A method for constructing a fault arc warning model for distribution cabinet based on Fourier transform

Abstract

The present invention discloses a method for constructing a distribution cabinet fault arc warning model based on Fourier transform, and the steps of the method include: S1, collecting relevant data of the distribution cabinet, S2, preliminary warning identification, S3, deep warning analysis, S4, selective warning report; the technical key points are: through an auxiliary supervision model, preliminary warning identification is performed on the fault arc, the heating condition of the fault arc is judged, and a first warning event and a second warning event are generated, and at the same time, a corresponding warning supervision mechanism is allocated, and the first warning event and the second warning event are deeply analyzed to evaluate the abnormal degree of the fault arc; in this process, Fourier transform is performed on the power parameters to obtain a time-frequency diagram, and the constructed dual-channel time-frequency model is used to train the power parameters and the corresponding time-frequency diagram, so as to further strengthen the correlation between the time domain and the frequency domain, improve the accuracy of the warning, and reduce the false alarm rate.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution cabinet detection, and in particular to a method for constructing a fault arc warning model of a power distribution cabinet based on Fourier transform. Background Art

[0002] Fault arc refers to abnormal current discharge caused by insulation damage, poor contact, overload, etc. in electrical equipment or power systems. This form of discharge usually occurs under high voltage or high current conditions, which may cause equipment damage, fire or other safety hazards. However, it still needs to be paid attention to when the equipment is running for a long time under low current or low voltage conditions.

[0003] Traditional models often select a single infrared image or current signal or voltage signal to input into the machine learning model for judgment, and extract the corresponding single energy entropy as a feature variable to classify normal arc or fault arc; currently, in terms of fault arc warning, a deep learning model has been introduced, and the time domain and frequency domain are extracted based on artificially constructed signals, and the signals are input into the deep learning model to further accurately determine the category of the fault arc; it is not adaptive and ignores the correlation between the time domain and the frequency domain, which may lead to poor accuracy of the warning and a high false alarm rate. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the prior art, the present invention provides a method for constructing a fault arc warning model of a distribution cabinet based on Fourier transform. Through an auxiliary supervision model, a preliminary warning identification is performed on the fault arc, the heating condition of the fault arc is judged, and a first warning event and a second warning event are generated. At the same time, a corresponding warning supervision mechanism is allocated, and an in-depth analysis is performed on the first warning event and the second warning event to evaluate the abnormality of the fault arc. In this process, a Fourier transform is performed on the power parameters to obtain a time-frequency diagram, and the power parameters and the corresponding time-frequency diagram are trained using the constructed dual-channel time-frequency model, which further strengthens the correlation between the time domain and the frequency domain, improves the accuracy of the warning, reduces the false alarm rate, and solves the problems in the background technology.

[0006] (II) Technical solution

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A method for constructing a distribution cabinet fault arc warning model based on Fourier transform, comprising the following steps:

[0009] S1. Collect relevant data of the power distribution cabinet: collect infrared images inside the power distribution cabinet through infrared imaging technology, and obtain image data of the power distribution cabinet in a heating state and in normal operation; collect power parameters of the power distribution cabinet through sensors, and the power parameters include voltage signals and current signals when the power distribution cabinet has a fault arc and is in normal operation;

[0010] S2. Preliminary warning identification: Build a dual-channel time-frequency model with an auxiliary supervision model built in it to perform preliminary warning identification of fault arcs on image data, use the auxiliary supervision model to identify warning signals, and build a rule engine to classify warning signals and generate warning supervision mechanisms corresponding to the first warning event and the second warning event;

[0011] S3, deep warning analysis: according to the known early warning supervision mechanism, supervision processing is carried out, and the corresponding power parameters are retrieved, and the time-frequency diagram is obtained through Fourier transform, and the corresponding features are extracted at the same time. The power parameters and the corresponding time-frequency diagram are trained using a dual-channel time-frequency model, and the fault arc or normal arc is predicted according to the extracted features; at the same time, the first warning event is deeply analyzed to evaluate the abnormality of the fault arc and execute the corresponding strategy; the second warning event is temporarily stored; wherein the extracted features include time, positive sequence component, negative sequence component, zero sequence component, voltage fluctuation signal and current fluctuation signal;

[0012] S4. Selective early warning report: including early warning supervision mechanism category, early warning event category, abnormal assessment coefficient, and time.

[0013] Furthermore, the input of the auxiliary supervision model is set as the image data of the power distribution cabinet in the heating state and normal operation, and the output is set as the warning event category, specifically:

[0014] Step A1, data preparation and processing:

[0015] Preprocessing: Perform histogram equalization and Gaussian filtering on image data;

[0016] Fourier transform: Perform a two-dimensional Fourier transform on the processed image data to generate a complex element matrix, and convert the DC low-frequency part corresponding to the matrix to the middle position, and then obtain the centralized spectrum through complex mapping; and normalize the spectrum, and mark the processed spectrum as P0;

[0017] Step A2, data analysis: preset time period T0, for the spectrum graph P0, obtain the set A of values ​​corresponding to each frequency position in the time period T0, and construct the maximum subset and the minimum subset of the set A, and mark the difference between the maximum subset and the minimum subset as the maximum span value KD; at the same time, obtain the standard deviation coefficient σ corresponding to each frequency position, and mark the product between the maximum span value KD corresponding to each frequency position and the corresponding standard deviation coefficient σ as the warning value yg; wherein the standard deviation coefficient is the standard deviation value obtained for the set A within the preset time period T0;

[0018] Step A3, rule division: preset the standard threshold value YG1 of the warning value yg, and compare and analyze the warning value yg with the standard threshold value YG1:

[0019] When the warning value yg ≥ the standard threshold value YG1, a first warning signal is generated, and the warning value yg is marked as jz1, and edited as a first-level character, and jz1 and the first-level character are combined to generate a first warning event;

[0020] When the warning value yg is less than the standard threshold value YG1, a second warning signal is generated, and the warning value yg is marked as jz2, and edited as a secondary character at the same time, and jz2 and the secondary character are combined to generate a second warning event;

[0021] Furthermore, a first warning event is matched with a first-class warning supervision mechanism, and a calculation period T1 is set, and the number of times the power parameters are collected within the calculation period T1 is nc1; a second warning event is matched with a second-class warning supervision mechanism, and a calculation period T2 is set, and the number of times the power parameters are collected within the calculation period T2 is nc2; wherein nc1 and nc2 are positive integers.

[0022] Furthermore, the construction process of the dual-channel time-frequency model is as follows:

[0023] Step B1, data preparation and processing: power parameters under the corresponding early warning supervision mechanism collected by the auxiliary supervision model;

[0024] Step B2, feature construction:

[0025] In channel 1, the power parameters are subjected to multi-layer wavelet decomposition to obtain high-frequency signals and low-frequency signals at different levels, which are then extracted and marked as feature Z1; in channel 2, the power parameters are subjected to Fourier transform to obtain the corresponding time-frequency diagram, which are then extracted and marked as feature Z2; feature Z1 and feature Z2 are fused and marked as fused feature Z3; wherein fused feature Z3 includes time, positive sequence component, negative sequence component, zero sequence component, voltage fluctuation signal and current fluctuation signal;

[0026] Step B3, model training: channel 1 includes constructing wavelet basis functions and wavelet convolutional neural networks; channel 2 includes constructing Fourier basis functions and Fourier convolutional neural networks;

[0027] Step B4, model prediction: Use the trained model to predict new warning events and obtain fault arcs or normal arcs.

[0028] Furthermore, the process of in-depth analysis of the first warning event is as follows:

[0029] Step C1, evaluate the abnormality of the first warning event, obtain the parameter data set evaluated under the first warning supervision mechanism, and construct a feature space in a preset time period T3 to generate a corresponding abnormality evaluation coefficient PG;

[0030] Step C2, presetting a standard threshold value YG2 of the abnormal evaluation coefficient PG, and generating an alarm signal when the abnormal evaluation coefficient PG exceeds the preset standard threshold value YG2.

[0031] Furthermore, the parameter data set includes voltage imbalance, current imbalance, the time between the occurrence time of the warning event and the current time, and the number of times the power parameters are collected during the time; the process of obtaining the voltage imbalance and current imbalance is:

[0032] Firstly, the positive sequence component, negative sequence component and zero sequence component are extracted from the extracted features and defined as evaluation parameters;

[0033] Then, a three-dimensional coordinate system is established with the positive sequence component F1 as the X-axis, the negative sequence component F2 as the Y-axis, and the zero sequence component F3 as the Z-axis, and a feature space S0 is constructed, a standard feature space S1 of the feature space S0 is preset, and the feature space S0 is compared with the standard feature space S1 to obtain their corresponding horizontal and vertical screenshots, and the total overlap area cmj and the total non-overlap area fmj are generated by comparison; wherein the total non-overlap area fmj is the fluctuation amount of the three-phase current or the three-phase voltage;

[0034] Finally, the obtained total overlap area cmj and the total non-overlap area fmj are combined to obtain the corresponding voltage imbalance or current imbalance;

[0035] According to the formula: ;

[0036] Wherein, Dj is the calculated voltage imbalance or current imbalance, β0 is the weight correction coefficient, and β0 is greater than 0;

[0037] Generate the abnormal assessment coefficient PG according to the formula: ;

[0038] Wherein, cs represents the number of times the power parameters are collected, and cs≤nc1, ch is the time between the occurrence time of the warning event and the current time, and ε0 is the time constant.

[0039] Furthermore, the second warning event is deeply analyzed according to the measurement cycle T2, and the process is processed according to step C1 to step C2.

[0040] (III) Beneficial effects

[0041] The present invention provides a method for constructing a distribution cabinet fault arc warning model based on Fourier transform, which has the following beneficial effects:

[0042] 1. The integrity of the data is ensured by collecting image data and power parameters, providing data support for subsequent early warning event judgment and fault arc category judgment, which improves the accuracy of the analysis to a certain extent; the auxiliary supervision model built into the dual-channel time-frequency model is used to perform preliminary early warning identification of fault arcs. The auxiliary supervision model converts image data into a spectrum diagram based on Fourier transform and uses it as auxiliary supervision for the model training to distinguish between power distribution cabinets in a heating state and in normal operation, and preliminarily judges the possible existence of a fault arc, which effectively improves the accuracy of the dual-channel time-frequency model;

[0043] 2. Conduct in-depth analysis of the first warning event through the first type of supervision mechanism, conduct in-depth analysis of the second warning event through the second type of supervision mechanism, and evaluate the abnormality assessment coefficient by integrating multiple parameters to ensure the accuracy of the abnormality assessment coefficient and accurately evaluate the abnormality of the fault arc;

[0044] 3. Adopting a reasonable deep learning network architecture, adaptively adjusting the Fourier basis function and wavelet basis function in the processing process, further strengthens the correlation between the time domain and the frequency domain, which is conducive to subsequent feature extraction, improves the accuracy of early warning, and reduces the false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A schematic diagram of the process of the present invention is shown. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Example: See Figure 1, this embodiment provides a method for constructing a distribution cabinet fault arc warning model based on Fourier transform;

[0048] The steps of this method are as follows:

[0049] S1. Collect relevant data of the power distribution cabinet: collect infrared images inside the power distribution cabinet through infrared imaging technology, and obtain image data of the power distribution cabinet in a heating state and in normal operation; collect power parameters of the power distribution cabinet through sensors, and the power parameters include voltage signals and current signals when the power distribution cabinet has a fault arc and is in normal operation;

[0050] Among them, infrared imaging technology is a technology that uses infrared radiation to generate images, which detects overheating of the distribution cabinet to prevent fault arcs; the sensor includes a voltage sensor and a current sensor, which are used to detect voltage signals and current signals respectively; and the above devices are not drawn in the figure;

[0051] S2. Preliminary warning identification: Build a dual-channel time-frequency model with an auxiliary supervision model built in it to perform preliminary warning identification of fault arcs on image data, use the auxiliary supervision model to identify warning signals, and build a rule engine to classify warning signals and generate warning supervision mechanisms corresponding to the first warning event and the second warning event;

[0052] The input of the auxiliary supervision model is set as the image data of the power distribution cabinet in the heating state and normal operation, and the output is set as the warning event category, specifically:

[0053] Step A1, data preparation and processing:

[0054] Preprocessing: Perform histogram equalization and Gaussian filtering on image data to remove noise interference and improve the accuracy of Fourier transform;

[0055] Histogram equalization: Based on the formula: ;

[0056] Among them, T(r) represents the new pixel value after preprocessing, r is the original pixel value, L represents the number of gray levels of the pixel, and M*N represents the size of the original image. Indicates the pixel value is n i The number of images in the original image, i represents the pixel value;

[0057] Gaussian filtering: Based on the formula: ;

[0058] Where G(x, y) represents the Gaussian function; (x, y) represents the coordinates of each pixel in the image; σ represents the standard deviation of the Gaussian kernel;

[0059] Fourier transform: Perform a two-dimensional Fourier transform on the processed image data to generate a complex element matrix, and convert the DC low-frequency part corresponding to the matrix to the middle position, and then obtain the centralized spectrum through complex mapping; and normalize the spectrum, and mark the processed spectrum as P0;

[0060] Step A2, data analysis: preset time period T0, for the spectrum graph P0, obtain the set A of corresponding values ​​of each frequency position in the time period T0, and construct the maximum subset and the minimum subset of the set A, and mark the difference between the maximum subset and the minimum subset as the maximum span value KD; at the same time, obtain the standard deviation coefficient σ corresponding to each frequency position, and mark the product between the maximum span value KD corresponding to each frequency position and the corresponding standard deviation coefficient σ as the warning value yg; wherein the standard deviation coefficient is the standard deviation value obtained for the set A within the preset time period T0; it should be noted that the larger the value of the standard deviation coefficient σ, the greater the degree of variation of each frequency position; the maximum span value KD reflects the variability or dynamic range of the frequency component, and the larger its value, the more obvious the frequency component is abnormal; in summary, the larger the corresponding warning value yg, the more obvious the fever abnormality;

[0061] Step A3, rule division: preset the standard threshold value YG1 of the warning value yg, and compare and analyze the warning value yg with the standard threshold value YG1:

[0062] When the warning value yg ≥ the standard threshold value YG1, a first warning signal is generated, and the warning value yg is marked as jz1, and edited as a first-level character, and jz1 and the first-level character are combined to generate a first warning event;

[0063] When the warning value yg is less than the standard threshold value YG1, a second warning signal is generated, and the warning value yg is marked as jz2, and edited as a secondary character at the same time, and jz2 and the secondary character are combined to generate a second warning event;

[0064] Among them, the standard threshold value YG1 is set as an adaptive threshold value, which is adaptively set according to the characteristics of the spectrum graph, specifically:

[0065] Based on historical data: collect the early warning value data of the distribution cabinet in historical operation, and obtain the data range of the distribution cabinet in the heating state and the data range under normal operation;

[0066] Statistical analysis: Collect the warning values ​​of the power distribution cabinet under normal operation and heating state; perform local statistical analysis on the collected data, such as: perform local interception on the points corresponding to the normal operation or heating state, determine the local mean and local standard deviation for the data under local normal operation, set the standard threshold value YG1 according to the statistical results, and the calculation formula of the standard threshold value YG1 is as follows:

[0067] ;

[0068] Among them, mean is the mean of the local area, sttdev is the standard deviation of the local area, and v is the adjustment constant;

[0069] It should be noted that the fault arc will cause abnormal heating in the surrounding area, and will also cause heating in the surrounding non-fault area. The first warning event and the second warning event are determined through the above analysis. The first warning event indicates that there is obvious abnormal heating;

[0070] A first-class early warning supervision mechanism is matched for the first early warning event, and a calculation period T1 is set, and the number of times the power parameters are collected during the calculation period T1 is nc1; a second-class early warning supervision mechanism is matched for the second early warning event, and a calculation period T2 is set, and the number of times the power parameters are collected during the calculation period T2 is nc2; wherein nc1 and nc2 are positive integers;

[0071] S3, deep warning analysis: according to the known early warning supervision mechanism, supervision processing is carried out, and the corresponding power parameters are retrieved, and the time-frequency diagram is obtained through Fourier transform, and the corresponding features are extracted at the same time. The power parameters and the corresponding time-frequency diagram are trained using a dual-channel time-frequency model, and the fault arc or normal arc is predicted according to the extracted features; at the same time, the first warning event is deeply analyzed to evaluate the abnormality of the fault arc and execute the corresponding strategy; the second warning event is temporarily stored; wherein the extracted features include time, positive sequence component, negative sequence component, zero sequence component, voltage fluctuation signal and current fluctuation signal;

[0072] Among them, the construction process of the dual-channel time-frequency model is:

[0073] Step B1, data preparation and processing: power parameters under the corresponding early warning supervision mechanism collected by the auxiliary supervision model;

[0074] Step B2, feature construction:

[0075] In channel 1, the power parameters are subjected to multi-layer wavelet decomposition to obtain high-frequency signals and low-frequency signals at different levels, which are then extracted and marked as feature Z1; in channel 2, the power parameters are subjected to Fourier transform to obtain the corresponding time-frequency diagram, which are then extracted and marked as feature Z2; feature Z1 and feature Z2 are fused and marked as fused feature Z3;

[0076] Step B3, model training:

[0077] Input: The input is set as power parameters, which are labeled as a sequence X(T) = [x1, x2, x3, …];

[0078] Channel 1 includes constructing trainable wavelet basis functions and wavelet convolutional neural networks;

[0079] Training process:

[0080] The wavelet basis function calculation model is:

[0081] ;

[0082] ;

[0083] Among them, λ i+1 To decompose the low-frequency intermediate variables step by step, η i+1 is to decompose the high-frequency intermediate variables step by step, σ(·) is the sigmoid activation function, b i (l) and b i (h) are the bias of the i-th layer, W i (l) and W i (h) represent the low-frequency subsequence and high-frequency subsequence generated by the i-th layer, and W i (l) and W i (h)∈R q*q , X i (l) with X i (h) is the sequence of input power parameters of the i-th layer, q is X i (l) is the sequence length, l and h are the low-frequency filter coefficients and high-frequency filter coefficients respectively;

[0084] Training of wavelet convolutional neural network:

[0085] In this embodiment, a three-layer wavelet basis function is used for the power parameters, and then:

[0086] First layer: Get low-frequency coefficients λ 1 and high frequency coefficient η 1 , and obtain the corresponding sequence X through pooling operation 1 (l) and sequence X 1 (h), the sequence X 1 (l) and sequence X 1 (h) Perform splicing to obtain the splicing vector c 1 ;

[0087] Second layer: Get low-frequency coefficient λ 2 and high frequency coefficient η 2 , and obtain the corresponding sequence X through pooling operation 2 (l) and sequence X 2 (h), the sequence X 2 (l) and sequence X 2 (h) Perform splicing to obtain the splicing vector c 2 ;

[0088] The third layer: Get the low-frequency coefficient λ3 and high frequency coefficient η 3 , and obtain the corresponding sequence X through pooling operation 3 (l) and sequence X 3 (h), the sequence X 3 (l) and sequence X 3 (h) Perform splicing to obtain the splicing vector c 3 ;

[0089] The vector will c 1 、c 2 and c 3 Input to the convolution pooling layer of channel 1 for feature extraction to obtain feature Z1;

[0090] Channel 2 includes constructing trainable Fourier basis functions and Fourier convolutional neural networks;

[0091] Fourier basis function calculation model:

[0092] ;

[0093] Where X(t, f) is the complex value at time t and frequency f, X(T) is the input power parameter, ω(Tt) is the window function at t, and e -j2πfT is a complex exponential function, and j is an imaginary unit;

[0094] In this embodiment, a three-layer Fourier transform is performed on the power parameters, and then:

[0095] First layer: obtain the time-frequency graph u1;

[0096] The second layer: obtain the time-frequency graph u2;

[0097] The third layer: obtain the time-frequency graph u3;

[0098] Input the spectrum graph u1, spectrum graph u2 and spectrum graph u3 into the convolution pooling layer of channel 2 for feature extraction to obtain feature Z2; and fuse the feature Z1 of channel 1 and the feature Z2 of channel 2 to form a fused feature Z3, which is output through the pooling layer and mapped to the category space through the fully connected layer and the classification layer; wherein the fused feature Z3 includes time, positive sequence component, negative sequence component, zero sequence component, voltage fluctuation signal and current fluctuation signal;

[0099] Step B4, model prediction: use the trained model to predict new warning events and obtain fault arcs or normal arcs;

[0100] The process of in-depth analysis of the first warning event is as follows:

[0101] Step C1, evaluate the abnormality of the first warning event, obtain the parameter data set evaluated under the first warning supervision mechanism, and construct a feature space in a preset time period T3 to generate a corresponding abnormality evaluation coefficient PG;

[0102] The parameter data set includes voltage imbalance, current imbalance, the time between the occurrence time of the warning event and the current time, and the number of times the power parameters are collected during the time. The process of obtaining voltage imbalance and current imbalance is as follows:

[0103] Firstly, the positive sequence component, negative sequence component and zero sequence component are extracted from the extracted features and defined as evaluation level parameters;

[0104] Positive sequence component: In a fault arc, the positive sequence component usually does not change significantly because the arc fault is asymmetric and the positive sequence component mainly reflects the normal working state;

[0105] Negative sequence component: In a fault arc, the negative sequence component will increase, because arc faults often cause system asymmetry, and the negative sequence component is used to detect and analyze these asymmetric phenomena;

[0106] Zero-sequence component: In a ground fault arc, the zero-sequence component increases significantly; arc faults usually involve the ground wire or ground, causing a large zero-sequence current to flow, which is used to detect ground faults and their severity;

[0107] According to the formula: ;

[0108] Where F1, F2 and F3 are positive sequence component, negative sequence component and zero sequence component respectively, Fa, Fb and Fc represent the current or voltage in the three-phase system respectively; α is the complex unity root, and , j represents the imaginary unit;

[0109] Then, a three-dimensional coordinate system is established with the positive sequence component F1 as the X-axis, the negative sequence component F2 as the Y-axis, and the zero sequence component F3 as the Z-axis, and a feature space S0 is constructed, a standard feature space S1 of the feature space S0 is preset, and the feature space S0 is compared with the standard feature space S1 to obtain their corresponding horizontal and vertical screenshots, and the total overlap area cmj and the total non-overlap area fmj are generated by comparison; wherein the total non-overlap area fmj is the fluctuation amount of the three-phase current or the three-phase voltage;

[0110] Finally, the obtained total overlap area cmj and the total non-overlap area fmj are combined to obtain the corresponding voltage imbalance or current imbalance;

[0111] According to the formula: ;

[0112] Wherein, Dj is the calculated voltage imbalance or current imbalance, β0 is the weight correction coefficient, and β0 is greater than 0;

[0113] Generate the abnormal assessment coefficient PG according to the formula: ;

[0114] Wherein, cs represents the number of times the power parameters are collected, and cs≤nc1, ch is the time between the occurrence time of the warning event and the current time, and ε0 is a time constant; it should be noted that in this embodiment, a three-phase system is selected. When a fault arc occurs, the three-phase current or three-phase voltage will fluctuate, and the negative sequence component F2 and the zero sequence component F3 will change significantly, often showing an increase; the larger the abnormal evaluation coefficient PG is, the greater the abnormality of the fault arc is;

[0115] Step C2, presetting a standard threshold value YG2 of the abnormal evaluation coefficient PG, when the abnormal evaluation coefficient PG exceeds the preset standard threshold value YG2, generating an alarm signal to remind the staff to respond;

[0116] The specific process of obtaining the standard threshold value YG2 is as follows:

[0117] Based on historical data: collect the abnormal degree assessment data of the distribution cabinet in historical operation, and obtain the abnormal assessment coefficient range under the presence of fault arc and the abnormal assessment coefficient range under normal operation;

[0118] Statistical analysis: Collect the abnormal evaluation coefficient PG of the distribution cabinet under fault arc and normal operation, perform statistical analysis on the collected data, and determine the mean and standard deviation under normal operation; set the standard threshold value YG2 according to the statistical results, and the calculation formula of the standard threshold value YG2 is as follows:

[0119] ;

[0120] Among them, Avg is the mean, dev is the standard deviation, μ is the multiple, and the value of μ is 2~3;

[0121] Perform an in-depth analysis on the second warning event according to the measurement cycle T2, and the process is processed according to steps C1 to C2;

[0122] S4. Selective early warning report; including early warning supervision mechanism category, early warning event category, abnormal assessment coefficient, and time.

[0123] In summary, the steps of the present invention include: S1, collecting relevant data of the distribution cabinet, S2, preliminary warning identification, S3, in-depth warning analysis, S4, selective warning report; through the auxiliary supervision model, preliminary warning identification of the fault arc is performed, the heating conditions of the fault arc are judged, and the first warning event and the second warning event are generated, and the corresponding warning supervision mechanism is allocated at the same time, and the first warning event and the second warning event are deeply analyzed to evaluate the abnormal degree of the fault arc; in this process, the power parameters are Fourier transformed to obtain the time-frequency diagram, and the power parameters and the corresponding time-frequency diagram are trained using the constructed dual-channel time-frequency model, which further strengthens the correlation between the time domain and the frequency domain, improves the effectiveness and accuracy of the fault arc warning model, reduces the false alarm and missed alarm rate during the use of the model, and reduces the conflict between the sensitivity and the false alarm rate.

[0124] The size of the interval and threshold is set for the convenience of comparison. The size of the threshold depends on the amount of sample data and the number of bases set by technicians in this field for each group of sample data; as long as it does not affect the proportional relationship between the parameter and the quantized value;

[0125] The above formulas are all dimensionless and numerical calculations. The formula is a formula obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0126] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0127] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for constructing a distribution cabinet fault arc warning model based on Fourier transform, characterized in that: The following steps are involved: S1. Collecting relevant data of the power distribution cabinet: using infrared imaging technology to collect infrared images inside the power distribution cabinet, and obtaining image data of the power distribution cabinet in a heating state and in normal operation; The power parameters of the distribution cabinet are collected by sensors, including voltage and current signals when the distribution cabinet has a fault arc and operates normally; S2. Preliminary warning identification: Build a dual-channel time-frequency model with an auxiliary supervision model built in it to perform preliminary warning identification of fault arcs on image data, use the auxiliary supervision model to identify warning signals, and build a rule engine to classify warning signals and generate warning supervision mechanisms corresponding to the first warning event and the second warning event; Among them, the construction process of the dual-channel time-frequency model is: Data preparation and processing: Power parameters under the corresponding early warning supervision mechanism collected by the auxiliary supervision model; Feature construction: In channel 1, the power parameters are subjected to multi-layer wavelet decomposition to obtain high-frequency signals and low-frequency signals at different levels, which are then extracted and marked as feature Z1; in channel 2, the power parameters are subjected to Fourier transform to obtain the corresponding time-frequency diagram, which are then extracted and marked as feature Z2; feature Z1 and feature Z2 are fused and marked as fused feature Z3; wherein fused feature Z3 includes time, positive sequence component, negative sequence component, zero sequence component, voltage fluctuation signal and current fluctuation signal; Model training: Channel 1 includes constructing wavelet basis functions and wavelet convolutional neural networks; Channel 2 includes constructing Fourier basis functions and Fourier convolutional neural networks; Model prediction: Use the trained model to predict new warning events and obtain fault arcs or normal arcs; The input of the auxiliary supervision model is set as the image data of the power distribution cabinet in the heating state and normal operation, and the output is set as the warning event category, specifically: Step A1, data preparation and processing: Preprocessing: Perform histogram equalization and Gaussian filtering on image data; Fourier transform: Perform a two-dimensional Fourier transform on the processed image data to generate a complex element matrix, and convert the DC low-frequency part corresponding to the matrix to the middle position, and then obtain the centralized spectrum through complex mapping; and normalize the spectrum, and mark the processed spectrum as P0; Step A2, data analysis: preset time period T0, for the spectrum graph P0, obtain the set A of values ​​corresponding to each frequency position in the time period T0, and construct the maximum subset and the minimum subset of the set A, and mark the difference between the maximum subset and the minimum subset as the maximum span value KD; at the same time, obtain the standard deviation coefficient σ corresponding to each frequency position, and mark the product between the maximum span value KD corresponding to each frequency position and the corresponding standard deviation coefficient σ as the warning value yg; wherein the standard deviation coefficient is the standard deviation value obtained for the set A within the preset time period T0; Step A3, rule division: preset the standard threshold value YG1 of the warning value yg, and compare and analyze the warning value yg with the standard threshold value YG1: When the warning value yg ≥ the standard threshold value YG1, a first warning signal is generated, and the warning value yg is marked as jz1, and edited as a first-level character, and jz1 and the first-level character are combined to generate a first warning event; When the warning value yg is less than the standard threshold value YG1, a second warning signal is generated, and the warning value yg is marked as jz2, and edited as a secondary character at the same time, and jz2 and the secondary character are combined to generate a second warning event; S3, deep warning analysis: according to the known early warning supervision mechanism, supervision processing is carried out, and the corresponding power parameters are retrieved, and the time-frequency diagram is obtained through Fourier transform, and the corresponding features are extracted at the same time. The power parameters and the corresponding time-frequency diagram are trained using a dual-channel time-frequency model, and the fault arc or normal arc is predicted according to the extracted features; at the same time, the first warning event is deeply analyzed to evaluate the abnormality of the fault arc and execute the corresponding strategy; the second warning event is temporarily stored; wherein the extracted features include time, positive sequence component, negative sequence component, zero sequence component, voltage fluctuation signal and current fluctuation signal; Among them, the known early warning and regulatory mechanisms are: A first-class early warning supervision mechanism is matched for the first early warning event, and a calculation period T1 is set, and the number of times the power parameters are collected during the calculation period T1 is nc1; a second-class early warning supervision mechanism is matched for the second early warning event, and a calculation period T2 is set, and the number of times the power parameters are collected during the calculation period T2 is nc2; wherein nc1 and nc2 are positive integers; S4. Selective early warning report: early warning supervision mechanism category, early warning event category, abnormal assessment coefficient, and time.

2. The method for constructing a distribution cabinet fault arc warning model based on Fourier transform according to claim 1, characterized in that: The process of in-depth analysis of the first warning event is as follows: Step C1, evaluate the abnormality of the first warning event, obtain the parameter data set evaluated under the first warning supervision mechanism, and construct a feature space in a preset time period T3 to generate a corresponding abnormality evaluation coefficient PG; Step C2, presetting a standard threshold value YG2 of the abnormal evaluation coefficient PG, and generating an alarm signal when the abnormal evaluation coefficient PG exceeds the preset standard threshold value YG2.

3. The method for constructing a distribution cabinet fault arc warning model based on Fourier transform according to claim 2, characterized in that: The parameter data set includes voltage imbalance, current imbalance, the time between the occurrence time of the warning event and the current time, and the number of times the power parameters are collected during the time. The process of obtaining voltage imbalance and current imbalance is as follows: Firstly, the positive sequence component, negative sequence component and zero sequence component are extracted from the extracted features and defined as evaluation parameters; Then, a three-dimensional coordinate system is established with the positive sequence component F1 as the X-axis, the negative sequence component F2 as the Y-axis, and the zero sequence component F3 as the Z-axis, and a feature space S0 is constructed, a standard feature space S1 of the feature space S0 is preset, and the feature space S0 is compared with the standard feature space S1 to obtain their corresponding horizontal and vertical screenshots, and the total overlap area cmj and the total non-overlap area fmj are generated by comparison; wherein the total non-overlap area fmj is the fluctuation amount of the three-phase current or the three-phase voltage; Finally, the obtained total overlap area cmj and the total non-overlap area fmj are combined to obtain the corresponding voltage imbalance or current imbalance; According to the formula: ; Wherein, Dj is the calculated voltage imbalance or current imbalance, β0 is the weight correction coefficient, and β0 is greater than 0; Generate the abnormal assessment coefficient PG according to the formula: ; Wherein, cs represents the number of times the power parameters are collected, and cs≤nc1, ch is the time between the occurrence time of the warning event and the current time, and ε0 is the time constant.

4. The method for constructing a distribution cabinet fault arc warning model based on Fourier transform according to claim 3 is characterized in that: The second warning event is deeply analyzed according to the measurement cycle T2, and the process is processed according to steps C1 to C2.

Images