Component Thermal Simulation Method, Neural Network Training Method, Apparatus and Device

Abstract

The present disclosure provides a component thermal simulation method, a training method, device, and equipment for a neural network, relating to the field of artificial intelligence technology, and particularly to technical fields such as thermal simulation and deep learning. The component thermal simulation method includes: determining a target position in a target component; determining at least one target configuration condition among a plurality of preset configuration conditions, and determining at least one configuration condition information corresponding to the at least one target configuration condition, wherein each of the plurality of preset configuration conditions designates at least one relevant position in the target component, and the configuration condition information characterizes the thermal simulation information of at least one relevant position in the target component designated by the corresponding target configuration condition; determining configuration selection information, the configuration selection information indicating the selection of the at least one target configuration condition among the plurality of preset configuration conditions; and inputting the target position, the at least one configuration condition information, and the configuration selection information into a neural network for component thermal simulation to obtain a thermal simulation result of the target position.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, particularly to technical fields such as thermal simulation and deep learning. Specifically, it relates to a method for component thermal simulation, a method for training a neural network, a component thermal simulation device, a neural network training device, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] Artificial intelligence is a discipline that studies the use of computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), with both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include several major directions such as natural language processing technology, computer vision technology, speech recognition technology, and machine learning / deep learning, big data processing technology, and knowledge graph technology.

[0003] Currently, the high-performance processor industry faces increasingly prominent thermal challenges. Due to the uneven distribution of power consumption within functional units, it leads to unpredictable and uncontrollable temperature unevenness, triggering thermal problems. This thermal problem seriously affects the reliability and performance of the device, accelerating the occurrence of failure mechanisms. As the chip size continues to decrease and the transistor density increases, the thermal problem becomes more serious, especially in three-dimensional stacked chips. Traditional heat dissipation methods have proven insufficient to cope with the increasing thermal density. Therefore, component thermal simulation has become an important part of the integrated circuit design process. Accurately predicting the temperature distribution helps analyze the thermal characteristics of components and provides a basis for power optimization, layout planning, etc.

[0004] The methods described in this section are not necessarily methods that have been previously conceived or adopted. Unless otherwise specified, no method described in this section should be considered prior art solely because it is included in this section. Similarly, unless otherwise specified, the problems mentioned in this section should not be considered to have been recognized in any prior art. Summary of the Invention

[0005] The present disclosure provides a method for component thermal simulation, a method for training a neural network, a component thermal simulation device, a neural network training device, an electronic device, a computer-readable storage medium, and a computer program product.

[0006] According to one aspect of the present disclosure, there is provided a method for component thermal simulation, including: determining a target position in a target component; determining at least one target configuration condition among a plurality of preset configuration conditions, and determining at least one configuration condition information corresponding to the at least one target configuration condition, wherein each of the plurality of preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes thermal simulation information of at least one relevant position in the target component specified by the corresponding target configuration condition; determining configuration selection information, the configuration selection information indicating a selection of at least one target configuration condition among the plurality of preset configuration conditions; and inputting the target position, the at least one configuration condition information, and the configuration selection information into a neural network for component thermal simulation to obtain a thermal simulation result of the target position.

[0007] According to another aspect of the present disclosure, there is provided a method for training a neural network, including: determining a plurality of sample positions in a target component; determining a plurality of preset configuration conditions, and determining a plurality of configuration condition information corresponding to the plurality of preset configuration conditions, wherein each of the plurality of preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes thermal simulation information of at least one relevant position in the target component specified by the corresponding preset configuration condition; determining a plurality of configuration selection information corresponding to the plurality of sample positions respectively, the configuration selection information indicating a selection of at least one target configuration condition among the plurality of preset configuration conditions, and the plurality of configuration selection information covering the plurality of preset configuration conditions; inputting the plurality of sample positions, the plurality of configuration selection information, and at least one configuration condition information corresponding to at least one target configuration condition indicated by each of the plurality of configuration selection information into the neural network to obtain a plurality of thermal simulation results corresponding to the plurality of sample positions respectively and corresponding to the plurality of configuration selection information; and adjusting parameters of the neural network based on the plurality of thermal simulation results corresponding to the plurality of sample positions respectively and corresponding to the plurality of configuration selection information to obtain a neural network for target component thermal simulation.

[0008] According to another aspect of the present disclosure, there is provided a component thermal simulation device, including: a first determination unit configured to determine a target position and a plurality of preset configuration conditions in a target component; a second determination unit configured to determine at least one target configuration condition among the plurality of preset configuration conditions and determine at least one configuration condition information corresponding to the at least one target configuration condition, wherein each of the plurality of preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes the thermal simulation information of at least one relevant position in the target component specified by the corresponding target configuration condition; a third determination unit configured to determine configuration selection information, where the configuration selection information indicates the selection of at least one target configuration condition among the plurality of preset configuration conditions; and a first simulation unit configured to input the target position, the at least one configuration condition information, and the configuration selection information into a neural network for component thermal simulation to obtain a thermal simulation result of the target position.

[0009] According to another aspect of the present disclosure, there is provided a neural network training device, including: a fourth determination unit configured to determine a plurality of sample positions in a target component; a fifth determination unit configured to determine a plurality of preset configuration conditions and determine a plurality of configuration condition information corresponding to the plurality of preset configuration conditions, wherein each of the plurality of preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes the thermal simulation information of at least one relevant position in the target component specified by the corresponding preset configuration condition; a sixth determination unit configured to determine a plurality of configuration selection information corresponding to the plurality of sample positions respectively, where the configuration selection information indicates the selection of at least one target configuration condition among the plurality of preset configuration conditions, and the plurality of configuration selection information covers the plurality of preset configuration conditions; a second simulation unit configured to input the plurality of sample positions, the plurality of configuration selection information, and at least one configuration condition information corresponding to at least one target configuration condition indicated by each of the plurality of configuration selection information into the neural network to obtain a plurality of thermal simulation results corresponding to the plurality of sample positions respectively and corresponding to the plurality of configuration selection information; and a parameter adjustment unit configured to adjust the parameters of the neural network based on the plurality of thermal simulation results corresponding to the plurality of sample positions respectively and corresponding to the plurality of configuration selection information to obtain a neural network for target component thermal simulation.

[0010] According to another aspect of the present disclosure, there is provided an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the above method.

[0011] According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the above method.

[0012] According to another aspect of the present disclosure, there is provided a computer program product including a computer program, wherein the computer program, when executed by a processor, implements the above method.

[0013] According to one or more embodiments of the present disclosure, the present disclosure inputs the target position in the target component, at least one target configuration condition determined among a plurality of preset configuration conditions, and configuration selection information indicating the selection of at least one target configuration condition among the plurality of preset configuration conditions into a neural network for component thermal simulation, so that the neural network can learn the target configuration condition corresponding to the received configuration condition information, and thus can perform thermal simulation on the target component according to the target configuration condition to obtain the thermal simulation result of the target position. Through the above method, it is realized to use a single neural network to process the thermal simulation of a certain position of the target component under different configuration conditions, so that it is not necessary to train or deploy multiple neural networks for different configuration conditions, reducing the training cost and deployment cost.

[0014] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings exemplarily show embodiments and form a part of the description, and are used together with the written description of the description to explain the exemplary embodiments of the embodiments. The shown embodiments are only for illustrative purposes and do not limit the scope of the claims. In all the drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0016] Figure 1 A schematic diagram of an exemplary system in which the various methods described herein can be implemented according to an embodiment of the present disclosure is shown;

[0017] Figure 2 A flowchart of a method for component thermal simulation according to an embodiment of the present disclosure is shown;

[0018] Figure 3 A flowchart of a process in which a neural network generates thermal simulation results of multiple target positions according to an embodiment of the present disclosure is shown;

[0019] Figure 4 A flowchart of a method for training a neural network according to an embodiment of the present disclosure is shown;

[0020] Figure 5 A flowchart of a process in which a neural network generates multiple thermal simulation results of multiple target positions according to an embodiment of the present disclosure is shown;

[0021] Figure 6 Shows a schematic diagram of a neural network for component thermal simulation according to an embodiment of the present disclosure;

[0022] Figure 7 Shows a schematic diagram of the test results of a neural network for component thermal simulation according to an embodiment of the present disclosure;

[0023] Figure 8 Shows a probability distribution histogram of RMSE values corresponding to test data and training data respectively according to an embodiment of the present disclosure;

[0024] Figure 9 Shows a structural block diagram of a component thermal simulation device according to an embodiment of the present disclosure;

[0025] Figure 10 Shows a structural block diagram of a training device for a neural network according to an embodiment of the present disclosure; and

[0026] Figure 11 Shows a structural block diagram of an exemplary electronic device capable of implementing the embodiments of the present disclosure. Detailed implementation manners

[0027] The following makes an explanation of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0028] In the present disclosure, unless otherwise specified, the terms "first", "second", etc. are used to describe various elements and do not intend to limit the positional relationship, timing relationship or importance relationship of these elements. Such terms are only used to distinguish one element from another. In some examples, the first element and the second element may refer to the same instance of the element, and in certain cases, based on the context description, they may also refer to different instances.

[0029] In the description of various examples in the present disclosure, the terms used are only for the purpose of describing specific examples and are not intended to be restrictive. Unless the context clearly indicates otherwise, if the number of elements is not specifically limited, the element may be one or more. In addition, the term "and / or" used in the present disclosure covers any one of the listed items and all possible combinations.

[0030] In the related art, a single neural network cannot perform thermal simulation for different configuration conditions.

[0031] To solve the above problems, the present disclosure inputs the target position in the target component, at least one target configuration condition determined among multiple preset configuration conditions, and configuration selection information indicating the selection of at least one target configuration condition among multiple preset configuration conditions into a neural network for component thermal simulation, enabling the neural network to learn the target configuration condition corresponding to the received configuration condition information, so as to perform thermal simulation on the target component according to the target configuration condition and obtain the thermal simulation result of the target position. Through the above method, it is realized to use a single neural network to process the thermal simulation of a certain position of the target component under different configuration conditions, so that it is not necessary to train or deploy multiple neural networks for different configuration conditions, reducing the training cost and deployment cost.

[0032] Embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0033] Figure 1 FIG. shows a schematic diagram of an exemplary system 100 in which the various methods and apparatuses described herein can be implemented according to an embodiment of the present disclosure. Referring to Figure 1 , the system 100 includes one or more client devices 101, 102, 103, 104, 105, and 106, a server 120, and one or more communication networks 110 that couple the one or more client devices to the server 120. The client devices 101, 102, 103, 104, 105, and 106 can be configured to execute one or more application programs.

[0034] In an embodiment of the present disclosure, the server 120 can run one or more services or software applications that enable the execution of the methods of the present disclosure. In certain embodiments, the server 120 can also provide other services or software applications, which can include non-virtual environments and virtual environments. In certain embodiments, these services can be provided as web-based services or cloud services, for example, provided to users of the client devices 101, 102, 103, 104, 105, and / or 106 under a software as a service (SaaS) model.

[0035] In Figure 1 the configuration shown, the server 120 can include one or more components that implement the functions performed by the server 120. These components can include software components, hardware components, or a combination thereof that can be executed by one or more processors. Users operating the client devices 101, 102, 103, 104, 105, and / or 106 can in turn use one or more client applications to interact with the server 120 to utilize the services provided by these components. It should be understood that various different system configurations are possible, which can be different from the system 100. Therefore, Figure 1is an example of a system for implementing the various methods described herein and is not intended to be limiting.

[0036] Users can use client devices 101, 102, 103, 104, 105, and / or 106 to perform human-computer interactions. The client devices can provide an interface that enables the users of the client devices to interact with the client devices. The client devices can also output information to the users via the interface. Although Figure 1 only six client devices are depicted, those skilled in the art will be able to understand that the present disclosure can support any number of client devices.

[0037] Client devices 101, 102, 103, 104, 105, and / or 106 can include various types of computer devices, such as portable handheld devices, general-purpose computers (such as personal computers and laptop computers), workstation computers, wearable devices, smart screen devices, self-service terminal devices, service robots, gaming systems, thin clients, various messaging devices, sensors, or other sensing devices, etc. These computer devices can run various types and versions of software applications and operating systems, such as MICROSOFT Windows, APPLE iOS, UNIX-like operating systems, Linux or Linux-like operating systems (such as GOOGLE Chrome OS); or include various mobile operating systems, such as MICROSOFT WindowsMobile OS, iOS, Windows Phone, Android. Portable handheld devices can include cellular phones, smartphones, tablets, personal digital assistants (PDAs), etc. Wearable devices can include head-mounted displays (such as smart glasses) and other devices. Gaming systems can include various handheld gaming devices, Internet-enabled gaming devices, etc. The client devices are capable of executing various different applications, such as various Internet-related applications, communication applications (such as email applications), short message service (SMS) applications, and can use various communication protocols.

[0038] Network 110 can be any type of network known to those skilled in the art, which can support data communication using any one of a variety of available protocols (including but not limited to TCP / IP, SNA, IPX, etc.). By way of example only, one or more networks 110 can be a local area network (LAN), an Ethernet-based network, token ring, wide area network (WAN), the Internet, virtual network, virtual private network (VPN), intranet, extranet, blockchain network, public switched telephone network (PSTN), infrared network, wireless network (such as Bluetooth, WIFI), and / or any combination of these and / or other networks.

[0039] Server 120 may include one or more general-purpose computers, dedicated server computers (such as PC (Personal Computer) servers, UNIX servers, midrange servers), blade servers, mainframes, server clusters, or any other suitable arrangement and / or combination. Server 120 may include one or more virtual machines running a virtual operating system, or other computing architectures involving virtualization (such as one or more flexible pools of logical storage devices that can be virtualized to maintain virtual storage devices of the server). In various embodiments, Server 120 may run one or more services or software applications that provide the functions described below.

[0040] The computing units in Server 120 may run one or more operating systems including any of the above-mentioned operating systems and any commercially available server operating systems. Server 120 may also run any one of a variety of additional server applications and / or middleware applications, including HTTP servers, FTP servers, CGI servers, JAVA servers, database servers, etc.

[0041] In some embodiments, Server 120 may include one or more applications to analyze and merge data feeds and / or event updates received from users of client devices 101, 102, 103, 104, 105, and / or 106. Server 120 may also include one or more applications to display data feeds and / or real-time events via one or more display devices of client devices 101, 102, 103, 104, 105, and / or 106.

[0042] In some embodiments, Server 120 may be a server of a distributed system, or a server incorporating a blockchain. Server 120 may also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology. A cloud server is a host product in the cloud computing service system, which solves the defects of high management difficulty and weak business scalability existing in traditional physical hosts and virtual private server (VPS, Virtual Private Server) services.

[0043] System 100 may also include one or more databases 130. In some embodiments, these databases can be used to store data and other information. For example, one or more of databases 130 can be used to store information such as audio files and video files. Databases 130 can reside in various locations. For example, the database used by server 120 can be local to server 120, or can be remote from server 120 and can communicate with server 120 via a network-based or dedicated connection. Databases 130 can be of different types. In some embodiments, the database used by server 120 can be, for example, a relational database. One or more of these databases can store, update, and retrieve data to and from the database in response to commands.

[0044] In some embodiments, one or more of databases 130 can also be used by an application to store application data. The database used by the application can be a different type of database, such as a key-value store, an object store, or a conventional store supported by a file system.

[0045] Figure 1 System 100 can be configured and operated in various ways to enable the application of the various methods and apparatuses described in this disclosure.

[0046] According to one aspect of the present disclosure, a method for component thermal simulation is provided. As Figure 2 shown, method 200 includes: step S201, determining a target position in a target component; step S202, determining at least one target configuration condition among a plurality of preset configuration conditions, and determining at least one configuration condition information corresponding to the at least one target configuration condition, wherein each of the plurality of preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes the thermal simulation information of at least one relevant position in the target component specified by the corresponding target configuration condition; step S203, determining configuration selection information, the configuration selection information indicating the selection of the at least one target configuration condition among the plurality of preset configuration conditions; and step S204, inputting the target position, the at least one configuration condition information, and the configuration selection information into a neural network for component thermal simulation to obtain a thermal simulation result of the target position.

[0047] Thus, by inputting the target position in the target component, at least one target configuration condition determined among multiple preset configuration conditions, and configuration selection information indicating the selection of the at least one target configuration condition among the multiple preset configuration conditions into a neural network for component thermal simulation, the neural network can learn the target configuration condition corresponding to the received configuration condition information, so that the thermal simulation of the target component can be performed according to the target configuration condition, and the thermal simulation result of the target position can be obtained. Through the above method, the thermal simulation of a certain position of the target component under different configuration conditions is realized by using a single neural network, so that it is not necessary to train or deploy multiple neural networks for different configuration conditions, reducing the training cost and deployment cost.

[0048] In some embodiments, the target component may be a chip, an integrated circuit, or other electronic components. The target component may be two-dimensional or three-dimensional; it may be square, rectangular, or have other geometric structures, which are not limited herein. In this disclosure, the chip will be mainly used as an exemplary embodiment of the target component to illustrate the solution, but it is not intended to limit the scope of this disclosure. The methods and devices of this disclosure can also be applied to other components, which are not limited herein.

[0049] In some embodiments, the target position of the target component determined in step S201 may be the position in the target component where the thermal simulation result is required. The thermal simulation of the target component may be, for example, the simulation of its temperature field, and the thermal simulation result of the target position may include, for example, the temperature field of the target position. The target position may be located, for example, on the boundary of the target component or inside the target component.

[0050] In an exemplary embodiment, the target component may be represented as a square geometry with a grid size of 20×20, which represents a 2D chip of 1mm×1mm. The target component has a total of 400 grid points, and each grid point corresponds to a position in the target component. The component has a network of 18×18 inside, including a total of 324 internal positions; the component also includes 76 boundary positions.

[0051] In some embodiments, a control equation can be represented in the following general format:

[0052]

[0053] Among them, N is the symbolic representation of the simplified control equation, and the temperature field is represented by s. Ω can represent all positions on the component, and x can represent one of the positions. The specific temperature field of the given target component is determined by various configuration conditions (e.g., internal heat source and boundary conditions). In the present disclosure, the configuration conditions can also be referred to as Partial Differential Equation (PDE) configurations. In some embodiments, it is assumed that there are m related PDE configurations, including parametric and non-parametric ones, i.e., u1, u2, …, u m . These m related PDE configurations form a family of PDE configurations, and then the temperature field of the target component is predicted at the corresponding position x, obtaining the final form expressed as s(u1, u2, …, u m )(x).

[0054] The neural network for component thermal simulation learns the function mapping G θ (θ represents all learnable parameters in the neural network, i.e., weights and biases), and this function mapping can be expressed as:

[0055]

[0056] where, U: U1×U2×…×U m represents the function space spanned by the above-mentioned family of PDE configurations, and S is the function space spanned by the temperature s(u1, u2, …, u m )(x). Such a mapping means that a trained neural network can accurately predict the temperature field distribution of the target component under any PDE configuration given and extracted from the function space U.

[0057] In some embodiments, different PDE configurations each specify at least one relevant position in the target component. For example, the heat source distribution configuration condition specifies multiple internal positions in the target component, and the boundary configuration condition specifies multiple boundary positions in the target component. And for some positions in the target component, there may be multiple PDE models. For example, the boundary configuration condition can include various boundary conditions such as Dirichlet boundary condition, Neumann boundary condition, convective boundary condition, radiative boundary condition, etc.

[0058] In some embodiments, in step S203, at least one target configuration condition among multiple preset configuration conditions is determined, and at least one configuration condition information corresponding to at least one target configuration condition is determined.

[0059] In some embodiments, the multiple preset configuration conditions may be that the neural network supports processing all PDE configurations. In other words, the neural network is trained using sample data corresponding to the multiple preset configuration conditions. At least one target configuration condition may be determined based on a user's input or selection, or may be determined from the multiple preset configuration conditions according to a preset rule, or determined by other means, which is not limited herein.

[0060] In some embodiments, the thermal simulation information represented by the configuration condition information may have different physical meanings according to different corresponding target configuration conditions. In an exemplary embodiment, the thermal simulation information of at least one relevant position specified by the heat source distribution configuration condition may be the heat source values at multiple internal positions of the target component; the thermal simulation information of at least one relevant position specified by the boundary configuration condition may be the boundary function values at multiple boundary positions of the target component. It can be understood that the configuration condition information and the thermal simulation information it represents may also have other forms and corresponding physical meanings, which is not limited herein.

[0061] In some embodiments, the multiple preset configuration conditions may include at least two preset configuration conditions that specify the same position in the target component, and at least one relevant position specified by each of the at least one target configuration condition may not overlap. In other words, when performing a thermal simulation on the target component, there are conflicts between multiple preset configuration conditions for the same position. When determining at least one target configuration condition for performing a thermal simulation on the target component, one of these conflicting preset configuration conditions needs to be selected; multiple non-conflicting preset configuration conditions can be used simultaneously.

[0062] In some embodiments, in step S203, configuration selection information is determined.

[0063] In order to enable the neural network for component thermal simulation of a given target component to output a corresponding thermal simulation result based on any input PDE configuration, configuration selection information may be output to the neural network to indicate at least one target configuration condition corresponding to the input at least one configuration condition information.

[0064] In some embodiments, index numbers may be set for the multiple preset configuration conditions, and the index numbers of at least one target configuration condition may be used as configuration selection information and input into the neural network. In addition to the above method, configuration selection information may also be determined by other means, which is not limited herein.

[0065] According to some embodiments, the multiple preset configuration conditions may include a heat source distribution configuration condition, and at least one target configuration condition selected from the multiple preset configuration conditions for performing component thermal simulation may include a heat source distribution configuration condition. The heat source distribution configuration condition may specify multiple internal positions of the target component, and the configuration condition information corresponding to the heat source distribution configuration condition may include heat source values at the multiple internal positions.

[0066] Thus, by using the heat source distribution configuration condition, the neural network can output more accurate thermal simulation results based on the heat source values at the multiple internal positions of the target component.

[0067] In some embodiments, the heat source values at the multiple internal positions in the target component can be obtained. For example, the heating elements and the heat conduction conditions at different positions in the target component can be determined according to the structural design of the target component, and the heat source value at each position in the component can be obtained accordingly.

[0068] According to some embodiments, the multiple preset configuration conditions may include a boundary configuration condition, and at least one target configuration condition selected from the multiple preset configuration conditions for performing component thermal simulation may include a boundary configuration condition. The boundary configuration condition may specify multiple boundary positions of the target component, and the configuration condition information corresponding to the boundary configuration condition may include boundary function values at the multiple boundary positions.

[0069] Thus, by using the boundary configuration condition, the neural network can output more accurate thermal simulation results based on the boundary function values at the multiple boundary positions of the target component.

[0070] In some embodiments, the boundary configuration condition may be a Dirichlet boundary condition, a Neumann boundary condition, a convective boundary condition, or a radiative boundary condition, or other boundary conditions. The boundary function values at the multiple boundary positions of the target component can be obtained according to the boundary function Q(x) corresponding to these boundary conditions.

[0071] For the Dirichlet boundary condition, Q(x) = q d (x), where q d is the temperature field; for the Neumann boundary condition, Q(x) = q n (x) / k, where q n is the temperature flux and k is the thermal conductivity; for the convective boundary condition, Q(x) = h / k, where h is the surface convective coefficient and k is the thermal conductivity; for the radiative boundary condition, Q(x) = ∈σ / k, where σ is the Stefan - Boltzmann coefficient and ∈ is the thermal radiation coefficient.

[0072] According to some embodiments, the multiple preset configuration conditions may include multiple preset boundary configuration conditions corresponding to multiple boundary conditions. In other words, the neural network can support thermal simulation under multiple boundary conditions. The multiple boundary conditions may include, for example, the above-mentioned Dirichlet boundary condition, Neumann boundary condition, convective boundary condition, and / or radiative boundary condition, and may also include other boundary conditions.

[0073] In some embodiments, at least one target configuration condition selected from the multiple preset configuration conditions for performing thermal simulation includes a boundary configuration condition. The boundary configuration condition may include a preset boundary configuration condition corresponding to one of the multiple boundary conditions, and the configuration condition information corresponding to the boundary configuration condition may include boundary function values of multiple boundary positions based on this boundary condition.

[0074] In some embodiments, the configuration selection information may indicate the selection of the preset boundary configuration condition corresponding to this boundary condition among the multiple preset boundary configuration conditions corresponding to the multiple boundary conditions. In other words, the configuration selection information may indicate to the neural network which configuration condition information among the at least one configuration condition information input into the neural network includes the configuration condition information corresponding to the preset boundary configuration condition corresponding to which boundary condition.

[0075] Thus, in the above manner, it is realized to use a single neural network to process thermal simulations under different boundary conditions.

[0076] In some embodiments, in step S204, the target position, at least one configuration condition information, and configuration selection information are input into the neural network for component thermal simulation to obtain the thermal simulation result of the target position. The neural network used in step S204 may be trained, for example, by using the method 400 to be described below.

[0077] According to some embodiments, the neural network may include a position sub-network, at least one configuration sub-network corresponding to at least one configuration condition information, and a selection sub-network corresponding to the configuration selection information. Figure 3 The flowchart of process 300 for the neural network to generate the thermal simulation result of the target position according to an embodiment of the present disclosure is shown. Process 300 may be used to implement step S204 in method 200. Process 300 may include: step S301, inputting the target position into the position sub-network to obtain a position feature vector; step S302, inputting at least one configuration condition information into at least one configuration sub-network respectively to obtain at least one configuration condition feature vector; step S303, inputting the configuration selection information into the selection sub-network to obtain a configuration selection feature vector; and step S304, obtaining the thermal simulation result based on the position feature vector, at least one configuration condition feature vector, and configuration selection feature vector.

[0078] Thus, by setting a position sub-network corresponding to a target position, at least one configuration sub-network corresponding to at least one configuration condition information, and a selection sub-network corresponding to configuration selection information, a position feature vector representing the position information of the target position, at least one configuration condition feature vector representing the at least one configuration condition information, and a configuration selection feature vector representing the selection of at least one target configuration condition can be obtained. Furthermore, an accurate thermal simulation result can be obtained based on these feature vectors.

[0079] According to some embodiments, the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector may have the same dimension. Step S304, obtaining a thermal simulation result based on the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector may include: obtaining an intermediate feature vector based on the Hadamard product of the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector; and summing the intermediate feature vectors to obtain a scalar output of the predicted temperature field of the target position.

[0080] Thus, in the above manner, different feature vectors output by multiple sub-networks of the neural network can be effectively combined, and an accurate temperature field prediction result can be obtained.

[0081] In some embodiments, multiple configuration sub-networks corresponding to multiple preset configuration conditions may be set in the neural network, and each configuration sub-network is used to process the configuration condition information corresponding to the corresponding preset configuration condition.

[0082] In some embodiments, a classification method may be introduced to reduce the number of configuration sub-networks, and thus reduce the amount of training data. Different preset configuration conditions may be divided into multiple non-overlapping subsets according to the input dimensions of the configuration sub-networks, and the input dimensions of all preset configuration conditions in the same subset are the same. As described above, each preset configuration condition specifies at least one relevant position in the target component. In some embodiments, the input dimension of the configuration sub-network is consistent with the number of relevant positions specified by the corresponding preset configuration condition, and different preset configuration conditions specifying the same number of relevant positions usually specify the same at least one relevant position. For example, among multiple preset configuration conditions, only the number of relevant positions specified by multiple preset boundary configuration conditions is the number of multiple boundary positions in the target component.

[0083] Therefore, the above setting method can enable the configuration condition information corresponding to conflicting preset configuration conditions to share a configuration sub-network, thereby realizing the reuse of the configuration sub-network and reducing the amount of training data. In other words, the configuration condition information corresponding to two or more preset configuration conditions specifying the same number of relevant positions shares the same configuration sub-network, or the configuration condition information corresponding to two or more preset configuration conditions specifying the same at least one relevant position shares the same configuration sub-network.

[0084] In an exemplary embodiment, the input dimension of the position sub-network can be 2, corresponding to the horizontal and vertical coordinates of the target position, and can include 6 fully connected layers, with each layer including 128 neurons. At least one configuration sub-network in the neural network includes a heat source distribution configuration sub-network and a boundary configuration sub-network. The input dimension of the heat source distribution configuration sub-network can be consistent with the number of internal positions of the target component, and the input dimension of the boundary configuration sub-network can be consistent with the number of boundary positions of the target component.

[0085] In an exemplary embodiment where the target component is represented as a square geometry with a grid size of 20×20, the input dimension of the heat source distribution configuration sub-network can be 324, and the input dimension of the boundary configuration sub-network can be 76. The input dimension of the selection sub-network can be consistent with the number of multiple boundary conditions. In an exemplary embodiment where the multiple boundary conditions include 4 boundary conditions: Dirichlet boundary condition, Neumann boundary condition, convective boundary condition, and radiative boundary condition, the input dimension of the selection sub-network can be 1, and the configuration selection information uses 0, 1, 2, and 3 to represent the Dirichlet boundary condition, Neumann boundary condition, convective boundary condition, and radiative boundary condition respectively.

[0086] The heat source distribution configuration sub-network, the boundary configuration sub-network, and the selection sub-network can adopt the same network structure, that is, they can include 9 fully connected layers, with each layer including 256 neurons. The output dimensions of the position sub-network, the heat source distribution configuration sub-network, the boundary configuration sub-network, and the selection sub-network can all be 128. The neural network can adopt the Swish activation function to obtain better results.

[0087] The neural network built using the above architecture parameters can generate accurate component thermal simulation results. In some embodiments, the above model architecture parameters can be adjusted, or the neural network can be built in other ways to realize obtaining the thermal simulation result of the target position based on the target position, at least one configuration condition information, and configuration selection information using the neural network, which is not limited herein.

[0088] According to another aspect of the present disclosure, a training method for a neural network is provided. As Figure 4As shown, method 400 includes: step S401, determining a plurality of sample positions in a target component; step S402, determining a plurality of preset configuration conditions and determining a plurality of configuration condition information corresponding to the plurality of preset configuration conditions, wherein each of the plurality of preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes the thermal simulation information of at least one relevant position in the target component specified by the corresponding preset configuration condition; step S403, determining a plurality of configuration selection information corresponding to each of the plurality of sample positions, the configuration selection information indicating the selection of at least one target configuration condition among the plurality of preset configuration conditions, and the plurality of configuration selection information covering the plurality of preset configuration conditions; step S404, inputting the plurality of sample positions, the plurality of configuration selection information, and at least one configuration condition information corresponding to at least one target configuration condition indicated by each of the plurality of configuration selection information into a neural network to obtain a plurality of thermal simulation results corresponding to each of the plurality of sample positions and the plurality of configuration selection information; and step S405, adjusting the parameters of the neural network based on the plurality of thermal simulation results corresponding to each of the plurality of sample positions and the plurality of configuration selection information to obtain a neural network for thermal simulation of the target component.

[0089] It can be understood that the operations of steps S401 - S404 in method 400 can refer to the descriptions of steps S201 - S204 in method 200 above.

[0090] Thus, through the above method, the trained neural network can process the thermal simulation of a certain position of the target component under different configuration conditions, and thus it is not necessary to train or deploy multiple neural networks for different configuration conditions, reducing the training cost and deployment cost.

[0091] In some embodiments, the plurality of sample positions determined in step S401 may, for example, cover all positions in the target component or may be partial positions in the target component. In an exemplary embodiment, for a target component that can be represented as a square geometry with a grid size of 20×20, the plurality of positions may include all 400 positions in the grid.

[0092] In some embodiments, the plurality of preset configuration conditions determined in step S402 may be the configuration conditions that the trained neural network is expected to be able to process. In the training phase, since it is necessary to train the neural network's processing ability for each preset configuration condition, it is necessary to determine the corresponding configuration condition information for each of the plurality of preset configuration conditions.

[0093] In some embodiments, in step S403, for each of the plurality of sample positions, a plurality of configuration selection information needs to be determined, and these configuration selection information cover the plurality of preset configuration conditions determined in step S402, so that the trained neural network can process different configuration selection information and corresponding preset configuration conditions, so as to generate a thermal simulation result that meets the preset configuration conditions corresponding to the received configuration selection information in the prediction stage.

[0094] According to some embodiments, at least one target configuration condition may include a heat source distribution configuration condition, and the heat source distribution configuration condition may specify a plurality of internal positions of the target component, and the configuration condition information corresponding to the heat source distribution configuration condition may include the heat source values of the plurality of internal positions.

[0095] According to some embodiments, at least one target configuration condition may include a boundary configuration condition, and the boundary configuration condition may specify a plurality of boundary positions of the target component, and the configuration condition information corresponding to the boundary configuration condition may include the boundary function values of the plurality of boundary positions.

[0096] In some embodiments, a random heat source distribution S(x) may be set inside the target component to obtain the heat source values of a plurality of internal positions of the target component, and a boundary function Q(s) may be set on the boundary to obtain the boundary function values of a plurality of boundary positions. In an exemplary embodiment, a two-dimensional correlated and scale-invariant Gaussian random field may be used to generate the random heat source distribution S(x) and the boundary function Q(x). The correlation can be explained by a scale-free spectrum, that is:

[0097]

[0098] where L is the number of coordinate points on the boundary, that is, the number of the plurality of boundary positions. The smoothness of the sampling function is determined by the length scale coefficient α. The larger the value of α, the smoother the obtained random heat source distribution S(x) and the boundary function Q(x). In an exemplary embodiment, α = 4 is adopted.

[0099] According to some embodiments, the plurality of preset configuration conditions may include a plurality of preset boundary configuration conditions corresponding to a plurality of boundary conditions, and the plurality of configuration selection information corresponds to the plurality of boundary conditions. For each of the plurality of configuration selection information, the boundary configuration condition included in the at least one target configuration condition indicated by the configuration selection information may include the preset boundary configuration condition corresponding to the boundary condition corresponding to the configuration selection information, and the configuration condition information corresponding to the boundary configuration condition may include the boundary function values of a plurality of boundary positions based on the corresponding boundary condition.

[0100] In some embodiments, at step S404, for one of the multiple sample positions and one of the multiple configuration selection information corresponding to the sample position, the sample position, the configuration selection information, and at least one configuration condition information corresponding to at least one target configuration condition indicated by the configuration selection information can be input into a neural network to obtain a thermal simulation result for the sample position and the configuration selection information.

[0101] According to some embodiments, the neural network may include a position sub-network, at least one configuration sub-network corresponding to at least one configuration condition information, and a selection sub-network corresponding to the configuration selection information. Figure 5 The flowchart of process 500 for the neural network to generate thermal simulation results for multiple target positions according to an embodiment of the present disclosure is shown. Process 500 can be used to implement step S404 in method 400. Process 500 may include: step S501, for one of the multiple sample positions and one of the multiple configuration selection information corresponding to the sample position, input the sample position into the position sub-network to obtain a position feature vector; step S502, input at least one configuration condition information corresponding to at least one target configuration condition indicated by the configuration selection information into at least one configuration sub-network respectively to obtain at least one configuration condition feature vector; step S503, input the configuration selection information into the selection sub-network to obtain a configuration selection feature vector; and step S504, based on the position feature vector, at least one configuration condition feature vector, and the configuration selection feature vector, obtain a thermal simulation result for the sample position and the configuration selection information.

[0102] Thus, by setting a position sub-network corresponding to the target position, at least one configuration sub-network corresponding to at least one configuration condition information, and a selection sub-network corresponding to the configuration selection information, and training in the above manner, the trained neural network can obtain a position feature vector that accurately represents the position information of the target position, at least one configuration condition feature vector that accurately represents at least one configuration condition information, and a configuration selection feature vector that accurately represents the selection of at least one target configuration condition. Furthermore, an accurate thermal simulation result can be obtained based on these feature vectors.

[0103] According to some embodiments, the position feature vector, at least one configuration condition feature vector, and the configuration selection feature vector may have the same dimension. Step S504, obtaining a thermal simulation result for the sample position and the configuration selection information based on the position feature vector, at least one configuration condition feature vector, and the configuration selection feature vector may include: obtaining an intermediate feature vector based on the Hadamard product of the position feature vector, at least one configuration condition feature vector, and the configuration selection feature vector; and summing the intermediate feature vectors to obtain a scalar output of the predicted temperature field for the sample position and the configuration selection information.

[0104] Thus, in the above manner, it is possible to effectively combine different feature vectors output by multiple sub-networks of the neural network, and enable the trained neural network to output accurate temperature field prediction results.

[0105] In some embodiments, multiple configuration sub-networks corresponding to multiple preset configuration conditions may be set in the neural network, and each configuration sub-network is used to process the configuration condition information corresponding to the corresponding preset configuration condition.

[0106] In some embodiments, a classification method may be introduced to reduce the number of configuration sub-networks, thereby reducing the amount of training data. Different preset configuration conditions may be divided into multiple non-overlapping subsets according to the input dimension of the configuration sub-networks, and the input dimensions of all preset configuration conditions in the same subset are the same. As described above, each preset configuration condition specifies at least one relevant position in the target component. In some embodiments, the input dimension of the configuration sub-network is consistent with the number of relevant positions specified by the corresponding preset configuration condition, and different preset configuration conditions specifying the same number of relevant positions usually specify the same at least one relevant position. For example, among multiple preset configuration conditions, only the number of relevant positions specified by multiple preset boundary configuration conditions is the number of multiple boundary positions in the target component.

[0107] Therefore, the above setting method can enable the configuration condition information corresponding to conflicting preset configuration conditions to share a configuration sub-network, thereby realizing the reuse of the configuration sub-network and reducing the amount of training data. In other words, the configuration condition information corresponding to two or more preset configuration conditions specifying the same number of relevant positions shares the same configuration sub-network, or the configuration condition information corresponding to two or more preset configuration conditions specifying the same at least one relevant position shares the same configuration sub-network.

[0108] In an exemplary embodiment, the input dimension of the position sub-network can be 2, corresponding to the horizontal and vertical coordinates of the target position, and it can include 6 fully connected layers, with each layer including 128 neurons. At least one configuration sub-network in the neural network includes a heat source distribution configuration sub-network and a boundary configuration sub-network. The input dimension of the heat source distribution configuration sub-network can be consistent with the number of internal positions of the target component, and the input dimension of the boundary configuration sub-network can be consistent with the number of boundary positions of the target component.

[0109] In an exemplary embodiment where the target component is represented as a square geometry with a grid size of 20×20, the input dimension of the heat source distribution configuration sub-network can be 324, and the input dimension of the boundary configuration sub-network can be 76. The input dimension of the selection sub-network can be consistent with the number of multiple boundary conditions. In an exemplary embodiment where the multiple boundary conditions include 4 boundary conditions: Dirichlet boundary condition, Neumann boundary condition, convective boundary condition, and radiative boundary condition, the input dimension of the selection sub-network can be 1, and the configuration selection information uses 0, 1, 2, and 3 to represent the Dirichlet boundary condition, Neumann boundary condition, convective boundary condition, and radiative boundary condition respectively.

[0110] The heat source distribution configuration sub-network, the boundary configuration sub-network, and the selection sub-network can adopt the same network structure, that is, they can include 9 fully connected layers, with each layer including 256 neurons. The output dimensions of the position sub-network, the heat source distribution configuration sub-network, the boundary configuration sub-network, and the selection sub-network can all be 128. The neural network can adopt the Swish activation function to obtain better results.

[0111] The neural network built using the above architecture parameters can generate accurate thermal simulation results of the component. In some embodiments, the above model architecture parameters can be adjusted, or the neural network can be built in other ways to achieve obtaining the thermal simulation result of the target position based on the target position, at least one configuration condition information, and configuration selection information using the neural network, which is not limited herein.

[0112] In some embodiments, in step S405, the loss value can be determined based on the multiple thermal simulation results corresponding to the multiple sample positions and the multiple configuration selection information, and the parameters of the neural network can be adjusted based on the loss value to obtain the neural network for thermal simulation of the target component. In an exemplary embodiment, the loss value can be determined based on the multiple thermal simulation results of the multiple sample positions and the ground truth of the temperature field of the multiple sample positions. In addition to the above method, the loss function can also be designed according to requirements to be used to determine the loss value and adjust the parameters of the neural network in step S405.

[0113] According to some embodiments, step S405, adjusting the parameters of the neural network based on multiple thermal simulation results corresponding to multiple sample positions and multiple configuration selection information to obtain a neural network for thermal simulation of a target component may include: obtaining a global loss based on multiple thermal simulation results of multiple internal positions, the heat source values of multiple internal positions, and the static temperature field control equation; and adjusting the parameters of the neural network based on the global loss.

[0114] In some embodiments, it is difficult or even impossible to obtain the true value of the temperature field at the internal positions of the target component. By the above method, it is possible to establish supervision over the thermal simulation results of the target component without knowing the temperature field at the internal positions of the target component, thereby realizing the constraint on the training process of the neural network.

[0115] In some embodiments, the static temperature field control equation can be expressed as:

[0116] kΔT(x)+S(x)=0, for Ω,

[0117] where k represents the material property of the target component, Ω represents multiple internal positions, ΔT(x) represents one of the multiple thermal simulation results at position x, and S(x) represents the heat source value at position x.

[0118] According to some embodiments, step S405, adjusting the parameters of the neural network based on multiple thermal simulation results corresponding to multiple sample positions and multiple configuration selection information to obtain a neural network for thermal simulation of a target component may include: for each of multiple boundary conditions, obtaining a loss corresponding to the boundary condition based on the thermal simulation results corresponding to the configuration selection information corresponding to multiple boundary positions and the physical constraint corresponding to the boundary condition; and adjusting the parameters of the neural network based on the losses corresponding to multiple boundary conditions.

[0119] Thus, by the above method, it is possible to establish supervision over the thermal simulation results at the boundary positions of the target component, thereby realizing the constraint on the training process of the neural network.

[0120] In some embodiments, for the Dirichlet boundary condition, the physical constraint can be expressed as:

[0121] T=q d

[0122] The meaning of which is that the temperature field on the surface is fixed at q d .

[0123] For the Neumann boundary condition, the physical constraint can be expressed as:

[0124]

[0125] where k is the thermal conductivity, and the above equation means that the temperature flux on the surface is fixed at q n .

[0126] For the convective boundary condition, the physical constraint can be expressed as:

[0127]

[0128] where h is the surface convection coefficient, and T amb is the ambient temperature of the surface. The above equation means the balance between heat conduction and convection in the same direction on the surface.

[0129] For the radiative boundary condition, the physical constraint can be expressed as:

[0130]

[0131] where σ is the Stefan-Boltzmann coefficient and ∈ is the thermal radiation coefficient. The above equation means the electromagnetic wave radiation generated by the temperature difference on the surface.

[0132] In some embodiments, the static temperature field control equation and / or physical constraint can also be used to construct the loss function. In an exemplary embodiment, the root mean squared error (RMSE) can be used as the loss function. Among them, the predicted value uses one end term of the equation of the static temperature field control equation and / or physical constraint (or the rewritten static temperature field control equation and / or physical constraint), and the true value uses the other end term of the equation. The RMSE formula can be written as:

[0133]

[0134] In an exemplary embodiment, the neural network can be trained iteratively 20,000 times to ensure convergence. At the same time, 500 groups of data are sampled from the given Gaussian random field as the input functions of the heat source distribution S(x) and the boundary function Q(x), as well as the configuration condition information corresponding to 4 groups of preset boundary configuration conditions, and input them into the corresponding configuration sub-networks. 400 sample positions in the entire region Ω of the target component are input into the position sub-network. For the neural network to be trained, there are inputs of 400000000×1 (configuration selection information), 400000000×324 (heat source values at multiple internal positions), 400000000×76 (boundary function values at multiple boundary positions), and 400000000×2 (coordinates of sample positions). The initial learning rate can be determined as 1e-3.

[0135] Figure 6A schematic diagram of a neural network for component thermal simulation according to an embodiment of the present disclosure is shown. The neural network 600 may include a selection sub-network 612, configuration sub-networks 614 and 616 or more configuration sub-networks, and a position sub-network 618. The input of the selection sub-network 612 is configuration selection information 602. The input of the configuration sub-network 614 is configuration condition information 604 corresponding to the heat source distribution configuration condition. The input of the configuration sub-network 616 is configuration condition information 606 corresponding to the boundary configuration condition. In addition to the above configuration condition information, a configuration sub-network or more sub-networks corresponding to configuration condition information 608 may also be set. The input of the position sub-network 618 is the target position 610. After calculating the Hadamard product 620 and summing 622 for the feature vectors respectively output by the sub-network 612, the configuration sub-networks 614 and 616 or more configuration sub-networks, and the position sub-network 618, thermal simulation information 624 can be obtained.

[0136] In the training stage, the loss 634 can be calculated by using configuration selection constraints 626 corresponding to the configuration selection information, configuration condition constraints 628 corresponding to the heat source distribution configuration condition, configuration condition constraints 630 corresponding to the boundary configuration condition, the static temperature field control equation 632 or more constraints, and the neural network 600 can be tuned 636 based on the loss 634 to obtain a trained neural network.

[0137] Figure 7 A schematic diagram of the test results of a neural network for component thermal simulation according to an embodiment of the present disclosure is shown. Figure 7 The first row 710 shows the generation of three sets of random heat source distributions S(x) through a Gaussian random field. The second to sixth rows 720 - 760 respectively show the predicted results of the temperature field distributions corresponding to five types of boundary conditions. During the test, let k = 100, h = 100, T amb = 1, ∈σ = 5.6×10 -7 . From Figure 7 it can be seen that although there are significant differences in the random heat source distribution S(x) and the boundary conditions among the test samples, the neural network can correctly predict the two-dimensional diffusion property solutions on the interior and boundary controlled by the heat conduction equation.

[0138] 2500 sets of random heat source distributions S(x) are randomly generated through a Gaussian random field, and then combined with four boundary types to finally form 10,000 sets of test data, which are tested using a neural network. After obtaining 10,000 sets of temperature field distributions, the RMSE values on the boundary and the RMSE values of the temperature field distributions are calculated according to the RMSE formula. Figure 8 A probability distribution histogram of the RMSE values corresponding to the test data and the training data according to an embodiment of the present disclosure is shown. From Figure 8As can be seen, most of the RMSE values corresponding to the test data and the training data are relatively small, which also confirms the quantitative correctness of the prediction results generated by the neural network for component thermal simulation proposed in the present disclosure.

[0139] According to another aspect of the present disclosure, there is provided a component thermal simulation device. As Figure 9 shown, the device 900 includes: a first determination unit 910 configured to determine a target position in a target component; a second determination unit 920 configured to determine at least one target configuration condition among a plurality of preset configuration conditions and determine at least one configuration condition information corresponding to the at least one target configuration condition, wherein each of the plurality of preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes thermal simulation information of at least one relevant position specified by the corresponding target configuration condition in the target component; a third determination unit 930 configured to determine configuration selection information, the configuration selection information indicating a selection of the at least one target configuration condition among the plurality of preset configuration conditions; and a first simulation unit 940 configured to input the target position, the at least one configuration condition information, and the configuration selection information into a neural network for component thermal simulation to obtain a thermal simulation result of the target position.

[0140] It can be understood that the operations of units 910 - 940 in the device 900 can refer to the descriptions of the operations of steps S201 - S204 in the method 200 above, and will not be elaborated here.

[0141] According to some embodiments, the at least one target configuration condition may include a heat source distribution configuration condition, the heat source distribution configuration condition may specify a plurality of internal positions of the target component, and the configuration condition information corresponding to the heat source distribution configuration condition may include heat source values of the plurality of internal positions.

[0142] According to some embodiments, the at least one target configuration condition may include a boundary configuration condition, the boundary configuration condition may specify a plurality of boundary positions of the target component, and the configuration condition information corresponding to the boundary configuration condition may include boundary function values of the plurality of boundary positions.

[0143] According to some embodiments, the plurality of preset configuration conditions may include a plurality of preset boundary configuration conditions corresponding to a plurality of boundary conditions, the boundary configuration condition may include a preset boundary configuration condition corresponding to one of the plurality of boundary conditions, and the configuration condition information corresponding to the boundary configuration condition may include boundary function values of the plurality of boundary positions based on this boundary condition. The configuration selection information may indicate a selection of the preset boundary configuration condition corresponding to this boundary condition among the plurality of preset boundary configuration conditions corresponding to the plurality of boundary conditions.

[0144] According to some embodiments, the neural network may include a position sub-network, at least one configuration sub-network corresponding to the at least one configuration condition information, and a selection sub-network corresponding to the configuration selection information. The first simulation unit may include: a first position sub-unit configured to input the target position into the position sub-network to obtain a position feature vector; a first configuration sub-unit configured to input the at least one configuration condition information into the at least one configuration sub-network respectively to obtain at least one configuration condition feature vector; a first selection sub-unit configured to input the configuration selection information into the selection sub-network to obtain a configuration selection feature vector; and a first simulation sub-unit configured to obtain the thermal simulation result based on the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector.

[0145] According to some embodiments, the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector may have the same dimension. The simulation sub-unit may include: a first calculation sub-unit configured to obtain an intermediate feature vector based on the Hadamard product of the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector; and a second calculation sub-unit configured to sum the intermediate feature vector to obtain a scalar output of the predicted temperature field of the target position.

[0146] According to another aspect of the present disclosure, there is provided a training device for a neural network. As Figure 10 shown, the device 1000 includes: a fourth determination unit 1010 configured to determine a plurality of sample positions in the target component; a fifth determination unit 1020 configured to determine a plurality of preset configuration conditions and determine a plurality of configuration condition information corresponding to the plurality of preset configuration conditions, wherein each of the plurality of preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes the thermal simulation information of at least one relevant position in the target component specified by the corresponding preset configuration condition; a sixth determination unit 1030 configured to determine a plurality of configuration selection information corresponding to the plurality of sample positions respectively, the configuration selection information indicating the selection of at least one target configuration condition among the plurality of preset configuration conditions, and the plurality of configuration selection information covering the plurality of preset configuration conditions; a second simulation unit 1040 configured to input the plurality of sample positions, the plurality of configuration selection information, and at least one configuration condition information corresponding to at least one target configuration condition indicated by the plurality of configuration selection information into the neural network to obtain a plurality of thermal simulation results corresponding to the plurality of sample positions respectively and the plurality of configuration selection information; and a parameter adjustment unit 1050 configured to adjust the parameters of the neural network based on the plurality of thermal simulation results corresponding to the plurality of sample positions respectively and the plurality of configuration selection information to obtain a neural network for thermal simulation of the target component.

[0147] It can be understood that the operations of units 1010 - 1050 in apparatus 1000 can be referred to the descriptions of the operations of steps S401 - S405 in method 400 above, and will not be elaborated here.

[0148] According to some embodiments, the at least one target configuration condition may include a heat source distribution configuration condition, and the heat source distribution configuration condition specifies a plurality of internal positions of the target element, and the configuration condition information corresponding to the heat source distribution configuration condition may include heat source values at the plurality of internal positions.

[0149] According to some embodiments, the at least one target configuration condition may include a boundary configuration condition, and the boundary configuration condition specifies a plurality of boundary positions of the target element, and the configuration condition information corresponding to the boundary configuration condition may include boundary function values at the plurality of boundary positions.

[0150] According to some embodiments, the plurality of preset configuration conditions may include a plurality of preset boundary configuration conditions corresponding to a plurality of boundary conditions, and the plurality of configuration selection information may correspond to the plurality of boundary conditions. For each piece of configuration selection information in the plurality of configuration selection information, the boundary configuration condition included in the at least one target configuration condition indicated by the configuration selection information may include a preset boundary configuration condition corresponding to the boundary condition corresponding to the configuration selection information, and the configuration condition information corresponding to the boundary configuration condition may include boundary function values of the plurality of boundary positions based on the corresponding boundary condition.

[0151] According to some embodiments, the neural network may include a position sub - network, at least one configuration sub - network corresponding to the at least one configuration condition information, and a selection sub - network corresponding to the configuration selection information. The second simulation unit may include: a second position sub - unit configured to input the sample position into the position sub - network for one of the plurality of sample positions and one of the plurality of configuration selection information corresponding to the sample position to obtain a position feature vector; a second configuration sub - unit configured to input the at least one configuration condition information corresponding to the at least one target configuration condition indicated by the configuration selection information into the at least one configuration sub - network respectively to obtain at least one configuration condition feature vector; a second selection sub - unit configured to input the configuration selection information into the selection sub - network to obtain a configuration selection feature vector; and a second simulation sub - unit configured to obtain a thermal simulation result for the sample position and the configuration selection information based on the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector.

[0152] According to some embodiments, the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector may have the same dimension. The second simulation subunit may include: a third calculation subunit configured to obtain an intermediate feature vector based on the Hadamard product of the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector; and a fourth calculation subunit configured to sum the intermediate feature vectors to obtain a scalar output of the predicted temperature field for the sample position and the configuration selection information.

[0153] According to some embodiments, the parameter tuning unit may include: a global loss calculation subunit configured to obtain a global loss based on the multiple thermal simulation results of the respective multiple internal positions, the heat source values of the multiple internal positions, and the static temperature field control equation; and a first parameter tuning subunit configured to adjust the parameters of the neural network based on the global loss.

[0154] According to some embodiments, the parameter tuning unit may include: a boundary loss calculation subunit configured to obtain, for each of the multiple boundary conditions, a loss corresponding to the boundary condition based on the thermal simulation results corresponding to the configuration selection information corresponding to the respective multiple boundary positions for the boundary condition and the physical constraints corresponding to the boundary condition; and a second parameter tuning subunit configured to adjust the parameters of the neural network based on the losses corresponding to the respective multiple boundary conditions.

[0155] In the technical solution of the present disclosure, the processing of the collection, storage, use, processing, transmission, provision, and disclosure of the user's personal information involved all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0156] According to an embodiment of the present disclosure, an electronic device, a readable storage medium, and a computer program product are further provided.

[0157] Reference Figure 11 , a block diagram of an electronic device 1100 that can be used as a server or a client of the present disclosure will now be described. It is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent various forms of digital electronic computer devices, such as, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0158] As Figure 11As shown, the electronic device 1100 includes a computing unit 1101, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1102 or a computer program loaded from a storage unit 1108 into a random access memory (RAM) 1103. In the RAM 1103, various programs and data required for the operation of the electronic device 1100 can also be stored. The computing unit 1101, the ROM 1102, and the RAM 1103 are connected to each other via a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0159] A plurality of components in the electronic device 1100 are connected to the I / O interface 1105, including: an input unit 1106, an output unit 1107, a storage unit 1108, and a communication unit 1109. The input unit 1106 can be any type of device capable of inputting information into the electronic device 1100. The input unit 1106 can receive input digital or character information and generate key signal inputs related to user settings and / or function controls of the electronic device, and can include, but is not limited to, a mouse, a keyboard, a touch screen, a trackpad, a trackball, a joystick, a microphone, and / or a remote control. The output unit 1107 can be any type of device capable of presenting information and can include, but is not limited to, a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 1108 can include, but is not limited to, a magnetic disk and an optical disk. The communication unit 1109 allows the electronic device 1100 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks, and can include, but is not limited to, a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset, such as a Bluetooth device, an 802.11 device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0160] The computing unit 1101 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1101 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1101 executes the various methods, processes, and / or operations described above. For example, in some embodiments, these methods, processes, and / or operations can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 1108. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 1100 via the ROM 1102 and / or the communication unit 1109. When the computer program is loaded into the RAM 1103 and executed by the computing unit 1101, one or more steps of the methods, processes, and / or operations described above can be executed. Alternatively, in other embodiments, the computing unit 1101 can be configured to execute these methods, processes, and / or operations in any other suitable manner (e.g., by means of firmware).

[0161] Various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGA), application specific integrated circuits (ASIC), application specific standard products (ASSP), system-on-a-chip systems (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special or general-purpose programmable processor, and can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0162] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0163] In the context of this disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0164] In order to provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, speech input, or tactile input).

[0165] The systems and techniques described herein can be implemented in a computing system that includes backend components (such as, for example, a data server), or a computing system that includes middleware components (such as, for example, an application server), or a computing system that includes frontend components (such as, for example, a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (such as, for example, a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0166] A computer system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The relationship between the client and the server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, and solves the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services ("Virtual Private Server", or simply "VPS" for short). The server may also be a server of a distributed system or a server combined with a blockchain.

[0167] It should be understood that various forms of the processes shown above can be used, steps can be reordered, added or deleted. For example, the steps described in the present disclosure can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, and no limitation is made herein.

[0168] Although the embodiments or examples of the present disclosure have been described with reference to the accompanying drawings, it should be understood that the above methods, systems and devices are merely exemplary embodiments or examples, and the scope of the present invention is not limited by these embodiments or examples, but is only limited by the authorized claims and their equivalent scope. Various elements in the embodiments or examples can be omitted or replaced by their equivalent elements. In addition, the steps can be executed in an order different from that described in the present disclosure. Further, various elements in the embodiments or examples can be combined in various ways. Importantly, with the evolution of technology, many of the elements described herein can be replaced by equivalent elements that emerge after the present disclosure.

Claims

1. A method for component thermal simulation, comprising: Determining a target position in a target component; Determining at least one target configuration condition among a plurality of preset configuration conditions, and determining at least one configuration condition information corresponding to the at least one target configuration condition, wherein each of the plurality of preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes thermal simulation information of at least one relevant position in the target component specified by the corresponding target configuration condition; Determining configuration selection information, the configuration selection information indicating the selection of the at least one target configuration condition among the plurality of preset configuration conditions; and Inputting the target position, the at least one configuration condition information, and the configuration selection information into a neural network for component thermal simulation to obtain a thermal simulation result of the target position, wherein the neural network includes a position sub-network, at least one configuration sub-network corresponding to the at least one configuration condition information, and a selection sub-network corresponding to the configuration selection information, wherein inputting the target position, the at least one configuration condition information, and the configuration selection information into a neural network for component thermal simulation to obtain a thermal simulation result of the target position includes: Inputting the target position into the position sub-network to obtain a position feature vector; Inputting the at least one configuration condition information into the at least one configuration sub-network respectively to obtain at least one configuration condition feature vector; Inputting the configuration selection information into the selection sub-network to obtain a configuration selection feature vector; and Obtaining the thermal simulation result based on the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector.

2. The method according to claim 1, wherein, The position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector have the same dimension, wherein obtaining the thermal simulation result based on the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector includes: Obtaining an intermediate feature vector based on the Hadamard product of the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector; and Summing the intermediate feature vector to obtain a scalar output of the predicted temperature field of the target position.

3. The method according to claim 1 or 2, wherein The at least one target configuration condition includes a heat source distribution configuration condition, the heat source distribution configuration condition specifying a plurality of internal positions of the target component, and the configuration condition information corresponding to the heat source distribution configuration condition includes heat source values of the plurality of internal positions.

4. The method according to claim 1 or 2, wherein The at least one target configuration condition includes a boundary configuration condition, the boundary configuration condition specifying a plurality of boundary positions of the target component, and the configuration condition information corresponding to the boundary configuration condition includes boundary function values of the plurality of boundary positions.

5. The method according to claim 4, wherein, The multiple preset configuration conditions include multiple preset boundary configuration conditions corresponding to multiple boundary conditions. The boundary configuration conditions include a preset boundary configuration condition corresponding to one of the multiple boundary conditions, and the configuration condition information corresponding to the boundary configuration condition includes boundary function values of the multiple boundary positions based on this boundary condition. Among them, the configuration selection information indicates the selection of the preset boundary configuration condition corresponding to this boundary condition among the multiple preset boundary configuration conditions corresponding to the multiple boundary conditions.

6. A training method for a neural network, comprising: Determine multiple sample positions in a target component; Determine multiple preset configuration conditions and determine multiple configuration condition information corresponding to the multiple preset configuration conditions. Among them, each of the multiple preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes the thermal simulation information of at least one relevant position specified by the corresponding preset configuration condition in the target component; Determine multiple configuration selection information corresponding to each of the multiple sample positions. The configuration selection information indicates the selection of at least one target configuration condition among the multiple preset configuration conditions, and the multiple configuration selection information covers the multiple preset configuration conditions; Input the multiple sample positions, the multiple configuration selection information, and at least one configuration condition information corresponding to at least one target configuration condition indicated by each of the multiple configuration selection information into the neural network to obtain multiple thermal simulation results corresponding to each of the multiple sample positions and the multiple configuration selection information; and Based on the multiple thermal simulation results corresponding to each of the multiple sample positions and the multiple configuration selection information, adjust the parameters of the neural network to obtain a neural network for thermal simulation of the target component. Among them, the neural network includes a position sub-network, at least one configuration sub-network corresponding to the at least one configuration condition information, and a selection sub-network corresponding to the configuration selection information. Among them, inputting the multiple sample positions, the multiple configuration selection information, and at least one configuration condition information corresponding to at least one target configuration condition indicated by each of the multiple configuration selection information into the neural network to obtain multiple thermal simulation results corresponding to each of the multiple sample positions and the multiple configuration selection information includes: For one of the multiple sample positions and one of the multiple configuration selection information corresponding to this sample position, Input this sample position into the position sub-network to obtain a position feature vector; Input at least one configuration condition information corresponding to at least one target configuration condition indicated by this configuration selection information into the at least one configuration sub-network respectively to obtain at least one configuration condition feature vector; Input this configuration selection information into the selection sub-network to obtain a configuration selection feature vector; and Based on the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector, obtain a thermal simulation result for this sample position and this configuration selection information.

7. The method according to claim 6, wherein The position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector have the same dimension. Wherein, obtaining a thermal simulation result for the sample position and the configuration selection information based on the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector includes: Obtaining an intermediate feature vector based on the Hadamard product of the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector; and Summing the intermediate feature vectors to obtain a scalar output of the predicted temperature field for the sample position and the configuration selection information.

8. The method according to claim 6 or 7, wherein The at least one target configuration condition includes a heat source distribution configuration condition, and the heat source distribution configuration condition specifies multiple internal positions of the target component. The configuration condition information corresponding to the heat source distribution configuration condition includes heat source values at the multiple internal positions.

9. The method according to claim 8, wherein, Adjusting the parameters of the neural network based on the multiple thermal simulation results corresponding to the multiple sample positions and the multiple configuration selection information to obtain a neural network for thermal simulation of the target component includes: Obtaining a global loss based on the multiple thermal simulation results corresponding to the multiple internal positions respectively, the heat source values at the multiple internal positions, and the static temperature field control equation; and Adjusting the parameters of the neural network based on the global loss.

10. The method according to claim 6 or 7, wherein The at least one target configuration condition includes a boundary configuration condition, and the boundary configuration condition specifies multiple boundary positions of the target component. The configuration condition information corresponding to the boundary configuration condition includes boundary function values at the multiple boundary positions.

11. The method according to claim 10, wherein, The multiple preset configuration conditions include multiple preset boundary configuration conditions corresponding to multiple boundary conditions, and the multiple configuration selection information corresponds to the multiple boundary conditions. For each configuration selection information among the multiple configuration selection information, the boundary configuration condition included in the at least one target configuration condition indicated by the configuration selection information includes a preset boundary configuration condition corresponding to the boundary condition corresponding to the configuration selection information, and the configuration condition information corresponding to the boundary configuration condition includes boundary function values of the multiple boundary positions based on the corresponding boundary condition.

12. The method according to claim 11, wherein, Adjusting the parameters of the neural network based on the multiple thermal simulation results corresponding to the multiple sample positions and the multiple configuration selection information to obtain a neural network for thermal simulation of the target component includes: For each boundary condition among the multiple boundary conditions, obtaining a loss corresponding to the boundary condition based on the thermal simulation results corresponding to the configuration selection information corresponding to the boundary condition for the multiple boundary positions respectively and the physical constraints corresponding to the boundary condition; and Adjusting the parameters of the neural network based on the losses corresponding to the multiple boundary conditions respectively.

13. A component thermal simulation device, comprising: A first determination unit configured to determine a target position in a target component; A second determination unit, configured to determine at least one target configuration condition among a plurality of preset configuration conditions, and determine at least one configuration condition information corresponding to the at least one target configuration condition, wherein each of the plurality of preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes thermal simulation information of at least one relevant position in the target component specified by the corresponding target configuration condition; A third determination unit, configured to determine configuration selection information, where the configuration selection information indicates a selection of the at least one target configuration condition among the plurality of preset configuration conditions; and A first simulation unit, configured to input the target position, the at least one configuration condition information, and the configuration selection information into a neural network for component thermal simulation to obtain a thermal simulation result of the target position, where the neural network includes a position sub-network, at least one configuration sub-network corresponding to the at least one configuration condition information, and a selection sub-network corresponding to the configuration selection information, wherein the first simulation unit includes: A first position sub-unit, configured to input the target position into the position sub-network to obtain a position feature vector; A first configuration sub-unit, configured to input the at least one configuration condition information into the at least one configuration sub-network respectively to obtain at least one configuration condition feature vector; A first selection sub-unit, configured to input the configuration selection information into the selection sub-network to obtain a configuration selection feature vector; and A first simulation sub-unit, configured to obtain the thermal simulation result based on the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector.

14. The apparatus according to claim 13, wherein, The position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector have the same dimension, wherein the simulation sub-unit includes: A first calculation sub-unit, configured to obtain an intermediate feature vector based on the Hadamard product of the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector; and A second calculation sub-unit, configured to sum the intermediate feature vector to obtain a scalar output of the predicted temperature field of the target position.

15. The device according to claim 13 or 14, wherein, The at least one target configuration condition includes a heat source distribution configuration condition, where the heat source distribution configuration condition specifies a plurality of internal positions of the target component, and the configuration condition information corresponding to the heat source distribution configuration condition includes heat source values of the plurality of internal positions.

16. The device according to claim 13 or 14, wherein The at least one target configuration condition includes a boundary configuration condition, where the boundary configuration condition specifies a plurality of boundary positions of the target component, and the configuration condition information corresponding to the boundary configuration condition includes boundary function values of the plurality of boundary positions.

17. The device according to claim 16, wherein, The plurality of preset configuration conditions include a plurality of preset boundary configuration conditions corresponding to a plurality of boundary conditions, the boundary configuration condition includes a preset boundary configuration condition corresponding to one of the plurality of boundary conditions, and the configuration condition information corresponding to the boundary configuration condition includes boundary function values of the plurality of boundary positions based on this boundary condition, Wherein, the configuration selection information indicates the selection of the preset boundary configuration condition corresponding to the boundary condition from among multiple preset boundary configuration conditions corresponding to multiple boundary conditions.

18. A training device for a neural network, comprising: A fourth determination unit configured to determine a plurality of sample positions in a target component; A fifth determination unit configured to determine a plurality of preset configuration conditions and determine a plurality of configuration condition information corresponding to the plurality of preset configuration conditions, wherein each of the plurality of preset configuration conditions specifies at least one relevant position in the target component, and the configuration condition information characterizes thermal simulation information of at least one relevant position specified by the corresponding preset configuration condition in the target component; A sixth determination unit configured to determine a plurality of configuration selection information respectively corresponding to the plurality of sample positions, the configuration selection information indicating the selection of at least one target configuration condition from among the plurality of preset configuration conditions, and the plurality of configuration selection information covering the plurality of preset configuration conditions; A second simulation unit configured to input the plurality of sample positions, the plurality of configuration selection information, and at least one configuration condition information corresponding to at least one target configuration condition indicated by the plurality of configuration selection information into the neural network to obtain a plurality of thermal simulation results respectively corresponding to the plurality of sample positions and the plurality of configuration selection information; and A parameter adjustment unit configured to adjust parameters of the neural network based on the plurality of thermal simulation results respectively corresponding to the plurality of sample positions and the plurality of configuration selection information to obtain a neural network for thermal simulation of the target component, wherein the neural network includes a position sub-network, at least one configuration sub-network corresponding to the at least one configuration condition information, and a selection sub-network corresponding to the configuration selection information, Wherein, the second simulation unit includes: A second position sub-unit configured to input, for one of the plurality of sample positions and one of the plurality of configuration selection information corresponding to the sample position, the sample position into the position sub-network to obtain a position feature vector; A second configuration sub-unit configured to input at least one configuration condition information corresponding to at least one target configuration condition indicated by the configuration selection information into the at least one configuration sub-network respectively to obtain at least one configuration condition feature vector; A second selection sub-unit configured to input the configuration selection information into the selection sub-network to obtain a configuration selection feature vector; and A second simulation sub-unit configured to obtain a thermal simulation result for the sample position and the configuration selection information based on the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector.

19. The device according to claim 18, wherein, The position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector have the same dimension. Wherein, the second simulation sub-unit includes: A third computing subunit, configured to obtain an intermediate feature vector based on a Hadamard product of the position feature vector, the at least one configuration condition feature vector, and the configuration selection feature vector; and A fourth computing subunit, configured to sum the intermediate feature vectors to obtain a scalar output of a predicted temperature field for the sample position and the configuration selection information.

20. The apparatus according to claim 18 or 19, wherein, The at least one target configuration condition includes a heat source distribution configuration condition that specifies a plurality of internal positions of the target component, and the configuration condition information corresponding to the heat source distribution configuration condition includes heat source values at the plurality of internal positions.

21. The apparatus according to claim 20, wherein, The parameter tuning unit includes: A global loss calculation subunit, configured to obtain a global loss based on a plurality of thermal simulation results of each of the plurality of internal positions, the heat source values at the plurality of internal positions, and a static temperature field control equation; and A first parameter tuning subunit, configured to adjust parameters of the neural network based on the global loss.

22. The apparatus according to claim 18 or 19, wherein The at least one target configuration condition includes a boundary configuration condition that specifies a plurality of boundary positions of the target component, and the configuration condition information corresponding to the boundary configuration condition includes boundary function values at the plurality of boundary positions.

23. The apparatus according to claim 22, wherein, The plurality of preset configuration conditions includes a plurality of preset boundary configuration conditions corresponding to a plurality of boundary conditions, and the plurality of configuration selection information corresponds to the plurality of boundary conditions. For each piece of configuration selection information in the plurality of configuration selection information, the boundary configuration condition included in the at least one target configuration condition indicated by the configuration selection information includes a preset boundary configuration condition corresponding to a boundary condition corresponding to the configuration selection information, and the configuration condition information corresponding to the boundary configuration condition includes boundary function values of the plurality of boundary positions based on the corresponding boundary condition.

24. The apparatus according to claim 23, wherein, The parameter tuning unit includes: A boundary loss calculation subunit, configured to obtain, for each boundary condition in the plurality of boundary conditions, a loss corresponding to the boundary condition based on thermal simulation results corresponding to the configuration selection information corresponding to the boundary condition of each of the plurality of boundary positions and physical constraints corresponding to the boundary condition; and A second parameter tuning subunit, configured to adjust parameters of the neural network based on the losses corresponding to the plurality of boundary conditions respectively.

25. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1-12.

26. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the method according to any one of claims 1-12.

27. A computer program product comprising a computer program, wherein, The computer program, when executed by a processor, implements the method according to any one of claims 1-12.

Images