Control parameter setting value optimization method and device, and computer device

Abstract

The application discloses a control parameter setting value optimization method and device, computer equipment and a storage medium. Specifically, the application is a control parameter setting value optimization method, which comprises the following steps: training a machine learning model based on real historical data, wherein it is judged whether the machine learning model can keep consistent and reasonable prediction results under random interference; under the premise of ensuring the robustness of the machine learning model, finding a control parameter setting value capable of maximizing target benefits in a first environment (i.e. a system state) by using an optimization algorithm based on the machine learning model; and evaluating other alternative schemes (such as control parameter setting values manually adjusted by on-site personnel) in an interactive mode, which comprises the following steps: predicting the target benefits of the scheme recommended by the optimization algorithm and the alternative schemes based on the machine learning model, and calculating a second environment (i.e. the prediction target benefit of the alternative scheme is the maximum in the system state) in which the alternative scheme is the optimal decision. The scheme recommended by the optimization algorithm is different from the alternative scheme, and the first environment is different from the second environment. Through the above method, more information about decision optimization (expected benefits of different decisions and environments suitable for each decision) is provided, so as to assist the user to understand the logic of decision optimization and judge the rationality of the decision.

Description

Control parameter setting value optimization method, device, and computer equipment Technical Field

[0001] The present application relates to the field of optimization, and in particular, to a method, apparatus, computer device and storage medium for optimizing control parameter setting values. Background Art

[0002] Data-driven decision optimization has been widely used to solve industrial problems, such as product design and system control. Compared to rule-based or empirical decision-making, data-driven methods can handle nonlinear and time-varying parameters in dynamic systems, identify long-term or cumulative rewards, and automatically learn from updated data. However, most data-driven methods are black-box algorithms, and users may not understand or trust the recommended decisions in practice. The lack of interpretability of optimization algorithms can hinder the adoption of data-driven solutions.

[0003] Summary of the Invention

[0004] This summary is provided to introduce some selected concepts in a simplified form, which will be further described in the detailed description below. This summary is not intended to identify any key features or essential features of the claimed subject matter, nor is it intended to be used to help determine the scope of the claimed subject matter.

[0005] Based on this, an embodiment of the present application discloses a method for optimizing control parameter setting values, including:

[0006] Training a machine learning model based on real historical data, where the machine learning model is used to predict system dynamics; using statistical tests to evaluate the robustness of the machine learning model, where it is determined whether the machine learning model maintains consistent and reasonable prediction results under random interference;

[0007] Under the premise of ensuring the robustness of the machine learning model, finding the control parameter setting value that maximizes the target benefit based on the machine learning model through an optimization algorithm in the first environment;

[0008] Evaluate alternative plans in an interactive manner, including: predicting the target benefits of the plan recommended by the optimization algorithm and the alternative plan based on the machine learning model, and calculating the alternative plan as a second environment for the optimal decision; wherein the plan recommended by the optimization algorithm is different from the alternative plan, the first environment is different from the second environment, and the plan recommended by the optimization algorithm and the alternative plan include control parameter setting values.

[0009] Through the above method, more information on decision optimization (the benefits of different decisions and the applicable environment of each decision) is provided, thereby assisting users to understand the logic of decision optimization and judge the rationality of the decision.

[0010] Furthermore, statistical tests are used to evaluate the robustness of the machine learning model, including: calculating a first expected return sample group of historical actual control parameter setting values ​​through the machine learning model; calculating a second expected return sample group of randomly generated control parameter setting values ​​through the machine learning model; and comparing whether the first expected return sample group and the second expected return sample group belong to the same distribution.

[0011] In the above manner, the robustness of the machine learning model can be detected through the distribution situation. The method is relatively simple, direct and effective, which is conducive to the evaluation and consideration of the machine learning model, and also lays the foundation for the evaluation of decision-making plans in subsequent actual production.

[0012] Furthermore, the second environment in which the alternative solution is calculated as the optimal decision includes: finding a temporary applicable environment for the alternative solution through reverse calculation of the optimization algorithm; finding a solution recommended by the optimization algorithm through the optimization algorithm in the temporary applicable environment; based on the machine learning model, if the target benefit of the alternative solution in the temporary applicable environment is greater than the target benefit of the solution recommended by the optimization algorithm in the temporary applicable environment, then the temporary applicable environment is the second environment.

[0013] By this method, the second environment in which the alternative solution is the optimal decision can be iteratively calculated. This facilitates the subsequent evaluation and comparison of alternative solutions, further considers the differences from the system environment or the first environment, and is more conducive to making decisions on control parameters.

[0014] In addition, an embodiment of the present application discloses a control parameter setting value optimization device, which includes:

[0015] A prediction model training and evaluation module is used to train a machine learning model based on real historical data, wherein the machine learning model is used to predict system dynamic changes; and to evaluate the robustness of the machine learning model using statistical tests, wherein it is determined whether the machine learning model maintains consistent and reasonable prediction results under random interference;

[0016] An optimal control parameter setting value solving module is used to find the control parameter setting value that maximizes the target benefit based on the machine learning model through an optimization algorithm under the first environment while ensuring the robustness of the machine learning model;

[0017] A reverse solution module is used to interactively evaluate alternative solutions, including: based on the machine learning model, predicting the target benefits of the solution recommended by the optimization algorithm and the alternative solution, and calculating the alternative solution as the second environment for the optimal decision; wherein the solution recommended by the optimization algorithm is different from the alternative solution, the first environment is different from the second environment, and the solution recommended by the optimization algorithm and the alternative solution include control parameter setting values.

[0018] The present application also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the above method when executing the computer program.

[0019] The present application also provides a computer-readable storage medium having a computer program stored thereon, and the computer program implements the above method when executed by a processor.

[0020] The present application also provides a computer program product, which is tangibly stored on a computer-readable medium and includes computer-executable instructions. When the computer-executable instructions are executed, the computer-executable instructions cause at least one processor to perform the method described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Implementations of the present disclosure are illustrated by way of example and not limitation in the figures of the accompanying drawings in which like references indicate the same or similar parts.

[0022] FIG1 is a schematic diagram of a process of a control parameter setting value optimization method according to an embodiment of the present application.

[0023] FIG2 is a schematic diagram of a control parameter setting value optimization device according to an embodiment of the present application.

[0024] FIG3 is a schematic diagram of a computer device for optimizing control parameter settings according to an embodiment of the present application.

[0025] FIG4 is a schematic diagram of a control parameter setting value optimization system according to an embodiment of the present application

[0026] The reference numerals are as follows: S101-S103 Step 200: Apparatus 201: Module 202: Module 203: Module 204: Module 300: Computer device 302: Processor 304: Memory DETAILED DESCRIPTION

[0027] In the following description, for the purpose of explanation, a large number of specific details are set forth. However, it is understood that the present invention can be implemented without these specific details. In other examples, well-known circuits, structures, and technologies are not shown in detail so as not to affect the understanding of the description.

[0028] References throughout this specification to "an implementation," "an implementation," "an exemplary implementation," "some implementations," "various implementations," etc., indicate that the implementations of the invention being described may include particular features, structures, or characteristics. However, it does not imply that every implementation must include those particular features, structures, or characteristics. Furthermore, some implementations may have some, all, or none of the features described for other implementations.

[0029] In order to make data-driven decision optimization explainable and trustworthy in practical applications, an explainable decision optimization algorithm should help users understand which characteristics of the decision environment make a decision optimal, and which changes will lead to another decision becoming optimal.

[0030] The embodiments in this application propose a method to identify environmental features and environmental changes that will lead to different optimal decisions. Data-driven optimization methods usually use efficient search algorithms to find the optimal solution that maximizes the target benefit under given environmental and constraint conditions. The concept of counterfactual explanation is to indirectly explain the logic of decision optimization by answering the following question: Under what circumstances, alternative solutions (such as rule-based or experience-based solutions) will be better than the solutions given by data-driven algorithms. For example, a bank customer applies for a loan but is rejected. Counterfactual explanation can suggest which information the customer should modify to get the loan application approved. By providing environmental information applicable to different decisions, counterfactual explanation can assist users in understanding the logic of data-driven decision optimization and judging the rationality of the decision.

[0031] Since 2017, counterfactual explanations have been widely used to explain black-box classifiers such as support vector machines (SVMs), logistic regressors, and random forests. In 2021, inverse constraint programming, a hybrid approach combining mixed integer programming and cutting plane algorithms, was proposed for discrete optimization of linear optimization objectives. In 2023, a nearest neighbor prediction algorithm based on integer programming was developed to obtain nearest explanations for risk aversion problems.

[0032] These methods are not suitable for explaining continuous and nonlinear optimization problems. Therefore, in the embodiment of the present application, a method is proposed to obtain counterfactual explanations for all data-driven optimization problems (discrete, continuous, linear and nonlinear).

[0033] As shown in FIG1 , an embodiment of the present application provides a method for optimizing control parameter setting values, which includes:

[0034] S101, training a machine learning model based on real historical data, wherein the machine learning model is used to predict dynamic changes in the system; using statistical tests to evaluate the robustness of the machine learning model, wherein it is determined whether the machine learning model maintains consistent and reasonable prediction results under random interference.

[0035] Specifically, this application can be trained using any machine learning model with recognized performance.

[0036] Secondly, after the machine learning model is trained, the robustness of the machine learning model needs to be evaluated. Specifically, the Wald-Wolfowitz method is used for evaluation. The specific evaluation method is: for a given environment (i.e., system state), the corresponding control parameter setting values ​​are first selected from historical actual samples, and the expected benefits (fitness score or reward) of the control parameter setting values ​​are calculated using the machine learning model to obtain sample group 1; then, a control parameter setting value is randomly generated from the feasible domain (search space) of the control parameter, and the expected benefits (fitness score or reward) of the randomly generated control parameter setting value are calculated using the machine learning model to obtain sample group 2. The Wald-Wolfowitz test is performed on sample groups 1 and 2. If there is a statistically significant difference between the two sample groups (i.e., they come from different distributions), it proves that the machine learning model can still correctly predict system behavior under random interference (i.e., correctly distinguish between reasonable decisions and random guesses). Then, the machine learning model can be further used to calculate and solve the optimal control parameter setting values.

[0037] Furthermore, another expression of the Wald-Wolfowitz test is as follows:

[0038] p-value=Wald-Wolfowitz runs test(Array 1,Array 2)

[0039] If p-value <= 0.05, it means that there is a statistically significant difference between the two sample groups (i.e., they come from different distributions). Array1 represents the above sample group 1. Array2 represents the above sample group 2.

[0040] S102, under the premise of ensuring the robustness of the machine learning model, find the control parameter setting value that maximizes the target benefit through an optimization algorithm based on the machine learning model in the first environment.

[0041] The first environment refers to the current state of the system.

[0042] The above description can be expressed in another way as follows:

[0043] find X * stF(X * |Y∈Ψ)=Maximize F X∈Φ (X|Y∈Ψ).

[0044] in,

[0045] X represents the setting value of the control parameter to be solved.

[0046] X * Represents the optimal control parameter setting value recommended by the algorithm.

[0047] Y∈Ψ represents the first environment (i.e., the current state of the system).

[0048] Φ represents the feasible region of X.

[0049] F() represents the machine learning model of expected returns.

[0050] Maximize F() means maximizing expected returns.

[0051] The above formula can also be used to indicate that, under the first environment, the control parameter setting value recommended by the optimization algorithm has the greatest expected benefit.

[0052] In some embodiments, the machine learning model mentioned in S101 can be used to predict the benefits of different control parameter setting values

[0053] S103, interactively evaluating alternative solutions, including: predicting the target benefits of the solution recommended by the optimization algorithm and the alternative solution based on the machine learning model, and calculating a second environment in which the alternative solution is the optimal decision; wherein the solution recommended by the optimization algorithm is different from the alternative solution, and the first environment is different from the second environment. The solution recommended by the optimization algorithm includes control parameter settings calculated by the optimization algorithm; and the alternative solution includes subjectively given control parameter settings.

[0054] The above description can be expressed in another way as follows:

[0055] find Ψ * stF(X A |Y∈Ψ * )=Maximize F X∈Φ (X|Y∈Ψ * )

[0056] in,

[0057] X A Represents alternatives, which may be subjective control parameter settings given based on experience or according to rules.

[0058] Y∈Ψ * Indicates the environment in which the alternative is applicable, that is, the environment in which the expected benefit of the alternative is maximized

[0059] The above formula indicates that by solving the problem in reverse, we can obtain the environment in which the alternative plan is applicable and in which the expected benefit of the alternative plan is the largest.

[0060] The reverse solution is an iterative process. Each iteration verifies whether the alternative solution is the optimal solution under the current environment. If it is, the iteration stops; otherwise, it continues. Specifically, the evaluation criteria can be calculated as follows:

[0061] It is assumed that the return value is always non-negative. The above else case of -1 is explained Less than This means that the expected benefit of the alternative solution in the current environment is not maximized, and the alternative solution is not the optimal solution. This iterative process continues until an environment is found where the expected benefit of the alternative solution exceeds the expected benefit of the control parameter settings recommended by the optimization algorithm. This environment is the applicable environment for the alternative solution, also known as the second environment.

[0062] The iterative process is expressed as follows:

[0063] in:

[0064] Indicates the possible applicable environment of the found alternative solutions;

[0065] X ^ It represents the optimal control parameter setting value obtained by the optimization algorithm (S102) under the found environment.

[0066] The above step a indicates that, through the reverse calculation of the optimization algorithm, for alternative solution X A and the recommended result set X * Conduct temporary alternative X A Find the applicable environment and get

[0067] The above step b means that through the optimization algorithm, the optimal control parameter setting value X in the applicable environment found in step a is found. ^ .

[0068] The meaning of the above step c is to confirm that the alternative solution is the best in the environment found in step a. If the expected benefit of the alternative solution is greater than or equal to the expected benefit of the optimal control parameter setting value found by the optimization algorithm in step b, the iteration is stopped, and the environment is now the applicable environment of the alternative solution. If the above greater than or equal situation does not occur, it means that the applicable environment of the alternative solution has not been found, and it is necessary to continue searching. The optimal control parameter setting value X found by the optimization algorithm in step b ^ Will be included in the recommendation result set X * , participate in the calculation of step a.

[0069] The above steps are explained with a specific embodiment:

[0070] The background of this embodiment is: for testing in the production and manufacturing of automobile steering wheels, its control parameters include the outward reversal (Break-in-Force-Out), inward reversal (Break-in-Force-In), pre-tightening torque (Pre-Torque), and pre-tightening angle (Pre-Angle) of the first screw (FIV); and the outward reversal, inward reversal, pre-tightening torque, and pre-tightening angle of the second screw (RIV); a total of 8 parameters.

[0071] The response (predicted value, target benefit) includes the outward force (Strok-out-Force) and inward force (Stroke-in-Force) of the first screw and the outward force and inward force of the second screw.

[0072] The environment is a standard in which the outward force and the inward force of the first screw and the outward force and the inward force of the second screw pass actual engineering tests.

[0073] The above example can be further simplified as follows:

[0074] Model input: "Break-in-Force-Out", "Break-in-Force-In", "Pre-Torque", "Pre-Angle", "FIV" / "RIV"

[0075] Model output: "Stroke-in-Force", "Strok-out-Force"

[0076] Fitness score:SUM(1if model output in the interval,0else)

[0077] Among them, the interval of FIV is [180,200], and the interval of RIV is [120,140].

[0078] In this embodiment, according to the above steps, first, 3 permutation invariance Gaussian Process Neural Networks are trained as a machine learning model (ensemble of 3permutation invariance Gaussian Process Neural Networks is trained). The target benefit is predicted using the average value of the output distribution, and the predicted target benefit result is consistent with the actual system change pattern. That is, the expected benefit of the actual control parameter setting value and the expected benefit of the randomly generated control parameter setting value are calculated using the machine learning model. The expected benefit of the actual control parameter setting value is significantly higher than the expected benefit of the randomly generated control parameter setting value, indicating that the machine learning model can correctly predict the benefit and pass the robustness evaluation.

[0079] Secondly, based on the machine learning model and optimization algorithm, the optimal control parameter setting values ​​under the maximum benefit (i.e., one-time test passing) are found.

[0080] In this embodiment, the optimization algorithm adopts the covariance matrix adaptation evolution strategy (CMA-ES) to find the optimal control parameter setting value under the current first environment (system environment). In some embodiments, there can be multiple optimization algorithms (such as manual adjustment based on experience, rule-based, model-based, etc.), each of which provides an optimal control parameter setting value recommended by the current optimization algorithm under the first environment. The recommended control parameter setting values ​​of multiple optimization algorithms can be evaluated by the machine learning model (i.e., the target benefit is predicted), thereby determining the optimal control parameter setting value, whose target benefit is the largest among the control parameter setting values ​​recommended by multiple optimization algorithms.

[0081] Optimized configuration includes:

[0082] Decision variables, such as eight control parameter settings;

[0083] Initial point, for example, the average value of historical samples is [100, 100, 0, 400, 80, 100, 0, 0]);

[0084] The feasible region of the decision variables, for example, [100, 100, 0, 400, 80, 100, 9, 0], [300, 500, 10, 800, 180, 300, 10, 100]];

[0085] The standard deviation of the population, for example, [30, 30, 1, 50, 20, 30, 1, 10];

[0086] The size of each population, for example, 101,

[0087] The number of iterations, for example 50.

[0088] Then, after 20 iterations, the model found the optimal control parameter settings. Furthermore, the optimal control parameter settings found by the optimization algorithm were verified to be reasonable and reliable, as follows:

[0089] - The predicted responses (target gains) for the optimal control parameter settings lie within the distribution of actual responses across the tests.

[0090] - The optimal control parameter setting lies within the distribution of actual setting values ​​tested.

[0091] The last step is counterfactual explanation, reverse optimization problem or finding an applicable environment for the alternative solution (i.e., the test pass standard in this embodiment).

[0092] For some control parameters, the range of control parameter settings based on experience is very large. For example, the range of RIV Break-in-Force-In is [80, 160], while the algorithm recommends a value of 80. If the alternative solution considers RIV Break-in-Force-In to be 120, the situation is as follows:

[0093] Assume that the control parameter settings in the device selection scheme are as follows:

[0094] FIV=[203.2,341.39,200,703.49]

[0095] RIV=[120, 237.1, 3.83, 30]

[0096] Compared with the control parameter settings recommended by the optimization algorithm:

[0097] FIV=[203.2,341.39,7.87,703.49]

[0098] RIV=[81.14,237.1,3.83,5.26]

[0099] The specific parameters that are different are marked in bold.

[0100] Next, we analyze how the applicable environment can change so that the expected benefits of the alternative solutions are better than the expected benefits of the control parameter settings recommended by the optimization algorithm.

[0101] Using the above step S103, it can be found that:

[0102] For the first environment (the applicable environment for the algorithm recommendation solution), that is, the standard for passing the test (for example, FIV is [180, 200], RIV is [120, 140]), the response of the alternative solution (predicted value, target return) will exceed the acceptance area, and the test will fail, which proves that the optimization algorithm recommendation solution is better than the alternative solution.

[0103] -For the applicable environment of the alternative solution (for example, FIV is [156,242], RIV is [126,151]), the response (predicted value, target return) of the algorithm-recommended solution exceeds the acceptance region, the test fails, and the alternative solution is better than the algorithm-recommended solution.

[0104] - The changes in the circumstances under which different decisions are applied suggest that increasing the pre-tightening torque (PreTorque) of FIV may reduce the outward force (Stroke-out-Force), while increasing the inward change (Break-in-Force-In) and pre-tightening torque (PreTorque) of RIV may increase both the outward force (Stroke-out-Force) and the inward force (Stroke-in-Force).

[0105] The effective effects and advantages of this application are as follows:

[0106] Given two decisions, counterfactual explanations can reveal the circumstances under which one decision would be better than the other, and then comparatively explain and emphasize the environmental characteristics that made a particular decision optimal. Counterfactual explanations can help data-driven optimization methods become more transparent and trustworthy in practice, and promote the application of AI-driven control optimization in industrial operations.

[0107] Existing counterfactual explanation methods are designed and developed for explaining black-box classifiers, nearest neighbor prediction algorithms, or discrete optimization with linear objectives. Currently, there are no methods designed and developed for continuous and nonlinear optimization problems.

[0108] The methods provided in the embodiments of the present application can obtain robust optimization results and counterfactual explanations for all types of data-driven optimization methods (discrete, continuous, linear, and nonlinear). In addition, the framework is very flexible and can be incorporated into any machine learning model and optimization algorithm.

[0109] How to check whether the embodiments described in this application are used can be done by the following method:

[0110] - Whether the solution verifies the robustness of the decision optimization process to random disturbances

[0111] - Whether the solution provides a counterfactual explanation

[0112] -Does the solution provide a visual counterfactual explanation?

[0113] - Whether the solution is limited to certain specific optimization problems, such as linear, discrete, etc.

[0114] FIG2 provides an optimization device 200 for setting values ​​of control parameters. The device 200 includes:

[0115] The prediction model training and evaluation module 201 is used to train a machine learning model based on real historical data, wherein the machine learning model is used to predict dynamic changes in the system; and use statistical tests to evaluate the robustness of the machine learning model, wherein it is determined whether the machine learning model maintains consistent and reasonable prediction results under random interference.

[0116] An optimal control parameter setting value solving module 202 is configured to find a control parameter setting value that maximizes a target benefit based on the machine learning model under a first environment through an optimization algorithm while ensuring the robustness of the machine learning model;

[0117] The reverse solution module 203 is used to interactively evaluate alternative solutions, including: based on the machine learning model, predicting the target benefits of the solution recommended by the optimization algorithm and the alternative solution, and calculating the alternative solution as the second environment for the optimal decision; wherein the solution recommended by the optimization algorithm is different from the alternative solution, and the first environment is different from the second environment.

[0118] Furthermore, the prediction model training and evaluation module 201 is used to:

[0119] Calculating a first expected return sample group of historical actual control parameter setting values ​​using the machine learning model;

[0120] A second expected return sample group of control parameter setting values ​​randomly generated by calculating the machine learning model;

[0121] Compare the first expected return sample group and the second expected return sample group to see whether they belong to the same distribution.

[0122] Furthermore, the reverse solution module 203 is further used to:

[0123] Finding a temporary applicable environment for the alternative solution through reverse calculation of the optimization algorithm;

[0124] Under the temporary applicable environment, searching for a solution recommended by the optimization algorithm through the optimization algorithm;

[0125] Based on the machine learning model, if the target benefit of the alternative solution in the temporary applicable environment is greater than the target benefit of the solution recommended by the optimization algorithm in the temporary applicable environment, then the temporary applicable environment is the second environment.

[0126] As shown in Figure 4, the present application provides an embodiment of the application of the control parameter setting value optimization device in an actual system. The Plant represents the device actually controlled by the control parameters. The Perception Subsystem extracts information data from the Plant and its surrounding environment through sensors, and performs signal processing of the sensor array and data fusion. In the Cognition Subsystem, corresponding decisions are made, such as the setting values ​​of the control parameters recommended by the above-mentioned optimization algorithm, the explanation of the optimization strategy (reverse optimization problem), and other strategy planning and model learning. The machine learning model is also in this subsystem. In the Actuation Subsystem, the control parameters are executed on the above-mentioned Plant.

[0127] It should be noted that the apparatus may include more or fewer modules to implement the described functionality. For example, at least one module in FIG. 2 may be further divided into a plurality of different submodules, each of which is configured to perform at least a portion of the operations described herein in conjunction with the corresponding module. Furthermore, in some examples, the apparatus 200 may further include additional modules for performing other operations already described in the specification. Furthermore, those skilled in the art will appreciate that the exemplary apparatus 200 may be implemented using software, hardware, firmware, or any combination thereof.

[0128] Figure 3 provides a computer device. According to one embodiment, the computer device 300 may include a processor 302, which executes a computer program stored in a memory 304. When the computer program is executed by the processor, the above method is implemented.

[0129] It should be understood that although the various steps in the flowchart of FIG1 are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in FIG1 may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0130] Those skilled in the art will understand that the structure shown in Figure 3 is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0131] Those skilled in the art will appreciate that all or part of the processes in the methods for implementing the above-mentioned embodiments can be accomplished by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include processes for the implementation of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the various embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0132] The present application also provides a computer-readable storage medium having a computer program stored thereon, which implements the above steps when the computer program is executed by a processor.

[0133] The present application also provides a computer program product, which is tangibly stored on a computer-readable medium and includes computer-executable instructions. When the computer-executable instructions are executed, the computer-executable instructions cause at least one processor to perform the above method.

[0134] Furthermore, the computer program can be stored and run in the cloud to perform the method. Furthermore, the components of the program can be deployed on multiple devices and the cloud. For example, the corresponding steps can be deployed and run on a local or local computer, or run on different cloud devices, and transmit signals through a communication connection, or can also be deployed and run on a local or local computer. This application does not limit the manner or method, and the corresponding technology can be flexibly deployed to make full use of equipment and technologies such as the cloud, big data, and supercomputing capabilities to execute and complete the method.

[0135] Some implementations of the present disclosure may include articles of manufacture. Articles of manufacture may include storage media for storing logic. Examples of storage media may include one or more types of computer-readable storage media capable of storing electronic data, including volatile or non-volatile memory, removable or non-removable memory, erasable or non-erasable memory, writable or rewritable memory, and the like. Examples of logic may include various software units, such as software components, programs, applications, computer programs, application programs, system programs, machine programs, operating system software, middleware, firmware, software modules, routines, subroutines, functions, methods, processes, software interfaces, application program interfaces (APIs), instruction sets, computing codes, computer codes, code segments, computer code segments, words, values, symbols, or any combination thereof. In some implementations, for example, articles of manufacture may store executable computer program instructions that, when executed by a processor, cause the processor to perform the methods and / or operations described herein. Executable computer program instructions may include any suitable type of code, such as source code, compiled code, interpreted code, executable code, static code, dynamic code, and the like. Executable computer program instructions can be implemented according to a predefined computer language, method or syntax for commanding a computer to perform a specific function. The instructions can be implemented using any appropriate high-level, low-level, object-oriented, visual, compiled and / or interpreted programming language.

[0136] What has been described above includes examples of the disclosed architecture. It is, of course, not possible to describe every conceivable combination of components and / or methodologies, but those skilled in the art will appreciate that many other combinations and permutations are possible. Therefore, the novel architecture is intended to embrace all such alternatives, modifications, and variations that fall within the spirit and scope of the appended claims.

Claims

1. A method for optimizing control parameter settings, wherein: include: Training a machine learning model based on real historical data, where the machine learning model is used to predict system dynamics; using statistical tests to evaluate the robustness of the machine learning model, where it is determined whether the machine learning model maintains consistent and reasonable prediction results under random interference; Under the premise of ensuring the robustness of the machine learning model, finding the control parameter setting value that maximizes the target benefit based on the machine learning model through an optimization algorithm in the first environment; Evaluate alternative plans in an interactive manner, including: predicting the target benefits of the plan recommended by the optimization algorithm and the alternative plan based on the machine learning model, and calculating the alternative plan as a second environment for the optimal decision; wherein the plan recommended by the optimization algorithm is different from the alternative plan, the first environment is different from the second environment, and the plan recommended by the optimization algorithm and the alternative plan include control parameter setting values.

2. The method according to claim 1, wherein Statistical tests were used to evaluate the robustness of the machine learning model, including: Calculating a first expected return sample group of historical actual control parameter setting values ​​using the machine learning model; A second expected return sample group of control parameter setting values ​​randomly generated by calculating the machine learning model; Compare the first expected return sample group and the second expected return sample group to see whether they belong to the same distribution.

3. The method according to claim 1, wherein The second environment for calculating alternative solutions to the optimal decision includes: Finding a temporary applicable environment for the alternative solution through reverse calculation of the optimization algorithm; Under the temporary applicable environment, searching for a solution recommended by the optimization algorithm through the optimization algorithm; Based on the machine learning model, if the target benefit of the alternative solution in the temporary applicable environment is greater than the target benefit of the solution recommended by the optimization algorithm in the temporary applicable environment, then the temporary applicable environment is the second environment.

4. An optimization device (200) for setting a control parameter, wherein: include: A prediction model training and evaluation module (201) is used to train a machine learning model based on real historical data, wherein the machine learning model is used to predict system dynamic changes; and to evaluate the robustness of the machine learning model using statistical tests, wherein it is determined whether the machine learning model maintains consistent and reasonable prediction results under random interference; An optimal control parameter setting value solving module (202) is used to find the control parameter setting value that maximizes the target benefit based on the machine learning model through an optimization algorithm under the first environment while ensuring the robustness of the machine learning model; A reverse solution module (203) is used to interactively evaluate alternative solutions, including: based on the machine learning model, predicting the target benefits of the solution recommended by the optimization algorithm and the alternative solution, and calculating the alternative solution as the second environment for the optimal decision; wherein the solution recommended by the optimization algorithm is different from the alternative solution, the first environment is different from the second environment, and the solution recommended by the optimization algorithm and the alternative solution include control parameter setting values.

5. The apparatus (200) of claim 4, wherein the prediction model training and evaluation module (201) is further configured to: Calculating a first expected return sample group of historical actual control parameter setting values ​​using the machine learning model; A second expected return sample group of control parameter setting values ​​randomly generated by calculating the machine learning model; Compare the first expected return sample group and the second expected return sample group to see whether they belong to the same distribution.

6. The device according to claim 4, wherein The reverse solution module (203) is further used to: Finding a temporary applicable environment for the alternative solution through reverse calculation of the optimization algorithm; Under the temporary applicable environment, searching for a solution recommended by the optimization algorithm through the optimization algorithm; Based on the machine learning model, if the target benefit of the alternative solution in the temporary applicable environment is greater than the target benefit of the solution recommended by the optimization algorithm in the temporary applicable environment, then the temporary applicable environment is the second environment.

7. [Corrected 15.04.2024 under Rule 91] A computer device comprising a memory and a processor, the memory storing a computer program, wherein When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.

8. [Corrected 15.04.2024 in accordance with Rule 91] A computer-readable storage medium having a computer program stored thereon, wherein When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

9. [Corrected 15.04.2024 according to Rule 91] A computer program product tangibly stored on a computer-readable medium and comprising computer-executable instructions which, when executed, cause at least one processor to perform the method according to any one of claims 1 to 3.

Images