Material growth simulation method, device and equipment and storage medium

Abstract

The invention provides a material growth simulation method, device and equipment and a storage medium, and relates to the technical field of simulation, and the method comprises the steps: calling material growth simulation software to execute simulation based on initial simulation parameters; calling material growth simulation software to execute simulation based on the initial simulation parameters; the initial simulation parameter is obtained by screening from a first simulation parameter interval input by a user according to a first model obtained by supervised learning, or is obtained by optimizing the first simulation parameter input by the user according to a second model obtained by reinforcement learning; when it is determined that simulation fails, the initial simulation parameters are optimized according to a second model, and target simulation parameters are obtained; and taking the target simulation parameter as an initial simulation parameter, and continuing to call the material growth simulation software to execute simulation based on the initial simulation parameter until it is determined that simulation succeeds. According to the method, automatic closed loop of simulation, verification and optimization is realized, so that the optimization efficiency of simulation parameters is improved.

Description

Technical Field

[0001] This invention relates to the field of simulation technology, and in particular to a material growth simulation method, apparatus, equipment, and storage medium. Background Technology

[0002] Materials growth simulation is a crucial part of modern materials research and development, widely used for performance prediction and process optimization of novel materials. However, traditional materials growth simulation processes often rely on researchers' personal experience for parameter settings. Researchers need to manually set a large number of simulation parameters, such as temperature, air pressure, and component concentration, based on theoretical knowledge and past experience.

[0003] This method is inefficient. After a simulation is completed, if the results do not meet the expected performance targets, researchers must intervene again, spending a lot of time analyzing the current results and manually adjusting the parameters, and then restarting the next simulation until an ideal set of parameters is found. Summary of the Invention

[0004] This invention provides a material growth simulation method, apparatus, equipment, and storage medium to solve the problem of low efficiency caused by the reliance on manual experience for parameter optimization in existing material growth simulation processes.

[0005] This invention provides a material growth simulation method, the method comprising: The material growth simulation software is invoked to perform the simulation based on the initial simulation parameters; the initial simulation parameters are obtained by filtering from the first simulation parameter range input by the user according to the first model obtained by supervised learning, or by optimizing the first simulation parameters input by the user according to the second model obtained by reinforcement learning. If the simulation fails, the initial simulation parameters are optimized based on the second model to obtain the target simulation parameters; The target simulation parameters are used as the initial simulation parameters, and the material growth simulation software is called again to perform simulation based on the initial simulation parameters until the simulation is confirmed to be successful.

[0006] In some embodiments, the step of invoking the material growth simulation software to perform the simulation based on initial simulation parameters includes: Obtain images of the parameter input box and the process start button from the material growth simulation startup interface of the material growth simulation software; Determine the first coordinates of the parameter input box image and the second coordinates of the process start button image; Based on the first coordinates, a first human-computer interaction operation is simulated in the material growth simulation startup interface; the first human-computer interaction operation is used to submit the initial simulation parameters to the parameter input box; Based on the second coordinate, a second human-computer interaction operation is simulated in the material growth simulation startup interface; the second human-computer interaction operation is used to start the simulation process.

[0007] In some embodiments, determining that the simulation failed includes: Based on the log files of the materials growth simulation software, monitor the execution status of the simulation tasks corresponding to the initial simulation parameters; If the execution state is abnormal, the simulation is determined to have failed; or, When the execution state is normal, obtain the simulation results output by the material growth simulation software; If the simulation results do not meet the preset simulation requirements, the simulation is determined to have failed.

[0008] In some embodiments, the simulation results are obtained in the following manner: Obtain the simulation result interface image after the material growth simulation software has finished running; Obtain the optical character recognition results of the simulation result interface image, and extract the simulation performance peak from the optical character recognition results according to a regular expression; Key points of the simulation curve are identified by performing key point identification on the characteristic curves of the material growth process in the simulation result interface image. Semantic segmentation is performed on the simulation heatmap in the simulation result interface image to obtain the target simulation heatmap region; The simulation results are obtained based on at least one of the simulation performance peak, the key points of the simulation curve, and the target simulation heatmap region.

[0009] In some embodiments, when the user input is a first simulation parameter range, the initial simulation parameters are obtained as follows: Multiple first candidate simulation parameters are sampled from the first simulation parameter range; Based on the first model, the first predicted simulation result corresponding to each of the first candidate simulation parameters is obtained; Based on the first predicted simulation results and preset simulation requirements, initial simulation parameters are selected from multiple first candidate simulation parameters.

[0010] In some embodiments, when the user input is a first simulation parameter, the initial simulation parameter is obtained as follows: The first simulation parameters are input into the second model to obtain the initial simulation parameters output by the second model; The second model was trained in the following way: Based on the policy function of the second model, determine the parameter adjustment amount corresponding to the training data of the first simulation parameters; The first simulation parameter training data is updated according to the parameter adjustment amount to obtain the updated second simulation parameter training data. Based on the first model, the second predicted simulation result corresponding to the training data of the second simulation parameters is obtained; Based on the second prediction simulation results and the reward function of the second model, determine the reward value corresponding to the training data of the second simulation parameters; The policy function of the second model is updated according to the reward value, and the training data of the second simulation parameters is used as the training data of the first simulation parameters. The process continues to execute the policy function of the second model to determine the parameter adjustment amount corresponding to the training data of the first simulation parameters until the iteration termination condition is met.

[0011] In some embodiments, the step of continuing to execute the parameter adjustment amount corresponding to the first simulation parameter training data according to the policy function of the second model until the iteration termination condition is met further includes: If it is determined that the parameter adjustment amount is less than the preset adjustment amount under a preset number of consecutive iterations before the current iteration round, and the parameter adjustment amount under the current iteration round is less than the preset adjustment amount, a penalty value under the current iteration round is generated; Update the reward value for the current iteration based on the penalty value for the current iteration.

[0012] In some embodiments, the simulation parameter training data for the first model and / or the second model are constructed in the following manner: Obtain the second simulation parameter range input by the user; Multiple second candidate simulation parameters are sampled from the second simulation parameter range; The material growth simulation software is invoked to perform simulations based on multiple second candidate simulation parameters, and the execution status of the simulation task corresponding to each second candidate simulation parameter is monitored according to the log file of the material growth simulation software. From a plurality of second candidate simulation parameters, a third candidate simulation parameter whose execution state is normal is selected to construct simulation parameter training data.

[0013] The present invention also provides a material growth simulation device, the device comprising: The first simulation unit is used to call the material growth simulation software to perform simulation based on the initial simulation parameters; the initial simulation parameters are obtained by filtering from the first simulation parameter range input by the user according to the first model obtained by supervised learning, or by optimizing the first simulation parameters input by the user according to the second model obtained by reinforcement learning. The simulation optimization unit is used to optimize the initial simulation parameters according to the second model to obtain the target simulation parameters when the simulation is determined to fail. The second simulation unit is used to take the target simulation parameters as the initial simulation parameters and continue to call the material growth simulation software to perform simulation based on the initial simulation parameters until the simulation is determined to be successful.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the material growth simulation method as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the material growth simulation method as described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the material growth simulation method as described above.

[0017] The material growth simulation method provided by this invention first uses a first model to filter the simulation parameters input by the user or uses a second model to perform preliminary optimization of the simulation parameters input by the user, thereby quickly determining effective initial simulation parameters and avoiding the blindness of traditional manual trial and error. On this basis, the second model is used to continuously optimize the parameters and iteratively verify them, realizing an automatic closed loop of simulation, verification and optimization, thereby improving the optimization efficiency of simulation parameters. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is one of the flowcharts of the material growth simulation method provided by the present invention.

[0020] Figure 2 This is the second flowchart of the material growth simulation method provided by the present invention.

[0021] Figure 3 This is a schematic diagram of the material growth simulation device provided by the present invention.

[0022] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] Materials growth simulation is a crucial part of modern materials research and development. It mainly uses core technologies such as molecular dynamics simulation and first-principles calculations to predict the performance of new materials and optimize their processes, providing precise digital support for the research and development process. It is a core technological means to improve research and development efficiency.

[0025] Current mainstream simulation workflows rely on expert experience, with parameter settings depending on the judgment of R&D personnel or simple linear regression analysis based on limited experimental data. At the execution level, simulation software calls are completed through simple scripts or manual operation. This model has a foundation for application in the early development of simple material systems, but its inherent shortcomings have gradually become apparent as R&D needs have evolved.

[0026] On the one hand, manual trial-and-error parameter optimization is time-consuming and inefficient, failing to meet the demands of high-throughput material screening, especially in complex material systems where R&D efficiency constraints are particularly pronounced. On the other hand, unreasonable parameter settings directly lead to a waste of computational resources, disrupt the R&D iteration rhythm, and may even generate misleading results, affecting the reliability and safety of product design and causing losses in human and material resources. Furthermore, in existing technologies, simulation operation and manual intervention are asynchronous. When simulation interruptions, anomalies, or unsatisfactory results occur, the software is often in a stagnant or erroneous state while manual parameter optimization is being conducted, resulting in significant waste of time and resources.

[0027] To address the above problems, embodiments of the present invention provide a material growth simulation method. Figure 1 This is one of the flowcharts illustrating the material growth simulation method provided by this invention, such as... Figure 1 As shown, the method includes the following steps 110, 120 and 130.

[0028] Step 110: Call the material growth simulation software to perform simulation based on the initial simulation parameters; the initial simulation parameters are obtained by filtering from the first simulation parameter range input by the user according to the first model obtained by supervised learning, or by optimizing the first simulation parameters input by the user according to the second model obtained by reinforcement learning.

[0029] Materials growth simulation software refers to computational software used to simulate the growth process of materials at the atomic or molecular scale. Materials growth simulation software can be based on molecular dynamics principles, first-principles calculations, or other computer-aided engineering software used to simulate the growth processes of specific material systems, such as semiconductors, metals, and ceramics; there are no restrictions on the types.

[0030] To automate the simulation process, this embodiment can employ a method that simulates manual operation. For example, a pre-built image template library of various function buttons in the software interface of the material growth simulation software can be constructed, such as image templates for buttons like "Start Simulation," "Modeling," and "Submit." When the material growth simulation software is automatically invoked to execute the simulation process, image matching technology, such as template matching functions, is used to locate the image positions of these function buttons in real time within the current software interface, simulating human-computer interaction operations such as mouse operation, keyboard operation, and touch operation to complete the automated simulation process.

[0031] In this step, the initial simulation parameters refer to a set of specific values ​​used to initiate the first or a new round of simulation calculations. These parameters collectively define the simulation environment and process conditions for material growth. The specific parameter types are related to the material growth being simulated. For example, in chemical vapor deposition simulations, the initial simulation parameters may include, but are not limited to, growth temperature, chamber pressure, and substrate material type.

[0032] In one example, the initial simulation parameters can be obtained by filtering from the range of first simulation parameters input by the user based on the first model.

[0033] Here, the first simulation parameter range refers to a user-defined multidimensional parameter space, where each dimension corresponds to a simulation parameter and is defined by an upper and lower limit. For example, the user can input a temperature range of [100℃, 300℃] and a pressure range of [200MPa, 500MPa].

[0034] The pre-trained first model is typically obtained through supervised learning training based on a large amount of historical simulation data and the corresponding simulation results. As a specific example, a deep feedforward neural network can be constructed, where the number of neurons in the input layer corresponds to the dimension of the simulation parameters. For instance, when the simulation parameters are growth temperature, chamber pressure, and substrate material type, the number of neurons in the input layer is 3. Similarly, the number of neurons in the output layer corresponds to the dimension of the simulation results. The hidden layers can adopt a progressively dimensionality-reduced funnel structure, such as 128 neurons in the first hidden layer, 64 neurons in the second, 32 neurons in the third, and 16 neurons in the fourth, using nonlinear activation functions such as Tanh to fit the complex nonlinear relationship between simulation parameters and simulation results. Through training, this neural network model can predict the simulation results generated by any given simulation parameters.

[0035] In this step, the system generates a large number of discrete parameter points within a user-defined first simulation parameter range, and then calls the first model to predict the corresponding simulation results for each of these discrete parameter points. Finally, the system selects the simulation result that is closest to the target simulation result or the optimal simulation result from these simulation results, and uses its corresponding parameter combination as the initial simulation parameters.

[0036] In one example, the initial simulation parameters could also be obtained by optimizing a first simulation parameter input by the user using a second model.

[0037] Here, the user inputs a specific set of parameter values, not a range. For example, the temperature is 250℃ and the air pressure is 350MPa.

[0038] It should be understood that the pre-trained second model is trained through reinforcement learning to perform a decision-making task: automatically and strategically generating a set of optimized parameters that are more likely to achieve the goal when the simulation results do not meet the requirements. This model learns and executes an optimal parameter adjustment strategy through interaction with the materials growth simulation software.

[0039] In this embodiment, the state of the second model represents the environmental information upon which the model bases its decisions. The state can include any data that describes the current simulation situation. In one specific embodiment, the state is a vector combination that includes at least: the current simulation parameters, for example, an n-dimensional vector containing values ​​such as temperature, pressure, and flow rate; and the simulation results obtained from these parameters, for example, one or more scalar values ​​characterizing material properties.

[0040] It should be understood that the action of the second model represents the decision instruction output by the model based on the current state, i.e., the specific adjustment amount to the simulation parameters. In one embodiment, the action space can be continuous, meaning the model can output adjustment values ​​with arbitrary precision. In another embodiment, the action space can also be discrete, meaning the model selects from a set of preset adjustment step sizes.

[0041] Furthermore, in one embodiment, the reward function can be a multi-level reward. For example, if the optimization objective is to minimize a certain metric value U, and the preset simulation requirement is U < U_target, then the reward function R can be designed as follows: when a new simulation result U_new < U_target, a large positive reward is given, indicating that the objective has been successfully achieved. When U_new < U_previous, where U_previous is the result of the previous step, i.e., the objective has not yet been achieved, a smaller positive reward is given to incentivize the model to optimize in the correct direction. When U_new ≥ U_previous, a negative reward or zero reward is given, for example, R = -1, to penalize actions that lead to a deterioration in the result. In this way, a hierarchical reward mechanism can effectively guide the model to converge quickly.

[0042] As a specific implementation, the second model can be trained using one or more reinforcement learning algorithms. For example, it can employ the Proximal Policy Optimization (PPO) algorithm, which is suitable for handling continuous action spaces, or the Deep Deterministic Policy Gradient Algorithm. Through extensive offline or online training, the second model learns a mapping policy from states to optimal actions, thereby guiding the iterative optimization of parameters.

[0043] In this step, the system inputs the first simulation parameters from the user as the initial state into the second model. The second model outputs one or more actions based on the strategy it has learned, thereby obtaining a set of optimized parameters, i.e., the initial simulation parameters.

[0044] After obtaining the initial simulation parameters, the system will automatically call the simulation software to execute and automatically extract the simulation results after the simulation ends.

[0045] It should be understood that simulation results refer to the data output by the simulation software after execution, used to characterize the various properties of the grown virtual material. In practical applications, simulation results can be presented as one or more scalar values, such as the average thickness of the film, average stress value, and total number of defects. They can also be a two-dimensional or three-dimensional data field, such as a thermogram formed on the substrate surface, visually displaying thickness distribution, temperature distribution, component concentration distribution, or defect density distribution. Furthermore, they can be a comprehensive evaluation index, i.e., a comprehensive score calculated by weighting multiple performance indicators.

[0046] Step 120: If the simulation fails, optimize the initial simulation parameters according to the second model to obtain the target simulation parameters.

[0047] It should be noted that simulation failure refers to the inability to complete the simulation process normally when executing the simulation based on the initial simulation parameters, or the simulation results not meeting the preset simulation requirements after completion. For example, the simulation program may crash due to the divergence of the initial simulation parameters, the simulation results may show non-physical values ​​(such as negative concentration), the calculation process of the simulation results may not converge, or although the simulation is completed, the simulation results (such as growth morphology and thickness) may differ greatly from the preset physical targets.

[0048] Here, preset simulation requirements refer to one or a set of specific, quantifiable objectives preset by the user. They serve as the standard for evaluating whether the simulation results are acceptable and are also the termination condition for the entire optimization process. For example, preset simulation requirements can be a single threshold requirement, such as film thickness non-uniformity less than 1% or film average stress less than 50 MPa; preset simulation requirements can also be multiple objective requirements, which can be flexibly set according to actual needs.

[0049] In this embodiment, the system includes a judgment module to monitor the output and simulation status of the material growth simulation software to determine whether the simulation has failed. When a simulation failure is determined, the system automatically provides the initial simulation parameters that caused the failure as input to the second model. The second model then outputs a set of adjusted target simulation parameters based on its learned strategies.

[0050] Step 130: Using the target simulation parameters as the initial simulation parameters, continue to call the material growth simulation software to perform simulation based on the initial simulation parameters until the simulation is confirmed to be successful.

[0051] Specifically, the system treats the target simulation parameters output by the second model as new initial simulation parameters, then calls the simulation software again to perform the simulation based on these new initial parameters and obtains new simulation results. Subsequently, the system continues to monitor the output and simulation status of the material growth simulation software to determine if the simulation has failed. If the simulation still fails, the above optimization and simulation cycle is repeated. Once an iteration of the simulation is successful, the cycle stops, and the system outputs the current set of simulation parameters as the final high-quality result.

[0052] The material growth simulation method provided in this invention first uses a first model to filter the simulation parameters input by the user or uses a second model to perform preliminary optimization of the simulation parameters input by the user, thereby quickly determining effective initial simulation parameters and avoiding the blindness of traditional manual trial and error. On this basis, the second model obtained by reinforcement learning is used to continuously optimize the parameters and iteratively verify them, realizing an automatic closed loop of simulation, verification and optimization, thereby improving the optimization efficiency of simulation parameters.

[0053] In some embodiments, determining that the simulation failed includes: Based on the log files of the materials growth simulation software, monitor the execution status of the simulation tasks corresponding to the initial simulation parameters; If the execution state is abnormal, the simulation is determined to have failed; or, When the execution state is normal, obtain the simulation results output by the material growth simulation software; If the simulation results do not meet the preset simulation requirements, the simulation is determined to have failed.

[0054] Here, log files refer to one or more text files continuously written by the material growth simulation software during its operation. The log files record in detail the convergence status, resource usage, warning messages, and critical error messages for each step of the simulation. The execution status is a label assigned in real-time by the monitoring program based on the log file content, describing the current stage or condition of the simulation task.

[0055] In one implementation, the execution status can be simply categorized as running, completed, and error-prone. In another implementation, the execution status can be further refined, for example, initialization, mesh generation, solver iteration, computation convergence, computation divergence, insufficient memory, and successful completion.

[0056] In this embodiment, a background monitoring script is pre-configured. This script reads the latest content of the log file at preset time intervals. By using string matching or regular expressions, the script can search for specific keywords to determine the current execution status.

[0057] It should be understood that an abnormal state refers to an execution state in which the characterization simulation cannot be successfully completed or cannot obtain meaningful results.

[0058] In this step, once the monitoring script detects a signal in the log file that triggers an abnormal state, it will immediately execute a preset termination procedure. This termination procedure will forcibly terminate the corresponding simulation software process via operating system commands.

[0059] Here, preset simulation requirements refer to one or a set of specific, quantifiable objectives preset by the user. They serve as the standard for evaluating whether the simulation results are acceptable and are also the termination condition for the entire optimization process. For example, preset simulation requirements can be a single threshold requirement, such as film thickness non-uniformity less than 1% or film average stress less than 50 MPa; preset simulation requirements can also be multiple objective requirements, which can be flexibly set according to actual needs.

[0060] In this step, if the monitoring script does not detect any signals triggering abnormal states in the log file, it will further analyze the simulation results output by the material growth simulation software. The simulation results are a series of data describing the final state of material growth, such as two-dimensional or three-dimensional data fields representing film thickness distribution, component concentration distribution, stress distribution, etc. Next, the simulation results are compared with preset simulation requirements. If the simulation results do not meet the preset requirements, the system also determines that the simulation has failed. Examples include: the calculated film thickness or component concentration non-uniformity is higher than the preset upper limit; the calculated average growth rate is much lower or much higher than the expected physical range; or negative concentration or thickness values ​​appear in certain regions.

[0061] The material growth simulation method provided in this invention can promptly detect and terminate invalid simulation tasks by monitoring the log files and simulation results of the simulation tasks in real time, thereby avoiding the waste of resources in invalid simulation tasks.

[0062] In some embodiments, reference Figure 2 , Figure 2 This is the second flowchart of the material growth simulation method provided by the present invention, as shown below. Figure 2 As shown, the method includes the following steps 210, 220, 230, 240 and 250.

[0063] Step 210: Obtain the simulation result interface image after the material growth simulation software has finished running.

[0064] It should be understood that the simulation result interface image refers to a screenshot or rendered image displayed on the graphical user interface of the material growth simulation software after completing a calculation task, containing simulation result information. In one example, this includes, but is not limited to, a screenshot of the entire computer screen, a screenshot of a specific application window, or a static image generated by the simulation software through its image export function.

[0065] In one example, you can use a screenshot tool or the built-in result export function of the material growth simulation software. After the simulation software finishes running and the result interface is fully displayed, a screenshot will be automatically triggered to obtain the simulation result interface image.

[0066] Specifically, after the simulation execution phase is completed, the system's built-in monitoring module will detect that the simulation software's running status has changed to "completed" or "stopped." At this time, the control module immediately triggers a screenshot command to capture the complete image of the current simulation software window and save it as a digital image file in a specified memory buffer or temporary file for subsequent processing.

[0067] Step 220: Obtain the optical character recognition result of the simulation result interface image, and extract the simulation performance peak from the optical character recognition result according to the regular expression; Here, peak simulation performance refers to the maximum value reached by the core performance indicators during the simulation process, such as the maximum value reached by core performance indicators like material crystallinity, film uniformity, and growth rate.

[0068] In this step, the acquired simulation result interface image can be input into an optical character recognition (OCR) engine. The OCR engine analyzes the entire image and outputs the original text string containing all the recognized characters, i.e., the OCR result. Examples include menu items, axis labels, and result values.

[0069] It should be noted that since the optical character recognition results contain a large number of irrelevant strings, this embodiment uses one or more preset regular expressions to parse the optical character recognition results, and then extracts the specific values ​​and units of the corresponding simulation performance peaks from the optical character recognition results and saves them as structured data.

[0070] Step 230: Identify the key points of the material growth process characteristic curve in the simulation result interface image to obtain the key points of the simulation curve.

[0071] Here, the characteristic curve of the material growth process refers to the curve displayed in graphical form on the simulation results interface, which describes the change of a physical quantity with another variable. Examples include the curve of substrate temperature changing with time, the curve of film thickness changing with wafer radius, or the curve of growth rate changing with reaction time. The key points of the simulation curve are the critical points on the characteristic curve of the material growth process, including but not limited to the start point, end point, inflection point, peak point, and valley point. These key points reflect the crucial changes in the material growth process.

[0072] In one example, the characteristic curve region of the material growth process in the simulation results interface image can be cropped. For instance, the chart area containing the characteristic curve can be located within the simulation results interface image. Within this area, image processing techniques, such as color thresholding and edge detection, can be used to extract the pixel trajectory of the curve. This pixel trajectory data is then analyzed. For example, local extrema and inflection points can be located by calculating the first and second derivatives. Alternatively, the coordinates of points intersecting with a preset horizontal line can be found by scanning pixel coordinates. Finally, the coordinates of each key point are output and associated with the corresponding numerical values, thus completing the extraction of key points from the simulation curve.

[0073] Step 240: Perform semantic segmentation on the simulation heatmap in the simulation result interface image to obtain the target simulation heatmap region.

[0074] A simulated heatmap is a graph that visualizes the distribution of a physical quantity, such as temperature, thickness, stress, or component concentration, in two-dimensional space using color gradients. Typically, red represents high values ​​and blue represents low values. A target simulated heatmap region refers to the area within the simulated heatmap that meets specific pre-defined conditions; for example, an overheated area where the temperature exceeds a certain threshold, or an excessively thin area where the film thickness is below design specifications.

[0075] In this embodiment, a semantic segmentation algorithm is used to perform semantic segmentation on the simulation heatmap in the simulation result interface image. Here, the semantic segmentation algorithm is used to assign a category label to each pixel in the image. In one specific implementation, a semantic segmentation model, such as the U-Net network model, is pre-trained to recognize the physical meaning corresponding to different colors in the heatmap. Based on the pre-trained semantic segmentation model, the simulation heatmap is classified at the pixel level to identify the specific region of interest to the user, i.e., the target simulation heatmap region.

[0076] Step 250: Obtain simulation results based on at least one of the simulation performance peak, the key points of the simulation curve, and the target simulation heatmap region.

[0077] Finally, the quantitative information extracted by one or more of the above methods, namely the peak performance of the simulation, the key points of the simulation curve, and the geometric features of the target simulation heatmap region, are combined to form a structured numerical value as the simulation result.

[0078] The material growth simulation method provided in this invention automatically reads and parses key simulation result data from the user graphical interface of the simulation software using machine vision and text recognition. In addition, by comprehensively applying multiple visual algorithms such as optical character recognition, key point recognition, and semantic segmentation, it achieves accurate extraction and quantification of different types of simulation results.

[0079] In some embodiments, the step of invoking the material growth simulation software to perform the simulation based on initial simulation parameters includes: Obtain images of the parameter input box and the process start button from the material growth simulation startup interface of the material growth simulation software; Determine the first coordinates of the parameter input box image and the second coordinates of the process start button image; Based on the first coordinates, a first human-computer interaction operation is simulated in the material growth simulation startup interface; the first human-computer interaction operation is used to submit the initial simulation parameters to the parameter input box; Based on the second coordinate, a second human-computer interaction operation is simulated in the material growth simulation startup interface; the second human-computer interaction operation is used to start the simulation process.

[0080] Here, the material growth simulation startup interface refers to the graphical user interface window or panel in the simulation software used to set the initial simulation conditions, such as temperature, pressure, gas flow rate, etc.

[0081] The parameter input box image is a visual representation of various controls in the simulation software interface used to receive user input parameters. In this embodiment, the parameter input box image not only refers to a screenshot of a text input box, but can also refer to the image of any interface element used for parameter setting, such as drop-down menus, slider controls, checkboxes, etc. In a specific implementation, images of these parameter input controls on the screen can be manually captured in advance and saved as template images.

[0082] The process start button image refers to the visual representation of the button control used to trigger the start of the simulation calculation. Typically, the process start button image is a button image labeled with words such as Compute, Run, Start, or Start Calculation. In specific implementations, images labeled with these words can be manually captured beforehand and saved as template images.

[0083] In one implementation, the system can also automatically configure template images by combining image feature recognition and optical character recognition technologies. For example, the system can scan the interface to find areas that conform to a specific shape, such as a rectangle, and have text labels in adjacent locations, and define a template image corresponding to this area based on the text label.

[0084] In this embodiment, the system's automated program first captures a screenshot of the entire material growth simulation software's interface. Then, it uses pre-saved parameter input box images and process start button images as templates to perform a sliding scan and comparison on the screen image. When the similarity between the template image and a certain region in the screen image reaches its maximum or exceeds a preset threshold, the coordinates of that region are determined as the target coordinates, i.e., the first coordinate or the second coordinate.

[0085] Here, the first and second coordinates refer to the pixel positions of the template image in the current real-time screenshot. These coordinates can be precise pixel coordinates, such as the top-left corner, center, or other predefined locations, or they can be a range of coordinates for a rectangular area; there are no restrictions on this.

[0086] After locating the first coordinates, the program code controls the movement and clicking behavior of the system's mouse cursor to simulate the user's first human-computer interaction operation of submitting initial simulation parameters to the parameter input box. This first human-computer interaction operation refers to the exchange of information, transmission of instructions, and task collaboration between the user and the system through specific methods, including but not limited to: keyboard input, touchscreen input, voice input, mouse input, and gesture input, without restriction.

[0087] In one example, the program first calls the operating system API or automation library to move the mouse cursor to the first coordinate determined in the previous step, and then simulates a left mouse click to activate the parameter input box. Next, the program simulates keyboard input, typing the corresponding values ​​from the initial simulation parameters character by character into the input box. For multiple parameters, this process is repeated, that is, locating each parameter input box and filling in the corresponding value in turn.

[0088] Similarly, after locating the second coordinates, the program code controls the movement and clicking behavior of the system's mouse cursor to simulate the second human-computer interaction operation that initiates the simulation process. This second human-computer interaction operation refers to the exchange of information, transmission of instructions, and task collaboration between the human and the system through specific methods, including but not limited to: keyboard input, touchscreen input, voice input, mouse input, and gesture input, without restriction.

[0089] In one example, after all parameters have been successfully entered via the first mouse operation, the program will move the mouse cursor to the second coordinate, i.e., the location of the process start button, and simulate a left mouse click. This click triggers the simulation software to begin calculations. Afterward, the automated program can enter a waiting state until the simulation is detected as complete.

[0090] The material growth simulation method provided in this embodiment of the invention achieves an automated simulation process by employing an automated technology based on image template matching.

[0091] In some embodiments, when the user input is a first simulation parameter range, the initial simulation parameters are obtained as follows: Multiple first candidate simulation parameters are sampled from the first simulation parameter range; Based on the first model, the first predicted simulation result corresponding to each of the first candidate simulation parameters is obtained; Based on the first predicted simulation results and preset simulation requirements, initial simulation parameters are selected from multiple first candidate simulation parameters.

[0092] It should be understood that the first simulation parameter interval refers to a user-defined multidimensional parameter space, where each dimension corresponds to a simulation parameter and is defined by an upper and lower bound. In this embodiment, a large number of discrete parameter combinations are generated by sampling from the multidimensional parameter interval; these combinations are referred to as the first candidate simulation parameters.

[0093] In practical applications, sampling methods can be varied. For example, grid sampling can be used, which divides each parameter dimension into several equal parts according to a preset sampling step size, and then takes the parameter combination at all grid points. Random sampling or Latin hypercube sampling can also be used to achieve more uniform coverage throughout the entire interval, and there are no restrictions on this.

[0094] As mentioned above, the first model is a pre-trained model using a large number of simulation parameters and corresponding data pairs of real simulation results. Here, for each first candidate simulation parameter, the first model outputs a predicted simulation result to replace the time-consuming physical simulation.

[0095] Then, the system iterates through all the first candidate simulation parameters and their corresponding first predicted simulation results. These predicted results are then compared with preset simulation requirements to select the initial simulation parameters that meet the requirements from all the first candidate simulation parameters.

[0096] Here, preset simulation requirements refer to one or a set of specific, quantifiable objectives preset by the user. They serve as the standard for evaluating whether the simulation results are acceptable and are also the termination condition for the entire optimization process. For example, preset simulation requirements can be a single threshold requirement, such as film thickness non-uniformity less than 1% or film average stress less than 50 MPa; preset simulation requirements can also be multiple objective requirements, which can be flexibly set according to actual needs.

[0097] For example, if the preset simulation requirement is to minimize non-uniformity, the system will select the first candidate simulation parameter that results in the lowest predicted non-uniformity value as the initial simulation parameter. If the preset simulation requirement is to maximize the growth rate under a stress of less than 100 MPa, the system will first eliminate all first candidate simulation parameters whose predicted stress exceeds 100 MPa, and then select the one with the highest predicted growth rate from the remaining first candidate simulation parameters as the initial simulation parameter. If multiple objectives are involved, the system can perform a weighted summation of the prediction results according to preset weights to obtain a comprehensive score, and select the one with the highest score as the initial simulation parameter.

[0098] The material growth simulation method provided in this invention achieves efficient parameter optimization by combining parameter space sampling and fast neural network prediction, thereby increasing the likelihood of approaching or reaching the target in the first simulation.

[0099] In some embodiments, when the user input is a first simulation parameter, the initial simulation parameter is obtained as follows: The first simulation parameters are input into the second model to obtain the initial simulation parameters output by the second model; The second model was trained in the following way: Based on the policy function of the second model, determine the parameter adjustment amount corresponding to the training data of the first simulation parameters; The first simulation parameter training data is updated according to the parameter adjustment amount to obtain the updated second simulation parameter training data. Based on the first model, the second predicted simulation result corresponding to the training data of the second simulation parameters is obtained; Based on the second prediction simulation results and the reward function of the second model, determine the reward value corresponding to the training data of the second simulation parameters; The policy function of the second model is updated according to the reward value, and the training data of the second simulation parameters is used as the training data of the first simulation parameters. The process continues to execute the policy function of the second model to determine the parameter adjustment amount corresponding to the training data of the first simulation parameters until the iteration termination condition is met.

[0100] Here, the first simulation parameter training data refers to the simulation parameter training data used as the starting point for optimization in the current iteration. For example, in the first iteration, the first simulation parameter training data is the simulation parameter training data where the simulation results do not meet the preset simulation requirements. In subsequent iterations, it becomes the simulation parameter training data updated in the previous iteration.

[0101] The policy function of the second model is a core component of the reinforcement learning model. Based on the current state—the training data of the first simulation parameters and the associated simulation results—it outputs an action. In this embodiment of the invention, this action is a parameter adjustment. For example, the policy function might output an action that increases the temperature parameter by 5°C and decreases the pressure parameter by 10 MPa.

[0102] In one example, the policy function can be any function that can implement a mapping from state to action, such as a table-based, linear function-based, or deep neural network-based policy, whose input is a parameter vector and output is an adjustment vector.

[0103] In one specific embodiment, the parameter adjustment amount can be a set of incremental values, a new set of absolute values, or a probability distribution from which the second model can sample a specific adjustment amount.

[0104] In this embodiment, after obtaining the parameter adjustment amount, the system applies the parameter adjustment amount obtained in the previous step to the first simulation parameter training data to obtain the second simulation parameter training data. Then, the first model is used to replace the simulation calculation process of calling the material growth simulation software, and the second predicted simulation result corresponding to the second simulation parameter training data output by the first model is directly obtained.

[0105] It should be understood that the reward function is used to evaluate the quality of an action. The system compares the second prediction simulation result predicted by the first model with the preset simulation requirements and calculates a reward value according to the preset reward function. For example, if the prediction result is closer to the preset simulation requirements than the previous round, a positive reward is calculated according to the preset reward function; otherwise, a negative reward is calculated according to the preset reward function.

[0106] The second model then uses this reward value to adjust the internal parameters of its policy function, such as the weights of the neural network. Typically, if the reward is high, the policy is adjusted to increase the probability of taking that action in a similar situation in the future; conversely, if the reward is low (or penalized), the probability of taking that action is reduced. This allows the second model to learn from experience, and its decision-making ability evolves iteratively, becoming increasingly inclined to make decisions that yield high rewards.

[0107] The system takes the second simulation parameter to be optimized obtained at the moment as the first simulation parameter to be optimized in the next round of internal iteration, and then returns to start a new round of internal iteration until the iteration termination condition is met, such as reaching the maximum number of iterations, the reward value converging, or obtaining a predicted simulation result that meets the preset simulation conditions.

[0108] The material growth simulation method provided in this invention embeds a first model into the decision loop of a second model, enabling the second model to perform a large number of virtual simulations at extremely low cost before making a final decision, which greatly accelerates its learning and convergence speed and improves the efficiency of parameter optimization.

[0109] In some embodiments, the step of continuing to execute the parameter adjustment amount corresponding to the first simulation parameter training data according to the policy function of the second model until the iteration termination condition is met further includes: If it is determined that the parameter adjustment amount is less than the preset adjustment amount under a preset number of consecutive iterations before the current iteration round, and the parameter adjustment amount under the current iteration round is less than the preset adjustment amount, a penalty value under the current iteration round is generated; Update the reward value for the current iteration based on the penalty value for the current iteration.

[0110] It should be understood that in each internal iteration of the model, the system records the magnitude of the parameter adjustment given by the policy function. In the current internal iteration round, the system first calculates the current parameter adjustment magnitude. Then, the system checks whether the parameter adjustment magnitudes in the most recent N times in the historical record are all less than the preset adjustment magnitude, and whether the parameter adjustment magnitude calculated this time is also less than the preset adjustment magnitude. If this condition is met, it indicates that the second model has made small adjustments in the parameter space for (N+1) consecutive times, and may have fallen into a local optimum. At this point, the system triggers a penalty mechanism.

[0111] Specifically, the system generates a negative value as a penalty. This penalty value can be a fixed value, such as a preset negative constant; or it can be a dynamic value, where the magnitude of the penalty value is related to the severity of the stagnation. For example, the penalty value can be set to P = -C * number of consecutive stagnations, where C is a preset conventional penalty coefficient.

[0112] Finally, this generated penalty value is applied to the reward value of the current iteration, for example, by simply adding them together. This updated reward value then provides negative feedback to the second model, incentivizing its policy function to attempt a larger parameter adjustment in the next iteration.

[0113] The material growth simulation method provided in this invention improves the probability of the algorithm finding the global optimal solution by imposing a penalty on continuously unchanged parameters, thereby improving the accuracy of the simulation parameters after the second model is optimized.

[0114] In some embodiments, the simulation parameter training data for the first model and / or the second model are constructed in the following manner: Obtain the second simulation parameter range input by the user; Multiple second candidate simulation parameters are sampled from the second simulation parameter range; The material growth simulation software is invoked to perform simulations based on multiple second candidate simulation parameters, and the execution status of the simulation task corresponding to each second candidate simulation parameter is monitored according to the log file of the material growth simulation software. From a plurality of second candidate simulation parameters, a third candidate simulation parameter whose execution state is normal is selected to construct simulation parameter training data.

[0115] Here, the second simulation parameter range is a pre-defined range of parameters used for data acquisition.

[0116] Typically, the range of the second simulation parameter range is larger than that of the first simulation parameter range, so as to cover as comprehensively as possible the various process conditions that may affect the material growth results, thereby giving the trained model better generalization ability.

[0117] In this step, the system generates a large number of specific parameter combinations within the second simulation parameter range. In specific implementations, methods such as random sampling or uniform grid sampling can be used to sample the second candidate simulation parameters from the second simulation parameter range, or Latin hypercube sampling can be used to sample the second candidate simulation parameters from the second simulation parameter range; there are no restrictions on this.

[0118] After generating multiple second candidate simulation parameters, the system uses an automated script to submit each of these parameters, generated in the previous step, as input to the materials growth simulation software for batch calculation tasks. It should be understood that not all parameter combinations will allow the simulation software to run normally and produce meaningful results. Certain extreme or unreasonable parameter combinations may lead to numerical calculation non-convergence, program crashes, or memory overflows. Therefore, this embodiment monitors the execution status of each simulation task by parsing the log files generated by the materials growth simulation software during operation. Specifically, the execution status includes, but is not limited to: Normal state: The simulation task completes all calculation steps within the preset time, no fatal errors are reported, and the result file is successfully generated; Error exit: The simulation task terminates prematurely due to internal errors, such as mesh errors or physical model setting conflicts; Computational divergence: During the iterative computation process, the solution of the simulation task fails to converge to a stable value, resulting in computational failure. Timeout: The simulation task has exceeded a preset threshold, such as 24 hours.

[0119] The system iterates through all second-candidate simulation parameters and their corresponding execution states, retaining only the parameter combinations whose execution state is normal. These retained parameter combinations that have successfully completed the simulation constitute the third-candidate simulation parameters, thus effectively eliminating all invalid and unstable data.

[0120] Next, for each set of third candidate simulation parameters, the system automatically parses its corresponding simulation output file, extracts key simulation results, such as thin film thickness distribution data, and calculates its non-uniformity percentage. Based on this, a paired dataset containing valid simulation parameters and their simulation results can be obtained.

[0121] Furthermore, to facilitate supervised learning of the subsequent model, the system compares the simulation results of each third candidate simulation parameter with the user-defined preset simulation requirements. If the simulation results meet the preset requirements, the corresponding simulation parameters and their simulation results are marked as positive sample data; if they do not meet the preset requirements, they are marked as negative sample data, thus constructing the simulation parameter training data.

[0122] Furthermore, in one embodiment, each simulation parameter and its simulation results can be standardized to eliminate the influence of differences in dimensions between different parameters, reduce the interference of extreme data on model training, and ensure the stable convergence of the model.

[0123] The material growth simulation method provided in this invention ensures the effectiveness and reliability of training data by automatically executing a large number of simulations and combining log monitoring for data filtering, thereby improving the effectiveness and reliability of subsequent model prediction results.

[0124] Based on any of the above embodiments, the present invention also provides a material growth simulation device, with reference to... Figure 3 The device includes: a first simulation unit 310, a simulation optimization unit 320, and a second simulation unit 330.

[0125] The first simulation unit 310 is used to call the material growth simulation software to perform simulation based on the initial simulation parameters; the initial simulation parameters are obtained by filtering from the first simulation parameter range input by the user according to the first model obtained by supervised learning, or by optimizing the first simulation parameters input by the user according to the second model obtained by reinforcement learning. The simulation optimization unit 320 is used to optimize the initial simulation parameters according to the second model to obtain the target simulation parameters when the simulation is determined to fail. The second simulation unit 330 is used to take the target simulation parameters as the initial simulation parameters and continue to call the material growth simulation software to perform simulation based on the initial simulation parameters until the simulation is determined to be successful.

[0126] The material growth simulation device provided by this invention first uses a first model to filter the simulation parameters input by the user or uses a second model to perform preliminary optimization of the simulation parameters input by the user, thereby quickly determining effective initial simulation parameters and avoiding the blindness of traditional manual trial and error. On this basis, the second model is used to continuously optimize the parameters and perform iterative verification, realizing an automatic closed loop of simulation, verification and optimization, thereby improving the optimization efficiency of simulation parameters.

[0127] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a material growth simulation method, the method including: The material growth simulation software is invoked to perform the simulation based on the initial simulation parameters; the initial simulation parameters are obtained by filtering from the first simulation parameter range input by the user according to the first model obtained by supervised learning, or by optimizing the first simulation parameters input by the user according to the second model obtained by reinforcement learning. If the simulation fails, the initial simulation parameters are optimized based on the second model to obtain the target simulation parameters; The target simulation parameters are used as the initial simulation parameters, and the material growth simulation software is called again to perform simulation based on the initial simulation parameters until the simulation is confirmed to be successful.

[0128] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0129] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the material growth simulation method provided by the above methods, the method comprising: The material growth simulation software is invoked to perform the simulation based on the initial simulation parameters; the initial simulation parameters are obtained by filtering from the first simulation parameter range input by the user according to the first model obtained by supervised learning, or by optimizing the first simulation parameters input by the user according to the second model obtained by reinforcement learning. If the simulation fails, the initial simulation parameters are optimized based on the second model to obtain the target simulation parameters; The target simulation parameters are used as the initial simulation parameters, and the material growth simulation software is called again to perform simulation based on the initial simulation parameters until the simulation is confirmed to be successful.

[0130] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the material growth simulation method provided by the methods described above, the method comprising: The material growth simulation software is invoked to perform the simulation based on the initial simulation parameters; the initial simulation parameters are obtained by filtering from the first simulation parameter range input by the user according to the first model obtained by supervised learning, or by optimizing the first simulation parameters input by the user according to the second model obtained by reinforcement learning. If the simulation fails, the initial simulation parameters are optimized based on the second model to obtain the target simulation parameters; The target simulation parameters are used as the initial simulation parameters, and the material growth simulation software is called again to perform simulation based on the initial simulation parameters until the simulation is confirmed to be successful.

[0131] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0132] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A material growth simulation method, characterized in that, The method includes: The material growth simulation software is invoked to perform the simulation based on the initial simulation parameters; the initial simulation parameters are obtained by filtering from the first simulation parameter range input by the user according to the first model obtained by supervised learning, or by optimizing the first simulation parameters input by the user according to the second model obtained by reinforcement learning. If the simulation fails, the initial simulation parameters are optimized based on the second model to obtain the target simulation parameters; The target simulation parameters are used as the initial simulation parameters, and the material growth simulation software is called again to perform simulation based on the initial simulation parameters until the simulation is confirmed to be successful.

2. The material growth simulation method according to claim 1, characterized in that, The process of calling the material growth simulation software to perform simulations based on initial simulation parameters includes: Obtain images of the parameter input box and the process start button from the material growth simulation startup interface of the material growth simulation software; Determine the first coordinates of the parameter input box image and the second coordinates of the process start button image; Based on the first coordinates, a first human-computer interaction operation is simulated in the material growth simulation startup interface; the first human-computer interaction operation is used to submit the initial simulation parameters to the parameter input box; Based on the second coordinate, a second human-computer interaction operation is simulated in the material growth simulation startup interface; the second human-computer interaction operation is used to start the simulation process.

3. The material growth simulation method according to claim 1, characterized in that, The determination of simulation failure includes: Based on the log files of the materials growth simulation software, monitor the execution status of the simulation tasks corresponding to the initial simulation parameters; If the execution state is abnormal, the simulation is determined to have failed; or, When the execution state is normal, obtain the simulation results output by the material growth simulation software; If the simulation results do not meet the preset simulation requirements, the simulation is determined to have failed.

4. The material growth simulation method according to claim 3, characterized in that, The simulation results were obtained in the following way: Obtain the simulation result interface image after the material growth simulation software has finished running; Obtain the optical character recognition results of the simulation result interface image, and extract the simulation performance peak from the optical character recognition results according to a regular expression; Key points of the simulation curve are identified by performing key point identification on the characteristic curves of the material growth process in the simulation result interface image. Semantic segmentation is performed on the simulation heatmap in the simulation result interface image to obtain the target simulation heatmap region; The simulation results are obtained based on at least one of the simulation performance peak, the key points of the simulation curve, and the target simulation heatmap region.

5. The material growth simulation method according to claim 1, characterized in that, When the user input is a first simulation parameter range, the initial simulation parameters are obtained as follows: Multiple first candidate simulation parameters are sampled from the first simulation parameter range; Based on the first model, the first predicted simulation result corresponding to each of the first candidate simulation parameters is obtained; Based on the first predicted simulation results and preset simulation requirements, initial simulation parameters are selected from multiple first candidate simulation parameters.

6. The material growth simulation method according to claim 1, characterized in that, When the user inputs the first simulation parameters, the initial simulation parameters are obtained as follows: The first simulation parameters are input into the second model to obtain the initial simulation parameters output by the second model; The second model was obtained through the following training: Based on the policy function of the second model, determine the parameter adjustment amount corresponding to the training data of the first simulation parameters; The first simulation parameter training data is updated according to the parameter adjustment amount to obtain the updated second simulation parameter training data. Based on the first model, the second predicted simulation result corresponding to the training data of the second simulation parameters is obtained; Based on the second prediction simulation results and the reward function of the second model, determine the reward value corresponding to the training data of the second simulation parameters; The policy function of the second model is updated according to the reward value, and the training data of the second simulation parameters is used as the training data of the first simulation parameters. The process continues to execute the policy function of the second model to determine the parameter adjustment amount corresponding to the training data of the first simulation parameters until the iteration termination condition is met.

7. The material growth simulation method according to claim 6, characterized in that, The step of continuing to execute the parameter adjustment amount corresponding to the first simulation parameter training data based on the strategy function of the second model until the iteration termination condition is met also includes: If it is determined that the parameter adjustment amount is less than the preset adjustment amount under a preset number of consecutive iterations before the current iteration round, and the parameter adjustment amount under the current iteration round is less than the preset adjustment amount, a penalty value under the current iteration round is generated; Update the reward value for the current iteration based on the penalty value for the current iteration.

8. The material growth simulation method according to claim 1, characterized in that, The simulation parameter training data for the first model and / or the second model were constructed in the following manner: Obtain the second simulation parameter range input by the user; Multiple second candidate simulation parameters are sampled from the second simulation parameter range; The material growth simulation software is invoked to perform simulations based on multiple second candidate simulation parameters, and the execution status of the simulation task corresponding to each second candidate simulation parameter is monitored according to the log file of the material growth simulation software. From a plurality of second candidate simulation parameters, a third candidate simulation parameter whose execution state is normal is selected to construct simulation parameter training data.

9. A material growth simulation device, characterized in that, include: The first simulation unit is used to call the material growth simulation software to perform simulations based on the initial simulation parameters. The initial simulation parameters are obtained by filtering from the range of first simulation parameters input by the user based on a first model obtained through supervised learning, or by optimizing the first simulation parameters input by the user based on a second model obtained through reinforcement learning. The simulation optimization unit is used to optimize the initial simulation parameters according to the second model to obtain the target simulation parameters when the simulation is determined to fail. The second simulation unit is used to take the target simulation parameters as the initial simulation parameters and continue to call the material growth simulation software to perform simulation based on the initial simulation parameters until the simulation is determined to be successful.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the material growth simulation method as described in any one of claims 1 to 8.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the material growth simulation method as described in any one of claims 1 to 8.

Images