Substation equipment grounding fault early warning method based on deep learning and storage medium

Abstract

The invention discloses a substation equipment grounding fault early warning method based on deep learning and a storage medium, and the method comprises the steps: arranging a plurality of electromagnetic monitoring points around a substation, and collecting the direction, intensity and frequency spectrum parameters of electromagnetic radiation; based on an electromagnetic scattering theory and an inverse problem solving method, constructing an electromagnetic environment model under normal operation and fault states by using the collected data, mining specific electromagnetic characteristics of the grounding fault, and quantifying to form a characteristic data set; training a deep learning model by using the feature data set; in the actual operation process of the transformer substation, electromagnetic environment data are collected in real time, the electromagnetic environment model at the current moment and the specific electromagnetic characteristics of the corresponding ground fault are obtained, and whether the transformer substation equipment is in the ground fault state or not is predicted and judged through the model. According to the invention, by using the electromagnetic environment data of the transformer substation, the specific electromagnetic characteristics of the ground fault are mined, and the accuracy and timeliness of early warning of the ground fault are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of fault warning technology, and in particular to a substation equipment grounding fault warning method and storage medium based on deep learning. Background Art

[0002] As a key facility in the power system, the safe and stable operation of substations is crucial to ensuring power supply. The electromagnetic environment inside substations is complex. Various electrical equipment will generate electromagnetic radiation during operation, forming a unique electromagnetic field. When a ground fault occurs in substation equipment, this electromagnetic environment will change significantly, producing specific electromagnetic characteristics. Therefore, by monitoring and analyzing the electromagnetic environment data of substations, effective early warning of equipment ground faults can be achieved.

[0003] Traditional detection and early warning methods for substation equipment grounding faults have certain limitations. They mainly rely on single electrical parameter monitoring, such as current and voltage. These methods often ignore the fault information contained in changes in the electromagnetic environment. Due to the failure to fully utilize electromagnetic environment data, these methods are insufficient in terms of fault detection timeliness and early warning accuracy. In addition, although there are some studies on electromagnetic environment monitoring, most of them remain at the level of data collection and simple analysis, and fail to deeply explore the intrinsic connection between electromagnetic environment data and grounding faults. In particular, the method of using substation electromagnetic environment reconstruction and combining deep learning to achieve early warning of grounding faults has not been widely used in existing patents and technologies, resulting in less than ideal early warning effects for substation equipment grounding faults.

[0004] The existing substation equipment grounding fault early warning method is optimized. By utilizing the substation electromagnetic environment data, the unique electromagnetic characteristics of grounding faults are mined to achieve accurate and early warning of grounding faults. Therefore, it is of great significance to develop a substation equipment grounding fault early warning method based on deep learning that can comprehensively realize the above characteristics. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a substation equipment grounding fault early warning method and storage medium based on deep learning. By utilizing the substation electromagnetic environment data, the unique electromagnetic characteristics of the grounding fault are mined, and the accuracy and timeliness of the grounding fault early warning are significantly improved.

[0006] To solve the above technical problems, as one aspect of the present invention, 1. A substation equipment grounding fault early warning method based on deep learning is provided, characterized in that it includes the following steps:

[0007] Step S10: Arrange multiple electromagnetic monitoring points around the substation to collect the direction, intensity, and spectrum parameters of electromagnetic radiation;

[0008] Step S11: Based on electromagnetic scattering theory and inverse problem solving methods, the collected data is used to construct electromagnetic environment models under normal operation and fault conditions, and the unique electromagnetic characteristics of ground faults are mined through comparative analysis and quantified to form a feature data set.

[0009] Step S12, using the quantized feature data set to train a deep learning model, and improving the model prediction accuracy by adjusting model parameters and optimizing algorithms;

[0010] Step S13: During the actual operation of the substation, electromagnetic environment data is collected in real time, and the electromagnetic environment model at the current moment and the corresponding electromagnetic characteristics unique to the ground fault are obtained. The model is used to predict whether the substation equipment is in a ground fault state.

[0011] Preferably, in step S10, arranging a plurality of electromagnetic monitoring points around the substation in a predetermined layout further includes:

[0012] Assume that the substation area is a rectangular area on a two-dimensional plane with a length of L and a width of W. It is planned to arrange N electromagnetic monitoring points. The monitoring point layout optimization formula based on the principle of electromagnetic coverage uniformity is used to determine the location of the monitoring points. The uniformity index U of the monitoring point layout is defined as follows:

[0013]

[0014] Among them, d ij It represents the distance between the i-th monitoring point and the j-th monitoring point, which is obtained by the following calculation formula:

[0015] Among them, (x i ,y i ) and (x j ,y j ) are the coordinates of the i-th monitoring point and the j-th monitoring point on the two-dimensional plane. By adjusting the coordinate positions of the monitoring points, the uniformity index U reaches the minimum value, thereby achieving the optimal layout of the monitoring points around the substation.

[0016] Preferably, the step S11 further includes:

[0017] Based on electromagnetic scattering theory and inverse problem solving methods, the electromagnetic environment data collected during normal operation and before and after a fault occurs are comprehensively processed. For the data under normal operation, mathematical algorithms and model building methods are used to establish a benchmark model that can accurately characterize the normal electromagnetic environment characteristics of the substation;

[0018] Based on the collected data before and after the fault, a model is constructed based on electromagnetic scattering theory and inverse problem solving methods to reflect the changes in the electromagnetic environment caused by interference when the fault occurs, including changes in the direction of electromagnetic radiation, abnormal fluctuations in intensity, and new characteristics of the spectrum.

[0019] In the process of constructing the above two models, the data differences in the model construction process under normal operation and fault conditions and the differences in the model characteristics formed are directly compared and analyzed to mine the unique electromagnetic characteristics of grounding faults. The mined unique electromagnetic characteristics of grounding faults are quantified and converted into a data format that can be recognized and processed by the computer system to form a feature data set.

[0020] Preferably, in step S11, for the data under normal operating conditions, a reference model capable of accurately characterizing the normal electromagnetic environment characteristics of the substation is established through mathematical algorithms and model building means, further comprising:

[0021] Assume that there are M devices in the substation, each device can be equivalent to a point charge source, and the equivalent point charge of the i-th device is Q i , whose coordinates in space are (x i ,y i ,z i ), for any point in space, let its coordinates be (x; y, z) and the electric field intensity at that point be According to electromagnetic scattering theory, the electric field strength is calculated using the following formula:

[0022]

[0023] in, is the vector pointing from the equivalent point charge position of the i-th device to the point in space where the electric field strength is to be determined, yes The unit vector of ,∈0 is the dielectric constant of vacuum;

[0024] The magnetic field strength is calculated using the following formula:

[0025]

[0026] in, It is the equivalent motion velocity of the equivalent point charge of the i-th device, and the state of the electromagnetic environment under normal operating conditions is described by calculating the electric field strength and magnetic field strength.

[0027] Preferably, in step S11, for the data collected before and after the fault occurs, a model is constructed based on electromagnetic scattering theory and an inverse problem solving method, further comprising:

[0028] Assume that the coordinates in the substation are (x f,y f ,z f ) occurs, the fault point is regarded as a new equivalent charge source, and its charge is Q f , and the fault will cause the dielectric constant of the electromagnetic medium around the fault point to become ∈ f , the conductivity becomes σ f , for any point in space, let its coordinates be (x, y, z) and the electric field intensity at that point be According to electromagnetic scattering theory, the electric field strength after considering the fault effect is calculated using the following formula:

[0029]

[0030] in, is the vector pointing from the fault point to the point in space where the electric field strength is to be determined, yes The principal unit vector of is the vector pointing from the equivalent point charge position of the i-th device to the point in space where the electric field strength is to be determined, yes The unit vector of

[0031] Regarding the magnetic field strength, according to the electromagnetic scattering theory, the magnetic field strength after considering the fault effect is calculated using the following formula:

[0032]

[0033] in, It is the equivalent movement speed of the equivalent charge source at the fault point. The state of the electromagnetic environment is described by calculating the electric field strength and magnetic field strength, combined with the actual layout of the equipment in the substation, the equivalent charge of the equipment, and the location and characteristics of the fault point.

[0034] Preferably, in step S11, by directly comparing and analyzing the data differences in the model construction process under normal operation and fault state and the differences in model features formed, the unique electromagnetic features of the ground fault are mined, specifically:

[0035] Assume that the electromagnetic radiation intensity of a monitoring point at a certain time t1 under normal operating conditions is I n (t1), the electromagnetic radiation intensity of the same monitoring point at time t2 after the fault occurs is I f (t2), the rate of change of electromagnetic radiation intensity at the monitoring point ΔI is calculated using the following formula: rate :

[0036]

[0037] And calculate the change rate of electromagnetic radiation intensity at each monitoring point. If ΔI rate Greater than the preset threshold ΔIMax , it can be preliminarily determined that there are electromagnetic characteristic changes related to ground faults in the area where the monitoring point is located.

[0038] Preferably, the step S12 further includes:

[0039] A deep learning model is constructed, and the quantified electromagnetic feature dataset unique to ground faults and the corresponding substation operating status labels are divided into training set, validation set and test set in proportion. The deep learning model is trained using the training set. By adjusting the model parameters, a loss function is used to measure the difference between the model prediction results and the actual labels during the training process, and the model parameters are optimized using an optimization algorithm based on the difference.

[0040] Preferably, in step S12, constructing a deep learning model using a neural network algorithm further includes:

[0041] Assume that the input electromagnetic feature dataset image is X, whose size is M×N, the convolution kernel is K, whose size is m×n, and the step size is s, then the convolution layer output Y is calculated using the following formula:

[0042]

[0043] Where X represents the electromagnetic feature dataset input to the convolution layer, presented in image form, and contains the unique electromagnetic feature information of the ground fault obtained from the electromagnetic environment model reconstruction and fault feature mining steps. K is the convolution kernel, whose element values are trainable weight parameters. The weights are continuously adjusted during the training process to extract local features of different scales in the input data. l is the position index of the convolution kernel K on a certain dimension of itself. M and N are the number of rows and columns of the input data image X, respectively, which are used to determine the size of the input data. m and n are the number of rows and columns of the convolution kernel, respectively, which are used to determine the size of the convolution kernel. Y is the output of the convolution layer. After the convolution operation, the new data with extracted local features is obtained, which will be further processed as the input of the pooling layer.

[0044] For the calculation of the pooling layer, let the convolution layer output be Y, whose size is P×Q, the pooling window size is p×q, and the step size is t. Then the pooling layer output Z is calculated using the following formula:

[0045]

[0046] Among them, Y is the input data of the pooling layer, that is, the output result of the convolutional layer, p and q are the number of rows and columns of the pooling window, respectively, which are used to determine the size of the pooling window, and Z is the output result of the pooling layer. The result obtained after the pooling operation will be used as the input of the fully connected layer to continue row processing;

[0047] For the calculation of the fully connected layer, let the output of the pooling layer be Z, whose size is R×S, and let Z ij For the elements in the pooling layer output data, is the connection weight between the kth neuron in the fully connected layer and the output element of the pooling layer, b k is the bias of the kth neuron, then the output a of the kth neuron in the fully connected layer is calculated using the following formula: k :

[0048]

[0049] Among them, Z is the input data of the fully connected layer, that is, the output result of the pooling layer, a k It is the output result of the kth neuron in the fully connected layer. After the fully connected layer operation, the new data with integrated features is obtained, which will be further processed as the input of the output layer. The electromagnetic feature data is gradually processed in depth to achieve the purpose of identifying the fault state.

[0050] For the output layer calculation, let the output of the fully connected layer be a k , then the output layer output y is calculated using the following formula:

[0051]

[0052] Among them, a k It is the output result of the fully connected layer, which contains the new data after the features are integrated by the fully connected layer. y is the output result of the output layer, and its value is between 0 and 1. When y ≥ 0.5, it is judged to be a fault state, and when y < 0.5, it is judged to be a normal operation state.

[0053] Preferably, in step S12, a loss function is used to measure the difference between the model prediction result and the actual label, further comprising:

[0054] Suppose there are N samples in the training set, and for the i-th sample, its actual label is y i , the model prediction result is The following formula is used to calculate the cross entropy loss function L:

[0055]

[0056] Among them, N represents the number of samples in the training set, which is used to determine the range of samples that need to be traversed when calculating the cross entropy loss function, and y i is the actual label of the i-th sample, is the model prediction result of the i-th sample. The predicted value obtained after the model processes the input electromagnetic feature data is used to compare with the actual label. L is the calculation result of the cross entropy loss function. The larger its value, the greater the difference between the model prediction result and the actual label.

[0057] As another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0058] The implementation of the present invention has the following beneficial effects:

[0059] This invention provides a deep learning-based ground fault early warning method and storage medium for substation equipment. By deploying multiple electromagnetic monitoring points around the substation, comprehensive electromagnetic environment data is collected. Using electromagnetic scattering theory and inverse problem-solving methods, the electromagnetic environment model is reconstructed and fault characteristics are mined, accurately capturing changes in the substation's electromagnetic environment. By analyzing and processing these characteristics using a deep learning model, the complex relationship between electromagnetic characteristics and substation operating status can be automatically learned, significantly improving the accuracy and timeliness of ground fault early warnings.

[0060] This embodiment of the present invention determines whether substation equipment is in a ground fault state and provides detailed early warning information. This allows operations and maintenance personnel to quickly locate the fault point based on this warning information and take targeted measures to address it, thereby significantly improving the efficiency of fault handling. Furthermore, through the training and optimization of deep learning models, the timeliness and accuracy of early warnings are further improved, providing a strong technical guarantee for the safe operation of substations and effectively enhancing the safety and reliability of substation operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those with ordinary skill in the art, other drawings derived from these drawings without inventive effort still fall within the scope of the present invention.

[0062] Figure 1 This is a main flow chart of an embodiment of a substation equipment grounding fault early warning method based on deep learning provided by the present invention;

[0063] Figure 2 for Figure 1 A more detailed flowchart of

[0064] Figure 3 for Figure 2 A more detailed flowchart of the electromagnetic environment model reconstruction and fault feature mining steps in . DETAILED DESCRIPTION

[0065] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be described in further detail below with reference to the accompanying drawings.

[0066] like Figure 1 As shown, it shows a main process diagram of an embodiment of a substation equipment grounding fault early warning method based on deep learning provided by the present invention. Figure 2 and Figure 3 As shown, in this embodiment, the method includes at least the following steps:

[0067] Step S10: Arrange multiple electromagnetic monitoring points around the substation to collect the direction, intensity, and spectrum parameters of electromagnetic radiation;

[0068] Specifically, in step S10, a plurality of electromagnetic monitoring points are arranged around the substation according to a predetermined layout, further comprising:

[0069] Assume that the substation area is a rectangular area on a two-dimensional plane with a length of L and a width of W. It is planned to deploy N electromagnetic monitoring points in this area. The monitoring point layout optimization formula based on the principle of electromagnetic coverage uniformity is used to determine the location of the monitoring points. The uniformity index U of the monitoring point layout is defined as follows:

[0070]

[0071] Among them, d ij It represents the distance between the i-th monitoring point and the j-th monitoring point, which is obtained by the following calculation formula:

[0072] Among them, (x i ,y i ) and (x j ,y j ) are the coordinates of the i-th monitoring point and the j-th monitoring point on the two-dimensional plane. By adjusting the coordinate positions of the monitoring points, the uniformity index U reaches the minimum value, thereby achieving the optimal layout of the monitoring points around the substation.

[0073] In the actual calculation process, iterative algorithms such as simulated annealing algorithm are used to find the monitoring point layout scheme that can minimize U. At the same time, taking into account the importance of different areas in the substation and the characteristics of electromagnetic environment changes, for some areas where important equipment is concentrated or where the electromagnetic environment is prone to change, on the basis of arranging monitoring points according to the above-mentioned uniform layout scheme, the density of monitoring points in these key areas is increased. According to the specific needs of electromagnetic environment monitoring in substations, electromagnetic sensors with high precision, high sensitivity and wide bandwidth characteristics are selected to enable them to accurately collect multiple key parameters of electromagnetic radiation, including the direction, intensity and spectrum of electromagnetic radiation. The selected electromagnetic sensors are arranged according to the pre- The designed layout plan is accurately installed at each electromagnetic monitoring point. During the installation process, in order to ensure that there is no electromagnetic interference between the sensor and the surrounding environment and other equipment, the relevant electrical installation specifications and electromagnetic compatibility requirements must be strictly followed to ensure that the sensor can operate in a stable and reliable state, thereby continuously and accurately collecting electromagnetic environment data. At the same time, in order to facilitate subsequent data collection and management, each installed sensor must be clearly marked and numbered so that it can accurately correspond to the sensor data of each monitoring point when processing the data later. For the determination of data collection frequency, assume that the change rate of a key electromagnetic parameter in the electromagnetic environment under normal operating conditions is ρ n , the maximum expected change rate of this parameter at the moment of fault occurrence is ρ f , considering the critical time period T after the failure occurs c Sufficient data samples reflecting changes in the electromagnetic environment can be collected within a certain period of time, so as to accurately capture the dynamic changes of the electromagnetic environment before and after the fault occurs. The data collection frequency f is determined according to the following formula:

[0074]

[0075] Among them, ΔI minIt is the minimum detectable change of a key electromagnetic parameter set in order to effectively detect changes in the electromagnetic environment. By determining the minimum detectable change and the known change rate of key electromagnetic parameters under normal and fault conditions, the critical time period of the fault and other information, the data acquisition frequency f that meets the requirements is calculated. According to the acquisition frequency determined by the above calculation, the electromagnetic sensors installed at each monitoring point are started to collect data. During the collection process, the electromagnetic environment data of the substation under normal operation and during the period when a ground fault may occur are continuously collected. The collected data should include detailed electromagnetic parameter information such as the electromagnetic radiation direction, intensity, and spectrum of each monitoring point at different times, and these data should be arranged in time series. At the same time, corresponding to different monitoring points, a complete electromagnetic parameter data set is formed. In order to facilitate the accurate tracing of the data collection time during subsequent data processing and analysis, an accurate timestamp is added to each data sample.

[0076] Step S11: Based on electromagnetic scattering theory and inverse problem solving methods, the collected data is used to construct electromagnetic environment models under normal operation and fault conditions, and the unique electromagnetic characteristics of ground faults are mined through comparative analysis and quantified to form a feature data set.

[0077] Specifically, the step S11 further includes:

[0078] Step S110: Based on electromagnetic scattering theory and inverse problem solving methods, the collected electromagnetic environment data under normal operation and before and after the fault occurs are comprehensively processed. For the data under normal operation, mathematical algorithms and model building methods are used to establish a benchmark model that can accurately characterize the normal electromagnetic environment characteristics of the substation;

[0079] More specifically, in step S110, for the data under normal operating conditions, a reference model capable of accurately characterizing the normal electromagnetic environment characteristics of the substation is established through mathematical algorithms and model building methods, further comprising:

[0080] Assume that there are M devices in the substation, each device can be equivalent to a point charge source, and the equivalent point charge of the i-th device is Q i , whose coordinates in space are (x i ,y i ,z i ), for any point in space, let its coordinates be (x; y, z) and the electric field intensity at that point be According to electromagnetic scattering theory, the electric field strength is calculated using the following formula:

[0081]

[0082] in, is the vector pointing from the equivalent point charge position of the i-th device to the point in space where the electric field strength is to be determined, yes The unit vector of ,∈0 is the dielectric constant of vacuum;

[0083] The magnetic field strength is calculated using the following formula:

[0084]

[0085] in, It is the equivalent motion velocity of the equivalent point charge of the i-th device, and the state of the electromagnetic environment under normal operating conditions is described by calculating the electric field strength and magnetic field strength.

[0086] For stationary equipment, it can usually be set as the zero vector. Through this formula, combined with the actual layout of the equipment in the substation and parameters such as the equivalent charge and equivalent movement speed of the equipment, the magnetic field strength at any point in space under normal operating conditions is calculated. The calculated electric field strength and magnetic field strength Combined with the actual layout of the equipment in the substation (i.e. the actual coordinate position of each device) and parameters such as the equivalent charge and equivalent movement speed of the equipment, an electromagnetic environment model of the substation at each point in space under normal operating conditions is constructed.

[0087] Step S111: Based on the collected data before and after the fault occurs, a model is constructed based on electromagnetic scattering theory and inverse problem solving methods to reflect changes in the electromagnetic environment caused by interference when the fault occurs, including changes in the direction of electromagnetic radiation, abnormal fluctuations in intensity, and new characteristics of the spectrum.

[0088] More specifically, in step S111, a model is constructed based on the collected data before and after the fault occurs according to electromagnetic scattering theory and an inverse problem solving method, further comprising:

[0089] Assume that the coordinates in the substation are (x f ,y f ,z f ) occurs, the fault point is regarded as a new equivalent charge source, and its charge is Q f , and the fault will cause the dielectric constant of the electromagnetic medium around the fault point to become ∈ f , the conductivity becomes σ f , for any point in space, let its coordinates be (x, y, z) and the electric field intensity at that point be According to electromagnetic scattering theory, the electric field strength after considering the fault effect is calculated using the following formula:

[0090]

[0091] in, is the vector pointing from the fault point to the point in space where the electric field strength is to be determined, yes The principal unit vector of is the vector pointing from the equivalent point charge position of the i-th device to the point in space where the electric field strength is to be determined, yes The unit vector of

[0092] Regarding the magnetic field strength, according to the electromagnetic scattering theory, the magnetic field strength after considering the fault effect is calculated using the following formula:

[0093]

[0094] in, It is the equivalent movement speed of the equivalent charge source at the fault point. For ground faults, it can usually be set to the zero vector. The state of the electromagnetic environment is described by calculating the electric field strength and magnetic field strength, combined with the actual layout of the equipment in the substation, the equivalent charge of the equipment, the location of the fault point and its characteristics (charge strength, dielectric constant, conductivity, etc.).

[0095] Step S112: In the process of constructing the above two models, the data differences and model feature differences formed during the model construction process under normal operation and fault conditions are directly compared and analyzed to mine the electromagnetic features unique to ground faults. The mined electromagnetic features unique to ground faults are quantified and converted into a data format that can be recognized and processed by a computer system to form a feature data set.

[0096] More specifically, in step S112, by directly comparing and analyzing the data differences in the model construction process under normal operation and fault state and the differences in model features formed, the unique electromagnetic features of the ground fault are mined, specifically:

[0097] An electromagnetic environment model of the substation at each point in space under a fault state can be constructed, which can reflect the electromagnetic environment characteristics of each point in the space around the substation when a fault occurs. The electromagnetic environment data before and after the fault occurs are collected, including the electromagnetic radiation direction, intensity, spectrum and other parameters of each monitoring point. These data are substituted as known conditions into the mathematical model established above that considers the fault factors. The unknown parameters in the model are solved through the inverse problem solving algorithm (such as based on the least squares method, regularization method, etc.), and a fault state electromagnetic environment model is constructed that can reflect the changes caused by the interference of the electromagnetic environment when the fault occurs, including changes in the direction of electromagnetic radiation, abnormal fluctuations in intensity and new characteristics of the spectrum. In the process of constructing the above two models, the data differences and the differences in the model characteristics formed during the model construction process under normal operation and fault state are directly compared and analyzed to mine the unique electromagnetic characteristics of grounding faults. Suppose the electromagnetic radiation intensity of a monitoring point at a certain time t1 under normal operation is I n (t1), the electromagnetic radiation intensity of the same monitoring point at time t2 after the fault occurs is I f (t2), the rate of change of electromagnetic radiation intensity at the monitoring point ΔI is calculated using the following formula: rate :

[0098]

[0099] And calculate the change rate of electromagnetic radiation intensity at each monitoring point. If ΔI rate Greater than the preset threshold ΔI Max , it can be preliminarily determined that there are changes in electromagnetic characteristics related to ground faults in the area where the monitoring point is located. The excavated electromagnetic characteristics unique to ground faults are quantified and converted into a data format that can be recognized and processed by a computer system to form a feature data set.

[0100] Step S12, using the quantized feature data set to train a deep learning model, and improving the model prediction accuracy by adjusting model parameters and optimizing algorithms;

[0101] Specifically, the step S12 further includes:

[0102] A deep learning model is constructed. The quantified ground fault-specific electromagnetic feature dataset and the corresponding substation operation status labels are divided into training, validation, and test sets in proportion. The deep learning model is trained using the training set. By adjusting the model parameters, a loss function is used during the training process to measure the difference between the model prediction results and the actual labels. The model parameters are then optimized using an optimization algorithm based on the difference. For details, please refer to Figure 3 shown.

[0103] More specifically, in an actual example, in step S12, building a deep learning model using a neural network algorithm further includes:

[0104] Based on the characteristics of substation electromagnetic environment data, such as the temporal nature of the data (the electromagnetic environment changes over time), spatial correlation (the electromagnetic environments at different monitoring points are interconnected), and the complexity of fault characteristics (ground fault characteristics have diverse manifestations), a convolutional neural network (CNN) was selected from existing deep learning architectures as the deep learning model architecture. The quantized electromagnetic feature dataset unique to ground faults and the corresponding substation operating status labels (normal operation or fault state) were divided into training, validation, and test sets in a ratio of 7:2:1. For model training, the input electromagnetic feature dataset image is X, whose size is M×N, the convolution kernel is K, whose size is m×n, and the step size is s. The convolution layer output Y is calculated using the following formula:

[0105]

[0106] Where X represents the electromagnetic feature dataset input to the convolution layer, presented in image form, and contains the unique electromagnetic feature information of the ground fault obtained from the electromagnetic environment model reconstruction and fault feature mining steps. K is the convolution kernel, whose element values are trainable weight parameters. The weights are continuously adjusted during the training process to extract local features of different scales in the input data. l is the position index of the convolution kernel K on a certain dimension of itself. M and N are the number of rows and columns of the input data image X, respectively, which are used to determine the size of the input data. m and n are the number of rows and columns of the convolution kernel, respectively, which are used to determine the size of the convolution kernel. Y is the output of the convolution layer. After the convolution operation, the new data with extracted local features is obtained, which will be further processed as the input of the pooling layer.

[0107] For the calculation of the pooling layer, let the convolution layer output be Y, whose size is P×Q, the pooling window size is p×q, and the step size is t. Then the pooling layer output Z is calculated using the following formula:

[0108]

[0109] Among them, Y is the input data of the pooling layer, that is, the output result of the convolutional layer, p and q are the number of rows and columns of the pooling window, respectively, which are used to determine the size of the pooling window, and Z is the output result of the pooling layer. The result obtained after the pooling operation will be used as the input of the fully connected layer to continue row processing;

[0110] For the calculation of the fully connected layer, let the output of the pooling layer be Z, whose size is R×S, and let Z ij For the elements in the pooling layer output data, is the connection weight between the kth neuron in the fully connected layer and the output element of the pooling layer, b k is the bias of the kth neuron, then the output a of the kth neuron in the fully connected layer is calculated using the following formula: k :

[0111]

[0112] Among them, Z is the input data of the fully connected layer, that is, the output result of the pooling layer, a k It is the output result of the kth neuron in the fully connected layer. After the fully connected layer operation, the new data with integrated features is obtained, which will be further processed as the input of the output layer. The electromagnetic feature data is gradually processed in depth to achieve the purpose of identifying the fault state.

[0113] For the output layer calculation, let the output of the fully connected layer be a k , then the output layer output y is calculated using the following formula:

[0114]

[0115] Among them, a k It is the output result of the fully connected layer, which contains the new data after the features are integrated by the fully connected layer. y is the output result of the output layer, and its value is between 0 and 1. When y ≥ 0.5, it is judged to be a fault state, and when y < 0.5, it is judged to be a normal operation state.

[0116] More specifically, in step S12, a loss function is used to measure the difference between the model prediction result and the actual label, further comprising:

[0117] Suppose there are N samples in the training set, and for the i-th sample, its actual label is y i , the model prediction result is The following formula is used to calculate the cross entropy loss function L:

[0118]

[0119] Among them, N represents the number of samples in the training set, which is used to determine the range of samples that need to be traversed when calculating the cross entropy loss function, and y i is the actual label of the i-th sample, is the model prediction result of the i-th sample. The predicted value obtained after the model processes the input electromagnetic feature data is used to compare with the actual label. L is the calculation result of the cross entropy loss function. The larger its value, the greater the difference between the model prediction result and the actual label.

[0120] And according to the difference, the optimization algorithm is used to optimize the model parameters. Specifically, let the model parameters be θ and the learning rate be η. Then, when the model parameters are updated each time, the calculation formula of the stochastic gradient descent algorithm is:

[0121]

[0122] Among them, θ represents all the trainable parameters of the model, which are objects that need to be continuously adjusted through the training process. η is the learning rate, which is a pre-set positive number used to control the step size of each iteration to update the model parameters. is the gradient of the cross entropy loss function L with respect to the model parameters θ, new is the updated model parameter. Through iterative updating, the model parameters are continuously optimized to improve the prediction accuracy and performance of the model. old It is the model parameter before the update, which serves as the basis for calculating the updated parameters. In each iterative update process, starting from the old parameters, it is updated according to the gradient information and learning rate to gradually optimize the model parameters.

[0123] Step S13: During the actual operation of the substation, electromagnetic environment data is collected in real time, and the electromagnetic environment model at the current moment and the corresponding electromagnetic characteristics unique to the ground fault are obtained. The model is used to predict whether the substation equipment is in a ground fault state.

[0124] If the deep learning model predicts a ground fault state, the early warning mechanism is triggered immediately. The early warning mechanism can send early warning information to the substation operation and maintenance personnel through the communication interface connected to the substation operation and maintenance system using a preset communication protocol (such as TCP / IP protocol, etc.). The early warning information should contain necessary fault-related information, such as the possible location of the fault inferred based on the location of the electromagnetic monitoring point and the electromagnetic characteristic analysis, and the severity of the fault preliminarily judged based on indicators such as the intensity of the electromagnetic characteristics, so that the operation and maintenance personnel can take appropriate measures in a timely manner to troubleshoot and handle the fault.

[0125] As another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program realizes the following Figures 1 to 3 For more details, please refer to and combine the above Figures 1 to 3 The description is not repeated here.

[0126] The implementation of the present invention has the following beneficial effects:

[0127] This invention provides a deep learning-based ground fault early warning method and storage medium for substation equipment. By deploying multiple electromagnetic monitoring points around the substation, comprehensive electromagnetic environment data is collected. Using electromagnetic scattering theory and inverse problem-solving methods, the electromagnetic environment model is reconstructed and fault characteristics are mined, accurately capturing changes in the substation's electromagnetic environment. By analyzing and processing these characteristics using a deep learning model, the complex relationship between electromagnetic characteristics and substation operating status can be automatically learned, significantly improving the accuracy and timeliness of ground fault early warnings.

[0128] This embodiment of the present invention determines whether substation equipment is in a ground fault state and provides detailed early warning information. This allows operations and maintenance personnel to quickly locate the fault point based on this warning information and take targeted measures to address it, thereby significantly improving the efficiency of fault handling. Furthermore, through the training and optimization of deep learning models, the timeliness and accuracy of early warnings are further improved, providing a strong technical guarantee for the safe operation of substations and effectively enhancing the safety and reliability of substation operations.

[0129] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, units, or computer program products. Thus, the present invention may take the form of a complete hardware embodiment, a complete 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-ROM, optical storage, etc.) containing computer-usable program code.

[0130] 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 unit of functionality specified in a box or multiple boxes.

[0131] The above disclosure is only a preferred embodiment of the present invention and certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A ground fault early warning method for substation equipment based on deep learning, characterized in that: The following steps are involved: Step S10: Arrange multiple electromagnetic monitoring points around the substation to collect the direction, intensity, and spectrum parameters of electromagnetic radiation; Step S11: Based on electromagnetic scattering theory and inverse problem solving methods, the collected data is used to construct electromagnetic environment models under normal operation and fault conditions, and the unique electromagnetic characteristics of ground faults are mined through comparative analysis and quantified to form a feature data set. Step S12, using the quantized feature data set to train a deep learning model, and improving the model prediction accuracy by adjusting model parameters and optimizing algorithms; Step S13: During the actual operation of the substation, electromagnetic environment data is collected in real time, and the electromagnetic environment model at the current moment and the corresponding electromagnetic characteristics unique to the ground fault are obtained. The model is used to predict whether the substation equipment is in a ground fault state.

2. The method according to claim 1, wherein In the step S10, a plurality of electromagnetic monitoring points are arranged around the substation according to a predetermined layout, further comprising: Assume that the substation area is a rectangular area on a two-dimensional plane with a length of L and a width of W. It is planned to arrange N electromagnetic monitoring points. The monitoring point layout optimization formula based on the principle of electromagnetic coverage uniformity is used to determine the location of the monitoring points. The uniformity index U of the monitoring point layout is defined as follows: Among them, d ij It represents the distance between the i-th monitoring point and the j-th monitoring point, which is obtained by the following calculation formula: Among them, (x i ,y i ) and (x j ,y j ) are the coordinates of the i-th monitoring point and the j-th monitoring point on the two-dimensional plane. By adjusting the coordinate positions of the monitoring points, the uniformity index U reaches the minimum value, thereby achieving the optimal layout of the monitoring points around the substation.

3. The method according to claim 2, wherein The step S11 further comprises: Based on electromagnetic scattering theory and inverse problem solving methods, the electromagnetic environment data collected during normal operation and before and after a fault occurs are comprehensively processed. For the data under normal operation, mathematical algorithms and model building methods are used to establish a benchmark model that can accurately characterize the normal electromagnetic environment characteristics of the substation; Based on the collected data before and after the fault, a model is constructed based on electromagnetic scattering theory and inverse problem solving methods to reflect the changes in the electromagnetic environment caused by interference when the fault occurs, including changes in the direction of electromagnetic radiation, abnormal fluctuations in intensity, and new characteristics of the spectrum. In the process of constructing the above two models, the data differences in the model construction process under normal operation and fault conditions and the differences in the model characteristics formed are directly compared and analyzed to mine the unique electromagnetic characteristics of grounding faults. The mined unique electromagnetic characteristics of grounding faults are quantified and converted into a data format that can be recognized and processed by the computer system to form a feature data set.

4. The method according to claim 3, wherein In step S11, for the data under normal operating conditions, a reference model capable of accurately characterizing the normal electromagnetic environment characteristics of the substation is established through mathematical algorithms and model building methods, further comprising: Assume that there are M devices in the substation, each device can be equivalent to a point charge source, and the equivalent point charge of the i-th device is Q i , whose coordinates in space are (x i ,y i ,z i ), for any point in space, let its coordinates be (x; y, z) and the electric field intensity at that point be According to electromagnetic scattering theory, the electric field strength is calculated using the following formula: in, is the vector pointing from the equivalent point charge position of the i-th device to the point in space where the electric field strength is to be determined, yes The unit vector of ,∈0 is the dielectric constant of vacuum; The magnetic field strength is calculated using the following formula: in, It is the equivalent motion velocity of the equivalent point charge of the i-th device, and the state of the electromagnetic environment under normal operating conditions is described by calculating the electric field strength and magnetic field strength.

5. The method according to claim 4, wherein In step S11, a model is constructed based on the collected data before and after the fault occurs according to electromagnetic scattering theory and an inverse problem solving method, further comprising: Assume that the coordinates in the substation are (x f ,y f ,z f ) occurs, the fault point is regarded as a new equivalent charge source, and its charge is Q f , and the fault will cause the dielectric constant of the electromagnetic medium around the fault point to become ∈ f , the conductivity becomes σ f , for any point in space, let its coordinates be (x, y, z) and the electric field intensity at that point be According to electromagnetic scattering theory, the electric field strength after considering the fault effect is calculated using the following formula: in, is the vector pointing from the fault point to the point in space where the electric field strength is to be determined, yes The principal unit vector of is the vector pointing from the equivalent point charge position of the i-th device to the point in space where the electric field strength is to be determined, yes The unit vector of Regarding the magnetic field strength, according to the electromagnetic scattering theory, the magnetic field strength after considering the fault effect is calculated using the following formula: in, It is the equivalent movement speed of the equivalent charge source at the fault point. The state of the electromagnetic environment is described by calculating the electric field strength and magnetic field strength, combined with the actual layout of the equipment in the substation, the equivalent charge of the equipment, and the location and characteristics of the fault point.

6. The method according to claim 5, wherein In step S11, by directly comparing and analyzing the data differences in the model construction process under normal operation and fault state and the differences in the model characteristics formed, the unique electromagnetic characteristics of the ground fault are mined, specifically: Assume that the electromagnetic radiation intensity of a monitoring point at a certain time t1 under normal operating conditions is I n (t1), the electromagnetic radiation intensity of the same monitoring point at time t2 after the fault occurs is I f (t2), the rate of change of electromagnetic radiation intensity at the monitoring point ΔI is calculated using the following formula: rate : And calculate the change rate of electromagnetic radiation intensity at each monitoring point. If ΔI rate Greater than the preset threshold ΔI Max , it can be preliminarily determined that there are electromagnetic characteristic changes related to ground faults in the area where the monitoring point is located.

7. The method according to claim 6, wherein The step S12 further comprises: A deep learning model is constructed, and the quantified electromagnetic feature dataset unique to ground faults and the corresponding substation operating status labels are divided into training set, validation set and test set in proportion. The deep learning model is trained using the training set. By adjusting the model parameters, a loss function is used to measure the difference between the model prediction results and the actual labels during the training process, and the model parameters are optimized using an optimization algorithm based on the difference.

8. The method according to claim 7, wherein In step S12, a deep learning model is constructed using a neural network algorithm, further comprising: Assume that the input electromagnetic feature dataset image is X, whose size is M×N, the convolution kernel is K, whose size is m×n, and the step size is s, then the convolution layer output Y is calculated using the following formula: Where X represents the electromagnetic feature dataset input to the convolution layer, presented in image form, and contains the unique electromagnetic feature information of the ground fault obtained from the electromagnetic environment model reconstruction and fault feature mining steps. K is the convolution kernel, whose element values are trainable weight parameters. The weights are continuously adjusted during the training process to extract local features of different scales in the input data. l is the position index of the convolution kernel K on a certain dimension of itself. M and N are the number of rows and columns of the input data image X, respectively, which are used to determine the size of the input data. m and n are the number of rows and columns of the convolution kernel, respectively, which are used to determine the size of the convolution kernel. Y is the output of the convolution layer. After the convolution operation, the new data with extracted local features is obtained, which will be further processed as the input of the pooling layer. For the calculation of the pooling layer, let the convolution layer output be Y, whose size is P×Q, the pooling window size is p×q, and the step size is t. Then the pooling layer output Z is calculated using the following formula: Among them, Y is the input data of the pooling layer, that is, the output result of the convolutional layer, p and q are the number of rows and columns of the pooling window, respectively, which are used to determine the size of the pooling window, and Z is the output result of the pooling layer. The result obtained after the pooling operation will be used as the input of the fully connected layer to continue row processing; For the calculation of the fully connected layer, let the output of the pooling layer be Z, whose size is R×S, and let Z ij For the elements in the pooling layer output data, is the connection weight between the kth neuron in the fully connected layer and the output element of the pooling layer, b k is the bias of the kth neuron, then the output a of the kth neuron in the fully connected layer is calculated using the following formula: k : Among them, Z is the input data of the fully connected layer, that is, the output result of the pooling layer, a k It is the output result of the kth neuron in the fully connected layer. After the fully connected layer operation, the new data with integrated features is obtained, which will be further processed as the input of the output layer, and the electromagnetic feature data will be gradually processed in depth to achieve the purpose of identifying the fault status. For the output layer calculation, let the output of the fully connected layer be a k , then the output layer output y is calculated using the following formula: Among them, a k It is the output result of the fully connected layer, which contains the new data after the features are integrated by the fully connected layer. y is the output result of the output layer, and its value is between 0 and 1. When y ≥ 0.5, it is judged to be a fault state, and when y < 0.5, it is judged to be a normal operation state.

9. The method according to claim 8, wherein In step S12, a loss function is used to measure the difference between the model prediction result and the actual label, further comprising: Suppose there are N samples in the training set, and for the i-th sample, its actual label is y i , the model prediction result is The following formula is used to calculate the cross entropy loss function L: Among them, N represents the number of samples in the training set, which is used to determine the range of samples that need to be traversed when calculating the cross entropy loss function, and y i is the actual label of the i-th sample, is the model prediction result of the i-th sample. The predicted value obtained after the model processes the input electromagnetic feature data is used to compare with the actual label. L is the calculation result of the cross entropy loss function. The larger its value, the greater the difference between the model prediction result and the actual label.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 9.

Images