An intelligent monitoring method for internal faults in box-type substations

Abstract

The present invention relates to the field of power equipment monitoring and discloses a method for intelligent monitoring of internal faults in box-type substations. The method comprises: collecting electrical signals from the high- and low-voltage sides of the transformer, calculating the conductivity difference in the 1-10kHz frequency band and dynamically compensating for it through temperature-load; constructing a dynamic matrix model for magnetic leakage to update coefficients in real time and output the mean value of magnetic field changes; fusing the conductivity and magnetic field characteristics, generating a fault membership vector through fuzzy logic and inputting it into a deep confidence network, outputting the fault probability and triggering hierarchical control. The innovation lies in: high-frequency conductivity difference improves fault feature recognition, millisecond-level magnetic leakage updates capture early mechanical deformation, and fuzzy-deep learning synergistically enhances diagnostic robustness, overcoming the problems of misjudgment and missed detection in traditional monitoring. The present invention significantly improves the timeliness of fault warnings and is suitable for real-time monitoring of hidden dangers within transformers.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment status monitoring, and in particular to an intelligent monitoring method for internal faults of a box-type substation. Background Art

[0002] As core equipment for terminal power distribution in modern power systems, modular substations are widely used in urban power grids, industrial parks, and renewable energy access due to their compact structure and flexible deployment. The reliability of their core components, particularly transformers, is directly related to power supply security and system stability. However, transformers are susceptible to electrical, thermal, and mechanical stresses over long-term operation, leading to frequent internal faults such as inter-turn short circuits, insulation degradation, and core deformation. These faults have subtle initial characteristics but rapidly develop. If not detected in time, they can easily cause equipment damage or even grid failures. Therefore, real-time, accurate, and sensitive monitoring of the internal status of modular substations is crucial for ensuring the safe and stable operation of power systems.

[0003] The core bottleneck of current monitoring technology lies in the following: Traditional methods rely on a single parameter (using only power-frequency conductivity or static magnetic flux leakage testing), making it difficult to distinguish between conditions with highly similar electromagnetic characteristics, such as magnetizing inrush current and inter-turn short circuits. This leads to frequent misjudgments in practical applications. Furthermore, existing magnetic flux leakage models, with update speeds exceeding 100 milliseconds, are unable to capture the transient micro-deformations (millimeter-level displacements) caused by short-circuit electrodynamics at the initial stages of a fault, significantly increasing the risk of missed early fault detection. These two major flaws—misjudgments caused by a single parameter and missed detections due to response lags—combinedly limit the reliability of fault monitoring in box-type substations.

[0004] Therefore, this paper proposes an intelligent internal fault monitoring method for box-type substations. This method uses high-frequency conductivity differences to distinguish easily confused operating conditions, addressing the problem of misjudgment due to single parameter errors. It also utilizes millisecond-level updates of the magnetic flux leakage matrix to capture transient deformations, eliminating the risk of missed detections due to response lags. Furthermore, it integrates fuzzy logic and a deep belief network to output fault classification results. Summary of the Invention

[0005] The main purpose of the present invention is to provide an intelligent monitoring method for internal faults of a box-type substation, aiming to solve the technical problems in the prior art of high misjudgment rate caused by single fault detection parameters and missed early fault detection caused by dynamic response lag.

[0006] To achieve the above objectives, the present invention provides a box-type substation fault diagnosis method based on high-frequency conductance difference and dynamic magnetic leakage matrix, the method comprising:

[0007] Real-time collection of voltage and current signals on the high and low voltage sides of the transformer in the box-type substation, and extraction of conductance values ​​in the 1kHz and 10kHz frequency bands;

[0008] Calculate high-frequency conductance difference , where the conductivity value and After three-dimensional temperature-load dynamic correction;

[0009] Constructing a dynamic matrix model of leakage magnetic induction intensity , solve the Poisson equation with a 1ms period to update the matrix coefficients , and calculate the mean value of magnetic field change ;

[0010] Will and Input the fuzzy logic module to generate the fault membership vector, and then input the deep belief network classifier to output the fault probability;

[0011] Trigger hierarchical control based on the output failure probability:

[0012] When the fault probability is less than 0.3, a normal signal is output;

[0013] When the failure probability is ≥0.3 and ≤0.7, an early warning signal is output;

[0014] When the failure probability is greater than 0.7, an emergency stop command is output.

[0015] As a further improvement of the present invention, the three-dimensional temperature-load dynamic correction process includes:

[0016] The original conductance value is calculated based on the improved T-type equivalent circuit model. The calculation expression is:

[0017]

[0018] in, 、 are the effective values ​​of the high and low voltage side currents of the transformer respectively; 、 are the effective values ​​of the high and low voltage sides of the transformer respectively; is the high frequency leakage capacitor; is the angular frequency;

[0019] right Perform dynamic correction, the calculation expression is:

[0020]

[0021] in, is the reference temperature, Indicates the rated load rate, is the temperature coefficient, and is the load influence coefficient;

[0022] Extract the corrected conductance values ​​of 10kHz and 1kHz bands as and .

[0023] As a further improvement of the present invention, the updating of the matrix coefficient C in step S30 includes:

[0024] The matrix coefficients The Poisson equation is solved with a 1ms period, and its element calculation formula is:

[0025]

[0026] in, , characterizes the magnetic field response coefficient of the measuring point m under the current component n, is the equivalent inductance of the winding, is the width correction factor, and is the spatial harmonic coefficient, Axial length of winding;

[0027] When the coefficient deviation exceeds 10%, an early warning is triggered.

[0028] As a further improvement of the present invention, the mean value of the magnetic field change The calculation formula is:

[0029]

[0030] in, is the total number of measurement points, For the The initial fault-free reference value of the measuring point.

[0031] As a further improvement of the present invention, the fuzzy logic module performs the following operations:

[0032] definition Trapezoidal membership function: normal (-∞, 5%), warning [5%, 15%], fault (15%, +∞);

[0033] definition Trapezoidal membership function: normal (-∞, 0.1T), warning [0.1T, 0.3T], fault (0.3T, +∞);

[0034] Output preliminary failure probability based on 9 IF-THEN rules.

[0035] As a further improvement of the present invention, the fuzzy logic module includes:

[0036] Input layer, receiving ∈[0%,30%] and ∈[0T,0.5T];

[0037] a membership function layer, applying the trapezoidal membership function;

[0038] Rule base, stores 9 IF-THEN rules: IF for AND for THEN = ,in, , ∈{normal, warning, fault}, ∈{0.2,0.5,0.9};

[0039] At the output layer, the initial probability of failure is calculated using the centroid method: , .

[0040] As a further improvement of the present invention, the rule base is trained by the following steps:

[0041] Build an expert rule base based on IEC60599 fault diagnosis guidelines;

[0042] Using genetic algorithm to optimize the conclusion value of rules , the chromosome is encoded as a 9-dimensional real vector, and the classification accuracy is the fitness function;

[0043] The training data is injected with Gaussian noise with a signal-to-noise ratio of 20 dB and ±2 Hz power frequency interference, and is enhanced in the temperature range of −20°C to 60°C and the load rate range of 20% to 120%.

[0044] As a further improvement of the present invention, the vector input to the deep belief network is:

[0045] X=[ μ normal , μ warn , μ Fault , Δ G T , Δ B x , PCA 1 , PCA 2 , PCA 3 ]

[0046] Among them, μ is the fuzzy membership; The first three principal components of the magnetic flux leakage matrix with a cumulative variance contribution rate ≥ 95% after principal component analysis.

[0047] As a further improvement of the present invention, the deep belief network classifier includes:

[0048] Input layer, 8 nodes receive the vector ;

[0049] Hidden layer, 3-layer structure, the number of nodes in each layer is 256, 128, and 64 respectively, using ReLU activation function;

[0050] Output layer, 3 nodes output probability through Softmax;

[0051] Training constraints, L2 regularization coefficient , Dropout rate 0.5.

[0052] In addition, to achieve the above-mentioned purpose, the present invention also provides an intelligent monitoring device for internal faults of a box-type substation, comprising:

[0053] The data acquisition module is equipped with a 0.2-level voltage transformer, a current transformer, and a Butterworth second-order anti-aliasing filter, and supports 20kHz synchronous sampling.

[0054] Electromagnetic analysis module, integrating high-frequency conductivity difference calculation unit and leakage magnetic dynamic matrix solution unit, with calculation delay less than 10ms;

[0055] Multi-feature fusion module, based on FPGA, implements parallel computing of fuzzy logic and deep belief network, with classification delay less than 30ms;

[0056] Alarm storage module, equipped with three-level alarm output interface and SQLite local database.

[0057] As a further improvement of the present invention, the data acquisition module includes:

[0058] The data acquisition module includes a 0.2-level mutual inductor installed on the three phases of the high / low voltage side of the transformer, a six-channel anti-aliasing filter and a 20kHz synchronous sampling unit.

[0059] As a further improvement of the present invention, the electromagnetic-magnetic analysis module accelerates the solution of the Poisson equation through FPGA (update period 1ms) and is integrated into the Xilinx Zynq-7000 SoC chip, with a calculation delay of less than 8ms.

[0060] As a further improvement of the present invention, the multi-feature fusion module includes:

[0061] Fuzzy logic operator, pre-stored trapezoidal membership function parameters, supports parallel reasoning of 9 rules;

[0062] DBN accelerator, solidifies the pre-trained model, and uses sparse matrix multipliers to achieve 30% weight pruning;

[0063] It interacts with the electromagnetic-magnetic analysis module through the PCIe interface, and the classification delay is less than 25ms.

[0064] As a further improvement of the present invention, the alarm storage module includes:

[0065] Three-level alarm trigger:

[0066] Normal state, failure probability <0.3, output green LED signal;

[0067] In the early warning state, the fault probability is 0.3≤≤0.7, and the yellow LED and buzzer will output an intermittent alarm.

[0068] In emergency state, if the failure probability is greater than 0.7, the red LED and buzzer will continuously sound an alarm and send a shutdown command;

[0069] Feature extraction memory, from 、 The fault feature map is extracted from the output of the DBN hidden layer and compressed and stored in the SQLite database.

[0070] As a further improvement of the present invention, the alarm storage module performs:

[0071] Normal state, probability <0.3, triggering the green LED;

[0072] Warning state, 0.3≤probability≤0.7, triggering yellow LED and buzzer intermittent alarm;

[0073] Emergency state, probability>0.7, triggers the red LED, buzzer continuous alarm and shutdown command;

[0074] from , And the DBN hidden layer extracts the fault feature map and compresses and stores it.

[0075] In addition, to achieve the above-mentioned purpose, the present invention also provides an intelligent monitoring system for internal faults of a box-type substation, comprising: a memory, a processor, and a fault diagnosis program stored in the memory and runnable on the processor. When the fault diagnosis program is executed by the processor, the steps of the box-type substation fault diagnosis method based on high-frequency conductivity difference and dynamic magnetic leakage matrix as described above are implemented.

[0076] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, which includes a memory, a processor, and a fault diagnosis program stored on the memory and runnable on the processor. When the fault diagnosis is executed by the processor, the steps of the box-type substation fault diagnosis method based on high-frequency conductivity difference and dynamic magnetic leakage matrix as described above are implemented.

[0077] The technical solution provided by the present invention can have the following beneficial effects:

[0078] During use, the present invention fundamentally solves the core defect of traditional monitoring technology, which coexists with misjudgment and missed detection, through the collaborative innovation of high-frequency conductivity dynamic compensation and real-time leakage magnetic field sensing. Its core value lies in: the fault fingerprint extraction mechanism constructed based on high-frequency conductivity difference significantly enhances the ability to distinguish between inter-turn short circuits and electromagnetic transient processes, and completely overcomes the stubborn problem of misjudgment caused by a single power frequency parameter; relying on the dynamic leakage magnetic field matrix updated at the millisecond level, it realizes the instantaneous capture of tiny deformations in the winding, and improves the early fault detection time to a hundred times that of traditional methods, effectively eliminating the risk of missed detection caused by response lag; the innovative fusion architecture of fuzzy logic and deep belief network retains the interpretability of physical characteristics and endows the intelligent identification capability of complex fault modes, maintaining stable diagnostic accuracy under extreme temperature and load fluctuations; the entire system is optimized at the hardware level of the edge computing platform, while meeting the stringent real-time requirements of power equipment, significantly reducing deployment costs, and ultimately forming a complete technical system for accurate fault warning, rapid intervention, and closed-loop management, providing reliable protection for the safe operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention with reference to the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present invention.

[0080] Figure 1 Schematic diagram of the hardware operating environment of the box-type substation fault real-time diagnosis system involved in an embodiment of the present invention;

[0081] Figure 2 This is a flow chart of a first embodiment of the intelligent monitoring method for internal faults in a box-type substation according to the present invention;

[0082] Figure 3 This is a schematic diagram of the overall process of the second embodiment of the intelligent monitoring method for internal faults of a box-type substation according to the present invention;

[0083] Figure 4 A graph showing the conductance change and magnetic flux leakage offset of the second embodiment of the intelligent monitoring method for internal faults in a box-type substation according to the present invention;

[0084] Figure 5 This is a flow chart of dynamic updating of the magnetic flux leakage matrix according to the second embodiment of the intelligent monitoring method for internal faults of a box-type substation of the present invention;

[0085] Figure 6 Diagnostic interface of the second embodiment of the intelligent monitoring method for internal faults of box-type substations of the present invention Figure 1 ;

[0086] Figure 7 Diagnostic interface of the second embodiment of the intelligent monitoring method for internal faults of box-type substations of the present invention Figure 2 ;

[0087] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0088] The present invention provides an intelligent monitoring method for internal faults of box-type substations. It can solve the industry bottleneck of high misjudgment rate due to single parameter and missed detection of early faults caused by response lag in box-type substation fault monitoring by high-frequency conductivity difference. Dynamic compensation in the 1-10kHz frequency band improves insulation fault sensitivity, combined with millisecond-level real-time updates of the leakage magnetic induction intensity matrix (solving the Poisson equation in a 1ms period) to capture mechanical deformation characteristics, and integrates fuzzy logic and deep belief network algorithms to achieve accurate fault classification. Ultimately, the system achieved breakthrough results in reducing the misjudgment rate from 28%-40% to <5%, the early fault missed detection rate from >15% to <3%, and the response delay to less than 30ms. At the same time, it significantly reduced deployment costs based on standard PT / CT reuse.

[0089] First embodiment

[0090] To better understand the above technical solutions, exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0091] As an implementation solution, Figure 1 This is a schematic diagram of the architecture of the hardware operating environment of the intelligent monitoring system for internal faults of a box-type substation involved in an embodiment of the present invention.

[0092] like Figure 1As shown, the intelligent monitoring system for internal faults in a box-type substation may include: a processor 1001, such as a CPU, a memory 1005, a user interface 1003, a network interface 1004, and a communication bus 1002. Communication bus 1002 is used to facilitate communication between these components. User interface 1003 may include a display and an input unit, such as a keyboard. Optionally, user interface 1003 may also include a standard wired interface or a wireless interface. Network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). Memory 1005 may be high-speed RAM or non-volatile memory, such as disk storage. Memory 1005 may also be a storage device independent of processor 1001.

[0093] Those skilled in the art will understand that Figure 1 The architecture of the box-type substation internal fault intelligent monitoring system shown in the figure does not constitute a limitation on the box-type substation internal fault intelligent monitoring system, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0094] like Figure 1 As shown, memory 1005, a storage medium, may include an operating system, a network communication module, a user interface module, and a real-time diagnostic program for box-type substation faults based on high-frequency conductance difference and dynamic magnetic leakage matrix. The operating system is a program that manages and controls the hardware and software resources of the box-type substation fault diagnosis system, and the real-time diagnostic program for box-type substation faults based on high-frequency conductance difference and dynamic magnetic leakage matrix, as well as other software or programs.

[0095] exist Figure 1 In the box-type substation internal fault intelligent monitoring system shown, the user interface 1003 is mainly used to connect to the terminal and communicate data with the terminal; the network interface 1004 is mainly used for the background server and communicates data with the background server; the processor 1001 can be used to call the box-type substation internal fault intelligent monitoring program stored in the memory 1005.

[0096] In this embodiment, the box-type substation internal fault intelligent monitoring system includes: a memory 1005, a processor 1001, and a box-type substation internal fault intelligent monitoring program stored in the memory and executable on the processor, wherein:

[0097] When the processor 1001 calls the real-time fault diagnosis program for box-type substations based on high-frequency conductance difference and dynamic magnetic leakage matrix stored in the memory 1005, the following operations are performed:

[0098] Real-time collection of voltage and current signals on the high and low voltage sides of the transformer in the box-type substation, and extraction of conductance values ​​in the 1kHz and 10kHz frequency bands;

[0099] Calculate high-frequency conductance difference , where the conductivity value and After three-dimensional temperature-load dynamic correction;

[0100] Constructing a dynamic matrix model of leakage magnetic induction intensity , solve the Poisson equation with a 1ms period to update the matrix coefficients , and calculate the mean value of magnetic field change ;

[0101] Will and Input the fuzzy logic module to generate the fault membership vector, and then input the deep belief network classifier to output the fault probability;

[0102] Trigger hierarchical control based on the output failure probability:

[0103] When the fault probability is less than 0.3, a normal signal is output;

[0104] When the failure probability is ≥0.3 and ≤0.7, an early warning signal is output;

[0105] When the failure probability is greater than 0.7, an emergency stop command is output.

[0106] Based on the hardware architecture of the above-mentioned box-type substation internal fault intelligent monitoring system, an embodiment of the box-type substation internal fault intelligent monitoring method of the present invention is proposed.

[0107] Reference Figure 2 In the first embodiment, the method for intelligently monitoring internal faults of a box-type substation includes the following steps:

[0108] Step S10, collecting the voltage and current signals of the high and low voltage sides of the transformer in the box-type substation in real time, and extracting the conductance values ​​in the 1kHz and 10kHz frequency bands;

[0109] In this embodiment, 0.2-level PT / CTs are installed on the high-voltage side (A / B / C phases) and low-voltage side (a / b / c phases) of the box-type substation transformer. The signals are pre-processed by a six-channel Butterworth second-order anti-aliasing filter (cutoff frequency 10kHz), and voltage and current data are collected synchronously at a sampling rate of 20kHz. FFT is used to separate the fundamental wave and the 2nd to 5th harmonic components, and the data are stored in the edge computing node (equipped with Xilinx Zynq-7000 SoC). This ensures the integrity of the high-frequency signal. Calculation provides a basis.

[0110] Step S20, calculating the high frequency conductivity difference , where the conductivity value and Corrected by three-dimensional temperature-load dynamics.

[0111] In this embodiment, based on the improved T-type equivalent circuit model, the branch conductance value is calculated and susceptance , when an internal fault occurs, Significantly increased unchanged, when the excitation inrush current occurs, Increased The calculation formula is unchanged:

[0112]

[0113] in, It is a high-frequency leakage capacitor, which is determined by offline frequency sweep calibration and temperature compensation. Sine excitation is applied in the 1-10kHz frequency band and the capacitance is measured at different frequencies. The imaginary part , fitting The mean of 、 is the effective value of the high and low voltage side current (measured by CT); 、 is the effective value of the high and low voltage side voltage (measured by PT); is the angular frequency, (f is the signal frequency);

[0114] Furthermore, the corrected conductance value is obtained through the three-dimensional correction model, and the calculation expression is:

[0115]

[0116] in, represents the corrected conductivity value, is the original conductance value, and the extraction formula is = , that is, through Calculated, and They are the temperature correction factor and the load rate correction factor respectively.

[0117] Furthermore, the temperature correction factor The determination is based on an in-depth study of the conductivity offset at different temperatures. Through a large number of rigorous experiments, the conductivity offset data corresponding to each temperature is accurately calibrated and mathematically fitted into an exponential function:

[0118]

[0119] In this formula, It is the inherent temperature coefficient of the material, reflecting the sensitivity of the material's conductivity to temperature changes; Represents the reference temperature, usually set to 25°C, which is used as the basis for calculating the temperature correction factor.

[0120] Furthermore, the load factor correction factor The construction of fully considers the impact of load rate on leakage current. Based on this, the strategy of piecewise linear correction is adopted, and its specific expression is:

[0121]

[0122] in, Indicates the rated load rate, and is the load influence coefficient, which corresponds to the degree of influence of load rate change on conductance correction when the load rate is less than or equal to the rated load rate and when it is greater than the rated load rate.

[0123] Step S30: Constructing a dynamic matrix model of leakage magnetic induction intensity , solve the Poisson equation with a 1ms period to update the matrix coefficients , and calculate the mean value of magnetic field change ;

[0124] In this embodiment, a dynamic matrix model of leakage magnetic induction intensity at multiple measuring points is constructed, and the radial leakage magnetic induction intensity at each measuring point is solved by Poisson's equation ( ), when the winding is deformed or short-circuited between turns, the matrix coefficient When an offset occurs, the model expression is:

[0125]

[0126] in, , characterizes the magnetic field response coefficient of the measuring point m under the current component n; : Related to the winding inductance, corresponding to the equivalent inductance value in different harmonic modes; : The coefficient of the harmonic order of different current components at different winding equivalent widths; , : is the spatial harmonic coefficient, calibrated by experiment; b: is the axial length of the winding.

[0127] Furthermore, the leakage flux offset ( ): Based on the dynamic model of the leakage magnetic matrix, the mean value of the magnetic field change at each measuring point is calculated. The calculation expression is:

[0128]

[0129] in, is the total number of measurement points, Indicates the magnetic field reference value in the initial fault-free state.

[0130] Step S40: and The fuzzy logic module is input to generate the fault membership vector, which is then input into the deep belief network classifier to output the fault probability.

[0131] In this embodiment, the fuzzy logic module performs the following operations:

[0132] definition Trapezoidal membership function: normal (-∞, 5%), warning [5%, 15%], fault (15%, +∞);

[0133] definition Trapezoidal membership function: normal (-∞, 0.1T), warning [0.1T, 0.3T], fault (0.3T, +∞);

[0134] Output preliminary failure probability based on 9 IF-THEN rules.

[0135] Among them, the change in conductance ( ): It is obtained by calculating the difference of high-frequency conductivity spectrum. The specific formula is: , and The conductance values ​​corresponding to the 10kHz and 1kHz frequency bands are obtained by combining step S10 with step S20. ) is obtained through step S30.

[0136] Furthermore, the fuzzy logic module includes:

[0137] Input layer, receiving ∈[0%,30%] and ∈[0T,0.5T];

[0138] a membership function layer, applying the trapezoidal membership function;

[0139] Rule base, stores 9 IF-THEN rules: IF for AND for THEN = ,in, , ∈{normal, warning, fault}, ∈{0.2,0.5,0.9};

[0140] At the output layer, the initial probability of failure is calculated using the centroid method: , .

[0141] Furthermore, the rule base is trained by the following steps:

[0142] Build an expert rule base based on IEC60599 fault diagnosis guidelines;

[0143] Using genetic algorithm to optimize the conclusion value of rules , the chromosome is encoded as a 9-dimensional real vector, and the classification accuracy is the fitness function;

[0144] The training data is injected with Gaussian noise with a signal-to-noise ratio of 20 dB and ±2 Hz power frequency interference, and is enhanced in the temperature range of −20°C to 60°C and the load rate range of 20% to 120%.

[0145] Furthermore, the vector input to the deep belief network is:

[0146] X=[ μ normal , μ warn , μ Fault , Δ G T , Δ B x , PCA 1 , PCA 2 , PCA 3 ]

[0147] Among them, μ is the fuzzy membership; The first three principal components of the magnetic flux leakage matrix with a cumulative variance contribution rate ≥ 95% after principal component analysis.

[0148] Furthermore, the deep belief network classifier includes:

[0149] Input layer, 8 nodes receive the vector ;

[0150] Hidden layer, 3-layer structure, the number of nodes in each layer is 256, 128, and 64 respectively, using ReLU activation function;

[0151] Output layer, 3 nodes output probability through Softmax;

[0152] Training constraints, L2 regularization coefficient , Dropout rate 0.5.

[0153] Step S50: triggering hierarchical control according to the output failure probability:

[0154] When the fault probability is less than 0.3, a normal signal is output;

[0155] When the failure probability is ≥0.3 and ≤0.7, an early warning signal is output;

[0156] When the failure probability is greater than 0.7, an emergency stop command is output.

[0157] In this embodiment, the alarm and storage module triggers three levels of alarms (normal / warning / emergency) and uses an SQLite database to store fault feature data.

[0158] The present invention fundamentally solves the core defect of traditional monitoring technology, which suffers from both misjudgment and missed detection, through the collaborative innovation of high-frequency conductivity dynamic compensation and real-time leakage magnetic field sensing. Its core value lies in: the fault fingerprint extraction mechanism constructed based on high-frequency conductivity difference significantly enhances the ability to distinguish between inter-turn short circuits and electromagnetic transient processes, and completely overcomes the stubborn problem of misjudgment caused by a single power frequency parameter; relying on the dynamic leakage magnetic field matrix updated at the millisecond level, it realizes the instantaneous capture of tiny deformations in the winding, and improves the early fault detection time to a hundred times that of traditional methods, effectively eliminating the risk of missed detection caused by response lag; the innovative fusion architecture of fuzzy logic and deep confidence network retains the interpretability of physical characteristics and gives the intelligent identification capability of complex fault modes, maintaining stable diagnostic accuracy under extreme temperature and load fluctuations; the entire system is optimized at the hardware level of the edge computing platform, meeting the stringent real-time requirements of power equipment while significantly reducing deployment costs, ultimately forming a complete technical system for accurate fault warning, rapid intervention, and closed-loop management, providing reliable protection for the safe operation of the power grid.

[0159] Second embodiment

[0160] Based on the method proposed in the above embodiment, in the actual production of box-type substation transformer monitoring, the system strictly follows Figure 3 The overall process shown here performs fault diagnosis. High-precision transformers installed on the high and low voltage sides collect electrical signals in real time, which are then transmitted to the edge computing platform after anti-aliasing filtering. A dynamic temperature-load compensation model processes the 1kHz / 10kHz frequency band conductivity values ​​(eliminating interference from high temperatures or overloads in summer) to generate the core indicator, the high-frequency conductivity difference. .like Figure 4 As shown in the upper curve, this difference can keenly capture insulation anomalies (such as A sudden increase of more than 15% indicates a risk of inter-turn short circuit).

[0161] Furthermore, the synchronous operation of the magnetic flux leakage monitoring system (perform Figure 5The dynamic update process shown in the figure solves the electromagnetic field equation every millisecond through the FPGA chip. When a sudden change in the magnetic field caused by a small displacement of the winding (such as a 0.2T instantaneous offset) is detected, the mean value of the magnetic field change is immediately calculated. ( Figure 4 Its millisecond-level response capability successfully identifies early mechanical deformations that are missed by traditional methods.

[0162] Furthermore, the conductivity and magnetic field data are fed into the fusion diagnosis core, and the fuzzy logic module will and Mapping to three-level membership (such as The "fault" membership degree is triggered by exceeding 0.3T), and a preliminary diagnosis is generated by combining 9 expert rules; the deep belief network fuses the fuzzy output and the main component of magnetic leakage to output the final probability through a three-layer neural network. Figure 6 and Figure 7 The diagnostic interface shown displays the process in real time, including the rule triggering status and probability evolution curve.

[0163] Furthermore, the system links the interface and hardware according to the probability value. When the probability is less than 0.3, the interface remains green and the green LED lights up; when the probability reaches the range of 0.3-0.7, the interface switches to yellow and triggers an audible and visual alarm; for emergency conditions with a probability greater than 0.7 (such as severe inter-turn short circuit), the interface turns red and executes a shutdown command, while storing the fault characteristic map.

[0164] Industrial validation has shown that this solution reduces the false positive rate from 35% to 4.2%, achieves a 97% early fault detection rate, and effectively prevents accidents from escalating with a 30-millisecond response time. By reusing existing PT / CT equipment, the cost of intelligent transformation has been reduced by over 40%.

[0165] Furthermore, those skilled in the art will appreciate that all or part of the steps in the method of the above-described embodiment can be implemented by instructing the relevant hardware through a computer program. The computer program includes program instructions, which can be stored in a computer-readable storage medium. The program instructions are executed by at least one processor in the real-time fault diagnosis system for box-type substations to implement the steps in the above-described method embodiment.

[0166] Therefore, the present invention also provides a computer-readable storage medium, which stores a real-time diagnosis program for box-type substation faults. When the real-time diagnosis program for box-type substation faults is executed by a processor, it implements the various steps of the intelligent monitoring method for internal faults of box-type substations based on high-frequency conductivity difference and dynamic magnetic leakage matrix as described in the above embodiment.

[0167] The computer-readable storage medium may be any computer-readable storage medium that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0168] It should be noted that since the storage medium provided in the embodiments of the present invention is the storage medium used to implement the method of the embodiments of the present invention, the specific structure and variations of the storage medium will be understood by those skilled in the art based on the method described in the embodiments of the present invention, and therefore will not be described in detail here. All storage media used in the methods of the embodiments of the present invention fall within the scope of protection of the present invention.

[0169] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0170] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0171] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0172] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0173] It should be noted that in the claims, any reference signs placed between parentheses shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claim. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The use of the words first, second, third etc. does not indicate any order. These words may be interpreted as names.

[0174] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0175] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for intelligent monitoring of internal faults in a box-type substation, characterized in that: The following steps are involved: Real-time collection of voltage and current signals on the high and low voltage sides of the transformer in the box-type substation, and extraction of conductance values ​​in the 1kHz and 10kHz frequency bands; Calculate high-frequency conductance difference , where the conductivity value and After three-dimensional temperature-load dynamic correction; Constructing a dynamic matrix model of leakage magnetic induction intensity , solve the Poisson equation with a 1ms period to update the matrix coefficients , and calculate the mean value of magnetic field change ; Will and Input the fuzzy logic module to generate the fault membership vector, and then input the deep belief network classifier to output the fault probability; Trigger hierarchical control based on the output failure probability: When the fault probability is less than 0.3, a normal signal is output; When the failure probability is ≥0.3 and ≤0.7, an early warning signal is output; When the failure probability is greater than 0.7, an emergency stop command is output.

2. The method according to claim 1, characterized in that The three-dimensional temperature-load dynamic correction process includes: The original conductance value is calculated based on the improved T-type equivalent circuit model. The calculation expression is: in, 、 are the effective values ​​of the high and low voltage side currents of the transformer respectively; 、 are the effective values ​​of the high and low voltage sides of the transformer respectively; is the high frequency leakage capacitor; is the angular frequency; Perform dynamic correction, the calculation expression is: in, =25℃ is the reference temperature, Indicates the rated load rate, is the temperature coefficient, and is the load influence coefficient; extract the corrected conductance values ​​of 10kHz and 1kHz frequency bands as and .

3. The method according to claim 1, characterized in that The update formula of the matrix coefficient C is: , in, is the equivalent inductance of the winding, is the width correction factor, and is the spatial harmonic coefficient, Winding axial length; when the coefficient offset exceeds 10%, an early warning is triggered.

4. The method according to claim 1, wherein The mean value of the magnetic field change The calculation formula is: in, is the total number of measurement points, For the The initial fault-free reference value of the measuring point.

5. The method according to claim 1, characterized in that The fuzzy logic module performs the following operations: right Define the trapezoidal membership function: normal (-∞, 5%), warning [5%, 15%], fault (15%, +∞); right Define the trapezoidal membership function: normal (-∞, 0.1T), warning [0.1T, 0.3T], fault (0.3T, +∞); Output preliminary failure probability based on 9 IF-THEN rules.

6. The method according to claim 5, characterized in that The fuzzy logic module includes: Input layer, receiving ∈[0%,30%] and ∈[0T,0.5T]; a membership function layer, applying the trapezoidal membership function; Rule base, stores 9 IF-THEN rules: IF for AND for THEN = ,in, , ∈{normal, warning, fault}, ∈{0.2,0.5,0.9}; At the output layer, the initial probability of failure is calculated using the centroid method: , .

7. The method according to claim 6, characterized in that The rule base is trained by the following steps: Build an expert rule base based on IEC60599 fault diagnosis guidelines; Using genetic algorithm to optimize the conclusion value of rules , the chromosome is encoded as a 9-dimensional real vector, and the classification accuracy is the fitness function; The training data is injected with Gaussian noise with a signal-to-noise ratio of 20 dB and ±2 Hz power frequency interference, and is enhanced in the temperature range of −20°C to 60°C and the load rate range of 20% to 120%.

8. The method according to claim 1, characterized in that The vector input to the deep belief network is: Among them, μ is the fuzzy membership; The first three principal components of the magnetic flux leakage matrix with a cumulative variance contribution rate ≥ 95% after principal component analysis.

9. The method according to claim 8, characterized in that The deep belief network classifier includes: Input layer, 8 nodes receive the vector ; Hidden layer, 3-layer structure, the number of nodes in each layer is 256, 128, and 64 respectively, using ReLU activation function; Output layer, 3 nodes output probability through Softmax; Training constraints, L2 regularization coefficient , Dropout rate 0.5.

Images