Excitation surge current identification method and system based on time sequence early classification algorithm

Abstract

The invention discloses an excitation surge current identification method and system based on a time sequence early classification algorithm, and the method comprises the steps: collecting the electrical quantity of each side of a transformer, and extracting the characteristic quantities of different dimensions, such as second harmonic, waveform symmetry degree, time domain equivalent impedance, phase current change rate, and the like through analyzing the phase current, differential current, excitation impedance and the like of each side; according to the method, actual data and a large amount of simulation data are fused to establish a data set, and a magnetizing inrush current identification model based on a time sequence early classification algorithm is trained, so that magnetizing inrush current is identified, and a basis is provided for latching protection of a transformer. According to the method, waveform analysis is performed in a sliding time window mode based on an excitation surge current identification model, a classification result is obtained by taking a half power frequency period as a sliding step length and result updating analysis data, and meanwhile, credible confidence is obtained based on a multi-confidence fusion method, so that the final state of the transformer is judged. And the method can be well combined with the existing protection logic.

Description

Technical Field

[0001] The present invention belongs to the technical field of relay protection of electric power systems, and in particular, relates to a method and system for identifying magnetizing inrush current based on a time series early classification algorithm. Background Art

[0002] As the core equipment in the power grid, transformers have certain requirements for their main protection. Differential protection such as longitudinal differential and phase differential has the characteristics of simple principle and easy fault location, and is therefore widely used. When the power transformer is switched on to the grid without load, due to the saturation of the transformer core magnetic flux and the nonlinear characteristics of the core material, a large amplitude of excitation surge current will be generated, which may cause the transformer differential protection to malfunction.

[0003] As the main protection of the transformer, the differential protection has the risk of refusal to operate and false operation when identifying the excitation inrush current and internal fault. In order to improve the reliability of transformer operation, it is necessary to distinguish between the fault current and the excitation inrush current in the protection zone, so it is necessary to explore the method of identifying the excitation inrush current. The mainstream protection algorithms currently used in practice mostly use single principles such as second harmonics and interruption angle as the judgment criteria, which reduces the accuracy of the excitation inrush current identification.

[0004] Prior art document 1 (CN104967097A) discloses a method for identifying excitation inrush current based on a support vector classifier. Seven characteristic quantities, namely, second harmonic, third harmonic, current interruption angle, wave width, waveform distortion, waveform correlation coefficient and excitation side measurement impedance, are selected as the input of the support vector machine, and then various operating states of the transformer are trained to construct a decision function for identifying excitation inrush current and fault current. When an accident occurs in the transformer, the seven characteristic quantities are calculated from the data collected by the protection device acquisition system and substituted into the decision function for judging the excitation inrush current and fault current.

[0005] Prior art document 2 (CN114398983A) discloses a classification prediction method, which includes: obtaining at least two data for a target classification task, wherein each data corresponds to a modality; obtaining the confidence corresponding to at least two modalities, respectively, the confidence is used to indicate the classification prediction probability of the modality in the target classification task; based on the confidence corresponding to at least two modalities, weighted fusion of data features of at least two data is performed to obtain fusion features, and the fusion features are predicted to obtain the classification prediction result corresponding to the target classification task.

[0006] However, the shortcoming of the prior art document 1 is that the excitation surge current and fault current are judged after the fact, and it is impossible to judge in advance. The relay protection device cannot make a timely judgment on whether to perform a protection action based on this, and the judgment accuracy is not high.

[0007] The shortcoming of the prior art document 2 is that each confidence has a corresponding confidence network, and the fusion feature is obtained by weighted fusion, which only considers the characteristics of the static data itself. Summary of the invention

[0008] In order to solve the deficiencies in the prior art, the present invention provides a method and system for identifying excitation inrush current based on a time series early classification algorithm, which improves the accuracy of identifying excitation inrush current.

[0009] The present invention takes into account the transformer excitation inrush current state, out-of-zone fault state, and in-zone fault state. The electrical quantities in the three situations show different distribution characteristics in the signal sampling value distribution. In order to quantify this difference, feature quantities such as second harmonic, waveform symmetry, time domain equivalent impedance, and phase current change rate are selected. At the same time, feature extraction is performed on the transformer fault data from different dimensions. Within a given time window, it is classified and identified based on the time series early classification algorithm, and the multi-confidence fusion method is used to improve the accuracy of identification.

[0010] The present invention adopts the following technical solution.

[0011] A first aspect of the present invention provides a method for identifying an excitation inrush current based on a time series early classification algorithm, comprising:

[0012] S1. Under the conditions of different transformer capacities, different closing initial phase angles, different residual magnetism and different winding connection methods, collect the voltage and current waveforms of the transformer under the conditions of magnetizing inrush current, external fault state and internal fault state;

[0013] S2. Extracting features from the voltage waveform and the current waveform to obtain feature quantities, wherein the feature quantities include waveform second harmonic ratio, waveform symmetry characteristics, time domain equivalent impedance, and phase current change rate, and establishing a multi-dimensional excitation inrush current identification data set based on the feature quantities;

[0014] S3. Establish a time series data analysis window based on the multi-dimensional excitation inrush current identification data set;

[0015] S4, inputting the time series in the continuous multiple time series analysis windows into the excitation inrush current identification model to obtain multiple original confidences, wherein the excitation inrush current identification model is obtained based on the training of the time series early classification algorithm, and the original confidences include the confidence that the transformer is in a normal state, the confidence that the transformer is in an excitation inrush current state, and the confidence that the transformer is in an internal fault state;

[0016] S5. Input multiple original confidences into the neural network, obtain the time confidences and offsets corresponding to the multiple original confidences through neural network training, calculate the comprehensive confidence based on the multiple original confidences and their corresponding time confidences and offsets, and select the category corresponding to the maximum confidence in the comprehensive confidence as the excitation inrush current identification result.

[0017] Optionally, in S2, feature extraction is performed on the voltage waveform and the current waveform to obtain waveform symmetry features, including:

[0018] The current waveform is sampled to obtain the following series X(k):

[0019] X(k)=|I(k)+I(k+n)| / (I(k)|+|I(k+n)),k=1,2...n

[0020] Among them, I(k) is the three-phase differential current derivative, and the number of sampling points per cycle is 2n;

[0021] Based on the sampling points of a week, the fuzzy closeness N is calculated as follows, which is the waveform symmetry feature:

[0022]

[0023] Where A[X(k)] is the membership function constructed based on the sequence X(k), and n is the number of sampling points in a half cycle.

[0024] Optionally, in S2, feature extraction is performed on the voltage waveform and the current waveform to obtain phase current change rate features, including:

[0025] The phase current change rate is calculated according to the following formula, which is the phase current change rate characteristic:

[0026]

[0027] Among them, K I is the current change rate, t ph is the time interval between the maximum and minimum values ​​of the first-order differential phase current in a sampling period, and T is the power frequency period.

[0028] Optionally, in S2, feature extraction is performed on the voltage waveform and the current waveform according to the following formula to obtain equivalent excitation impedance features:

[0029]

[0030] Among them, u1 is the voltage across the primary winding of the transformer, r is the equivalent resistance, L is the equivalent excitation inductance, that is, the equivalent excitation impedance characteristic, which is i d is the differential current.

[0031] Optionally, the time series data analysis window in S3 is a sliding time series data analysis window, and the data in a time series data analysis window is represented by the following matrix X:

[0032]

[0033] Where X={X1,X2,...,X t ,...,X T} represents the multi-dimensional excitation inrush current identification dataset, X t It means that the feature dimension at time t is D.

[0034] Optionally, the sliding step size of the time series data analysis window is 1 / 2 power frequency period.

[0035] Optionally, the neural network in S5 is an LSTM network.

[0036] Optionally, in S5, the comprehensive confidence S is calculated according to the following formula:

[0037] S=W*P+B

[0038] Among them, P is the original confidence, W is the time confidence, and B is the offset.

[0039] Optionally, when collecting voltage waveforms and current waveforms in the transformer excitation inrush current state, out-of-zone fault state and in-zone fault state, the voltage waveform data or current waveform data collected in each state is not less than 1200.

[0040] A second aspect of the present invention provides an excitation inrush current identification system, comprising:

[0041] The acquisition module is used to collect the voltage and current waveforms of the transformer under the conditions of different transformer capacities, different closing initial phase angles, different residual magnetism and different winding connection modes, in the state of magnetizing inrush current, out-of-zone fault state and in-zone fault state;

[0042] A feature extraction module is used to extract features from the voltage waveform and the current waveform to obtain feature quantities, wherein the feature quantities include waveform second harmonic ratio, waveform symmetry characteristics, time domain equivalent impedance and phase current change rate, and establish a multi-dimensional excitation inrush current identification data set based on the feature quantities;

[0043] Establish a module for establishing a time series data analysis window based on a multi-dimensional excitation inrush current identification data set;

[0044] A classification module, used for inputting the time series in a plurality of continuous time series analysis windows into an excitation inrush current identification model, wherein the excitation inrush current identification model is obtained by training based on an early classification algorithm for the time series, and obtaining a plurality of original confidences, wherein the original confidences include the confidence that the transformer is in a normal state, the confidence that the transformer is in an excitation inrush current state, and the confidence that the transformer is in an internal fault state;

[0045] The fusion module is used to input multiple original confidences into the neural network, obtain the time confidences and offsets corresponding to the multiple original confidences through neural network training, calculate the comprehensive confidence based on the multiple original confidences and their corresponding time confidences and offsets, and select the category corresponding to the maximum confidence in the comprehensive confidence as the excitation inrush current identification result.

[0046] Compared with the prior art, the beneficial effects of the present invention include at least:

[0047] Compared with the prior art document 1, the significant difference of the present invention is that it extracts features based on a sliding window and uses a time series early classification algorithm to identify results in advance; thereby identifying the excitation inrush current in advance, providing a basis for whether the transformer should be locked for protection, and can be well combined with existing relay protection methods.

[0048] Compared with the prior art document 2, the present invention is significantly different in that, in order to increase the accuracy of the early classification algorithm of the time series, the present invention uses an LSTM network to obtain the time confidence of multiple consecutive analysis results, and obtains the final confidence by fusing the original confidence with the time confidence. The change of data over time, that is, the dynamic characteristics, is taken into account, based on which early fault identification can be achieved and the results can be dynamically updated. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0050] Figure 1 A schematic flow chart of a method for identifying an excitation inrush current based on a time series early classification algorithm provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0052] Combination Figure 1 As shown, Embodiment 1 of the present invention provides a method for identifying an excitation inrush current based on a time series early classification algorithm, the method comprising:

[0053] S1. Under the conditions of different transformer capacities, different closing initial phase angles, different residual magnetism and different winding connection methods, the voltage waveform and current waveform of the transformer under the state of magnetizing inrush current, out-of-zone fault state and in-zone fault state are collected.

[0054] S2. Extract features from the voltage waveform and the current waveform to obtain feature quantities, wherein the feature quantities include waveform second harmonic ratio, waveform symmetry characteristics, time domain equivalent impedance and phase current change rate, and establish a multi-dimensional excitation inrush current identification data set based on the feature quantities.

[0055] The following is a further explanation of the four feature quantities.

[0056] Feature 1: Waveform second harmonic ratio.

[0057] The waveform second harmonic ratio is the differential current second harmonic ratio. The differential current second harmonic ratio refers to the ratio of the second harmonic to the fundamental wave in the three-phase differential current, which is expressed by the following formula:

[0058]

[0059] in,

[0060] I dφ2 is the second harmonic component in the differential current,

[0061] K xb.2 is the differential current second harmonic ratio,

[0062] I dφ is the fundamental component in the differential current.

[0063] Feature 2: Waveform symmetry feature.

[0064] Assume the three-phase differential current derivative is I(k), the number of sampling points per cycle is 2n, and the logarithmic sequence is:

[0065] X(k)=|I(k)+I(k+n) / (I(k)|+I(k+n)),k=1,2...n

[0066] It can be understood that the smaller X(k), the more fault information the point contains, that is, the greater the credibility of the fault; conversely, the larger X(k), the more information about the surge contained in the point, that is, the greater the credibility of the surge. Take a membership function, set as A[X(k)], integrate the information of one week, for k=1,2...n, and obtain the fuzzy closeness N, which is the waveform symmetry feature:

[0067]

[0068] Find the threshold value N set , when N>N set When N <N set It is considered as magnetizing inrush current.

[0069] Feature 3: Equivalent excitation impedance characteristic, expressed as follows:

[0070]

[0071] Where u1 is the voltage across the primary winding of the transformer, r is the equivalent resistance, L is the equivalent excitation inductance, i.e. the equivalent excitation impedance characteristic, d is the differential current.

[0072] According to experiments, the excitation inductance has the following characteristics: when the transformer is operating normally or has an out-of-zone fault, the iron core operates in the non-saturation zone and the excitation inductance has a large value; when the transformer generates an excitation surge current, the iron core switches back and forth between the saturation zone and the non-saturation zone, and the excitation inductance varies between the normal value and the saturation value, with obvious fluctuations; when an internal fault occurs, the excitation inductance has a small value.

[0073] Feature 4: Phase current change rate characteristics.

[0074] According to the characteristic of the magnetizing inrush current waveform with a discontinuity angle, in the non-fault state, the time interval between the maximum and minimum current values ​​within one cycle should be less than or equal to half the power frequency cycle. The fault current waveform is similar to a sine wave. This feature is used to identify the magnetizing inrush current:

[0075]

[0076] Among them, t ph is the time interval between the maximum and minimum values ​​of the first-order differential phase current in a sampling period, T is the power frequency period, K I is the rate of change of current.

[0077] Optionally, when collecting voltage waveforms and current waveforms in the transformer excitation inrush current state, out-of-zone fault state and in-zone fault state, the voltage waveform data or current waveform data collected in each state is not less than 1200.

[0078] In S2, a multi-dimensional excitation inrush current recognition data set is established based on the feature quantity. The recognition data set can be divided into a training set and a test set. Each recognition data set includes 4 types of feature data and label data in S2. The label data includes the transformer being in a normal state, the transformer being in an excitation inrush current state, and the transformer being in an internal fault state.

[0079] The time series data analysis window in S3 is a sliding time series data analysis window.

[0080] By establishing a sliding time series data analysis window, the changes in data over time, that is, the dynamic characteristics, are taken into account. Based on this, early fault identification can be achieved and the results can be dynamically updated.

[0081] Optionally, the sliding step size of the time series data analysis window is 1 / 2 power frequency period.

[0082] Specifically, combined with actual business needs, 1 / 2 power frequency cycle is used as the sliding data window (sampling rate is 1200 Hz, corresponding to 12 sampling points), and the feature set obtained by S2 is X = {X1, X2, ..., X t ,...,X T},in X t It represents the feature vector with feature dimension D (here, D=4) at time t. A time series window data can finally be represented by a matrix as shown below:

[0083]

[0084] S4. Input the time series in multiple consecutive time series analysis windows into an excitation inrush current identification model to obtain multiple original confidences, wherein the excitation inrush current identification model is trained based on an early classification algorithm for the time series, and the original confidences include the confidence that the transformer is in a normal state, the confidence that the transformer is in an excitation inrush current state, and the confidence that the transformer is in an internal fault state.

[0085] In order to give the transformer status as early as possible and ensure high accuracy, it is necessary to train a multi-classifier to analyze each time series analysis window and give the classification result and the original confidence P, where P = [p 1, p 2, p 3, ], p1 is the confidence that the transformer is in a normal state, p2 is the confidence that the transformer is in an excitation inrush current state, and p3 is the confidence that the transformer is in an internal fault state. At the same time, the classification results and original confidence of the next time series analysis window are obtained.

[0086] The present invention trains multiple classifiers based on lightGBM (Light Gradient Boosting Machine) to identify the three states of the transformer: magnetizing inrush current state, out-of-zone fault state and in-zone fault state. The advantage of the present invention is that it can not only meet the demand for shortening the model calculation time in the protection and control field, but also reduce the use of memory by data, and more data can be processed in the same time.

[0087] S5. Input multiple original confidences into the neural network, obtain the time confidences and offsets corresponding to the multiple original confidences through neural network training, calculate the comprehensive confidence based on the multiple original confidences and their corresponding time confidences and offsets, and select the category corresponding to the maximum confidence in the comprehensive confidence as the excitation inrush current identification result.

[0088] Optionally, the neural network in S5 is an LSTM network.

[0089] Optionally, in S5, the comprehensive confidence S is calculated according to the following formula:

[0090] S=W*P+B

[0091] Among them, P is the original confidence, W is the time confidence, and B is the offset.

[0092] Specifically, in order to ensure the earlyness and accuracy of the recognition results, we cannot rely solely on the classification results and their original confidence P obtained in a single time window, where It is also necessary to use the time confidence of the time window. According to business needs, 1 / 2 cycle is used as the analysis object. Therefore, the present invention integrates the original confidence obtained from 12 consecutive data analysis windows, and trains the time confidence W and its offset B corresponding to the original confidence P based on the LSTM network, where W = [w 1, w 2, w 3, …,w 11, w 12 ],B=[b 1, b 2, b 3, …,b 11, b 12 ]. Finally, the result of the operation based on the product of the time confidence W and the original confidence P and the offset B is taken as the comprehensive confidence S, and the calculation formula is: S = W*P+B, where Finally, the category corresponding to the element with the largest value in S is selected as the final recognition result.

[0093] Embodiment 2 of the present invention provides an excitation inrush current identification system, which runs the excitation inrush current identification method based on the time series early classification algorithm as described in Embodiment 1, and the system comprises:

[0094] The acquisition module is used to collect the voltage and current waveforms of the transformer under the conditions of different transformer capacities, different closing initial phase angles, different residual magnetism and different winding connection modes, in the state of magnetizing inrush current, out-of-zone fault state and in-zone fault state;

[0095] A feature extraction module is used to extract features from the voltage waveform and the current waveform to obtain feature quantities, wherein the feature quantities include waveform second harmonic ratio, waveform symmetry characteristics, time domain equivalent impedance and phase current change rate, and establish a multi-dimensional excitation inrush current identification data set based on the feature quantities;

[0096] Establish a module for establishing a time series data analysis window based on a multi-dimensional excitation inrush current identification data set;

[0097] A classification module, used for inputting the time series in a plurality of continuous time series analysis windows into an excitation inrush current identification model, wherein the excitation inrush current identification model is obtained by training based on an early classification algorithm for the time series, and obtaining a plurality of original confidences, wherein the original confidences include the confidence that the transformer is in a normal state, the confidence that the transformer is in an excitation inrush current state, and the confidence that the transformer is in an internal fault state;

[0098] The fusion module is used to input multiple original confidences into the neural network, obtain the time confidences and offsets corresponding to the multiple original confidences through neural network training, calculate the comprehensive confidence based on the multiple original confidences and their corresponding time confidences and offsets, and select the category corresponding to the maximum confidence in the comprehensive confidence as the excitation inrush current identification result.

[0099] Compared with the prior art, the beneficial effects of the present invention include at least:

[0100] Compared with the prior art document 1, the significant difference of the present invention is that it extracts features based on a sliding window and uses a time series early classification algorithm to identify results in advance; thereby identifying the excitation inrush current in advance, providing a basis for whether the transformer should be locked for protection, and can be well combined with existing relay protection methods.

[0101] Compared with the prior art document 2, the present invention is significantly different in that, in order to increase the accuracy of the early classification algorithm of the time series, the present invention uses an LSTM network to obtain the time confidence of multiple consecutive analysis results, and obtains the final confidence by fusing the original confidence with the time confidence. The change of data over time, that is, the dynamic characteristics, is taken into account, based on which early fault identification can be achieved and the results can be dynamically updated.

[0102] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0103] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0104] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0105] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0106] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for identifying magnetizing inrush current based on a time series early classification algorithm, characterized in that: include: S1. Under the conditions of different transformer capacities, different closing initial phase angles, different residual magnetism and different winding connection methods, collect the voltage and current waveforms of the transformer under the conditions of magnetizing inrush current, external fault state and internal fault state; S2. Extracting features from the voltage waveform and the current waveform to obtain feature quantities, wherein the feature quantities include waveform second harmonic ratio, waveform symmetry characteristics, time domain equivalent impedance, and phase current change rate, and establishing a multi-dimensional excitation inrush current identification data set based on the feature quantities; S3. Establish a time series data analysis window based on the multi-dimensional excitation inrush current identification data set; S4, inputting the time series in a plurality of continuous time series analysis windows into an excitation inrush current identification model to obtain a plurality of original confidences, wherein the excitation inrush current identification model is obtained based on early classification training of the time series, and the original confidences include the confidence that the transformer is in a normal state, the confidence that the transformer is in an excitation inrush current state, and the confidence that the transformer is in an internal fault state; S5. Input multiple original confidences into the neural network, obtain the time confidences and offsets corresponding to the multiple original confidences through neural network training, calculate the comprehensive confidence based on the multiple original confidences and their corresponding time confidences and offsets, and select the category corresponding to the maximum confidence in the comprehensive confidence as the excitation inrush current identification result.

2. The method for identifying magnetizing inrush current based on a time series early classification algorithm according to claim 1, characterized in that: In S2, feature extraction is performed on the voltage waveform and the current waveform to obtain waveform symmetry features, including: The current waveform is sampled to obtain the following series X(k): X(k)=|I(k)+I(k+n)| / (I(k)|+|I(k+n)),k=1,2...n Among them, I(k) is the three-phase differential current derivative, and the number of sampling points per cycle is 2n; Based on the sampling points of a week, the fuzzy closeness N is calculated as follows, which is the waveform symmetry feature: Where A[X(k)] is the membership function constructed based on the sequence X(k), and n is the number of sampling points in a half cycle.

3. The method for identifying magnetizing inrush current based on a time series early classification algorithm according to claim 1, characterized in that: In S2, feature extraction is performed on the voltage waveform and the current waveform to obtain phase current change rate features, including: The phase current change rate is calculated according to the following formula, which is the phase current change rate characteristic: Among them, K I is the current change rate, t ph is the time interval between the maximum and minimum values ​​of the first-order differential phase current in a sampling period, and T is the power frequency period.

4. The method for identifying magnetizing inrush current based on a time series early classification algorithm according to claim 1, characterized in that: In S2, the voltage waveform and the current waveform are feature extracted according to the following formula to obtain the equivalent excitation impedance feature: Among them, u1 is the voltage across the primary winding of the transformer, r is the equivalent resistance, L is the equivalent excitation inductance, that is, the equivalent excitation impedance characteristic, which is i d is the differential current.

5. The method for identifying magnetizing inrush current based on a time series early classification algorithm according to claim 1, characterized in that: The time series data analysis window in S3 is a sliding time series data analysis window. The data in a time series data analysis window is represented by the following matrix X: Where X={X1,X2,…,X t ,…,X T } represents the multi-dimensional excitation inrush current identification dataset, X t Indicates that the feature dimension at time t is D.

6. The method for identifying magnetizing inrush current based on a time series early classification algorithm according to claim 5, characterized in that: The sliding step of the time series data analysis window is 1 / 2 power frequency period.

7. The method for identifying magnetizing inrush current based on a time series early classification algorithm according to claim 1, characterized in that: The neural network in S5 is an LSTM network.

8. The method for identifying magnetizing inrush current based on time series early classification algorithm according to claim 1, characterized in that: In S5, the comprehensive confidence S is calculated according to the following formula: S=W*P+B Among them, P is the original confidence, W is the time confidence, and B is the offset.

9. The method for identifying magnetizing inrush current based on a time series early classification algorithm according to claim 1, characterized in that: When collecting voltage waveforms and current waveforms under the transformer excitation inrush current state, out-of-zone fault state and in-zone fault state, the voltage waveform data or current waveform data collected under each state shall be no less than 1,200 items.

10. An excitation inrush current identification system using the excitation inrush current identification method based on the time series early classification algorithm according to claims 1 to 9, characterized in that: include: The acquisition module is used to collect the voltage and current waveforms of the transformer under the conditions of different transformer capacities, different closing initial phase angles, different residual magnetism and different winding connection modes, in the state of magnetizing inrush current, out-of-zone fault state and in-zone fault state; A feature extraction module is used to extract features from the voltage waveform and the current waveform to obtain feature quantities, wherein the feature quantities include waveform second harmonic ratio, waveform symmetry characteristics, time domain equivalent impedance and phase current change rate, and establish a multi-dimensional excitation inrush current identification data set based on the feature quantities; Establish a module for establishing a time series data analysis window based on a multi-dimensional excitation inrush current identification data set; A classification module, used for inputting the time series in a plurality of continuous time series analysis windows into an excitation inrush current identification model, wherein the excitation inrush current identification model is obtained by training based on an early classification algorithm for the time series, and obtaining a plurality of original confidences, wherein the original confidences include the confidence that the transformer is in a normal state, the confidence that the transformer is in an excitation inrush current state, and the confidence that the transformer is in an internal fault state; The fusion module is used to input multiple original confidences into the neural network, obtain the time confidences and offsets corresponding to the multiple original confidences through neural network training, calculate the comprehensive confidence based on the multiple original confidences and their corresponding time confidences and offsets, and select the category corresponding to the maximum confidence in the comprehensive confidence as the excitation inrush current identification result.

Images