Intelligent excitation surge current identification method based on machine learning

Abstract

The invention relates to the technical field of transformers, in particular to an intelligent excitation surge current identification method based on machine learning, which comprises the following steps of: firstly, constructing a simulation model of a high-impedance transformer, and optimizing the model so that the model can output an oscillogram of excitation surge current according with an experimental result according to specified input parameters; a simulation experiment is carried out on the basis, so that the difficulty and the time cost of a manual experiment are greatly reduced; processing and analyzing the data of the simulation experiment to obtain a feature vector; deep learning is carried out on the feature vectors to find the relation between the feature vectors of different parameters; and finally, decision-level fusion is carried out on the multi-modal data set, each parameter independently obtains a decision result and then is fused together, and a final waveform result is output according to positive and negative correlation gains. The prediction and identification of the excitation inrush current are realized; and a basis is provided for suppression of excitation inrush current.

Description

Technical Field

[0001] The present invention relates to the technical field of transformers, and in particular to an intelligent identification method of excitation inrush current based on machine learning. Background Art

[0002] Inrush current refers to the sudden high current phenomenon in the circuit in the power system, especially when the circuit parameters are not matched or the load changes suddenly. The main reasons for the inrush current are as follows:

[0003] Transformer core saturation:

[0004] When the transformer is switched on without load or the voltage is restored after the external fault is removed, the maximum magnetic flux in the core cannot change suddenly, and due to the influence of residual magnetism, the maximum magnetic flux will far exceed the saturation magnetic flux of the transformer, causing the excitation current to increase sharply. The magnetization characteristics of the transformer are closely related. The more saturated the core is, the greater the excitation current required to generate a certain magnetic flux.

[0005] Circuit parameters do not match:

[0006] When the inductive components in the circuit (such as transformers, coils, etc.) do not match the circuit parameters of the entire system, it is easy to generate a peak voltage when the load changes suddenly, which in turn causes an instantaneous surge in current.

[0007] System voltage changes:

[0008] When the system voltage changes, the excitation voltage is affected, resulting in excitation inrush current. This change may be caused by factors such as sudden load changes and instability of the power grid.

[0009] Phase angle at closing moment:

[0010] The phase angle of the voltage during closing will also affect the magnitude of the excitation inrush current. At the most unfavorable closing moment, that is, when the instantaneous voltage value is zero, the maximum value of the magnetic flux density in the core can reach 2Φm. At this time, the saturation of the core will be very serious, and the value of the excitation current will increase significantly.

[0011] Conventional methods for detecting excitation inrush current include a second harmonic braking method, a waveform recognition method, and a voltage braking method. For example, a Chinese patent application with publication number CN105021920A discloses a multi-feature recognition method for main transformer excitation inrush current, which mainly uses the second harmonic braking method to identify the excitation inrush current. The above methods all identify the excitation inrush current after it occurs and then take action, and are unable to predict the occurrence of the excitation inrush current in advance. Based on this, the present application proposes an intelligent recognition method for excitation inrush current based on machine learning. Summary of the invention

[0012] The present invention is proposed in view of the above-mentioned background problems.

[0013] Therefore, the problem to be solved by the present invention is how to construct a prediction model of excitation inrush current based on experimental data.

[0014] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0015] A first aspect of the present invention provides a method for intelligently identifying an excitation inrush current based on machine learning, comprising the following specific steps:

[0016] S1. Construct a simulation model in the simulation software according to the geometric electrical parameters of the transformer;

[0017] S2. Data collection: Collect the electrical data of the transformer during normal operation, the electrical data of the transformer when it is switched on without load, and the electrical data when the voltage is restored after the external fault is removed;

[0018] S3, data preprocessing: cleaning, denoising and standardizing the collected raw data;

[0019] S4, using the preprocessed data as input to conduct a simulation experiment in the simulation model, and comparing the results of the simulation experiment with those obtained from the actual experiment, optimizing the parameters of the simulation model until the fitting degree between the simulation experiment result and the actual experiment result is greater than 0.9;

[0020] S5. Use the adjusted simulation model to conduct multiple simulation experiments to obtain experimental results under different parameters, and use them to construct a simulation data set R;

[0021] S6. Extract features from the simulation data set R, establish a multi-modal feature vector of the excitation inrush current and the opening angle, closing angle, core saturation, time, temperature, and residual magnetism; and map the multi-modal feature vectors to the same feature space to construct a multi-modal data set;

[0022] S7. Perform deep learning on multimodal datasets to learn the potential correlation between datasets of different modalities;

[0023] S8. Perform decision-level fusion on multimodal data sets, build a prediction model for excitation inrush current, and verify the model;

[0024] S9. Use the prediction model to identify the excitation inrush current during transformer operation.

[0025] The geometrical electrical parameters of the transformer in step S1 include the transformer model, core material, core structure, winding structure, size and basic material quantity; and also include rated current, nominal frequency, voltage transformation ratio, current transformation ratio, efficiency and loss, insulation performance and temperature rise performance;

[0026] The core material includes density, magnetic permeability, and saturation magnetic induction intensity;

[0027] The winding structure also includes the number of turns, coil material and winding resistance;

[0028] And set boundary conditions according to the actual nameplate parameters of the transformer; specifically including magnetic field boundary, current boundary and voltage boundary.

[0029] In step S2, electrical data of the transformer during normal operation is collected, including operating voltage and operating current;

[0030] The electrical data of the transformer when it is closed without load includes the opening angle, residual magnetism, closing angle, and bias magnetism;

[0031] The electrical data when the voltage is restored after the external fault is removed include opening angle, residual magnetism, closing angle, and bias magnetism.

[0032] The data preprocessing steps in step S3 include data consistency check and deduplication, outlier processing, and normalization processing.

[0033] In step S4, the fitting degree T satisfies the following formula:

[0034]

[0035] Among them, y a is the actual value, y p is the predicted value, is the mean of the actual values; and satisfies T 2 Greater than 0.9. Establish the characteristic vector data set Q1 when the transformer is switched on without load:

[0036] Q1=[θ d ,θ c ,S,t,T,B r ]

[0037] In the formula, θ d is the opening angle, θ c is the closing angle; S is the core saturation, t is the time, and T is the temperature;

[0038] The opening angle θ d , closing angle θ c Through current waveform extraction; time t is obtained through the built-in timer, and temperature T is obtained through the temperature sensor; the calculation formula of S core saturation is:

[0039]

[0040] Among them, S is the core saturation; B is the current magnetic flux density; B sat The saturation flux density of the material can be obtained by consulting the core material and database data;

[0041] B=μH

[0042] μ=μ0μ r

[0043] Among them, core material: μ0 is the vacuum magnetic permeability; μ r is the relative magnetic permeability of the material; H is the magnetic field intensity; calculated by the formula:

[0044]

[0045] Where N is the number of turns in the coil; I is the current through the coil, and L is the average length of the magnetic circuit;

[0046] B r is the residual magnetism; B r =μ0(μ r -1)H c ; where H c For coercive force.

[0047] Establish the data set Q2 when the voltage is restored after the transformer external fault is removed:

[0048] Establish the characteristic vector data set Q1 when the transformer is closed without load:

[0049] Q2=[θ d ,θ c ,S,t,T,B r ]

[0050] In the formula, θ d is the opening angle, θ c is the closing angle; S is the core saturation, t is the time, and T is the temperature;

[0051] The opening angle θ d , closing angle θ c Through current waveform extraction; time t is obtained through the built-in timer, and temperature T is obtained through the temperature sensor; the calculation formula of S core saturation is:

[0052]

[0053] Among them, S is the core saturation; B is the current magnetic flux density; B sat The saturation flux density of the material can be obtained by consulting the core material and database data;

[0054] B=μH

[0055] μ=μ0μ r

[0056] Among them, core material: μ0 is the vacuum magnetic permeability; μ r is the relative magnetic permeability of the material; H is the magnetic field intensity; calculated by the formula:

[0057]

[0058] Where N is the number of turns in the coil; I is the current through the coil, and L is the average length of the magnetic circuit;

[0059] B r is the residual magnetism; B r =μ0(μ r -1)H c ; where H c For coercive force.

[0060] The convolutional neural network (CNN) is used to learn the established data sets Q1 and Q2 respectively to determine the correlation between different modality data sets;

[0061] The loss function is expressed as mean square error MSE:

[0062]

[0063] Where n is the total number of samples, yi is the actual value of the i-th sample, is the predicted value of the i-th sample.

[0064] A second aspect of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned method for intelligent identification of excitation inrush current based on machine learning when executing the computer program.

[0065] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned method for intelligent identification of excitation inrush current based on machine learning.

[0066] The beneficial effects of the present invention are:

[0067] The present invention first constructs a simulation model of a high-impedance transformer and then optimizes the model so that it can output a waveform diagram of the excitation inrush current that meets the experimental results according to the specified input parameters; then a simulation experiment is carried out based on this, which greatly reduces the difficulty and time cost of manual experiments; the data of the simulation experiment are processed, analyzed and a feature vector is obtained; the relationship between the feature vectors of different parameters is found by deep learning of the feature vectors; finally, a decision-level fusion is performed on the multimodal data set, and the decision results of each parameter are obtained separately and then fused together to output the final waveform result according to the positive and negative correlated gains; the prediction and identification of the excitation inrush current are realized; and the suppression of the excitation inrush current is used as a basis, which is conducive to improving the safety of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of 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 work.

[0069] Figure 1 A 3D simulation model of a medium and high impedance transformer with intelligent identification method of magnetizing inrush current based on machine learning.

[0070] Figure 2 It is a 2D simulation model of medium and high impedance transformers with intelligent identification method of excitation inrush current based on machine learning.

[0071] Figure 3 This is a flow chart of the intelligent identification method of excitation inrush current based on machine learning. DETAILED DESCRIPTION

[0072] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0073] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0074] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0075] Example 1

[0076] A first aspect of the present invention provides a method for intelligently identifying an excitation inrush current based on machine learning, comprising the following specific steps:

[0077] S1. Construct a simulation model in the simulation software according to the geometric electrical parameters of the transformer;

[0078] The geometrical electrical parameters of the transformer in step S1 include the transformer model, core material, core structure, winding structure, size and basic material quantity; and also include rated current, nominal frequency, voltage transformation ratio, current transformation ratio, efficiency and loss, insulation performance and temperature rise performance;

[0079] The core material includes density, magnetic permeability, and saturation magnetic induction intensity;

[0080] The winding structure also includes the number of turns, coil material and winding resistance;

[0081] And set boundary conditions according to the actual nameplate parameters of the transformer; specifically including magnetic field boundary, current boundary and voltage boundary.

[0082] This article takes the high-voltage built-in high-impedance transformer as an example to introduce the solution in detail:

[0083] First, the high-voltage built-in transformer includes an iron core, a high-voltage winding, a medium-voltage winding, a voltage regulating winding and a low-voltage winding; by setting a reactor inside the transformer oil tank, the impedance is increased by increasing the reactance at the transformer inlet; the key to this method is to effectively shield the leakage magnetic field generated by the reactor to reduce the stray loss generated in the structural parts and prevent local overheating; the hysteresis model of the high-voltage built-in high-impedance transformer is:

[0084]

[0085] Where: B is the magnetic induction intensity; H is the magnetic field intensity; B s is the saturation magnetic induction intensity; H c is the coercive force, which indicates the magnetic field strength required for the material to return to a non-magnetic state after the external magnetic field is removed. The adjustment model is constructed with K as the adjustment coefficient, as follows:

[0086] B′(H)=KB(H)

[0087] Among them, B′(H) is the adjusted hysteresis model, K is the adjustment coefficient;

[0088]

[0089] k1≤K≤k2;

[0090] In the above formula, H0 is a material characteristic constant, which is used to adjust the relationship between magnetic induction intensity and magnetic field intensity in the equation; generally, for different types of materials, such as silicon steel and ferrite, the material characteristic constant H0 is not the same, and the value of H0 can be directly obtained by querying the database or textbook to select data matching the transformer material in the embodiment. Calibration can also be performed by experimental measurement; for example, the relationship between the magnetic induction intensity B and the magnetic field intensity H under different magnetic field intensities can be recorded by measuring the hysteresis loop of the material, and the relationship between the magnetic induction intensity B and the magnetic field intensity H under different temperatures, different frequencies and different material uniformities can be collected by performing a control experiment; a specific magnetic field intensity value H1 is extracted from the experimental data as a preliminary estimate of the H0 value; the experimental data is fitted using a nonlinear regression method, and the magnetic field intensity value H2 that best fits the experimental data is calculated.

[0091] For instances where the material characteristic constant H0 cannot be obtained by querying a database or textbook, verification is performed only by experiment;

[0092]

[0093] k1≤K≤k2.

[0094] S2. Data collection: Collect the electrical data of the transformer during normal operation, the electrical data of the transformer when it is switched on without load, and the electrical data when the voltage is restored after the external fault is removed;

[0095] In step S2, electrical data of the transformer during normal operation is collected, including operating voltage and operating current;

[0096] The electrical data of the transformer when it is closed without load includes the opening angle, residual magnetism, closing angle, and bias magnetism;

[0097] The electrical data when the voltage is restored after the external fault is removed include opening angle, residual magnetism, closing angle, and bias magnetism.

[0098] By collecting the electrical data of the transformer during normal operation, the relationship between the operating voltage, operating current, the phase difference between the voltage and current, time t, and temperature T can be analyzed; and by conducting a specific analysis of the electrical data under the two special conditions of transformer no-load closing and external fault removal, the correlation between various parameters such as the opening angle, residual magnetism, closing angle, and bias magnetism can be further explored.

[0099] S3, data preprocessing: cleaning, denoising and standardizing the collected raw data;

[0100] The data preprocessing steps in step S3 include data consistency check and deduplication, outlier processing, and normalization processing.

[0101] S4, using the preprocessed data as input to conduct a simulation experiment in the simulation model, and comparing the results of the simulation experiment with those obtained from the actual experiment, optimizing the parameters of the simulation model until the fitting degree between the simulation experiment result and the actual experiment result is greater than 0.9;

[0102] In step S4, the fitting degree T satisfies the following formula:

[0103]

[0104] Among them, y a is the actual value, y p is the predicted value, is the mean of the actual values; and satisfies T 2 Greater than 0.9.

[0105] S5. Use the adjusted simulation model to conduct multiple simulation experiments to obtain experimental results under different parameters, and use them to construct a simulation data set R;

[0106] S6. Extract features from the simulation data set R, establish a multi-modal feature vector of the excitation inrush current and the opening angle, closing angle, core saturation, time, temperature, and residual magnetism; and map the multi-modal feature vectors to the same feature space to construct a multi-modal data set;

[0107] Establish the characteristic vector data set Q1 when the transformer is closed without load:

[0108] Q1=[θ d ,θ c ,S,t,T,B r ]

[0109] In the formula, θ d is the opening angle, θ c is the closing angle; S is the core saturation, t is the time, and T is the temperature;

[0110] The opening angle θ d , closing angle θ c Through current waveform extraction; time t is obtained through the built-in timer, and temperature T is obtained through the temperature sensor; the calculation formula of S core saturation is:

[0111]

[0112] Among them, S is the core saturation; B is the current magnetic flux density; B sat The saturation flux density of the material can be obtained by consulting the core material and database data;

[0113] B=μH

[0114] μ=μ0μ r

[0115] Among them, core material: μ0 is the vacuum permeability; μ r is the relative magnetic permeability of the material; H is the magnetic field intensity; calculated by the formula:

[0116]

[0117] Where N is the number of turns in the coil; I is the current through the coil, and L is the average length of the magnetic circuit;

[0118] B r is the residual magnetism; B r =μ0(μ r -1)H c ; where H c For coercive force.

[0119] Establish the data set Q2 when the voltage is restored after the transformer external fault is removed:

[0120] Establish the characteristic vector data set Q1 when the transformer is closed without load:

[0121] Q2=[θ d ,θ c ,S,t,T,B r ]

[0122] In the formula, θ d is the opening angle, θ c is the closing angle; S is the core saturation, t is the time, and T is the temperature;

[0123] The opening angle θ d , closing angle θ c Through current waveform extraction; time t is obtained through the built-in timer, and temperature T is obtained through the temperature sensor; the calculation formula of S core saturation is:

[0124]

[0125] Among them, S is the core saturation; B is the current magnetic flux density; B sat The saturation flux density of the material can be obtained by consulting the core material and database data;

[0126] B=μH

[0127] μ=μ0μ r

[0128] Among them, core material: μ0 is the vacuum magnetic permeability; μ r is the relative magnetic permeability of the material; H is the magnetic field intensity; calculated by the formula:

[0129]

[0130] Where N is the number of turns in the coil; I is the current through the coil, and L is the average length of the magnetic circuit;

[0131] B r is the residual magnetism; B r =μ0(μ r -1)H c ; where H c For coercive force.

[0132] S7, deep learning is performed on multimodal data sets to learn the potential correlation between different modal data sets; convolutional neural network CNN is used to learn the established data sets Q1 and Q2 respectively to determine the correlation between different modal data sets;

[0133] The loss function is expressed as mean square error MSE:

[0134]

[0135] Where n is the total number of samples, yi is the actual value of the i-th sample, is the predicted value of the i-th sample.

[0136] S8. Perform decision-level fusion on multimodal data sets, build a prediction model for excitation inrush current, and verify the model;

[0137] S9. Use the prediction model to identify the excitation inrush current during transformer operation.

[0138] In this embodiment, a plurality of high-impedance transformer models with different core structures are first established by simulation software, and then a hysteresis model is constructed for each high-impedance transformer model based on the model to describe the hysteresis characteristics of the core; then, various parameters of the actual transformer are input into the established hysteresis model, and the magnetic induction intensity saturation value B is determined according to the experimental data. s , coercive force H c The value is substituted into the model to output the image of the hysteresis loop and compared with the hysteresis loop obtained from the actual experiment, so as to verify the model;

[0139] Then, the magnetic flux density and magnetization intensity in the iron core are calculated based on the hysteresis model and solution algorithm; the magnetic flux density B = μH, μ is the magnetic permeability of the iron core material, and H is the magnetic field intensity; the simulation model is run to obtain the distribution of the magnetic flux density and magnetization intensity in the iron core; the change of the magnetic flux density over time is calculated considering the hysteresis characteristics of the iron core, and the state quantity in the hysteresis model (magnetic induction intensity saturation value B s , coercive force H c value) and continuously optimize the adjustment coefficient K so that the image of the hysteresis loop output by the model and the hysteresis loop obtained by the actual transformer experiment have a fitting degree greater than 0.9 to obtain a high impedance transformer simulation model B′(H) that takes into account the hysteresis characteristics.

[0140] Using this model as the basis, simulation experiments are carried out to explore the influence of various parameters on the excitation inrush current phenomenon; specifically, electrical data of the transformer during normal operation, electrical data of the transformer when no-load closing, and electrical data when the voltage is restored after the external fault is removed are collected; including opening angle, residual magnetism, closing angle, and bias magnetism data; then the original data is cleaned, denoised, and standardized through data preprocessing, and each data is unified into the same dimension; the preprocessed data is used as input to carry out simulation experiments in the simulation model, and the results obtained from the simulation experiments are compared with those obtained from the actual experiments, and the parameters of the simulation model are optimized until the fit between the simulation experiment results and the actual experimental results is greater than 0.9; the final simulation model is obtained.

[0141] A multivariate simulation experiment is performed using the adjusted simulation model to obtain a simulation data set R; feature extraction is performed based on the simulation data set R; it is known that during the closing process, the opening angle, residual magnetism, closing angle, and bias magnetism data will affect the waveform and amplitude of the excitation inrush current; this embodiment uses the above parameters as feature quantities to establish different feature vector data sets for the two states of transformer no-load closing and voltage recovery after the transformer external fault is removed; they are expressed as:

[0142] Q1=[θ d ,θ c ,S,t,T,B r ]

[0143] Q2=[θ d ,θ c ,S,t,T,B r ]

[0144] In the formula, θ d is the opening angle, θ c is the closing angle; S is the core saturation, t is the time, and T is the temperature;

[0145] The opening angle θ d , closing angle θ c Through current waveform extraction; time t is obtained through the built-in timer, and temperature T is obtained through the temperature sensor; the calculation formula of S core saturation is:

[0146]

[0147] Among them, S is the core saturation; B is the current magnetic flux density; B sat The saturation flux density of the material can be obtained by consulting the core material and database data;

[0148] B=μH

[0149] μ=μ0μ r

[0150] Among them, core material: μ0 is the vacuum permeability; μ r is the relative magnetic permeability of the material; H is the magnetic field intensity; calculated by the formula:

[0151]

[0152] Where N is the number of turns in the coil; I is the current through the coil, L is the average length of the magnetic circuit; B r is the residual magnetism; B r =μ0(μ r -1)H c ; where H c For coercive force.

[0153] The convolutional neural network CNN is used to learn the established data sets Q1 and Q2 respectively to determine the correlation between different modal data sets; the value range of each parameter is determined according to several common excitation inrush current waveforms; for example, if the closing angle is at the rising edge of the sine wave of the voltage waveform (between 0-90°), the waveform of the excitation inrush current will be steeper and the amplitude will be higher; in addition, the angular position at the time of opening will affect the residual magnetism. If the magnetic flux is in a saturated state at the time of opening, the excitation inrush current will be larger and the waveform will be steeper when closing; based on this, the user can divide different accuracies for each excitation inrush current waveform with typical characteristics as needed, and select a specific value range for each parameter; finally, the multi-modal data set is fused at the decision level; the decision result is first obtained through each parameter separately, and then the decision results of multiple parameters are fused to obtain the final result and output it.

[0154] For example, the opening angle includes four states (A1, A2, A3, A4), the closing angle includes four states (B1, B2, B3, B4); the core saturation includes three states (C1, C2, C3), the time is divided into multiple levels with a gradient of 30s, the temperature is divided into multiple levels with a gradient of 100 degrees, and the residual magnetism includes four states (D1, D2, D3, D4); the corresponding common excitation inrush current waveforms are a, b, c, d; according to the above simulation, the opening angle A1 A3 and the closing angle B1 B3 will output waveform a; the opening angle A2 A4 and the closing angle B2 B4 will output waveform b; and as time goes by, it will affect the temperature and residual magnetism, which are negatively correlated, so it will produce negative gain on the waveform; and the core saturation and residual magnetism will both produce positive gain on the waveform; therefore, each parameter obtains a decision result separately and then merges them together to output the final waveform result according to the positive and negative correlated gains.

[0155] Example 2

[0156] This embodiment provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the intelligent identification method of excitation inrush current based on machine learning described in Example 1 are implemented.

[0157] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0158] Example 3

[0159] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the intelligent identification method of excitation inrush current based on machine learning described in Embodiment 1 are implemented.

[0160] The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination of them, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0161] In summary, the intelligent identification method of excitation inrush current based on machine learning proposed in the present invention can quickly predict the waveform of the excitation inrush current based on a simulation model through a small number of sensor measurement parameters, thereby providing a research basis for the subsequent suppression of the excitation inrush current.

[0162] 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 preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for intelligent identification of excitation inrush current based on machine learning, characterized in that: The specific steps include: S1. Construct a simulation model in the simulation software according to the geometric electrical parameters of the transformer; S2. Data collection: Collect the electrical data of the transformer during normal operation, the electrical data of the transformer when it is switched on without load, and the electrical data when the voltage is restored after the external fault is removed; S3, data preprocessing: cleaning, denoising and standardizing the collected raw data; S4, using the preprocessed data as input to conduct a simulation experiment in the simulation model, and comparing the results of the simulation experiment with those obtained from the actual experiment, optimizing the parameters of the simulation model until the fitting degree between the simulation experiment result and the actual experiment result is greater than 0.9; S5. Use the adjusted simulation model to conduct multiple simulation experiments to obtain experimental results under different parameters, and use them to construct a simulation data set R; S6. Extract features from the simulation data set R, establish a multi-modal feature vector of the excitation inrush current and the opening angle, closing angle, core saturation, time, temperature, and residual magnetism; and map the multi-modal feature vectors to the same feature space to construct a multi-modal data set; S7. Perform deep learning on multimodal datasets to learn the potential correlation between datasets of different modalities; S8. Perform decision-level fusion on multimodal data sets, build a prediction model for excitation inrush current, and verify the model; S9. Use the prediction model to identify the excitation inrush current during transformer operation.

2. The method for intelligent identification of excitation inrush current based on machine learning according to claim 1, characterized in that: The geometrical electrical parameters of the transformer in step S1 include the transformer model, core material, core structure, winding structure, size and basic material quantity; and also include rated current, nominal frequency, voltage transformation ratio, current transformation ratio, efficiency and loss, insulation performance and temperature rise performance; The core material includes density, magnetic permeability, and saturation magnetic induction intensity; The winding structure also includes the number of turns, coil material and winding resistance; And set boundary conditions according to the actual nameplate parameters of the transformer; specifically including magnetic field boundary, current boundary and voltage boundary.

3. The method for intelligent identification of excitation inrush current based on machine learning according to claim 1, characterized in that: In step S2, electrical data of the transformer during normal operation is collected, including operating voltage and operating current; The electrical data of the transformer when it is closed without load includes the opening angle, residual magnetism, closing angle, and bias magnetism; The electrical data when the voltage is restored after the external fault is removed include opening angle, residual magnetism, closing angle, and bias magnetism.

4. The method for intelligent identification of excitation inrush current based on machine learning according to claim 1, characterized in that: The data preprocessing steps in step S3 include data consistency check and deduplication, outlier processing, and normalization processing.

5. The method for intelligent identification of excitation inrush current based on machine learning according to claim 1, characterized in that: In step S4, the fitting degree T satisfies the following formula: Among them, y a is the actual value, y p is the predicted value, is the mean of the actual values; and satisfies T 2 Greater than 0.

9.

6. The method for intelligent identification of excitation inrush current based on machine learning according to claim 3, characterized in that: Establish the characteristic vector data set Q1 when the transformer is closed without load: Q1=[θ d ,θ c ,S,t,T,B r ] In the formula, θ d is the opening angle, θ c is the closing angle; S is the core saturation, t is the time, and T is the temperature; The opening angle θ d , closing angle θ c Through current waveform extraction; time t is obtained through the built-in timer, and temperature T is obtained through the temperature sensor; the calculation formula of S core saturation is: Among them, S is the core saturation; B is the current magnetic flux density; B sat The saturation flux density of the material can be obtained by consulting the core material and database data; B=μH μ=μ0μ r Among them, core material: μ0 is the vacuum magnetic permeability; μ r is the relative magnetic permeability of the material; H is the magnetic field intensity; calculated by the formula: Where N is the number of turns in the coil; I is the current through the coil, and L is the average length of the magnetic circuit; B r is the residual magnetism; B r =μ0(μ r -1)H c ; where H c For coercive force.

7. The method for intelligent identification of excitation inrush current based on machine learning according to claim 3, characterized in that: Establish the data set Q2 when the voltage is restored after the transformer external fault is removed: Establish the characteristic vector data set Q1 when the transformer is closed without load: Q2=[θ d ,θ c ,S,t,T,B r ] In the formula, θ d is the opening angle, θ c is the closing angle; S is the core saturation, t is the time, and T is the temperature; The opening angle θ d , closing angle θ c Through current waveform extraction; time t is obtained through the built-in timer, and temperature T is obtained through the temperature sensor; the calculation formula of S core saturation is: Among them, S is the core saturation; B is the current magnetic flux density; B sat The saturation flux density of the material can be obtained by consulting the core material and database data; B=μH μ=μ0μ r Among them, core material: μ0 is the vacuum magnetic permeability; μ r is the relative magnetic permeability of the material; H is the magnetic field intensity; calculated by the formula: Where N is the number of turns in the coil; I is the current through the coil, and L is the average length of the magnetic circuit; B r is the residual magnetism; B r =μ0(μ r -1)H c ; where H c For coercive force.

8. The method for intelligent identification of excitation inrush current based on machine learning according to claim 6 or 7, characterized in that: The convolutional neural network (CNN) is used to learn the established data sets Q1 and Q2 respectively to determine the correlation between different modality data sets; The loss function is expressed as mean square error MSE: Where n is the total number of samples, yi is the actual value of the i-th sample, is the predicted value of the i-th sample.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent identification method of excitation inrush current based on machine learning described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent identification method of excitation inrush current based on machine learning described in any one of claims 1 to 8 are implemented.

Images