A multi-objective alloy optimization design method and system based on preference modulation

Abstract

The application discloses a kind of multi-objective alloy optimization design method and system based on preference modulation, the method obtains the historical experimental data and design space configuration information of target alloy system, determines alloy composition variable, process variable, constraint condition and multiple performance optimization target;Preference representation is constructed according to preference vector, and input is super network to generate preference related strategy parameter, which is fused with basic strategy parameter to obtain reinforcement learning strategy network;Candidate alloy composition-process design point is iteratively generated using the strategy network, and target evaluation results and uncertainty information are obtained through performance evaluation module, and multi-objective reward vector is constructed in combination with constraint condition;Constraint verification and Pareto non-dominated screening are carried out on the candidate scheme obtained by searching, and candidate alloy composition-process scheme is output.The application can realize strategy adaptation under different performance preferences, and improve the coverage range and screening efficiency of candidate scheme.

Description

Technical Field

[0001] This invention belongs to the fields of computational materials science, artificial intelligence-aided materials design, and alloy composition and process optimization, specifically relating to a multi-objective alloy optimization design method and system based on preference modulation. Background Technology

[0002] The properties of alloy materials are typically determined by their composition, manufacturing process, and their interactions. In alloy optimization design, it is often necessary to simultaneously consider multiple performance indicators such as strength, plasticity, electrical conductivity, phase transformation temperature, and microstructure stability. However, different performance indicators usually have mutually restrictive relationships; for example, increasing alloy strength may lead to a decrease in electrical conductivity or plasticity. Therefore, it is necessary to weigh multiple performance objectives and screen candidate solutions that meet application requirements from a large composition-process design space.

[0003] Existing alloy optimization design methods mainly include empirical trial-and-error methods, single-objective optimization methods based on performance prediction models, and fixed-weight multi-objective optimization methods. Empirical trial-and-error methods rely on researchers' experience, obtaining design solutions through repeated adjustments to alloy composition and process parameters and experimental verification. This approach suffers from problems such as numerous experiments, long development cycles, and low candidate screening efficiency. Single-objective optimization methods typically establish predictive models for a single performance index, making it difficult to handle the coupling relationships between multiple performance objectives and to form a set of candidate solutions with different performance trade-offs.

[0004] Fixed-weight multi-objective optimization methods typically linearly weight multiple performance objectives according to pre-defined weights, transforming a multi-objective optimization problem into a single-objective optimization problem. However, fixed weights can only reflect the trade-offs required under a specific optimization task. When the design focus changes, it is usually necessary to reset the weights and re-execute the optimization, making it difficult to flexibly switch between different preference conditions within the same optimization framework.

[0005] Existing multi-objective reinforcement learning methods often express design preferences by statically weighting the reward output or action value output. Although these methods can select actions based on weights, the preference information is not incorporated into the internal parameter structure of the reinforcement learning policy network. Under different preference conditions, they still mainly rely on the same policy parameters, making it difficult to develop suitable search strategies for different tasks such as biased strength, biased conductivity, or equilibrium optimization.

[0006] Furthermore, existing candidate evaluation and ranking processes typically rely primarily on predicted performance values, failing to fully utilize the uncertainty information output by performance prediction models. This makes it difficult for the search process to distinguish between high-risk areas with low confidence and high-uncertainty areas with potential exploration value, easily leading to problems such as candidate solutions being concentrated in local target areas, insufficient Pareto front expansion, and low search robustness.

[0007] Therefore, it is necessary to provide a multi-objective alloy optimization design method that can uniformly characterize different performance trade-off preferences, dynamically adjust the internal parameters of the strategy according to the preferences, and explicitly incorporate prediction uncertainty into the reward construction and candidate selection process. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a multi-objective alloy optimization design method and system based on preference modulation, which addresses the shortcomings of the prior art. This method solves the technical problems in existing multi-objective alloy optimization methods, such as the difficulty in uniformly expressing and flexibly switching the trade-off preferences of different performance objectives, the fact that preference information only acts on the output and cannot drive the dynamic adaptation of internal parameters of the strategy, the failure to fully utilize the uncertainty of performance prediction in the search process, and the limited coverage of the generated candidate solutions in the target space and insufficient search robustness.

[0009] The present invention adopts the following technical solution: A multi-objective alloy optimization design method based on preference modulation includes the following steps: S1. Obtain historical experimental data and design space configuration information of the target alloy system, and determine alloy composition variables, process variables, constraints and at least two alloy performance optimization targets based on the design space configuration information; S2. Obtain the preference vector corresponding to the alloy performance optimization target, and construct a preference representation based on the preference vector; S3. Input the preference representation into the super network to generate preference-related policy parameters, and fuse the preference-related policy parameters with the preset basic policy parameters to obtain a reinforcement learning policy network; S4. Iteratively adjust the alloy composition variables and the process variables using the reinforcement learning strategy network to generate candidate alloy composition-process design points; S5. Based on the historical experimental data, the target evaluation results and uncertainty information of the candidate alloy composition-process design point are obtained through the performance evaluation module. A multi-objective reward vector is constructed according to the target evaluation results, the uncertainty information and the constraints, and iterative search is continued based on the multi-objective reward vector. S6. Perform constraint verification and Pareto non-dominated screening on the candidate schemes obtained by iterative search, and output the candidate alloy composition-process scheme.

[0010] Furthermore, the historical experimental data includes alloy composition parameters, process parameters, and performance test results corresponding to each historical sample; the design space configuration information includes the range of composition variables, the range of process variables, equilibrium element rules, dispersion length, continuous boundary, performance target, and constraint conditions.

[0011] Furthermore, the preference vector is generated from at least one of user input, a preset optimization mode, or sampling results during the training phase, and the preference vector is normalized to obtain the preference representation.

[0012] Furthermore, the supernetwork takes the preference representation as input and outputs preference-related policy parameters corresponding to the preference representation; the preference-related policy parameters and the basic policy parameters are fused according to a preset fusion coefficient.

[0013] Furthermore, the reinforcement learning strategy network outputs multi-objective action value vectors corresponding to each alloy performance optimization objective for candidate actions in the current state, scalarizes the multi-objective action value vectors according to the preference representation, and selects candidate actions for adjusting the alloy composition variable or the process variable based on the scalarization result.

[0014] Furthermore, the performance evaluation module includes at least one of a proxy model, a rule model, and an experimental database matching module, used to obtain the target evaluation results corresponding to each candidate alloy composition-process design point.

[0015] Furthermore, the performance evaluation module includes multiple sub-models, and the uncertainty information is determined based on the degree of dispersion between the prediction results of the multiple sub-models for the same candidate alloy composition-process design point.

[0016] Furthermore, each objective reward term in the multi-objective reward vector is determined based on the normalized evaluation value, normalized uncertainty, and constraint penalty term of the corresponding alloy performance optimization objective.

[0017] Furthermore, the Pareto candidate schemes obtained through the Pareto non-dominated screening are ranked according to at least one of preference score, comprehensive reward score, uncertainty level, and hypervolume contribution.

[0018] Secondly, embodiments of the present invention provide a multi-objective alloy optimization design system based on preference modulation, comprising: The data input module is used to acquire historical experimental data and design space configuration information of the target alloy system, and to determine alloy composition variables, process variables, constraints, and at least two alloy performance optimization objectives. The preference characterization module is used to obtain the preference vector corresponding to the alloy performance optimization target and construct the preference characterization. The hypernetwork module is used to generate preference-related policy parameters based on the preference representation; The parameter fusion module is used to fuse the preference-related policy parameters with preset basic policy parameters to obtain a reinforcement learning policy network. The reinforcement learning search module is used to iteratively adjust the alloy composition variables and the process variables using the reinforcement learning policy network to generate candidate alloy composition-process design points. The performance evaluation module is used to obtain the target evaluation results and uncertainty information of the candidate alloy composition-process design point; The reward construction module is used to construct a multi-objective reward vector based on the target evaluation result, the uncertainty information, and the constraints. The Pareto screening module is used to perform constraint verification and Pareto non-dominated screening on candidate schemes, and output candidate alloy composition-process schemes.

[0019] Thirdly, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the above-described multi-objective alloy optimization design method based on preference modulation.

[0020] Fourthly, embodiments of the present invention provide a computer-readable storage medium including a computer program, which, when executed by a processor, implements the steps of the above-described multi-objective alloy optimization design method based on preference modulation.

[0021] Fifthly, a chip includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the above-described multi-objective alloy optimization design method based on preference modulation.

[0022] In a sixth aspect, embodiments of the present invention provide an electronic device including a computer program, which, when executed by the electronic device, implements the steps of the above-described multi-objective alloy optimization design method based on preference modulation.

[0023] Compared with the prior art, the present invention has at least the following beneficial effects: A multi-objective alloy optimization design method based on preference modulation is proposed. This method establishes a unified optimization space among alloy composition, process variables, performance objectives, and constraints by acquiring historical experimental data and design space configuration information. Preference representations are constructed using preference vectors, and preference-related policy parameters are generated using a hypernetwork, allowing different performance trade-offs to directly influence the reinforcement learning policy parameters. By fusing preference-related policy parameters with basic policy parameters, a search strategy corresponding to the current preference is formed while retaining general search capabilities. Furthermore, objective evaluation results, uncertainty information, and constraints are incorporated into the multi-objective reward construction, ensuring the search process considers prediction performance, prediction reliability, and scheme feasibility. Finally, Pareto non-dominated screening outputs multiple sets of alloy composition-process schemes with different performance trade-offs, expanding the selection range of candidate schemes.

[0024] Furthermore, by simultaneously configuring alloy composition parameters, process parameters, equilibrium element rules, variable boundaries, and performance targets, the alloy design space can be structurally constrained, avoiding the generation of invalid candidate schemes that exceed the composition equilibrium relationship, process range, or actual constraints, and facilitating the transfer of the same optimization framework to different alloy systems.

[0025] Furthermore, preference vectors can be generated through user input, preset optimization modes, or training sampling methods, which can be adapted to manually specified design focuses, standardized optimization tasks, and multi-preference strategy training, respectively; normalization processing ensures that different target preferences are on a uniform scale, avoiding the influence of differences in the numerical dimensions or amplitudes of preferences on strategy modulation.

[0026] Furthermore, the preference representation is mapped to preference-related policy parameters through a hypernetwork, and then fused with the basic policy parameters according to the fusion coefficient. This allows preference information to no longer only affect the final reward or action value output, but to enter the parameter structure of the policy network. The fusion coefficient can adjust the ratio between general search knowledge and preference-specific policies.

[0027] Furthermore, by outputting a multi-objective action value vector, the impact of candidate actions on the performance objectives of each alloy can be evaluated separately; and by scalarizing based on preference characterization, action selection can be made consistent with the current performance trade-off requirements, avoiding the use of the same action decision criteria under all preference conditions.

[0028] Furthermore, performance evaluation of candidate design points can be conducted using proxy models, rule models, or experimental database matching modules. Different evaluation methods can be selected based on data scale, material system, and model availability, reducing the reliance on conducting physical experiments for each candidate scheme and improving the evaluation efficiency of large-scale composition-process space.

[0029] Furthermore, by using multiple sub-models to predict the same candidate design point and determining the uncertainty information based on the dispersion of the prediction results, the credibility of the performance evaluation results can be reflected, providing data support for distinguishing known high-performance areas, predicted risk areas, and potential high-value exploration areas.

[0030] Furthermore, the normalized performance evaluation value, normalized uncertainty, and constraint penalty term are used together to construct the target reward term, so that reinforcement learning search not only considers the performance level of candidate solutions, but also the prediction uncertainty and constraint satisfaction, thereby reducing high-scoring but unfeasible or unreliable candidate solutions.

[0031] Furthermore, Pareto candidate schemes can be ranked by preference score, comprehensive reward score, uncertainty level, or hypervolume contribution. Candidate results can be organized from different dimensions such as preference matching degree, overall performance, prediction credibility, and frontier extension value, which makes it easier for researchers to further screen them according to specific application needs.

[0032] It is understood that the beneficial effects of the second to sixth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0033] In summary, this invention uses preference vectors to uniformly represent multiple performance trade-offs, generates preference-related policy parameters through a hypernetwork and integrates them with basic policy parameters to achieve dynamic policy modulation under different preference conditions. At the same time, it incorporates performance evaluation uncertainties and constraints into the reward construction, and outputs multiple sets of candidate alloy composition-process schemes through Pareto screening, taking into account search adaptability, candidate coverage and engineering feasibility.

[0034] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0035] Figure 1 This is an overall flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the system module structure of the present invention; Figure 3 This invention is compared with the target range of ordinary reinforcement learning algorithms; Figure 4 A distribution diagram of dual-objective candidates in the copper alloy system; Figure 5 Distribution map of dual-target candidates for shape memory alloys; Figure 6 A dual-target candidate distribution map for aluminum bronze alloys; Figure 7 A schematic diagram of a computer device provided in an embodiment of the present invention; Figure 8This is a block diagram of a chip provided according to an embodiment of the present invention.

[0036] Among them, 60. Computer equipment; 61. Processor; 62. Memory; 63. Computer program; 600. Electronic device; 610. Processing unit; 620. Storage unit; 6201. Random access memory unit; 6202. Cache memory unit; 6203. Read-only memory unit; 6204. Program / utility; 6205. Program module; 630. Bus; 640. Display unit; 650. Input / output interface; 660. Network adapter; 700. External device. Detailed Implementation

[0037] The technical solutions of the embodiments of the present 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 the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0039] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0040] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this invention generally indicates that the preceding and following objects have an "or" relationship.

[0041] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0042] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0043] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0044] This invention provides a multi-objective alloy optimization design method based on preference modulation. It employs preference vectors to uniformly represent the trade-offs between different alloy performance objectives. These preference representations are input into a hypernetwork to generate preference-related policy parameters, which are then fused with the basic policy parameters. This allows preference information to directly influence the internal parameters of the reinforcement learning policy network, rather than merely participating in the weighting of end-action values. Secondly, a performance evaluation module simultaneously obtains the target evaluation results and uncertainty information of candidate alloy composition-process design points, and these, along with constraints, are used to construct a multi-objective reward. Finally, the search results are constrained and subjected to Pareto non-dominated screening. Therefore, it can adapt to different design requirements such as biased strength, biased conductivity, and equilibrium optimization within the same policy framework, improving the search adaptability under multi-preference tasks. Simultaneously, it balances high-performance region utilization with high-uncertainty region exploration, reducing ineffective or high-risk candidate solutions and expanding the coverage of the Pareto candidate frontier, providing more optional results with different performance trade-offs for subsequent alloy scheme screening.

[0045] Please see Figure 1 The present invention discloses a multi-objective alloy optimization design method based on preference modulation, comprising the following steps: S1. Read the historical experimental data and design space configuration of the target alloy system to determine the composition variables, process variables, constraints and optimization objectives; First, historical experimental data of the target alloy system is read, including composition parameters, process parameters, and corresponding performance test results for multiple samples. Simultaneously, the design space configuration file is read to determine the range of composition variables, the range of process variables, equilibrium element rules, dispersion lengths, continuity boundaries, and the name of the performance target.

[0046] S2. Construct a preference representation based on the user-input preference vector, preset optimization mode, or sampling results from the training phase. A preference representation is generated based on the user's input optimization focus. In a two-objective implementation, the preference representation can be represented as a two-dimensional normalized vector; in a multi-objective implementation, the preference representation can be represented as a multi-dimensional normalized vector. The preference representation can be directly input by the user or automatically generated by the system based on a preset template or training sampling mechanism.

[0047] S3. Input the preference representation into the supernetwork to generate preference-related policy parameters, and fuse the preference-related policy parameters with the basic policy parameters to obtain the reinforcement learning policy network under the current preference conditions; The preference representation is input into the hypernetwork to obtain the preference-related policy parameters. If the hypernetwork is denoted as H(·), then the preference-related policy parameters satisfy:

[0048] The system also maintains basic policy parameters θbase, which represent the general search knowledge shared across preferences. Then, the reinforcement learning policy network under the current preference conditions is obtained through parameter fusion, and the fusion relationship is as follows:

[0049] Here, α is the fusion coefficient. This structure allows different preferences to no longer only affect the weighting at the output end, but also enter the internal parameter structure of the policy, realizing super-network modulation from weights to policy.

[0050] Under the influence of the fused policy parameters, the policy network outputs a multi-objective action value vector for each action a in state s:

[0051] Then, the action value vector is scalarized based on the preference representation ω:

[0052] And select the action according to the following formula:

[0053] S4. With the support of the performance evaluation module, the reinforcement learning policy network performs multiple rounds of search on the component variables and process variables, and constructs a multi-objective reward vector based on the target evaluation results, uncertainty information, and constraints corresponding to the candidate design points. The uncertainty information is output synchronously by the performance evaluation module when predicting the candidate design points, and is represented by the standard deviation of the prediction results from multiple sub-models. With the support of the performance evaluation module, the reinforcement learning agent performs multiple rounds of search in the design space. The agent forms candidate design points by progressively adjusting component and process variables, and then evaluates these candidate design points against objectives. The performance evaluation module can employ a surrogate model, a rule-based model, an experimental database matching module, or a combination thereof. Based on the evaluation results of each objective of the candidate design points, uncertainty information, and constraint satisfaction, a multi-objective reward vector is constructed to guide action value evaluation and subsequent search processes.

[0054] The reward for the i-th objective is:

[0055] in, For the first The normalized evaluation value of each objective; For the first The normalized uncertainty of each objective is determined by the degree of dispersion of predictions of the same candidate design point by multiple sub-models in the performance evaluation module, and is the normalized result of the prediction standard deviation. For uncertainty weighting coefficients; To constrain penalty items.

[0056] This approach enables the system not only to seek highly rated candidate points, but also to identify high-uncertainty regions with exploratory value.

[0057] S5. Perform deduplication, constraint verification, Pareto non-dominated screening and sorting on the candidate solution set obtained by the search, and output the candidate alloy composition-process scheme.

[0058] After obtaining the candidate solution set, deduplication, constraint verification, and Pareto non-dominated screening are performed on the candidate solution set to obtain the Pareto candidate set. Further, the Pareto candidate set can be sorted according to indicators such as user preference score, comprehensive reward score, uncertainty level, and hypervolume contribution, and candidate solutions can be output.

[0059] Please see Figure 2 In another embodiment of the present invention, a multi-objective alloy optimization design system based on preference modulation is provided. This system can be used to implement the above-mentioned multi-objective alloy optimization design method based on preference modulation. Specifically, the multi-objective alloy optimization design system based on preference modulation includes a data input module, a preference representation module, a supernetwork module, a parameter fusion module, a reinforcement learning search module, a performance evaluation module, a reward construction module, and a Pareto screening module.

[0060] The data input module is used to acquire historical experimental data and design space configuration information of the target alloy system, and to determine alloy composition variables, process variables, constraints and at least two alloy performance optimization objectives. The preference characterization module is used to obtain the preference vector corresponding to the alloy performance optimization target and construct the preference characterization. The hypernetwork module is used to generate preference-related policy parameters based on the preference representation; The parameter fusion module is used to fuse the preference-related policy parameters with preset basic policy parameters to obtain a reinforcement learning policy network. The reinforcement learning search module is used to iteratively adjust the alloy composition variables and the process variables using the reinforcement learning policy network to generate candidate alloy composition-process design points. The performance evaluation module is used to obtain the target evaluation results and uncertainty information of the candidate alloy composition-process design point; The reward construction module is used to construct a multi-objective reward vector based on the target evaluation result, the uncertainty information, and the constraints. The Pareto screening module is used to perform constraint verification and Pareto non-dominated screening on candidate schemes, and output candidate alloy composition-process schemes.

[0061] This invention provides a terminal device comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment can be used for the operation of a preference-modulated multi-objective alloy optimization design method, including: The process involves: acquiring historical experimental data and design space configuration information of the target alloy system; determining alloy composition variables, process variables, constraints, and at least two alloy performance optimization objectives based on the design space configuration information; obtaining preference vectors corresponding to the alloy performance optimization objectives and constructing preference representations based on these preference vectors; inputting these preference representations into a supernetwork to generate preference-related policy parameters, and fusing these preference-related policy parameters with preset basic policy parameters to obtain a reinforcement learning policy network; iteratively adjusting the alloy composition variables and process variables using the reinforcement learning policy network to generate candidate alloy composition-process design points; obtaining target evaluation results and uncertainty information for the candidate alloy composition-process design points based on the historical experimental data through a performance evaluation module; constructing a multi-objective reward vector based on the target evaluation results, the uncertainty information, and the constraints; and continuing iterative search based on the multi-objective reward vector; performing constraint verification and Pareto non-dominated screening on the candidate schemes obtained from the iterative search to output candidate alloy composition-process schemes.

[0062] Please see Figure 7 The terminal device is a computer device. In this embodiment, the computer device 60 includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When executed by the processor 61, the computer program 63 implements the preference-modulated multi-objective alloy optimization design method of the embodiment. To avoid repetition, details are omitted here. Alternatively, when executed by the processor 61, the computer program 63 implements the functions of each model / unit in the preference-modulated multi-objective alloy optimization design system of the embodiment. To avoid repetition, details are omitted here.

[0063] Computer device 60 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. Computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will understand that... Figure 7 This is merely an example of computer device 60 and does not constitute a limitation on computer device 60. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer device may also include input / output devices, network access devices, buses, etc.

[0064] The processor 61 may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0065] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or memory of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the computer device 60.

[0066] Furthermore, the memory 62 may include both internal storage units of the computer device 60 and external storage devices. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.

[0067] Please see Figure 8 The terminal device is an electronic device 600, which is manifested in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including storage unit 620 and processing unit 610), a display unit 640, etc.

[0068] The storage unit stores program code, which can be executed by the processing unit 610 to perform the steps described in the method section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 610 can perform actions such as... Figure 1 The steps are shown in the figure.

[0069] Storage unit 620 may include readable media in the form of volatile storage units, such as random access memory (RAM) 6201 and / or cache memory 6202, and may further include read-only memory (ROM) 6203.

[0070] Storage unit 620 may also include a program / utility 6204 having a set (at least one) program module 6205, such program module 6205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0071] Bus 630 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the multiple bus structures.

[0072] Electronic device 600 can also communicate with one or more external devices 700 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 600, and / or with any device that enables electronic device 600 to communicate with one or more other computing devices (e.g., router, modem). This communication can be performed via input / output interface 650. Furthermore, electronic device 600 can also communicate with one or more networks (e.g., local area network, wide area network, and / or public network, such as the Internet) via network adapter 660. Network adapter 660 can communicate with other modules of electronic device 600 via bus 630. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms.

[0073] This invention also provides a storage medium, specifically a computer-readable storage medium, which is a memory device in a terminal device for storing programs and data. It is understood that the computer-readable storage medium here can include both built-in storage media in the terminal device and extended storage media supported by the terminal device; it can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). More specific examples of the computer-readable storage medium include: an electrical connection with one or more wires, a portable disk, a hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical fiber, portable compact disk read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0074] Computer-readable storage media also include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium can also be any readable medium other than a readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, radio frequency, etc., or any suitable combination thereof.

[0075] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0076] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the multi-objective alloy optimization design method based on preference modulation in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor to perform the following steps: The process involves: acquiring historical experimental data and design space configuration information of the target alloy system; determining alloy composition variables, process variables, constraints, and at least two alloy performance optimization objectives based on the design space configuration information; obtaining preference vectors corresponding to the alloy performance optimization objectives and constructing preference representations based on these preference vectors; inputting these preference representations into a supernetwork to generate preference-related policy parameters, and fusing these preference-related policy parameters with preset basic policy parameters to obtain a reinforcement learning policy network; iteratively adjusting the alloy composition variables and process variables using the reinforcement learning policy network to generate candidate alloy composition-process design points; obtaining target evaluation results and uncertainty information for the candidate alloy composition-process design points based on the historical experimental data through a performance evaluation module; constructing a multi-objective reward vector based on the target evaluation results, the uncertainty information, and the constraints; and continuing iterative search based on the multi-objective reward vector; performing constraint verification and Pareto non-dominated screening on the candidate schemes obtained from the iterative search to output candidate alloy composition-process schemes.

[0077] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0078] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present 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 the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0079] Taking the multi-objective optimization design task of copper alloys as an example, this paper compares and explains the method of this invention with ordinary reinforcement learning algorithms. The comparison index does not focus on the absolute performance value of a single recommended point, but rather on the ability of the candidate frontier to expand in the target space. This is because, for multi-objective optimization tasks, whether the method can form a wider candidate trade-off interval directly affects the operability of subsequent manual screening, process constraint superposition, and engineering decision-making.

[0080] Please see Figure 3 Under the same comparative conditions, the span of the method of this invention in the tensile strength direction is 61.48 MPa, while the span of the ordinary reinforcement learning algorithm is 26.67 MPa; the span of the method of this invention in the conductivity direction is 13.23% IACS, while the span of the ordinary reinforcement learning algorithm is 10.36% IACS. This shows that the Pareto candidate front formed by the method of this invention has a larger expansion range in both target directions, indicating that this method can output a wider multi-objective trade-off interval.

[0081] The above results indicate that the beneficial effects of the method of the present invention are mainly reflected in: First, it can cover a wider range of performance combinations in the process of multi-objective optimization, rather than focusing on a narrower local area. Secondly, it can retain more candidate solutions with different trade-offs for subsequent researchers, making it easier to further screen them by combining manufacturing costs, process feasibility and application scenarios; Third, it is more suitable as a candidate generation tool for engineering design, rather than just outputting a single high-scoring solution.

[0082] Please see Figure 4 In the copper alloy system, using tensile strength (UTS) and electrical conductivity (EC) as dual optimization objectives, a comparative distribution diagram of the predicted samples in the search space and the candidate outputs of the method of this invention is constructed as follows: Figure 4 As shown in the figure. The results show that the candidate points output by the method of the present invention are mainly distributed in the outer envelope region of the target space, and form a clear expansion along the intensity-conductivity tradeoff direction, indicating that the method can identify representative non-dominated candidate schemes in this system.

[0083] Please see Figure 5 In shape memory alloy systems, using recoverable strain RS and austenite transformation completion temperature Af as dual optimization objectives, a comparison distribution diagram of the predicted samples in the search space and the candidate output by the method of this invention is constructed as follows. Figure 5 As shown in the figure. The results show that the output candidates of the method of the present invention are mainly located in the boundary region of the sample distribution in the search space, indicating that the method can identify more valuable trade-off boundaries under multi-objective conditions, rather than just staying in the sample set region.

[0084] Please see Figure 6 In a high-strength, high-ductility aluminum bronze alloy system, using tensile strength (UTS) and elongation (EL) as dual optimization objectives, a comparative distribution diagram of the search space prediction samples and the candidate outputs of the method of this invention is constructed as follows: Figure 6 As shown in the figure. The results show that the candidate outputs of the method of the present invention form a relatively clear front distribution in the strength-plasticity objective space and cover multiple different trade-off positions, indicating that the method can be adapted to dual-objective optimization tasks of different material systems.

[0085] The method of this invention does not rely on a single alloy system or a single target combination. Under different variable spaces, different target definitions and different material scenarios, it can output multiple sets of candidate solutions located near the boundary of the target space. Therefore, it has good versatility, scalability and engineering application value.

[0086] This invention addresses the problems in multi-objective alloy optimization, such as the difficulty in uniformly expressing different task preferences, the limited response of fixed weighting methods to preferences, and the difficulty in balancing performance improvement and exploration of potential high-value regions during the search process. It proposes a multi-objective alloy optimization design method based on preference modulation and supernetwork reinforcement learning.

[0087] This method uses preference vectors to uniformly express the trade-offs for different optimization tasks, enabling the same algorithm framework to perform multi-objective search under different preference conditions. Unlike simple reinforcement learning methods that only perform linear weighting at the output, this method uses a hypernetwork to generate preference-related policy parameters and fuses them with the basic policy parameters, allowing preference information to enter the policy parameter layer, thereby improving the policy adaptability under different preference conditions.

[0088] Meanwhile, this method explicitly introduces uncertainty information into the reward construction, enabling the search process to balance the utilization of known high-performance regions with the exploration of potentially high-value regions, thus improving search robustness. The method outputs a Pareto non-dominated candidate solution set, facilitating subsequent screening from different objective trade-off perspectives. Since the design variable space, target system, and constraints can all be defined through configuration files, this method also possesses good task reconfiguration capabilities and potential for expansion across alloy systems.

[0089] In summary, this invention presents a multi-objective alloy optimization design method and system based on preference modulation. It uses preference vectors to uniformly represent the trade-offs between different alloy performance objectives and utilizes a hypernetwork to generate preference-related policy parameters, fusing these parameters with the basic policy parameters. This allows preference information to directly enter the reinforcement learning policy network, rather than being statically weighted at the reward output, thereby achieving dynamic policy adaptation for different tasks such as strength-biased, conductivity-biased, plasticity-biased, or equilibrium optimization. Furthermore, this invention simultaneously obtains the target evaluation results and uncertainty information of candidate alloy composition-process design points through a performance evaluation module, and constructs multi-objective rewards based on constraints, ensuring that the search process considers performance improvement, prediction reliability, and scheme feasibility. After constraint verification and Pareto non-dominated screening, multiple sets of candidate schemes covering different performance trade-offs can be output, expanding the candidate frontier and reducing the problem of candidates concentrating in local areas, providing richer and more reliable design basis for subsequent experimental verification, process screening, and engineering decisions.

[0090] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0091] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0092] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0093] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0094] 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0095] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0096] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random-access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0097] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0098] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0099] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0100] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A multi-objective alloy optimization design method based on preference modulation, characterized in that, Includes the following steps: S1. Obtain historical experimental data and design space configuration information of the target alloy system, and determine alloy composition variables, process variables, constraints and at least two alloy performance optimization targets based on the design space configuration information; S2. Obtain the preference vector corresponding to the alloy performance optimization target, and construct a preference representation based on the preference vector; S3. Input the preference representation into the super network to generate preference-related policy parameters, and fuse the preference-related policy parameters with the preset basic policy parameters to obtain a reinforcement learning policy network; S4. Iteratively adjust the alloy composition variables and the process variables using the reinforcement learning strategy network to generate candidate alloy composition-process design points; S5. Based on the historical experimental data, the target evaluation results and uncertainty information of the candidate alloy composition-process design point are obtained through the performance evaluation module. A multi-objective reward vector is constructed according to the target evaluation results, the uncertainty information and the constraints, and iterative search is continued based on the multi-objective reward vector. S6. Perform constraint verification and Pareto non-dominated screening on the candidate schemes obtained by iterative search, and output the candidate alloy composition-process scheme.

2. The multi-objective alloy optimization design method based on preference modulation according to claim 1, characterized in that, The historical experimental data includes alloy composition parameters, process parameters, and performance test results corresponding to each historical sample; the design space configuration information includes the range of composition variables, the range of process variables, equilibrium element rules, dispersion length, continuous boundary, performance target, and constraint conditions.

3. The multi-objective alloy optimization design method based on preference modulation according to claim 1, characterized in that, The preference vector is generated from at least one of user input, preset optimization mode, or sampling results during the training phase, and the preference vector is normalized to obtain the preference representation.

4. The multi-objective alloy optimization design method based on preference modulation according to claim 1, characterized in that, The supernetwork takes the preference representation as input and outputs preference-related policy parameters corresponding to the preference representation; the preference-related policy parameters and the basic policy parameters are fused according to a preset fusion coefficient.

5. The multi-objective alloy optimization design method based on preference modulation according to claim 1, characterized in that, The reinforcement learning strategy network outputs multi-objective action value vectors corresponding to each alloy performance optimization objective for candidate actions in the current state. The multi-objective action value vectors are scalarized according to the preference representation, and candidate actions for adjusting the alloy composition variable or the process variable are selected based on the scalarization result.

6. The multi-objective alloy optimization design method based on preference modulation according to claim 1, characterized in that, The performance evaluation module includes at least one of a proxy model, a rule model, and an experimental database matching module, used to obtain the target evaluation results corresponding to each candidate alloy composition-process design point.

7. The multi-objective alloy optimization design method based on preference modulation according to claim 1, characterized in that, The performance evaluation module includes multiple sub-models, and the uncertainty information is determined based on the degree of dispersion among the prediction results of the multiple sub-models for the same candidate alloy composition-process design point.

8. The multi-objective alloy optimization design method based on preference modulation according to claim 1, characterized in that, Each objective reward term in the multi-objective reward vector is determined based on the normalized evaluation value, normalized uncertainty, and constraint penalty term of the corresponding alloy performance optimization objective.

9. The multi-objective alloy optimization design method based on preference modulation according to claim 1, characterized in that, The Pareto candidate schemes obtained through the Pareto non-dominated screening are ranked according to at least one of the following: preference score, comprehensive reward score, uncertainty level, and hypervolume contribution.

10. A multi-objective alloy optimization design system based on preference modulation, characterized in that, include: The data input module is used to acquire historical experimental data and design space configuration information of the target alloy system, and to determine alloy composition variables, process variables, constraints, and at least two alloy performance optimization objectives. The preference characterization module is used to obtain the preference vector corresponding to the alloy performance optimization target and construct the preference characterization. The hypernetwork module is used to generate preference-related policy parameters based on the preference representation; The parameter fusion module is used to fuse the preference-related policy parameters with preset basic policy parameters to obtain a reinforcement learning policy network. The reinforcement learning search module is used to iteratively adjust the alloy composition variables and the process variables using the reinforcement learning policy network to generate candidate alloy composition-process design points. The performance evaluation module is used to obtain the target evaluation results and uncertainty information of the candidate alloy composition-process design point; The reward construction module is used to construct a multi-objective reward vector based on the target evaluation result, the uncertainty information, and the constraints. The Pareto screening module is used to perform constraint verification and Pareto non-dominated screening on candidate schemes, and output candidate alloy composition-process schemes.

Images