Electrical equipment state determination method and apparatus, computer device, and storage medium

Abstract

The application relates to an electrical equipment state determination method and device, computer equipment and a storage medium. An initial coordinate of a target electrical equipment in a first number of dimensions at a target sampling point is acquired; the initial coordinate is split into a first number of one-dimensional coordinates; each one-dimensional coordinate is input into a corresponding target neural network unit to obtain a second number of sub-feature vectors output by each target neural network unit; a first predicted physical characteristic value of the target electrical equipment is determined according to each sub-feature vector; and the state of the target electrical equipment is determined according to the first predicted physical characteristic value. According to the above scheme, the high-dimensional initial coordinate of the target electrical equipment is split into one-dimensional coordinates, and each one-dimensional coordinate is processed by a corresponding target neural network unit, so that the overall processing of high-dimensional features is converted into independent parallel processing of multiple one-dimensional features, the problem of high calculation complexity of traditional high-dimensional coordinates is solved, and the feature processing efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, computer device, and storage medium for determining the state of electrical equipment. Background Technology

[0002] The operating status of each piece of electrical equipment in a power system directly determines the overall performance and reliability of the power system. Therefore, analyzing and predicting the operating status of each piece of electrical equipment is of great significance for the stable operation of the power system.

[0003] In traditional techniques, the high computational complexity of processing the high-dimensional feature vectors of each sampling point of power equipment reduces the efficiency of power equipment status analysis. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, and storage medium for determining the state of electrical equipment to address the aforementioned technical problems, thereby improving the efficiency of power equipment state analysis.

[0005] In a first aspect, this application provides a method for determining the state of electrical equipment, including:

[0006] Obtain the initial coordinates of the target electrical equipment in the first dimension of the target sampling point;

[0007] The initial coordinates are split into a first number of one-dimensional coordinates;

[0008] Each one-dimensional coordinate is input into the corresponding target neural network unit to obtain the sub-feature vector of the second dimension output by each target neural network unit;

[0009] Based on each sub-feature vector, determine the first predicted physical feature value of the target electrical equipment;

[0010] The state of the target electrical equipment is determined based on the first predicted physical characteristic value.

[0011] In one embodiment, determining the first predicted physical feature value of the target electrical device based on each sub-feature vector includes:

[0012] The sub-feature vectors are fused to obtain the target fused feature vector;

[0013] Based on the target fusion feature vector, the first predicted physical feature value of the target electrical equipment is determined.

[0014] In one embodiment, the fusion of the sub-feature vectors to obtain the target fused feature vector includes:

[0015] Element-wise multiplication of each sub-feature vector yields the target fusion feature vector of the second quantity dimension; wherein the second quantity dimension is consistent with the dimension of the first predicted physical feature value.

[0016] In one embodiment, the weight parameters and bias parameters of different target neural network units are the same, and at least one activation function of different target neural network units is different.

[0017] In one embodiment, each target neural network unit is trained in the following manner:

[0018] Obtain the sample coordinates of the electrical equipment in the first dimension of the sample sampling points;

[0019] Based on the sample coordinates and the first number of initial neural network units, a second predicted physical feature value of the sample electrical equipment at the sample sampling point is determined; wherein, the network structure of each initial neural network unit is the same, and at least one layer of activation function is different;

[0020] The training loss is determined based on the second predicted physical quantity feature value and the sample physical quantity feature value corresponding to the sample coordinates;

[0021] Based on the training loss, any initial neural network unit is trained to obtain the corresponding target neural network unit; and,

[0022] Using the weight and bias parameters of the obtained target neural network units, the parameters of the untrained initial neural network units are configured to obtain the remaining target neural network units.

[0023] In one embodiment, determining the training loss based on the second predicted physical quantity feature value and the sample physical quantity feature value corresponding to the sample coordinates includes:

[0024] The data loss is determined based on the deviation between the second predicted physical feature value and the sample physical feature value;

[0025] The partial differential equation (PDE) physical loss is determined based on the degree to which the second predicted physical feature value deviates from the partial differential equation corresponding to the sample physical feature value.

[0026] The initial loss is determined based on the deviation between the second predicted physical feature value of the initial state sampling point and the physical feature value of the sample.

[0027] The boundary loss is determined based on the deviation between the second predicted physical feature value and the physical feature value of the sampled points in the boundary region of the sampled points.

[0028] The training loss is determined based on the data loss, PDE physical loss, the initial loss, and the boundary loss.

[0029] In one embodiment, the initial coordinates include spatial coordinates and temporal coordinates, and obtaining the initial coordinates of the target electrical equipment in a first dimension of the target sampling point includes:

[0030] Based on the three-dimensional simulation model of the target electrical equipment, the spatial coordinates of the target electrical equipment in the first dimension at the target sampling point are obtained; and...

[0031] Based on the operating time of the target electrical equipment, obtain the time coordinates of the target electrical equipment at the target sampling point in the first dimension.

[0032] Secondly, this application also provides an electrical equipment condition analysis device, comprising:

[0033] The acquisition module is used to acquire the initial coordinates of the target electrical equipment in the first dimension of the target sampling point;

[0034] A splitting module is used to split the initial coordinates into a first number of one-dimensional coordinates;

[0035] The output module is used to input each one-dimensional coordinate into the corresponding target neural network unit to obtain the second-dimensional sub-feature vector output by each target neural network unit;

[0036] The first determining module is used to determine the first predicted physical feature value of the target electrical equipment based on each sub-feature vector;

[0037] The second determining module is used to determine the state of the target electrical equipment based on the first predicted physical characteristic value.

[0038] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0039] Obtain the initial coordinates of the target electrical equipment in the first dimension of the target sampling point;

[0040] The initial coordinates are split into a first number of one-dimensional coordinates;

[0041] Each one-dimensional coordinate is input into the corresponding target neural network unit to obtain the sub-feature vector of the second dimension output by each target neural network unit;

[0042] Based on each sub-feature vector, determine the first predicted physical feature value of the target electrical equipment;

[0043] The state of the target electrical equipment is determined based on the first predicted physical characteristic value.

[0044] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0045] Obtain the initial coordinates of the target electrical equipment in the first dimension of the target sampling point;

[0046] The initial coordinates are split into a first number of one-dimensional coordinates;

[0047] Each one-dimensional coordinate is input into the corresponding target neural network unit to obtain the sub-feature vector of the second dimension output by each target neural network unit;

[0048] Based on each sub-feature vector, determine the first predicted physical feature value of the target electrical equipment;

[0049] The state of the target electrical equipment is determined based on the first predicted physical characteristic value.

[0050] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0051] Obtain the initial coordinates of the target electrical equipment in the first dimension of the target sampling point;

[0052] The initial coordinates are split into a first number of one-dimensional coordinates;

[0053] Each one-dimensional coordinate is input into the corresponding target neural network unit to obtain the sub-feature vector of the second dimension output by each target neural network unit;

[0054] Based on each sub-feature vector, determine the first predicted physical feature value of the target electrical equipment;

[0055] The state of the target electrical equipment is determined based on the first predicted physical characteristic value.

[0056] The aforementioned method, apparatus, computer equipment, and storage medium for determining the state of electrical equipment acquire the initial coordinates of the target electrical equipment in a first-order dimension at the target sampling point; decompose the initial coordinates into a first-order number of one-dimensional coordinates; input each one-dimensional coordinate into a corresponding target neural network unit to obtain a sub-feature vector of a second-order dimension output by each target neural network unit; determine the first predicted physical feature value of the target electrical equipment based on each sub-feature vector; and determine the state of the target electrical equipment based on the first predicted physical feature value. This scheme, by decomposing the high-dimensional initial coordinates of the target electrical equipment into one-dimensional coordinates and processing them separately by the corresponding target neural network units, transforms the overall processing of high-dimensional features into independent parallel processing of multiple one-dimensional features, solving the problem of high computational complexity in traditional high-dimensional coordinate calculations and improving feature processing efficiency. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart illustrating a method for determining the state of electrical equipment in one embodiment;

[0059] Figure 2 This is a flowchart illustrating the training process for each target neural network unit in one embodiment;

[0060] Figure 3 This is a schematic diagram of the process for determining the training loss in one embodiment;

[0061] Figure 4 This is a structural block diagram of an electrical equipment state determination device in one embodiment;

[0062] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0064] The method for determining the state of electrical equipment provided in this application can be applied to scenarios involving the analysis of the operating state of electrical equipment. This method can be executed by a server or a terminal. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc.

[0065] In one exemplary embodiment, such as Figure 1 As shown, a method for determining the state of electrical equipment is provided. Taking the application of this method to a server as an example, the method includes the following steps:

[0066] S101, Obtain the initial coordinates of the target electrical equipment in the first dimension of the target sampling point.

[0067] The target electrical equipment refers to the electrical equipment in the power system to be analyzed, such as oil-immersed transformers, switchgear, and instrument transformers. The target sampling points are feature points selected within the physical structure or spatiotemporal range of the electrical equipment for state analysis. These points can be determined through a 3D simulation model of the equipment or by actual sensor data collection, such as winding measurement points or tank measurement points inside an oil-immersed transformer. The initial coordinates of the first dimension are the high-dimensional coordinates of the target sampling points in the spatiotemporal coordinate system. The first quantity is the number of dimensions of the coordinates, denoted as d. The initial coordinates include spatial and temporal coordinates, such as a 4-dimensional coordinate (x, y, z, t) in three-dimensional space and one-dimensional time, where the first quantity is 4. A one-dimensional coordinate is a single-dimensional coordinate obtained by splitting the high-dimensional initial coordinates according to their dimensions. For example, x, y, z, and t after splitting the 4-dimensional coordinates are each a one-dimensional coordinate.

[0068] For example, spatiotemporal data of sampling points can be collected using sensors to obtain initial coordinates. Alternatively, spatial coordinates can be extracted from a 3D simulation model of the target electrical equipment (such as a finite element simulation model), combined with temporal coordinates collected during the actual operation of the electrical equipment, to obtain the first quantitative dimension of the target electrical equipment at the target sampling point.

[0069] In some optional implementations, the initial coordinates may include spatial coordinates and temporal coordinates. The spatial coordinates of the target electrical equipment at the target sampling point can be obtained based on the three-dimensional simulation model of the target electrical equipment; and the temporal coordinates of the target electrical equipment at the target sampling point can be obtained based on the operating time of the target electrical equipment.

[0070] The spatial coordinates represent the coordinates of the target sampling point in three-dimensional physical space, reflecting the physical location of the sampling point, such as (x, y, z) in a Cartesian coordinate system, and are the spatial dimension of the initial coordinates. The time coordinates represent the operating time of the electrical equipment corresponding to the target sampling point, reflecting the temporal characteristics of the sampling point, such as t=1h, t=2h, and are the time dimension of the initial coordinates. The three-dimensional simulation model is a three-dimensional physical model of the target electrical equipment established through simulation software, which can accurately reproduce the structure, size, and physical characteristics of the electrical equipment and can be used to extract the spatial coordinates of the sampling points.

[0071] For example, a three-dimensional finite element simulation model of the target oil-immersed transformer can be established using COMSOL simulation software to accurately restore the dimensions and positions of the transformer's windings, core, tank, and other structures; 100 target sampling points (such as the middle layer of the windings, the middle of the tank, etc.) can be selected in the simulation model; and the three-dimensional spatial coordinates (x, y, z) of each target sampling point can be extracted from the simulation model. At this time, the dimension of the spatial coordinates is 3, which is part of the first quantity.

[0072] The actual operating time of the target electrical equipment can also be collected through the operation monitoring system, such as the transformer's operating time from startup to the current time being 2.5 hours; and the same time coordinate can be configured for all target sampling points, or different time coordinates can be configured according to the sampling time difference. The dimension of the time coordinate is 1, which is another part of the first quantity; then the spatial coordinates and time coordinates are integrated to obtain the initial coordinates of the first quantity dimension.

[0073] For example, if the running time is 2.5h and the time coordinate is t=2.5h, the spatial coordinates (0.5, 0.3, 0.4) are integrated with the time coordinate t=2.5h to obtain the 4-dimensional initial coordinates (0.5, 0.3, 0.4, 2.5), and the first quantity is 4.

[0074] In the above embodiments, the spatiotemporal attributes of the initial coordinates are clearly defined, and the spatial and temporal coordinates are obtained through the three-dimensional simulation model and the equipment running time, respectively. This ensures both the accuracy of the spatial coordinates and the real-time nature of the temporal coordinates, enabling the initial coordinates to fully characterize the spatiotemporal physical field of the electrical equipment. This, in turn, facilitates the improvement of the model's accuracy in learning the inherent laws of spatiotemporal dimensions and physical characteristics.

[0075] S102, split the initial coordinates into the first number of one-dimensional coordinates.

[0076] For example, the initial coordinates of the first number of dimensions can be split according to the coordinate dimensions, with each dimension corresponding to an independent one-dimensional coordinate, and the number of one-dimensional coordinates after splitting is equal to the first number.

[0077] For example, the initial 4-dimensional coordinates (x, y, z, t) can be split into 4 one-dimensional coordinates, namely x (horizontal coordinate), y (vertical coordinate), z (vertical coordinate), and t (time coordinate).

[0078] S103, input each one-dimensional coordinate into the corresponding target neural network unit to obtain the second-order sub-feature vector output by each target neural network unit.

[0079] The target neural network unit is a trained neural network module used to process one-dimensional coordinates. It is a multi-layer perceptron (MLP) neural network, and each target neural network unit is a sub-network with shared parameters and differentiated activation functions. The sub-feature vector of the second dimension is the feature vector output by the target neural network unit after processing the one-dimensional coordinates. The second dimension is the number of dimensions of the feature vector, denoted as r. For example, the second dimension can be determined based on the dimension of the first predicted physical feature value.

[0080] For example, a unique target neural network unit can be configured for each one-dimensional coordinate. The one-dimensional coordinate is passed as input to the corresponding target neural network unit, and the sub-feature vector corresponding to that dimension coordinate is output through the forward propagation calculation of the neural network. Among them, the parallel calculation of each target neural network unit can improve the processing efficiency.

[0081] For example, target neural network units MLP1, MLP2, MLP3, and MLP4 are configured for the four one-dimensional coordinates x, y, z, and t, respectively. x is input into MLP1, y into MLP2, z into MLP3, and t into MLP4. After forward propagation, each unit outputs sub-feature vectors Rx, Ry, Rz, and Rt of the second order of magnitude.

[0082] S104, determine the first predicted physical characteristic value of the target electrical equipment based on each sub-feature vector.

[0083] Among them, the first predicted physical feature value is the predicted value of the physical features of the electrical equipment at the target sampling point obtained by fusing the sub-feature vectors, such as magnetic induction intensity, temperature, electric field intensity, liquid flow velocity, heat source intensity, etc.

[0084] For example, feature fusion can be performed on the sub-feature vectors output by each target neural network unit. The resulting feature vector has the same dimension as the first predicted physical feature value, and the first predicted physical feature value is calculated based on the fused feature vector.

[0085] In some alternative implementations, the sub-feature vectors can be fused to obtain the target fused feature vector; and the first predicted physical feature value of the target electrical equipment can be determined based on the target fused feature vector.

[0086] For example, element-wise operations can be used to fuse the sub-feature vectors, including element-wise multiplication and element-wise addition, to obtain the target fused feature vector.

[0087] In some alternative implementations, element-wise multiplication of each sub-feature vector can be performed to obtain a target fusion feature vector with a second dimension; wherein the second dimension is consistent with the dimension of the first predicted physical feature value.

[0088] In this process, element-wise multiplication of each sub-feature vector can be performed to obtain the target fusion feature vector of the second dimension. For example, taking the four 8-dimensional sub-feature vectors of an oil-immersed transformer, Rx=[Rx1, Rx2, ..., Rx8], Ry=[Ry1, Ry2, ..., Ry8], Rz=[Rz1, Rz2, ..., Rz8], and Rt=[Rt1, Rt2, ..., Rt8], element-wise multiplication yields the 8-dimensional target fusion feature vector R=[Rx1×Ry1×Rz1×Rt1, Rx2×Ry2×Rz2×Rt2, ..., Rx8×Ry8×Rz8×Rt8].

[0089] For example, the dimensions of all sub-feature vectors can be pre-set to be the second quantity dimension, and the second quantity dimension can be precisely matched with the dimension of the first predicted physical feature value; for example, if the first predicted physical feature value is three-dimensional (such as temperature, liquid flow rate, heat source intensity), then the second quantity is set to 3, and all sub-feature vectors are 3-dimensional.

[0090] Furthermore, the sub-feature vectors of the same dimension output by each target neural network unit can be aligned according to their positions, and the elements at the same positions can be multiplied sequentially. The new feature vector obtained after the operation is the target fusion feature vector, and its dimension remains the second order dimension.

[0091] In the above embodiments, the fusion method of the sub-feature vectors is limited to element-wise multiplication and the dimension of the fused feature vector is consistent with the dimension of the predicted physical feature value. This maximizes the capture of nonlinear correlation features between dimensions and eliminates the dimension reduction calculation step. While improving the prediction accuracy, it further simplifies the calculation process and improves the calculation efficiency.

[0092] For example, if it is necessary to predict the three-dimensional first predicted physical characteristic values ​​[temperature, liquid flow rate, heat source intensity] of an oil-immersed transformer, then the second quantity is set to 3, and each target neural network unit outputs a 3-dimensional sub-feature vector: Rx=[Rx_temperature, Rx_flow, Rx_heat], Ry=[Ry_temperature, Ry_flow, Ry_heat], Rz=[Rz_temperature, Rz_flow, Rz_heat], Rt=[Rt_temperature, Rt_flow, Rt_heat]. Element-wise multiplication is then performed on the four sub-feature vectors:

[0093] Multiplying the elements of the temperature dimension: Rx_temperature × Ry_temperature × Rz_temperature × Rt_temperature = R_temperature;

[0094] Flow velocity element-wise multiplication: Rx_flow × Ry_flow × Rz_flow × Rt_flow = R_flow;

[0095] Multiplying the elements of the heat source dimension: Rx_heat × Ry_heat × Rz_heat × Rt_heat = R_heat.

[0096] The 3D target fusion feature vector R = [R_temperature, R_flow, R_heat] is obtained. At this time, the second quantity dimension (3D) is completely consistent with the dimension (3D) of the first predicted physical feature value. R can be directly used as the first predicted physical feature value, or the final predicted value can be obtained after simple normalization.

[0097] In the above embodiments, the target fused feature vector is obtained by fusing sub-feature vectors and then the predicted physical feature value is determined. This fully explores the correlation features of coordinates in each dimension, avoids the isolation of single-dimensional feature information, and achieves the effect of more sufficient basis for predicting physical feature values, thereby improving the prediction accuracy.

[0098] S105, determine the state of the target electrical equipment based on the first predicted physical characteristic value.

[0099] The state of electrical equipment refers to the operating state of the electrical equipment as determined by the first predicted physical characteristic value, such as normal operation, abnormal high temperature, partial discharge, abnormal medium flow rate, etc.

[0100] For example, normal threshold ranges for each physical characteristic of electrical equipment can be preset. The first predicted physical characteristic value is compared with the threshold range. If all physical characteristic values ​​are within the threshold, it is determined to be normal operation; if any physical characteristic value exceeds the threshold, it is determined to be an abnormal state of the corresponding type.

[0101] For example, assuming the preset normal temperature threshold for an oil-immersed transformer is 293K~310K, the liquid flow velocity threshold is 0.1m / s~0.8m / s, and the heat source intensity threshold is 50W / m³~150W / m³, if the first predicted physical characteristic value is 305K, 0.5m / s, and 100W / m³, then the transformer is determined to be operating normally; if the predicted temperature is 315K, then the transformer is determined to be abnormally high temperature.

[0102] In the above embodiments, by splitting the high-dimensional initial coordinates of the target electrical equipment into one-dimensional coordinates and processing them separately by the corresponding target neural network units, the overall processing of high-dimensional features is transformed into the independent parallel processing of multiple one-dimensional features, which solves the problem of high computational complexity of traditional high-dimensional coordinates and improves feature processing efficiency.

[0103] In some optional implementations, different target neural network units share the same weight and bias parameters, but at least one layer of activation functions differs. The weight parameters are the connection coefficients between neurons in each layer of the neural network, determining the strength of input feature transmission between neurons; they are the trainable parameters of the neural network. The bias parameters are the offsets of neurons in the neural network, used to adjust the output threshold of neurons, and together with the weight parameters, determine the computational result of the neural network. The activation function is a non-linear function introduced into the neural network to overcome the limitations of linear models, enabling the neural network to fit complex non-linear features. Activation functions can be Tanh, ReLU, Sigmoid, Leaky ReLU, etc. Parameter sharing means that different neural network units use exactly the same weight and bias parameters; that is, the trainable parameters of each unit are completely identical, differing only in non-trainable parameters (such as activation functions).

[0104] In this way, different target neural network units share weights and bias parameters and have differentiated activation functions, which reduces the number of model training parameters, lowers training costs and dependence on the amount of training data, and allows each neural network unit to adapt to the feature patterns of the corresponding dimension, thereby improving the targeting of dimensional feature learning.

[0105] Based on this, see Figure 2 , Figure 2 A flowchart illustrating the training process for each target neural network unit is provided, specifically including the following steps:

[0106] S201, Obtain the sample coordinates of the sample electrical equipment in the first dimension of the sample sampling point.

[0107] Among them, the sample electrical equipment is the same type of electrical equipment as the target electrical equipment used for model training. Its operating status data has been collected and labeled, such as the test transformer of the same model as the target oil-immersed transformer.

[0108] The sample sampling points are feature points selected from the sample electrical equipment for training. The number of these points can be set according to training requirements (e.g., 1000). The selection rules for sample sampling points are consistent with those for target sampling points.

[0109] The sample coordinates are the spatiotemporal coordinates of the first dimension of the sample sampling point, and their dimension is consistent with the initial coordinates of the target electrical equipment.

[0110] For example, a test transformer of the same model as the target oil-immersed transformer can be used as the sample electrical equipment. 1,000 sample sampling points can be selected, and the three-dimensional spatial coordinates of the sample sampling points can be obtained through a finite element simulation model. Combined with the time coordinates collected during the test run, the four-dimensional sample coordinates (x, y, z, t) can be obtained, that is, the first quantity is 4.

[0111] S202, based on the sample coordinates and the first number of initial neural network units, determine the second predicted physical feature value of the sample electrical equipment at the sample sampling point.

[0112] In this system, all initial neural network units have the same network structure, but at least one activation function is different. The initial neural network units are those configured before training, with the same network structure and differentiated activation functions, and the weight and bias parameters are set to their initial values.

[0113] The second predicted physical feature value is the physical feature prediction value obtained after the initial neural network unit processes the sample coordinates. It is an intermediate prediction value during the model training process.

[0114] For example, the 4-dimensional sample coordinates can be split into 4 one-dimensional coordinates x, y, z, and t; and 4 initial neural network units IN1-IN4 can be configured. The network structure has 3 hidden layers, each layer includes 64 neurons, the activation function is configured differently, and the weights and biases are configured with Xavier initial values; then each one-dimensional coordinate is input into the corresponding initial neural network unit to obtain sub-feature vectors, which are then fused to output the second predicted physical feature value.

[0115] S203, determine the training loss based on the second predicted physical quantity feature value and the sample physical quantity feature value corresponding to the sample coordinates.

[0116] Among them, the sample physical feature value is the actual physical feature value at the sample sampling point, which can be obtained through sensor acquisition or precise calculation by simulation model, and serves as the label value for model training. The training loss is an indicator used to measure the deviation between the second predicted physical feature value and the sample physical feature value, and serves as the basis for updating the model parameters.

[0117] For example, the actual physical feature values ​​of the sample sampling points can be collected by experimental sensors, i.e., the sample physical feature values; then, a loss function (such as mean square error MSE) can be used to calculate the deviation between the second predicted physical feature value and the sample physical feature value to obtain the training loss.

[0118] S204, train any initial neural network unit based on the training loss to obtain the corresponding target neural network unit; and, using the weight parameters and bias parameters of the obtained target neural network unit, configure the parameters of the untrained initial neural network unit to obtain the remaining target neural network units.

[0119] For example, any initial neural network unit (such as IN1) can be selected, and its weight parameters and bias parameters can be updated according to the training loss using a gradient descent algorithm (such as the Adam algorithm). The training is iterated repeatedly until the training loss drops to a preset threshold (such as loss < 1). The IN1 after training is the target neural network unit TN1.

[0120] Then, the weight parameters W and bias parameters b after training TN1 are directly copied to the untrained initial neural network units IN2, IN3, and IN4 to complete the parameter configuration and obtain the target neural network units TN2, TN3, and TN4; finally, four target neural network units TN1-TN4 with shared parameters and different activation functions are obtained.

[0121] In the above embodiments, all target neural network units are obtained by training only one initial neural network unit and then synchronizing the parameters to the remaining initial neural units. This avoids repeated training of multiple units, improves model training efficiency, and also makes the model parameters of each neural unit consistent, thus achieving parameter sharing.

[0122] In some alternative implementations, see [link to relevant documentation]. Figure 3 , Figure 3 A flowchart illustrating the process of determining training loss is provided, which includes the following steps:

[0123] S301, determine the data loss based on the deviation between the second predicted physical feature value and the sample physical feature value.

[0124] Among them, data loss is used to measure the loss of the direct deviation between the second predicted physical feature value and the sample physical feature value, reflecting the degree of fit of the model to the measured data.

[0125] For example, the mean squared error (MSE) can be used to calculate the deviation between the second predicted physical feature value and the sample physical feature value for all sample sampling points, thus obtaining the data loss. The calculation formula is as follows:

[0126]

[0127] in, The data loss is represented by n, where n is the number of sample points. The second predicted physical feature value for the i-th sampling point. Let be the physical feature value of the sample at the i-th sampling point.

[0128] S302, determine the physical loss of the partial differential equation (PDE) based on the degree to which the second predicted physical feature value deviates from the partial differential equation corresponding to the sample physical feature value.

[0129] Among them, the physical loss of the partial differential equation (PDE) is used to measure the degree to which the second predicted physical characteristic value deviates from the partial differential equation corresponding to the physical law of the electrical equipment, such as Maxwell's equations, heat conduction equations, Navier-Stokes equations, etc., and reflects the degree to which the model obeys physical constraints. When the physical loss of PDE is 0, the predicted value fully conforms to the physical law.

[0130] For example, the corresponding partial differential equation can be determined based on the physical characteristic values ​​of the sample. For instance, the corresponding partial differential equation for heat conduction can be determined based on the heat conduction characteristics of an oil-immersed transformer:

[0131]

[0132] in, For density, For isobaric specific heat, For temperature, For flow rate, For heat flux density, As the main heat source, As an additional heat source.

[0133] For example, the partial differential equation for temperature characteristic values ​​can be the heat conduction equation; the partial differential equation for electric field strength or potential characteristic values ​​can be the Poisson equation or the Laplace equation; the partial differential equation for fluid velocity or pressure characteristic values ​​can be the Navier-Stokes equation (NS equation); the partial differential equation for magnetic induction intensity or magnetic field strength characteristic values ​​can be the Maxwell equations; the partial differential equation for heat flux density characteristic values ​​can be Fourier's law of heat conduction; and the partial differential equation for concentration characteristic values ​​(such as dissolved gases in insulating oil) can be the diffusion equation.

[0134] The formula for calculating PDE physical loss is as follows:

[0135]

[0136] in, For PDE physical loss, The residual is the result of substituting the predicted value of the i-th sampling point into the partial differential equation.

[0137] S303, determine the initial loss based on the deviation between the second predicted physical feature value of the initial state sampling point and the physical feature value of the sample.

[0138] The initial state sampling points are the sampling points corresponding to the initial operating state of the electrical equipment (such as start-up time and initial temperature) among the sample sampling points, and are the initial constraint points for model training. The initial loss is used to measure the loss of the deviation between the predicted value and the actual value at the initial state sampling points, and is the initial constraint loss of the model.

[0139] For example, initial state sampling points (such as 50 sampling points at the transformer start-up time t=0) can be selected from the sample sampling points, and the deviation between the predicted value and the actual value of this type of sampling point can be calculated using MSE to obtain the initial loss.

[0140] The formula for calculating the initial loss is as follows:

[0141]

[0142] in, For the initial loss, The initial number of sampling points. The second predicted physical feature value is the i-th initial state sampling point. Let be the sample physical feature value of the i-th initial state sampling point.

[0143] S304, determine the boundary loss based on the deviation between the second predicted physical feature value of the sampling point in the boundary region and the physical feature value of the sample.

[0144] Among them, the boundary region sampling points are the sampling points corresponding to the physical boundaries of electrical equipment (such as the inner wall of the transformer tank or the edge of the winding) in the sample sampling points, and are the boundary constraint points for model training. The boundary loss is used to measure the loss of the deviation between the predicted value and the actual value at the boundary region sampling points, and is the boundary constraint loss of the model.

[0145] For example, boundary region sampling points (such as 100 sampling points on the inner wall of a transformer tank) can be selected from the sample sampling points, and the deviation between the predicted value and the actual value of such sampling points can be calculated using MSE to obtain the boundary loss.

[0146] The formula for calculating boundary loss is as follows:

[0147]

[0148] in, For boundary loss, The number of sampling points in the boundary area. The second predicted physical feature value is the sampling point of the i-th boundary region. Let be the sample physical feature value of the i-th boundary region sampling point.

[0149] S305 determines the training loss based on the data loss, PDE physical loss, initial loss, and boundary loss.

[0150] For example, the data loss, PDE physical loss, initial loss, and boundary loss can be fused using a weighted summation method to obtain the training loss. The weight of each loss can be set according to actual needs. For example, if physical laws are emphasized, the weight of the PDE physical loss can be increased, and the sum of the weights can be 1.

[0151] The formula for calculating training loss is as follows:

[0152]

[0153] in, For training loss, , , , As weight, .

[0154] In the above embodiments, the training loss is decomposed into data loss, PDE physical loss, initial loss and boundary loss and then weighted and fused. This allows the model training to not only fit the measured data and comply with the physical laws of electrical equipment, but also to accurately constrain the prediction results of the initial state and boundary region. The multi-dimensional loss constraints make the model prediction values ​​more realistic, improve the accuracy of physical feature value prediction and the physical interpretability of the model.

[0155] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0156] Based on the same inventive concept, this application also provides an electrical equipment state determination apparatus for implementing the above-described electrical equipment state determination method. The solution provided by this apparatus is similar to the implementation described in the above-described method; therefore, the specific limitations in one or more embodiments of the electrical equipment state determination apparatus provided below can be found in the limitations of the electrical equipment state determination method described above, and will not be repeated here.

[0157] In one exemplary embodiment, such as Figure 4 As shown, a state determination device for electrical equipment is provided, comprising:

[0158] The acquisition module 10 is used to acquire the initial coordinates of the target electrical equipment in the first dimension of the target sampling point;

[0159] The splitting module 20 is used to split the initial coordinates into a first number of one-dimensional coordinates;

[0160] The output module 30 is used to input each one-dimensional coordinate into the corresponding target neural network unit to obtain the second-dimensional sub-feature vector output by each target neural network unit;

[0161] The first determining module 40 is used to determine the first predicted physical feature value of the target electrical equipment based on each sub-feature vector;

[0162] The second determining module 50 is used to determine the state of the target electrical equipment based on the first predicted physical characteristic value.

[0163] The above scheme solves the problem of high computational complexity of traditional high-dimensional coordinates by splitting the high-dimensional initial coordinates of the target electrical equipment into one-dimensional coordinates, which are then processed separately by the corresponding target neural network units. This transforms the overall processing of high-dimensional features into the independent parallel processing of multiple one-dimensional features, thereby improving the efficiency of feature processing.

[0164] In one embodiment, the first determining module 40 is specifically used for:

[0165] The sub-feature vectors are fused to obtain the target fused feature vector; based on the target fused feature vector, the first predicted physical feature value of the target electrical equipment is determined.

[0166] In one embodiment, the first determining module 40 is specifically used for:

[0167] Element-wise multiplication of each sub-feature vector yields the target fusion feature vector with a second dimension; where the second dimension is consistent with the dimension of the first predicted physical feature value.

[0168] In one embodiment, the weight parameters and bias parameters of different target neural network units are the same, and at least one activation function of different target neural network units is different.

[0169] In one embodiment, the device further includes a training module for:

[0170] The process involves: acquiring sample coordinates of the sample electrical equipment at the sample sampling points in a first-order dimension; determining the second predicted physical feature value of the sample electrical equipment at the sample sampling points based on the sample coordinates and a first-order number of initial neural network units; wherein each initial neural network unit has the same network structure and at least one layer of activation function is different; determining the training loss based on the second predicted physical feature value and the sample physical feature value corresponding to the sample coordinates; training any initial neural network unit based on the training loss to obtain the corresponding target neural network unit; and configuring the untrained initial neural network units using the weight parameters and bias parameters of the obtained target neural network units to obtain the remaining target neural network units.

[0171] In one embodiment, the training module is specifically used for:

[0172] The data loss is determined based on the deviation between the second predicted physical feature value and the sample physical feature value; the partial differential equation (PDE) physical loss is determined based on the degree to which the second predicted physical feature value deviates from the partial differential equation corresponding to the sample physical feature value; the initial loss is determined based on the deviation between the second predicted physical feature value and the sample physical feature value of the initial state sampling point in the sample sampling points; the boundary loss is determined based on the deviation between the second predicted physical feature value and the sample physical feature value of the boundary region sampling point in the sample sampling points; and the training loss is determined based on the data loss, PDE physical loss, initial loss, and boundary loss.

[0173] In one embodiment, the initial coordinates include spatial coordinates and temporal coordinates, and the acquisition module 10 is specifically used for:

[0174] Based on the three-dimensional simulation model of the target electrical equipment, the spatial coordinates of the target electrical equipment in the first dimension of the target sampling point are obtained; and, based on the running time of the target electrical equipment, the temporal coordinates of the target electrical equipment in the first dimension of the target sampling point are obtained.

[0175] Each module in the aforementioned electrical equipment status determination device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0176] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores the coordinate data of electrical equipment at sampling points. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a method for determining the state of electrical equipment.

[0177] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0178] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for determining the state of electrical equipment described in any of the above embodiments.

[0179] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for determining the state of electrical equipment described in any of the above embodiments.

[0180] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method for determining the state of electrical equipment described in any of the above embodiments.

[0181] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0182] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0183] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0184] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for determining the state of electrical equipment, characterized in that, The method includes: Obtain the initial coordinates of the target electrical equipment in the first dimension of the target sampling point; The initial coordinates are split into a first number of one-dimensional coordinates; Each one-dimensional coordinate is input into the corresponding target neural network unit to obtain the sub-feature vector of the second dimension output by each target neural network unit; Based on each sub-feature vector, determine the first predicted physical feature value of the target electrical equipment; The state of the target electrical equipment is determined based on the first predicted physical characteristic value.

2. The method according to claim 1, characterized in that, The step of determining the first predicted physical feature value of the target electrical equipment based on each sub-feature vector includes: The sub-feature vectors are fused to obtain the target fused feature vector; Based on the target fusion feature vector, the first predicted physical feature value of the target electrical equipment is determined.

3. The method according to claim 2, characterized in that, The process of fusing the sub-feature vectors to obtain the target fused feature vector includes: Element-wise multiplication of each sub-feature vector yields the target fusion feature vector of the second quantity dimension; wherein the second quantity dimension is consistent with the dimension of the first predicted physical feature value.

4. The method according to any one of claims 1-3, characterized in that, Different target neural network units have the same weight and bias parameters, but at least one activation function is different.

5. The method according to claim 4, characterized in that, Each target neural network unit is trained in the following way: Obtain the sample coordinates of the electrical equipment in the first dimension of the sample sampling points; Based on the sample coordinates and the first number of initial neural network units, a second predicted physical feature value of the sample electrical equipment at the sample sampling point is determined; wherein, the network structure of each initial neural network unit is the same, and at least one layer of activation function is different; The training loss is determined based on the second predicted physical quantity feature value and the sample physical quantity feature value corresponding to the sample coordinates; Based on the training loss, any initial neural network unit is trained to obtain the corresponding target neural network unit; and, Using the weight and bias parameters of the obtained target neural network units, the parameters of the untrained initial neural network units are configured to obtain the remaining target neural network units.

6. The method according to claim 5, characterized in that, The step of determining the training loss based on the second predicted physical quantity feature value and the sample physical quantity feature value corresponding to the sample coordinates includes: The data loss is determined based on the deviation between the second predicted physical feature value and the sample physical feature value; The partial differential equation (PDE) physical loss is determined based on the degree to which the second predicted physical feature value deviates from the partial differential equation corresponding to the sample physical feature value. The initial loss is determined based on the deviation between the second predicted physical feature value of the initial state sampling point and the physical feature value of the sample. The boundary loss is determined based on the deviation between the second predicted physical feature value and the physical feature value of the sampled points in the boundary region of the sampled points. The training loss is determined based on the data loss, PDE physical loss, the initial loss, and the boundary loss.

7. The method according to any one of claims 1-4, characterized in that, The initial coordinates include spatial coordinates and temporal coordinates. Obtaining the initial coordinates of the target electrical equipment in the first dimension of the target sampling point includes: Based on the three-dimensional simulation model of the target electrical equipment, the spatial coordinates of the target electrical equipment in the first dimension at the target sampling point are obtained; and... Based on the operating time of the target electrical equipment, obtain the time coordinates of the target electrical equipment at the target sampling point in the first dimension.

8. An electrical equipment condition analysis device, characterized in that, The device includes: The acquisition module is used to acquire the initial coordinates of the target electrical equipment in the first dimension of the target sampling point; A splitting module is used to split the initial coordinates into a first number of one-dimensional coordinates; The output module is used to input each one-dimensional coordinate into the corresponding target neural network unit to obtain the second-dimensional sub-feature vector output by each target neural network unit; The first determining module is used to determine the first predicted physical feature value of the target electrical equipment based on each sub-feature vector; The second determining module is used to determine the state of the target electrical equipment based on the first predicted physical characteristic value.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images