Cloud simulation cost prediction method, device, electronic device and medium

Abstract

The present application discloses a cloud simulation cost prediction method, device, electronic device and medium. The method analyzes the multi-dimensional characteristics that affect the cloud simulation cost of a model to be simulated executed by a cloud simulation platform, obtains corresponding target characteristic values ​​through reasonable quantification processing, and directly inputs the trained decision tree regression model. From each of multiple decision trees, a candidate prediction value represented by an end node of a prediction path that matches these target characteristic values ​​is identified, and then a target prediction value is determined from multiple candidate prediction values, thereby realizing nonlinear prediction of the model to be simulated, comprehensively weighing various factors affecting the cloud simulation cost, and efficiently and reliably predicting the resource demand / consumption of cloud simulation of the model to be simulated under the corresponding calculation equations, material complexity, simulation calculation step size and model grid volume indicated by each target characteristic value, thereby accurately and reliably evaluating the cloud simulation cost of the model to be simulated and reliably avoiding potential financial losses.

Description

Technical Field

[0001] The present application relates to the field of model simulation applications, and in particular to a cloud simulation cost prediction method, device, electronic device and medium. Background Art

[0002] CAE (Computer-Aided Engineering) cloud simulation platform is a high-performance simulation platform based on cloud computing technology. It can provide powerful computing and storage capabilities to meet the needs of large-scale engineering simulation and analysis, thereby reducing product design iterations, shortening product development cycles, reducing costs, and improving the reliability and performance of developed products.

[0003] In practical applications, since the use of CAE cloud simulation platform is charged, it is usually necessary to rely on empirical parameters to predict the resource consumption / demand of the simulation task, and to evaluate the cloud simulation cost of the simulation task based on this to give a reasonable quotation.

[0004] However, the current cloud simulation cost prediction method that relies on empirical parameters often underestimates the cloud simulation cost due to its inability to accurately capture the resource requirements of complex simulation tasks, thus causing financial losses. Summary of the invention

[0005] In view of the above problems, this application provides the following solutions:

[0006] A first aspect of the present application provides a cloud simulation cost prediction method, the method comprising:

[0007] Acquire feature quantification information of a model to be simulated for a cloud simulation platform; the feature quantification information includes target feature values ​​of multi-dimensional features that affect the cloud simulation cost of the cloud simulation platform executing the model to be simulated, and the target feature values ​​of the multi-dimensional features respectively indicate a target calculation equation, material complexity, simulation calculation step, and model grid amount of the model to be simulated;

[0008] Input the target feature value of the multi-dimensional feature into a trained decision tree regression model, and determine the target prediction value for the model to be simulated from a plurality of candidate prediction values ​​identified to match the target feature value of the multi-dimensional feature; each of the candidate prediction values ​​is a prediction value represented by an end node of a prediction path that matches the target feature value of the multi-dimensional feature in a corresponding decision tree in the decision tree regression model;

[0009] The target prediction value is used to output a cloud simulation cost prediction result for the model to be simulated.

[0010] In a possible implementation, the step of obtaining the characteristic quantitative information of the model to be simulated on the cloud simulation platform includes:

[0011] Obtain a model to be simulated for a cloud simulation platform;

[0012] Obtaining a target equation identifier corresponding to a target calculation equation used by the model to be simulated; the target calculation equation identifier is determined as a target characteristic value of a first dimensional characteristic, and the target calculation equation includes at least one characteristic value of a continuous medium mechanics equation;

[0013] Determining the material complexity of the model to be simulated by using the material model complexity and the number of material properties of the model to be simulated; the material complexity is determined as the target feature value of the second dimensional feature;

[0014] Determine the inverse of the simulation calculation step configured for the to-be-simulated model as the target feature value of the third dimensional feature;

[0015] The model grid amount of the model to be simulated is scaled to obtain a target feature value of the fourth dimensional feature.

[0016] In a possible implementation, determining the material complexity of the model to be simulated by using the material model complexity and the number of material properties of the model to be simulated includes:

[0017] Determine the number of material properties corresponding to each material model included in the model to be simulated;

[0018] Determine the target material stress-strain relationship described by each of the material models, and the target behavior model to which it belongs; the target behavior model is used to describe the mechanical behavior of the corresponding material under the action of external force;

[0019] Querying the correspondence between different types of material stress-strain relationships and behavior models and different indicator identifiers, and obtaining a first indicator identifier corresponding to the target material stress-strain relationship and a second indicator identifier corresponding to the target behavior model;

[0020] The material complexity of the model to be simulated is obtained by using the material property quantity, the first index identifier and the second index identifier.

[0021] In one possible implementation, where:

[0022] Different types of material stress-strain relationships include linear relationships and nonlinear relationships, and the indicator identifier corresponding to the linear relationship is a first value, and the indicator identifier corresponding to the nonlinear relationship is a second value;

[0023] Different types of behavior models include elastic models and plastic models, the indicator identifier corresponding to the configuration of the plastic model is the first value, and the indicator identifier corresponding to the configuration of the elastic model is the second value;

[0024] Obtaining the material complexity of the model to be simulated by using the material attribute quantity, the first index identifier, and the second index identifier, including:

[0025] The material property quantity, the first index identifier and the second index identifier are summed to obtain the material complexity of the corresponding material model in the model to be simulated.

[0026] In a possible implementation, the step of obtaining a target equation identifier corresponding to a target calculation equation used by the model to be simulated includes:

[0027] Obtaining a computational complexity classification relationship between equation identifiers of different computational equations and different types of simulation models; the computational complexity classification relationship is determined by a machine learning model based on historical data of the simulation model executed by the cloud simulation platform;

[0028] The computational complexity classification relationship is queried to determine the target equation identifier corresponding to the model type to which the model to be simulated belongs.

[0029] In a possible implementation, the simulation calculation step size is configured based on the calculation accuracy requirement and calculation efficiency requirement of the model to be simulated;

[0030] Scaling the model grid amount of the to-be-simulated model to obtain the characteristic value of the fourth dimension includes:

[0031] Determine a reduction order of magnitude of the model grid amount for the model to be simulated; the reduction order of magnitude includes ten thousand levels;

[0032] The ratio of the model grid volume to the reduction order of magnitude is determined as the eigenvalue of the fourth dimension.

[0033] In a possible implementation, the training process of the decision tree regression model includes:

[0034] Acquire historical data of the simulation model executed by the cloud simulation platform;

[0035] Determine, based on the historical data, various historical feature values ​​corresponding to different dimensional features, as well as historical simulation conditions and historical prediction values ​​corresponding to different historical feature values;

[0036] Constructing a plurality of initial decision trees using the historical feature values ​​corresponding to the different dimensional features, the historical simulation conditions and the historical prediction values;

[0037] For each internal node of each of the initial decision trees, by evaluating the information gain of features of different dimensions, a one-dimensional feature is selected from the internal node as a split feature;

[0038] Obtaining the mean square error of two sub-nodes corresponding to each splitting point corresponding to the splitting feature;

[0039] Taking the splitting point corresponding to the minimum mean square error as the target splitting point, and performing the splitting feature segmentation on the corresponding internal node;

[0040] It is determined that each new decision tree meets a termination condition, and each new decision tree is determined as a decision tree regression model.

[0041] A second aspect of the present application provides a cloud simulation cost prediction device, the device comprising:

[0042] A feature quantification information acquisition module is used to acquire feature quantification information of a model to be simulated on a cloud simulation platform; the feature quantification information includes target feature values ​​of multi-dimensional features that affect the cloud simulation cost of the cloud simulation platform executing the model to be simulated, and the target feature values ​​of the multi-dimensional features respectively indicate a target calculation equation, material complexity, simulation calculation step size, and model grid amount of the model to be simulated;

[0043] A prediction module, used for inputting the target feature value of the multi-dimensional feature into a trained decision tree regression model, and determining the target prediction value for the model to be simulated from a plurality of candidate prediction values ​​identified to match the target feature value of the multi-dimensional feature; each of the candidate prediction values ​​is a prediction value represented by an end node of a prediction path matching the target feature value of the multi-dimensional feature in a corresponding decision tree in the decision tree regression model;

[0044] The output module is used to output the cloud simulation cost prediction result for the model to be simulated by using the target prediction value.

[0045] A third aspect of the present application provides an electronic device, the electronic device comprising at least one communication element, at least one memory and at least one processor;

[0046] The communication element is used to obtain the model to be simulated;

[0047] The memory is used to store computer instructions;

[0048] The processor is used to load and execute the computer instructions to implement the cloud simulation cost prediction method as described above.

[0049] A fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by an electronic device, the cloud simulation cost prediction method as described above is implemented.

[0050] It can be seen that in the cloud simulation cost prediction method provided in the present application, by analyzing the multi-dimensional characteristics that affect the cloud simulation cost of the cloud simulation platform executing the model to be simulated, the corresponding target feature values ​​are obtained through reasonable quantification processing to constitute the feature quantification information of the model to be simulated, and the target feature values ​​of these multiple dimensional features are directly input into the trained decision tree regression model. From each of the multiple decision trees included in the model, the candidate prediction value represented by the end node of a prediction path that matches these target feature values ​​is quickly and accurately identified, and then the target prediction value for the model to be simulated is determined from these multiple candidate decision values, that is, the nonlinear prediction of the model to be simulated is realized, and the various factors affecting the cloud simulation cost are comprehensively weighed, and the resource requirements / consumption of the cloud simulation of the model to be simulated under the corresponding calculation equations, material complexity, simulation calculation step size and model grid volume indicated by each target feature value are efficiently and reliably predicted, so as to accurately and reliably evaluate the cloud simulation cost of the model to be simulated using the target prediction value, ensure the reliability and accuracy of the cloud simulation quotation given based on this, and reliably avoid potential financial losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0052] Figure 1 A flowchart of the cloud simulation cost prediction method proposed in Example 1 of the present application;

[0053] Figure 2 This is a flow chart of the cloud simulation cost prediction method proposed in Example 2 of the present application;

[0054] Figure 3 This is a flow chart of the cloud simulation cost prediction method proposed in Example 3 of the present application;

[0055] Figure 4 This is a flow chart of the cloud simulation cost prediction method proposed in the fourth embodiment of the present application;

[0056] Figure 5 A schematic diagram of the initial decision tree splitting process in the cloud simulation cost prediction method proposed in the embodiment of the present application;

[0057] Figure 6 A schematic diagram of the structure of a cloud simulation cost prediction device proposed in an embodiment of the present application;

[0058] Figure 7 A schematic diagram of the hardware structure of an electronic device proposed in an embodiment of the present application. DETAILED DESCRIPTION

[0059] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation mode of the present application are only used to explain the specific embodiments of the present application and are not intended to limit the present application. The embodiments of the present application will be described below in conjunction with the drawings. It is known to those of ordinary skill in the art that with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0060] Moreover, the terms "first", "second" etc. in the full text description content and the accompanying drawings of the present application are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, and this is only to describe the distinction mode adopted by the objects of the same attributes when describing in the embodiments of the present application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0061] Regarding the description in the background technology section, when using the CAE (Computer-Aided Engineering) cloud simulation platform to perform simulation tasks of different models, the complexity of the entire simulation process will be affected by the differences in the complexity of different models (such as different physical models and geometric models, the difference in the amount of grids and the complexity of solving physical formulas, etc.), as well as factors such as the time step set for the corresponding simulation task, resulting in changes in the consumption of computing resources and storage resources of the CAE cloud simulation platform during the simulation process. As a result, the empirical prediction method that relies on the proportional coefficient, such as the weighted sum of various prediction factors, to predict the resource requirements of the CAE cloud simulation platform for different simulation tasks has poor prediction accuracy, which in turn leads to inaccurate assessment of cloud simulation costs and difficulty in avoiding financial losses.

[0062] In order to improve the above problems, this application proposes a cloud simulation cost prediction method, which reasonably quantifies multiple dimensional features of the cloud simulation cost that can affect the cloud simulation platform to perform model simulation tasks, obtains the corresponding quantified feature values, and directly inputs them into a high-performance decision tree regression model to automatically and accurately predict the resource requirements of the model simulation task for the CAE cloud simulation platform, so as to reliably and accurately obtain the corresponding cloud simulation cost prediction results, and reliably avoid potential financial losses. The implementation steps of the cloud simulation cost prediction method proposed in the embodiment of this application will be introduced in detail below.

[0063] Reference Figure 1, is a flow chart of the cloud simulation cost prediction method proposed in the first embodiment of the present application. The method proposed in the embodiment of the present application can be applied to electronic devices, such as terminal devices or servers with data processing capabilities, and the cloud simulation cost prediction method can also be implemented by the terminal device and the server in cooperation. Based on this, Figure 1 As shown, the cloud simulation cost prediction method proposed in this embodiment may include but is not limited to the following steps:

[0064] Step S11, obtaining feature quantification information of the model to be simulated for the cloud simulation platform; the feature quantification information includes target feature values ​​of multi-dimensional features that affect the cloud simulation cost of the cloud simulation platform executing the model to be simulated, and the target feature values ​​of the multi-dimensional features respectively indicate the target calculation equation, material complexity, simulation calculation step size, and model grid amount of the model to be simulated;

[0065] In the actual application of the present application, the historical data of the cloud simulation platform (such as the CAE platform) executing the historical simulation model can be classified to determine the various dimensional features that can affect the cloud simulation cost during the cloud simulation platform's execution of the historical simulation model. The four dimensional features that have the greatest impact on the computing resources of the cloud simulation platform can be analyzed, such as the target calculation equation used by the model to be simulated (which can be determined based on the level of computational complexity, and can be an equation derived from the field of continuous medium mechanics to describe the behavior of engineering product materials under different physical conditions), material complexity, simulation calculation step (the time interval / sampling interval between two iterative simulation calculations, the size of which directly affects the simulation accuracy and efficiency) and model mesh quantity (the number or degree of refinement of the meshes in the model to be simulated, which directly affects the resource requirements of the simulation).

[0066] Among them, the simulation model executed by the CAE platform can be an engineering product model, such as a material model, a geometric model, a network model, a physical model, a boundary condition and loading model, a multi-physics field coupling model, an optimization model, and a control and response model. One or more of them are used to jointly constitute a complete simulation environment to simulate and analyze the performance of engineering products under actual working conditions without actually manufacturing and testing engineering products. Usually, the performance of the corresponding engineering product can be simulated and determined through one or more functions such as structural analysis, fluid dynamics analysis, thermal analysis, electromagnetic field analysis, multi-physics field analysis, durability and fatigue analysis, collision and impact analysis, manufacturing process simulation, material analysis, noise and vibration analysis, control and system behavior simulation. Therefore, for the simulation model that needs to be executed by the CAE platform in this application, the various influencing factors of the cloud simulation cost can be determined in combination with the corresponding engineering product type analysis, and the implementation process is not described in detail in this application.

[0067] Based on this, according to any currently determined model to be simulated (such as the various engineering product models listed above), the four dimensional features determined above can be analyzed and quantified to obtain the eigenvalues ​​of the corresponding dimensions, that is, the quantified eigenvalue of the dimensional feature representing the target calculation equation used by the model to be simulated, the quantified eigenvalue of the dimensional feature representing the material complexity of the model to be simulated, the quantified eigenvalue of the dimensional feature representing the simulation calculation step of the model to be simulated, and the quantified eigenvalue of the dimensional feature representing the model grid amount of the model to be simulated, etc., so as to facilitate the subsequent use of the quantified eigenvalues ​​of each dimensional feature to quickly and accurately predict the resource requirements of the cloud simulation platform to execute the model to be simulated through the trained model.

[0068] It should be noted that the quantitative acquisition method of the four dimensional features listed above in this application can be determined according to actual needs, and this application does not impose any restrictions. Of course, with the upgrade or change of the version of the CAE cloud simulation platform, the consumption of computing resources and storage resources for executing the model to be simulated may change, and the influencing factors affecting the consumption of these resources (that is, the influencing factors affecting the cloud simulation cost, such as the above-mentioned different dimensional features) may also change. This application can flexibly adjust the corresponding feature quantification information based on these changes to ensure that the cloud simulation cost prediction results are reliable and accurate. This application does not limit the implementation method of how to analyze and determine the various influencing factors that affect the cloud simulation cost.

[0069] In order to improve prediction efficiency and reliability and reduce interference from human factors, in the process of acquiring the above-mentioned feature quantification information, appropriate machine learning models can be called to obtain certain dimensional feature values, such as grading the computational complexity of the simulation model to achieve the classification of simulation resource types, etc. The implementation process is not described in detail in this application.

[0070] Step S12, inputting the target feature value of the multi-dimensional feature into the trained decision tree regression model, and determining the target prediction value for the model to be simulated from a plurality of candidate prediction values ​​that are identified to match the target feature value of the multi-dimensional feature; each candidate prediction value is a prediction value represented by an end node of a prediction path that matches the target feature value of the multi-dimensional feature in a corresponding decision tree in the decision tree regression model;

[0071] In the embodiment of the present application, the decision tree regression model is a decision tree structure used for regression analysis in machine learning. As a high-performance prediction model for predicting the cloud simulation cost / resource demand / resource consumption of the cloud simulation platform of the model to be simulated, a prediction value can be generated at the end node (i.e., leaf node) of the decision tree, indicating the resource demand / consumption of the model to be simulated under the simulation condition constraints of each dimensional feature value. In order to ensure the prediction performance of the decision tree regression model, the multi-dimensional features extracted from the historical data of the simulation model executed on the cloud simulation platform can be analyzed to determine a sample data set to implement the training and testing of the decision tree regression model. The present application does not limit the training and testing implementation process.

[0072] Based on this, this application uses the quantized feature values ​​of each dimension as different inputs x of the model. i , that is, the target feature value of the i-th dimensional feature, is directly input into the trained decision tree regression model. Since the decision tree regression model contains multiple trained decision trees, the target feature values ​​of each dimensional feature input are judged according to the simulation conditions represented by the internal nodes of each decision tree in the decision tree regression model (which can be determined according to the actual use of the engineering product corresponding to the model to be simulated), so as to determine which child node under the internal node to enter, and then continue to judge the simulation conditions represented by the child node, and so on, to determine a prediction path for the target feature values ​​of each dimensional feature input in the decision tree, that is, to advance layer by layer from each decision tree root node of the decision tree regression model until a path is formed by connecting the end nodes, and the prediction values ​​represented by the end nodes of the prediction paths corresponding to multiple decision trees are determined as candidate prediction values, thereby obtaining multiple candidate prediction values ​​that match the target feature values ​​of the multi-dimensional feature, and then, from these multiple candidate prediction values, the target prediction value for the model to be simulated can be determined.

[0073] Among them, when the decision tree regression model contains multiple trained decision trees, after determining the candidate prediction values ​​represented by the terminal nodes of multiple prediction paths (different prediction paths belong to different decision trees) according to the above method, a voting strategy (using the majority voting strategy as an example) can be adopted to elect the candidate prediction value with the largest number of votes as the target prediction value, but it is not limited to this determination method. The target prediction value can also be determined by using strategies such as the Bully algorithm, the Raft algorithm, cluster voting, or consensus voting. The appropriate voting strategy can be flexibly selected according to actual needs, and the implementation process is not described in detail in this application. It can be seen that for different models to be simulated, the target feature values ​​of the multi-dimensional features listed above can be directly obtained, and input into the decision tree regression model, so as to quickly and accurately predict the resource requirements / consumption of the cloud simulation platform to execute the model to be simulated. Compared with the weighted summation of the resource consumption of each dimension of manual analysis, this application adopts a nonlinear prediction method implemented by a machine learning algorithm to comprehensively measure the influence of each dimensional feature on the cloud simulation cost, thereby improving the accuracy and reliability of the target prediction value, solving the problem of inaccurate prediction results caused by deviations in manual prediction analysis, especially for the prediction of cloud simulation costs of large-scale complex models, and reducing the ability requirements for forecasters.

[0074] Step S13: using the target prediction value, output the cloud simulation cost prediction result for the model to be simulated.

[0075] Combined with the above analysis, the predicted value represented by each leaf node in the decision tree regression model can be the predicted demand / consumption of computing resources, which can be represented by resource configuration parameters such as the number of computing cores. According to the needs, it can also include the predicted demand / consumption of storage resources, which can be represented by storage capacity, etc. Of course, it can also be cloud simulation cost, etc. This application does not restrict the content of the predicted value represented by each leaf node, and it can be flexibly determined according to the actual situation. You can even select the predicted value content according to actual needs, and then perform cloud simulation cost prediction on the simulation model. The implementation process is not described in detail in this application.

[0076] In this way, according to the method described above, after quickly and accurately obtaining the target prediction value for the model to be simulated, the target prediction value can be directly determined as the cloud simulation cost prediction result output, or the target prediction value can be processed according to the cloud simulation cost prediction result output requirements, such as output format or content requirements, and the cloud simulation cost prediction result that meets the requirements is obtained and then output. It can be sent to a display device (such as a display screen of an electronic device or a designated display device connected to an electronic device, etc.) for display output, or sent to an audio playback device (such as a speaker configured by the electronic device itself or an external independent audio player, etc.) for voice playback output, or sent to a printing device for print output, etc. This application does not limit the output method of the cloud simulation cost prediction result.

[0077] It should be understood that the cloud simulation cost prediction results in different output modes as listed above can be processed into cloud simulation cost prediction results in an output format that matches the output mode, and output according to the corresponding output mode. The contents of the cloud simulation cost prediction results output in different output modes can be the same or different, and can be determined in combination with the output content requirements of the corresponding output mode.

[0078] In summary, for the simulation task of any engineering product model (referred to as the model to be simulated), before using the CAE cloud simulation platform to execute the model to be simulated for performance analysis, it is necessary to first predict the corresponding cloud simulation cost in order to give a reasonable quotation. To this end, the embodiment of the present application will analyze the multi-dimensional features that affect the cloud simulation cost of the cloud simulation platform executing the model to be simulated, and obtain the target feature values ​​of the corresponding dimensional features through reasonable quantification processing, which constitute the feature quantification information of the model to be simulated. When participating in the cloud simulation cost prediction of the model to be simulated, the feature values ​​of each dimensionality are directly input into the trained decision tree regression model, and the multiple decision trees contained therein are directly identified in each decision tree. The terminal node of a prediction path that matches these inputs represents a candidate prediction value, and then the target prediction value for the model to be simulated is determined from the multiple identified candidate prediction values, that is, nonlinear prediction of the model to be simulated is realized, and various factors affecting the cloud simulation cost (i.e., influencing factors) are comprehensively weighed, and the resource requirements / consumption of cloud simulation for the model to be simulated under the corresponding calculation equations, material complexity, simulation calculation step size, and model grid volume indicated by each target eigenvalue are efficiently and reliably predicted, so as to accurately and reliably evaluate the cloud simulation cost of the model to be simulated using the target prediction value, ensure the reliability and accuracy of the cloud simulation quotation given based on this, and effectively avoid potential financial losses.

[0079] Reference Figure 2 , is a flow chart of the cloud simulation cost prediction method proposed in the second embodiment of the present application. This embodiment can further describe the process of obtaining the task quantization information in the method proposed above, but is not limited to the feature quantization implementation method described in this embodiment. Figure 2 As shown, the cloud simulation cost prediction method proposed in this embodiment may include:

[0080] Step S21, obtaining a model to be simulated for the cloud simulation platform;

[0081] Step S22, obtaining a target equation identifier corresponding to a target calculation equation used by the model to be simulated; the target equation identifier is determined as a target feature value of a first dimensional feature, and the target calculation equation includes at least one continuous medium mechanics equation;

[0082] In the embodiment of the present application, the model to be simulated for the CAE platform can be an engineering product model as listed above, so as to simulate and analyze different physical phenomena and engineering problems through different models. The calculation equations involved in the simulation and analysis process can be derived from the field of continuous medium mechanics, such as the three major fields of elasticity / plasticity mechanics, fluid dynamics and thermal conductivity. Therefore, the above-mentioned continuous medium mechanics equations can include one or more of the elasticity / plasticity mechanics equations, fluid dynamics equations and thermal conductivity equations. The target calculation equation used by the model to be simulated in the present application can be at least one of these three types of calculation equations, which can be determined according to the type of engineering product and its simulation performance analysis requirements.

[0083] In a possible implementation, when it is necessary to perform structural analysis on the simulation model, the target calculation equations that can be used may include elastic / plastic mechanics equations, which can be specifically determined according to the material property category. For the performance analysis of the fluid's operating state, the target calculation equation may include at least one fluid dynamics equation, such as the Navier-Stokes equations (momentum conservation analysis), the continuity equation (mass conservation analysis), etc. For the analysis of heat transfer in the material, the target calculation equation may include a heat conduction equation, which may involve steady-state and transient heat conduction analysis, and may be implemented based on the Fourier heat conduction equation, the energy conservation equation (heat equation), etc. This application does not limit the type of target calculation equation.

[0084] Among them, in order to facilitate the identification of different calculation equations (i.e., continuous medium mechanics equations), a one-to-one corresponding equation identifier can be configured for each calculation equation (continuum medium mechanics equation), such as numbering all calculation equations and using them as the target characteristic value of a dimensional feature to participate in subsequent steps, etc. This application does not restrict the content of the equation identifier.

[0085] It should be noted that the description content / simulation function of each calculation equation in the field of continuous medium mechanics listed above, as well as the calculation complexity in the simulation model are often different. For example, in the turbulence model to be simulated, the calculation complexity of the Navier-Stokes equations used for structural analysis is very high. The calculation complexity of the heat conduction equation is usually low when used for steady-state and transient heat conduction analysis, but the calculation complexity will increase when multiple physical fields are coupled. It can be seen that the calculation complexity of the same calculation equation in different simulation models, or in the same simulation model to achieve different performance analyses, may be different, and the higher the calculation complexity, the more computing resources are often required. The dynamic change of the calculation complexity of the calculation equation can easily lead to inaccurate calculation complexity of the target calculation equation determined by manual experience, thereby leading to inaccurate resource requirements predicted based on this.

[0086] Therefore, in order to accurately predict the cloud simulation cost of the model to be simulated and avoid over / underpredicting resource requirements, the present application can use machine learning models to continue analyzing historical data to determine the computational complexity of the calculation equations used in different performance analysis processes of different simulation models. Based on this, all calculation equations in the fields listed above are graded in computational complexity, so that each computational complexity level corresponds to one or more calculation equations. The present application does not elaborate on the implementation method of computational complexity grading.

[0087] Step S23, determining the material complexity of the model to be simulated by using the material model complexity and the number of material properties of the model to be simulated; the material complexity is determined as the target feature value of the second dimensional feature;

[0088] The material properties of the model to be simulated on the CAE platform are key parameters for defining and simulating material behavior, and are crucial to the accuracy and reliability of the simulation results. The various material properties of the model to be simulated can be determined by experimental measurement or by consulting a material manual, and defined in the model to be simulated before simulation. The implementation process is not described in detail in this application. It should be understood that different combinations of material properties may be required for different model simulation tasks, and the material property categories required for the model to be simulated can be flexibly determined based on actual needs. This application does not limit the content of the material model in the simulation model.

[0089] The material properties of the model to be simulated may include but are not limited to density (in fluid dynamics and structural dynamics simulation, density can be used to calculate the effects of mass and force), elastic modulus (describes the proportional relationship between stress and strain of the material within the elastic range. In elastic-plastic mechanics simulation, elastic modulus is a key parameter that determines the stiffness of the material), Poisson's ratio (describes the proportional relationship between lateral and longitudinal deformation of the material when it is stretched or compressed. It can be used together with the elastic modulus to calculate the stress-strain behavior of the material), thermal conductivity (describes the ability of the material to transfer heat energy. It is a key parameter for calculating temperature distribution in thermal conduction simulation) and thermal expansion coefficient (describes the rate of change of volume or length of the material when the temperature changes. In thermal-structural coupling simulation, the thermal expansion coefficient affects the thermomechanical behavior of the material). For some advanced material models to be simulated, it may also be necessary to define material parameters such as creep parameters and damage parameters to accurately simulate the complex behavior of the material.

[0090] By analyzing the historical data of the simulation model executed on the CAE platform, it is determined that the material complexity of the simulation model directly affects the simulation calculation efficiency. The embodiment of the present application can determine the material complexity of the model to be simulated from two aspects: the material model complexity and the number of material properties, to measure the simulation calculation efficiency of the model to be simulated.

[0091] Among them, due to the mechanical behavior of the material described by the material model and the stress-strain relationship of the material under the action of external force, the present application can determine the complexity of the material model of the model to be simulated by determining the target material stress-strain relationship and the target behavior model described by the material model in the model to be simulated. The implementation process is not described in detail in this application. In order to facilitate subsequent processing, the present application can pre-configure corresponding indicator identifiers, such as numbers or characters, for different material stress-strain relationships and different behavior models (i.e., the mechanical behavior of the material), and then directly use the indicator identifier for processing, which is conducive to improving processing efficiency compared to the corresponding material stress-strain relationship or behavior model represented by the indicator identifier.

[0092] The number of material properties of the model to be simulated can be obtained by counting the number of material properties defined in the model to be simulated. Combined with the above description of material properties, different material properties have different descriptions to simulate different material behaviors and realize the analysis of corresponding performance. It can be seen that the larger the number of material properties, the longer the time it takes for the CAE platform to execute the model to be simulated, which will affect the efficiency of simulation calculation.

[0093] Step S24, determining the inverse of the simulation calculation step length configured for the to-be-simulated model as the target feature value of the third dimensional feature;

[0094] In this application, the simulation calculation step refers to the minimum time interval or spatial interval for the solver to perform calculations during the execution of the model to be simulated by the CAE platform. The simulation calculation step has different meanings in different analysis types. For example, in structural dynamics analysis, the simulation calculation step can be the time step in transient analysis, such as solving a set of target calculation equations in each time step, and its size directly affects the simulation calculation accuracy and stability. If the time step is too small, the calculation amount and calculation time will increase. If the time step is too large, the target calculation equation will be unstable or the accuracy will decrease. This application needs to reasonably select the time step according to the actual simulation requirements.

[0095] In spatial domain simulation, the simulation calculation step can be the spatial step, that is, the size of the grid unit, which affects the accuracy and computational cost of the solution. In cases involving complex physical processes (such as contact, plastic deformation, etc.), the simulation calculation step can be a physical sub-step to capture physical phenomena more finely within the main time step. In nonlinear analysis, the simulation calculation step can be the solution step, which can refer to the step size for updating the solution in each iteration to control the convergence of the nonlinear solution.

[0096] Based on the above analysis, the present application can comprehensively consider the purpose of the simulation, accuracy requirements, computing resources and characteristics of the physical process, and select an appropriate simulation calculation step. For example, in structural dynamics analysis, the time step can be less than or equal to a certain fraction of the system's minimum natural period to ensure numerical stability. In thermal analysis, the spatial step can be small enough to capture changes in temperature gradients. In practical applications, the tools and auxiliary functions provided by the CAE platform can be used to pre-configure a suitable simulation calculation step for the model to be simulated. The present application does not limit the value of the simulation calculation step and its configuration method.

[0097] In combination with the above problems caused by various simulation calculation steps being too large or too small, the embodiments of the present application can measure the calculation accuracy and efficiency, select a suitable simulation calculation step, and then use the inverse of the simulation calculation step to participate in subsequent processing. Compared with directly using the simulation calculation step to participate in subsequent calculations, this can improve the reliability of calculation cost measurement.

[0098] Step S25, scaling the model grid amount of the to-be-simulated model to obtain a target feature value of the fourth dimensional feature;

[0099] Since the model grid volume refers to the number or refinement of the grids in the model to be simulated, this directly affects the computational requirements and storage requirements of the simulation. Different refinement levels have a great impact on the simulation computational volume, directly affecting the computational time. For example, the finer the grid, the more units there are in the model to be simulated, and each unit takes a certain amount of time to calculate, resulting in a large amount of computation for the entire model to be simulated; and the refined grid will also increase the amount of data that the solver needs to process, thereby increasing the computational time. In addition, the refined grid will also increase the memory requirements, which can easily lead to insufficient memory in electronic devices, further slowing down the calculation speed. In some cases, although finer grids can improve the accuracy of simulation results, this may require more iterative steps to reach convergence, which will also increase the computational time, especially in the nonlinear analysis process. The finer the degree of mesh, the more complicated the solution process will be, increasing the computational time.

[0100] According to the above analysis, this application can comprehensively consider the actual requirements of the simulation and reasonably set the model grid volume. However, since the model grid volume is usually in the hundreds of thousands, it is directly involved in the subsequent processing, especially when the model grid volumes of different models (such as the various engineering product models listed above) are quite different, it will highlight the role of model features with larger values ​​in the entire analysis and weaken the role of model features with larger values ​​in the entire analysis, affecting the accuracy of the analysis results of the entire model to be simulated. In this regard, this application proposes to scale the model grid volume of the model to be simulated, and the target eigenvalue of the fourth dimensional feature that measures the calculation time of the model to be simulated participates in the subsequent calculation. This application does not limit the order of magnitude of the scaling, and can be determined based on the order of magnitude of the grid volumes of each model contained in the model to be simulated.

[0101] In one possible implementation, when the grid volume of each model is in the tens of thousands, the reduction order of magnitude of the model grid volume for the model to be simulated can be determined accordingly, such as in the tens of thousands, and the ratio of the model grid volume to the reduction order of magnitude is determined as the target characteristic value of the fourth dimension feature, such as model grid volume / 10000, so as to measure the model grid volume of the model to be simulated, and then measure the calculation time, but it is not limited to this reduction order of magnitude. In the model to be simulated, the scale of each model is small, and the grid volume of each model is in the thousands of data volumes, and the reduction order of magnitude can be set to the thousands; similarly, if the scale of each model in the model to be simulated is larger, such as in the millions or even hundreds of millions, the reduction order of magnitude of the corresponding order of magnitude can be configured.

[0102] Step S26, inputting the target feature value of each dimensional feature into the trained decision tree regression model, and identifying a candidate prediction value represented by the terminal node of a prediction path that matches the target feature value of each dimensional feature from each decision tree included in the decision tree regression model;

[0103] Step S27, determining a target prediction value for the model to be simulated from the candidate prediction values ​​of each of the identified multiple decision trees;

[0104] Step S28: using the target prediction value, output the cloud simulation cost prediction result for the model to be simulated.

[0105] Following the above analysis, according to the method described above, the eigenvalues ​​of the four dimensional features are obtained after quantization, which can be recorded as x 1 、x 2 、x 3 、x 4, you can directly input the trained decision tree regression model, and use the multiple tree structures (decision trees) in the decision tree regression model for prediction. Among them, each internal node of each decision tree represents the simulation conditions for the characteristics of a certain dimension, such as performance requirements and various constraints, etc. By comparing with the input feature value, it is determined which child node under the internal node to enter, and then further judgment is made based on the simulation conditions represented by the child node, and then continue to enter the next layer of nodes, and query layer by layer until the end node is reached, and the predicted value represented by the end node is determined as a candidate predicted value of the model to be simulated. Repeat the identification of multiple decision trees in this way to determine the candidate predicted value in each decision tree that matches the target feature value of the input multi-dimensional feature.

[0106] Among them, in the above-mentioned layer-by-layer query process for each decision tree, the node connections passed through constitute a prediction path for the model to be simulated, and a target prediction path is determined from the characteristic data that meets the simulation conditions represented by each node contained in the prediction path to achieve accurate prediction of the resource demand / consumption / cloud simulation cost of the model to be simulated, that is, the candidate test value represented by the terminal node in a prediction path determined by each decision tree is identified, and then the target prediction value is determined from multiple candidate prediction values, thereby obtaining the cloud simulation cost prediction result that meets the output requirements and then outputting it. It should be noted that due to the different data sets used for training different decision trees, there will be certain differences in the meaning and relationship represented by each node in the obtained tree structure. The candidate prediction values ​​represented by the terminal nodes of different prediction paths obtained according to the above-mentioned layer-by-layer query method are often different. The present application comprehensively considers the prediction results of each decision tree representation strategy, improves the accuracy of the obtained target prediction value, and thus improves the reliability and accuracy of cloud simulation cost estimation.

[0107] Moreover, compared with weighted combination of various influencing factors affecting cloud simulation costs based on empirical weight ratios, the present application will quantify the features of multiple dimensions that have a greater impact on cloud simulation costs, and use the feature values ​​of the corresponding dimensions to measure the relevant physical quantities that affect the cloud simulation cost of the model to be simulated, such as model grid volume, target calculation equation, material complexity, and simulation calculation step length. After that, they are input into the trained decision tree regression model for prediction, which improves the prediction efficiency. Moreover, since the decision tree regression model is trained based on the historical data of the simulation model executed on the CAE platform, it can more accurately analyze the impact of different feature values ​​of each of these multi-dimensional features on the predicted value, so as to accurately and reliably determine a reasonable prediction path and obtain a reasonable prediction value, thereby solving the technical problems existing in manual analysis, improving the reliability and accuracy of cloud simulation cost prediction, and effectively avoiding financial losses caused by too low a cloud simulation cost assessment.

[0108] Reference Figure 3, is a flow chart of the cloud simulation cost prediction method proposed in the third embodiment of the present application. This embodiment can describe an optional detailed implementation method of how to determine the material complexity of the model to be simulated in the cloud simulation cost prediction method proposed above, such as Figure 3 As shown, the method may include but is not limited to:

[0109] Step S31, determining the number of material properties in each material model included in the model to be simulated;

[0110] Combined with the above description of material properties and material models, in the CAE platform, the simulation calculation of the model to be simulated depends on the material model and material properties. Material properties are specific parameters in the material model, which quantify the physical and mechanical properties of the material and are used to define and distinguish different material models. For example, for an elastic material model, the material properties that need to be specified are elastic modulus and Poisson's ratio; for a plastic material model, the material properties that need to be specified are yield strength and hardening parameters, etc. This application does not limit the material properties that need to be specified for different material models, which can be determined according to the situation.

[0111] Step S32, determining the stress-strain relationship of the target material described by each material model, and the target behavior model to which it belongs; the target behavior model is used to describe the mechanical behavior of the material under the action of an external force;

[0112] Step S34, querying the correspondence between different types of material stress-strain relationships and behavior models and different indicator identifiers, and obtaining a first indicator identifier corresponding to the target material stress-strain relationship and a second indicator identifier corresponding to the target behavior model;

[0113] In an embodiment of the present application, the above-mentioned different types of material stress-strain relationships include linear relationships and nonlinear relationships. For the convenience of processing, different material stress-strain relationships can be configured with unique indicator identifiers, such as values ​​such as 1 or 0. For example, the indicator identifier corresponding to the linear relationship is configured as a first value, and the indicator identifier corresponding to the nonlinear relationship is configured as a second value.

[0114] In addition, the types of mechanical behaviors generated by the material under the action of external forces can be divided into elasticity, plasticity, viscoelasticity, creep, damage and fracture, etc. The embodiment of the present application will analyze the two mechanical behaviors of elasticity and plasticity. The models that simulate different mechanical behaviors can be called different types of behavior models, such as elastic models and plastic models. To facilitate subsequent processing, the indicator identifier corresponding to the configuration of the plastic model can be a first value, and the indicator identifier corresponding to the configuration of the elastic model can be a second value.

[0115] It can be seen that from the above-mentioned material stress-strain relationship and mechanical behavior analysis, each material model included in the simulation model can be a linear / nonlinear, and elastic / plastic material model, thereby determining the complexity of the material model. For the indicator identification corresponding to different characteristic configurations, the indicator identification method described above is used for differentiation and representation, such as the first value is 0 and the second value is 1. It can also be adjusted to different character representations according to actual conditions. This application does not limit the content of the indicator identification.

[0116] In practical applications, various material stress-strain relationships of various material models and behavior models representing various mechanical behaviors can be configured with one-to-one corresponding indicator identifiers, and the corresponding relationship between each indicator identifier and the corresponding mechanical characteristics of the material model can be stored. In this way, according to the above steps, after determining whether the stress-strain relationship of the target material described by a certain material model is a linear relationship or a nonlinear relationship, and whether the mechanical behavior of the corresponding material under the action of external force is elastic or plastic, the stored corresponding relationship can be queried to determine the indicator identifier corresponding to each mechanical characteristic to measure the complexity of the corresponding material model.

[0117] For example, if material A used in an engineering product is elastically linear, the method described above can be used to determine that the first index of the material A model is 0 and the second index is 0. These two values ​​can be directly used in subsequent processing, which is simple and clear compared to the detailed description of the stress-strain relationship of the material, and is conducive to shortening the calculation time and reducing the consumption of computing resources.

[0118] Step S35, using the material property quantity, the first index identifier and the second index identifier, obtain the material complexity of the corresponding material model in the to-be-simulated model.

[0119] In a possible implementation, combined with the above description of the first / second indicator identification, the number of material properties, the first indicator identification, and the second indicator identification corresponding to the same material model can be directly summed to obtain the material complexity of the corresponding material model. For example, material B has two material shapes, elastic modulus and Poisson's ratio (that is, the number of material properties is 2), and is plastic (its corresponding second indicator identification is 1) and nonlinear (its corresponding second indicator identification is 1). According to the processing method described in the example above, the complexity of material B can be 4, that is, 2+1+1=4, and 4 can be used as the target characteristic value of the second dimension feature. Similarly, the material complexity measurement process for other material models is similar, and this application does not give detailed examples one by one.

[0120] It can be seen that the present application analyzes the complexity of each material model in the model to be simulated, as well as the corresponding number of material properties, and obtains the characteristic value representing the corresponding material complexity from the corresponding indicator identifier and the number of material properties, so as to accurately measure the computational efficiency of the model to be simulated and participate in subsequent processing. Compared with directly using each material property and each material model, it is simpler and easier to calculate.

[0121] In some embodiments, in combination with the above description of the process of obtaining the eigenvalues ​​of the first dimension, each computational equation in the field of continuous medium mechanics is configured with a unique corresponding equation identifier. After obtaining the historical data of the cloud simulation platform executing the simulation model, the historical data can be analyzed through a machine learning model to determine the computational complexity grading relationship between the equation identifiers of different computational equations and different types of simulation models, that is, to achieve automatic grading of the computational complexity of each computational equation used for different types of simulation models, and determine which one or more computational equations correspond to each computational complexity level, and each computational equation is represented by a corresponding configured equation identifier.

[0122] In this way, after determining the model type to which the model to be simulated belongs, the computational complexity classification relationship obtained above can be queried accordingly to determine the target equation identifier corresponding to the model type, and the computational equation with the target equation identifier is the target computational equation used by the model to be simulated. It should be noted that this application does not restrict the representation and storage method of the computational complexity classification relationship, such as table storage or storage according to certain rules, etc. The corresponding query method can be used to determine the target equation identifier, and the implementation process is not described in detail in this application.

[0123] Reference Figure 4 , is a flow chart of the cloud simulation cost prediction method proposed in the fourth embodiment of the present application. This embodiment can describe the training process of the decision tree regression model used in the cloud simulation cost prediction method proposed above, such as Figure 4 As shown, the training process may include but is not limited to:

[0124] Step S41, obtaining historical data of the simulation model executed by the cloud simulation platform;

[0125] In this embodiment, the historical data of the CAE platform executing the simulation model can be obtained by reading log files or databases, that is, the data obtained by executing various historical simulation tasks, such as the model mesh volume, calculation equation complexity, material complexity, simulation calculation step size, calculation time, number of computing cores, computing resource consumption and storage resource consumption, etc. The content of the historical data can be determined based on the cloud simulation cost prediction requirements, and this application does not impose any restrictions.

[0126] Step S42, determining historical feature values ​​corresponding to different dimensional features, and historical simulation conditions and historical prediction values ​​corresponding to different historical feature values ​​based on historical data;

[0127] Step S43, constructing multiple initial decision trees using historical feature values, historical simulation conditions, and historical prediction values ​​corresponding to features of different dimensions;

[0128] Combined with the above description of the constituent nodes of the tree structure of the decision tree regression model, the present application can make a prediction of the cloud simulation cost of the simulation model by constructing a decision process of the tree structure. During the construction process, appropriate features can be selected as nodes of the decision tree to reduce the prediction error. In an embodiment of the present application, a training data set suitable for the CAE platform simulation model can be obtained according to the method described above. Combined with the above description of the characteristic values ​​of the multi-dimensional features for predicting the cloud simulation cost of the model to be simulated, various combinations of the characteristic values ​​of each dimensional feature (such as calculation equations, material property quantity, simulation calculation step size or model grid quantity) in each simulation model can be obtained to construct different data sets (i.e., sample sets) for training different initial decision trees. Among them, the characteristic value range of the same dimensional feature in different data sets can be different, and can even include characteristic values ​​of different dimensional features. The random forest method can be used to determine the data sets used by each of the constructed multiple initial decision trees from all the characteristic values ​​of the dimensional features to train the corresponding initial decision trees.

[0129] For example, the equation identifiers of each calculation equation under the dimensional feature of the calculation equation are used as the historical characteristic values ​​of the dimensional feature to realize automatic grading of the calculation complexity; the historical material property quantities corresponding to different material models are determined as the historical characteristic values ​​of the dimensional feature of the material property to determine the material complexity; various calculation step values ​​are used as the historical characteristic values ​​of the dimensional feature of the simulation calculation step to select the appropriate simulation calculation step; various model grid quantities of different simulation models are obtained to determine the historical characteristic values ​​of the dimensional feature of the model grid quantity to select the appropriate degree of model grid refinement.

[0130] In addition, for different simulation models, the boundary conditions and external loads of the simulation objects (such as engineering products), as well as performance requirements and constraints, etc. are defined. They can be used as historical simulation conditions, i.e., decision conditions, to realize the decision of the characteristic values ​​of each dimension, and they can be used as internal nodes of the initial decision tree. Each internal node contains all the historical characteristic values, i.e., sample sets, corresponding to the characteristics of each dimension as described above; the cloud simulation cost / resource consumption (such as the number of computing cores, etc.) of each simulation model is used as the corresponding historical prediction value, and it is used as the leaf node of the corresponding initial decision tree. On this basis, the feature data sets (i.e., sample sets) corresponding to each internal node are divided into smaller subsets through recursion and other methods, that is, by recursively performing split feature selection and split point selection, the sub-nodes are continuously divided, and finally a complete decision tree is constructed.

[0131] It should be noted that since the present application adopts the random forest method to construct multiple decision trees for regression prediction of cloud simulation costs, the accuracy and robustness of the overall model can be improved by combining the prediction results of multiple decision trees (such as the target prediction value determined from the candidate prediction values ​​represented by the end nodes of a prediction path corresponding to each decision tree). After constructing / training multiple decision trees according to the method described above, the multiple decision trees finally obtained can be determined as a decision tree regression model for predicting cloud simulation costs.

[0132] Step S44, for each internal node of each initial decision tree, by evaluating the information gain of features of different dimensions, a one-dimensional feature is selected from the internal node as a split feature;

[0133] Among them, information gain is an indicator to measure the contribution of a feature to the classification result. In the embodiment of the present application, the size of the information gain represents the accuracy of the cloud simulation cost prediction of the model according to the corresponding dimensional feature split estimation. Usually, the dimensional feature with the larger information gain is selected as the split feature for processing, and the more accurate the cloud simulation cost prediction result estimated by the model is. Therefore, the dimensional feature with the largest information gain in the same internal node can be selected as the split feature, and then the appropriate split point (the historical feature value with the smallest mean square error) can be selected from its historical feature values ​​to realize the classification of the feature data set of the dimensional feature under the internal node, that is, all historical feature values, so as to reduce the mean square error of the two child nodes after the feature data set is split, that is, the variance of the target variable value of each subset after the split. The present application does not limit the method for selecting the split point.

[0134] It can be seen that for each internal node, the decision tree can evaluate which dimensional feature can most effectively reduce the model prediction error at the internal node. For example, the dimensional feature of "simulation calculation step length" as a split feature can significantly reduce the prediction error (that is, the cloud simulation cost prediction result estimated by the model is more accurate). This dimensional feature can be selected as a split feature, and then a suitable split point is selected from the eigenvalues ​​of the dimensional feature to split all eigenvalues ​​of the dimensional feature of the internal node into two subsets, and the internal node is split into two corresponding child nodes. The selection of split features can be achieved by evaluating the information gain of each dimensional feature. In a possible implementation, the information gain can be calculated using the following formula:

[0135] ; (1)

[0136] In formula (1), D can represent the set of all samples of a certain dimension feature (i.e., all historical feature values ​​under this dimension feature), and x is the dimension feature identifier (such as the feature name). Taking the dimension feature of calculation equation as an example, after analyzing historical data, it is known that the total number of calculation equations used by different simulation models is M, D m It can represent the sample set of m in the dimension feature of "calculation equation" among all samples D. Gain(D, x) can represent the information gain of the dimension feature x in all samples D. Ent(D) and Ent(D m ) represent the information entropy of the corresponding sample set, which can be calculated using the following formula:

[0137] ; (2)

[0138] In formula (2), |y| can represent the sample type, such as the number of types of features in the dimension of "calculation equation". For example, if there are 20 types of calculation equations, |y|=20, k is 1 and counts to 20, p k It can represent the probability / proportion of the kth calculation equation among all calculation equations. For example, if the kth calculation equation is the heat conduction equation, and the heat conduction equation accounts for 18 / 100 in all samples, then p k 18 / 100. Log 2 Function represents the logarithmic function with base 2. ∑ represents the summation operation. For Ent(D m ) is calculated in a similar manner to Ent(D) and will not be described in detail in this application.

[0139] In an embodiment of the present application, the information gain calculation method described above can be used to calculate the information gain of the dimensional features described above for each internal node. After comparison, a one-dimensional feature with the maximum information gain is selected as the split feature, and the corresponding split point is selected according to the method described below to achieve binary classification of the historical feature values ​​of the dimensional feature under the internal node, thereby obtaining two corresponding child nodes.

[0140] Step S45, obtaining the mean square error of two child nodes corresponding to each splitting point corresponding to the splitting feature;

[0141] Step S46, taking the splitting point corresponding to the minimum mean square error as the target splitting point, and performing splitting feature segmentation on the corresponding internal nodes;

[0142] After selecting the split feature according to the method described above, the sample set consisting of each historical feature value under the split feature can be classified to split the corresponding internal node into two sub-nodes, each of which corresponds to a historical feature value range of the split feature. In order to obtain the best split point, that is, a historical feature value that can minimize the prediction error, all possible split points can be traversed, and the mean square error between the two sub-nodes obtained after splitting the split feature of the internal node according to the split point can be calculated to select the split point corresponding to the minimum mean square error as the target split point, so as to split the feature value under the split feature according to the split point.

[0143] For each possible split point s traversed, the mean square error MSE(s) of the two child nodes after the split can be calculated using the following formula:

[0144] ; (3)

[0145] In formula (3), t1 and t2 represent the two child nodes after an internal node in the decision tree is split, n t Indicates the total number of samples contained in the split feature represented by the internal node t, n t1 and n t1 They respectively represent the number of samples belonging to the split feature obtained by segmenting the child node t1 and the child node t2, that is, the number of subset samples, and MSE (t1) and MSE (t2) respectively represent the mean square error of the corresponding child nodes. This application does not elaborate on the calculation process. It should be noted that other error calculation methods can also be used for the selection of the split point, and are not limited to this mean square error calculation method.

[0146] For example, if the dimension feature of the computational equation is selected as the splitting feature, all computational equations are traversed according to different classifications (i.e., computational complexity grading), and finally the equation identifiers of each computational equation at different computational complexity levels are determined, the automatic grading of the computational complexity of all computational equations is achieved. Compared with the manual grading method, the grading efficiency is improved, as well as the reliability and accuracy in cloud simulation cost prediction.

[0147] Similarly, when other dimensional features such as the number of material properties, simulation calculation step size, and model grid size are selected as splitting features, the method described above can also be used to determine the reasonable number of material properties under different material models, so as to determine the corresponding material complexity according to the method described above; select a reasonable simulation calculation step size and a model grid size of appropriate data level, etc. The implementation process will not be described in detail with examples one by one in this application.

[0148] Step S47, determining that each new decision tree meets the termination condition, and determining each new decision tree as a decision tree regression model.

[0149] Combination Figure 5 As shown in the decision tree splitting process diagram, for each initial decision tree, after determining the best splitting feature and its corresponding best splitting point (i.e., the target splitting point) s according to the method described above, all historical feature values ​​contained in the splitting feature represented by the internal node, the historical feature values ​​less than or equal to s can be divided into a subset, and the historical feature values ​​greater than s can be divided into another subset. Each subset represents a child node, and the output value of the node feature of this dimension can be the mean of each historical feature value of the feature of this dimension represented by the node.

[0150] It should be understood that for each initial decision tree, after completing the splitting of an internal node, it can be determined whether the new decision tree obtained by the split meets the termination condition, such as whether the depth of the new decision tree reaches the preset value, and / or whether the number of samples of each leaf node is lower than the threshold. If not, it means that the termination condition is not met, and each child node after the split can still be split according to the method described above. Through this recursive processing, until it is determined that the new decision tree obtained by the split meets the termination condition, such as the depth of the new decision tree reaches the preset value, or the number of samples of each leaf node is lower than the threshold, the splitting process is terminated. After completing the processing of multiple initial decision trees and obtaining the corresponding multiple new decision trees that meet the termination condition, the multiple finally generated new decision trees are determined as the decision tree regression model required for cloud simulation cost prediction, so as to achieve accurate prediction of the cloud simulation cost of the model to be simulated through multiple decision trees. It should be noted that the termination condition content includes but is not limited to the above example content, and the termination condition content can be flexibly configured or adjusted according to actual needs. This application does not give detailed examples one by one.

[0151] In practical applications, for the decision tree regression model obtained by the above training, the characteristic values ​​of the four dimensions of the calculation equation identifier, material complexity, inverse of the calculation step length, and scaled model grid volume used by a certain simulation model can be obtained based on historical data, recorded as test characteristic values, and input into the decision tree regression model, so as to determine multiple test paths matching the test characteristic values ​​of the four dimensions input according to the tree structure of the multiple decision trees generated when the termination conditions are met, determine the predicted value represented by the end node of each test path, and then use the voting strategy to select the appropriate predicted value, and calculate the prediction loss between the predicted value and the true value of the corresponding simulation model. If the prediction loss is less than the loss threshold, it means that the prediction performance of the decision tree regression model meets the prediction requirements and can be directly used for the prediction of the cloud simulation cost of the model to be simulated; otherwise, it means that the prediction performance of the decision tree regression model does not meet the prediction requirements, and the split optimization can be continued according to the training method described above until the prediction performance of the optimized decision tree regression model meets the prediction requirements, and the final decision tree regression model is stored for realizing the prediction of the cloud simulation cost of the model to be simulated.

[0152] It should be understood that after using the CAE platform to execute the model to be simulated and obtaining the actual cloud simulation cost, the loss between it and the cloud simulation cost prediction result of the model to be simulated using the decision tree regression model can be calculated, and the decision tree regression model can be optimized based on the loss. The implementation process is not described in detail in this application.

[0153] A cloud simulation cost prediction method provided in an embodiment of the present application is introduced above, and a device for executing the above cloud simulation cost prediction method will be introduced below.

[0154] Reference Figure 6 , is a schematic diagram of the structure of a cloud simulation cost prediction device proposed in an embodiment of the present application, such as Figure 6 As shown, the cloud simulation cost prediction device may include:

[0155] A feature quantification information acquisition module 61 is used to acquire feature quantification information of a model to be simulated on a cloud simulation platform; the feature quantification information includes target feature values ​​of multi-dimensional features that affect the cloud simulation cost of the cloud simulation platform executing the model to be simulated, and the target feature values ​​of the multi-dimensional features respectively indicate a target calculation equation, material complexity, simulation calculation step size, and model grid amount of the model to be simulated;

[0156] A prediction module 62 is used to input the target feature value of the multi-dimensional feature into a trained decision tree regression model, and determine the target prediction value for the model to be simulated from a plurality of candidate prediction values ​​that are identified to match the target feature value of the multi-dimensional feature; each of the candidate prediction values ​​is a prediction value represented by an end node of a prediction path that matches the target feature value of the multi-dimensional feature in a corresponding decision tree in the decision tree regression model;

[0157] The output module 63 is used to output the cloud simulation cost prediction result for the model to be simulated using the target prediction value.

[0158] In some embodiments, the feature quantitative information acquisition module 61 may include:

[0159] A to-be-simulated model acquisition unit, used to acquire a to-be-simulated model for a cloud simulation platform;

[0160] A target equation identifier acquisition unit is used to acquire a target equation identifier corresponding to a target calculation equation used by the model to be simulated; the target calculation equation identifier is determined as a target characteristic value of a first dimensional characteristic, and the target calculation equation includes at least one characteristic value of a continuous medium mechanics equation;

[0161] A material complexity determination unit, configured to determine the material complexity of the model to be simulated by using the material model complexity and the number of material properties of the model to be simulated; the material complexity is determined as a target feature value of the second dimensional feature;

[0162] A characteristic value determination unit, used to determine the inverse of the simulation calculation step configured for the to-be-simulated model as a target characteristic value of the third dimensional characteristic;

[0163] The feature scaling unit is used to scale the model grid amount of the model to be simulated to obtain the target feature value of the fourth dimensional feature.

[0164] In a possible implementation, the material complexity determination unit may include:

[0165] A first determining unit, used to determine the number of material properties corresponding to each material model included in the model to be simulated;

[0166] A second determination unit is used to determine the stress-strain relationship of the target material described by each of the material models, and the target behavior model to which it belongs; the target behavior model is used to describe the mechanical behavior of the corresponding material under the action of an external force;

[0167] A first query unit is used to query the corresponding relationship between different types of material stress-strain relationships and behavior models and different indicator identifiers, and obtain a first indicator identifier corresponding to the target material stress-strain relationship and a second indicator identifier corresponding to the target behavior model;

[0168] The first obtaining unit is used to obtain the material complexity of the model to be simulated by using the material property quantity, the first indicator identifier and the second indicator identifier.

[0169] Among them, different types of material stress-strain relationships include linear relationships and nonlinear relationships, and the indicator identifier configured corresponding to the linear relationship is a first value, and the indicator identifier configured corresponding to the nonlinear relationship is a second value. Different types of behavior models include elastic models and plastic models, and the indicator identifier configured corresponding to the plastic model is the first value, and the indicator identifier configured corresponding to the elastic model is the second value.

[0170] Based on this, the first obtaining unit includes:

[0171] A summing subunit is used to sum the material property quantity, the first index identifier and the second index identifier to obtain the material complexity of the corresponding material model in the model to be simulated.

[0172] In a possible implementation, the target equation identification obtaining unit may include:

[0173] A first acquisition unit is used to acquire a computational complexity classification relationship between equation identifiers of different computational equations and different types of simulation models; the computational complexity classification relationship is determined by a machine learning model based on historical data of the simulation model executed by the cloud simulation platform;

[0174] The second query unit is used to query the computational complexity classification relationship and determine the target equation identifier corresponding to the model type to which the model to be simulated belongs.

[0175] In a possible implementation, the simulation calculation step size is configured based on the calculation accuracy requirement and the calculation efficiency requirement of the model to be simulated, and the implementation process is not described in detail in this application.

[0176] The feature scaling unit may include:

[0177] A reduction order of magnitude determining unit, used to determine a reduction order of magnitude of a model grid amount for a model to be simulated; the reduction order of magnitude includes ten thousand;

[0178] The third determining unit is used to determine the ratio of the model grid volume to the reduction order of magnitude as the characteristic value of the fourth dimension.

[0179] In some embodiments, a training module for training a decision tree regression model may include:

[0180] A historical data acquisition unit, used to acquire historical data of the simulation model executed by the cloud simulation platform;

[0181] A fourth determination unit is used to determine, based on the historical data, each historical characteristic value corresponding to different dimensional characteristics, and historical simulation conditions and historical prediction values ​​corresponding to different historical characteristic values; the different dimensional characteristics include calculation equations, material property quantities, simulation calculation steps, and model grid quantities;

[0182] A construction unit, configured to construct a plurality of initial decision trees by using the historical feature values ​​corresponding to the features of different dimensions, the historical simulation conditions and the historical prediction values;

[0183] A split feature selection unit, configured to select a one-dimensional feature from each internal node of each of the initial decision trees as a split feature by evaluating the information gain of features of different dimensions;

[0184] A mean square error acquisition unit, used to acquire the mean square error of two sub-nodes corresponding to each splitting point corresponding to the splitting feature;

[0185] A splitting unit, used to take the splitting point corresponding to the minimum mean square error as the target splitting point, and perform the splitting feature segmentation on the corresponding internal node;

[0186] The fifth determining unit is used to determine whether each new decision tree meets the termination condition and determine each new decision tree as a decision tree regression model.

[0187] An embodiment of the present application also provides a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any one of the cloud simulation cost prediction methods provided in the embodiments of the present application.

[0188] Wherein, when the computer program instructions are loaded and executed on a computer (electronic device), the above-mentioned process or function according to the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, training equipment, or data center to another website, computer, training equipment, or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0189] Reference Figure 7 , is a schematic diagram of the structure of an electronic device proposed in an embodiment of the present application. The electronic device may be a terminal device or a server. The electronic device is a server as an example for explanation. The server may be one or more physical servers, or a cloud server supporting cloud computing, such as Figure 7 As shown, the electronic device may include: at least one communication element 710, at least one memory 720 and at least one processor 730, wherein:

[0190] At least one communication element 710, at least one memory 720 and at least one processor 730 can communicate with each other via a bus. The bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0191] The communication element 710 can be used to obtain the model to be simulated, which can be obtained from other devices or platforms or local storage devices. Based on this, the communication element 710 may include a first communication element that supports wireless network communication between the electronic device and other devices (such as a server or database, etc.), and the first communication element includes but is not limited to: supporting wireless communication methods such as wifi, 5G / 6G mobile communication, Bluetooth and / or near-field channels, GSM (Global System of Mobile communication) communication, GPRS (General Packet Radio Service) communication, etc. corresponding to one or more communication elements.

[0192] It should be understood that in order to transmit data between the various components inside the electronic device, the communication element 710 may also include one or more second communication elements that support wired communication, such as a general-purpose input / output (GPIO) interface, a USB interface, a universal asynchronous receiver / transmitter (UART) interface, and the like. The present application does not limit the type of communication element 710 contained in the electronic device and its communication transmission mechanism, which may be determined according to the circumstances.

[0193] The memory 720 can be used to store computer instructions and data, and may include an instruction storage area and a data storage area. The data storage area can store various data, such as the model to be simulated and the intermediate data and results generated in the cloud simulation cost prediction process; the instruction storage area can store software units such as computer instructions required for an operating system, an application, and at least one function (such as a display function, a decision tree building function, etc.). In an embodiment of the present application, the memory 720 can provide the processor 730 with multiple computer instructions for implementing the cloud simulation cost prediction method of an embodiment of the present application, as well as data or other software and hardware resources required during the execution of the cloud simulation cost prediction method.

[0194] The processor 730, as the control center of the electronic device, can use various interfaces and lines to connect various parts of the entire electronic device, and implement the various steps of the cloud simulation cost prediction method proposed in the embodiment of the present application by loading and executing multiple computer instructions stored in the memory 720. Other components of the electronic device (such as one or more input components and output components, etc.) can also be called to implement corresponding functions.

[0195] In the embodiment of the present application, the memory 720 may include a volatile memory (volatile memory), such as a random access memory (random access memory, RAM), and may also include a non-volatile memory (non-volatile memory), such as a read-only memory (read-only memory, ROM), a flash memory, a mechanical hard disk (hard disk drive, HDD) or a solid state drive (solid state drive, SSD) and other storage media in one or more forms. The processor 730 may include any one or more of a central processing unit (central processing unit, CPU), a graphics processing unit (graphics processing unit, GPU), a microprocessor (micro processor, MP), a digital signal processor (digital signal processor, DSP), an application specific integrated circuit (application specific integrated circuit, ASIC) or a field programmable gate array (field-programmable gate array, FPGA) and other processors. The present application does not limit the types of the memory 720 and the processor 730 and their working principles, and can be flexibly configured according to actual needs.

[0196] It should be understood that Figure 7The structure of the electronic device shown does not constitute a limitation on the electronic device in the embodiment of the present application. In practical applications, the electronic device may include Figure 7 More or less components as shown, or a combination of some components, especially when the electronic device is any one or more terminal devices such as a smart phone, a tablet computer, a wearable device, an augmented reality (AR) / virtual reality (VR) device, a laptop computer, an ultra-mobile personal computer (UMPC) or a robot, the electronic device may also include at least one input component such as a touch sensing unit for sensing touch events on a touch display panel, a keyboard, a mouse, a camera, a microphone, etc.; at least one output component such as a display, a speaker, a vibration mechanism, an indicator light, etc.; an antenna, a radio frequency unit; a variety of sensors; a power module, etc. The hardware structure can be determined according to the type of electronic device and its functional requirements, and this application does not list them one by one here.

[0197] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any cloud simulation cost prediction method provided in the embodiment of the present application.

[0198] The computer-readable storage medium may be any available medium that can be stored by a computer or a data storage device such as a training device or a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state disk (SSD)).

[0199] In addition, it should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines.

[0200] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course can also be implemented by dedicated hardware including dedicated integrated circuits, dedicated CPUs, dedicated memories, dedicated components, etc. That is to say, in the above embodiments, all or part of the implementation can be achieved by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product.

Claims

1. A cloud simulation cost prediction method, characterized in that: The method comprises: Acquire feature quantification information of a model to be simulated for a cloud simulation platform; the feature quantification information includes target feature values ​​of multi-dimensional features that affect the cloud simulation cost of the cloud simulation platform executing the model to be simulated, and the target feature values ​​of the multi-dimensional features respectively indicate a target calculation equation, material complexity, simulation calculation step, and model grid amount of the model to be simulated; Input the target feature value of the multi-dimensional feature into a trained decision tree regression model, and determine the target prediction value for the model to be simulated from a plurality of candidate prediction values ​​identified to match the target feature value of the multi-dimensional feature; each of the candidate prediction values ​​is a prediction value represented by an end node of a prediction path that matches the target feature value of the multi-dimensional feature in a corresponding decision tree in the decision tree regression model; The target prediction value is used to output a cloud simulation cost prediction result for the model to be simulated.

2. The method according to claim 1, characterized in that The step of obtaining the characteristic quantitative information of the model to be simulated on the cloud simulation platform includes: Obtain a model to be simulated for a cloud simulation platform; Obtaining a target equation identifier corresponding to a target calculation equation used by the model to be simulated; the target equation identifier is determined as a target characteristic value of a first dimensional characteristic, and the target calculation equation includes at least one characteristic value of a continuous medium mechanics equation; Determining the material complexity of the model to be simulated by using the material model complexity and the number of material properties of the model to be simulated; the material complexity is determined as the target feature value of the second dimensional feature; Determine the inverse of the simulation calculation step configured for the to-be-simulated model as the target feature value of the third dimensional feature; The model grid amount of the model to be simulated is scaled to obtain a target feature value of the fourth dimensional feature.

3. The method according to claim 2, characterized in that The determining the material complexity of the model to be simulated by using the material model complexity and the number of material properties of the model to be simulated includes: Determine the number of material properties corresponding to each material model included in the model to be simulated; Determine the target material stress-strain relationship described by each of the material models, and the target behavior model to which it belongs; the target behavior model is used to describe the mechanical behavior of the corresponding material under the action of external force; Querying the correspondence between different types of material stress-strain relationships and behavior models and different indicator identifiers, and obtaining a first indicator identifier corresponding to the target material stress-strain relationship and a second indicator identifier corresponding to the target behavior model; The material complexity of the model to be simulated is obtained by using the material property quantity, the first index identifier and the second index identifier.

4. The method according to claim 3, characterized in that in: Different types of material stress-strain relationships include linear relationships and nonlinear relationships, and the indicator identifier corresponding to the linear relationship is a first value, and the indicator identifier corresponding to the nonlinear relationship is a second value; Different types of behavior models include elastic models and plastic models, the indicator identifier corresponding to the configuration of the plastic model is the first value, and the indicator identifier corresponding to the configuration of the elastic model is the second value; Obtaining the material complexity of the model to be simulated by using the material attribute quantity, the first index identifier, and the second index identifier, including: The material property quantity, the first index identifier and the second index identifier are summed to obtain the material complexity of the corresponding material model in the model to be simulated.

5. The method according to claim 2, characterized in that: The step of obtaining a target equation identifier corresponding to a target calculation equation used by the model to be simulated includes: Obtaining a computational complexity classification relationship between equation identifiers of different computational equations and different types of simulation models; the computational complexity classification relationship is determined by a machine learning model based on historical data of the simulation model executed by the cloud simulation platform; The computational complexity classification relationship is queried to determine the target equation identifier corresponding to the model type to which the model to be simulated belongs.

6. The method according to claim 2, characterized in that The simulation calculation step size is configured based on the calculation accuracy requirement and calculation efficiency requirement of the model to be simulated; Scaling the model grid amount of the to-be-simulated model to obtain the characteristic value of the fourth dimension includes: Determine a reduction order of magnitude of the model grid amount for the model to be simulated; the reduction order of magnitude includes ten thousand levels; The ratio of the model grid volume to the reduction order of magnitude is determined as the eigenvalue of the fourth dimension.

7. The method according to any one of claims 1 to 6, characterized in that: The training process of the decision tree regression model includes: Acquire historical data of the simulation model executed by the cloud simulation platform; Determine, based on the historical data, various historical feature values ​​corresponding to different dimensional features, as well as historical simulation conditions and historical prediction values ​​corresponding to different historical feature values; Constructing a plurality of initial decision trees using the historical feature values ​​corresponding to the different dimensional features, the historical simulation conditions and the historical prediction values; For each internal node of each of the initial decision trees, by evaluating the information gain of features of different dimensions, a one-dimensional feature is selected from the internal node as a split feature; Obtaining the mean square error of two sub-nodes corresponding to each splitting point corresponding to the splitting feature; Taking the splitting point corresponding to the minimum mean square error as the target splitting point, and performing the splitting feature segmentation on the corresponding internal node; It is determined that each new decision tree meets a termination condition, and each new decision tree is determined as a decision tree regression model.

8. A cloud simulation cost prediction device, characterized in that: The device comprises: A feature quantification information acquisition module is used to acquire feature quantification information of a model to be simulated on a cloud simulation platform; the feature quantification information includes target feature values ​​of multi-dimensional features that affect the cloud simulation cost of the cloud simulation platform executing the model to be simulated, and the target feature values ​​of the multi-dimensional features respectively indicate a target calculation equation, material complexity, simulation calculation step size, and model grid amount of the model to be simulated; A prediction module, used for inputting the target feature value of the multi-dimensional feature into a trained decision tree regression model, and determining the target prediction value for the model to be simulated from a plurality of candidate prediction values ​​identified to match the target feature value of the multi-dimensional feature; each of the candidate prediction values ​​is a prediction value represented by an end node of a prediction path matching the target feature value of the multi-dimensional feature in a corresponding decision tree in the decision tree regression model; The output module is used to output the cloud simulation cost prediction result for the model to be simulated by using the target prediction value.

9. An electronic device, characterized in that: The electronic device comprises at least one communication element, at least one memory and at least one processor; The communication element is used to obtain the model to be simulated; The memory is used to store computer instructions; The processor is used to load and execute the computer instructions to implement the cloud simulation cost prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by an electronic device, the cloud simulation cost prediction method according to any one of claims 1 to 7 is implemented.

Images