) represents the score of the discriminator on the historical sample data x, and D(G(z)) represents the score of the discriminator on the generated sample data G(z);

[0130] S34, input the normalized time series data of wind power equipment operating in non-steady state in the historical data set into the WGAN model for iterative training, adjust the learning rate parameters and decay rate of the WGAN model based on the optimizer until the WGAN model reaches a convergence state, and the WGAN model outputs the generated non-steady-state similar sample data.

[0131] It can be understood that WGAN is a special generative adversarial network. The WGAN model consists of a generator G and a discriminator D. The goal of the generator G is to generate samples that are as close to the real data distribution as possible, while the discriminator D tries to distinguish between the generated samples and the real samples. The core of WGAN is to use the Wasserstein distance as the loss function to measure the distribution difference between the generated samples and the real samples. The loss function L of the discriminator is D Aims to maximize the Wasserstein distance estimate between real samples and generated samples, while the generator's loss function L G Then try to minimize this distance.

[0132] In this embodiment, the RMSProp optimizer is used to adjust the model parameters to minimize the loss function, and the WGAN model is optimized to generate high-quality samples similar to the real non-steady-state data.

[0133] It should be noted that the WGAN model is used to perform data augmentation training on the non-steady-state data in the historical data set, generating more non-steady-state similar sample data, effectively solving the problem of scarcity of non-steady-state data in practice, increasing the diversity and representativeness of training data, and helping to improve the generalization ability of the model under non-steady-state conditions.

[0134] like Figure 3 As shown, S4, according to the time series data running under non-steady state in the historical data set, the generated non-steady-state similar sample data is reordered based on the Reshuffling time series learning to obtain non-steady-state similar time series data.

[0135] Specifically, step S4 in this embodiment includes the following sub-steps:

[0136] S41, constructing a first time series and a second time series, wherein the length of the first time series X is T, wherein each monitoring point of the first time series X has q variables, and the first time series X corresponds to the time series data running under non-steady state in the historical data set, and the first time series expression is:

[0137]

[0138] In the formula, x Tq is the monitoring value of the monitoring point at the Tth moment of the qth variable of the first time series X;

[0139] The length of the second time series Y is T, where each monitoring of the second time series Y has q variables, and the expression of the first time series is:

[0140]

[0141] In the formula, y Tq is the monitoring value of the monitoring point at the Tth moment of the qth variable of the second time series Y;

[0142] S42, respectively obtaining a sequential sequence of the first time series and the second time series;

[0143] The order sequence calculation expression of the first time series is:

[0144]

[0145] In the formula, For X j The observed value at the i-th moment, For X j 'The i′th moment observation value;

[0146] The order sequence calculation expression of the second time series is:

[0147]

[0148] In the formula, Y j The observed value at the kth moment, Y j 'The k'th moment observation value;

[0149] S43, according to X j 'The index i of each observation' for the first time series X j The corresponding i-th observation value is marked to obtain the temporal trend characteristics of the first time series, which is expressed as:

[0150]

[0151] S44, the second time series learns the temporal trend characteristics of the first time series, reorders the generated non-steady-state similar sample data, and obtains non-steady-state similar time series data, which is expressed as:

[0152]

[0153] In the formula, Y j The value of "at the kth moment", is the value of the temporal trend feature of the first time series at the k'th moment.

[0154] It can be understood that this step utilizes the numerical arrangement of the monitoring time series to extract features that can represent the ordering of the monitoring time series as timing features, and reorders the generated non-steady-state similar sample data according to the extracted timing features to obtain non-steady-state similar time series data.

[0155] It should be noted that, by reshuffling the generated non-steady-state similar sample data, non-steady-state similar time series data that is more in line with the actual operating conditions is obtained, thereby solving the problem of data imbalance between steady-state data and non-steady-state data caused by insufficient samples of non-steady-state operation of wind power equipment, and improving the identification accuracy of the identification model.

[0156] like Figure 4 As shown in S5, a wind power equipment state identification model is constructed based on the LSTM long short-term memory network, and the non-steady-state similar time series data and the historical data set are merged as training data and input into the wind power equipment state identification model for training, and the wind power equipment operation state measurement index is output.

[0157] It should be noted that the LSTM model is good at processing time series data and can effectively capture long-term dependencies in the data. It is particularly suitable for dynamic systems such as wind power equipment. Therefore, the wind power equipment state identification model constructed based on the LSTM long short-term memory network in this embodiment is trained by combining historical data sets and amplified non-steady-state similar time series data, which significantly improves the accuracy and robustness of the model in identifying the non-stable operating state of wind power equipment.

[0158] like Figure 4 As shown, S6, a steady-state range threshold is preset, and the operating state of the wind power equipment is determined according to the wind power equipment operating state measurement index.

[0159] Specifically, the steady-state range threshold in step S6 in this embodiment is:

[0160]

[0161] In the formula, is the maximum value of the steady-state range threshold at the i-th moment, is the minimum value of the steady-state range threshold at the i-th moment, μ j is the average value of the j-th variable metric, σ j is the variance of the j-th variable metric;

[0162] When the wind power equipment operating state measurement indicators output by each variable are all within the corresponding steady-state range threshold, the wind power equipment is determined to be operating in a stable state, otherwise the wind power equipment is determined to be operating in an unstable state.

[0163] The wind power equipment is operating in a stable state as c1, and the wind power equipment is operating in an unsteady state as c2.

[0164] like Figure 5 As shown, in addition, this embodiment also includes step S7, locating the unstable operation state variable based on the residual spectrum to determine the cause of the unstable operation state of the wind power equipment, including the following sub-steps:

[0165] S71, obtaining real-time data of the wind power equipment based on real-time data sliding window sampling, performing variable similarity distance matrix calculation, and obtaining a real-time distance matrix;

[0166] S72, calculating a variable similarity distance matrix based on the time series data of the wind power equipment operating in a steady state to obtain an original distance matrix;

[0167] S73, based on the trained WGAN model, inputting the time series data of the wind power equipment operating in a steady state to obtain steady-state generated sample data, performing variable similarity distance matrix calculation based on the steady-state generated sample data, and obtaining a generated distance matrix;

[0168] S74, performing difference calculation on corresponding elements according to the original distance matrix and the standard distance matrix to obtain a standard residual matrix;

[0169] S75, converting the standard residual matrix and the real-time distance matrix into a standard residual spectrum and a real-time residual spectrum respectively, and identifying the key variables causing the unstable operation state of the wind power equipment according to the difference between the two, and determining the location of the cause of the unstable operation state of the wind power equipment.

[0170] It should be noted that the sliding window sampling technology can obtain the latest operating data of wind power equipment in real time, ensuring the real-time monitoring of the equipment status. By calculating the difference between the original distance matrix and the generated distance matrix, the standard residual matrix is ​​obtained, which quantifies the difference between the model-generated data and the real steady-state data, and obtains a standard residual matrix. The standard residual matrix is ​​compared with the real-time distance matrix. The real-time residual matrix can intuitively show the difference between the current state and the steady state of the equipment, which is convenient for timely detection of abnormalities. After converting the residual matrix into a residual spectrum, by comparing the difference between the standard residual spectrum and the real-time residual spectrum, the key variables that cause the unstable operation state of the wind power equipment can be accurately identified. The identification of key variables provides a direct basis for determining the cause of the unstable operation state of the wind power equipment, which helps to quickly locate the fault point and reduce maintenance time and cost. Therefore, the real-time monitoring, accurate identification and cause location of the unstable operation state of wind power equipment are realized, which not only improves the efficiency and accuracy of fault diagnosis, but also provides data support for the preventive maintenance and optimized operation of wind power equipment, and improves the reliability and availability of equipment.

[0171] The variable similarity distance matrix calculation expression in this embodiment is:

[0172]

[0173] In the formula, x t1 and x t2 are two time step variables, and n is the total number of samples;

[0174] The distance matrix expression is:

[0175]

[0176] In addition, this embodiment also includes step S8, which is an adaptive update of the steady-state range threshold based on unsupervised clustering and incremental learning, which includes the following sub-steps:

[0177] S81, normalizing the real-time monitoring data of the wind power equipment and inputting it into the wind power equipment state identification model in time sequence to obtain output measurement indicators corresponding to each data;

[0178] S82, obtaining a state metric index threshold library, in which the steady-state range threshold corresponding to each variable and each moment is stored, and comparing each data output metric index with the corresponding steady-state range threshold;

[0179] The normalized data set of the real-time monitoring data of wind power equipment is X'=(X t ', X t+1 ',…,X t+δ '), corresponding to the output metric D = {D 1 ,D 2 ,…,D n}, the state metric threshold library is

[0180] S83, if the output metric corresponding to a state category in the real-time monitoring data within a time window is in the corresponding steady-state range threshold If the number of values ​​in the steady-state range is greater than half, the steady-state range threshold in the state metric threshold library is To update, the expression is:

[0181]

[0182] Otherwise, the real-time monitoring data at that moment will be marked as an outlier and stored in the outlier database;

[0183] S83, preset a response threshold, use a counter to count the number of outliers in the outlier library, when the number of outliers in the outlier library reaches the preset response threshold, use a clustering algorithm to cluster the outlier library to obtain outlier data sets of different category states, if the threshold is not reached, the currently detected outlier is regarded as a noise point, and the system operation status continues to be analyzed;

[0184] In this embodiment, the K-MENS clustering algorithm is used to classify the data in the outlier database.

[0185] S84, determining whether the sample size in the outlier data set of each category state reaches the update threshold of the wind power equipment state identification model. If the sample size of a certain category reaches the update threshold, the wind power equipment state identification model is trained based on the outlier data set of the category state until convergence.

[0186] S85, based on the trained wind power equipment state identification model, calculate the steady-state threshold corresponding to the state, and add it to the state measurement index threshold library. The updated state measurement index threshold library is

[0187] in,

[0188] The steady-state threshold expression corresponding to this state is:

[0189]

[0190] It should be noted that the dynamic update of the steady-state range threshold can reflect the changes in the actual operating state of the wind power equipment, and can improve the accuracy of state identification, the adaptability and robustness of the system.

[0191] In a second aspect, the present invention further provides a system for identifying the unstable operation state of the service quality of wind power equipment, which is implemented by using the method for identifying the unstable operation state of the service quality of wind power equipment according to any one of the above claims, and the system comprises:

[0192] A data acquisition module, used to obtain a historical data set of wind power equipment in service, wherein the historical data set includes time series data of the wind power equipment operating in a steady state and an unsteady state;

[0193] The first generation module is used to train the data in the historical data set based on the CAN and MMD evaluation indicators of the generative adversarial network to determine the minimum sample size during model training;

[0194] The second generation module is used to perform data augmentation training on the non-steady-state data in the historical data set based on the WGAN model according to the minimum sample size and the historical data set, and generate non-steady-state similar sample data;

[0195] The sample reorganization module is used to reorder the generated non-steady-state similar sample data based on the time series data running in the non-steady state in the historical data set based on the Reshuffling time series learning to obtain non-steady-state similar time series data;

[0196] The model building module is used to build a wind power equipment state identification model based on the LSTM long short-term memory network, merge the non-steady-state similar time series data with the historical data set as training data and input it into the wind power equipment state identification model for training, and output the wind power equipment operation state measurement indicators;

[0197] The state determination module is used to preset a steady-state range threshold and determine the operating state of the wind power equipment according to the wind power equipment operating state measurement index.

[0198] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above-mentioned methods for identifying the unstable operating state of the service quality of wind power equipment when executing the computer program.

[0199] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying unstable operation state of service quality of wind power equipment, characterized in that: The method comprises the following sub-steps: S1, obtaining a historical data set of wind power equipment in service, wherein the historical data set includes time series data of the wind power equipment operating in a steady state and an unsteady state; S2, training the data in the historical data set based on the CAN and MMD evaluation indicators of the generative adversarial network to determine the minimum sample size for model training; S3, based on the minimum sample size and historical data set, perform data augmentation training on the non-steady-state data in the historical data set based on the WGAN model to generate non-steady-state similar sample data; S4, according to the time series data running under non-steady state in the historical data set, the generated non-steady-state similar sample data is reordered based on the Reshuffling time series learning to obtain non-steady-state similar time series data; S5, construct a wind power equipment state identification model based on LSTM long short-term memory network, merge the non-steady-state similar time series data with the historical data set as training data and input it into the wind power equipment state identification model for training, and output the wind power equipment operation state measurement index; S6, preset a steady-state range threshold, and determine the operating state of the wind power equipment according to the wind power equipment operating state measurement index.

2. The method for identifying the unstable operation state of the service quality of wind power equipment according to claim 1, characterized in that: Step S2 includes the following sub-steps: S21, constructing a generative adversarial network (CAN) model, and initializing the parameters of the generator and discriminator of the generative adversarial network (CAN) model; S22, presetting a plurality of learning rate parameters within the learning rate range, and presetting a plurality of groups of different alternation times; S23, training using the historical data set of the wind power equipment in service according to each learning rate parameter, and performing an iterative training process for each group of alternation times to obtain a generated data set; S24, after each iterative training, the MMD indicator is used to calculate the distance between the distribution of the generated sample data set and the historical data set, and the sample size at which the MMD indicator value begins to stabilize and approaches the minimum is selected as the minimum sample size for model training. The calculation expression is: In the formula, φ(·) represents the mapping function of the original feature space X to the Hilbert space H, X (s) represents the distribution of historical data set samples, X (t) represents the distribution of the generated dataset samples, represents samples drawn from historical data sets, represents the samples drawn from the generated dataset; n s Represents the sample size of the historical data set, n t represents the sample size of the generated dataset, ||·|| H represents the norm in Hilbert space.

3. The method for identifying the unstable operation state of the service quality of wind power equipment according to claim 2, characterized in that: Step S3 includes the following sub-steps: S31, preprocessing the data in the historical data set of wind power equipment service based on the data normalization method, the formula for data normalization is: In the formula, x ij ′ is the historical sample data of the i-th variable after normalization at the j-th moment, x ij is the historical sample data of the ith variable at the jth moment, max(X j ),min(X j ) are the maximum and minimum values ​​of the jth variable in all historical sample data respectively; S32, constructing a WGAN model, and randomly initializing the parameters of the generator G and the discriminator D of the WGAN model; S33, define the loss function and optimizer, and set the learning rate parameters and decay rate. The loss function expression is: Where, L D represents the loss function of the discriminator, L G represents the loss function of the generator, D(x) represents the score of the discriminator on the historical sample data x, and D(G(z)) represents the score of the discriminator on the generated sample data G(z); S34, input the normalized time series data of wind power equipment operating in non-steady state in the historical data set into the WGAN model for iterative training, adjust the learning rate parameters and decay rate of the WGAN model based on the optimizer until the WGAN model reaches a convergence state, and the WGAN model outputs the generated non-steady-state similar sample data.

4. The method for identifying the unstable operation state of the service quality of wind power equipment according to claim 3, characterized in that: The step S4 comprises the following sub-steps: S41, constructing a first time series and a second time series, wherein the length of the first time series X is T, wherein each monitoring point of the first time series X has q variables, and the first time series X corresponds to the time series data running under non-steady state in the historical data set, and the first time series expression is: In the formula, x Tq is the monitoring value of the monitoring point at the Tth moment of the qth variable of the first time series X; The length of the second time series Y is T, where each monitoring of the second time series Y has q variables, and the expression of the first time series is: In the formula, y Tq is the monitoring value of the monitoring point at the Tth moment of the qth variable of the second time series Y; S42, respectively obtaining a sequential sequence of the first time series and the second time series; The order sequence calculation expression of the first time series is: In the formula, For X j The observed value at the i-th moment, For X j 'The i′th moment observation value; The order sequence calculation expression of the second time series is: In the formula, Y j The observed value at the kth moment, Y j 'The k'th moment observation value; S43, according to X j 'The index i of each observation' for the first time series X j The corresponding observation value i is marked to obtain the temporal trend characteristics of the first time series, which is expressed as: S44, the second time series learns the temporal trend characteristics of the first time series, reorders the generated non-steady-state similar sample data, and obtains non-steady-state similar time series data, which is expressed as: In the formula, Y j The value of "at the kth moment", is the value of the temporal trend feature of the first time series at the k'th moment.

5. The method for identifying the unstable operation state of the service quality of wind power equipment according to claim 1, characterized in that: The steady-state range threshold in step S6 is: In the formula, is the maximum value of the steady-state range threshold at the i-th moment, is the minimum value of the steady-state range threshold at the i-th moment, μ j is the average value of the j-th variable metric, σ j is the variance of the j-th variable metric; When the wind power equipment operating state measurement indicators output by each variable are all within the corresponding steady-state range threshold, the wind power equipment is determined to be operating in a stable state, otherwise the wind power equipment is determined to be operating in an unstable state.

6. The method for identifying the unstable operation state of the service quality of wind power equipment according to claim 5, characterized in that: The method further includes step S7, locating the unstable operation state variable based on the residual spectrum to determine the cause of the unstable operation state of the wind power equipment, including the following sub-steps: S71, obtaining real-time data of the wind power equipment based on real-time data sliding window sampling, performing variable similarity distance matrix calculation, and obtaining a real-time distance matrix; S72, calculating a variable similarity distance matrix based on the time series data of the wind power equipment operating in a steady state to obtain an original distance matrix; S73, based on the trained WGAN model, inputting the time series data of the wind power equipment operating in a steady state to obtain steady-state generated sample data, performing variable similarity distance matrix calculation based on the steady-state generated sample data, and obtaining a generated distance matrix; S74, performing difference calculation on corresponding elements according to the original distance matrix and the standard distance matrix to obtain a standard residual matrix; S75, converting the standard residual matrix and the real-time distance matrix into a standard residual spectrum and a real-time residual spectrum respectively, and identifying the key variables causing the unstable operation state of the wind power equipment according to the difference between the two, and determining the location of the cause of the unstable operation state of the wind power equipment.

7. The method for identifying the unstable operation state of the service quality of wind power equipment according to claim 6, characterized in that: The variable similarity distance matrix calculation expression is: In the formula, x t1 and x t2 are two time step variables, and n is the total number of samples; The distance matrix expression is:

8. The method for identifying the unstable operation state of the service quality of wind power equipment according to claim 7, characterized in that: The method further includes step S8, adaptively updating the steady-state range threshold based on unsupervised clustering and incremental learning, which includes the following sub-steps: S81, normalizing the real-time monitoring data of the wind power equipment and inputting it into the wind power equipment state identification model in time sequence to obtain output measurement indicators corresponding to each data; S82, obtaining a state metric index threshold library, in which the steady-state range threshold corresponding to each variable and each moment is stored, and comparing each data output metric index with the corresponding steady-state range threshold; S83, if the number of output metrics corresponding to a state category in the real-time monitoring data within a time window that is within the corresponding steady-state range threshold is greater than half, then the steady-state range threshold in the state metric threshold library is updated as follows: Otherwise, the real-time monitoring data at that moment will be marked as an outlier and stored in the outlier database; S83, preset a response threshold, use a counter to count the number of outliers in the outlier library, when the number of outliers in the outlier library reaches the preset response threshold, use a clustering algorithm to cluster the outlier library to obtain outlier data sets of different category states, if the threshold is not reached, the currently detected outlier is regarded as a noise point, and the system operation status continues to be analyzed; S84, determining whether the sample size in the outlier data set of each category state reaches the update threshold of the wind power equipment state identification model. If the sample size of a certain category reaches the update threshold, the wind power equipment state identification model is trained based on the outlier data set of the category state until convergence. S85, based on the trained wind power equipment state identification model, calculating the steady-state threshold corresponding to the state, and adding it to the state measurement indicator threshold library; The steady-state threshold expression corresponding to this state is:

9. A system for identifying the unstable operation state of the service quality of wind power equipment, implemented by the method for identifying the unstable operation state of the service quality of wind power equipment according to any one of claims 1 to 8, characterized in that: The system comprises: A data acquisition module, used to obtain a historical data set of wind power equipment in service, wherein the historical data set includes time series data of the wind power equipment operating in a steady state and an unsteady state; The first generation module is used to train the data in the historical data set based on the CAN and MMD evaluation indicators of the generative adversarial network to determine the minimum sample size during model training; The second generation module is used to perform data augmentation training on the non-steady-state data in the historical data set based on the WGAN model according to the minimum sample size and the historical data set, and generate non-steady-state similar sample data; The sample reorganization module is used to reorder the generated non-steady-state similar sample data based on the time series data running in the non-steady state in the historical data set based on the Reshuffling time series learning to obtain non-steady-state similar time series data; The model building module is used to build a wind power equipment state identification model based on the LSTM long short-term memory network, merge the non-steady-state similar time series data with the historical data set as training data and input it into the wind power equipment state identification model for training, and output the wind power equipment operation state measurement indicators; The state determination module is used to preset a steady-state range threshold and determine the operating state of the wind power equipment according to the wind power equipment operating state measurement index.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for identifying the unstable operating state of the service quality of wind power equipment according to any one of claims 1 to 8 when executing the computer program.

Images