CN117350152A - Electromagnetic field modeling method, system, equipment and medium based on mechanism data fusion - Google Patents
Electromagnetic field modeling method, system, equipment and medium based on mechanism data fusion Download PDFInfo
- Publication number
- CN117350152A CN117350152A CN202311278715.9A CN202311278715A CN117350152A CN 117350152 A CN117350152 A CN 117350152A CN 202311278715 A CN202311278715 A CN 202311278715A CN 117350152 A CN117350152 A CN 117350152A
- Authority
- CN
- China
- Prior art keywords
- electromagnetic field
- neural network
- modeling
- field distribution
- physical constraint
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000005672 electromagnetic field Effects 0.000 title claims abstract description 131
- 238000000034 method Methods 0.000 title claims abstract description 71
- 230000007246 mechanism Effects 0.000 title claims abstract description 33
- 230000004927 fusion Effects 0.000 title claims abstract description 28
- 238000009826 distribution Methods 0.000 claims abstract description 101
- 238000012549 training Methods 0.000 claims abstract description 77
- 238000013528 artificial neural network Methods 0.000 claims abstract description 53
- 230000006870 function Effects 0.000 claims abstract description 39
- 238000003062 neural network model Methods 0.000 claims abstract description 39
- 239000011159 matrix material Substances 0.000 claims abstract description 34
- 238000013178 mathematical model Methods 0.000 claims description 16
- 238000003860 storage Methods 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 12
- 238000004088 simulation Methods 0.000 claims description 11
- 230000035699 permeability Effects 0.000 claims description 7
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 238000013135 deep learning Methods 0.000 abstract description 15
- 238000004364 calculation method Methods 0.000 abstract description 8
- 230000008569 process Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 8
- 238000012545 processing Methods 0.000 description 7
- 230000006872 improvement Effects 0.000 description 6
- 238000006467 substitution reaction Methods 0.000 description 6
- 238000010276 construction Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 239000000463 material Substances 0.000 description 4
- 230000015572 biosynthetic process Effects 0.000 description 3
- 230000004907 flux Effects 0.000 description 3
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical group [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000006698 induction Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000000746 body region Anatomy 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000005684 electric field Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000003449 preventive effect Effects 0.000 description 1
- 238000007670 refining Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0499—Feedforward networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Hardware Design (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Geometry (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the technical field of deep learning and electromagnetic field distribution modeling, and discloses an electromagnetic field modeling method, system, equipment and medium based on mechanism data fusion; the electromagnetic field modeling method based on mechanism data fusion comprises the following steps: acquiring data required by electromagnetic field distribution modeling based on a transformer to be subjected to electromagnetic field distribution modeling; based on the acquired data required by the electromagnetic field distribution modeling, predicting by using a trained physical constraint neural network model to obtain a vector magnetic position matrix predicted value; calculating to obtain an electromagnetic field quantity matrix, and further obtaining electromagnetic field distribution; the loss function adopted in training is a physical constraint loss function constructed based on a vector magnetic potential equation, a boundary condition, an initial condition and data information. The invention can solve the technical problems of low modeling speed, low efficiency and poor interpretability and reliability of the traditional data driving method such as the traditional artificial neural network and the like in the traditional numerical calculation method.
Description
Technical Field
The invention belongs to the technical field of deep learning and electromagnetic field distribution modeling, and particularly relates to an electromagnetic field modeling method, system, equipment and medium based on mechanism data fusion.
Background
The reliable operation of the transformer is the basis of the safety and stability of the power grid; the operation experience shows that by means of the existing equipment state evaluation technology of means such as online monitoring, electrified detection, preventive test and the like, only partial problems can be deduced, but the evolution of the internal state of the equipment cannot be completely deduced. Inside the transformer, electromagnetic field distribution is vital, and can produce important influence to temperature distribution and structural force distribution in the equipment, in time master the evolution mechanism of the internal electromagnetic state of the transformer, improve the digitization degree of the transformer, realize the digitization and the intelligent operation and maintenance of the core equipment transformer of the power grid, and the method is a key problem which is urgently needed to be solved in the safe operation of the current power grid.
For modeling of the electromagnetic field distribution condition of the transformer, a conventional method is a numerical calculation method, such as a finite element method, a finite difference method, a finite volume method and the like, and the method of the conventional method is to approximately solve a partial differential equation through the numerical method, so that a great deal of time is consumed, and the requirement of timely mastering the electromagnetic field distribution condition is difficult to meet.
In recent years, with the rise of artificial intelligence, some people also model electromagnetic fields by adopting a deep learning method; the method is characterized in that a numerical calculation or experimental substitution model is built based on deep learning, so that modeling efficiency can be effectively improved, and time required by modeling is shortened. However, existing deep learning-based surrogate models have their own limitations, mainly including: deep learning is used as a pure data driving model, a large amount of high-quality data is required for training, and the phenomenon of data resource waste can be generated; in addition, because the inside of the black box model has poor interpretability, the performance of the model in a truly open scene is not guaranteed, and the robustness is poor.
Disclosure of Invention
The invention aims to provide an electromagnetic field modeling method, system, equipment and medium based on mechanism data fusion, so as to solve one or more technical problems. The technical scheme provided by the invention is particularly a physical-data hybrid driving transformer electromagnetic field distribution modeling scheme, which can solve the technical problems of low modeling speed and low efficiency of the traditional numerical calculation method and the technical problems of poor interpretability and insufficient reliability of the traditional data driving method of the traditional artificial neural network and the like.
In order to achieve the above purpose, the invention adopts the following technical scheme:
the electromagnetic field modeling method based on mechanism data fusion provided by the first aspect of the invention comprises the following steps:
acquiring data required by electromagnetic field distribution modeling based on a transformer to be subjected to electromagnetic field distribution modeling; the data required by electromagnetic field distribution modeling comprises electromagnetic field distribution modeling area coordinates and preset modeling time;
based on the acquired data required by the electromagnetic field distribution modeling, predicting by using a trained physical constraint neural network model to obtain a vector magnetic position matrix predicted value;
calculating to obtain an electromagnetic field quantity matrix based on the obtained vector magnetic bit matrix predicted value, and obtaining electromagnetic field distribution based on the electromagnetic field quantity matrix;
the loss function adopted by the trained physical constraint neural network model in training is a physical constraint loss function constructed based on a vector magnetic level equation, a boundary condition, an initial condition and data information.
The invention discloses a further improvement of the method, which is characterized in that the physical constraint neural network model adopts a multi-layer fully-connected neural network and comprises an input layer, a plurality of hidden layers and an output layer; the input layer is used for inputting position coordinates and time, and the output layer is used for outputting vector magnetic bit matrix predicted values.
The invention discloses a further improvement of the method, which is characterized in that the training step of the trained physical constraint neural network model comprises the following steps:
performing partial simulation based on the transformer to be subjected to electromagnetic field distribution modeling or the transformers of the same type to obtain an electromagnetic field distribution model sample;
acquiring a training sample set based on the electromagnetic field distribution model sample; each training sample in the training sample set comprises a distribution point boundary condition position coordinate, a moment and a vector magnetic position label;
training the physical constraint neural network model based on the training sample set, and obtaining the trained physical constraint neural network model after reaching a preset convergence condition; during training, a physical constraint loss function is calculated based on a selected training sample, a gradient between the output and the input of the physical constraint neural network is calculated, the feedforward neural network is counter-propagated according to the physical constraint loss function according to a gradient descent method, and weights and biases in the network are iteratively updated.
A further improvement of the disclosed method is that the physical constraint loss function MSE is expressed as,
where MSE b Indicating the degree to which the predicted value meets the boundary condition; MSE (mean square error) f Representing the error of substituting the predicted value obtained by the input feedforward neural network into the mathematical model; i represents an i-th data node; n (N) b The matching point quantity for the selected boundary condition;representing the setpoint at the boundary condition; />Representing a vector magnetic potential predicted value obtained after the training of the input neural network; />Is the true value of the vector magnetic bit; n (N) f Representing the amount of data points used;representing the mathematical model used; />The vector magnetic potential predicted value obtained after the training of the input neural network is obtained; mu is magnetic permeability; sigma is conductivity; j (J) S Representing the source current density; x and y are the horizontal and vertical coordinates of the adopted planar field respectively; t is time.
The electromagnetic field modeling system based on mechanism data fusion provided by the second aspect of the invention comprises:
the data acquisition module is used for acquiring data required by electromagnetic field distribution modeling based on the transformer to be subjected to electromagnetic field distribution modeling; the data required by electromagnetic field distribution modeling comprises electromagnetic field distribution modeling area coordinates and preset modeling time;
the predicted value acquisition module is used for predicting by utilizing the trained physical constraint neural network model based on the acquired data required by the electromagnetic field distribution modeling to obtain a vector magnetic position matrix predicted value;
the electromagnetic field distribution acquisition module is used for calculating and acquiring an electromagnetic field quantity matrix based on the acquired vector magnetic potential matrix predicted value and acquiring electromagnetic field distribution based on the electromagnetic field quantity matrix;
the loss function adopted by the trained physical constraint neural network model in training is a physical constraint loss function constructed based on a vector magnetic level equation, a boundary condition, an initial condition and data information.
The invention discloses a further improvement of the system, which is characterized in that in the predicted value acquisition module, the physical constraint neural network model adopts a multi-layer fully-connected neural network and comprises an input layer, a plurality of hidden layers and an output layer; the input layer is used for inputting position coordinates and time, and the output layer is used for outputting vector magnetic bit matrix predicted values.
The invention discloses a further improvement of the system, wherein in the predicted value acquisition module, the training step of the trained physical constraint neural network model comprises the following steps:
performing partial simulation based on the transformer to be subjected to electromagnetic field distribution modeling or the transformers of the same type to obtain an electromagnetic field distribution model sample;
acquiring a training sample set based on the electromagnetic field distribution model sample; each training sample in the training sample set comprises a distribution point boundary condition position coordinate, a moment and a vector magnetic position label;
training the physical constraint neural network model based on the training sample set, and obtaining the trained physical constraint neural network model after reaching a preset convergence condition; during training, a physical constraint loss function is calculated based on a selected training sample, a gradient between the output and the input of the physical constraint neural network is calculated, the feedforward neural network is counter-propagated according to the physical constraint loss function according to a gradient descent method, and weights and biases in the network are iteratively updated.
A further improvement of the disclosed system is that, in the predictor acquisition module, the physical constraint loss function MSE is expressed as,
where MSE b Indicating the degree to which the predicted value meets the boundary condition; MSE (mean square error) f Representing the error of substituting the predicted value obtained by the input feedforward neural network into the mathematical model; i represents an i-th data node; n (N) b The matching point quantity for the selected boundary condition;representing the setpoint at the boundary condition; />Representing a vector magnetic potential predicted value obtained after the training of the input neural network; />Is the true value of the vector magnetic bit; n (N) f Representing the amount of data points used;representing the mathematical model used; />The vector magnetic potential predicted value obtained after the training of the input neural network is obtained; mu is magnetic permeability; sigma is conductivity; j (J) S Representing the source current density; x and y are the horizontal and vertical coordinates of the adopted planar field respectively; t is time.
An electronic device provided in a third aspect of the present invention includes:
at least one processor; the method comprises the steps of,
a memory communicatively coupled to the at least one processor; wherein,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the electromagnetic field modeling method based on mechanism data fusion as described in any one of the first aspects of the invention.
A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the electromagnetic field modeling method based on mechanism data fusion according to any one of the first aspects of the present invention.
Compared with the prior art, the invention has the following beneficial effects:
in the electromagnetic field modeling method based on mechanism data fusion, the vector magnetic potential matrix predicted value is obtained based on the trained physical constraint neural network model, so that the electromagnetic field distribution can be obtained rapidly; the physical constraint neural network model introduces physical information constraint conditions on the basis of the original data driving model, can reduce the data scale required by model training, can solve the technical problems of low modeling speed and low efficiency of the traditional numerical calculation method, and can also solve the technical problems of poor interpretability and insufficient reliability of the traditional data driving method of the traditional artificial neural network and the like.
The method is characterized by high response speed and high calculation instantaneity, and is used for solving the problems of low modeling speed and low efficiency of the traditional numerical calculation method and remarkably improving the modeling efficiency of electromagnetic field distribution.
The method is specifically and explanatory, and aims at the problems of poor interpretability and insufficient reliability of the traditional data driving method such as the existing artificial neural network, and the technical scheme of the invention adopts the technical means of the physical information neural network, and the means can introduce physical mechanisms such as Maxwell equations in the model training process, so that the model training is constrained by training data and physical rules, the reliability and the reliability of modeling of the neural network model in a real open scene can be remarkably improved by solving the problems, and the reliability of the model in the real open scene can be improved by considering physical mechanisms such as electromagnetic field control equations.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following description of the embodiments or the drawings used in the description of the prior art will make a brief description; it will be apparent to those of ordinary skill in the art that the drawings in the following description are of some embodiments of the invention and that other drawings may be derived from them without undue effort.
FIG. 1 is a schematic flow chart of an electromagnetic field modeling method based on mechanism data fusion provided by an embodiment of the invention;
FIG. 2 is a schematic diagram of a training process of a physical information neural network according to an embodiment of the present invention;
fig. 3 is a schematic diagram of an electromagnetic field modeling system based on mechanism data fusion according to an embodiment of the present invention.
Detailed Description
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
The invention is described in further detail below with reference to the attached drawing figures:
referring to fig. 1, an electromagnetic field modeling method based on mechanism data fusion provided by an embodiment of the invention includes the following steps:
step 1, acquiring data required by electromagnetic field distribution modeling based on a transformer to be subjected to electromagnetic field distribution modeling; the data required by electromagnetic field distribution modeling comprises electromagnetic field distribution modeling area coordinates and preset modeling time;
step 2, based on the data required by the electromagnetic field distribution modeling obtained in the step 1, predicting by using a trained physical constraint neural network model to obtain a vector magnetic position matrix predicted value; the method comprises the steps that a trained physical constraint neural network model is trained, wherein a loss function adopted in the training is a physical constraint loss function constructed based on a vector magnetic level equation, a boundary condition, an initial condition and data information;
and 3, calculating and obtaining an electromagnetic field quantity matrix based on the vector magnetic potential matrix predicted value obtained in the step 2, and obtaining electromagnetic field distribution based on the electromagnetic field quantity matrix.
In the technical scheme provided by the embodiment of the invention, the vector magnetic position matrix predicted value is obtained based on the trained physical constraint neural network model, so that the electromagnetic field distribution is obtained; the loss function adopted in training is a physical constraint loss function constructed based on a vector magnetic potential equation, a boundary condition, an initial condition and data information, and the physical constraint neural network model introduces a physical information constraint condition on the basis of the original data driving model, so that the data size required by model training can be reduced.
Referring to fig. 2, in a further preferred embodiment of the present invention, the physical constraint neural network model adopts a multi-layer full-connected Neural Network (NN), including an input layer, a plurality of hidden layers, and an output layer; the input layer comprises 3 neurons which respectively represent x, y and t and are used for inputting position coordinates and time; the output layer is used for outputting the vector magnetic bit matrix predicted value.
The training step of the trained physical constraint neural network model comprises the following steps:
performing partial simulation based on the transformer to be subjected to electromagnetic field distribution modeling or the transformers of the same type to obtain an electromagnetic field distribution model sample; for example, a transformer model can be built by adopting a simulation development environment of simdroid software, and a transformer electromagnetic field distribution model sample can be calculated and created;
acquiring a training sample set based on the electromagnetic field distribution model sample; wherein each training sample comprises: position coordinates, time and vector magnetic position labels of point boundary conditions;
during training, for a selected training sample, calculating a physical constraint loss function of the training sample, calculating a gradient between the output and the input of the physical constraint neural network, carrying out back propagation on the feedforward neural network according to the physical constraint loss function according to a gradient descent method, iteratively updating weights and biases in the network, and obtaining a trained physical constraint neural network model after reaching a preset convergence condition; further illustratively, the model training process can be further optimized by adopting an optimization method such as a random gradient descent method.
In the embodiment of the invention, the physical constraint loss function is constructed based on a vector magnetic bit equation, a boundary condition, an initial condition and data information, which is specifically expressed as,
where MSE b Representing the degree to which the predicted value meets boundary conditions, namely boundary condition loss terms, wherein the boundary conditions are frame coordinates of the transformer core; i represents an i-th data node;representing the setpoint at the boundary condition;representing a vector magnetic potential predicted value obtained after the training of the input neural network; />Is the true value of the vector magnetic bit; n (N) b The matching point quantity for the selected boundary condition; MSE (mean square error) f Representing the error of substituting the predicted value obtained by inputting the feedforward neural network into the mathematical model, namely controlling the equation loss term; n (N) f Representing the amount of data points it uses; />Representing the mathematical model used, the vector magnetic potential predicted value obtained after the training of the input neural network is obtained; mu is magnetic permeability, and the unit is H/m; sigma is the conductivity of the material in S/m.
Further illustratively, the mathematical model is established for the transformer flux density distribution, comprising:
according to maxwell's equationsAnd the relation between the corresponding field amounts +.>
Wherein B is a magnetic induction intensity vector; h represents a magnetic field strength vector; e represents an electric field intensity vector; d represents an electric displacement vector; ρ represents the charge density; j represents current density; j (J) S Representing the source current density; j (J) C Indicating the induced eddy current density; epsilon represents the dielectric constant of the medium; mu represents permeability; gamma represents conductivity;
in the time harmonic analysis, it is very difficult to directly solve the field quantities E and H; thus, vector magnetic bits are usedThe magnetic flux distribution is obtained from the vector magnetic potential as the electromagnetic field distribution.
By way of specific example, the solution to the intra-domain partial differential equation and boundary conditions can be derived using the finite element method, where the mathematical model is constructed using parallel planar fields (x, y):
further processing the mathematical model to obtain:
the left formula of the mathematical model under the non-boundary condition is the final mathematical model used by the loss term of the control equation, wherein x and y are respectively the horizontal coordinate and the vertical coordinate of the adopted plane field (x and y), t is time, mu is material magnetic conductivity, and the unit is H/m; sigma is the conductivity of the material, and the unit is S/m; a is vector magnetic position; j (J) S Is the source current density, the unit is A/m 2 。
The embodiment of the invention is further explanatory, the influence of a physical mechanism is considered in the process of constructing the deep learning substitution model, and the explanatory property of the deep learning model can be enhanced while the electromagnetic field distribution modeling efficiency is greatly improved. Specifically, the embodiment of the invention combines physical information such as transformer data information, maxwell equation information, boundary condition information and the like to construct a physical constraint neural network as a substitute model for modeling electromagnetic field distribution; compared with the traditional numerical simulation method, the technical scheme of the embodiment of the invention can greatly improve the modeling efficiency; compared with the traditional data driving method, the technical scheme of the embodiment of the invention can improve the reliability of the model in a real open scene by considering physical mechanisms such as an electromagnetic field control equation, greatly improves the modeling precision and the interpretability, and has strong application value.
Further explanation of the technical scheme of the embodiment of the invention is that the embedded physical knowledge neural network is an application method of a scientific machine in the traditional numerical field, and is particularly used for solving various problems related to Partial Differential Equations (PDEs), including equation solving, parameter inversion, model discovery, control and optimization and the like. The principle of the physical information neural network is to approximate the solution of PDE by training the neural network to minimize the loss function, the term of which includes the residual terms of the initial and boundary conditions, and the partial differential equation residuals at selected points in the region (which should be conventionally referred to as "fitting points"), and the values at the space-time points are obtained by performing an Inference (information) after the training is completed.
The embodiment of the invention is specifically and exemplarily, for each training sample, the method comprises the following steps: further explaining the position coordinates, time and vector magnetic position labels, wherein the position coordinates are the position coordinates of the selected points, and the vector magnetic position labels are the vector magnetic position sizes; the vector magnetic bit size is NxT because the position coordinate information (x, y) is corresponding to each other in pairs to form a group; preferably, in order to meet the learning characteristics of the artificial neural network, the artificial neural network is flattened into a size of NT multiplied by 1, and the size of input information is correspondingly processed into a two-dimensional matrix with the size of NT multiplied by 3 during training.
The embodiment of the invention is specifically and exemplarily used for further explaining the selection of the matching points when the training sample set is acquired, and comprises the following steps:
10000 distribution points can be selected in a control area based on a Latin hypercube sampling method; at boundary conditions, 3000 distribution points are selected; at initial conditions, 3000 formulation points were selected.
In the embodiment of the invention, after the physical constraint neural network model is trained, different position information and time information are input according to the model, so that the vector magnetic potential distribution in the control body at the time can be predicted, and then B is obtained through the vector magnetic potential A, thereby rapidly modeling the formation and distribution of magnetic fluxes in the control body region;
wherein for a parallel planar field, the relationship of a to B is expressed as:
wherein, x and y are the horizontal and vertical coordinates of the adopted plane field (x and y), A represents the vector magnetic position, and B represents the magnetic induction intensity vector.
Further specifically, the embodiment of the invention provides an electromagnetic field distribution modeling technology based on deep learning, which comprises an electromagnetic field distribution substitution model construction strategy based on deep learning, and a physical constraint neural network construction training strategy; in the technical scheme of the embodiment of the invention, a mathematical physical model is obtained based on the actual electromagnetic field distribution condition, and comprises the steps of determining the size of a control area, the position of an iron core, a control equation, boundary conditions, time and the like; secondly, preprocessing input data, and determining the properties of the iron core, relative magnetic permeability, conductivity, original current density and the like; then, constructing a neural network model according to the mathematical physical model, constructing a loss function, and determining the weight corresponding to each item in the loss function; finally, training the neural network based on partial simulation, and after training, rapidly modeling the actual electromagnetic field distribution and the formation process by using the neural network, thereby solving the problems of low modeling speed and low efficiency of the traditional numerical simulation.
The embodiment of the invention is particularly exemplary, and compared with the traditional numerical calculation methods such as finite elements, the technical scheme of the embodiment of the invention can obviously improve the modeling efficiency of the electromagnetic field distribution of the transformer. Compared with a pure data driving method, the method can remarkably reduce the number of samples required by training and improve modeling accuracy, and the comparison result is shown in table 1.
TABLE 1 comparison results
Referring to table 1, the training data in table 1 can be understood as the required field data acquisition, that is, the data acquisition of vector magnetic bits is performed on each coordinate inside the transformer core, when the invention is used, I only need to perform acquisition training on the data on the boundary conditions, and the mathematical model deduced by the invention can know that the distribution point values on the boundary conditions are all the same constant 0, that is, after knowing the distribution point coordinates on the frame of the transformer core, the true value of the vector magnetic bit A can be marked as 0 by itself, and the field vector magnetic bit A value acquisition is not needed.
In summary, the embodiment of the invention relates to an electromagnetic field distribution modeling technology based on deep learning, in particular to an electromagnetic field distribution modeling technology based on a physical constraint neural network. According to a mathematical physical model of an electromagnetic field distribution process, a roll neural network point magnetic field distribution modeling technology based on physical information constraint is provided, training is carried out on a neural network based on partial simulation, the actual electromagnetic field forming and distribution process can be rapidly modeled by using the neural network after training is finished, the problems of low speed and low efficiency of traditional numerical simulation modeling are solved, and meanwhile the problem of poor interpretability of a traditional data driving method is solved. Further illustratively, the embodiment of the invention solves the technical problem of low solving efficiency of the three-dimensional electromagnetic field of the large-scale transformer caused by complex model, difficult acquisition of nonlinear characteristics of materials and large calculated amount of solving by constructing the electromagnetic field modeling agent model based on mechanism data fusion based on physical constraint deep learning. Furthermore, according to the modeling result, the electromagnetic field distribution in the transformer can be evaluated in real time and quantitatively, possible risk hidden danger in the transformer can be analyzed, operation and maintenance basis and reference are provided for transformer operation and detection personnel, and the digitization degree of the transformer is improved.
The following are device embodiments of the present invention that may be used to perform method embodiments of the present invention. For details not disclosed in the apparatus embodiments, please refer to the method embodiments of the present invention.
Referring to fig. 3, in still another embodiment of the present invention, an electromagnetic field modeling system based on mechanism data fusion is provided, including:
the data acquisition module is used for acquiring data required by electromagnetic field distribution modeling based on the transformer to be subjected to electromagnetic field distribution modeling; the data required by electromagnetic field distribution modeling comprises electromagnetic field distribution modeling area coordinates and preset modeling time;
the predicted value acquisition module is used for predicting by utilizing the trained physical constraint neural network model based on the acquired data required by the electromagnetic field distribution modeling to obtain a vector magnetic position matrix predicted value;
the electromagnetic field distribution acquisition module is used for calculating and acquiring an electromagnetic field quantity matrix based on the acquired vector magnetic potential matrix predicted value and acquiring electromagnetic field distribution based on the electromagnetic field quantity matrix;
the loss function adopted by the trained physical constraint neural network model in training is a physical constraint loss function constructed based on a vector magnetic level equation, a boundary condition, an initial condition and data information.
The system provided by the embodiment of the invention comprises an electromagnetic field distribution substitution model construction strategy based on deep learning and a neural network construction strategy of physical constraint; the invention to be protected is characterized by the following two points: (1) Constructing a strategy based on a deep learning electromagnetic field distribution substitution model; the electromagnetic field distribution substitution model based on deep learning is constructed by refining and summarizing the electromagnetic field formation and distribution process, so that the speed and efficiency of constructing the electromagnetic field distribution model are improved, and the problem of low prediction speed of the traditional numerical simulation method is solved; (2) a neural network construction strategy of physical constraint; the method comprises the steps of combining vector magnetic bit data information, control equation information, boundary condition information and initial condition information to construct a physical constraint neural network, and solving the problem of poor interpretability of the traditional deep learning method by introducing physical constraint.
In yet another embodiment of the present invention, a computer device is provided that includes a processor and a memory for storing a computer program including program instructions, the processor for executing the program instructions stored by the computer storage medium. The processor may be a central processing unit (Central Processing Unit, CPU), but may also be other general purpose processor, digital signal processor (Digital Signal Processor, DSP), application specific integrated circuit (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc., which are the computational core and control core of the terminal adapted to implement one or more instructions, in particular to load and execute one or more instructions within a computer storage medium to implement a corresponding method flow or a corresponding function; the processor disclosed by the embodiment of the invention can be used for operating an electromagnetic field modeling method based on mechanism data fusion.
In yet another embodiment of the present invention, a storage medium, specifically a computer readable storage medium (Memory), is a Memory device in a computer device, for storing a program and data. It is understood that the computer readable storage medium herein may include both built-in storage media in a computer device and extended storage media supported by the computer device. The computer-readable storage medium provides a storage space storing an operating system of the terminal. Also stored in the memory space are one or more instructions, which may be one or more computer programs (including program code), adapted to be loaded and executed by the processor. The computer readable storage medium herein may be a high-speed RAM memory or a non-volatile memory (non-volatile memory), such as at least one magnetic disk memory. One or more instructions stored in a computer-readable storage medium may be loaded and executed by a processor to implement the corresponding steps of the electromagnetic field modeling method in the above-described embodiments with respect to mechanism data fusion.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical aspects of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those of ordinary skill in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which is intended to be covered by the claims.
Claims (10)
1. An electromagnetic field modeling method based on mechanism data fusion is characterized by comprising the following steps:
acquiring data required by electromagnetic field distribution modeling based on a transformer to be subjected to electromagnetic field distribution modeling; the data required by electromagnetic field distribution modeling comprises electromagnetic field distribution modeling area coordinates and preset modeling time;
based on the acquired data required by the electromagnetic field distribution modeling, predicting by using a trained physical constraint neural network model to obtain a vector magnetic position matrix predicted value;
calculating to obtain an electromagnetic field quantity matrix based on the obtained vector magnetic bit matrix predicted value, and obtaining electromagnetic field distribution based on the electromagnetic field quantity matrix;
the loss function adopted by the trained physical constraint neural network model in training is a physical constraint loss function constructed based on a vector magnetic level equation, a boundary condition, an initial condition and data information.
2. The electromagnetic field modeling method based on mechanism data fusion according to claim 1, wherein the physical constraint neural network model adopts a multi-layer fully-connected neural network, and comprises an input layer, a plurality of hidden layers and an output layer; the input layer is used for inputting position coordinates and time, and the output layer is used for outputting vector magnetic bit matrix predicted values.
3. The electromagnetic field modeling method based on mechanism data fusion of claim 1, wherein the training step of the trained physical constraint neural network model comprises:
performing partial simulation based on the transformer to be subjected to electromagnetic field distribution modeling or the transformers of the same type to obtain an electromagnetic field distribution model sample;
acquiring a training sample set based on the electromagnetic field distribution model sample; each training sample in the training sample set comprises a distribution point boundary condition position coordinate, a moment and a vector magnetic position label;
training the physical constraint neural network model based on the training sample set, and obtaining the trained physical constraint neural network model after reaching a preset convergence condition; during training, a physical constraint loss function is calculated based on a selected training sample, a gradient between the output and the input of the physical constraint neural network is calculated, the feedforward neural network is counter-propagated according to the physical constraint loss function according to a gradient descent method, and weights and biases in the network are iteratively updated.
4. An electromagnetic field modeling method based on mechanism data fusion as defined in claim 1, wherein the physical constraint loss function MSE is expressed as,
where MSE b Indicating the degree to which the predicted value meets the boundary condition; MSE (mean square error) f Representing the error of substituting the predicted value obtained by the input feedforward neural network into the mathematical model; i represents an i-th data node; n (N) b The matching point quantity for the selected boundary condition;representing the setpoint at the boundary condition; />Representing a vector magnetic potential predicted value obtained after the training of the input neural network; />Is the true value of the vector magnetic bit; n (N) f Representing the amount of data points used; /> Representing the mathematical model used; />The vector magnetic potential predicted value obtained after the training of the input neural network is obtained; mu is magnetic permeability; sigma is conductivity; j (J) S Representing the source current density; x and y are the horizontal and vertical coordinates of the adopted planar field respectively; t isTime.
5. An electromagnetic field modeling system based on mechanism data fusion, comprising:
the data acquisition module is used for acquiring data required by electromagnetic field distribution modeling based on the transformer to be subjected to electromagnetic field distribution modeling; the data required by electromagnetic field distribution modeling comprises electromagnetic field distribution modeling area coordinates and preset modeling time;
the predicted value acquisition module is used for predicting by utilizing the trained physical constraint neural network model based on the acquired data required by the electromagnetic field distribution modeling to obtain a vector magnetic position matrix predicted value;
the electromagnetic field distribution acquisition module is used for calculating and acquiring an electromagnetic field quantity matrix based on the acquired vector magnetic potential matrix predicted value and acquiring electromagnetic field distribution based on the electromagnetic field quantity matrix;
the loss function adopted by the trained physical constraint neural network model in training is a physical constraint loss function constructed based on a vector magnetic level equation, a boundary condition, an initial condition and data information.
6. The electromagnetic field modeling system based on mechanism data fusion according to claim 5, wherein in the predicted value acquisition module, the physical constraint neural network model adopts a multi-layer fully-connected neural network, and comprises an input layer, a plurality of hidden layers and an output layer; the input layer is used for inputting position coordinates and time, and the output layer is used for outputting vector magnetic bit matrix predicted values.
7. The electromagnetic field modeling system based on mechanism data fusion of claim 5, wherein the training step of the trained physical constraint neural network model in the predicted value acquisition module comprises:
performing partial simulation based on the transformer to be subjected to electromagnetic field distribution modeling or the transformers of the same type to obtain an electromagnetic field distribution model sample;
acquiring a training sample set based on the electromagnetic field distribution model sample; each training sample in the training sample set comprises a distribution point boundary condition position coordinate, a moment and a vector magnetic position label;
training the physical constraint neural network model based on the training sample set, and obtaining the trained physical constraint neural network model after reaching a preset convergence condition; during training, a physical constraint loss function is calculated based on a selected training sample, a gradient between the output and the input of the physical constraint neural network is calculated, the feedforward neural network is counter-propagated according to the physical constraint loss function according to a gradient descent method, and weights and biases in the network are iteratively updated.
8. An electromagnetic field modeling system based on a mechanism data fusion as defined in claim 5, wherein in said predicted value acquisition module, said physical constraint loss function MSE is expressed as,
where MSE b Indicating the degree to which the predicted value meets the boundary condition; MSE (mean square error) f Representing the error of substituting the predicted value obtained by the input feedforward neural network into the mathematical model; i represents an i-th data node; n (N) b The matching point quantity for the selected boundary condition;representing the setpoint at the boundary condition; />Representing a vector magnetic potential predicted value obtained after the training of the input neural network; />Is the true value of the vector magnetic bit; n (N) f Representing data points usedAn amount of; /> Representing the mathematical model used; />The vector magnetic potential predicted value obtained after the training of the input neural network is obtained; mu is magnetic permeability; sigma is conductivity; j (J) S Representing the source current density; x and y are the horizontal and vertical coordinates of the adopted planar field respectively; t is time.
9. An electronic device, comprising:
at least one processor; the method comprises the steps of,
a memory communicatively coupled to the at least one processor; wherein,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the electromagnetic field modeling method based on mechanism data fusion of any one of claims 1 to 4.
10. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the electromagnetic field modeling method based on mechanism data fusion as defined in any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311278715.9A CN117350152A (en) | 2023-09-28 | 2023-09-28 | Electromagnetic field modeling method, system, equipment and medium based on mechanism data fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311278715.9A CN117350152A (en) | 2023-09-28 | 2023-09-28 | Electromagnetic field modeling method, system, equipment and medium based on mechanism data fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117350152A true CN117350152A (en) | 2024-01-05 |
Family
ID=89355158
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311278715.9A Pending CN117350152A (en) | 2023-09-28 | 2023-09-28 | Electromagnetic field modeling method, system, equipment and medium based on mechanism data fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117350152A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117709170A (en) * | 2024-02-05 | 2024-03-15 | 合肥工业大学 | Magnetic field rapid calculation method based on improved depth operator network |
| CN117952020A (en) * | 2024-03-26 | 2024-04-30 | 大连理工大学 | Multi-layer medium electromagnetic calculation method based on physical information neural network |
-
2023
- 2023-09-28 CN CN202311278715.9A patent/CN117350152A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117709170A (en) * | 2024-02-05 | 2024-03-15 | 合肥工业大学 | Magnetic field rapid calculation method based on improved depth operator network |
| CN117709170B (en) * | 2024-02-05 | 2024-04-19 | 合肥工业大学 | Magnetic field rapid calculation method based on improved depth operator network |
| CN117952020A (en) * | 2024-03-26 | 2024-04-30 | 大连理工大学 | Multi-layer medium electromagnetic calculation method based on physical information neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN117350152A (en) | Electromagnetic field modeling method, system, equipment and medium based on mechanism data fusion | |
| Jiang et al. | Two-stage structural damage detection using fuzzy neural networks and data fusion techniques | |
| Abbate et al. | Data-driven profile prediction for DIII-D | |
| Brekke et al. | Multimodal 3d object detection from simulated pretraining | |
| CN107292337A (en) | Ultralow order tensor data filling method | |
| CN113571122A (en) | Electronic density map determining method and device, electronic equipment and storage medium | |
| Mao et al. | Adaptive disassembly sequence planning for VR maintenance training via deep reinforcement learning | |
| CN116384244A (en) | Electromagnetic field prediction method based on physical enhancement neural network | |
| Zheng | Form finding and evaluating through machine learning: the prediction of personal design preference in polyhedral structures | |
| Xia et al. | Virtual comissioning of manufacturing system intelligent control | |
| Laddach et al. | An automatic selection of optimal recurrent neural network architecture for processes dynamics modelling purposes | |
| CN117875504A (en) | Dam break prediction and model training method and device, electronic equipment and storage medium | |
| CN117170980A (en) | Early warning method, device, equipment and storage medium for server hardware abnormality | |
| CN115062551B (en) | Wet physical process parameterization method based on time sequence neural network | |
| CN115688571A (en) | Method and device for monitoring running state of transformer, electronic equipment and storage medium | |
| CN114880929A (en) | Deep reinforcement learning-based multi-energy flow optimization intelligent simulation method and system | |
| Nasrollahzadeh et al. | Implementation of neuro-fuzzy system with modified high performance genetic algorithm on embedded systems | |
| CN105279320B (en) | A kind of method for generating FDTD grids | |
| Ma et al. | Application of BP neural networks in garment pattern design system | |
| Cooper et al. | Error homogenization in physics-informed neural networks for modeling in manufacturing | |
| CN111967596A (en) | Feature automatic intersection method based on deep learning in wind control scene | |
| Wei et al. | Study on battlefield situation assessment model of simulation entity based on stacked auto-encoder network | |
| Malov et al. | Synthetic data generation approach for face recognition system | |
| Wang et al. | Research on the Role of Hybrid Mesh Warm-up in Flow Prediction Based on Deep Learning | |
| Yang et al. | Computer aided product development and design based on information technology |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |