Adaptive hybrid parameter optimization method and device fusing multi-task migration and regional potential modeling

Abstract

The invention discloses a self-adaptive hybrid parameter optimization method and device fusing multi-task migration and regional potential modeling, and the method carries out the unified modeling of a hybrid parameter space based on the performance feedback obtained in the system operation process, and depicts the internal difference of the parameter space through regional management. By evaluating historical performance expressions and potential optimization capabilities of different parameter regions, parameter search is guided to adaptively focus on a high-potential region, so that the search efficiency is improved, and invalid exploration is reduced. Meanwhile, a multi-task migration mechanism based on a parameter search behavior is introduced, the initial search process of a new task is guided by utilizing regional evolution and search features in historical tasks, and optimization convergence is accelerated. According to the method, system operation interruption or manual intervention is not needed, cross-task optimization experience can be effectively utilized, and the method has good universality and engineering application value.

Description

Technical Field

[0001] This invention belongs to the field of computer system automatic optimization and adaptive configuration management technology, specifically relating to an adaptive hybrid parameter optimization method and apparatus that integrates multi-task migration and regional potential modeling. Background Technology

[0002] In recent years, with the continuous expansion of computer system scale and the increasing richness of application scenarios, the number of configuration parameters involved in the operation of software systems and computing platforms has been increasing, and the types of parameters and their interactions have become increasingly complex. System performance often depends heavily on the reasonable configuration of operating parameters. Under different application loads, different operating environments, and different hardware platforms, the optimal parameter configuration of a system varies significantly. Bolet et al. pointed out that in the OpenMP parallel optimization framework commonly used in high-performance computing, after a small amount of initial sampling, if appropriate optimization strategies are adopted, OpenMP parameter settings can bring performance improvements of up to tens of percentage points for certain loads. For example, in their experiments, under a systematic parameter scan of OpenMP loop scheduling and thread count parameters, some loads could reduce runtime by about 20% to 30% or more through automatic optimization based on the default strategy. Therefore, how to achieve automatic optimization of parameter configuration during system operation has become an important research direction in the field of computer system performance optimization.

[0003] In practical applications, system configuration parameters typically include both continuous and discrete parameters, such as the number of threads, cache size, scheduling policy, binding method, and operating mode selection, thus forming a high-dimensional, hybrid parameter space. This type of parameter space is not only enormous, but also often exhibits complex nonlinear coupling relationships between different parameters, making system performance highly uncertain in response to parameter changes. Traditional parameter tuning methods that rely on manual experience or rule-based configuration are not only inefficient but also struggle to achieve stable optimization results in complex scenarios.

[0004] To address these issues, existing research has proposed a series of parameter optimization and automatic parameter tuning methods. One type of method is based on offline modeling or static search strategies, which construct a performance model using a large amount of experimental data before the system runs, and determine the parameter configuration scheme accordingly. However, this type of method usually requires high upfront experimental costs, and assumes that the system performance characteristics remain relatively stable during operation. Once the application load or operating environment changes, the offline model often becomes unusable.

[0005] With the development of online optimization concepts, some research has begun to focus on dynamically adjusting parameter configurations based on real-time performance feedback during system operation. These methods achieve online parameter optimization by continuously collecting system performance metrics and updating optimization strategies during operation. However, when dealing with high-dimensional mixed parameter spaces, directly performing a global search across the entire parameter space often leads to low sampling efficiency, slow convergence of the optimization process, and difficulty in obtaining ideal configurations within limited time or resource budgets.

[0006] To address the aforementioned issues, some studies have attempted to introduce probabilistic modeling, heuristic search, or uncertainty-based sampling strategies to improve parameter search efficiency. For example, surrogate models can be constructed to predict the potential performance of parameter combinations, or a trade-off between exploration and exploitation can be made to reduce invalid sampling. However, such methods still face several challenges in practical applications: on the one hand, the cost of model construction and updating increases significantly with the increase in parameter dimensionality; on the other hand, the potential value differences exhibited by different parameter regions during the optimization process are not fully utilized, leading to an uneven allocation of computational resources.

[0007] Furthermore, most existing online parameter optimization methods typically assume that the optimization process is carried out for a single task or a fixed scenario, lacking systematic utilization of historical optimization information. When the system needs to be tuned for a new application load or a new operating environment, it often needs to start exploring from the initial parameter space again, resulting in redundant sampling and wasted computational resources. Although some research has begun to focus on the problem of information reuse across tasks or scenarios, under mixed parameter space conditions, how to effectively identify the similarity between historical tasks and current tasks, and reasonably transfer existing optimization experience to new optimization processes, still lacks a general and efficient solution.

[0008] On the other hand, in real-world system environments, parameter configuration optimization often needs to be completed without interrupting system operation. Some existing methods require pausing the system or performing costly restarts during parameter adjustments, limiting their application in online scenarios. How to achieve dynamic parameter adjustment and continuous performance improvement while ensuring continuous system operation is another important challenge facing existing technologies.

[0009] In summary, existing parameter optimization and automatic parameter tuning methods still have significant shortcomings in handling high-dimensional mixed parameter spaces, improving online search efficiency, effectively utilizing historical optimization information, and adapting to dynamic operating environments. Therefore, there is an urgent need for an online automatic configuration optimization method that can model the potential of the parameter space at the regional level and combine cross-task search behavior to achieve effective transfer, so as to achieve continuous optimization of parameter configuration and stable improvement of system performance without affecting the normal operation of the system. Summary of the Invention

[0010] The main objective of this invention is to overcome the shortcomings and deficiencies of existing technologies and provide an adaptive hybrid parameter optimization method and apparatus that integrates multi-task migration and regional potential modeling. This invention can perform regional modeling of the hybrid parameter space based on continuously acquired performance feedback information during operation, without stopping the target system or program. It also achieves adaptive adjustment of the parameter search strategy through dynamic evaluation and scheduling of the optimization potential of different parameter regions. Simultaneously, by analyzing and migrating historical task search behavior, it guides the parameter optimization process of new tasks, thereby significantly improving online optimization efficiency, reducing the time and computational costs caused by repeated tuning, and enhancing the overall system performance and resource utilization efficiency.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] In a first aspect, the present invention provides an adaptive hybrid parameter optimization method that integrates multi-task migration and regional potential modeling, comprising the following steps:

[0013] Performance monitoring and search trajectory construction steps: During the operation of the target program or system, collect the performance feedback information corresponding to the current parameter configuration; associate and store the performance feedback information with the corresponding parameter configuration to construct a parameter-performance historical dataset and parameter search trajectory for subsequent regional modeling and multi-task migration analysis.

[0014] Hybrid parameter space regionalization modeling and potential assessment steps: A unified representation model is performed on the hybrid parameter space containing continuous and discrete parameters, and the parameter space is divided into multiple parameter configuration regions; For each parameter configuration region, a regional proxy model is constructed based on the historical performance and model uncertainty within the region, and the corresponding regional optimization potential index is calculated;

[0015] Adaptive region scheduling steps based on region potential: Based on the optimization potential evaluation results of each parameter configuration region, adaptively execute region maintenance, expansion, contraction or switching operations, dynamically determine the target parameter configuration region for the current optimization stage, so as to achieve a balance between exploration and utilization;

[0016] The multi-task migration guidance steps based on search behavior are as follows: Analyze the parameter search behavior, region selection trajectory and performance feedback distribution in historical tasks, and evaluate the similarity between historical tasks and current tasks at the search behavior level; when similar historical tasks exist, transfer their high-potential region information or region evolution patterns to the current task to guide region selection and sampling decisions.

[0017] Adaptive update steps: Generate new parameter combinations within the selected parameter configuration area and apply them to the target program or system operation to obtain new performance feedback information; based on the performance feedback information, update the regional agent model, regional potential assessment results, and search behavior representation, and return to execute the performance monitoring and search trajectory construction steps to form a continuous online adaptive optimization process.

[0018] As a preferred technical solution, in the performance monitoring and search trajectory construction steps:

[0019] Collect performance feedback information at preset time intervals or as a trigger condition for the completion of each parameter execution;

[0020] The performance feedback information is represented as a performance feedback vector. ,in, Indicates the first Performance evaluation results corresponding to the secondary parameter configuration;

[0021] The performance feedback vector is associated with and stored in combination with the corresponding parameter configuration to construct a global performance dataset for subsequent optimization decisions. The global performance dataset includes parameters, historical performance dataset, and parameter search trajectory.

[0022] As a preferred technical solution, in the adaptive regional scheduling step based on regional potential:

[0023] The optimization potential of a region is determined by calculating the expected performance value and uncertainty index of each parameter configuration region.

[0024] The Region Expectation Improvement (REI) index is introduced to better evaluate the optimization potential of a region. When the optimization potential of the parameter configuration region is higher than a preset threshold, the region is determined to have further exploration value, and the region is maintained or expanded. When the optimization potential of the parameter configuration region is lower than the preset threshold, the region is determined to have limited optimization space, and the region is shrunk or switched.

[0025] As a preferred technical solution, REI is used to characterize the overall optimization potential gain of a certain region, and it is defined as the average level of the expected improvement value in that region:

[0026]

[0027] in, Indicates the first Parameter configuration area This represents the expected improvement value calculated based on the surrogate model.

[0028] As a preferred technical solution, in the multi-task migration guidance step based on search behavior:

[0029] Multi-task transfer is based on parameter search behavior and region evolution characteristics. The parameter search behavior includes at least the access order of parameter regions, sampling density distribution, region expansion or contraction frequency, and performance convergence trend. By matching the early search behavior of the current task with the corresponding behavioral characteristics of historical tasks, the similarity between tasks at the parameter space structure and optimization process level is evaluated. When the similarity meets the preset conditions, high-potential parameter region information or region-level proxy models in the corresponding historical tasks are selected as priors to guide the region initialization, region priority sorting, or sampling strategy setting of the current task, thereby realizing model reuse and transfer guidance based on search behavior.

[0030] As a preferred technical solution, in the adaptive update step:

[0031] Parameter sampling is based on the regionalization of the parameter space, where the parameter space is divided into multiple parameter configuration regions. Each parameter configuration region corresponds to a subspace in the parameter space, and historical performance statistics and uncertainty measures are maintained within that region.

[0032] Based on the optimization potential assessment results of each parameter configuration region, parameter configuration regions with high potential are selected for sampling, and new parameter combinations are generated within the selected regions based on regional proxy models or regional constraint sampling strategies.

[0033] When continuous sampling results within a region show performance improvement or convergence trend changes, the corresponding parameter configuration region is dynamically expanded, shrunk, or maintained to achieve region-level adaptive search and resource allocation.

[0034] As a preferred technical solution, the region's historical performance, optimal performance, and potential indicators are combined to form a region selection scoring function:

[0035] ;

[0036] in, Indicates the average performance of the region. Indicates the region's historical best performance. Indicating regional potential, These are the weighting coefficients.

[0037] Secondly, the present invention provides an adaptive hybrid parameter optimization system that integrates multi-task migration and regional potential modeling, which is applied to the aforementioned adaptive hybrid parameter optimization method that integrates multi-task migration and regional potential modeling. The system includes a performance monitoring and search trajectory construction module, a hybrid parameter spatial regionalization modeling and potential assessment module, an adaptive regional scheduling module based on regional potential, a multi-task migration guidance module based on search behavior, and an adaptive update module.

[0038] The performance monitoring and search trajectory construction module is used to collect performance feedback information corresponding to the current parameter configuration during the operation of the target program or system; and to associate and store the performance feedback information with the corresponding parameter configuration to construct a parameter-performance historical dataset and parameter search trajectory for subsequent regional modeling and multi-task migration analysis.

[0039] The hybrid parameter space regionalization modeling and potential assessment module is used to perform unified representation modeling of the hybrid parameter space containing continuous and discrete parameters, and divide the parameter space into multiple parameter configuration regions; for each parameter configuration region, a regional proxy model is constructed based on the historical performance and model uncertainty within the region, and the corresponding regional optimization potential index is calculated.

[0040] The adaptive region scheduling module based on region potential is used to adaptively perform region maintenance, expansion, contraction or switching operations according to the optimization potential evaluation results of each parameter configuration region, and dynamically determine the target parameter configuration region of the current optimization stage in order to achieve a balance between exploration and utilization.

[0041] The multi-task migration guidance module based on search behavior is used to analyze the parameter search behavior, region selection trajectory and performance feedback distribution in historical tasks, and to evaluate the similarity between historical tasks and current tasks at the search behavior level; when similar historical tasks exist, their high-potential region information or region evolution patterns are transferred to the current task to guide region selection and sampling decisions.

[0042] The adaptive update module is used to generate new parameter combinations within the selected parameter configuration area and apply them to the target program or system operation to obtain new performance feedback information; based on the new performance feedback information, it updates the regional agent model, regional potential assessment results, and search behavior representation, and returns to execute the performance monitoring and search trajectory construction steps to form a continuous online adaptive optimization process.

[0043] Thirdly, the present invention provides an electronic device, the electronic device comprising:

[0044] At least one processor; and,

[0045] A memory communicatively connected to the at least one processor; wherein,

[0046] The memory stores computer program instructions that can be executed by the at least one processor to enable the at least one processor to perform the adaptive hybrid parameter optimization method that integrates multi-task migration and regional potential modeling.

[0047] Fourthly, the present invention provides a computer-readable storage medium storing a program that, when executed by a processor, implements the adaptive hybrid parameter optimization method that integrates multi-task migration and regional potential modeling.

[0048] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0049] (1) This invention addresses the high-dimensional mixed parameter space containing continuous and discrete parameters by introducing regional modeling and regional potential assessment mechanisms. This elevates parameter optimization decision-making from the level of single parameter combination to the level of parameter region, guiding the search process to focus on parameter regions with high optimization potential. This effectively avoids blind global search in a large-scale parameter space and significantly improves online optimization efficiency.

[0050] (2) The present invention constructs a continuously updated online optimization closed loop based on operational performance feedback. During the operation of the system, the parameter configuration strategy is dynamically adjusted without stopping or restarting the target program or system. It can naturally adapt to the dynamic changes in the operating environment, input scale and load characteristics, and has good real-time performance and engineering feasibility.

[0051] (3) This invention achieves adaptive allocation of optimization resources in different regions by comprehensively evaluating the historical performance, uncertainty information and search progress of regions with different parameter configurations, reducing the number of invalid samples in low-potential regions, thereby reducing the computational overhead and time cost caused by repeated optimization.

[0052] (4) The present invention further introduces a multi-task migration mechanism based on search behavior. By analyzing the parameter search trajectory, region selection mode and performance feedback distribution in historical tasks, the similarity between different tasks is evaluated, and regional information or search strategies with migration value are introduced into the current optimization process. Under new tasks or new scenarios, the initial exploration stage is significantly shortened, the parameter configuration convergence speed is accelerated, and the overall optimization efficiency is improved.

[0053] (5) This invention does not rely on human experience or customized rules for specific applications. It is applicable to a variety of complex systems and application scenarios. It can achieve stable and efficient automatic configuration optimization under mixed parameter space conditions. It has strong versatility, scalability and promotion application value. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of an adaptive hybrid parameter optimization method that integrates multi-task migration and regional potential modeling, as described in an embodiment of the present invention.

[0056] Figure 2 This is a diagram illustrating the role of the regional agent model in this implementation.

[0057] Figure 3 This is a flowchart of the multi-task migration process based on search behavior in this embodiment;

[0058] Figure 4 This is a block diagram of an adaptive hybrid parameter optimization system that integrates multi-task migration and regional potential modeling, according to an embodiment of the present invention.

[0059] Figure 5 This is a structural diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0060] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0061] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0062] Currently, parameter configuration optimization has become a core technical issue in complex computing systems and software platforms, serving as a crucial foundation for supporting efficient system operation and continuous performance improvement. While various solutions have been proposed in research and engineering practice for automatic parameter tuning, most existing methods focus on the overall parameter space or static configuration, relying on global modeling, offline analysis, or unified search strategies for parameter optimization. In real-world operating environments, these methods often require periodic interruptions, redeployment, or reconfiguration of the system during optimization, making it difficult to achieve truly continuous online optimization.

[0063] Furthermore, existing methods typically use single parameter combinations as the basic decision-making unit, lacking a detailed characterization of the differences in the internal structure of the parameter space. They fail to systematically distinguish the potential value of different parameter subspaces in the optimization process, leading to low search efficiency and dispersed optimization resources in high-dimensional mixed parameter spaces. Meanwhile, most online optimization methods remain limited to single-task or single-scenario assumptions, failing to effectively utilize search process information from historical tasks. When facing new application loads or operating environments, they need to re-explore from scratch, further exacerbating sampling redundancy and computational overhead, limiting their practical application effectiveness in dynamic and complex scenarios. To address these issues, this invention systematically reconstructs the online optimization process in mixed parameter spaces by introducing a regional potential modeling mechanism and a multi-task migration strategy based on search behavior, achieving an automatic configuration optimization scheme that balances search efficiency, migration capability, and online adaptability.

[0064] like Figure 1 As shown in the figure, this embodiment provides an adaptive hybrid parameter optimization method that integrates multi-task migration and regional potential modeling. This method is suitable for dynamic and adaptive optimization of a configuration space containing both continuous and discrete parameters during continuous system operation. Specifically, it includes the following:

[0065] S1. Performance monitoring and search trajectory construction steps: During the operation of the target program or system, collect performance feedback information corresponding to the current parameter configuration; associate and store the performance feedback information with the corresponding parameter configuration to construct a parameter-performance historical dataset and parameter search trajectory for subsequent regional modeling and multi-task migration analysis; specifically:

[0066] During the operation of the target program or system, the performance monitoring module monitors the running status under the current parameter configuration. The performance monitoring module can be deployed inside the program or exist as an external monitoring component, used to collect system performance indicators periodically or in a manner triggered by parameter execution completion events.

[0067] Specifically, performance feedback information is collected at preset time intervals or as a trigger condition for the completion of each parameter execution;

[0068] The performance feedback information is represented as a performance feedback vector. ,in, Indicates the first Performance evaluation results corresponding to the secondary parameter configuration;

[0069] The performance feedback vector is associated with and stored in combination with the corresponding parameter configuration to construct a global performance dataset for subsequent optimization decisions.

[0070] In this embodiment, the performance indicators include, but are not limited to, runtime, throughput, energy consumption, resource utilization, or a comprehensive performance evaluation value formed by weighting multiple indicators. After each parameter configuration execution, the collected performance feedback is associated with the corresponding parameter configuration vector to form a parameter-performance sample record: ,in, Indicates the first Mixed parameter configuration for the next execution This indicates the corresponding performance feedback result.

[0071] The parameter-performance samples are stored in chronological order and used to construct parameter search trajectories, providing basic data support for subsequent regional potential modeling and cross-task search behavior analysis.

[0072] S2. Hybrid Parameter Space Regionalization Modeling and Potential Assessment Steps: A unified representation model is performed on the hybrid parameter space containing continuous and discrete parameters, and the parameter space is divided into multiple parameter configuration regions. For each parameter configuration region, a regional proxy model is constructed based on the historical performance and model uncertainty within the region, and the corresponding regional optimization potential index is calculated, such as... Figure 2 As shown. The parameter space includes both continuous parameters and discrete parameters, where continuous parameters are represented as intervals of real numbers and discrete parameters are represented as a finite set of values.

[0073] In this embodiment, discrete parameters are numerically encoded or embedded and mapped to form a unified parameter representation vector together with continuous parameters, thereby supporting the overall modeling of the mixed parameter space. Based on this, a probabilistic surrogate model is used to model the mapping relationship between parameter configuration and performance feedback, and outputs the performance prediction mean and corresponding uncertainty information.

[0074] As the number of parameter-performance samples increases, the surrogate model is continuously updated online, providing a basis for parameter region potential assessment and search decisions.

[0075] Furthermore, a unified representation model is used to model the mixed parameter space containing continuous and discrete parameters, and the parameter space is divided into multiple parameter configuration regions. For each parameter configuration region, a regional proxy model is constructed based on the historical performance and model uncertainty within the region, and the corresponding regional optimization potential index is calculated.

[0076] To improve search efficiency in high-dimensional mixed parameter spaces, this embodiment introduces a parameter space regionalization management and regional potential modeling mechanism. The parameter region management module divides the parameter space into multiple parameter configuration regions based on parameter distribution, historical search trajectories, or performance change trends. Each parameter configuration region corresponds to a subspace within the parameter space. For each parameter configuration region, historical performance statistics are maintained, including the region's average performance, region's best performance, number of samplings, and the uncertainty level of the surrogate model, and a corresponding region-level surrogate model is constructed.

[0077] In this embodiment, based on a regional agent model, a Regional Expected Improvement (REI) index is introduced to quantitatively evaluate regional potential. REI characterizes the overall optimization potential gain of a region and is defined as the average level of expected improvement values ​​within that region.

[0078] ;

[0079] in, Indicates the first Parameter configuration area This represents the expected improvement value calculated based on the surrogate model. By modeling regional potential, this invention upgrades the traditional "point evaluation" optimization strategy to "regional evaluation," significantly reducing the decision complexity in a high-dimensional parameter space.

[0080] S4. Multi-task migration guidance steps based on search behavior: such as... Figure 3 As shown, the parameter search behavior, region selection trajectory, and performance feedback distribution in historical tasks are analyzed to assess the similarity between historical tasks and current tasks at the search behavior level. When similar historical tasks exist, their high-potential region information or region evolution patterns are transferred to the current task to guide region selection and sampling decisions.

[0081] In this embodiment, the parameter optimization decision module adaptively selects among multiple parameter configuration regions based on the regional potential assessment results. Specifically, it integrates the region's historical performance, optimal performance, and regional potential indicators to construct a region selection scoring function:

[0082] ;

[0083] in, Indicates the average performance of the region. Indicates the region's historical best performance. Indicating regional potential, , representing weighting coefficients. By embedding regional potential into the multi-armed slot machine decision-making framework, the region selection process simultaneously considers current performance levels and future exploration value, thereby achieving a dynamic balance between exploration and utilization among different parameter regions. Once a target parameter configuration region is selected, a restricted sampling strategy is executed within that region to generate new parameter combinations, which are then applied to the target system operation.

[0084] The parameter sampling can generate new parameter combinations based on the existing sample distribution within the region, thereby improving the targeting of the sampling. Through this method, the parameter optimization decision module achieves a dynamic balance between exploration and utilization across different parameter regions, avoiding the concentration of search resources in locally optimal regions or their dispersion in low-value regions.

[0085] S5. Implementation steps for multi-task migration based on search behavior:

[0086] To further improve optimization efficiency, this embodiment introduces a multi-task migration mechanism based on search behavior. When the system faces new application loads or operating scenarios, the parameter optimization decision module first analyzes the initial search trajectory characteristics of the current task and performs a similarity assessment with the search behavior representations saved in historical tasks. These search behavior representations include region selection sequences, region potential change trends, and performance improvement paths from historical tasks. When it is determined that a historical task with similar search behavior to the current task exists, its corresponding high-potential region information or region evolution pattern is migrated to the current task to guide region initialization and region selection order.

[0087] Unlike directly migrating historical optimal parameter configurations, this invention migrates the search structure and region-level experience, thereby improving the robustness and generalization ability of multi-task migration under different parameter spaces and different task conditions.

[0088] S5. Online Closed-Loop Update Implementation Steps:

[0089] After parameter execution is complete, the performance monitoring module collects new performance feedback information and returns it to the parameter optimization decision module. Based on this, the parameter optimization decision module jointly updates the global agent model, the regional agent model, the regional potential assessment results, and the search behavior representation, and then enters the next round of parameter optimization loop.

[0090] Unlike traditional optimization methods based on global search or single-parameter combination decisions, this invention elevates the parameter optimization process to the parameter region level for decision-making. By constructing a region-level proxy model and a region potential assessment mechanism, it achieves efficient search in a high-dimensional hybrid parameter space. Simultaneously, by analyzing parameter search behavior and region evolution trajectories from historical tasks, valuable optimization experience is introduced into the current task, thereby reducing the blind spots in the initial exploration phase.

[0091] It should be noted that, for the sake of simplicity, the aforementioned method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously.

[0092] Based on the same idea as the adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling in the above embodiments, this invention also provides an adaptive hybrid parameter optimization system integrating multi-task migration and regional potential modeling. This system can be used to execute the aforementioned adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling. For ease of explanation, the structural diagram of the embodiment of the adaptive hybrid parameter optimization system integrating multi-task migration and regional potential modeling only shows the parts related to the embodiments of this invention. Those skilled in the art will understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0093] Please see Figure 4 In another embodiment of this application, an adaptive hybrid parameter optimization system 100 integrating multi-task migration and regional potential modeling is provided. The system includes a performance monitoring and search trajectory construction module 101, a hybrid parameter spatial regionalization modeling and potential assessment module 102, an adaptive regional scheduling module 103 based on regional potential, a multi-task migration guidance module 104 based on search behavior, and an adaptive update module 105.

[0094] The performance monitoring and search trajectory construction module 101 is used to collect performance feedback information corresponding to the current parameter configuration during the operation of the target program or system; and to associate and store the performance feedback information with the corresponding parameter configuration to construct a parameter-performance historical dataset and parameter search trajectory for subsequent regional modeling and multi-task migration analysis.

[0095] The hybrid parameter space regionalization modeling and potential assessment module 102 is used to perform unified representation modeling of the hybrid parameter space containing continuous and discrete parameters, and divide the parameter space into multiple parameter configuration regions; for each parameter configuration region, a regional proxy model is constructed based on the historical performance and model uncertainty within the region, and the corresponding regional optimization potential index is calculated.

[0096] The adaptive region scheduling module 103 based on region potential is used to adaptively perform region maintenance, expansion, contraction or switching operations according to the optimization potential evaluation results of each parameter configuration region, and dynamically determine the target parameter configuration region of the current optimization stage in order to achieve a balance between exploration and utilization.

[0097] The multi-task migration guidance module 104 based on search behavior is used to analyze the parameter search behavior, region selection trajectory and performance feedback distribution in historical tasks, and evaluate the similarity between historical tasks and current tasks at the search behavior level; when there are similar historical tasks, its high-potential region information or region evolution mode is transferred to the current task to guide region selection and sampling decisions.

[0098] The adaptive update module 105 is used to generate new parameter combinations within the selected parameter configuration area and apply them to the target program or system operation to obtain new performance feedback information; based on the new performance feedback information, it updates the regional agent model, regional potential assessment results, and search behavior representation, and returns to execute the performance monitoring and search trajectory construction steps to form a continuous online adaptive optimization process.

[0099] It should be noted that the adaptive hybrid parameter optimization system integrating multi-task migration and regional potential modeling of the present invention corresponds one-to-one with the adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling of the present invention. The technical features and beneficial effects described in the embodiments of the adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling are applicable to the embodiments of the adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling. For details, please refer to the description in the embodiments of the method of the present invention, which will not be repeated here.

[0100] Furthermore, in the implementation of the adaptive hybrid parameter optimization system integrating multi-task migration and regional potential modeling in the above embodiments, the logical division of each program module is only an example. In actual applications, the above functions can be assigned to different program modules as needed, for example, for the sake of corresponding hardware configuration requirements or the convenience of software implementation. That is, the internal structure of the adaptive hybrid parameter optimization system integrating multi-task migration and regional potential modeling is divided into different program modules to complete all or part of the functions described above.

[0101] Please see Figure 5 In one embodiment, an electronic device is provided that implements an adaptive hybrid parameter optimization method that integrates multi-task migration and regional potential modeling. The electronic device 200 may include a first processor 201, a first memory 202 and a bus, and may also include a computer program stored in the first memory 202 and executable on the first processor 201, such as an adaptive hybrid parameter optimization program 203 that integrates multi-task migration and regional potential modeling.

[0102] The first memory 202 includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the first memory 202 can be an internal storage unit of the electronic device 200, such as the portable hard drive of the electronic device 200. In other embodiments, the first memory 202 can also be an external storage device of the electronic device 200, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device 200. Furthermore, the first memory 202 can include both internal storage units and external storage devices of the electronic device 200. The first memory 202 can be used not only to store application software and various types of data installed on the electronic device 200, such as the code of the adaptive hybrid parameter optimization program 203 integrating multi-task migration and regional potential modeling, but also to temporarily store data that has been output or will be output.

[0103] In some embodiments, the first processor 201 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The first processor 201 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the first memory 202 and calls data stored in the first memory 202 to perform various functions of the electronic device 200 and process data.

[0104] Figure 5 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 5 The structure shown does not constitute a limitation on the electronic device 200, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0105] The adaptive hybrid parameter optimization program 203, which integrates multi-task migration and regional potential modeling and is stored in the first memory 202 of the electronic device 200, is a combination of multiple instructions. When run in the first processor 201, it can achieve the following:

[0106] Performance monitoring and search trajectory construction steps: During the operation of the target program or system, collect the performance feedback information corresponding to the current parameter configuration; associate and store the performance feedback information with the corresponding parameter configuration to construct a parameter-performance historical dataset and parameter search trajectory for subsequent regional modeling and multi-task migration analysis.

[0107] Hybrid parameter space regionalization modeling and potential assessment steps: A unified representation model is performed on the hybrid parameter space containing continuous and discrete parameters, and the parameter space is divided into multiple parameter configuration regions; For each parameter configuration region, a regional proxy model is constructed based on the historical performance and model uncertainty within the region, and the corresponding regional optimization potential index is calculated;

[0108] Adaptive region scheduling steps based on region potential: Based on the optimization potential evaluation results of each parameter configuration region, adaptively execute region maintenance, expansion, contraction or switching operations, dynamically determine the target parameter configuration region for the current optimization stage, so as to achieve a balance between exploration and utilization;

[0109] The multi-task migration guidance steps based on search behavior are as follows: Analyze the parameter search behavior, region selection trajectory and performance feedback distribution in historical tasks, and evaluate the similarity between historical tasks and current tasks at the search behavior level; when similar historical tasks exist, transfer their high-potential region information or region evolution patterns to the current task to guide region selection and sampling decisions.

[0110] Adaptive update steps: Generate new parameter combinations within the selected parameter configuration area and apply them to the target program or system operation to obtain new performance feedback information; based on the performance feedback information, update the regional agent model, regional potential assessment results, and search behavior representation, and return to execute the performance monitoring and search trajectory construction steps to form a continuous online adaptive optimization process.

[0111] Furthermore, if the modules / units integrated in the electronic device 200 are implemented as software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium. The computer-readable medium may 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, or a read-only memory (ROM).

[0112] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0113] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0114] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. An adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling, characterized in that, Includes the following steps: Performance monitoring and search trajectory construction steps: During the operation of the target program or system, collect the performance feedback information corresponding to the current parameter configuration; associate and store the performance feedback information with the corresponding parameter configuration to construct a parameter-performance historical dataset and parameter search trajectory for subsequent regional modeling and multi-task migration analysis. Hybrid parameter space regionalization modeling and potential assessment steps: A unified representation model is performed on the hybrid parameter space containing continuous and discrete parameters, and the parameter space is divided into multiple parameter configuration regions; For each parameter configuration region, a regional proxy model is constructed based on the historical performance and model uncertainty within the region, and the corresponding regional optimization potential index is calculated; Adaptive region scheduling steps based on region potential: Based on the optimization potential evaluation results of each parameter configuration region, adaptively execute region maintenance, expansion, contraction or switching operations, dynamically determine the target parameter configuration region for the current optimization stage, so as to achieve a balance between exploration and utilization; The multi-task migration guidance steps based on search behavior are as follows: Analyze the parameter search behavior, region selection trajectory and performance feedback distribution in historical tasks, and evaluate the similarity between historical tasks and current tasks at the search behavior level; when similar historical tasks exist, transfer their high-potential region information or region evolution patterns to the current task to guide region selection and sampling decisions. Adaptive update steps: Generate new parameter combinations within the selected parameter configuration area and apply them to the target program or system operation to obtain new performance feedback information; based on the performance feedback information, update the regional agent model, regional potential assessment results, and search behavior representation, and return to execute the performance monitoring and search trajectory construction steps to form a continuous online adaptive optimization process.

2. The adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling as described in claim 1, characterized in that, In the performance monitoring and search trajectory construction steps: Collect performance feedback information at preset time intervals or as a trigger condition for the completion of each parameter execution; The performance feedback information is represented as a performance feedback vector. ,in, Indicates the first Performance evaluation results corresponding to the secondary parameter configuration; The performance feedback vector is associated with and stored in combination with the corresponding parameter configuration to construct a global performance dataset for subsequent optimization decisions. The global performance dataset includes parameters, historical performance dataset, and parameter search trajectory.

3. The adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling as described in claim 1, characterized in that, In the adaptive region scheduling step based on region potential: The optimization potential of a region is determined by calculating the expected performance value and uncertainty index of each parameter configuration region. The Region Expectation Improvement (REI) index is introduced to better evaluate the optimization potential of a region. When the optimization potential of the parameter configuration region is higher than a preset threshold, the region is determined to have further exploration value, and the region is maintained or expanded. When the optimization potential of the parameter configuration region is lower than the preset threshold, the region is determined to have limited optimization space, and the region is shrunk or switched.

4. The adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling as described in claim 1, characterized in that, REI is used to characterize the overall optimization potential gain of a region, and it is defined as the average level of the expected improvement value within that region: in, Indicates the first Parameter configuration area This represents the expected improvement value calculated based on the surrogate model.

5. The adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling as described in claim 1, characterized in that, In the multi-task migration guidance steps based on search behavior: Multi-task transfer is based on parameter search behavior and region evolution characteristics. The parameter search behavior includes at least the access order of parameter regions, sampling density distribution, region expansion or contraction frequency, and performance convergence trend. By matching the early search behavior of the current task with the corresponding behavioral characteristics of historical tasks, the similarity between tasks at the parameter space structure and optimization process level is evaluated. When the similarity meets the preset conditions, high-potential parameter region information or region-level proxy models in the corresponding historical tasks are selected as priors to guide the region initialization, region priority sorting, or sampling strategy setting of the current task, thereby realizing model reuse and transfer guidance based on search behavior.

6. The adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling as described in claim 1, characterized in that, In the adaptive update step: Parameter sampling is based on the regionalization of the parameter space, where the parameter space is divided into multiple parameter configuration regions. Each parameter configuration region corresponds to a subspace in the parameter space, and historical performance statistics and uncertainty measures are maintained within that region. Based on the optimization potential assessment results of each parameter configuration region, parameter configuration regions with high potential are selected for sampling, and new parameter combinations are generated within the selected regions based on regional proxy models or regional constraint sampling strategies. When continuous sampling results within a region show performance improvement or convergence trend changes, the corresponding parameter configuration region is dynamically expanded, shrunk, or maintained to achieve region-level adaptive search and resource allocation.

7. The adaptive hybrid parameter optimization method integrating multi-task migration and regional potential modeling as described in claim 1, characterized in that, The region selection scoring function is composed of a combination of historical performance, best performance, and potential indicators. ； in, Indicates the average performance of the region. Indicates the region's historical best performance. Indicating regional potential, These are the weighting coefficients.

8. An adaptive hybrid parameter optimization system integrating multi-task migration and regional potential modeling, characterized in that, The adaptive hybrid parameter optimization method for integrating multi-task migration and regional potential modeling, as applied to any one of claims 1-7, includes a performance monitoring and search trajectory construction module, a hybrid parameter spatial regionalization modeling and potential assessment module, an adaptive regional scheduling module based on regional potential, a multi-task migration guidance module based on search behavior, and an adaptive update module. The performance monitoring and search trajectory construction module is used to collect performance feedback information corresponding to the current parameter configuration during the operation of the target program or system; and to associate and store the performance feedback information with the corresponding parameter configuration to construct a parameter-performance historical dataset and parameter search trajectory for subsequent regional modeling and multi-task migration analysis. The hybrid parameter space regionalization modeling and potential assessment module is used to perform unified representation modeling of the hybrid parameter space containing continuous and discrete parameters, and divide the parameter space into multiple parameter configuration regions; for each parameter configuration region, a regional proxy model is constructed based on the historical performance and model uncertainty within the region, and the corresponding regional optimization potential index is calculated. The adaptive region scheduling module based on region potential is used to adaptively perform region maintenance, expansion, contraction or switching operations according to the optimization potential evaluation results of each parameter configuration region, and dynamically determine the target parameter configuration region of the current optimization stage in order to achieve a balance between exploration and utilization. The multi-task migration guidance module based on search behavior is used to analyze the parameter search behavior, region selection trajectory and performance feedback distribution in historical tasks, and to evaluate the similarity between historical tasks and current tasks at the search behavior level; when similar historical tasks exist, their high-potential region information or region evolution patterns are transferred to the current task to guide region selection and sampling decisions. The adaptive update module is used to generate new parameter combinations within the selected parameter configuration area and apply them to the target program or system operation to obtain new performance feedback information. Based on the new performance feedback information, the regional agent model, regional potential assessment results, and search behavior representation are updated, and the performance monitoring and search trajectory construction steps are returned to be executed, forming a continuous online adaptive optimization process.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores computer program instructions that can be executed by the at least one processor to enable the at least one processor to perform the adaptive hybrid parameter optimization method for integrating multi-task migration and regional potential modeling as described in any one of claims 1-7.

10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the adaptive hybrid parameter optimization method for integrating multi-task migration and regional potential modeling as described in any one of claims 1-7.

Images