Fault Data Feature Extraction Method and Device for Wind Power Grid-Connected Inverters

Abstract

The present invention provides a method and device for extracting fault data characteristics of a wind power grid-connected inverter, belonging to the technical field of photovoltaic power generation. The method of the present invention includes the following steps: analyzing the faults of its inverter based on a wind power generation system to obtain fault data; performing variational mode decomposition on the fault data to decompose the fault data into different modes to obtain IMFs; solving the entropy of the IMFs to obtain multi-scale fuzzy entropy, which is used as an eigenvalue. The method of the present invention realizes the extraction of characteristics of open-circuit faults of a wind power grid-connected inverter, and the method of the present invention is also applicable to the processing of open-circuit fault data of a wind power grid-connected inverter with added white noise. Compared with the traditional VMD, the method of extracting fault eigenvalues by variational mode decomposition mentioned in the present invention takes into account the optimization algorithm to optimize the kernel function parameters and penalty parameters, which is beneficial to extracting effective data characteristics of faults and normal conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power generation, and particularly relates to a method and device for extracting fault data characteristics of a wind power grid-connected inverter. Background Art

[0002] A wind power generation system (WPGS) can serve as a sustainable solution to mitigate the impact of greenhouse gas emissions and meet the energy demands of a growing economy. In addition, the construction period of a wind farm is short, and the investment and construction scale have great flexibility. Therefore, wind energy has become one of the fastest-growing renewable energy sources in the past decade.

[0003] However, the large-scale development of wind power systems has brought two important problems: 1) the reliability of wind power systems; 2) the maintenance cost of wind power systems. In the development of wind power generation systems, not only the functions of wind power generation systems need to be considered, but also the reliability of wind power generation systems, so as to promote the safe operation of wind power generation systems, maximize the utilization of wind power, and significantly increase the penetration rate of wind power generation systems. On the other hand, the maintenance cost should be reduced. For a wind turbine with a service life of 20 years, it is estimated that the maintenance cost accounts for at least 10-15% of the total cost. Therefore, it is necessary to study effective methods to improve the reliability of wind power systems and reduce the maintenance cost.

[0004] Since the occurrence of converter faults in wind power systems is independent and random, and fault diagnosis is a feasible method for rapid fault detection and isolation, the engineering value of wind power converter fault diagnosis is as follows: (1) During the development process, fault diagnosis can provide a reliable theoretical basis for improving the design of converters, so as to adjust the structure of converters and optimize converter controllers. For example, higher fault diagnosis accuracy can ensure effective fault tolerance control and improve the continuous operation ability of wind turbines under fault conditions. (2) For operation, fault diagnosis can guide operators to take active and effective adjustment measures in the early stage of converter faults, minimize the occurrence of catastrophic faults, effectively avoid unnecessary shutdowns, and reduce the economic losses caused by faults. (3) In the maintenance management of wind power systems, fault diagnosis can quickly detect and identify faults; this is of great significance for making earlier decisions, reducing risks, and more effectively reducing costs.

[0005] Therefore, the present invention proposes a method and device for extracting fault data characteristics of a wind power grid-connected inverter. Summary of the Invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art, and provides a method and device for extracting fault data characteristics of a wind power grid-connected inverter.

[0007] On one hand, the present invention provides a method for extracting fault data features of a wind power grid-connected inverter, including the following steps:

[0008] Analyze the faults of the inverter based on the wind power generation system to obtain fault data;

[0009] Perform variational mode decomposition on the fault data, decompose the fault data into different modes to obtain IMFs;

[0010] Solve the entropy of the IMFs to obtain multi-scale fuzzy entropy, which is used as the eigenvalue.

[0011] Optionally, performing variational mode decomposition on the fault data, decomposing the fault data into different modes to obtain IMFs includes:

[0012] Perform variational mode decomposition on the fault data to decompose the fault data into fault data and normal data; and

[0013] When performing variational mode decomposition on the fault data, select an improved harmony search algorithm to optimize the quadratic penalty factor and the number of modal components of the variational mode decomposition, and the fitness function is the minimum envelope entropy.

[0014] Optionally, performing variational mode decomposition on the fault data includes:

[0015] Decompose the original signal p according to the preset number of modal components to obtain the intrinsic mode function, specifically as follows:

[0016]

[0017] In the formula, {u} is the K modal components obtained by decomposition, {u} = {u 1 , …, u k}; {ω} is the center frequency of each mode, {ω} = {ω 1 , …, ω k}; k = 1, 2, 3, ..., K; t represents time, ω k represents the frequency of the signal u k (t), represents taking the partial derivative, δ(t) represents the Dirac distribution function, and p represents the original signal;

[0018] Use the augmented Lagrangian function to process the variational constraint model and convert it into a saddle point solution problem of an unconstrained model:

[0019]

[0020] Among them, λ(t) is the Lagrange multiplier and α is the quadratic penalty factor.

[0021] Optionally, the improved harmony search algorithm is to add a Cauchy mutation operator to the harmony search algorithm, and the updated formula for obtaining the new harmony position is expressed as follows:

[0022]

[0023] In the formula: represents the spatial position of the j-th individual in the new population, Cauchy represents a random number of the standard Cauchy distribution, and its value is tan(π·(rand - 0.5)).

[0024] Optionally, the harmony search algorithm includes:

[0025] Generate N harmonies in the search space of the HS algorithm and store them in HM;

[0026] Adopt a pitch adjustment mechanism for the improvisation of new harmonies;

[0027] Compare the worst harmony stored in HM with the new harmony and replace the worst harmony;

[0028] When the predefined maximum number of iterations is satisfied, take the most pleasant harmony stored in HM as the optimal solution.

[0029] Optionally, the mathematical form of HM is as follows:

[0030] Harmony

[0031] where, ff 1 is the fitness function corresponding to harmony 1, ff 2 is the fitness function corresponding to harmony 2, ff N is the fitness function corresponding to harmony N; H i represents the i-th harmony in the population, represents the first position of the i-th harmony in the spatial dimension, and so on; and / or,

[0032] For the initialization of HM, the following formula can be used:

[0033]

[0034] where, u j and l j are the upper and lower bounds; and rand is a random number between 0 and 1.

[0035] Optionally, the improvisation of new harmonies using the pitch adjustment mechanism includes:

[0036] If the random number rand ≤ PAR parameter, the improvised note is transferred to an adjacent value using the following formula, specifically as follows:

[0037]

[0038] Wherein, represents the spatial position of the j-th individual in the new population, BW is the bandwidth, rand is a random number between (0,1), and |u j -l j | term controls the scale of the decision variable, u j is the upper bound of the spatial position of the j-th individual, l j is the upper bound of the spatial position of the j-th individual;

[0039] If rand > PAR, the improvised note remains unchanged; and / or,

[0040] Compare the worst harmony stored in HM with the new harmony and replace the worst harmony, including:

[0041] After calculating through the fitness function, when the harmony h is in 1 < h < N, it is determined as the worst harmony stored in HM, and if ff h is worse than ff new fitness, the new harmony will replace the harmony h in HM; otherwise, the new harmony is removed.

[0042] Optionally, the envelope entropy calculation formula is as follows:

[0043]

[0044] Wherein, E P为 is the envelope entropy, where a(j) is the envelope signal after Hilbert demodulation of the k modal components decomposed by VMD, p j is the probability distribution sequence obtained by calculating the normalization of a(j), and N is the number of sampling points.

[0045] Optionally, solving the entropy of the IMF to obtain the multi-scale fuzzy entropy and using it as an eigenvalue includes:

[0046] Consider performing coarse-grained processing on the time series X = {x(t), t = 1, 2,..., N'} composed of IMF to obtain the coarse-grained sequence y (τ) ={y d (τ) , 1 ≤ d ≤ N' / τ}, specifically as follows:

[0047]

[0048] In the formula: x(t) represents the IMF quantity at time t, y d(τ) represents the coarse-grained quantity at time t, where τ is the scale factor, generally a positive integer;

[0049] Calculate the fuzzy entropy of the coarse-grained sequence under each scale factor, and its calculation formula is:

[0050] MFE(X,τ,m,r) = FE(y (τ) ,m,r)

[0051] In the formula: MFE represents generating multi-scale fuzzy entropy, FE represents fuzzy entropy, m is the embedding dimension, r is the similarity tolerance, X is the multi-scale fuzzy entropy, and τ is the scale factor.

[0052] On the other hand, the present invention provides a device for extracting fault data characteristics of a wind power grid-connected inverter, including: a data acquisition module, a data decomposition module, and a feature value extraction module; wherein,

[0053] The data acquisition module is used to analyze the faults of its inverter based on the wind power generation system to obtain fault data;

[0054] The data decomposition module is used to perform variational mode decomposition on the fault data, decompose the fault data into different modes to obtain IMF;

[0055] The feature value extraction module is used to solve the entropy of the IMF to obtain multi-scale fuzzy entropy and use it as a feature value.

[0056] The present invention provides a method for extracting fault data characteristics of a wind power grid-connected inverter, including the following steps: analyzing the faults of its inverter based on the wind power generation system to obtain fault data; performing variational mode decomposition on the fault data, decomposing the fault data into different modes to obtain IMF; solving the entropy of the IMF to obtain multi-scale fuzzy entropy and using it as a feature value. The method of the present invention realizes the extraction of characteristics of open-circuit faults of wind power grid-connected inverters, and the method of the present invention is also applicable to processing open-circuit fault data of wind power grid-connected inverters with added white noise. The method of extracting fault feature values by variational mode decomposition mentioned in the present invention, compared with the traditional VMD, takes into account the optimization algorithm to optimize the kernel function parameters and penalty parameters, which is beneficial to extracting effective data characteristics of faults and normal conditions. Description of the Drawings

[0057] Figure 1 is a flowchart of a method for extracting fault data characteristics of a wind power grid-connected inverter according to an embodiment of the present invention;

[0058] Figure 2 is a schematic diagram of the topological structure of a doubly-fed wind power generation system according to another embodiment of the present invention;

[0059] Figure 3 Schematic diagram of a device for extracting fault data characteristics of a wind power grid-connected inverter according to another embodiment of the present invention. Detailed implementation manners

[0060] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0061] As Figure 1 shown, on one hand, the present invention provides a method S100 for extracting fault data characteristics of a wind power grid-connected inverter, including the following specific steps S110 to S130:

[0062] S110. Analyze the faults of the inverter based on the wind power generation system to obtain fault data.

[0063] It should be noted that the doubly-fed induction generator with a variable-speed constant-frequency wind turbine in this embodiment is the mainstream of the wind power generation system. Due to the characteristics of the DFIG such as reactive power regulation, active power regulation, constant-frequency variable-speed, etc., it is easy to be connected to the grid and has been widely used in many occasions. The topological structure of the DFIG wind power generation system is as Figure 2 shown.

[0064] As Figure 2 shown, the doubly-fed wind power generation system consists of a blade, a gearbox, a generator, a control system, a power converter, and a filter. The wind turbine converts wind energy into mechanical energy, drives the generator to rotate, converts the mechanical energy into electrical energy, and generates alternating current (variable amplitude and variable frequency); then the current is converted into constant-frequency alternating current through the power converter, and is sent to the grid by the transformer. The power converter usually adopts a converter combination with a back-to-back structure. The converter close to the rotor side is called the rotor-side converter, and the converter close to the grid side is called the grid-side converter. The rotor-side converter tracks the maximum wind energy, realizes constant-frequency variable-speed, and improves the operation efficiency of the power generation system. The grid-side converter keeps the DC-side voltage constant and prevents grid-side current harmonics. The two ends of this converter have the same structure as the space vector pulse width modulation (SVPWM) strategy. Taking the grid-side converter as an example, this converter includes three-phase bridge arms, and each bridge arm is composed of two power switches insulated gate bipolar transistors (IGBTs, denoted as Ti) and anti-parallel connected diodes (called reverse free-wheel diodes). The power switch Ti is controlled by a gate signal gi: when the gate signal gi is equal to 1, the switch is turned on. When the gate signal gi is equal to 0, the switch is turned off. The two IGBT switches on each bridge arm work in a complementary mode; only one IGBT switch is turned on to prevent the short circuit of the DC bus voltage.

[0065] Furthermore, the faults of the power converter can be divided into short - circuit (SC) faults and open - circuit (OC) faults. SC faults usually generate very large currents, triggering system protection (such as fuses, circuit breakers), and are easily detected. The initial SC faults will eventually transform into OC faults. OC faults do not generate excessive currents and do not cause system protection. Therefore, they are usually difficult to detect. Inverter OC faults do not immediately cause the system to collapse, but if the system is in the OC fault state for a long time, it is easy to cause secondary faults in other components and may damage the inverter. Therefore, this embodiment proposes a method for extracting wind power grid - connected inverter data.

[0066] S120. Perform variational mode decomposition on the fault data, decompose the fault data into different modes to obtain IMFs.

[0067] Specifically, in this embodiment, performing variational mode decomposition on the fault data and decomposing the fault data into different modes to obtain IMFs includes: performing variational mode decomposition on the fault data to decompose the fault data into fault data and normal data; and when performing variational mode decomposition on the fault data, an improved harmony search algorithm is used to optimize the quadratic penalty factor and the number of modal components of the variational mode decomposition, and the fitness function is the minimum envelope entropy.

[0068] It should be noted that the improved harmony search algorithm in this embodiment is adding a Cauchy mutation operator to the harmony search algorithm.

[0069] Specifically, the harmony search algorithm (HS) in this embodiment was proposed by Geem et al. in 2001. The process of musical improvisation is the basis for the design of the HS algorithm. One of the purposes of musicians in the process of improvisation is to seek pleasant harmony. Due to the similarity between the optimization process and the process of musical improvisation, the HS algorithm was created.

[0070] In the HS algorithm, each harmony is called a possible solution, and each decision variable of this harmony corresponds to a note. The algorithm consists of a harmony memory (HM), which stores a certain number of harmonies (N). If the goal of the optimization process is to maximize or minimize the fitness function (ff), the optimization problem is defined according to the number of decision variables as follows:

[0071] Minimize(Maximize)ff(x 1 ,x 2 ,...,x d )

[0072] Where ff and x i(i = 1, 2, ..., d) are the fitness function and decision variables. The steps of the algorithm are described below. Generally, the description of the harmony search algorithm is divided into five steps:

[0073] 1. Generate N harmonies in the search space of the HS algorithm and store them in HM. The following formula represents the mathematical form of HM:

[0074]

[0075] Harmony Similarly, for each harmony, its value ff i (i = 1, 2, ..., N) assigns the value of the fitness function, ff 1 is the fitness function corresponding to harmony 1, ff 2 is the fitness function corresponding to harmony 2, ff N is the fitness function corresponding to harmony N; H i represents the i-th harmony in the population, represents the first position of the i-th harmony in the spatial dimension, and so on. For example, ff 1 is the fitness function corresponding to harmony 1.

[0076] Therefore, for the initialization of HM, the following formula can be used:

[0077]

[0078] where u j and l j are the upper and lower bounds. And rand is a random number between 0 and 1.

[0079] 2. Improvisation of new harmonies

[0080] In this step, a new harmony is improvised x new = [x new.1 , x new.2 , ..., x new.d . Generating new harmonies using existing harmonies is one of the characteristics of the HS algorithm compared to other algorithms such as genetic algorithms.

[0081] The HS algorithm adopts a pitch adjustment mechanism, defined as the pitch adjustment rate parameter, to avoid local optima. This parameter varies between 0 and 1. Small and large values of the PAR parameter weaken and enrich the pitch adjustment mechanism respectively. Below, a random number (rand) uniformly distributed between 0 and 1 is generated to execute the pitch angle adjustment mechanism. If rand ≤ PAR, the following formula is used to transfer the improvised note to an adjacent value, otherwise (rand > PAR) the improvised note remains unchanged.

[0082]

[0083] Among them, represents the spatial position of the j-th individual in the new population, BW is the bandwidth, rand is a random number between (0, 1), and the term |u j -l j | controls the scale of the decision variable, u j is the upper bound of the spatial position of the j-th individual, and l j is the upper bound of the spatial position of the j-th individual. Then, calculate the fitness function (ff new ) of the new harmony.

[0084] 3. Replacement

[0085] Compare the worst harmony stored in HM with the new harmony. For example, assume that after calculating the fitness function, the harmony h (1 < h < N) is determined to be the worst harmony stored in HM. In this case, if ff h is worse than ff new in terms of fitness, a new harmony will replace the harmony h in HM. Otherwise, the new harmony is removed.

[0086] 4. Stopping Criterion

[0087] Generally, the predefined maximum number of iterations is the criterion for the meta-heuristic algorithm to stop. When this criterion is met, the algorithm proceeds to step 5. Otherwise, repeat steps 2 and 3.

[0088] 5. Final Result

[0089] Finally, select the most pleasant harmony stored in HM, that is, the fitness optimal solution, as the optimal solution.

[0090] Furthermore, due to the defect that the harmony search algorithm itself is prone to falling into local optimal solutions, an improvement strategy is added, a Cauchy mutation operator is added to increase the probability of jumping out of local optimal solutions.

[0091] The probability density function formula of the one-dimensional standard Cauchy distribution is expressed as:

[0092]

[0093] The Cauchy distribution has a large probability density at the origin, with a compact distribution, while the distributions at both ends are long and have small densities. It has a very long tail, which enables an individual to have a greater probability of jumping to a better position and escaping from the local optimum, and has the characteristic of a relatively high probability at both wings. Therefore, it can generate random numbers that are far from the origin, and the distribution range of the random numbers generated by Cauchy mutation is wider than that of Gaussian mutation. This means that the harmony after adopting the Cauchy mutation factor can effectively increase the diversity of the harmony population during the operation of the algorithm. In addition, the peak value of the Cauchy distribution is relatively low, which can shorten the time for the mutated harmony to search around the neighborhood, help the harmony quickly jump out of the local extreme value, and enable the algorithm to obtain better performance. By introducing the Cauchy mutation factor, the updated formula for obtaining the new harmony position is expressed as follows:

[0094]

[0095] In the formula: represents the spatial position of the j-th individual in the new population, and Cauchy represents a random number of the standard Cauchy distribution, whose value is tan(π·(rand - 0.5)).

[0096] Furthermore, Variational Mode Decomposition (VMD) is an adaptive signal processing method. Based on EMD and Local Mean Decomposition (LMD), it effectively reduces the mode mixing phenomenon and has a good theoretical basis. The overall framework of VMD is a variational problem. According to the preset number of mode components, the original signal p is decomposed to obtain the intrinsic mode functions, as shown in the following formula:

[0097]

[0098] In the formula, {u} is the K mode components obtained by decomposition, {u} = {u 1 , …, u k}; {ω} is the central frequency of each mode, {ω} = {ω 1 , …, ω k}; k = 1, 2, 3, …, K; t represents time, and ω k represents the frequency of the signal u k (t), represents taking the partial derivative, δ(t) represents the Dirac distribution function, and p represents the original signal;

[0099] By introducing the Lagrange multiplier λ(t) and the quadratic penalty factor α, the constrained variational problem is transformed into an unconstrained variational problem, that is, the extended Lagrangian function is used to handle the variational constraint model, and it is converted into a saddle point solution problem of an unconstrained model. Among them, α ensures the reconstruction accuracy of the signal in the presence of Gaussian noise, and λ(t) makes the solution of the variational problem maintain strict constraints. The extended Lagrangian expression is as follows:

[0100]

[0101] Among them, λ(t) is the Lagrange multiplier and α is the quadratic penalty factor.

[0102] The multiplicative operator alternating direction method is used to solve the above formula, and the "saddle point" of the extended Lagrangian is sought by alternately updating and The new and are updated as:

[0103]

[0104]

[0105]

[0106] Among them: is equivalent to the Wiener filter of the current residual; is the centroid of the power spectrum of the current mode function; performing the inverse Fourier transform on , and the real part of it is {u (ω)} k (ω)}

[0107] Furthermore, the improved harmony search algorithm is used to optimize the quadratic penalty factor and the number of mode components. The fitness function is the minimum envelope entropy. The envelope entropy represents the sparse characteristics of the original signal. When there is more noise and less characteristic information in the IMF, the envelope entropy value is larger; conversely, the envelope entropy value is smaller.

[0108] The calculation formula of the envelope entropy is as follows:

[0109]

[0110] Among them, E P为 is the envelope entropy. In the formula, a(j) is the envelope signal after Hilbert demodulation of the k mode components decomposed by VMD, and p j is the probability distribution sequence obtained by calculating the normalization of a(j). N is the number of sampling points, and the entropy value of the distribution sequence p j is the envelope entropy E P .

[0111] S130. Solve the entropy of the IMF to obtain the multi-scale fuzzy entropy, and use it as an eigenvalue.

[0112] Specifically, the multi-scale entropy (MSE) refers to the sample entropy at different scales. It measures the complexity of the time series from different scales, overcomes the defects of the traditional single-scale sample entropy, and can reflect the deeper pattern information of the time series. The fuzzy entropy (FE) uses an exponential function instead of a unit step function, overcomes the mutation of similarity measurement, and can better highlight the differences between signals. Therefore, replacing the sample entropy with the fuzzy entropy, the multi-scale fuzzy entropy is obtained, and its calculation steps are as follows:

[0113] 1) Consider performing coarse-graining processing on the time series X = {x(t), t = 1, 2,..., N'} composed of the IMF to obtain the coarse-grained sequence y (τ) ={y d (τ) , 1 ≤ d ≤ N' / τ}, specifically as follows:

[0114]

[0115] where: x(t) represents the IMF quantity at time t, y d (τ) represents the coarse-grained quantity at time t, τ is the scale factor, generally a positive integer;

[0116] 2) Calculate the fuzzy entropy of the coarse-grained sequence at each scale factor, and its calculation formula is:

[0117] MFE(X, τ, m, r) = FE(y (τ) , m, r)

[0118] where: MFE represents generating the multi-scale fuzzy entropy, FE represents the fuzzy entropy, m is the embedding dimension, r is the similarity tolerance, X is the multi-scale fuzzy entropy, and τ is the scale factor.

[0119] In this embodiment, the variational mode decomposition and the multi-scale fuzzy entropy are combined to more scientifically extract the fault eigenvalue. The proposed multi-fault feature extraction method for wind power grid-connected inverters by combining the optimization algorithm with the variational mode decomposition can effectively extract the eigenvalues of the fault data with noisy signals. Moreover, the method proposed in this embodiment is also applicable to processing the open-circuit fault data of wind power grid-connected inverters with added white noise.

[0120] Such as Figure 3As shown in the figure, on the other hand, the present invention provides a device 200 for extracting fault data characteristics of a wind power grid-connected inverter, including: a data acquisition module 210, a data decomposition module 220, and a feature value extraction module 230; wherein, the data acquisition module 210 is used to analyze the faults of its inverter based on the wind power generation system to obtain fault data; the data decomposition module 220 is used to perform variational mode decomposition on the fault data, decompose the fault data into different modes to obtain IMF; the feature value extraction module 230 is used to solve the entropy of the IMF to obtain multi-scale fuzzy entropy and use it as a feature value.

[0121] It should be noted that the device in this embodiment refers to the foregoing description for the method of extracting fault data characteristics of the wind power grid-connected inverter, and will not be elaborated herein.

[0122] The following will illustrate the method for extracting fault data characteristics of the wind power grid-connected inverter with specific embodiments:

[0123] Build a simulation of the grid-connected inverter of the wind power generation system based on the simulink platform of MATLAB, set an open-circuit fault for each wind power grid-connected inverter, extract 1000 groups of data for each type of open-circuit fault, randomly extract 200 groups from the 1000 groups of fault data and add a noise signal, and combine the data with the noise signal and the data without the noise signal as fault data.

[0124] 1. Perform variational mode decomposition on the fault data of each type, and decompose the fault data and normal data into different modes.

[0125] 2. Among them, when decomposing the data into modal components, in order to preserve the feature information as much as possible, optimize the quadratic penalty factor and the number of modal components of the variational mode decomposition. When optimizing, use the improved harmony search algorithm, and select the minimum envelope entropy as the fitness function.

[0126] 3. In the improved harmony search algorithm, initialize the relevant parameters.

[0127] 4. Generate N harmonies in the search space of the HS algorithm and store them in HM.

[0128]

[0129] Harmony Similarly, for each harmony, its value ff i (i = 1, 2,..., N) is used to specify the fitness function. For example, ff 1 is the fitness function corresponding to harmony 1. Initialize HM:

[0130]

[0131] where u j and l j are the upper and lower bounds. Also, rand is a random number between 0 and 1.

[0132] 5. After updating the harmony, introduce the Cauchy mutation factor to update the harmony position again; by introducing the Cauchy mutation factor, the update formula for obtaining the new harmony position is expressed as follows:

[0133]

[0134] In the formula: Cauchy represents a random number of the standard Cauchy distribution, and its value is tan(π·(rand - 0.5)).

[0135] 6. The HS algorithm updates the harmony using the pitch adjustment mechanism; a random number (rand) uniformly distributed between 0 and 1 is generated to execute the pitch angle adjustment mechanism. If rand ≤ PAR, the following formula is used to transfer the improvised note to an adjacent value, otherwise (rand > PAR) the improvised note remains unchanged.

[0136]

[0137] where BW is the bandwidth. The |u j - l j | term controls the scale of the decision variable. Then, calculate the fitness function (ff new ) of the new harmony.

[0138] 7. Compare the worst harmony stored in HM with the new harmony, and select the one with better fitness as the next-generation harmony population.

[0139] 8. When the maximum number of iterations is reached, output the obtained optimal quadratic penalty factor and the number of modal components, otherwise return to step 5 and loop again.

[0140] 9. For each type of fault data, after VMD processing with the obtained optimal quadratic penalty factor and the number of modal components, obtain the IMF, and solve the entropy of the IMF as the eigenvalue of the fault and normal data.

[0141] 10. When solving the entropy, consider performing coarse-graining processing on the time series X = {x(t), t = 1, 2,..., N′} composed of the IMF to obtain the coarse-grained sequence y (τ) = {y d (τ) , 1 ≤ d ≤ N′ / τ}, that is

[0142]

[0143] Based on the above coarse-grained sequence, calculate the fuzzy entropy of the coarse-grained sequence at each scale factor, and its calculation formula is

[0144] MFE(X, τ, m, r) = FE(y (τ) , m, r)

[0145] where: m is the embedding dimension; r is the similarity tolerance.

[0146] 11. The obtained fuzzy entropy is used as the eigenvalue.

[0147] The present invention provides a method and device for extracting fault data features of a wind power grid-connected inverter, which has the following beneficial effects compared with the prior art:

[0148] First, the present invention can provide a method for dealing with open-circuit faults of a wind power grid-connected inverter and a method for extracting fault features, providing a new idea for diagnosing and processing fault data of grid-connected inverters in a wind farm.

[0149] Second, the method for extracting fault eigenvalues by variational mode decomposition mentioned in the present invention, compared with the traditional VMD, takes into account the optimization algorithm to optimize the kernel function parameters and penalty parameters, which is beneficial to extracting effective data features of faults and normal conditions.

[0150] It can be understood that the above embodiments are only exemplary embodiments adopted to illustrate the principle of the present invention, and the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also regarded as the protection scope of the present invention.

Claims

1. A method for extracting fault data characteristics of a wind power grid-connected inverter, characterized in that, it includes the following steps: Analyze the faults of the inverter based on the wind power generation system to obtain fault data; Perform variational mode decomposition on the fault data, decompose the fault data into different modes to obtain IMFs, including: performing variational mode decomposition on the fault data, decompose the fault data into different modes to obtain IMFs, including: Perform variational mode decomposition on the fault data to decompose the fault data into fault data and normal data; and, When performing variational mode decomposition on the fault data, an improved harmony search algorithm is used to optimize the quadratic penalty factor and the number of modal components of the variational mode decomposition, and the fitness function is the minimum envelope entropy; The improved harmony search algorithm is to add a Cauchy mutation operator to the harmony search algorithm, and the update formula for obtaining the new harmony position is expressed as follows: In the formula: represents the spatial position of the j-th individual in the new population, and Cauchy represents a random number of the standard Cauchy distribution, and its value is tan(π·(rand - 0.5)); Solve the entropy of the IMFs to obtain multi-scale fuzzy entropy, and use it as the eigenvalue.

2. The method according to claim 1, characterized in that, Performing variational mode decomposition on the fault data includes: Decompose the original signal p according to the preset number of modal components to obtain the intrinsic mode function, specifically as follows: where {u} are the K modal components obtained by decomposition, {u} = {u 1 , …, u k}; {ω} are the central frequencies of each mode, {ω} = {ω 1 , …, ω k}; k = 1, 2, 3, ..., K; t represents time, ω k represents the frequency of the signal u k (t), represents taking the partial derivative, δ(t) represents the Dirac distribution function, and p represents the original signal; Use the augmented Lagrangian function to process the variational constraint model and convert it into the saddle point solution of the unconstrained model: where λ(t) is the Lagrange multiplier and α is the quadratic penalty factor.

3. The method according to claim 1, characterized in that, The harmony search algorithm includes: Generate N harmonies in the search space of the HS algorithm and store them in HM; Use the pitch adjustment mechanism to improvise new harmonies; Compare the worst harmony stored in HM with the new harmony and replace the worst harmony; When the predefined maximum number of iterations is met, take the most pleasant harmony stored in HM as the optimal solution.

4. The method according to claim 3, characterized in that, The mathematical form of HM is as follows: Harmony Among them, ff 1 is the fitness function corresponding to harmony 1, ff 2 is the fitness function corresponding to harmony 2, ff N is the fitness function corresponding to harmony N; H i represents the i-th harmony in the population, represents the first position of the i-th harmony in the spatial dimension, and so on; and / or, For the initialization of HM, the following formula can be used: where u j and l j are the upper and lower bounds; and rand is a random number between 0 and 1.

5. The method according to claim 3, characterized in that, The use of the pitch adjustment mechanism to improvise new harmonies includes: If the random number rand ≤ the PAR parameter, use the following formula to transfer the improvised note to the adjacent value, specifically as follows: Among them, represents the spatial position of the j j-th individual in the new population. BW is the bandwidth, rand is a random number between (0, 1), and the term |u j -l j | controls the scale of the decision variable. u j is the upper bound of the spatial position of the j-th individual, and l j is the lower bound of the spatial position of the j-th individual; If rand > PAR, the improvised note remains unchanged; and / or, Compare the worst harmony stored in HM with the new harmony and replace the worst harmony, including: After calculating through the fitness function, when the harmony h is in the range of 1 < h < N, it is determined as the worst harmony stored in HM, and if ff h is worse than ff new in terms of fitness, the new harmony will replace the harmony h in HM; otherwise, the new harmony is removed.

6. The method according to claim 1, characterized in that, The envelope entropy calculation formula is as follows: Among them, E P is the envelope entropy, where a(j) is the envelope signal after Hilbert demodulation of the k modal components decomposed by VMD, and p j is the probability distribution sequence obtained by calculating the normalization of a(j), and N is the number of sampling points.

7. The method according to claim 1, characterized in that, The solving of the entropy of the IMFs to obtain multi-scale fuzzy entropy and using it as the eigenvalue includes: Coarsely granulate the time series X = {x(t), t = 1, 2,..., N'} of the IMF composition to obtain the coarsely granulated sequence y (τ) ={y d (τ) , 1 ≤ d ≤ N' / τ}, specifically as follows: where: \(x(t)\) represents the IMF quantity at time \(t\), \(y\) d (τ) represents the coarse-grained quantity at time \(t\), \(\tau\) is the scale factor, generally a positive integer; Calculate the fuzzy entropy of the coarse-grained sequence under each scale factor, and its calculation formula is: MFE(X,τ,m,r) = FE(y (τ) ,m,r) In the formula: MFE represents generating multi-scale fuzzy entropy, FE represents fuzzy entropy, m is the embedding dimension, r is the similarity tolerance, X is the multi-scale fuzzy entropy, and τ is the scale factor.

8. A device for extracting fault data characteristics of a wind power grid-connected inverter, characterized in that, it includes: Data acquisition module, data decomposition module, eigenvalue extraction module; among them, The data acquisition module is used to analyze the faults of its inverter based on the wind power generation system to obtain fault data; The data decomposition module is used to perform variational mode decomposition on the fault data, decompose the fault data into different modes to obtain IMFs, including: performing variational mode decomposition on the fault data, decomposing the fault data into different modes to obtain IMFs, including: Performing variational mode decomposition on the fault data to decompose the fault data into fault data and normal data; and, When performing variational mode decomposition on the fault data, an improved harmony search algorithm is used to optimize the quadratic penalty factor and the number of modal components of the variational mode decomposition, and the fitness function is the minimum envelope entropy; The improved harmony search algorithm is to add a Cauchy mutation operator to the harmony search algorithm, and the update formula for obtaining the new harmony position is expressed as follows: In the formula: represents the spatial position of the j-th individual in the new population, Cauchy represents a standard Cauchy distribution random number, and its value is tan(π·(rand - 0.5)); The eigenvalue extraction module is used to solve the entropy of the IMF to obtain multi-scale fuzzy entropy and use it as an eigenvalue.

Images