Method for estimating physical quantities of an electrostatic induction device assembly
By combining time-dependent partial differential equations and neural network systems with temperature models and measurement data, the problem of accuracy in estimating physical quantities of components in electrostatic induction devices was solved, achieving high-accuracy estimation in liquid environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HITACHI ENERGY LTD
- Filing Date
- 2024-03-15
- Publication Date
- 2026-07-14
Smart Images
Figure CN120641727B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for estimating physical quantities of components in an electrostatic induction device. Furthermore, this invention relates to each of a computer program product, a non-transitory computer-readable storage medium, and a control unit. Background Technology
[0002] In electrostatic induction device assemblies (such as assemblies including transformers or shunt reactors), it may be desirable to obtain information indicating physical quantities of the assembly. Purely by way of example, it may be desirable to determine information related to heat losses associated with at least a portion of the electrostatic induction device assembly.
[0003] However, obtaining this information from numerical models of components such as electrostatic induction devices can be a challenging task. Furthermore, since electrostatic induction devices may include those surrounded by liquid within a casing (such as a container), determining this information experimentally can also be a challenging task. Summary of the Invention
[0004] In view of the above, the object of the present invention is to provide a method for estimating physical quantities of components of an electrostatic induction device, which provides appropriate and reliable results.
[0005] Thus, the present invention relates to a method for estimating physical quantities of an electrostatic induction device assembly. The electrostatic induction device assembly includes a housing, an electrostatic induction device, and a liquid, wherein the housing contains the electrostatic induction device and the liquid such that the electrostatic induction device is at least partially, preferably completely, immersed in the liquid. The method includes using measured temperature data obtained from a measuring component. The measured temperature data includes the temperature at each of a plurality of different locations of the electrostatic induction device assembly as a function of time within a reference time range, when the electrostatic induction device assembly is subjected to conditions where at least a portion of the electrostatic induction device generates heat during at least a portion of a reference time range.
[0006] The method further includes:
[0007] - Using time-dependent partial differential equations, which characterize the physical conditions of the electrostatic induction device assembly during the reference time range, wherein the physical quantity constitutes the source term of the partial differential equation.
[0008] - A temperature model that generates estimated temperature data corresponding to the estimated temperature at each of multiple different locations of the electrostatic induction device assembly as a function of time. This temperature model includes a first neural network characterizing both the estimated and measured temperature data.
[0009] - Estimate the physical quantity by training a neural network system that uses at least the following entities: time-dependent partial differential equations, information from a temperature model, and a second neural network for the physical quantity.
[0010] The method described above implies that the estimation of physical quantities has appropriate accuracy because it uses information from temperature models to estimate the physical quantities of interest.
[0011] As used herein, the term “source term” in the partial differential equation is intended to encompass terms relating to the net inflow or net outflow of physical entities within a portion of an electrostatic induction device assembly.
[0012] As an example, the Laplace equation can often be solved by utilizing appropriate boundary conditions. This is used to establish the steady-state temperature distribution of the object. However, if there is a net inflow or net outflow of a physical entity within a part of the object that can be characterized by a function J, then the Laplace equation can be reformulated as follows: Thus, the function J is the source term in the Laplace equation, which is thereby reformulated.
[0013] It should be noted that, depending on the partial differential equation and the characteristics of the net inflow or outflow, the function J may depend on one or more parameters, such as at least one of the following: time t, location x, and temperature T. Therefore, the source term can be formulated, for example, as follows: .
[0014] By way of example only, a time-dependent partial differential equation can be a thermal equation with temperature T as the unknown variable. This example can be applied to every embodiment of the invention.
[0015] Optionally, the method further includes establishing a set of partial differential equation entities associated with the probability distribution of solutions to time-dependent partial differential equations. The method further includes generating a partial differential equation cost function comprising these partial differential equation entities, wherein training the neural network system includes:
[0016] - Determine the entity of the set of partial differential equations such that the corresponding value of the cost function of the partial differential equation is within a predetermined range, and / or
[0017] - Change the entity of the set of partial differential equations until the predetermined stopping condition is obtained.
[0018] The aforementioned characteristics imply that training this neural network system may involve an iterative process that performs multiple iterations. This, in turn, means that the likelihood of obtaining reasonably accurate results is reasonably high.
[0019] Optionally, the cost function of the partial differential equation includes at least one of the following:
[0020] - A set of residual entities and a residual associated with a time-dependent partial differential equation, preferably, the residual entities comprising residual functionals, or alternatively composed of residual functionals;
[0021] - A set of boundary condition entities and at least one boundary condition associated with a time-dependent partial differential equation, preferably, the boundary condition entities comprising boundary condition functionals, or alternatively composed of boundary condition functionals, and
[0022] - A set of initial condition entities and at least one initial condition associated with a time-dependent partial differential equation, preferably, the initial condition entities comprising initial condition functionals, or alternatively composed of initial condition functionals.
[0023] As used herein, the term "functional" is intended to encompass a mapping from a space to a range of real or complex numbers. The space can include functions, where a functional can be referred to as a function that takes another function as its input. As a non-restrictive example, an integral can be an example of a functional because it takes a function as its input, and depending on, for example, the properties of the function input to the integral, the result of the integral can be a real or complex number.
[0024] Optionally, the residuals associated with the time-dependent partial differential equations are determined at least using information from the temperature model, preferably using estimated temperatures at each of several different locations of the electrostatic induction device assembly as a function of time. In this way, information from the temperature model (e.g., estimated temperatures) can be used to determine the aforementioned set of residual entities. This further means that measured temperature data characterized by the temperature model can be used when determining the aforementioned residuals. However, the measured temperature data itself is not required when determining the residuals. Instead, the measured temperature data can be characterized by a first neural network, which in turn forms part of the temperature model. This means that the accuracy level of the residuals is appropriately high because the temperature model can be associated with an accuracy level exceeding that of the measured temperature data itself. For example, the temperature model can provide temperature information for locations and / or instances where the temperature was not measured.
[0025] Optionally, the set of partial differential equation entities includes a set of temperature entities associated with the probability distribution of the solutions to the temperature model, wherein the cost function of the partial differential equations includes an addend and a temperature cost function, the addend including the temperature entities, and the temperature cost function including estimated temperature data and measured temperature data. Preferably, the temperature entities include temperature hyperparameters, or alternatively, are composed of temperature hyperparameters.
[0026] Optionally, the method includes using measured temperature data to train a first neural network to obtain a temperature model, wherein the temperature model is subsequently used to train the neural network system.
[0027] As shown above, using a temperature model instead of measuring temperature data to train a neural network system only means that the method has a reasonably high accuracy.
[0028] Optionally, the step of obtaining the temperature model includes establishing a set of temperature entities associated with the probability distribution of the solutions to the temperature model. Preferably, the temperature entities include temperature hyperparameters, or alternatively, are composed of temperature hyperparameters, wherein training the first neural network includes: generating a temperature cost function, which includes a set of temperature entities, estimated temperature data, and measured temperature data, and
[0029] - Determine the set of temperature entities such that the corresponding value of the temperature cost function is within a predetermined temperature range, and / or
[0030] - Change the temperature of this group of entities until the predetermined stopping condition is met.
[0031] In this way, temperature entities (such as temperature hyperparameters) can be determined, for example, using an iterative process that performs multiple iterations. This, in turn, means that the likelihood of obtaining reasonably accurate results is reasonably high.
[0032] Optionally, the physical quantity characterizes at least one of the following of the components of the electrostatic induction device: heat loss through the housing, stray loss in the metal part of the electrostatic induction device, and hot spot temperature associated with the electrostatic induction device.
[0033] Optionally, the electrostatic induction device includes a transformer and / or a shunt reactor.
[0034] Optionally, the neural network system further utilizes temperature-dependent material properties of at least a portion of the casing, such as thermal conductivity.
[0035] A second aspect of the invention relates to a method for evaluating an electrostatic induction device assembly. The electrostatic induction device assembly includes an electrostatic induction device located inside a housing. The assembly further includes a liquid located inside the housing, the liquid surrounding the electrostatic induction device. The method includes:
[0036] - Arrange the electrostatic induction device assembly in a condition where at least a portion of the electrostatic induction device generates heat;
[0037] - The measurement component is used to determine the measured temperature data, which includes: the temperature at each of a plurality of different locations of the electrostatic induction device assembly as a function of time within a reference time range, when the electrostatic induction device assembly is subjected to conditions where at least a portion of the electrostatic induction device generates heat during at least a portion of a reference time range; and
[0038] - Use the method according to the first aspect of the invention to estimate the physical quantities of the electrostatic induction device components.
[0039] Optionally, the housing has an outer surface facing away from the interior of the housing, wherein the measuring component includes one or more sensing devices adapted to sense the temperature at multiple different locations on the outer surface of the housing.
[0040] Optionally, the measuring component includes one or more thermal imaging cameras, and the step of determining the measured temperature data includes capturing an image of the outer surface using the one or more thermal imaging cameras.
[0041] Optionally, the step of determining the measured temperature data includes transforming images captured by one or more thermal imaging cameras onto the outer surface.
[0042] A third aspect of the invention relates to a computer program product comprising program code for performing the method of the first aspect of the invention when executed by a processor device.
[0043] A fourth aspect of the invention relates to a non-transitory computer-readable storage medium comprising instructions that, when executed by a processor device, cause the processor device to perform the method of the first aspect of the invention.
[0044] The fifth aspect of the invention relates to a control unit arranged to perform the method of the first aspect of the invention. Attached Figure Description
[0045] Referring to the accompanying drawings, the following is a more detailed description of embodiments of the present disclosure cited as examples.
[0046] In the attached diagram:
[0047] Figure 1 This is a schematic diagram of the electrostatic induction device assembly;
[0048] Figure 2 This is a schematic diagram of a method embodiment of the present invention, and
[0049] Figure 3 This is a schematic diagram of another method embodiment of the present invention. Detailed Implementation
[0050] Preferred embodiments of this disclosure will be discussed below with reference to the accompanying drawings.
[0051] Figure 1 An embodiment of an electrostatic induction device assembly 10 is schematically illustrated. The assembly includes a housing 14, an electrostatic induction device 12, and a liquid 18, wherein the housing 14 contains the electrostatic induction device 12 and the liquid 18 such that the electrostatic induction device 12 is at least partially, preferably completely, immersed in the liquid 18. Figure 1 In some embodiments, the electrostatic induction device 12 is completely immersed in the liquid 18. As a non-limiting example, the electrostatic induction device 12 may include a transformer and / or a shunt reactor, or even be composed of a transformer and / or a shunt reactor. Furthermore, as... Figure 1 As shown, the outer casing 14 can be defined by the outer casing wall 16.
[0052] By way of example only, the outer casing 16 may include a metallic material (such as steel), or may even be made of a metallic material.
[0053] Furthermore, as a non-limiting example, the casing 14 may be referred to as a can.
[0054] Furthermore, purely by way of example, liquid 18 may include a dielectric liquid (such as mineral oil), or even be composed of a dielectric liquid.
[0055] The electrostatic induction device assembly 10 can be adapted to conditions in which at least a portion of the electrostatic induction device 12 generates heat. This is purely by way of example and as... Figure 1 As indicated, when the electrostatic induction device 12 is operated, the electrostatic induction device can generate internal heat loss q. int As a non-limiting example, if the electrostatic induction device 12 includes a transformer and / or a shunt reactor, or is even composed of a transformer and / or a shunt reactor, then the internal heat loss q int This internal heat loss q may be caused by the core (not shown) and windings (not shown) of the electrostatic induction device 12. int It can propagate towards the outer shell wall 16 via the liquid 18.
[0056] In addition, heat can propagate through the outer casing wall 16, resulting in heat loss q in the outer casing wall. 壁 Purely by way of example, the magnitude of the heat loss q of the outer casing wall. 壁 This can depend on the temperature difference on the outer casing wall 16 (i.e., the temperature difference between the liquid 18 inside the outer casing wall and the temperature of the fluid (such as air) surrounding the outer casing wall 16), and the outer casing heat transfer parameters (such as the wall diffusion coefficient k, which indicates diffusion on the outer casing wall 16). 壁 ).
[0057] The method of the present invention includes using measured temperature data obtained from the measuring component 20. The measured temperature data includes: when the electrostatic induction device assembly 10 is in at least a portion of the electrostatic induction device 12 within a reference time range ΔT. ref When heat is generated during at least a portion of the period, the reference time range ΔT is used. ref The temperature T at each of the multiple different locations of the electrostatic induction device assembly 10 is a function of time t. 真实(x, t). For completeness, it should be noted that the position is usually represented by x in this article for clarification: for example, depending on the characteristics of the measured temperature data, the position may be represented in one, two, or three dimensions.
[0058] like Figure 1 As indicated, the measurement component 20 may include one or more measurement entities.
[0059] exist Figure 1 In one embodiment, the housing 14 has an outer surface 22 facing away from the interior of the housing 14. This is purely by way of example and as... Figure 1 As indicated, the outer surface 22 may be the outer surface of the housing wall 16. Furthermore, the measuring component 20 may include one or more sensing devices adapted to sense the temperature at multiple different locations on the outer surface 22 of the housing 14.
[0060] By way of example only, the measuring component 20 may include one or more thermal imaging cameras 24. Figure 1 (Only one camera is indicated in the text), and the steps to determine the measured temperature data may include capturing images of the outer surface using one or more thermal imaging cameras.
[0061] However, implementations of the measuring component 20 may include other types of sensors. For this purpose, please refer again... Figure 1 The array 26 of temperature sensors is located inside the housing 14, and each temperature sensor in the array 26 can, for example, measure the temperature of the liquid 18 located inside the housing 14.
[0062] For completeness, it should be noted that the implementation of the measurement component 20 may include multiple different sensors of different types.
[0063] Regardless of how the reference time range ΔT is determined... ref The temperature T at each of the multiple different locations of the electrostatic induction device assembly 10 is a function of time t. 真实 (x, t), the methods of the present invention all include:
[0064] - Using time-dependent partial differential equations, which characterize the electrostatic induction device assembly over a reference time range ΔT 参考 The physical conditions during the period, wherein the physical quantity constitutes the source term of the partial differential equation;
[0065] - Generate estimated temperature data T est The temperature model (x, t) estimates the temperature data T. est(x, t) corresponds to the estimated temperature at each of a plurality of different locations of the electrostatic induction device assembly 10 as a function of time t, the temperature model including the estimated temperature data T. est (x, t) and the measured temperature data T 真实 The first neural network NN1 with (x, t), and
[0066] - Estimate the physical quantity by training a neural network system that uses at least the following entities: time-dependent partial differential equations, information from a temperature model, and a second neural network NN2 for the physical quantity.
[0067] As a non-limiting example, the physical quantity can characterize at least one of the following of the electrostatic induction device assembly 10: heat loss through the housing 14, stray loss in the metal portion of the electrostatic induction device 12, and hot spot temperature associated with the electrostatic induction device 12.
[0068] Purely through examples, a neural network system can be called a physical information neural network.
[0069] also, Figure 1 Control unit 28 is indicated. Purely by way of example, control unit 28 can be arranged to perform the method of the first aspect of the invention. For this purpose, although purely by way of example, control unit 28 can be adapted to receive information from relevant entities (such as measuring component 20 and electrostatic induction device 12).
[0070] Purely through examples, the method further includes establishing a set of partial differential equation entities β associated with the probability distribution of solutions to time-dependent partial differential equations. PDE , ß IC , ß BC The method further includes generating entities including partial differential equations β. PDE , ß IC , ß BC Partial differential equation cost function L 总 Training the neural network system includes:
[0071] - Determine the entity β of this set of partial differential equations PDE , ß IC , ß BC This makes the cost function of the partial differential equation... L 总 The corresponding value is within the predetermined range, and / or
[0072] - Change the entity of the set of partial differential equations until the predetermined stopping condition is obtained.
[0073] As a non-restrictive example, the cost function of a partial differential equation may include at least one of the following:
[0074] - A set of residual entities associated with time-dependent partial differential equations ß PDE and a residual e PDE Preferably, residual entity ß PDE Including residual functionals, and alternatively composed of residual functionals;
[0075] - A set of boundary condition entities associated with time-dependent partial differential equations ß BC and at least one boundary condition L BC Preferably, the boundary condition entity includes a boundary condition functional, or alternatively, is composed of boundary condition functionals, and
[0076] - A set of initial condition entities associated with time-dependent partial differential equations ß IC and at least one initial condition L IC Preferably, the initial condition entity includes an initial condition functional, or alternatively, is composed of an initial condition functional.
[0077] As a non-restrictive example, residuals associated with time-dependent partial differential equations e PDE The estimated temperature T at each of several different locations x of the electrostatic induction device assembly 10, preferably as a function of time t, can be determined using at least information from a temperature model. est (x, t) can be used to determine this. Thus, although only by way of example, the estimated temperature data T can be used. est The residual is determined by at least a portion of (x, t). e PDE .
[0078] Purely by way of example, this set of partial differential equation entities comprises a set of temperature entities associated with the probability distribution of the solutions to the temperature model. ß T The cost function of this partial differential equation includes addends. ß T L T Temperature cost function L T The addend includes the temperature entity. ß T The temperature cost function includes estimated temperature data T. est(x, t) and measured temperature data T 真实 (x,t). Preferably, temperature entity. ß T This includes temperature hyperparameters, which can be alternatively composed of temperature hyperparameters.
[0079] Thus, in embodiments of the present invention, the first neural network NN1 and the neural network system can be trained simultaneously. Figure 2 Such an embodiment is schematically illustrated in the figure.
[0080] Optionally, the method may include using measured temperature data T 真实 (x, t) are used to train the first neural network NN1 to obtain a temperature model, which is then used to train the neural network system.
[0081] Thus, as Figure 3 As illustrated in the embodiments of the invention, a first neural network NN1 can be trained in the first step S10, and the result of the first step S10 will subsequently be used to train the neural network system, see [link to documentation]. Figure 3 Step S12 in the process.
[0082] In addition, such as Figure 3 As indicated, step S10 of obtaining the temperature model includes establishing a set of temperature entities associated with the probability distribution of the solutions to the temperature model. ß T Preferably, the temperature entity includes temperature hyperparameters, or alternatively, is composed of temperature hyperparameters, wherein training the first neural network NN1 includes: generating a temperature cost function. L 总 The temperature cost function includes the set of temperature entities. ß T Estimated temperature data T est (x, t) and measured temperature data T 真实 (x, t), and
[0083] - Determine the temperature entities in this group ß T This makes the temperature cost function L 总 The corresponding value is within the predetermined temperature range, and / or
[0084] - Change the temperature of this group of entities ß T Until the predetermined stopping condition is met.
[0085] Thus, temperature entity ß T (e.g., temperature over-parameter) ß TThis can be determined, for example, using an iterative process that performs multiple iterations. Subsequently, a temperature entity can be used. ß T In order to determine the estimated temperature data T est (x, t).
[0086] It should be noted that the process of training the first neural network NN1 can be performed in the manner presented in US 10,963,540 B2. It should also be noted that any of the training processes presented in US 10,963,540 B2 can be applied to the above-mentioned procedures. Figure 2 and Figure 3 Each of the embodiments presented.
[0087] Optionally, the neural network system further uses the following: temperature-dependent material properties D(T) of at least a portion of the shell 14, such as thermal conductivity. Figure 2 and Figure 3 Each of the terms indicates this possibility.
[0088] It should be noted that, although Figure 2 The embodiment simultaneously trains the first neural network NN1 and the neural network system and Figure 3 The embodiments of the invention train a first neural network NN1 in front of the neural network system, but other embodiments of the method of the invention are also envisioned.
[0089] By way of example only, embodiments of the method of the present invention are envisioned to employ an iterative process in which a first neural network NN1 is trained only partially, and the temperature model obtained from such partial training is subsequently used to train the neural network system. Thereafter, the first neural network NN1 can be further trained, and the resulting temperature model can be used to further train the neural network system.
[0090] Purely by way of example, the first neural network NN1 can be trained only partially by selecting a first predetermined temperature range and determining, for example, temperature entities. ß T This makes the temperature cost function L 总 The corresponding value falls within the first predetermined temperature range. Subsequently, in further training of the first neural network NN1, a second predetermined temperature range narrower than the first predetermined temperature range can be used, thereby enabling the determination of the temperature entity. ß T This makes the temperature cost function L 总 The corresponding value is within the second predetermined temperature range.
[0091] A second aspect of the invention relates to a method for evaluating an electrostatic induction device assembly 10. The electrostatic induction device assembly 10 includes an electrostatic induction device 12 located inside a housing 14. The assembly 10 further includes a liquid 18 located inside the housing 14, the liquid 18 at least partially surrounding the electrostatic induction device 12. The method includes:
[0092] - The electrostatic induction device assembly 10 is arranged under conditions where at least a portion of the electrostatic induction device 12 generates heat;
[0093] - Use measurement component 20 to determine the measured temperature data T 真实 (x, t), measured temperature data T 真实 (x, t) includes: when the electrostatic induction device assembly 10 is at least a portion of the electrostatic induction device 12 within the reference time range ΔT ref When heat is generated during at least a portion of the period, the reference time range ΔT is used. ref The temperature at each of the multiple different locations of the electrostatic induction device assembly, as a function of time, and
[0094] - Use the method according to the first aspect of the invention to estimate the physical quantities of the electrostatic induction device components.
[0095] Refer again Figure 1 The housing 14 may have an outer surface 22 facing away from the interior of the housing 14, wherein the measuring component includes one or more sensing devices adapted to sense the temperature at multiple different locations on the outer surface 22 of the housing 14.
[0096] As a non-limiting example, the step of determining the measured temperature data may include transforming images captured by one or more thermal imaging cameras 24 onto the outer surface.
[0097] The method according to the invention will now be illustrated by the following time-dependent partial differential equation:
[0098] Equation 1
[0099] in:
[0100] T = T(x, t) represents the temperature at each of several distinct locations (x) as a function of time (t);
[0101] D(T) characterizes at least a portion of the temperature-dependent material properties (e.g., thermal conductivity) of the outer shell 14, and
[0102] q 壁(I(t),T) characterizes the heat loss of the outer casing wall, which can depend on either the current I(t) fed to the electrostatic induction device 12 or the temperature T (e.g., the temperature of the inside of the outer casing wall 16).
[0103] As can be understood from the above, the physical quantities constituting the source terms of the partial differential equation presented in Equation 1 above are related to the heat loss q of the outer shell wall. 壁 Let's take (I(t),T) as an example.
[0104] Furthermore, as indicated in the above description, the method of the present invention includes a temperature model for generating estimated temperature data. The estimated temperature data corresponds to the estimated temperature T at each of a plurality of different locations x of the electrostatic induction device assembly 10 as a function of time t. est (x, t). The temperature model includes a representation of the estimated temperature data T. est (x, t) and measured temperature data T 真实 The first neural network NN1 of (x, t) (see...) Figure 2 and Figure 3 Each of them).
[0105] refer to Figure 3 Estimated temperature data T est (x, t) can be obtained in a separate step S10 by using the measured temperature data T. 真实 (x, t) are determined by training the first neural network NN1. Thus, Figure 3 The result of step S10 can be the estimated temperature data T. est (x, t).
[0106] Therefore and as Figure 3 The indicated steps for obtaining the temperature model include establishing a set of temperature entities associated with the probability distribution of the solutions to the temperature model. ß T .
[0107] Preferably, though not necessarily, temperature entity ß T This includes temperature hyperparameters, which can be alternatively composed of temperature hyperparameters, and training the first neural network NN1 includes generating a temperature cost function. L 总 The temperature cost function includes the set of temperature entities. ß T Estimated temperature data T est (x, t) and measured temperature data T 真实 (x, t). As a non-restrictive example, estimate the temperature data T. est (x, t) and measured temperature data T 真实(x, t) can be included in the cost function L T (T est (x, t), T 真实 In (x, t), the total cost function can be defined as follows: L 总 = ß T L T (T est (x, t), T 真实 (x, t)).
[0108] Furthermore, although this is done purely through examples, the process of obtaining a temperature model can include...
[0109] - Determine the temperature entity of this group ß T This makes the temperature cost function L 总 The corresponding value is within the predetermined temperature range, and / or
[0110] - Change the temperature of this group of entities ß T Until the predetermined stopping condition is met.
[0111] Regardless of how the estimated temperature data T is determined, it is based on this. est The temperature model (x, t) yields the estimated temperature data T. est (x, t) can then be entered into Equation 1 above according to the following:
[0112] Equation 2
[0113] As can be understood from the above, the estimated temperature data T in Equation 2 above can be used. est (x, t) and, for example, assuming the current I(t) is known, then the unknown entity in Equation 2 above and the heat loss q of the outer shell wall. 壁 (I(t),T) is related.
[0114] Heat loss q of outer casing wall 壁 (I(t),T) can be estimated by training a neural network system that uses at least the following entities: a time-dependent partial differential equation (see, for example, Equation 2), information from a temperature model (see, for example, Equation 2), and a second neural network NN2 for that physical quantity. Purely by way of example, the above training can be performed as follows: Figure 3 Performed in the indicated individual step S12.
[0115] In this way, a set of partial differential equation entities β can be established that are associated with the probability distribution of the solutions to time-dependent partial differential equations. PDE , ß IC , ß BC To determine the estimated value q of the heat loss of the outer casing wall. est (I(t),T).
[0116] This method may further include generating entities including partial differential equations β. PDE , ß IC , ß BC Partial differential equation cost function L 总 Training the neural network system can include:
[0117] - Determine the entity β of this set of partial differential equations PDE , ß IC , ß BC This makes the cost function of the partial differential equation... L 总 The corresponding value is within the predetermined range, and / or
[0118] - Change the entity β of this set of partial differential equations PDE , ß IC , ß BC Until the predetermined stopping condition is met.
[0119] In addition, for example Figure 3 The indicated cost function of the partial differential equation L 总 It may include at least one of the following:
[0120] - A set of residual entities associated with time-dependent partial differential equations ß PDE and a residual e PDE Preferably, residual entity e PDE Including residual functionals, and alternatively composed of residual functionals;
[0121] - A set of boundary condition entities associated with time-dependent partial differential equations ß BC and at least one boundary condition L BC Preferably, the boundary condition entity ß BC This includes boundary condition functionals, functionals that can be alternatively constructed from boundary condition functionals, and...
[0122] - A set of initial condition entities associated with time-dependent partial differential equations ß ICand at least one initial condition L IC Preferably, the initial condition entity ß IC It includes initial condition functionals, and can be alternatively composed of initial condition functionals.
[0123] exist Figure 3 In the example, the cost function of the partial differential equation L 总 This includes each of the entities listed in the three items above, but also considers the cost function of partial differential equations. L 总 Other implementations may include only one or two of the entities listed in the above three items.
[0124] Furthermore, in the partial differential equation cost function of the method of this invention L 总 Includes a set of residual entities ß PDE In this implementation, at least information from the temperature model can be used to determine the residuals associated with the time-dependent partial differential equations. e PDE .
[0125] Therefore, although purely by way of example, the method could include: testing various residual entities using a neural network system. ß PDE For example, various residual functionals. For each residual entity being evaluated ß PDE The residual entity evaluated in this way can be used. ß PDE and estimated temperature data T est (x, t) is used to determine the residual. e PDE Purely through examples, residuals e PDE It can be determined based on the following:
[0126] Equation 3
[0127] Alternatively:
[0128] Equation 4
[0129] By identifying appropriate entities (such as appropriate residual entities) ß PDE This allows us to determine an estimate of the heat loss q from the outer casing wall. est (I(t),T) are the physical quantities of interest in the above examples.
[0130] It should be noted that although the above example uses heat loss of the outer casing wall as an example, the method of the present invention can also be used for other examples of this physical quantity. As a non-limiting example, the physical entity can be related to stray losses in the metal portion of the electrostatic induction device 12 and / or the hot spot temperature associated with the electrostatic induction device 12.
[0131] In such an example, the process presented above, starting with Equation 1, can be used, but it can be achieved, for example, by replacing the heat loss q of the outer shell wall with another source term. 壁 (I(t),T) source term, or through heat loss q from the outer shell wall 壁 The source term (I(t),T) is used to add another source term to update the source term of the partial differential equation in Equation 1.
[0132] Furthermore, although the example presented above, beginning with Equation 1, includes two separate steps, S10 and S12, other examples of the method of the present invention can also be performed in a single step.
[0133] Finally, it should be understood that the characteristics of the described method for estimating physical quantities of an electrostatic induction device assembly are applicable to all embodiments of this method that fall within the scope of the appended claims.
Claims
1. A method for estimating physical quantities of an electrostatic induction device assembly (10), the electrostatic induction device assembly (10) comprising a housing (14), an electrostatic induction device (12), and a liquid (18), wherein, The housing (14) accommodates the electrostatic induction device (12) and the liquid (18) such that the electrostatic induction device (12) is at least partially immersed in the liquid (18). The physical quantity represents at least one of the following of the electrostatic induction device assembly (10): heat loss through the housing (14), stray loss in the metal portion of the electrostatic induction device (12), and hot spot temperature associated with the electrostatic induction device (12). The method includes using measured temperature data (T) obtained from the measuring component (20). 真实 (x, t)), the measured temperature data includes: when the electrostatic induction device assembly (10) is in at least a portion of the electrostatic induction device (12) within a reference time range (ΔT). ref When heat is generated during at least a portion of the period of ), the reference time range (ΔT) is used as the reference time range. ref The temperature at each of the multiple different locations (x) of the electrostatic induction device assembly (10) is a function of time (t) within a period of time. The method further includes: - Using time-dependent partial differential equations, the electrostatic induction device assembly (10) is characterized within the reference time range (ΔT). ref The physical conditions during the period, wherein the physical quantities constitute the source terms of the partial differential equation; - Generate estimated temperature data (T) est A temperature model (x, t) is provided, wherein the estimated temperature data corresponds to the estimated temperature at each of the plurality of different locations (x) of the electrostatic induction device assembly (10) as a function of time (t), and the temperature model includes a characterization of the estimated temperature data (T). est (x, t) and the measured temperature data (T) 真实 The first neural network (NN1) of (x,t) and - Estimate the physical quantity by training a neural network system that uses at least the following entities: the time-dependent partial differential equation, information from the temperature model, and an estimated value q characterizing the physical quantity. est The second neural network (NN2) of (I(t), T).
2. The method according to claim 1, wherein, The electrostatic induction device (12) is completely immersed in the liquid (18).
3. The method according to claim 1, wherein, The method further includes establishing a set of partial differential equation entities (β) associated with the probability distribution of the solutions to the time-dependent partial differential equation. PDE , ß IC , ß BC The method further includes generating a partial differential equation cost function that includes the entity of the partial differential equation. L 总 ), wherein training the neural network system includes: - Determine the entity of the set of partial differential equations (ß) PDE , ß IC , ß BC ), making the cost function of the partial differential equation ( L 总 The corresponding value of ) is within the predetermined range, and - Change the entity of the set of partial differential equations (ß) PDE , ß IC , ß BC (until the predetermined stopping condition is met).
4. The method according to claim 3, wherein, The partial differential equation cost function ( L 总 Includes at least one of the following: - A set of residual entities associated with the time-dependent partial differential equation ( ß PDE ) and a residual ( e PDE ); - A set of boundary condition entities associated with the time-dependent partial differential equation ( ß BC ) and at least one boundary condition ( L BC ),as well as - A set of initial condition entities associated with the time-dependent partial differential equation ( ß IC ) and at least one initial condition ( L IC ).
5. The method according to claim 4, wherein, The residual entity ( ß PDE This includes residual functionals.
6. The method according to claim 4, wherein, The boundary condition entity ( ß BC This includes boundary condition functionals.
7. The method according to claim 4, wherein, And the initial condition entity ( ß IC This includes the initial condition functional.
8. The method according to claim 4, wherein, The residuals associated with the time-dependent partial differential equation ( e PDE At least information from the temperature model should be used to determine this.
9. The method according to claim 8, wherein, The information from the temperature model is the estimated temperature data (T) at each of the plurality of different locations (x) of the electrostatic induction device assembly (10) as a function of time (t). est (x, t)).
10. The method according to any one of claims 3 to 9, wherein, The set of partial differential equation entities includes a set of temperature entities associated with the probability distribution of the solutions to the temperature model. ß T ), wherein the cost function of the partial differential equation includes addends ( ß T L T ) and temperature cost function ( L T The addend includes the temperature entity ( ß T The temperature cost function includes the estimated temperature data (T). est (x, t) and the measured temperature data (T) 真实 (x, t)).
11. The method according to claim 10, wherein, The temperature entity ( ß T This includes temperature hyperparameters.
12. The method according to any one of claims 1 to 9, wherein, The method includes using the measured temperature data (T) 真实 (x, t) is used to train the first neural network (NN1) to obtain the temperature model, wherein the temperature model is then used to train the neural network system.
13. The method of claim 12, wherein, The steps to obtain the temperature model include: establishing a set of temperature entities associated with the probability distribution of the solutions to the temperature model. ß T ), wherein training the first neural network includes generating a temperature cost function ( L 总 The temperature cost function includes the set of temperature entities ( ß T The estimated temperature data (T) est (x, t) and the measured temperature data (T) 真实 (x, t)), and - Determine the set of temperature entities ( ß T ), making the temperature cost function ( L 总 The corresponding value of ) is within the predetermined temperature range, and - Change the set of temperature entities ( ß T (until the predetermined stopping condition is met).
14. The method according to claim 13, wherein, The temperature entity includes temperature hyperparameters.
15. The method according to any one of claims 1 to 9, wherein, The neural network system further utilizes the temperature-dependent material properties (D(T)) of at least a portion of the housing (14).
16. The method according to claim 15, wherein, The temperature-related material property is thermal conductivity.
17. A method for evaluating an electrostatic induction device assembly (10), the electrostatic induction device assembly (10) including an electrostatic induction device (12) located inside a housing (14), the assembly further including a liquid (18) located inside the housing (14), the liquid (18) surrounding the electrostatic induction device (12), the method comprising: - The electrostatic induction device assembly (10) is arranged in a condition where at least a portion of the electrostatic induction device (12) generates heat; - Use the measuring component (20) to determine the measured temperature data (T) 真实 (x, t)), the measured temperature data (T) 真实 (x, t) includes: when the electrostatic induction device assembly (10) is at least a portion of the electrostatic induction device (12) within a reference time range (ΔT). ref When heat is generated during at least a portion of the period of ), the reference time range (ΔT) is used as the reference time range. ref The temperature at each of the multiple different locations (x) of the electrostatic induction device assembly (10) as a function of time (t) within a period of time, and - Use the method according to any one of the preceding claims to estimate the physical quantities of the electrostatic induction device assembly (10).
18. The method according to claim 17, wherein, The housing (14) has an outer surface (22) facing away from the interior of the housing (14), wherein the measuring component (20) includes one or more sensing devices adapted to sense the temperature at multiple different locations on the outer surface (22) of the housing (14).
19. The method according to claim 18, wherein, The measurement component (20) includes one or more thermal imaging cameras (24), and the step of determining the measured temperature data includes capturing an image of the outer surface (22) using the one or more thermal imaging cameras (24).
20. The method according to claim 19, wherein, The step of determining the measured temperature data includes transforming the image captured by the one or more thermal imaging cameras onto the outer surface (22).
21. A computer program product comprising program code for performing the method as described in any one of claims 1 to 16 when executed by a processor device.
22. A non-transitory computer-readable storage medium comprising instructions that, when executed by a processor device, cause the processor device to perform the method as described in any one of claims 1 to 16.
23. A control unit (28) arranged to perform the method according to any one of claims 1 to 16.