A method and system for intelligent diagnosis of operating conditions of a wind turbine

Abstract

A kind of intelligent diagnosis method and system for wind turbine operating state, based on big data platform realizes intelligent diagnosis of wind turbine operating state, potential hidden danger existing in wind turbine is excavated, effectively reduce the operation and maintenance cost of wind turbine, improve the economic benefit of wind power;The intelligent diagnosis of wind turbine operating state of the present application fuses Self-representation unsupervised feature selection and set pair analysis, the unsupervised feature selection method relieves the difficult problem of evaluation index selection and collection of labeled training samples, improves the precision, generalization performance and robustness of model;The set pair analysis method fully considers the relationship between power grid, generator, gear box and main shaft, cabin and variable pitch, can effectively analyze the uncertainty of system, realize multi-source information fusion, improve the rapid and accurate diagnosis of wind turbine operating state.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind turbine operation data analysis, and particularly relates to a wind turbine operation state intelligent diagnosis method and system. BACKGROUND

[0002] Considering the environment and load stability of wind farms, as the operation time of wind turbines increases, various types of faults may occur in the electrical, transmission and control systems of the wind turbines, resulting in abnormal operation of the wind turbines or even accidents. The SCADA system of the wind farm can only send an alarm when the monitoring parameters of the wind turbine exceed the set threshold, and at this time the components of the wind turbine may have been damaged. Therefore, in order to reduce losses, the existing wind turbine operation state diagnosis method is more based on artificial experience and shallow learning for analysis. The wind turbine operation state diagnosis problem not only involves the uncertainty of state parameters and information loss, but also involves the information fusion problem between subsystems. Therefore, a single monitoring method cannot comprehensively deal with the uncertainty and information fusion problems involved in the overall operation state monitoring of the wind turbine, and inaccurate state information also reduces the effectiveness of the analysis, resulting in a large number of false positives and false negatives, which cannot fully play the advantages of massive wind turbine data. SUMMARY

[0003] In order to solve the problems existing in the prior art, the present application provides a wind turbine operation state intelligent diagnosis method, comprising:

[0004] obtaining real-time operation data of a wind farm based on a pre-determined optimized wind turbine operation state diagnosis index system;

[0005] calculating a multi-element contact degree expression pre-constructed based on the real-time operation data of the wind farm to obtain a contact value of each object of the index system object layer and the wind turbine operation state;

[0006] when the wind turbine operation state is normal, taking the wind turbine operation state as the diagnosis result; otherwise, taking the contact value of each monitoring object as the diagnosis result, and determining a corresponding operation and maintenance scheme based on the contact value of each monitoring object;

[0007] wherein the index system determines a target layer with the wind turbine operation state as the target, and constructs a plurality of monitoring objects as the object layer for the target layer, and performs dimension reduction processing by using a Self-representation unsupervised feature selection method; the multi-element contact degree expression is constructed by using a set pair analysis method based on historical wind turbine operation states and historical operation data.

[0008] Preferably, the construction of the index system comprises:

[0009] determining a target layer with a wind turbine operating state as a target;

[0010] determining a plurality of monitoring objects as an object layer based on historical operating data provided by a plurality of data sources of a wind farm and fault conditions of the wind turbine;

[0011] determining a plurality of indexes for each object of the object layer respectively, thereby forming an index layer;

[0012] performing dimensionality reduction processing on each index of the index layer based on a Self-representation unsupervised feature selection method;

[0013] obtaining a multi-level index system of the wind turbine operating state based on the target layer, the object layer and the index layer after the dimensionality reduction processing.

[0014] Preferably, the monitoring objects at least include one or more of the following: a generator, a gearbox, a nacelle, a main bearing, a converter, a variable pitch, a power grid environment and a natural environment.

[0015] Preferably, the dimensionality reduction processing on each index of the index layer based on the Self-representation unsupervised feature selection method comprises:

[0016] constructing a feature quantity matrix X based on each index in the index layer;

[0017] substituting the feature quantity matrix X into a pre-split unconstrained optimization equation to obtain a plurality of indexes with the largest features in the feature quantity matrix X;

[0018] wherein the optimization equation is determined based on the Self-representation unsupervised feature selection method after splitting.

[0019] Preferably, the determination of the split unconstrained optimization equation comprises:

[0020] determining an optimization equation for index feature selection by using a Self-representation method;

[0021] obtaining the split unconstrained optimization equation by solving the optimization equation by using a split Bregman method.

[0022] Preferably, the unconstrained optimization equation is calculated as follows:

[0023]

[0024] In the formula: D is an auxiliary matrix variable; E is a self-representation error matrix variable; F is a gradient at W; W is a self-representation matrix variable; λ, β are regularization parameters; TL1 is a transformed L1 norm; φ is a penalty parameter; B is an auxiliary variable; k is the number of iterations; ||W||2 is a L2 norm. 2,1 -||W||2 is a L2 norm. F is a L 1,2-2 norm.

[0025] Preferably, the characteristic quantity matrix X is substituted into the pre-split optimization equation to obtain the first several characteristics with the largest characteristics in the characteristic quantity matrix X, including:

[0026] Based on the updating rule of the auxiliary variable B, the parameters in the unconstrained optimization equation are iteratively solved by using an alternating iteration method to obtain the values of the auxiliary matrix D, the self-representation error matrix E, the self-representation matrix W, and the gradient F at W at the current iteration number.

[0027] Based on the non-negativity of the self-representation matrix W, a projection operator equation is executed to update the self-representation matrix W.

[0028] Based on the values of the auxiliary matrix D, the self-representation error matrix E, the gradient F at W, and the updated self-representation matrix W at the current iteration number, the auxiliary variable B is updated, and the iterative calculation is continued until the convergence condition is met.

[0029] After the iteration is completed, the indicators in the characteristic quantity matrix X are sorted based on ||W||2, and the largest several indicators are selected. p ||2 is a L2 norm.

[0030] Preferably, the updating rule of the auxiliary variable B at k+1 iteration is as follows:

[0031] B k+1 = B k -(D k+1 -(X-XW k+1 -E k+1 ))

[0032] The calculation formula of the auxiliary matrix variable D at k+1 iteration is as follows:

[0033]

[0034] The calculation formula of the self-representation error matrix variable E at k+1 iteration is as follows:

[0035]

[0036] The calculation formula of the gradient F at W at k+1 iteration is as follows:

[0037] F k = λW k / ||W k || F

[0038] The calculation formula of the self-representation matrix variable W at k+1 iteration is as follows:

[0039]

[0040] The projection operator equation is as follows:

[0041]

[0042] In the formula, p and q represent the pth row and the qth column respectively.

[0043] Preferably, after obtaining the multi-level index system of the wind turbine operating state, the method further comprises:

[0044] Determining the constant weight value of each index in the index layer according to statistical data and expert opinions;

[0045] Calculating the variable weight coefficient of each index and the variable weight value of each monitoring object based on the constant weight coefficient.

[0046] Preferably, the calculation formula of the variable weight value of each index is as follows:

[0047]

[0048] In the formula, w m,n is the variable weight value of the evaluation index n under the monitoring object m; is the constant weight value of the evaluation index n under the monitoring object m; x m,n is the normalized value of the evaluation index n under the monitoring object m; a is the variable weight coefficient;

[0049] The calculation formula of the variable weight value of each monitoring object is as follows:

[0050]

[0051] In the formula, w' m is the variable weight value of the monitoring object m; is the constant weight value of the monitoring object m; is the score value of the monitoring object m, and

[0052] Preferably, the method for obtaining real-time operating data of the wind farm based on the pre-determined optimized wind turbine operating state diagnosis index system comprises:

[0053] The real-time operation data of the wind farm is obtained from operation data provided by multiple data sources of the wind farm on the big data platform based on an index layer of the index system.

[0054] Preferably, the construction of the multi-element contact degree expression comprises:

[0055] The operation state index layer of the wind turbine after the dimension reduction processing is taken as a data set.

[0056] The multi-element contact degree expression is determined based on the data set by using the set pair analysis method.

[0057] The multi-element contact degree expression comprises a same-different-contrary evaluation matrix and a same-different-contrary coefficient matrix.

[0058] Preferably, the multi-element contact degree expression of the object layer is as follows:

[0059]

[0060] In the formula, μ m is the multi-element contact degree of the monitoring object m; W m is a variable weight coefficient matrix of the monitoring object m corresponding to the index layer set; R m is a same-different-contrary evaluation matrix of the monitoring object m corresponding to the index layer set; and E is a same-different-contrary coefficient matrix.

[0061] The expression of the same-different-contrary evaluation matrix of the monitoring object m corresponding to the index layer set is as follows:

[0062] R m = [R m,1 , R m,2 , …, R m,n ] T ;

[0063] In the formula, R m,n is the same-different-contrary evaluation matrix of the nth index of the monitoring object m.

[0064] The expression of the variable weight coefficient matrix of the monitoring object m corresponding to the index layer set is as follows:

[0065] W m = [w m,1 , w m,2 , …, w m,n ]

[0066] In the formula, w m,n is the variable weight value of the nth index of the monitoring object m.

[0067] The expression of the same-different-contrary coefficient matrix is as follows:

[0068] E = [1, i1, …, ig-1, ig, ig+1, …, in-1, in] g-2 T

[0069] wherein: i g-2 is the difference coefficient of the wind turbine operating state level being the g-1th; and j is the opposition coefficient.

[0070] Preferably, the R m,n is expressed as follows:

[0071] R m,n = [r1(x m,n ), r2(x m,n ), …, r g (x m,n )]

[0072] wherein: r g is the membership function value of the wind turbine state level being g; and x m,n is the normalized value of the evaluation index.

[0073] Preferably, the multi-element connection degree of the wind turbine operating state layer is expressed as follows:

[0074]

[0075] wherein: μ' is the multi-element connection degree of the wind turbine operating state layer; W' is the variable weight coefficient matrix of the monitoring object set under the wind turbine operating state layer; R' is the same-different-opposite evaluation matrix of the monitoring object set under the wind turbine operating state layer; and E is the same-different-opposite coefficient matrix.

[0076] The same-different-opposite evaluation matrix of the monitoring object set under the wind turbine operating state layer is expressed as follows:

[0077] R' = [W1R1, W2R2, …, W m R m ] T ;

[0078] wherein: W m is the multi-element connection degree of the monitoring object m; and R m is the weight coefficient matrix of the index layer set corresponding to the mth monitoring object.

[0079] The variable weight coefficient matrix of the monitoring object set under the wind turbine operating state layer is expressed as follows:

[0080] W' = [w'1, w'2, …, w' m ]

[0081] wherein: w' m is the variable weight value of the monitoring object m.​

[0082] Preferably, the calculation of the multi-element connection degree expression based on the real-time operation data of the wind farm obtains the connection values of the wind turbine operation state and each object in the object layer of the index system, including:

[0083] The same-different-contrary coefficient matrix is obtained by using the equal division method based on the predetermined wind turbine state level;

[0084] The same-different-contrary evaluation matrix of each index under each monitoring object layer is determined based on the real-time operation data of the wind farm;

[0085] The connection values of each object are obtained by weighted summation based on the same-different-contrary evaluation matrix of each index, the same-different-contrary coefficient matrix, and the variable weight value corresponding to the index;

[0086] The same-different-contrary evaluation matrix of each monitoring object is determined based on the index values of each monitoring object;

[0087] The connection values of the target layer are obtained by weighted summation based on the same-different-contrary evaluation matrix of each monitoring object, the same-different-contrary coefficient matrix, and the variable weight value corresponding to the monitoring object.

[0088] Preferably, the data source at least includes one or more of the following: a wind farm second-level data SCADA system, a unit main control, a converter, an AGC / AVC, a primary frequency control device, and an energy management platform.

[0089] Based on the same invention concept, the present application also provides a wind turbine operation state intelligent diagnosis system, comprising:

[0090] A data acquisition layer is used to obtain real-time operation data of a wind farm based on a predetermined optimized wind turbine operation state diagnosis index system;

[0091] A data processing layer is used to calculate the connection values of the wind turbine operation state and each object in the object layer of the index system by calculating a multi-element connection degree expression based on the real-time operation data of the wind farm; when the wind turbine operation state is normal, the wind turbine operation state is taken as the diagnosis result; otherwise, the connection values of each monitoring object layer are taken as the diagnosis result, and the corresponding operation and maintenance scheme is determined based on the connection values of each monitoring object;

[0092] The index system determines a target layer with the wind turbine operation state as the target, and constructs a plurality of monitoring objects as the object layer, and uses the Self-representation unsupervised feature selection method for dimension reduction processing; the multi-element connection degree expression is constructed by using the set pair analysis method based on the historical wind turbine operation state and the historical operation data.

[0093] Preferably, the system further comprises a data storage layer;

[0094] The data storage layer comprises a big data platform and an algorithm library;

[0095] The algorithm library comprises a dimension reduction algorithm and a state diagnosis algorithm in the wind turbine operation state diagnosis index system;

[0096] The big data platform is used to store the real-time operation data of the wind farm provided by the collection layer based on the big data platform.

[0097] Preferably, the state diagnosis algorithm comprises at least one or more of the following: a multivariate contact degree expression of the object layer, an expression of a same-different-contrary evaluation matrix of the mth monitoring object, an expression of a weight coefficient matrix of the index layer set corresponding to the mth monitoring object, an expression of a same-different-contrary coefficient matrix, a multivariate contact degree expression of the wind turbine operation state layer, an expression of a same-different-contrary evaluation matrix of the wind turbine operation state layer, and an expression of a weight coefficient matrix of the index layer set corresponding to the mth monitoring object.

[0098] Compared with the prior art, the present application has the following advantages:

[0099] 1) The present application proposes a wind turbine operation state intelligent diagnosis method and system, which combines the Self-representation unsupervised feature selection and the set pair analysis wind turbine operation state diagnosis method. The unsupervised feature selection method alleviates the difficulty of selecting and collecting labeled training samples, and improves the precision, generalization performance and robustness of the model. The set pair analysis method fully considers the relationship between the power grid, the generator, the gearbox and the main shaft, the cabin and the variable pitch, can effectively analyze the uncertainty of the system, realize multi-source information fusion, and improve the rapid and accurate diagnosis of the wind turbine operation state.

[0100] 2) The present application proposes a wind turbine operation state intelligent diagnosis method and system, which realizes wind turbine operation state intelligent diagnosis based on a big data platform, excavates potential hidden dangers of wind turbines, effectively reduces the operation and maintenance cost of wind turbines, and improves the economic benefit of wind power. BRIEF DESCRIPTION OF DRAWINGS

[0101] Figure 1 A wind turbine operation state intelligent diagnosis method block diagram is provided for the present application;

[0102] Figure 2 A wind turbine operation state diagnosis index system is provided for the present application;

[0103] Figure 3 A membership function diagram corresponding to the wind turbine operation state and the deterioration degree is provided for the present application;

[0104] Figure 4 The application provides a wind turbine operation state intelligent diagnosis process.

[0105] Figure 5 The application provides a wind turbine operation state intelligent diagnosis system overall system architecture. DETAILED DESCRIPTION

[0106] The application aims at the problems of insufficient diagnosis accuracy and difficult selection of evaluation indexes of the current wind turbine operation state diagnosis, and provides a wind turbine operation state intelligent diagnosis method and system, so as to realize accurate discrimination of the wind turbine operation state.

[0107] In order to better understand the application, the content of the application is further described below in combination with the description, the accompanying drawings and examples.

[0108] Embodiment 1

[0109] The application provides a wind turbine operation state intelligent diagnosis method, which can intelligently diagnose the wind turbine operation state, for example, as shown in the figure, the method comprises the following steps. Figure 1

[0110] S1, obtaining wind farm real-time operation data based on a pre-determined optimized wind turbine operation state diagnosis index system;

[0111] S2, calculating a multi-element contact degree expression pre-constructed based on the wind farm real-time operation data to obtain a contact value of each object of the index system object layer and the wind turbine operation state;

[0112] S3, when the wind turbine operation state is a normal state, taking the wind turbine operation state as a diagnosis result; otherwise, taking the contact value of each monitoring object as a diagnosis result, and determining a corresponding operation and maintenance scheme based on the contact value of each monitoring object;

[0113] The index system determines a target layer with the wind turbine operation state as a target, constructs a plurality of monitoring objects as an object layer for the target layer, and performs dimension reduction processing by using a Self-representation unsupervised feature selection method; and the multi-element contact degree expression is constructed by using a set pair analysis method based on historical wind turbine operation states and historical operation data.

[0114] ​The overall concept of this invention lies in introducing feature selection and set pair analysis methods into the diagnosis of wind turbine operation status. Feature selection eliminates irrelevant and redundant monitoring indicators, minimizing data dimensionality and training time costs. Set pair analysis fully considers the inherent relationships between indicators, effectively analyzes system uncertainties, and achieves multi-source information fusion. In summary, by utilizing massive amounts of data for intelligent diagnosis of wind turbine operation status, fully exploring the value of wind turbine data, and establishing an intelligent diagnostic system for wind turbine operation status based on a big data platform, this invention achieves accurate identification of turbine status and provides technical support for the safe and efficient grid-connected operation of wind turbines.

[0115] The technical means used in this invention will be described below:

[0116] Establishment of a diagnostic index system for the operating status of wind turbine units

[0117] 1) Construct a multi-layered diagnostic index for generator set operation status.

[0118] Based on the actual fault conditions of wind turbine units during operation, appropriate condition monitoring information should be selected to establish a diagnostic index system for the operating status of wind turbine units. For example... Figure 2 As shown, the established indicators are divided into three levels: target layer, object layer, and indicator layer. The wind turbine operating status serves as the target layer. The object layer includes eight evaluation factors: generator, gearbox, nacelle, main bearing, converter, pitch, grid environment, and natural environment. Each evaluation factor contains several indicators. The indicator layer includes 49 indicators under eight object layers: line voltage, phase current, grid frequency, gearbox cooling temperature, generator speed, generator torque, tower base cabinet temperature, and active power. This constitutes a wind turbine operating status diagnostic indicator system. The established indicators comprehensively reflect the wind turbine operating status from different perspectives. Data sources include SCADA system, turbine main control, converter, AGC / AVC, primary frequency control device, and energy management platform.

[0119] 2) Unsupervised feature selection based on self-representation

[0120] Feature selection involves filtering features from a high-dimensional feature set (here, each indicator in the indicator layer is constructed as a feature set) according to specific rules to obtain a lower-dimensional feature set. This new feature set is a subset of the original set, and each feature retains its original numerical value and physical meaning. This application uses a self-representation-based unsupervised feature selection method to reduce the dimensionality of the indicators in the indicator layer. The main process is as follows:

[0121] Given a data matrix X = (X1, ..., X...) n )∈Rd× nThen the feature selection problem can be converted into the following optimization problem:

[0122]

[0123] where E represents the self-expression error matrix; ||X-XW-E||1is the data fidelity term, and the L1norm is used to ensure the robustness of the estimation; ||W||1 2,1 -||W|| F represents the L 1,2-2 norm, which is used to ensure the sparsity of W; TL1represents the Transformed L1norm, which is used to ensure the sparsity of E, and is defined as:

[0124]

[0125]

[0126] The equation (1) is solved by using the split Bregman method:

[0127]

[0128] The equation (4) can be converted into the following unconstrained optimization problem:

[0129] min{ψ(D,W,E)} (5)

[0130] where ψ(D,W,E) is defined as follows:

[0131]

[0132] The update rule of the auxiliary variable B is:

[0133] B k+1 =B k -(D k+1 -(X-XW k+1 -E k+1 )) (7)

[0134] Solving the equation (6) by using the alternating iteration method leads to the following sub-problems:

[0135] D k+1 =min{ψ(D,W k ,E k )} (8)

[0136] E k+1 =min{ψ(D k+1 ,W k ,E)} (9)

[0137] W k+1 =min{ψ(D k+1W, E k+1 )} (10)

[0138] According to equation (6), equations (8)-(10) can be specifically expressed as:

[0139]

[0140]

[0141]

[0142] Equations (11) and (12) can be easily solved. Equation (13) is a DC programming problem, which can be further relaxed as:

[0143]

[0144] In the formula, F k is ||W|| F The gradient at W k is defined as:

[0145] F k = λW k / ||W k || F (15)

[0146] To further simplify the calculation, ||W|| 2,1 is approximated as:

[0147]

[0148] In the formula, Ω pp is a diagonal matrix,

[0149] Finally, equation (14) can be rewritten as:

[0150]

[0151] It is obvious that equation (17) can be easily solved. In addition, in order to determine the non-negativity of the W matrix, we map the following projection operator:

[0152]

[0153] According to the above discussion, the proposed feature selection method can be summarized in algorithm 1.

[0154] The above D represents the auxiliary matrix; E represents the self-representation error matrix; F represents ||W|| FGradient at W; W represents the self-representation matrix; λ, β are regularization parameters; TL1 is the Transformed L1 norm; φ is the penalty parameter; B is the auxiliary variable; k is the iteration number; ||W|| 2,1 -||W|| F is L 1,2-2 norm.

[0155]

[0156]

[0157] Based on the set pair analysis, a multi-element connection degree expression is established to diagnose the operation state of a wind turbine.

[0158] (1) Set pair analysis method

[0159] The essence of set pair analysis is to simultaneously contain determinacy and uncertainty in a system, and to divide the system into three aspects of "degree of identity, degree of difference, and degree of opposition", and comprehensively depict the mutual connection between sets.

[0160] Suppose that two sets A and B form a set pair H = (A, B), and the characteristics of the set pair H are analyzed under the background of a specific problem Q, and P characteristics are obtained. Among them, X characteristics are commonly possessed by sets A and B, sets A and B are opposite on Z characteristics, and the remaining Y = P - X - Z characteristics are neither commonly possessed by the two sets nor opposite to each other. Then, the connection degree expression under the background of the specified problem Q can be established as:

[0161] μ (A,B) = a + bi + cj (19)

[0162] In the formula, μ (A,B) is the connection degree of sets A and B, and the value range is within the interval [-1, 1]; a, b, and c are the degree of identity, degree of difference, and degree of opposition of the two sets under the problem Q, respectively, a = X / P, b = Y / P, and c = Z / P; i is the difference degree coefficient, which is specified in the interval [-1, 1] according to different situations; and j is the opposition degree coefficient, which is usually a constant value -1.

[0163] In order to improve the accuracy of wind turbine operation state diagnosis, the difference degree b in the connection degree μ = a + bi + cj of set pair analysis can be further divided according to the scalability of set pair analysis. According to different situations, the μ is expanded to a certain level, and the corresponding multi-element connection degree is obtained, that is:

[0164] μ = a + b1i1 + b2i2 + … + b l-2 i l-2 +cj (20)

[0165] wherein, the matrix R = [a, b1, …, b l-2 ,c] is the same, different and reverse evaluation matrix, and the matrix E = [1, i1, …, i l-2 ,j] T is the same, different and reverse coefficient matrix.

[0166] The index under the index system of the wind turbine operating state diagnosis is U = {u1, u2, …, u r}. Assuming that the wind turbine operating state level is g levels, denoted as V = {v1, v2, …, v g}, the set pair H = (U, V) is formed, representing the corresponding relationship between the wind turbine state index and the state level, and the connection degree g is determined:

[0167] μ = r1 + r2i1 + r3i2 + … + r g-1 i g-2 +r g j (21)

[0168] The same, different and reverse evaluation matrix can be expressed as R = [r1r2…r g-1 r g ].

[0169] When the set pair analysis theory is used for wind turbine state diagnosis, the same, different and reverse evaluation matrix of each index layer is first determined, and the connection degree of each object layer is obtained by weighted summation combined with the weight coefficient, that is:

[0170]

[0171] wherein, μ m is the multi-element connection degree of the monitoring object m; W m is the variable weight coefficient matrix of the monitoring object m corresponding to the index layer set, W m = [w m,1 ,w m,2 ,…, w m,n ], w m,n is the variable weight value of the nth index of the monitoring object m; R m is the same, different and reverse evaluation matrix of the monitoring object m, R m = [R m,1 ,R m,2 ,…, R m,n ] T , R m,n is the same, different and reverse evaluation matrix of the nth index of the monitoring object m; E is the same, different and reverse coefficient matrix. Similarly, the connection degree of the overall operating state of the wind turbine can be obtained:

[0172]

[0173] In the formula, W' and R' are weight coefficient matrix and same and opposite evaluation matrix of object layer set of wind turbine operating state layer, respectively, W' = [w'1, w'2, …, w'N] and R' = [R'1, R'2, …, R'N], N is the number of object layer set of wind turbine operating state layer, and R'N is same and opposite evaluation matrix of object layer set of wind turbine operating state layer. m ] and R' = [R'1, R'2, …, R'N], N is the number of object layer set of wind turbine operating state layer, and R'N is same and opposite evaluation matrix of object layer set of wind turbine operating state layer. m R m T .

[0174] (2) Index data normalization processing

[0175] In view of the fact that wind turbine operating state indexes have their own physical meanings and the orders of magnitude of the indexes are quite different and the dimensions are different, in order to avoid the mismatch caused thereby, the indexes are normalized by using the deterioration degree, and the calculation formula is as follows:

[0176]

[0177] In the formula, x is the measured value of the index parameter, and [x min ,x max ] is the normal range of the index parameter.

[0178] (3) Evaluation grade division

[0179] The present embodiment is exemplified by taking the state grade as 4 levels, i.e. good, general, attention and abnormal, which are denoted as v1, v2, v3 and v4, respectively. Table 1 shows the corresponding relationship between the wind turbine operating state grade and the relative deterioration degree and the connection value interval, Figure 3 is a distribution membership function diagram corresponding to the wind turbine operating state grade and the deterioration degree. However, the state grade can be set to multiple levels according to the specific situation.

[0180] Table 1 Corresponding relationship between wind turbine operating state grade and relative deterioration degree and connection value interval

[0181]

[0182] The values in the above table are calculated according to the state grade g = 4, and 1 is divided into 4 intervals; when g takes other data, the calculation is the same.

[0183] According to g = 4 and the deterioration degree and the connection degree value corresponding to the operating state, the membership functions of each state can be obtained from Figure 3 , which are described as follows:

[0184] 1) When the state is good, the membership function is:

[0185]

[0186] 2) When the state is general, the membership function is:

[0187] ​

[0188] 3) In the attention state, the membership function is:

[0189]

[0190] 4) In the abnormal state, the membership function is:

[0191]

[0192] The index layer relative degradation degree calculated by formula (24) is substituted into formulas (25)-(28) to obtain the membership function values r1(x m,n ), r2(x m,n ), …, r4(x m,n ) of each state level, i.e., the same, different and opposite evaluation matrix of the index layer x m,n is:

[0193] R m,n = [r1(x m,n ), r2(x m,n ), …, r4(x m,n )] (29)

[0194] Then, the same, different and opposite evaluation matrix of the object layer F m can be expressed as:

[0195] R m = [R m,1 , R m,2 , …, R m,n ] T (30).

[0196] (4) Weight determination

[0197] The operation state of the wind turbine is greatly affected by the environment, and the fault rate data statistics of different regions and different types of wind turbines are different. Only by determining the constant weight value through the fault rate data statistics will bring a large error. The patent selects the wind turbine fault statistical data and expert opinions to obtain the constant weight value, and uses the following formula to calculate and process the variable weight, reflecting the relationship between the degree of deviation of the wind turbine operation state index parameter from the normal value and the index weight. The variable weight value of the weight vector W m = [w m,1 , w m,2 , …, w m,n ] of the index layer set corresponding to a object layer is calculated as:

[0198]

[0199] In the formula, w m,n is the evaluation index variable weight value under the monitoring object m; x represents the constant weight value of the evaluation index n for the monitored object m; m,n α is the normalized value of the evaluation index n under the monitoring object m; N is the number of evaluation indicators under the monitoring object m; α is the variable weight coefficient, which is generally taken as 0.2.

[0200] Then the object layer weight vector W' = [w'1, w'2, ..., w' m The variable weight value is:

[0201]

[0202] In the formula, w' m , and y m The object layer F m Variable weighting coefficients, constant weighting coefficients, and score values α is the variable weighting coefficient.

[0203] (5) Determination of the difference coefficient

[0204] The difference coefficients of the multivariate correlation degree are processed using the equal-division method, where i1, i2, ..., i n-2 The value of should be located at n-2 equal divisions (n-1) within the interval [-1, 1]. The degree of opposition coefficient j usually takes the value -1. Then i1, i2, ..., i n-2 The values ​​are taken in sequence as follows:

[0205]

[0206] The wind turbine generator set in this patent has a state level of 4, i.e., g = 4. Substituting into formula (33), we calculate i1 and i2 as 1 / 3 and -1 / 3 respectively, and obtain the inverse coefficient matrix E = [11 / 3 - 1 / 3 - 1]. T .

[0207] Based on the above description, the process of an intelligent diagnostic method for the operating status of wind turbine generators according to the present invention is as follows: Figure 4 As shown:

[0208] Before executing step S1, perform the following procedure:

[0209] 1) Based on historical data analysis of wind turbine failures, data is extracted from multiple sources such as SCADA system, main control unit, converter, AGC / AVC, primary frequency control device and energy management platform to establish a wind turbine status diagnosis index system, including three levels: target layer, object layer and index layer.

[0210] 2) Based on the Self-representation method for unsupervised feature selection, the final 3-layer index system is established, the constant weight value is determined by the method of statistical data combined with expert opinions, and the index layer and object layer variable weight values are calculated according to formula (31) and (32);

[0211] 3) The operating state of the wind turbine is graded, the element number of the connection degree is determined, and the connection degree expression is constructed;

[0212] Step S1 is executed to obtain real-time operation data of the wind farm based on a pre-determined optimized wind turbine operating state diagnosis index system;

[0213] Step S2 is executed to calculate the connection values of the wind turbine operating state and each object of the object layer of the index system based on the real-time operation data of the wind farm, which specifically includes:

[0214] 1) Based on real-time data, the same, different and opposite evaluation matrix of the index layer corresponding to the object layer is calculated by formula (29), and then the same, different and opposite evaluation matrix of the object layer is determined by formula (30);

[0215] 2) The difference degree coefficient calculated by formula (33) is substituted into formula (22) and (23) to obtain the connection values of the overall operating state of the wind turbine and each object layer, respectively;

[0216] Step S3 specifically includes:

[0217] According to Table 1, the operating state of the wind turbine is diagnosed. If it is judged to be normal, the result is output; if it is judged to be a fault state, the connection values of each object layer are compared, the fault reason of the wind turbine is diagnosed, and the result is output.

[0218] Example 2

[0219] The wind turbine operating state intelligent diagnosis method provided by the present application is introduced below by taking the wind turbine operating in a wind farm in a certain region as an example. The method takes the wind turbine data to be analyzed as input, analyzes the model, determines the wind turbine operating state, and realizes intelligent diagnosis of the operating state of the wind turbine.

[0220] (1) Extract the historical data of the wind farm second-level data SCADA system, unit main control, converter, AGC / AVC, primary frequency control device and energy management platform, etc. to establish a wind turbine operating state diagnosis index system.

[0221] (2) Based on the Self-representation method for index feature selection, the final 3-layer index system is established, and the constant weight coefficient is determined according to the statistical data and expert opinions, and the variable weight coefficient of each index is calculated;

[0222] (3) According to the determined 4-grade wind turbine state level, a contact degree expression is constructed;

[0223] (4) According to the real-time data of the wind farm, the index layer corresponding to each object layer and the same, different and opposite evaluation matrix of the object layer are calculated;

[0224] (5) The overall operation state of the wind turbine and the contact values of each object layer are calculated by formula, the operation state of the wind turbine is diagnosed, if it is judged to be a normal state, the result is output; if it is judged to be a fault state, the contact values of each object layer are compared, and the result is output to guide the on-site personnel to formulate a corresponding operation and maintenance scheme.

[0225] Embodiment 3:

[0226] Based on the same invention concept, the application further provides a wind turbine operation state intelligent diagnosis system, which applies a big data platform to the system, and the system can provide data acquisition access, real-time cleaning, real-time storage, analysis modeling, real-time monitoring, data interaction and visualization capabilities.

[0227] The overall architecture of the system is shown in Figure 5 The system is composed of a data acquisition layer, a data storage layer, a data processing layer and an application display layer. The functions of each part are as follows:

[0228] 1) Data acquisition layer: This layer is the source of system data, including SCADA system, unit main control, converter, AGC / AVC, primary frequency control device and energy management platform data, etc.

[0229] 2) Data storage layer: This layer mainly provides data storage services for the calculation layer and the application layer. The system uses HDFS to store data, mainly for the storage and query of all types of data (structured and unstructured), with the characteristics of mass-scale storage and fast query reading. On the basis of traditional underlying hardware and file system, technologies including relational database are used to support high-level data processing applications. The data storage layer also provides computing frameworks for batch computing and in-memory computing, as well as algorithms for data mining and multi-dimensional analysis, to provide an algorithm library for high-level data applications in the upper layer.

[0230] 3) Data processing layer: The data processing layer calls the algorithm library provided by the data storage layer, and uses unsupervised feature selection and set pair analysis methods to realize accurate wind turbine state discrimination according to the requirements of wind turbine state intelligent diagnosis.

[0231] 4) Application display layer: Based on the proposed wind turbine operation state intelligent diagnosis method, the wind turbine operation state is displayed in the system to facilitate remote monitoring by actual workers.

[0232] Obviously, the described embodiments are only a part of the embodiments of the present application, but not all of them. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort fall within the protection scope of the present application.

[0233] Those skilled in the art should clearly understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt a form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0234] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the functions specified in one or more blocks.

[0235] These computer program instructions can also be stored in a computer readable storage medium capable of guiding the computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer readable storage medium produce a product including instruction means, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the functions specified in one or more blocks.

[0236] These computer program instructions can also be loaded into a computer or other programmable data processing device, so that a series of operation steps are performed on the computer or other programmable data processing device to produce a computer implemented process, so that the instructions executed on the computer or other programmable data processing device provide a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the functions specified in one or more blocks.

[0237] The above merely illustrates the embodiments of the present application, but should not be taken as limitations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall fall into the protection scope of the present application.

Claims

1. A method for intelligent diagnosis of operating conditions of a wind turbine, characterized in that, The application comprises the following steps: obtaining real-time operation data of a wind farm based on a predetermined optimized wind turbine operation state diagnosis index system; calculating a multi-element connection degree expression based on the real-time operation data of the wind farm to obtain a connection value of each object in the index system object layer and the wind turbine operation state; when the wind turbine operation state is normal, taking the wind turbine operation state as the diagnosis result; otherwise, taking the connection value of each monitoring object as the diagnosis result, and determining a corresponding operation and maintenance scheme based on the connection value of each monitoring object; wherein the index system determines a target layer with the wind turbine operation state as the target, constructs a plurality of monitoring objects as the object layer for the target layer, and performs dimension reduction processing by using a Self-representation unsupervised feature selection method.

2. The method of claim 1, wherein, The construction of the index system comprises the following steps: determining a target layer with the wind turbine operation state as the target; determining a plurality of monitoring objects as the object layer based on historical operation data provided by a plurality of data sources of the wind farm and fault conditions of the wind turbine; determining a plurality of indexes for each object in the object layer to form an index layer; performing dimension reduction processing on each index in the index layer based on a Self-representation unsupervised feature selection method; obtaining a multi-level index system of the wind turbine operation state based on the target layer, the object layer, and the index layer after the dimension reduction processing.

3. The method of claim 2, wherein, The monitoring objects at least include one or more of the following: a generator, a gearbox, a nacelle, a main bearing, a converter, a variable pitch, a power grid environment, and a natural environment.

4. The method of claim 2, wherein, The dimension reduction processing on each index in the index layer based on the Self-representation unsupervised feature selection method comprises the following steps: Constructing a feature quantity matrix based on each index in the index layer ; The characteristic matrix is substituted into a previously split unconstrained optimization equation to solve, obtaining the characteristic matrix The characteristic matrix is substituted into a previously split unconstrained optimization equation to solve, obtaining the characteristic matrix The first several indexes with the largest characteristic in the characteristic matrix wherein the optimization equation is determined based on the Self-representation unsupervised feature selection method through splitting.

5. The method of claim 4, wherein, The determination of the split unconstrained optimization equation comprises the following steps: determining an optimization equation for index feature selection by using a Self-representation method; solving the optimization equation by using a split Bregman method to obtain a split unconstrained optimization equation.

6. The method of claim 5, wherein, The calculation formula of the unconstrained optimization equation is as follows: where: is the auxiliary matrix variable; is the self-representation error matrix variable; F is the gradient at is the self-representation matrix variable; , is the regularization parameter; is the Transformed L 1 norm; is the penalty parameter; is the auxiliary variable; is the number of iterations; is the L 1,2-2 norm; denotes the unconstrained optimization equation; is the 1 norm of L is the eigenvalue matrix.​​ 7. The method of claim 6, wherein, The calculation formula of the unconstrained optimization equation is as follows: Based on auxiliary variables The update rule is to use an alternating iterative method to solve for each parameter in the unconstrained optimization equation, and obtain the auxiliary matrix respectively. Self-representation error matrix Self-representation matrix and in gradient at The value at the current iteration number; based on the non-negativity of the self-representation matrix performing a projection operator equation to update the self-representation matrix ​ Based on the auxiliary matrix Self-representation error matrix ,exist gradient at F and the updated self-representation matrix The value of the auxiliary variable at the current iteration number Update the algorithm and continue iterative calculations until the convergence condition is met; After the iteration is completed, based on the feature matrix , each index is sorted, and a number of largest indexes are selected; wherein the self-representation matrix is the L2 norm of the first p row.

8. The method of claim 7, wherein, The updating rule of the auxiliary variable when the iteration is +1 is as follows: The auxiliary matrix variable At The calculation formula at the +1 iteration is as follows: the self-representation error matrix variable In The calculation formula at the +1 iteration is as follows: The in Gradient at F At The calculation formula at +1 iteration is as follows: The self-representing matrix variable At The calculation formula at the +1 iteration is as follows: The projection operator equation is as follows: wherein , represent the first p row and the first q column, respectively; represents a diagonal matrix.

9. The method of claim 2, wherein, After obtaining the multi-level index system of the wind turbine operation state, the following steps are further included: determining a constant weight coefficient of each index in the index layer according to statistical data and expert opinions; calculating a variable weight coefficient of each index and a variable weight coefficient of each monitoring object based on the constant weight coefficient.

10. The method of claim 9, wherein, The calculation formula of the variable weight coefficient of each index is as follows: In the formula, is a variable weight coefficient of the evaluation index under the monitoring object; is a constant weight coefficient of the evaluation index under the monitoring object; is a normalized value of the evaluation index under the monitoring object; is a variable weight coefficient of the evaluation index under the monitoring object; is a constant weight coefficient of the evaluation index under the monitoring object; is a normalized value of the evaluation index under the monitoring object; is a variable weight coefficient of the evaluation index under the monitoring object; is a constant weight coefficient of the evaluation index under the monitoring object; is a normalized value of the evaluation index under the monitoring object; is a variable weight coefficient of the evaluation index under the monitoring object; The calculation formula of the variable weight coefficient of each monitoring object is as follows: wherein is a variable weight coefficient for the monitoring object ; is a constant weight coefficient for the monitoring object ; is a score value for the monitoring object , and .

11. The method of claim 2, wherein, The method for obtaining real-time operation data of a wind farm based on a predetermined optimized wind turbine operation state diagnosis index system comprises the following steps: The real-time operation data of the wind farm is obtained from operation data provided by multiple data sources of the wind farm on the big data platform based on the index layer of the index system.

12. The method of claim 2, wherein, The construction of the multi-element contact degree expression includes: The multi-element contact degree expression is determined based on the data set by using a set pair analysis method. The multi-element contact degree expression includes a same-different-contrary evaluation matrix and a same-different-contrary coefficient matrix. The multi-element contact degree expression of the object layer is as follows:

13. The method of claim 12, wherein, The expression of the same-different-contrary coefficient matrix is as follows: In the formula, monitoring object the multi-element contact degree; monitoring object The variable weight coefficient matrix corresponding to the index layer set; monitoring object The same and opposite evaluation matrix corresponding to the index layer set; The same and opposite coefficient matrix; The monitoring object The expression of the same-different-reverse evaluation matrix corresponding to the index layer set is as follows: ； In the formula: is the first index of the subject to be monitored is the second index of the subject to be monitored is the third index of the subject to be monitored The monitoring object The expression of the variable weight coefficient matrix corresponding to the index layer set is as follows: In the formula: is a variable weight coefficient for monitoring the first index of the object .​ The expression is as follows: In the formula: is the difference degree coefficient of the wind turbine operating state level being the first is the difference degree coefficient of the wind turbine operating state level being the second is the opposition degree coefficient.

14. The method of claim 13, wherein, The multi-element contact degree expression of the wind turbine operation state layer is as follows: In the formula, is the membership function value of the wind turbine state level ; is the evaluation index normalized value.

15. The method of claim 12, wherein, The expression of the same-different-contrary evaluation matrix of the monitoring object set under the wind turbine operation state layer is as follows: In the formula, is the multi-element contact degree of the wind turbine operating state layer; is the variable weight coefficient matrix of the monitoring object set under the wind turbine operating state layer; is the same, different, and opposite evaluation matrix of the monitoring object set under the wind turbine operating state layer; is the same, different, and opposite coefficient matrix; The expression of the variable weight coefficient matrix of the monitoring object set under the wind turbine operation state layer is as follows: ； In the formula, monitoring objects ; monitoring objects ; The contact values of the wind turbine operation state and each object of the object layer of the index system are obtained by calculating the pre-constructed multi-element contact degree expression based on the real-time operation data of the wind farm, including: In the formula: to monitor the subject is a variable weight coefficient.

16. The method of claim 12, wherein, The same-different-contrary coefficient matrix is obtained by using a uniform distribution method based on the pre-determined wind turbine state level; The same-different-contrary evaluation matrix of each index under each monitoring object layer is determined based on the real-time operation data of the wind farm; The contact values of each object are obtained by weighted summation based on the same-different-contrary evaluation matrix and the same-different-contrary coefficient matrix of each index and the variable weight coefficient corresponding to the index; The same-different-contrary evaluation matrix of each monitoring object is determined based on the index values of each monitoring object; The contact values of the target layer are obtained by weighted summation based on the same-different-contrary evaluation matrix and the same-different-contrary coefficient matrix of each monitoring object and the variable weight coefficient corresponding to the monitoring object. The data source at least includes one or more of the following: a wind farm second-level data SCADA system, a unit main control, a converter, an AGC / AVC, a primary frequency control device, and an energy management platform.

17. The method of claim 2, wherein, The system includes:

18. A wind turbine generator operating condition intelligent diagnosis system, characterized in that, A data acquisition layer is configured to obtain real-time operation data of a wind farm based on a pre-determined optimized wind turbine operation state diagnosis index system; A data processing layer is configured to calculate contact values of a wind turbine operation state and each object of an object layer of the index system by calculating a pre-constructed multi-element contact degree expression based on the real-time operation data of the wind farm; When the wind turbine operation state is normal, the wind turbine operation state is taken as a diagnosis result; otherwise, the contact values of each monitoring object layer are taken as a diagnosis result, and a corresponding operation and maintenance scheme is determined based on the contact values of each monitoring object. The index system determines a target layer with the wind turbine operation state as a target, constructs multiple monitoring objects as an object layer for the target layer, and performs dimension reduction processing by using a Self-representation unsupervised feature selection method. The system further includes a data storage layer.

19. The system of claim 18, wherein, ​ The data storage layer comprises a big data platform and an algorithm library; The algorithm library comprises a dimension reduction algorithm and a state diagnosis algorithm in a wind turbine operation state diagnosis index system; The big data platform is configured to store real-time operation data of a wind farm provided by the collection layer based on the big data platform.

20. The system of claim 19, wherein, The state diagnosis algorithm comprises at least one or more of the following: a multivariate connection degree expression of the object layer, an expression of a same-different-opposite evaluation matrix of the first monitoring object, an expression of a weight coefficient matrix of a corresponding index layer set of the first monitoring object, an expression of a same-different-opposite coefficient matrix, a multivariate connection degree expression of the wind turbine operation state layer, an expression of a same-different-opposite evaluation matrix of the wind turbine operation state layer, and an expression of a weight coefficient matrix of a corresponding index layer set of the first monitoring object.

Images