Processing method and device of physical information neural network model, electronic equipment and storage medium

Abstract

A physical information neural network model processing method and device, electronic equipment and storage medium are disclosed. The method comprises: obtaining first data; inputting the first data into a first model; the first model is constructed based on a physical information neural network model, a physical loss is determined by at least one physical constraint corresponding to a physical constraint loss, each physical constraint loss is adjusted to the same order of magnitude and determined by output data of the first model; and at least one second data is obtained based on the first model. By using the technical solution of the present application, since each physical constraint loss of the first model is adjusted to the same order of magnitude, the problem of gradient imbalance in the training process caused by too large order of magnitude difference between multiple physical constraint losses can be avoided, each physical constraint loss is balanced, single loss is avoided from dominating the training, the overall solving accuracy of the first model is improved, and fast solving of the second data is realized.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of artificial intelligence and electrical engineering scientific computing, and in particular to a method, apparatus, electronic device, and storage medium for processing physical information neural network models. Background Technology

[0002] In the research, development, and operational evaluation of power transmission and transformation equipment, accurate simulation of internal physical processes is crucial. Physical information neural networks (PINs) guide the neural network to conform to physical laws while satisfying observed data by using the residuals of partial differential equations as part of the loss function. Therefore, PINs are an important tool for researching the research, development, and operation of power transmission and transformation equipment.

[0003] However, when solving the complex partial differential equations of power transmission and transformation equipment, the differences in the numerical magnitude of physical quantities and the huge differences in the coefficients of individual or coupled equations often lead to gradient imbalance during the training process, resulting in numerical rigidity problems, which seriously affect the convergence of the model and the simulation accuracy. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for processing physical information neural network models, in order to solve the problem that physical information neural network models have problems such as gradient explosion, making it difficult to accurately calculate the physical quantities of power transmission and transformation equipment.

[0005] According to one aspect of the present invention, a method for processing a physical information neural network model is provided, the method comprising:

[0006] Obtain the first data; the first data belongs to the first type of physical quantity related to power transmission and transformation equipment;

[0007] The first data is input into the first model; the first model is constructed based on the physical information neural network model and is trained based on the first loss function. The loss value of the first loss function is determined by the data loss and the physical loss. The physical loss is determined by the physical constraint loss corresponding to at least one physical constraint. The losses of each physical constraint are adjusted to the same order of magnitude and determined by the output data of the first model.

[0008] Based on the first model, at least one second data is obtained; the second data is a second type of physical quantity of power transmission and transformation equipment; the first type is different from the second type.

[0009] According to another aspect of the present invention, a processing apparatus for a physical information neural network model is provided, the apparatus comprising:

[0010] The first data acquisition module is used to acquire first data; the first data belongs to the first type of physical quantity related to power transmission and transformation equipment.

[0011] The first data input module is used to input the first data into the first model; the first model is constructed based on the physical information neural network model and is trained based on the first loss function. The loss value of the first loss function is determined by the data loss and the physical loss. The physical loss is determined by the physical constraint loss corresponding to at least one physical constraint. The losses of each physical constraint are adjusted to the same order of magnitude and determined by the output data of the first model.

[0012] The second data generation module is used to obtain at least one second data based on the first model; the second data is a second type of physical quantity of power transmission and transformation equipment; the first type is different from the second type.

[0013] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0014] At least one processor; and

[0015] A memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the processing method of the physical information neural network model according to any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the processing method of the physical information neural network model according to any embodiment of the present invention.

[0018] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the processing method of the physical information neural network model according to any embodiment of the present invention.

[0019] The technical solution of this invention involves acquiring first data; the first data belongs to a first type of physical quantity related to power transmission and transformation equipment; inputting the first data into a first model, wherein the first model is constructed based on a physical information neural network model, and the first model is trained based on a first loss function. The loss value of the first loss function is determined by data loss and physical loss. The physical loss is determined by the physical constraint loss corresponding to at least one physical constraint, and each physical constraint loss is adjusted to the same order of magnitude and determined by the output data of the first model; based on the first model, at least one second data is obtained, thereby realizing the prediction of unknown physical quantities of power transmission and transformation equipment; in addition, since the physical constraint losses of the first model are adjusted to the same order of magnitude, the gradient imbalance problem that occurs during training can be avoided when the order of magnitude difference of multiple physical constraint losses is too large. At this time, it can be ensured that during the training process, the physical laws equally constrain the training of the first model, balance the physical constraint losses, avoid a single loss dominating the training, and thus improve the overall solution accuracy of the first model.

[0020] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of a method for processing a physical information neural network model provided by the present invention;

[0023] Figure 2 This is a flowchart of the first model training provided by the present invention;

[0024] Figure 3 This is a flowchart of another first model training method provided by the present invention;

[0025] Figure 4 This is a schematic diagram of the structure of a processing device for a physical information neural network model provided by the present invention;

[0026] Figure 5 This is a schematic diagram of the structure of an electronic device that implements the processing method of the physical information neural network model of the present invention. Detailed Implementation

[0027] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] Figure 1 This is a flowchart of a physical information neural network model processing method provided by the present invention. It is applicable to the rapid prediction of physical quantities of power transmission and transformation equipment. This method can be executed by a physical information neural network model processing device, which can be implemented in hardware and / or software. This processing device can be configured in an electronic device with data processing capabilities. Figure 1 As shown, the method includes:

[0030] S110. Obtain the first data; the first data belongs to the first type of physical quantity related to power transmission and transformation equipment.

[0031] Power transmission and transformation equipment can be used for long-distance power transmission or voltage transformation. Power transmission and transformation equipment includes, but is not limited to: power transformers, high-voltage circuit breakers, high-voltage switchgear, current transformers, surge arresters, and relay protection devices. The power transmission and transformation equipment scenarios addressed in this application include, but are not limited to: calculating operating parameters of power transmission and transformation equipment during operation, predicting the remaining service life of power transmission and transformation equipment, optimizing operating parameters of power transmission and transformation equipment, predicting boundary conditions of abnormal operating conditions of power transmission and transformation equipment, and digital twin modeling of power transmission and transformation equipment; this application does not impose any limitations on these aspects.

[0032] The first data can be known physical quantities used to predict unknown physical quantities of power transmission and transformation equipment. The first data includes, but is not limited to, voltage, current, active power, reactive power, impedance, reactance, conductance, susceptance, partial discharge, dielectric loss, insulation resistance, electric field strength, winding temperature, oil surface temperature, core temperature, ambient temperature, thermal conductivity, convective heat transfer coefficient, heat source power, vibration amplitude, vibration frequency, stress, deformation, oil pressure, gas pressure, and local stress concentration. The first data can be acquired in real-time by sensors collecting data on the corresponding physical quantities of the power transmission and transformation equipment, or it can be obtained in real-time from a database. The first data can contain at least one physical quantity.

[0033] S120. Input the first data into the first model.

[0034] The first model is constructed based on the physical information neural network model. The first model is trained based on the first loss function. The loss value of the first loss function is determined by the data loss and the physical loss. The physical loss is determined by the physical constraint loss corresponding to at least one physical constraint. The losses of each physical constraint are adjusted to the same order of magnitude and determined by the output data of the first model.

[0035] The first model is constructed based on a physical information neural network model and consists of three layers: a normalization layer, a deep neural network backbone, and an inverse normalization layer. After training, the first model is used to predict unknown data related to power transmission and transformation equipment based on known input data. The normalization layer normalizes the input data, unifying the numerical magnitude of the input data. The deep neural network backbone predicts the unknown physical quantities based on the normalized input data and outputs the normalized prediction results. The inverse normalization layer inversely normalizes the normalized prediction results to obtain the prediction results for the unknown physical quantities.

[0036] The first loss function is the loss function of the first model during training, consisting of a data loss function and a physical loss function. Correspondingly, the loss value of the first loss function is also a weighted fusion of the data loss and physical loss. The data loss is used to compare the error between the first model's prediction and the corresponding true value, constraining the first model's prediction to closely match the corresponding true value during training. The physical loss is used to calculate the derivatives of the prediction using automatic differentiation, substitute them into the partial differential equation, and calculate the residuals where the partial differential equation does not satisfy physical laws, thus constraining the first model's calculations to conform to preset physical constraints.

[0037] The physical loss is determined by the physical constraint loss corresponding to at least one physical constraint, and these physical constraint losses are weighted and fused to generate the final physical loss. The physical constraint loss can be the magnitude of the error caused by violating the physical equation corresponding to the physical constraint. Each physical constraint corresponds to a separate physical equation. During training, the physical constraint losses of the first model are adjusted to the same order of magnitude.

[0038] After obtaining the first data, it can be input into the first model, which will then use the first data to predict the unknown data.

[0039] Since the losses of each physical constraint in the first model are adjusted to the same order of magnitude, the gradient imbalance problem during training is avoided when the orders of magnitude of the losses of multiple physical constraints are too different. At this time, it can be ensured that the training of the first model is equally constrained by various physical laws, the losses of each physical constraint are balanced, the training is not dominated by a single loss, and the overall solution accuracy of the first model is improved.

[0040] S130. Based on the first model, at least one second data is obtained; the second data is a second type of physical quantity of power transmission and transformation equipment; the first type is different from the second type.

[0041] After the first data is input into the first model, the first model will predict the unknown physical quantity and output the predicted second data.

[0042] The second data can be the prediction result obtained when predicting unknown physical quantities of power transmission and transformation equipment based on the first data. The second data includes, but is not limited to, voltage, current, active power, reactive power, impedance, reactance, conductance, susceptance, partial discharge, dielectric loss, insulation resistance, electric field strength, winding temperature, oil surface temperature, core temperature, ambient temperature, thermal conductivity, convective heat transfer coefficient, heat source power, vibration amplitude, vibration frequency, stress, deformation, oil pressure, gas pressure, and local stress concentration. The second data and the first data are of different data types.

[0043] The technical solution of this application involves acquiring first data, which belongs to a first type of physical quantity related to power transmission and transformation equipment; inputting the first data into a first model, wherein the first model is constructed based on a physical information neural network model and trained based on a first loss function. The loss value of the first loss function is determined by data loss and physical loss, and the physical loss is determined by the physical constraint loss corresponding to at least one physical constraint. The physical constraint losses are adjusted to the same order of magnitude and determined by the output data of the first model. Based on the first model, at least one second data is obtained, thereby enabling the prediction of unknown physical quantities of power transmission and transformation equipment. Furthermore, since the physical constraint losses of the first model are adjusted to the same order of magnitude, the gradient imbalance problem during training can be avoided when the order of magnitude difference of multiple physical constraint losses is too large. This ensures that during the training process, the physical laws equally constrain the training of the first model, balancing the physical constraint losses and avoiding a single loss dominating the training, thereby improving the overall solution accuracy of the first model.

[0044] Figure 2 This invention provides a flowchart for training a first model, further optimizing the training process of the aforementioned first model, and can be combined with the various optional solutions described above. For example... Figure 2 As shown, the training method for the first model may include the following steps:

[0045] S210. Determine the third data. The third data is the data output by the first model after the fourth data, which is input as sample data.

[0046] S220. Differentiate the third data to determine the statistical characteristics and physical data of each physical term in the physical equation; the physical equation is an expression that applies physical constraints to the first model.

[0047] When training the first model, the fourth data is input into the first model, and the first model will make predictions based on the input fourth data to generate the third data corresponding to the fourth data.

[0048] The automatic differentiation function differentiates the input third data, and substitutes the third data and the differentiation result into the physical terms in the physical equations to determine the physical term data and statistical characteristics of each physical term. When there are multiple physical equations, the above method is used to calculate the physical term data and statistical characteristics of each physical term for each equation. The statistical characteristics can be indices that reflect the magnitude of the corresponding physical term data. Statistical characteristics include, but are not limited to, standard deviation, variance, root mean square, absolute mean, and maximum norm.

[0049] Once the statistical features are determined, they can be used only during subsequent training without updating them, or they can be updated periodically or with momentum.

[0050] For example, taking Maxwell's equations-total current law as an example, the expression of Maxwell's equations-total current law is:

[0051] ;

[0052] in, Resistivity magnetic vector potential In the vertical direction Quantity; Current density In the vertical direction Quantity; Represents the horizontal direction; It represents the vertical direction.

[0053] and Based on the output of the first model, the physical terms of Maxwell's equations-current law are calculated using the automatic differentiation function:

[0054] ;

[0055] ;

[0056] ;

[0057] ;

[0058] ;

[0059] ;

[0060] in, The physical term data for the first physical term in the Maxwell equations-current law expression; The physical term data for the second physical term in the Maxwell equations-current law expression; This refers to the physical term data of the third physical term in the Maxwell equations-current law expression. This is the labeling difference of the first physical term in the Maxwell equations-current law expression; This is the labeling difference of the second physical term in the Maxwell equations-current law expression; This is the labeling difference of the third physical term in the Maxwell equations-current law expression.

[0061] Taking Maxwell's equations-Gauss's law for magnetic fields as an example, the expression for Maxwell's equations-Gauss's law for magnetic fields is:

[0062] ;

[0063] Taking the field-circuit coupling equation as an example, the expression of the field-circuit coupling equation is:

[0064] ;

[0065] In the formula, Excite the AC voltage at the ports of each phase winding; This represents the equivalent DC voltage drop across each phase winding. Let be the length of each phase winding in the z-direction; This refers to the number of turns in each phase winding; This represents the cross-sectional area of ​​the winding. For the winding region; For the corresponding winding DC resistance; For the winding region .

[0066] Correspondingly, the process for determining the physical terms of Maxwell's equations-Gauss's law for magnetic fields and the field-circuit coupling equations is the same as the process for determining the physical terms of Maxwell's equations-current law, and will not be elaborated upon here.

[0067] Accordingly, the expression for the physical loss function of the first equation is:

[0068] ;

[0069] in, Represents the physical loss function; express The weights; express The weights; express The weights; This represents the loss in Maxwell's equations - the total current law; This represents the loss in Maxwell's equations-Gauss's law for magnetic fields; This represents the loss in the field-path coupling equation.

[0070] The expression for the physical loss function in the first equation above is merely an example of the expression for the physical loss function in the presence of multiple physical constraints.

[0071] Furthermore, determine the statistical characteristics of each physical term in the physical equation, including steps A1-A2:

[0072] Step A1: Differentiate the third data to determine the physical term data of each physical term in the physical equation corresponding to the physical constraint.

[0073] Step A2: Based on the data of each physical item, determine the standard deviation of each physical item's data, and use the standard deviation as a statistical characteristic of the physical item.

[0074] The physical equations consist of partial differential equations to be solved, initial conditions, and boundary conditions, and can characterize the spatiotemporal evolution and conservation relationships of the physical field. The physical equations are used in the first model to automatically differentiate and calculate the equation residuals. By minimizing the residual loss, the network's predicted solutions are forced to strictly satisfy the physical equations, ensuring that the solution results conform to objective physical mechanisms. The physical equations include, but are not limited to, Maxwell's equations-current law, Maxwell's equations-Gauss's law for magnetic fields, and field-circuit coupling equations. During training, the physical equations are not input into the first model; instead, a physical constraint loss is calculated based on the physical equations, and the first model is trained using this physical constraint loss.

[0075] A physical term can be a term in a physical equation that represents a physical action or law. Physical terms include, but are not limited to, time derivative terms, spatial derivative terms, source terms, convection terms, and diffusion terms. Physical terms within the same physical equation are connected by plus or minus signs. Within the same physical equation, physical terms are interconnected through addition and subtraction operations, and the algebraic sum of each physical term satisfies a conservation equilibrium. During the first model training, the physical term data and statistical characteristics are calculated separately for each physical term.

[0076] The physics term data can be calculated by substituting the derivative of the first model's output into the physics equation. If, during the current training phase, the sum of the physics term data for each physics term in the same physics equation of the first model is not zero, it can be determined that the output of the first model does not satisfy the physics equation during this training phase, thus indicating a discrepancy between the first model's output and the corresponding true result.

[0077] Since the third data contains several sampling points, by differentiating the third data, we can calculate the data of each physical term in the physical equation corresponding to multiple physical constraints. Based on the data of each physical term, we can calculate the average value of the physical term. Based on the average value of the physical term and the data of each physical term, we can determine the statistical characteristic quantity of the physical term.

[0078] By applying the above method to each physical term in the physical equation corresponding to the physical constraint, the standard deviation of each physical term can be determined and used as a statistical characteristic of the physical term.

[0079] When calculating the statistical characteristics of each physical term in the physical equation, the physical term data corresponding to each physical term in the physical equation is determined by differentiating the third data. Based on the physical term data, the standard deviation of each physical term data is determined and used as the statistical characteristic of the physical term. This ensures the accuracy of the statistical characteristic. Furthermore, by selecting the standard deviation as the statistical characteristic, the problem of false convergence with a small mean and severe local physical distortion can be avoided when determining the physical constraint loss based on the standard deviation. It can also ensure the uniform and stable physical constraints throughout the solution domain by considering the discrete distribution of the entire domain.

[0080] For example, in the physical equation corresponding to the first model, five physical terms are calculated, namely 2, -2, 1, -1, and 0. Based on these five physical terms, the mean is determined to be 0, and the sample variance is 2.5. Taking the square root of the sample variance yields a standard deviation of 1.58.

[0081] For example, taking Maxwell's equations-total current law as an example, after balancing the various physical terms in Maxwell's equations-total current law, the expression for calculating the loss is as follows:

[0082] ;

[0083] Taking Maxwell's equations-Gauss's law for magnetic fields as an example, after balancing the various physical terms in Maxwell's equations-Gauss's law for magnetic fields, the expression for calculating the loss is as follows:

[0084] ;

[0085] Taking the field-circuit coupling equation as an example, after balancing the various physical terms in the field-circuit coupling equation, the expression for calculating the loss is as follows:

[0086] ;

[0087] In the formula, Indicates the number of data points; and As an incentive condition; , Here, t represents the grid space coordinates, and t represents the time variable. The area corresponding to the grid; and This is the output result for the first model.

[0088] Optionally, determine the statistical characteristics of each physical term in the physical equations, including steps B1-B2:

[0089] Step B1: Differentiate the third data to determine the physical term data of each physical term in the physical equation corresponding to the physical constraint.

[0090] Step B2: Based on the data of each physical item, determine the mean of each physical item's data, and use the mean as the statistical characteristic of the physical item.

[0091] Since the third set of data contains several sampling points, by differentiating the third set of data, we can calculate the data for each physical term in the physical equations corresponding to multiple physical constraints. Summing the data for each physical term and dividing the sum by the total number of data for that physical term yields the mean of the physical term's data. By processing each physical term in the above manner, we can obtain the mean of each physical term, thus determining the statistical characteristics of the physical term.

[0092] By differentiating the third data, the physical term data of each physical term in the physical equation corresponding to the physical constraint is determined. Based on the physical term data, the mean of each physical term data is determined. Finally, the mean is selected as the statistical characteristic quantity of the physical term, which can quickly characterize the overall average deviation of each physical term and clearly define the overall magnitude of different physical terms.

[0093] Optionally, determine the statistical characteristics of each physical term in the physical equation, including steps C1-C2:

[0094] Step C1: Differentiate the third data to determine the physical term data of each physical term in the physical equation corresponding to the physical constraint.

[0095] Step C2: Based on the data of each physical term, determine the maximum norm of each physical term's data, and use the maximum norm as a statistical characteristic of the physical term.

[0096] Since the third data contains several sampling points, by differentiating the third data, we can calculate the data of each physical term in the physical equation corresponding to multiple physical constraints. At this time, we can determine the physical term with the largest absolute value among the physical term data, obtain the maximum norm of the physical term, and realize the determination of the statistical characteristic quantity of the physical term.

[0097] By differentiating the third data, the physical terms data of each physical term in the physical equation corresponding to the physical constraints are determined. Based on the physical terms data, the maximum norm of each physical term data is determined, and the maximum norm is used as the statistical characteristic quantity of the physical term to constrain the worst position error in the solution domain, thus avoiding serious non-compliance of the physical equation in local areas, which would cause the first model training to fail to converge.

[0098] Optionally, determine the statistical characteristics of each physical term in the physical equations, including steps D1-D3:

[0099] Step D1: Determine the first number of statistical characteristic quantities.

[0100] Step D2: If the first count meets the preset conditions, redetermine the statistical characteristic quantity.

[0101] Step D3: If the first count does not meet the preset conditions, the statistical characteristic quantity will not be re-determined.

[0102] During the training of the first model, since the training of the first model is an iterative dynamic process, the magnitude of loss may vary greatly in different training stages. Therefore, if the statistical features are not updated after they are determined, the statistical features may be difficult to represent the magnitude of the physical data of the corresponding physical items in the subsequent training process. Therefore, the statistical features can be updated periodically to avoid this problem.

[0103] In response, when using statistical features, the first count of the statistical feature is determined, and it is determined whether the first count meets the preset conditions. If the first count meets the preset conditions, the statistical feature is redefined.

[0104] The "first number" can be either the number of iterations of the first model training that the statistical features have undergone in the current stage, or the remaining number of iterations of the first model training that the statistical features can use in the current stage. The preset conditions will be set accordingly when the meaning of the "first number" is different.

[0105] When the first number is the number of iterations of the first model training that the statistical feature has undergone in the current stage, the preset condition can be the maximum number of iterations allowed for the statistical feature. If the first number is equal to the maximum number of iterations, it can be determined that the statistical feature has been used too many times. In this case, the statistical feature will be regenerated and used in subsequent training to replace the previously calculated statistical feature.

[0106] When the first iteration is the number of iterations for the first model training that the remaining statistical features can be used in the current stage, the preset condition can be the minimum number of iterations that the statistical features are allowed to continue to be used. If the first iteration equals the minimum number of iterations, then it can be determined that the number of times the statistical feature can be used is 0. In this case, the statistical feature will be recalculated, and in subsequent training, the recalculated statistical feature will be used instead of the previously calculated statistical feature.

[0107] If the first count does not meet the preset conditions, the already determined statistical characteristics can continue to be used.

[0108] For example, when the first number is the number of iterations of the first model training that the statistical feature has undergone in the current stage, the first number is 500, and the maximum number of iterations allowed is 500. At this time, the first number is equal to the maximum number of iterations. Therefore, the statistical feature will be redefined to obtain a new statistical feature, and the first number of the new statistical feature will be determined to be 1.

[0109] When the first number is the number of iterations of the first model training that can be used for the remaining statistical features at the current stage, the first number is 1 and the minimum number of iterations is 1. At this time, the first number is equal to the minimum number of iterations, so the statistical features will be redefined to obtain new statistical features.

[0110] By periodically updating the statistical features, we can avoid the situation where the statistical features fail to reflect the order of magnitude of the physical terms corresponding to the statistical features due to a lack of updates during the training process. Furthermore, since the statistical features are updated periodically rather than in real time, we can reduce the computational resource burden caused by repeated calculations of the statistical features, reduce the operating pressure on computing devices, and minimize the impact on the training efficiency of the first model.

[0111] S230. Based on the statistical characteristics of each physical term and the physical term data, determine the physical constraint loss.

[0112] After obtaining the statistical characteristics and data of each physical item, the magnitude of the physical item data is adjusted using the statistical characteristics to bring the data of each physical item into the same magnitude. This magnitude adjustment can be achieved by dividing the physical item data by the statistical characteristics. Since some statistical characteristics may approach zero, leading to numerical overflow, a smoothing term can be added to the statistical characteristics when dividing the physical item data by the statistical characteristics.

[0113] For example, taking the Maxwell's equations-current law expression as an example, the physical loss function for the Maxwell's equations-current law expression can be constructed as follows:

[0114] ;

[0115] The physical loss function representing Maxwell's equations-total current law; For smoothing, used to avoid Approaching zero.

[0116] pass , , , able to The result is on the order of O(1).

[0117] S240. Based on the losses of each physical constraint, determine the physical loss, and train the first model based on the physical loss.

[0118] After obtaining the individual physical constraint losses, these losses can be integrated to generate the final physical loss. This integration can be achieved by weighting and fusing the individual physical constraint losses to balance them, ensuring accuracy. The physical loss is then used to train the first model, enabling it to reduce physical loss during subsequent training.

[0119] Optionally, the loss value of the first loss function is also determined by the boundary condition loss;

[0120] After determining the physical loss based on each physical constraint loss, the process also includes steps E1-E4:

[0121] Step E1: Determine the statistical characteristics of the physical terms corresponding to the boundary condition loss function and the physical term data of the physical terms corresponding to the boundary condition loss function.

[0122] Step E2: Determine the boundary condition loss based on the statistical characteristics of the physical terms corresponding to the boundary condition loss function and the physical term data of the physical terms corresponding to the boundary condition loss function.

[0123] Step E3: Determine the loss value of the first loss function based on the boundary condition loss and physical loss.

[0124] Step E4: Train the first model based on the loss value of the first loss function.

[0125] Boundary condition loss can be the sum of squared errors between the predicted values ​​of the first model at the boundary sampling points and the true boundary physical values. Boundary condition loss is used to force the first model to satisfy pre-defined boundary constraints. The boundary condition loss can be calculated using a boundary condition loss function. This function can be used to calculate the error between the first model's output and the given boundary theoretical values ​​at the boundary collocation points, thus constraining the first model to strictly satisfy the boundary physical laws at the solution domain edges.

[0126] As mentioned above, the process of determining the physical term data and statistical characteristic quantities of each physical term in the physical equation corresponding to the physical constraints is the same as that described above, and will not be repeated here.

[0127] After obtaining the boundary condition loss and physical loss, they can be combined to generate the loss value of the first loss function. The combination of boundary condition loss and physical loss can be achieved through weighted fusion. After generating the loss value of the first loss function, the first model is trained based on this loss value.

[0128] By determining the statistical features and physical term data corresponding to the physical terms of the boundary condition loss function, and based on these features and data, the boundary condition loss can be determined. This allows the magnitude of the boundary condition loss to be adjusted to be the same as that of the physical loss. Consequently, when weighting and fusing the boundary condition loss and physical loss, the sensitivity of the weights of both can be minimized, avoiding the problem of non-convergence in the first model training caused by a large difference in magnitude between the boundary condition loss and the physical loss.

[0129] Optionally, based on the boundary condition loss and physical loss, the loss value of the first loss function is determined, including steps F1-F2:

[0130] Step F1: Determine the first weight of the boundary condition loss and the second weight of the physical loss.

[0131] Step F2: Based on the first weight and the second weight, the boundary condition loss and the physical loss are fused to generate the loss value of the first loss function.

[0132] Determine the first weight for the boundary condition loss and the second weight for the physical loss. The specific values ​​of the first and second weights are the adjusted values ​​from the previous training stage. Multiplying the first weight by the boundary condition loss and adding the second weight by the physical loss yields the loss value of the first loss function.

[0133] By determining the first weight of the boundary condition loss and the second weight of the physical loss, and then fusing the boundary condition loss and the physical loss based on the first and second weights, a loss value of the first loss function is generated. This allows the loss value of the first loss function to integrate the boundary condition loss and the physical loss, thereby ensuring the accuracy of the final loss value of the first loss function. Consequently, after adjusting the parameters of the first model using the loss value of the first loss function, the accuracy of the first model can be improved.

[0134] By adopting the technical solution of this application, a third data point is determined, which is the output data of the first model after the fourth data, which is input as sample data; the derivative of the third data is calculated to determine the statistical characteristics and physical term data of each physical term in the physical equation; the physical equation is an expression for physical constraints on the first model; based on the statistical characteristics and physical term data of each physical term, the physical constraint loss is determined; based on the physical constraint loss, the physical loss is determined, and the first model is trained based on the physical loss, thereby achieving that the physical term data of each physical term in the physical equation corresponding to the physical constraints of the first model are adjusted to the same order of magnitude, realizing gradient balancing at the operator level, solving the numerical rigidity problem, and abandoning the tedious process of relying on manual experience to select feature scales and manually writing dimensionless equations in traditional methods, thus realizing rapid training of the first model.

[0135] Figure 3 This invention provides another flowchart for training the first model, which further optimizes the training process of the first model described above and can be combined with one or more of the aforementioned alternative solutions. For example... Figure 3 As shown, another method for training the first model may include the following steps:

[0136] S310, Obtain the fifth and sixth data.

[0137] Among them, the fifth data is the data transmitted by the first model to its denormalization layer after processing the fourth data; the sixth data is obtained by back-inferring the statistical features of the seventh data; the sixth data and the fifth data have the same order of magnitude; the fifth data is denormalized to obtain the third data; and the seventh data is the true value corresponding to the pre-labeled third data.

[0138] As mentioned above, the first model includes a normalization layer, a deep neural network backbone, and an anti-normalization layer. Therefore, the data at different stages in the first model differs in some aspects.

[0139] After the fourth data is input into the first model, it will first pass through the normalization layer of the first model for normalization processing to obtain the normalized fourth data. The normalized fourth data passes through the backbone of the deep neural network to generate the fifth data and send the fifth data to the denormalization layer. The fifth data passes through the denormalization layer for denormalization processing to generate the third data.

[0140] In addition to the fourth data used for prediction, it also includes the actual value corresponding to the third data, which is the seventh data.

[0141] After adjusting the order of magnitude of the physical terms in the physical equations corresponding to each physical constraint in the physical loss calculation, to avoid order-of-magnitude discrepancies between the data loss and the physical loss, it is necessary to ensure consistency in the order of magnitude between them. Therefore, this application does not directly use the predicted and actual values ​​input from the first model for processing. Instead, it uses the fifth data and the sixth data normalized from the seventh data for calculation.

[0142] To determine the sixth data point, we can first use the mean and standard deviation of the seventh data point, and then transform the seventh data point based on the mean and standard deviation to generate the sixth data point.

[0143] For example, the seventh data is transformed based on the following expression:

[0144] ;

[0145] in, This is the sixth data point; This is the seventh data point; The mean of the seventh data point; The standard deviation of the seventh data point.

[0146] Optionally, obtain the fourth and seventh data, including steps H1-H3:

[0147] Step H1: Determine the geometric model of the power transmission and transformation equipment.

[0148] Step H2: Based on the finite element analysis method, determine the simulation results of the geometric model under various preset excitation conditions.

[0149] Step H3: Based on the simulation results under various preset excitation conditions, determine the fourth and seventh data.

[0150] The geometric model of power transmission and transformation equipment can be a three-dimensional spatial structural model constructed according to the actual shape, size, and structure of the equipment. This geometric model defines the solution domain and boundary positions of the physical field when constructing the first model. Excitation conditions can be external actions applied to the equipment, driving changes in the distribution of electric, magnetic, and temperature fields. These external actions include, but are not limited to, voltage, current, load, temperature, and magnetic field loads. The excitation conditions constrain the evolution of the physical field, providing solution boundaries and input constraints for solving the first model. The finite element method (FEM) can be a method of simulating the power transmission and transformation equipment by dividing its geometric model into several small element meshes and solving the superimposed element equations. The FEM can cut the ensemble model of the power transmission and transformation equipment into several elements, with each element connected only by nodes, and each element approximating the physical field using a simple polynomial.

[0151] For the acquisition of the fourth and seventh data, a geometric model is established based on the configuration of the power transmission and transformation equipment, and simulation results under different excitation conditions are obtained based on the traditional finite element method. The simulation results corresponding to the same excitation condition are used as a data sample. Multiple sample data obtained by the above method are used as the dataset for model training, namely the fourth and seventh data.

[0152] By determining the geometric model of the power transmission and transformation equipment; based on the finite element analysis method, determining the simulation results of the geometric model under various preset excitation conditions; and based on the simulation results under various preset excitation conditions, determining the fourth and seventh data, the accuracy of the fourth and seventh data can be improved based on the precision of finite element simulation. Furthermore, obtaining the fourth and seventh data through simulation can improve data acquisition efficiency while reducing data acquisition costs.

[0153] Optionally, retrieve the fourth and seventh data, including:

[0154] The fourth and seventh data points of the power transmission and transformation equipment are collected based on sensors.

[0155] The type of sensor is determined based on the data types of the fourth and seventh data. Sensor types include, but are not limited to, temperature sensors, electric field sensors, voltage / current sensors, vibration sensors, magnetic field sensors, and partial discharge sensors. The data transmitted by the sensor can be used directly as the fourth and seventh data, or it can be used after data filtering.

[0156] The fourth and seventh data of the power transmission and transformation equipment are directly read by the sensors. After the sensors transmit the fourth and seventh data, outliers and data cleaning can be performed on the fourth and seventh data. After the data cleaning is completed, the fourth data with the corresponding seventh data is retained.

[0157] By collecting the fourth and seventh data points from power transmission and transformation equipment using sensors, it is possible to ensure that the fourth and seventh data points are as consistent as possible with the actual operating conditions of the power transmission and transformation equipment. When long-term data acquisition of the power transmission and transformation equipment is required, it is possible to detect the operating status of the power transmission and transformation equipment in real time, thereby continuously acquiring more accurate fourth and seventh data points. Furthermore, since the fourth and seventh data points are generated directly from the power transmission and transformation equipment by sensors, the errors in the fourth and seventh data points themselves are reduced, thus reducing the impact on the normal use of the first model.

[0158] S320. Based on the fifth and sixth data, determine the data loss.

[0159] S330. Based on data loss, train the first model.

[0160] After obtaining the fifth and sixth data points, the data loss can be calculated, and the first model can be trained based on the data loss.

[0161] The expression for data loss is:

[0162] ;

[0163] in, This indicates data loss.

[0164] By determining the data loss based on the fifth and sixth data points, since both the fifth and sixth data points are normalized results, the magnitude of the data loss can be adjusted to the O(1) order of magnitude, thereby accelerating the model's iterative convergence speed while ensuring model stability. Furthermore, by combining the magnitude of the physical loss with that of the data loss, the magnitudes of the data loss and physical loss are matched, thus reducing the sensitivity of the weights corresponding to the data loss and physical loss, and reducing the convergence difficulty of the first model.

[0165] The loss value of the first loss function can be determined by multiple losses, such as data loss, physical loss, and boundary condition loss, and the first model is trained based on the determined loss value of the first loss function.

[0166] By adopting the technical solution of this application, by acquiring the fifth and sixth data; determining the data loss based on the fifth and sixth data; and training the first model based on the data loss, it is possible to adjust the magnitude of the data loss to be the same as the magnitude of the physical loss, thereby reducing the sensitivity of the weights of the data loss and physical loss, achieving a balance between the data loss and physical loss, and improving the training effect of the first model.

[0167] Figure 4 This is a schematic diagram of the structure of a processing device for a physical information neural network model provided in an embodiment of the present invention. Figure 4 As shown, the device includes: a first data acquisition module 410, a first data input module 420, and a second data generation module 430. Wherein:

[0168] The first data acquisition module 410 is used to acquire first data; the first data belongs to the first type of physical quantity related to power transmission and transformation equipment.

[0169] The first data input module 420 is used to input the first data into the first model; the first model is constructed based on the physical information neural network model and is trained based on the first loss function. The loss value of the first loss function is determined by the data loss and the physical loss. The physical loss is determined by the physical constraint loss corresponding to at least one physical constraint. The physical constraint losses are adjusted to the same order of magnitude and determined by the output data of the first model.

[0170] The second data generation module 430 is used to obtain at least one second data based on the first model; the second data is a second type of physical quantity of power transmission and transformation equipment; the first type is different from the second type.

[0171] Based on the above embodiments, optionally, the training process of the first model includes:

[0172] The third data is determined; the third data is the output of the first model after the fourth data, which is used as the sample data, is input.

[0173] Differentiate the third data to determine the statistical characteristics and physical term data of each physical term in the physical equation; the physical equation is an expression that applies physical constraints to the first model;

[0174] Based on the statistical characteristics of each physical term and the physical term data, the physical constraint loss is determined.

[0175] Based on the losses of each physical constraint, the physical loss is determined, and the first model is trained based on the physical loss.

[0176] Based on the above embodiments, optionally, the statistical characteristic quantities of each physical term in the physical equation are determined, including:

[0177] Differentiate the third data to determine the physical term data of each physical term in the physical equation corresponding to the physical constraint;

[0178] Based on the data of each physical term, the standard deviation of each physical term is determined, and the standard deviation is used as a statistical characteristic of the physical term.

[0179] Based on the above embodiments, optionally, the training process of the first model includes:

[0180] The fifth and sixth data are obtained. The fifth data is the data transmitted by the first model to its denormalization layer after processing the fourth data. The sixth data is obtained by back-inferring the statistical features of the seventh data. The sixth data and the fifth data have the same order of magnitude. The fifth data is denormalized to obtain the third data. The seventh data is the true value corresponding to the pre-labeled third data.

[0181] Based on the fifth and sixth data points, determine the data loss;

[0182] The first model is trained based on data loss.

[0183] Based on the above embodiments, optionally, obtaining the fourth and seventh data includes:

[0184] Determine the geometric model of the power transmission and transformation equipment;

[0185] Based on the finite element analysis method, the simulation results of the geometric model under various preset excitation conditions are determined;

[0186] Based on the simulation results under various preset excitation conditions, the fourth and seventh data points are determined.

[0187] Based on the above embodiments, optionally, the loss value of the first loss function is also determined by the boundary condition loss;

[0188] This includes, after determining the physical loss based on the losses of each physical constraint, the following:

[0189] Determine the statistical characteristics of the physical terms corresponding to the boundary condition loss function and the physical term data corresponding to the boundary condition loss function;

[0190] The boundary condition loss is determined based on the statistical characteristics of the physical terms corresponding to the boundary condition loss function and the physical term data of the physical terms corresponding to the boundary condition loss function.

[0191] Based on boundary condition loss and physical loss, determine the loss value of the first loss function;

[0192] The first model is trained based on the loss value of the first loss function.

[0193] Based on the above embodiments, optionally, the loss value of the first loss function is determined based on boundary condition loss and physical loss, including:

[0194] Determine the first weight of the boundary condition loss and the second weight of the physical loss;

[0195] Based on the first and second weights, the boundary condition loss and physical loss are fused to generate the loss value of the first loss function.

[0196] The physical information neural network model processing device provided in the embodiments of the present invention can execute the physical information neural network model processing method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method execution.

[0197] Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of the present invention, such as... Figure 5The diagram illustrates a schematic representation of an electronic device 10 that can be used to implement embodiments of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0198] like Figure 5 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor 11, and the computer program is executed by the at least one processor 11 to enable the at least one processor 11 to perform the method provided by the present invention.

[0199] The processor 11 can perform various appropriate actions and processes based on a computer program stored in the read-only memory (ROM) 12 or a computer program loaded from the storage unit 18 into the random access memory (RAM) 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0200] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0201] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the methods provided in this invention.

[0202] In some embodiments, the methods provided herein may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to execute the methods by any other suitable means (e.g., by means of firmware).

[0203] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard parts (ASSPs), systems-on-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0204] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0205] In the context of this invention, a computer-readable storage medium stores computer instructions that are used to cause a processor to execute and implement the method provided by this invention.

[0206] The present invention also provides a computer program product comprising a computer program that, when executed by a processor, implements the method provided according to embodiments of the present invention. A computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. The computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, the computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0207] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0208] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0209] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0210] This invention also provides a computer program product, including a computer program that, when executed by a processor, can implement the methods provided in any embodiment of this application.

[0211] In the implementation of the computer program product, computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0212] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0213] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for processing a physical information neural network model, characterized in that, include: Get the first data; The first data belongs to the first type of physical quantity related to power transmission and transformation equipment; Input the first data into the first model; The first model is constructed based on a physical information neural network model. The first model is trained based on a first loss function. The loss value of the first loss function is determined by data loss and physical loss. The physical loss is determined by the physical constraint loss corresponding to at least one physical constraint. The physical constraint losses are adjusted to the same order of magnitude and determined by the output data of the first model. Based on the first model, at least one second data is obtained; the second data is a second type of physical quantity of the power transmission and transformation equipment; the first type is different from the second type.

2. The method according to claim 1, characterized in that, The training process of the first model includes: The third data is determined, which is the data output by the first model after the fourth data, which is input as sample data; The third data is differentiated to determine the statistical characteristics and physical term data of each physical term in the physical equation; the physical equation is an expression that applies physical constraints to the first model. Based on the statistical characteristics of each physical term and the physical term data, the physical constraint loss is determined. Based on the losses of each physical constraint, the physical loss is determined, and the first model is trained based on the physical loss.

3. The method according to claim 2, characterized in that, Determine the statistical characteristics of each physical term in the physical equations, including: Differentiate the third data to determine the physical term data of each physical term in the physical equation corresponding to the physical constraint; Based on the data of each physical term, the standard deviation of each physical term is determined, and the standard deviation is used as a statistical characteristic of the physical term.

4. The method according to claim 2, characterized in that, The training process of the first model includes: The fifth data and the sixth data are obtained. The fifth data is the data transmitted by the first model to its denormalization layer after processing the fourth data. The sixth data is obtained by back-inferring the statistical features of the seventh data. The sixth data has the same order of magnitude as the fifth data. The fifth data is denormalized to obtain the third data. The seventh data is the real value corresponding to the pre-labeled third data. Based on the fifth and sixth data points, determine the data loss; The first model is trained based on the data loss.

5. The method according to claim 4, characterized in that, Obtain the fourth and seventh data points, including: Determine the geometric model of the power transmission and transformation equipment; Based on the finite element analysis method, the simulation results of the geometric model under various preset excitation conditions are determined; Based on the simulation results under various preset excitation conditions, the fourth data and the seventh data are determined.

6. The method according to claim 2, characterized in that, The loss value of the first loss function is also determined by the boundary condition loss; This includes, after determining the physical loss based on the losses of each physical constraint, the following: Determine the statistical characteristics of the physical terms corresponding to the boundary condition loss function and the physical term data corresponding to the boundary condition loss function; The boundary condition loss is determined based on the statistical characteristics of the physical terms corresponding to the boundary condition loss function and the physical term data of the physical terms corresponding to the boundary condition loss function. Based on the boundary condition loss and the physical loss, determine the loss value of the first loss function; The first model is trained based on the loss value of the first loss function.

7. The method according to claim 6, characterized in that, Based on the boundary condition loss and the physical loss, the loss value of the first loss function is determined, including: Determine the first weight of the boundary condition loss and the second weight of the physical loss; Based on the first weight and the second weight, the boundary condition loss and the physical loss are fused to generate the loss value of the first loss function.

8. A processing device for a physical information neural network model, characterized in that, include: The first data acquisition module is used to acquire the first data. The first data belongs to the first type of physical quantity related to power transmission and transformation equipment; The first data input module is used to input the first data into the first model; the first model is constructed based on the physical information neural network model, and the first model is trained based on the first loss function. The loss value of the first loss function is determined by the data loss and the physical loss. The physical loss is determined by the physical constraint loss corresponding to at least one physical constraint. The physical constraint losses are adjusted to the same order of magnitude and determined by the output data of the first model. The second data generation module is used to obtain at least one second data based on the first model; the second data is a second type of physical quantity of the power transmission and transformation equipment; the first type is different from the second type.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the processing method of the physical information neural network model according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that are used to cause a processor to execute the processing method of the physical information neural network model according to any one of claims 1-7.

Images