An automatic deployment method for multidisciplinary distributed parallel optimization

Generate the executable program MDexe through model interpreter technology and realize the automatic deployment of multidisciplinary distributed optimization, solving the problems of cumbersome manual deployment and low resource utilization in traditional methods, and achieving rapid deployment of multidisciplinary distributed parallel optimization and Efficient calculation.

CN118860453BActive Publication Date: 2024-11-29CHENGDU AERONAUTIC POLYTECHNIC
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411368678.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-29
Publication Date
2024-11-29
Estimated Expiration
2044-09-29

AI Technical Summary

Technical Problem

It is difficult for the existing technology to realize the automatic deployment of multidisciplinary distributed optimization. The traditional method requires manual configuration of the simulator operation environment. The process is cumbersome and error-prone, the resource utilization rate is low, and the coordination between disciplines is difficult.

Method used

Model interpreter technology is used to interpret models for multidisciplinary engineering, generate executable program MDexe, and distribute it to multiple cloud nodes through a cloud distributed center to realize automated deployment and distributed parallel computing.

Benefits of technology

It realizes rapid deployment and efficient computing with multi-disciplinary distributed parallel optimization, solves the problems of cumbersomeness of manual deployment and low resource utilization, improves the degree of automation and resource utilization, and reduces computing costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118860453B_ABST
    Figure CN118860453B_ABST
Patent Text Reader

Abstract

The present invention discloses an automatic deployment method for multidisciplinary distributed parallel optimization, and relates to the technical field of multidisciplinary simulation optimization. The automatic deployment method for multidisciplinary distributed parallel optimization is used to solve the problems of difficult manual deployment, low efficiency, and cumbersome user operation of simulation software of various disciplines in the distributed parallel optimization process, so as to meet the design optimization application of large-scale engineering. The steps of the present invention are: the first step is to receive the multidisciplinary engineering input and output data packets provided by the third-party design platform; the second step is to analyze the data and generate an execution instruction set according to environmental conditions; the third step is to use the "model interpreter" technology to interpret the multidisciplinary engineering solution model, and generate an executable program for batch processing, referred to as MDexe; the fourth step is to use the cloud distributed center to send the MDexe corresponding to each instruction to multiple cloud nodes to complete the distributed deployment operation, and quickly iterate and calculate the data packet output output data of the first step.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of multidisciplinary simulation optimization, and in particular to an automatic deployment method for multidisciplinary distributed parallel optimization. Background Art

[0002] With the intelligent transformation of industrial design and production, more and more design and manufacturing industries need to use distributed optimization solutions, which are a strategy to effectively solve optimization problems through cooperation and coordination among multiple intelligent agents. It mainly targets large-scale and complex optimization problems that centralized algorithms are difficult to handle.

[0003] In actual multidisciplinary distributed optimization, users need to install and deploy various structural, thermal, electromagnetic and other multidisciplinary and multi-domain simulators to a large-scale computing cluster in the cloud in advance to realize distributed calls. Taking into account the differences in cloud node operating systems, the differences in environmental requirements of various simulators, and the differences and difficulties in the installation and deployment steps of various simulators, the deployment work greatly increases the difficulty of multidisciplinary distributed optimization. Due to limitations such as licenses, GUI support, and complex software installation, the traditional Docker-based deployment method cannot be applied to the field of multidisciplinary simulation optimization. At the same time, the traditional method requires manual configuration of the operating environment of each simulator, including the operating system, dependent libraries, etc. The process is cumbersome and error-prone, the resource utilization rate is low, and inter-disciplinary collaboration is difficult. In view of the shortcomings of the prior art, the present invention provides an automatic deployment method for multidisciplinary distributed parallel optimization to solve the above problems. Summary of the invention

[0004] In order to better apply multi-disciplinary distributed optimization software to industrial design and manufacturing and truly improve the efficiency of engineers' design and manufacturing, the present invention proposes an automatic deployment method for multi-disciplinary distributed parallel optimization. The method adopts the key technology "model interpreter" to perform model interpretation processing of various types of solver simulators on the projects to be optimized submitted by users and generate executable programs adapted to cloud computing nodes, thereby achieving rapid adaptation of multi-disciplinary parallel optimization software and completing one-click automated deployment tasks.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an automatic deployment method for multidisciplinary distributed parallel optimization, comprising the following steps:

[0006] Step S1, receiving data packets, communicating with the third-party design platform interface through HTTP protocol, and receiving input and output data packets of multidisciplinary engineering;

[0007] Step S2, analyzing the data packet content, using the pointer linked list method to generate sequence instructions for multiple design points, each instruction contains the simulator running project file, attached input data and specified output items, and the cloud distributed center allocates computing nodes according to the instruction type and system environment load requirements;

[0008] Step S3, using the model interpreter method to solve a small number of samples of the engineering model of each discipline and explain the solution calculation process of the model, so as to train a high-precision model that meets the parallel operation of a large number of samples and convert it into an executable running program so that it can be directly deployed and used in step S4;

[0009] Step S4, sends the high-precision executable running program MDexe corresponding to each discipline to multiple cloud nodes. The cloud nodes execute MDexe, quickly iterate and calculate the data packets generated by S2 and produce output data and return it to the third-party design platform, realizing multi-disciplinary distributed parallel computing and completing the automatic deployment operation of multi-disciplinary distributed parallel optimization.

[0010] Preferably, in step S1, the data package includes simulation engineering files of various disciplines and input and output data specified by the user.

[0011] Preferably, in step S1, the input and output data packets of the multidisciplinary engineering are composed of Json text and model files, wherein the Json text contains node connection relationship information and node configuration information, and the model files are model engineering files to be solved in each discipline. The file types vary according to the disciplines, for example, electromagnetic simulation is usually HFSS engineering files; circuit simulation is usually ADS engineering files.

[0012] Preferably, in step S2, the pointer linked list method refers to generating a composite node linked list structure according to the node connection relationship in the Json text provided in step S1, each node linked list structure contains an input node vector, an output node vector, and node configuration information, and then recursively generates a multidisciplinary solution instruction sequence based on the node linked list structure.

[0013] Preferably, in step S3, the model interpreter method refers to a general structure that uses a factory pattern of programming to provide an interface for producing MDexe, and different simulators will inherit and implement a specific MDexe production method.

[0014] Preferably, the MDexe production method is to monitor the transient data of a single run of a specific solver, usually the data of each node under the grid. The data type is different in the specific physical scenario, and the unknown partial differential equation matrix is ​​extracted and generated based on the assumed model of the specific solver type and the space-time derivative matrix of the transient data, referred to as the feature extraction description PDES, and the fuzzy matrix is ​​denoised using a short-term information neural network algorithm to complete high-precision refinement, and finally achieve an equation group matrix with the same accuracy as the simulator, and compile the grid partitioning tool and the equation group matrix to generate an executable running program MDexe, and MDexe submits the final single instruction set containing parameter configuration to the computing node to run and generate output data.

[0015] Preferably, the second-layer short-term neural network model trained in step S3 adopts the optimization process of the method customized in combination with the partial differential matrix data loss function. This method significantly improves the calculation accuracy. When the total loss value is less than 0.01, the training is launched, and the model is packaged and converted into an MDexe execution program using the pyinstaller library.

[0016] Preferably, in step S4, the cloud node refers to a computer network cluster connected by a switch at the physical level, and the control scheduling level is to send the task package consisting of the MDexe generated in step S3 and the instruction set generated in step S2 to the central server through the HTTP protocol, and the server then combines with a third-party load balancing tool to automatically forward the task package to the computer network cluster node for execution.

[0017] Preferably, in step S3, the model interpreter trains a two-layer neural network model through input data, wherein the first layer is a multi-layer perceptron MLP model, whose input is design parameters and output is spatiotemporal node data, and the second layer is a short-term neural network LSTM model, whose input is spatiotemporal node data and output is a specified performance index. The closed loop from input design parameters to output performance index is completed through the two-layer model, providing a high-precision data basis for engineering optimization.

[0018] Preferably, the multidisciplinary optimization platform software passes the Json data packet carrying the specific simulator version, project files, and input and output data packets to the automatic deployment method of the present invention. The automatic deployment method sends the instruction sequence to the load balancer, and uses the load balancer to allocate nodes to the automatic deployment method. The automatic deployment method communicates with the load balancer to obtain the computing nodes with idle resources, and sends the executable running program MDexe and input design parameters to the computing nodes. The computing nodes receive the instructions sent by the automatic deployment method, parse the path and specific input design parameters of the executable running program MDexe, call MDexe by constructing command line instructions, and pass the input design parameters to MDexe to execute the computing task, and finally send the generated output data back to the multidisciplinary optimization platform to achieve a complete closed loop of data.

[0019] The present invention discloses an automatic deployment method for multidisciplinary distributed parallel optimization, which has the following beneficial effects:

[0020] 1. The automatic deployment method of multidisciplinary distributed parallel optimization is used to realize a general deployment method of distributed parallel optimization of simulation software in various disciplines, and to perform rapid deployment calculations by interpreting models of large-scale CAE and EDA simulation software; it effectively solves the problems of difficult manual deployment, low efficiency, and cumbersome user operations of simulation software in various disciplines during distributed parallel optimization, thereby meeting the application of large-scale engineering practice; the model interpreter method of the present invention can quickly generate executable programs, which can be quickly deployed to various computing nodes for operation without relying on a specific operating system, and realizes a high-efficiency, high-precision, high-portability, and high-scalability distributed simulation automatic deployment function, providing a high-efficiency computing means with rapid iteration and rapid convergence for the field of multidisciplinary simulation optimization.

[0021] 2. The automatic deployment method of multidisciplinary distributed parallel optimization has a high degree of automation. From data packet reception to computing node execution, the entire process is highly automated, which reduces manual intervention and improves efficiency. It has strong flexibility and supports a variety of simulators and project file types. It can flexibly respond to different multidisciplinary optimization needs, resource optimization, intelligent allocation of computing resources, improved resource utilization, and reduced computing costs. It has high precision and realizes high-precision simulator replacement through model interpreter methods and neural network algorithms, ensuring the accuracy of optimization results.

[0022] 3. This automatic deployment method for multidisciplinary distributed parallel optimization has strong scalability, supports dynamic expansion of computing nodes, can easily cope with large-scale computing needs, is easy to maintain, simplifies the simulator update and maintenance process, reduces maintenance costs, has high security, and enhances the security and stability of the system through the management of the central server and load balancer. Closed-loop optimization realizes closed-loop optimization from input design parameters to output performance indicators through a two-layer neural network model, providing high-precision data basis for engineering optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 It is the overall flow chart of the present invention;

[0025] Figure 2 A diagram of components for implementing the present invention;

[0026] Figure 3 It is a flow chart of the steps of the present invention;

[0027] Figure 4 is a schematic diagram of the code of the present invention;

[0028] Figure 5 This is a data schematic diagram of the execution instruction of step S2 of the present invention;

[0029] Figure 6 This is a Json field illustration diagram exported by the multidisciplinary platform in step S1 of the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] The embodiment of the present application provides an automatic deployment method for multidisciplinary distributed parallel optimization, which solves the problem that in actual multidisciplinary distributed optimization, the deployment work greatly increases the difficulty of multidisciplinary distributed optimization. Due to limitations such as licenses, GUI support, and complex software installation, the traditional Docker-based deployment method cannot be applied to the field of multidisciplinary simulation optimization. At the same time, the traditional method requires manual configuration of the operating environment of each simulator, including the operating system, dependent libraries, etc. The process is cumbersome and error-prone, the resource utilization rate is low, and the problem of difficulty in inter-disciplinary collaboration.

[0032] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0033] The embodiment of the present invention discloses an automatic deployment method for multidisciplinary distributed parallel optimization, such as Figure 1 As shown, first, the multidisciplinary engineering input and output data packets provided by the third-party design platform are received; then the data is analyzed and an execution instruction set is generated according to environmental conditions; secondly, the "model interpreter" technology is used to interpret the solution model of the multidisciplinary engineering, and an executable program for batch processing, referred to as MDexe, is generated; finally, the cloud distributed center sends the MDexe corresponding to each instruction to multiple cloud nodes to complete the distributed deployment operation, and quickly iterates the data packet output output data of step S1.

[0034] like Figure 2 As shown, the present invention includes four major components: a third-party multidisciplinary optimization platform software, an automatic deployment method of the present invention, a load balancer, and a cloud node. The multidisciplinary optimization platform software passes a Json data packet carrying a specific simulator version, project files, and input and output data packets to the automatic deployment method of the present invention. The present invention connects other components and produces an executable running program MDexe for use by cloud nodes.

[0035] like Figure 3 The figure shows the step flow of the core technology "model interpreter" of the present invention. Generally speaking, the method trains a two-layer neural network model through input data: the first layer is an MLP (multi-layer perceptron) model, whose input is design parameters and output is spatiotemporal node data, and the second layer is an LSTM (short-term neural network) model, whose input is spatiotemporal node data and output is specified performance indicators; the two-layer model completes the closed loop from input design parameters to output performance indicators, providing high-precision data basis for engineering optimization.

[0036] like Figure 4The figure shows the core code schematic diagram of the present invention, which uses Python language to perform two-layer model training, including key steps such as fuzzy matrix generation function, noise processing, definition of LSTM model, definition of MLP model, definition of loss function, optimization algorithm, and packaging model;

[0037] according to Figure 4 From the comments in the code, we can see that the model interpreter is mainly divided into five steps: data preparation, fuzzy processing, partial differential calculation, neural network training, and iterative optimization of loss function. The actual test case can be fully reproduced by combining the following data table;

[0038]

[0039] The above table gives the sample data of these steps, where the letters in the shape column represent the dimension information of the matrix, (N,M) represents an N-row and M-column matrix, N represents the number of samples, and M represents the dimension information of the variable; (1,M) represents an M-dimensional column vector; (N,O) represents the O-dimensional output data corresponding to N samples; (N,O,M) represents the M partial differential values ​​of the O-dimensional output corresponding to each sample under N samples;

[0040] By adopting the above-mentioned automatic deployment method, the automatic deployment method of multidisciplinary distributed parallel optimization has a high degree of automation. From data packet reception to computing node execution, the entire process is highly automated, which reduces manual intervention and improves efficiency. It has strong flexibility and supports a variety of simulators and project file types. It can flexibly respond to different multidisciplinary optimization needs, resource optimization, intelligent allocation of computing resources, improve resource utilization, reduce computing costs, and high precision. Through the model interpreter method and neural network algorithm, high-precision simulator replacement is achieved to ensure the accuracy of the optimization results.

[0041] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention is further described below in conjunction with the accompanying drawings and ADS simulation examples:

[0042] Step 1: Communicate with the third-party design platform interface through the http protocol and receive the input and output data packets of multidisciplinary projects. The data content includes simulation project files of various disciplines and user-specified input and output data Json text (Json fields are specified by the third-party design platform). Take ADS discipline as an example. For specific content, refer to Figure 6 Json data diagram;

[0043] Step 2: Analyze the data packet content and use the pointer linked list method to generate sequence instructions for multiple design points. Each instruction contains a simulator running project file and its associated input data and specified output items. Its format is as follows: Figure 5As shown, multiple instruction sets have the complete information required for the operation of multidisciplinary optimization projects. The instruction sets will be allocated to specific computing nodes by the cloud distributed center according to the instruction type and the system environment load requirements. Before batch processing, the executable program of each instruction is generated and sent to the cloud computing node by step S3. Therefore, the work of the cloud distributed center is simplified to the iteration of intelligent algorithms and the distribution of load balancing. The executable program of each design point is provided by the "model interpreter" method of step S3, thereby completing the entire process of automatically deploying computing tasks.

[0044] Step 3: Generate executable running programs MDexe one by one according to the execution instruction set generated in step S2. Taking a single instruction set of an ADS project as an example, the instruction set format is as follows: Figure 5 As shown, it includes the simulator version number, project file name, input design parameters, output result name, and parameter information on which each result depends; the input design parameters refer to the optimizable variables of the circuit model in the simulation, such as inductance value, resistance value, microstrip line size, etc., the output result name refers to the optimizable targets such as noise figure, gain, 1dB compression point, etc., and the parameter information on which each result depends refers to the frequency value parameter on which the noise figure depends; the automatic deployment method generates a small number of jitter sample sets that satisfy the orthogonal space according to the dimensions of the input design parameters. For example, 3 design parameters will generate a total of 7 sets of sample sets that satisfy the central difference;

[0045]

[0046] The above table is a schematic table of the input sample set in step S3 of the present invention;

[0047] Step 4: The automatic deployment method calls the ADS simulator to complete the 9 sample sets generated in step S3 and collect node spatiotemporal data and output result data. The spatiotemporal data set refers to the transient response information of the node voltage, current, etc. of each node in the circuit, and then Figure 4 The code shown calls the library function of tensorflow to train the first-layer MLP model, realizing the conversion function from input design parameters to node spatiotemporal data;

[0048] Step 5: Automatic deployment method generates the second-layer LSTM model based on the 7 sets of spatiotemporal data sets and output result data collected. The method is to first transform the node frequency domain data into time domain data through IFFT transformation, and then use Figure 4The fuzzy processing formula shown generates a fuzzy matrix, and then performs noise processing, the purpose of which is to improve the impact of nonlinear noise in the real simulation scene on the accuracy of the results and improve the robustness of the model; then define the loss function combined with the partial differential matrix data, and its formula represents the weighted sum of the output error and the partial differential error. Finally, iteratively optimize the loss function, and exit the training when the total loss value is less than 0.01, and use the pyinstaller library to package the model to generate an exe, which can be directly run on the cloud computing node, without the need to tediously install the ADS software and then perform the simulation, and the calculation time is greatly reduced, which can be effectively used in rapid optimization and rapid deployment environments.

[0049] In the loss function, let: x be the input parameter vector;

[0050] y is the output response vector;

[0051] is the output response predicted by the neural network;

[0052] J is the real partial differential matrix, representing the partial differential of y with respect to x ;

[0053] is the partial differential matrix predicted by the neural network, representing Partial derivative with respect to x ;

[0054] The loss function can be defined as the weighted sum of the output error and the partial differential error:

[0055]

[0056] in,

[0057] is the output error (mean square error);

[0058] is the partial differential error (Frobenius norm);

[0059] α and β are weight factors used to balance the contribution of the two errors.

[0060] Step 6. The automatic deployment method communicates with the load balancer to obtain the computing node with idle resources, and sends the executable running program MDexe and the input design parameters to the computing node. The computing node receives the instructions sent by the automatic deployment method, parses the path of the executable running program MDexe and the specific input design parameters, calls MDexe by constructing command line instructions, and passes the input design parameters to MDexe to execute the computing task, and finally sends the generated output data back to the multidisciplinary optimization platform to achieve a complete closed loop of data.

[0061] exist Figure 1 In the process, the multidisciplinary platform user sends optimization tasks and obtains Json data packets. The system receives and converts the generated link execution instructions to determine whether the next level is empty;

[0062] If it is empty, the corresponding MDexe is generated and added to the MDexe collection, and then the MDexe and execution instructions are fed back to the multidisciplinary platform;

[0063] If it is not empty, MDexe and execution instructions are directly fed back to the multidisciplinary platform;

[0064] Next, the multidisciplinary platform uses load balancing to distribute computing nodes;

[0065] Send a single-subject single-sample instruction and the corresponding MDexe to the computing node to perform calculations. Send a single-subject single-sample instruction and the corresponding MDexe to the computing node to perform calculations;

[0066] Finally, the platform collects the resulting data.

[0067] This automatic deployment method for multidisciplinary distributed parallel optimization has strong scalability, supports dynamic expansion of computing nodes, can easily cope with large-scale computing needs, is easy to maintain, simplifies the simulator update and maintenance process, reduces maintenance costs, has high security, and enhances the security and stability of the system through the management of the central server and load balancer. Closed-loop optimization realizes closed-loop optimization from input design parameters to output performance indicators through a two-layer neural network model, providing high-precision data basis for engineering optimization.

[0068] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0069] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An automatic deployment method for multidisciplinary distributed parallel optimization, characterized in that: The steps include: Step S1, receiving data packets, communicating with the third-party design platform interface through HTTP protocol, and receiving input and output data packets of multidisciplinary engineering; Step S2, analyzing the data packet content, using the pointer linked list method to generate sequence instructions for multiple design points, each instruction contains the simulator running project file, attached input data and specified output items, and the cloud distributed center allocates computing nodes according to the instruction type and system environment load requirements; Step S3, using the model interpreter method to solve a small number of samples of the engineering model of each discipline and explain the solution calculation process of the model, so as to train a high-precision model that meets the parallel operation of a large number of samples and convert it into an executable running program so that it can be directly deployed and used in step S4; Step S4, sending the high-precision executable running program MDexe corresponding to each discipline to multiple cloud nodes, the cloud nodes execute MDexe, quickly iterate and calculate the data packets generated by S2 and output the output data and return it to the third-party design platform, realizing multi-disciplinary distributed parallel computing and completing the automatic deployment operation of multi-disciplinary distributed parallel optimization; In step S1, the input and output data packets of the multidisciplinary engineering are composed of Json text and model files, wherein the Json text contains node connection relationship information and configuration information of each node, and the model file is the model engineering file to be solved in each discipline; In step S2, the pointer linked list method refers to generating a composite node linked list structure according to the node connection relationship in the Json text provided in step S1, each node linked list structure contains an input node vector, an output node vector, and node configuration information, and recursively generates a multi-disciplinary solution instruction sequence according to the node linked list structure; In step S3, the model interpreter method refers to a general structure that uses a factory pattern of programming, provides an interface for producing MDexe, and different simulators will inherit and implement specific MDexe production methods; In step S3, the MDexe production method in the factory mode is to monitor the transient data of each subject at the time and space nodes during the 9 calculation processes, train the first-layer multi-layer perceptron MLP model, and extract the partial differential equation matrix and its noise-processed fuzzy matrix according to the time and space derivative matrix corresponding to the data to train the second-layer short-term neural network model, and complete the high-precision model that meets any sample set through the double-layer neural network model.

2. The automatic deployment method for multidisciplinary distributed parallel optimization according to claim 1, characterized in that: In step S1, the data package includes simulation project files of various disciplines and input and output data specified by the user.

3. The automatic deployment method for multidisciplinary distributed parallel optimization according to claim 1, characterized in that: In step S4, the cloud node refers to a computer network cluster connected by a switch at the physical level. The control scheduling level sends the task package consisting of the MDexe generated in step S3 and the instruction set generated in step S2 to the central server through the HTTP protocol. The server then automatically forwards the task package to the computer network cluster node for execution in conjunction with a third-party load balancing tool.

4. The automatic deployment method for multidisciplinary distributed parallel optimization according to claim 1, characterized in that: The second-layer short-term neural network model trained in step S3 uses a custom loss function optimization process combined with partial differential matrix data. When the total loss value is less than 0.01, the training is terminated, and the pyinstaller library is used to package the model and convert it into an MDexe executable program.

5. The automatic deployment method for multidisciplinary distributed parallel optimization according to claim 1, characterized in that: In step S3, the model interpreter trains a two-layer neural network model through input data. The first layer is a multi-layer perceptron MLP model, whose input is the design parameters and output is spatiotemporal node data. The second layer is a short-term neural network LSTM model, whose input is spatiotemporal node data and output is the specified performance indicator.

6. The automatic deployment method for multidisciplinary distributed parallel optimization according to claim 1, characterized in that: The multidisciplinary optimization platform software passes the Json data packet carrying the specific simulator version, project files, input and output data packets to the automatic deployment method. The automatic deployment method sends the instruction sequence to the load balancer, and uses the load balancer to allocate nodes to the automatic deployment method. The automatic deployment method communicates with the load balancer to obtain the computing nodes with idle resources, and sends the executable running program MDexe and input design parameters to the computing nodes. The computing nodes receive the instructions sent by the automatic deployment method, parse the path and specific input design parameters of the executable running program MDexe, call MDexe by constructing command line instructions, and pass the input design parameters to MDexe to execute the computing task. Finally, the generated output data is sent back to the multidisciplinary optimization platform to achieve a complete closed loop of data.

Citation Information

Patent Citations

  • Method for packaging and integrating simulation models of distributed simulation platform of aircraft

    CN103412972A

  • Critical care cross-institution collaborative platform system based on cloud computing

    CN117238458A