A new energy grid-connected transient weak fault high-frequency identification method and system

Abstract

The application discloses a new energy grid-connected transient weak fault high-frequency identification method and system, relates to the field of novel power system transient signal processing technology, and comprises the following steps: performing real-time tracking on signal energy mutation of a discrete digital signal sequence of three-phase current of new energy grid connection, and taking a transient sequence vector by taking a mutation point as a starting point; performing adaptive variational modal decomposition on the transient sequence vector based on improved composite target function optimization and performing inherent interference avoidance screening to obtain a plurality of effective modal vectors; based on the plurality of effective modal vectors, a multi-dimensional time-frequency high-order feature matrix including a plurality of preset feature dimensions is constructed; the plurality of preset feature dimensions at least include: inputting the multi-dimensional time-frequency high-order feature matrix into a trained fault identification model for safety identification to obtain a new energy grid-connected transient weak fault identification result. The application alleviates the technical problem of low identification accuracy of high-frequency transient characteristic signals in the new energy grid connection in the prior art.

Description

Technical Field

[0001] This invention relates to the field of novel power system transient signal processing technology, and in particular to a high-frequency identification method and system for transient weak faults in new energy grid connection. Background Technology

[0002] With the continuous advancement of new power system construction, the high proportion of distributed photovoltaic and wind power grid connection poses a severe challenge to the stable operation of the power grid. In response, power grids at all levels have successively issued mandatory regulations, explicitly requiring new energy power plants to possess interactive capabilities of "observability, measurability, controllability, and adjustability" (referred to as "four capabilities"). However, at present, the engineering implementation of the "four capabilities" policy is generally limited to the simple integration of hardware-level systems such as communication gateways and remote control units (RTUs). The "measurability" dimension of these conventional devices is limited by standard communication protocols, typically only able to acquire steady-state electrical parameters at the second or cycle level. This monitoring mechanism based on steady-state macroscopic data has significant technical blind spots when dealing with complex, weak transient faults.

[0003] In practical engineering scenarios, the data collection network of new energy power plants extensively uses cross-linked polyethylene (XLPE) cables, and contains a large number of AC / DC side joints. Due to long-term exposure to harsh outdoor environments, coupled with the continuous stress from the high-frequency pulse voltage output of the inverter, early water treeing or electrical treeing degradation of the XLPE cable insulation layer is easily induced. At the same time, line aging or mechanical vibration can easily trigger highly concealed series arcing faults. From a physical mechanism perspective, the partial discharge and series arcing caused by insulation degradation in the early stages of evolution both exhibit extremely weak high-frequency transient nonlinear distortions on the order of microseconds to milliseconds, and do not cause significant changes in the effective value of the macroscopic current of the system.

[0004] Even more challenging is that these weak, high-frequency transient characteristic signals severely overlap in frequency band with the high-frequency background switching noise generated by massive grid-connected inverters during normal operation. Traditional hardware-based "four-capable" systems rely on simple steady-state amplitude over-limit protection logic, which is completely unable to isolate and extract the aforementioned high-frequency distortion features from strong inverter background noise. More seriously, existing transient fault detection methods incorporating artificial intelligence are mostly limited to outputting deterministic classification results, lacking uncertainty assessment of model confidence. When faced with extreme unknown electromagnetic interference, they are highly susceptible to catastrophic maloperation of protection devices, failing to meet the stringent safety requirement of "zero maloperation" for power system relay protection. Summary of the Invention

[0005] To address the aforementioned technical problems in existing technologies, embodiments of the present invention provide a method and system for high-frequency identification of transient weak faults in new energy grid connection. The technical solution is as follows: On the one hand, a high-frequency identification method for transient weak faults in new energy grid connection is provided, which is applied to new energy grid connection. The method includes: real-time tracking of signal energy mutations in the discrete digital signal sequence of the three-phase current of the new energy grid connection, and extracting transient sequence vectors starting from the mutation points; adaptive variational mode decomposition of the transient sequence vectors based on an improved composite objective function to obtain multiple intrinsic mode function vectors; the improved composite objective function includes a composite objective function of envelope entropy and Pearson skewness coefficient; filtering the multiple intrinsic mode function vectors to avoid inherent interference to obtain multiple effective mode vectors; constructing a multidimensional time-frequency high-order feature matrix including multiple preset feature dimensions based on the multiple effective mode vectors; the multiple preset feature dimensions include at least: high-frequency energy ratio, sample entropy, skewness, and kurtosis; inputting the multidimensional time-frequency high-order feature matrix into a trained fault identification model for safety identification to obtain the transient weak fault identification result of the new energy grid connection.

[0006] Optionally, real-time tracking of signal energy mutations is performed on the discrete digital signal sequence of the three-phase current of the new energy grid-connected system, and transient sequence vectors are extracted starting from the mutation points. This includes: calculating the instantaneous energy characteristic value of each sampling point in the discrete digital signal sequence of the three-phase current of the new energy grid-connected system based on the nonlinear discrete Tegel energy operator; when N consecutive w When the instantaneous energy characteristic value of each sampling point is greater than the preset energy threshold, the discrete digital signal sequence is truncated based on an adaptive sliding time window of length L, starting from the mutation point, to obtain the original data column vector; the original data column vector is then standardized to obtain the transient sequence vector.

[0007] Optionally, an adaptive variational mode decomposition is performed on the transient sequence vector based on an improved composite evaluation function to obtain multiple intrinsic mode function vectors. This includes: performing initial mode decomposition on the transient sequence vector to obtain multiple initial intrinsic mode function components; extracting the envelope amplitude sequence of each initial intrinsic mode function component and normalizing it to obtain a probability distribution sequence; calculating the envelope entropy of the probability distribution sequence corresponding to each initial intrinsic mode function component, and calculating the Pearson skewness coefficient of each initial intrinsic mode function component; establishing a two-dimensional continuous optimization space with decomposition level and penalty factor, and establishing an improved composite objective function with the minimization of the envelope entropy and the maximization of the Pearson skewness coefficient as composite objectives; searching for the optimal parameter combination in the two-dimensional continuous optimization space based on the improved composite objective function using an objective optimization algorithm; and performing adaptive variational mode decomposition on the transient sequence vector under the optimal parameter combination to obtain multiple intrinsic mode function vectors.

[0008] Optionally, the improved composite objective function includes:

[0009] In the formula, K is the number of decomposition layers, α is the penalty factor, and w1 and w2 are the balancing weight coefficients. For the improved composite objective function, E k Let P be the envelope entropy. k is the Pearson skewness coefficient.

[0010] Optionally, the multiple intrinsic mode function vectors are screened to avoid inherent interference, resulting in multiple effective mode vectors, including: calculating the center frequency of each intrinsic mode function vector; removing interference modes in the multiple intrinsic mode function vectors whose center frequency is equivalent to the power frequency fundamental wave and the inherent switching frequency of the grid-connected inverter, and retaining the wideband effective components as effective mode vectors.

[0011] Optionally, based on the multiple effective modal vectors, a multidimensional time-frequency high-order feature matrix including multiple preset feature dimensions is constructed, including: calculating multiple preset feature dimensions for each effective modal component respectively; combining and cascading the multiple preset feature dimensions corresponding to each effective modal component to construct a multidimensional time-frequency high-order feature matrix.

[0012] Optionally, the fault identification model includes a multi-scale hybrid attention convolutional network with a Monte Carlo random deactivation mechanism.

[0013] On the other hand, a high-frequency identification system for transient weak faults in new energy grid connection is also provided, used to implement the high-frequency identification method for transient weak faults in new energy grid connection provided in the embodiments of the present invention; the system includes: a mutation tracking module, a mode decomposition module, an interference screening module, a matrix construction module, and a security identification module; wherein, the mutation tracking module is used to track the signal energy mutation of the discrete digital signal sequence of the three-phase current of the new energy grid connection in real time, and to extract the transient sequence vector with the mutation point as the starting point; the mode decomposition module is used to perform adaptive variational mode decomposition on the transient sequence vector based on the optimization of the improved composite objective function to obtain multiple intrinsic mode functions. The improved composite objective function includes a composite objective function of envelope entropy and Pearson skewness coefficient; the interference screening module is used to screen the multiple intrinsic mode function vectors to avoid inherent interference and obtain multiple effective mode vectors; the matrix construction module is used to construct a multi-dimensional time-frequency high-order feature matrix with multiple preset feature dimensions based on the multiple effective mode vectors; the multiple preset feature dimensions include at least: high-frequency energy ratio, sample entropy, skewness and kurtosis; the safety identification module is used to input the multi-dimensional time-frequency high-order feature matrix into the trained fault identification model for safety identification and obtain the transient weak fault identification result of the new energy grid connection.

[0014] On the other hand, an electronic device is also provided, including: a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the method provided in the embodiments of the present invention.

[0015] On the other hand, a computer-readable storage medium is also provided, which stores computer instructions that, when executed by a processor, implement the method provided in the embodiments of the present invention.

[0016] This invention provides a method and system for high-frequency identification of transient weak faults in new energy grid connection. It introduces a composite evaluation function combining envelope entropy and Pearson skewness coefficient. This mechanism effectively overcomes the defect of conventional VMD optimization algorithm getting trapped in local optima at high-proportion photovoltaic / wind power grid connection points due to excessive high-frequency switching noise energy of inverters. It achieves lossless decoupling and accurate stripping of extremely weak transient features, and alleviates the technical problem of low identification accuracy of high-frequency transient feature signals in new energy grid connection in existing technologies. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a high-frequency identification method for transient weak faults in new energy grid connection provided by an embodiment of the present invention; Figure 2 A flowchart of an adaptive variational mode decomposition based on local envelope entropy and Pearson skewness coefficients is provided for an embodiment of the present invention. Figure 3 A comparison diagram of the optimization trajectories of different objective functions under strong background noise, provided in the embodiments of the present invention; Figure 4 A comparison diagram of the optimization trajectories of different objective functions under strong background noise, provided in the embodiments of the present invention; Figure 5 This is a schematic diagram of a high-frequency identification system for transient weak faults in new energy grid connection provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] Figure 1 This is a flowchart of a high-frequency identification method for transient weak faults in new energy grid connection according to an embodiment of the present invention. This method is applied to new energy grid connection. Figure 1 The method shown specifically includes the following steps: Step S102: Real-time tracking of signal energy mutations in the discrete digital signal sequence of the three-phase current of the new energy grid-connected system, and extracting transient sequence vectors starting from the mutation points.

[0021] Step S104: Based on the optimization of the improved composite objective function, adaptive variational mode decomposition is performed on the transient sequence vector to obtain multiple intrinsic mode function vectors; the improved composite objective function includes the composite objective function of envelope entropy and Pearson skewness coefficient.

[0022] Step S106: Select multiple intrinsic mode function vectors to avoid inherent interference and obtain multiple effective mode vectors.

[0023] Step S108: Based on multiple effective modal vectors, construct a multi-dimensional time-frequency high-order feature matrix including multiple preset feature dimensions; the multiple preset feature dimensions include at least: high-frequency energy proportion, sample entropy, skewness and kurtosis.

[0024] Step S110: Input the multidimensional time-frequency high-order feature matrix into the trained fault identification model for safety identification, and obtain the transient weak fault identification results of new energy grid connection.

[0025] Specifically, step S102 further includes the following steps: Step S1021: Calculate the instantaneous energy characteristic value of each sampling point in the discrete digital signal sequence of the three-phase current of the new energy grid-connected grid based on the nonlinear discrete Tegel energy operator; Step S1022, when N consecutive w When the instantaneous energy characteristic value of each sampling point is greater than the preset energy threshold, the discrete digital signal sequence is truncated based on an adaptive sliding time window of length L, starting from the point of change, to obtain the original data column vector. Step S1023: Standardize the original data column vector to obtain the transient sequence vector.

[0026] Specifically, firstly, the synchronous phasor measurement device or high-frequency waveform recording module of the edge computing terminal at the grid connection point is used to sample at a frequency f. s The discrete digital signal sequence i(n) of the three-phase current is continuously acquired at a frequency of ≥100kHz, where n is the discrete time index of the sampling point. The instantaneous energy characteristic value of each sampling point is obtained in real time using the nonlinear discrete Tegel energy operator (TEO). :

[0027] The TEO operator has a secondary amplification effect on the instantaneous energy changes of high-frequency transient signals, which can effectively overcome the masking effect of strong PWM switching noise at the grid connection point of high-proportion new energy sources on weak arc mutation points and avoid missed detection.

[0028] Set the steady-state energy reference threshold (i.e., the preset energy threshold) to E. th When N consecutive w Instantaneous energy eigenvalues ​​of sampling points >E th At that time, an adaptive sliding time window is triggered. A column vector of length L of the original data is truncated from the point of change. To eliminate the difference in signal reference amplitude under different output conditions, X is Z-score normalized to obtain a transient sequence vector X that eliminates the influence of dimensions. * :

[0029] In the formula, μ and σ are the arithmetic mean and standard deviation of the original data column vector X, respectively; I is a column vector of length L consisting of all 1s.

[0030] Specifically, step S104 further includes the following steps: Step S1041: Perform initial mode decomposition on the transient sequence vector to obtain multiple initial intrinsic mode function components, extract the envelope amplitude sequence of each initial intrinsic mode function component and normalize it to obtain the probability distribution sequence. Step S1042: Calculate the envelope entropy of the probability distribution sequence corresponding to each initial intrinsic mode function component, and calculate the Pearson skewness coefficient of each initial intrinsic mode function component; Step S1043: Establish a two-dimensional continuous optimization space for the number of decomposition layers and the penalty factor, and establish an improved composite objective function with the minimization of envelope entropy and the maximization of Pearson skewness coefficient as composite objectives. Step S1044: Based on the improved composite objective function, the optimal parameter combination is searched in the two-dimensional continuous optimization space using an objective optimization algorithm. Step S1045: Under the optimal parameter combination, the variational mode decomposition algorithm is executed to perform adaptive variational mode decomposition on the transient sequence vector to obtain multiple eigenmode function vectors.

[0031] Specifically, in response to the technical problem that the single envelope entropy optimization in the existing technology is prone to getting trapped in local optima under extreme non-stationary noise, this invention constructs an adaptive optimization model with "minimizing envelope entropy and maximizing Pearson skewness coefficient" as the composite objective.

[0032] Figure 2 This is a flowchart of an adaptive variational mode decomposition based on local envelope entropy and Pearson skewness coefficients, provided according to an embodiment of the present invention. Figure 2 As shown, the two-dimensional continuous optimization space to be optimized is set as [K, α], where K is the number of decomposition levels and α is the quadratic penalty factor. For any k-th intrinsic mode function (IMF) component u obtained from the decomposition... k (t), its envelope amplitude sequence is extracted by Hilbert transform and then normalized to obtain the probability distribution sequence p. k,i Calculate the envelope entropy E k :

[0033] Simultaneously, the Pearson skewness coefficient P, which characterizes pulse asymmetry, was calculated. k :

[0034] In the formula, E[] represents the expected value. and They represent the components of the k-th intrinsic mode function, respectively. The mean and standard deviation.

[0035] Constructing an improved composite objective function :

[0036] In the formula, w1 and w2 are the balancing weight coefficients. The optimal parameter combination [K] that globally minimizes the composite objective function is determined through an optimization algorithm. opt ,α opt Preferably, the globally optimal parameter combination is locked using a particle swarm optimization algorithm. Under these parameters, the augmented Lagrangian function of the Variational Mode Decomposition (VMD) algorithm is constructed:

[0037]

[0038] In the formula, For the k-th intrinsic mode function component The center frequency, The symbol represents the Dirac impulse function, and "*" indicates convolution. Let be the Larlange multipliers, and <, > denote inner product operations.

[0039] By iteratively updating in the frequency domain using the Alternating Directional Multiplier Method (ADMM), broadband noisy signals are orthogonally decoupled into independent modes.

[0040] Figure 3 This is a comparison chart of the optimization trajectories of different objective functions under strong background noise, provided by embodiments of the present invention. Figure 3 It is known that conventional methods using a single envelope entropy for optimization are prone to getting trapped in local optima in low-frequency background noise, while the composite objective function optimization method used in this invention can accurately pinpoint the weak fault frequency band.

[0041] Specifically, step S106 further includes the following steps: Step S1061: Calculate the center frequency of each intrinsic mode function vector; Step S1062: Remove interference modes from multiple intrinsic mode function vectors whose center frequency is equal to the power frequency fundamental wave and the inherent switching frequency of the grid-connected inverter, and use the retained wideband effective components as effective mode vectors.

[0042] Specifically, in the obtained K opt In the eigenmode function vector, the center frequency is forcibly removed. Equivalent to the interference modes of the power frequency fundamental wave and the inverter's inherent switching frequency, the M broadband effective components characterizing the early aging and series arc of cross-linked polyethylene (XLPE) cables are retained as effective mode vectors.

[0043] Specifically, step S108 further includes the following steps: Step S1081: Calculate multiple preset feature dimensions for each effective modal component; Step S1082: Combine and cascade multiple preset feature dimensions corresponding to each effective modal component to construct a multidimensional time-frequency high-order feature matrix.

[0044] Specifically, for the retained m-th effective mode vector Calculate the following four preset feature dimensions: (1) High-frequency energy ratio R m : The proportion of high-frequency excitation energy generated by a weak electric arc to the total transient energy; among which,

[0045] (2) Sample Entropy (SampEn)m : The complexity of the quantized sequence in phase space; (3) Skewness Sk m : Calculate the asymmetry of the probability density distribution; (4) kurtosis m : Calculate the non-Gaussian steepness characteristics caused by partial discharge pulses.

[0046] The aforementioned preset feature dimensions are combined into the row feature vector of the m-th effective modality vector, and all M row feature vectors v are concatenated. m =[R m SampEn m Sk m Ku m Construct a two-dimensional time-frequency high-order feature matrix. .

[0047] Preferably, in this embodiment of the invention, the fault identification model includes a multi-scale hybrid attention convolutional network with a Monte Carlo random deactivation (MCDropout) mechanism.

[0048] Specifically, the network first uses a multi-scale receptive field residual block (Inception-ResNet-1D) containing parallel 1×1, 3×3 and 5×5 convolutional kernels to simultaneously capture the broadband envelope of cable degradation and the extremely narrow pulse of the arc, extract multi-scale features and perform residual fusion. The fused feature mapping matrix H enters the convolutional block attention module (CBAM-1D). This module uses channel attention to dynamically suppress residual background noise channel weights and uses a temporal attention mechanism to accurately lock the microsecond-level nodes of fault energy bursts within the adaptive sliding window.

[0049] Specifically, the convolutional block attention module uses the channel attention module to evaluate the global importance of features in each frequency band using a shared multilayer perceptron (MLP), outputs a channel weight vector, and performs the first background noise suppression. Then, through the temporal attention module, it performs convolution operations in the sliding time dimension to highlight the specific occurrence time of the abrupt pulse and performs the second precise locking.

[0050] To meet the extremely high safety requirements of "zero false tripping" for relay protection in new power systems, this invention performs MC Dropout sampling at the front end of the fully connected network classifier. Specifically, a Dropout layer with a dropout rate of p=0.4 is embedded in the fully connected network classifier. When deployed in the field, for each input feature matrix F, the network forces Dropout to perform N independent forward inferences, obtaining N independent softmax posterior probability distribution column vectors. .

[0051] Calculate the expected probability vector of the current time window state. With uncertainty variance vector :

[0052]

[0053] In the formula, These represent the expected confidence probabilities of the normal operating state, the early degradation state of the cross-linked polyethylene cable, and the initial state of the series arc fault, respectively.

[0054] This invention also introduces industrial safety decision logic: when the expected probability of the target fault type... Continuous N confirm Each sliding window is larger than the diagnostic threshold γ, and its uncertainty variance v3 is strictly less than the safety limit variance. Only when the system recognizes the fault as high confidence and issues an emergency cutoff command to the site's energy management system (EMS) can it be considered a high-confidence fault; if the variance v3 ≥ If the signal is strong, it indicates that the system has encountered an extreme distortion interference that it has not learned before. At this time, the system locks the trip signal and only sends a waveform alarm to the EMS to request manual review, thus completely ensuring the absolute safety and flexible "adjustable and controllable" nature of the new power system dispatch control.

[0055] Figure 4 This is a schematic diagram comparing the anti-disturbance performance of the identification algorithm under different signal-to-noise ratio conditions according to embodiments of the present invention. Figure 4 It is evident that the method provided in this embodiment of the invention can effectively suppress noise compared to existing deep learning methods and traditional physical methods.

[0056] As described above, this invention provides a high-frequency identification method for transient weak faults in new energy grid connection. From the two dimensions of underlying mathematical logic and industrial implementation safety, it achieves a substantial upgrade of the "four capabilities" of new energy grid connection in a transient manner, and has the following technical effects compared with the prior art: (1) High-sensitivity weak mutation tracking based on TEO operator: Existing technologies often use the cumulative sum (CUSUM) algorithm for mutation point detection, which is prone to hysteresis or missed detection when faced with strong background noise. This invention introduces the nonlinear discrete Tegel energy operator (TEO), which utilizes its physical characteristic of "secondary amplification" of instantaneous energy changes of high-frequency transient signals to effectively overcome the masking effect of strong PWM switching noise at high-proportion new energy grid connection points on weak arc / partial discharge mutation points, and realizes more sensitive adaptive sliding window interception.

[0057] (2) Decoupling of Composite Optimization Signals Overcoming Strong Masking Effect: To address the limitation of traditional methods that use a single envelope entropy as the optimization objective, this invention introduces a composite evaluation function combining envelope entropy and skewness coefficients. This mechanism effectively overcomes the defect of conventional VMD optimization algorithms getting trapped in local optima at high-proportion photovoltaic / wind power grid connection points due to excessive high-frequency switching noise energy of the inverter, achieving lossless decoupling and precise stripping of extremely weak transient features.

[0058] (3) Disturbance-resistant identification model based on multi-scale and hybrid attention: The conventional single-scale 1D-CNN structure was abandoned, and a customized network topology combining Inception-ResNet-1D and CBAM-1D was designed. The multi-scale residual structure solves the problem that a single convolutional kernel cannot take into account both low-frequency insulation degradation and high-frequency arc pulses; the dual attention mechanism gives the network a strong "autonomous disturbance resistance" capability, which effectively suppresses the interference of inverter PWM chopper harmonics on the classifier at the physical level.

[0059] (4) Model uncertainty assessment mechanism for the "zero malfunction" requirement of the power grid: Considering the stringent safety requirements of power system dispatch, this invention innovatively introduces the Monte Carlo Dropout mechanism into the industrial online inference process. The model not only outputs the confidence probability of the fault type, but also outputs the "uncertainty variance" of the current identification result in real time. This provides the upper-level energy management system (EMS) with a "knowing what it doesn't know" safety defense line - when facing extreme unknown strong noise conditions, the system can actively refuse to blindly trip based on the high variance, completely avoiding the common problem of AI algorithms easily causing catastrophic malfunctions in the industrial application of power systems, and achieving absolute safety and extremely fast "controllability".

[0060] Figure 5 This is a schematic diagram of a high-frequency identification system for transient weak faults in new energy grid connection according to an embodiment of the present invention. Figure 5 As shown, the system includes: a mutation tracking module 10, a mode decomposition module 20, an interference screening module 30, a matrix construction module 40, and a security identification module 50.

[0061] Specifically, the mutation tracking module 10 is used to track the signal energy mutation of the discrete digital signal sequence of the three-phase current of the new energy grid connection in real time, and to extract the transient sequence vector starting from the mutation point. The mode decomposition module 20 is used to perform adaptive variational mode decomposition on the transient sequence vector based on the optimization of the improved composite objective function to obtain multiple intrinsic mode function vectors; the improved composite objective function includes the composite objective function of envelope entropy and Pearson skewness coefficient; Interference filtering module 30 is used to filter multiple intrinsic mode function vectors to avoid inherent interference and obtain multiple effective mode vectors; The matrix construction module 40 is used to construct a multi-dimensional time-frequency high-order feature matrix based on multiple effective modal vectors, including multiple preset feature dimensions; the multiple preset feature dimensions include at least: high-frequency energy proportion, sample entropy, skewness and kurtosis; The safety identification module 50 is used to input the multi-dimensional time-frequency high-order feature matrix into the trained fault identification model for safety identification, and obtain the transient weak fault identification results of new energy grid connection.

[0062] The present invention also provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and running on the processor, wherein the processor executes the computer program to implement the method provided in the embodiments of the present invention.

[0063] The present invention also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the method provided in the embodiments of the present invention.

[0064] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0065] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A high-frequency identification method for transient weak faults in new energy grid connection, characterized in that, Applied to grid connection of new energy sources; the method includes: Real-time tracking of signal energy mutations in the discrete digital signal sequence of the three-phase current of the new energy grid-connected system is performed, and transient sequence vectors are extracted starting from the mutation points. Based on the optimization of the improved composite objective function, the transient sequence vector is subjected to adaptive variational mode decomposition to obtain multiple intrinsic mode function vectors; the improved composite objective function includes a composite objective function of envelope entropy and Pearson skewness coefficient; By filtering the multiple intrinsic mode function vectors to avoid inherent interference, multiple effective mode vectors are obtained; Based on the multiple effective modal vectors, a multi-dimensional time-frequency high-order feature matrix is ​​constructed, which includes multiple preset feature dimensions; the multiple preset feature dimensions include at least: high-frequency energy proportion, sample entropy, skewness and kurtosis. The multidimensional time-frequency high-order feature matrix is ​​input into the trained fault identification model for safety identification, and the transient weak fault identification result of the new energy grid connection is obtained.

2. The method according to claim 1, characterized in that: Real-time tracking of signal energy mutations in the discrete digital signal sequence of the three-phase current of the grid-connected new energy source, and extracting transient sequence vectors starting from the mutation points, including: The instantaneous energy characteristic value of each sampling point in the discrete digital signal sequence of the three-phase current of the new energy grid-connected device is calculated based on the nonlinear discrete Tegel energy operator. When consecutive N w When the instantaneous energy characteristic value of each sampling point is greater than the preset energy threshold, the discrete digital signal sequence is truncated based on an adaptive sliding time window of length L, starting from the point of change, to obtain the original data column vector; The original data column vector is standardized to obtain a transient sequence vector.

3. The method according to claim 1, characterized in that: Based on the optimization of the improved composite evaluation function, an adaptive variational mode decomposition is performed on the transient sequence vector to obtain multiple intrinsic mode function vectors, including: The transient sequence vector is subjected to initial mode decomposition to obtain multiple initial intrinsic mode function components. The envelope amplitude sequence of each initial intrinsic mode function component is extracted and normalized to obtain the probability distribution sequence. Calculate the envelope entropy of the probability distribution sequence corresponding to each initial intrinsic mode function component, and calculate the Pearson skewness coefficient of each initial intrinsic mode function component; A two-dimensional continuous optimization space for the number of decomposition layers and the penalty factor is established, and an improved composite objective function is established with the minimization of the envelope entropy and the maximization of the Pearson skewness coefficient as composite objectives. Based on the improved composite objective function, the optimal parameter combination is searched in the two-dimensional continuous optimization space using an objective optimization algorithm. Under the optimal parameter combination, the variational mode decomposition algorithm is executed to adaptively decompose the transient sequence vector to obtain multiple eigenmode function vectors.

4. The method according to claim 1, characterized in that: The improved composite objective function includes: In the formula, K is the number of decomposition layers, α is the penalty factor, and w1 and w2 are the balancing weight coefficients. For the improved composite objective function, E k Let P be the envelope entropy. k is the Pearson skewness coefficient.

5. The method according to claim 1, characterized in that: By filtering the multiple intrinsic mode function vectors to avoid inherent interference, multiple effective mode vectors are obtained, including: Calculate the center frequency of each intrinsic mode function vector; The interference modes whose center frequency is equivalent to the power frequency fundamental wave and the inherent switching frequency of the grid-connected inverter are removed from the multiple intrinsic mode function vectors, and the retained wideband effective components are taken as the effective mode vectors.

6. The method according to claim 1, characterized in that: Based on the multiple effective mode vectors, a multi-dimensional time-frequency high-order feature matrix is ​​constructed, including multiple preset feature dimensions, comprising: Calculate multiple preset feature dimensions for each effective modal component; Multiple preset feature dimensions corresponding to each effective modal component are combined and cascaded to construct a multidimensional time-frequency high-order feature matrix.

7. The method according to claim 1, characterized in that: The fault identification model includes a multi-scale hybrid attention convolutional network with a Monte Carlo random deactivation mechanism.

8. A high-frequency identification system for transient weak faults in new energy grid connection, characterized in that, This system is used to implement the high-frequency identification method for transient weak faults in new energy grid connection as described in any one of claims 1-7; the system includes: a mutation tracking module, a mode decomposition module, an interference screening module, a matrix construction module, and a security identification module; wherein... The mutation tracking module is used to track the signal energy mutation of the discrete digital signal sequence of the three-phase current of the new energy grid connection in real time, and to extract the transient sequence vector starting from the mutation point. The mode decomposition module is used to perform adaptive variational mode decomposition on the transient sequence vector based on an improved composite objective function to obtain multiple intrinsic mode function vectors; the improved composite objective function includes a composite objective function of envelope entropy and Pearson skewness coefficient; The interference filtering module is used to filter the multiple intrinsic mode function vectors to avoid inherent interference, thereby obtaining multiple effective mode vectors; The matrix construction module is used to construct a multi-dimensional time-frequency high-order feature matrix based on the multiple effective modal vectors, including multiple preset feature dimensions; the multiple preset feature dimensions include at least: high-frequency energy proportion, sample entropy, skewness and kurtosis; The safety identification module is used to input the multi-dimensional time-frequency high-order feature matrix into the trained fault identification model for safety identification, and obtain the transient weak fault identification result of the new energy grid connection.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the method as claimed in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the method as described in any one of claims 1-7.

Images