Industrial multi-objective hierarchical optimization method and device, electronic equipment and storage medium

Abstract

The application provides an industrial multi-objective hierarchical optimization method and device, electronic equipment and storage medium, belonging to the technical field of resource planning, distribution, allocation or scheduling of enterprises or organizations, comprising: obtaining initial real-time production data in an industrial production process, preprocessing the initial real-time production data to obtain target real-time production data; performing feature extraction on the target real-time production data to obtain decision variables, and determining a target function and a constraint condition based on the decision variables; and using an industrial multi-objective optimization algorithm based on dynamic multi-dimensional hierarchical and adaptive neighborhood information feedback, combining hardware accelerated parallel computing, to solve the target function, so as to realize industrial multi-objective hierarchical optimization. The application can enhance dynamic optimization performance and improve the efficiency and accuracy of industrial multi-objective optimization in a complex industrial production environment.

Description

Technical Field

[0001] This invention relates to the field of resource planning, allocation, distribution or scheduling technology for enterprises or organizations, and in particular to an industrial multi-objective hierarchical optimization method, device, electronic device and storage medium. Background Technology

[0002] With the continuous advancement of industrialization, multi-objective optimization has become a pressing technical challenge in high-energy-consuming and high-precision core industrial scenarios such as steel rolling, chemical synthesis, and automobile manufacturing. These complex production environments often face the synergistic challenge of multiple optimization demands—improving energy efficiency and ensuring product quality stability while simultaneously maintaining efficient production pace and meeting the performance standards of core materials. Since these objectives may conflict, finding the optimal solution by balancing multiple objectives while adhering to process constraints is crucial for promoting intelligent industrial upgrading.

[0003] Traditional multi-objective optimization methods, such as NSGA-II and NSGA-III based on genetic algorithms, still face many limitations in practical industrial applications. Firstly, traditional optimization algorithms typically use fixed hierarchical structures of objectives or static optimization models, which makes them poorly adaptable to dynamically changing production environments. For example, in steel rolling, process parameters such as temperature and pressure frequently change due to environmental variations or equipment malfunctions. These changes prevent algorithms based on static optimization models from adjusting and adapting in a timely manner, resulting in unsatisfactory optimization performance. Summary of the Invention

[0004] To address the above problems, this invention provides an industrial multi-objective hierarchical optimization method, apparatus, electronic device, and storage medium, which enhances dynamic optimization performance and improves the efficiency and accuracy of industrial multi-objective optimization in complex industrial production environments.

[0005] This invention provides an industrial multi-objective hierarchical optimization method, comprising:

[0006] Acquire initial real-time production data during the industrial production process, preprocess the initial real-time production data, and obtain target real-time production data;

[0007] Feature extraction is performed on the real-time production data of the target to obtain decision variables, and the objective function and constraints are determined based on the decision variables;

[0008] An industrial multi-objective optimization algorithm based on dynamic multi-dimensional hierarchical structure and adaptive neighborhood information feedback is used, combined with hardware-accelerated parallel computing, to solve the objective function and achieve industrial multi-objective hierarchical optimization.

[0009] The industrial multi-objective optimization algorithm based on dynamic multi-dimensional hierarchical layering and adaptive neighborhood information feedback comprises a dynamic multi-dimensional hierarchical layering mechanism and an adaptive neighborhood information feedback mechanism.

[0010] Optionally, the initial real-time production data in the industrial production process is obtained by obtaining temperature, pressure, speed, production progress, and equipment state data, and the initial real-time production data is preprocessed by data cleaning, denoising, missing value filling, standardization, and normalization.

[0011] Optionally, before determining the objective function and the constraint condition based on the decision variable, all possible solutions of the objective function are set to form a population, and the population is initialized wherein the population is composed of individuals, and the i-th individual is represented by , i = 1, 2, …, n, where n is the number of decision variables; , i = 1, 2, …, n, where n is the number of decision variables; is the j-th decision variable of the i-th individual.

[0012] Optionally, the objective function is:

[0013]

[0014] wherein is the i-th objective function, which is determined by different combinations of decision variables

[0015] The constraint condition is:

[0016]

[0017] wherein is the i-th process constraint, = 1, 2, …, m, and m is the number of inequality constraints.

[0018] Optionally, the division formula of the target subspace is:

[0019] ​​​​​​

[0020] In the formula, Indicates the first The objective subspace is a high-dimensional space composed of the weighted sum of the objective functions. This high-dimensional space contains all solutions that satisfy the specific objective constraints. To ensure that, under the current optimization conditions, the objective weighted combination in the solution space does not exceed a threshold. The set of all solutions; These are the dynamic weights of each objective function, representing the importance of different objective functions in the optimization process. A reference value is... And the sum of these values ​​is 1. It is the first The threshold of each objective subspace is used to limit the upper limit of the weighted objective function value. It is dynamically adjusted in each generation of optimization to control the range of the objective and ensure that the weighted sum of the objective function of each solution does not exceed the threshold. It is the space of decision variables.

[0021] Optionally, the formula for calculating the neighborhood distance is:

[0022]

[0023] In the formula, Individual with neighboring individuals Neighborhood distance between two individuals measures the similarity between them in the target space and the decision space. It is the number of individuals in the neighborhood; It is an index adjustment factor that controls the weighting of differences in the objective function; It is the sensitivity of controlling the spatial distance to the target; It is the first The weights of each decision variable reflect the importance of that decision variable in calculating the neighborhood distance; It controls the degree to which the differences in decision variables contribute to the neighborhood distance; It adjusts the influence of decision space differences on neighborhood distance calculation, and is used to adjust the sensitivity of the decision space; It is a weighting factor used to adjust the influence of the combined effect of the objective function and the decision space difference; These are non-linear weighting coefficients used to adjust the influence of differences between objective function values; It is a nonlinear influence coefficient that controls the differences in the decision space, used to strengthen or weaken the contribution of the differences between solutions in the decision space to distance calculation.

[0024] Optionally, the update formula for an individual is:

[0025]

[0026] In the formula, is the current time The decision variable of the individual; is the updated decision variable of the individual; is the updated decision variable of the individual; is the learning rate or step factor, which controls the step size in the updating process and determines the relative influence of the feedback from the objective space and the decision space in the updating.

[0027] Optionally, FPGA acceleration card is used for hardware acceleration and parallel computing.

[0028] The industrial multi-objective hierarchical optimization device provided by the application comprises a real-time production data acquisition module, an objective function determination module and an industrial multi-objective hierarchical optimization module.

[0029] The real-time production data acquisition module is used for acquiring initial real-time production data in an industrial production process, pre-processing the initial real-time production data and obtaining target real-time production data.

[0030] The objective function determination module is used for extracting features from the target real-time production data, obtaining decision variables and determining an objective function and constraint conditions based on the decision variables.

[0031] The industrial multi-objective hierarchical optimization module is used for solving the objective function by using an industrial multi-objective optimization algorithm based on dynamic multi-dimensional hierarchical and adaptive neighborhood information feedback and combining hardware acceleration and parallel computing, so as to realize industrial multi-objective hierarchical optimization.

[0032] The industrial multi-objective optimization algorithm based on dynamic multi-dimensional hierarchical and adaptive neighborhood information feedback comprises a dynamic multi-dimensional hierarchical mechanism and an adaptive neighborhood information feedback mechanism, the dynamic multi-dimensional hierarchical mechanism adopts a dynamic hierarchical strategy, adjusts the hierarchical structure of the objective function in each generation in real time according to the distribution of solutions in the objective space and actual optimization requirements, and divides the objective space based on the weighted sum of the objective function in each generation, so that each objective space is divided into a plurality of objective subspaces; the adaptive neighborhood information feedback mechanism realizes a search process based on the neighborhood distance between individuals in the objective subspaces and neighborhood individuals, and the neighborhood individuals are taken from the objective subspaces.

[0033] The electronic device provided by the application comprises a memory, a processor and a computer program stored in the memory, and the processor executes the computer program to realize the steps of the industrial multi-objective hierarchical optimization method.

[0034] The application provides a storage medium, which has a computer program stored thereon, and the computer program is executed by a processor to realize steps of the industrial multi-objective hierarchical optimization method.

[0035] The application provides an industrial multi-objective hierarchical optimization method, device, electronic equipment and storage medium, which can achieve the following beneficial effects:

[0036] 1. The industrial multi-objective optimization algorithm based on dynamic multi-dimensional hierarchical and adaptive neighborhood information feedback improves adaptability, calculation efficiency and optimization accuracy by introducing a dynamic multi-dimensional hierarchical mechanism, an adaptive neighborhood information feedback mechanism and hardware acceleration parallel computing.

[0037] 2. The dynamic multi-dimensional hierarchical mechanism avoids the limitation of fixed hierarchical structure on optimization by adjusting the division of the target space in real time. In each generation, the target subspace is dynamically divided according to the distribution of solutions in the target space and optimization requirements, so that the optimization of each subspace is more targeted, and redundant calculation and invalid search in global search are avoided. Through the adaptive neighborhood information feedback mechanism, the distance between individuals and neighborhood individuals is calculated in the search process, and the search direction is guided according to the information, further improving the global search ability, which can effectively avoid falling into a local optimal solution in the search process and improve the quality of the optimization solution.

[0038] 3. By introducing hardware acceleration-based parallel computing, the calculation speed of the individual target function and constraint condition evaluation is greatly improved. Through parallel computing, the target function and constraint condition of multiple individuals can be calculated simultaneously in each generation evolution process, thereby significantly shortening the calculation time and improving the overall calculation efficiency of the algorithm, especially when the population size is large or the target function calculation is complex, the acceleration effect is particularly significant. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 FIG. 1 is a flowchart of an industrial multi-objective hierarchical optimization method according to an embodiment of the application.

[0040] Figure 2 FIG. 2 is a structural schematic diagram of an industrial multi-objective hierarchical optimization device according to an embodiment of the application.

[0041] Legend: 10, real-time production data acquisition module; 20, target function determination module; 30, industrial multi-objective hierarchical optimization module. DETAILED DESCRIPTION

[0042] The following will be described in detail in combination with specific embodiments and accompanying drawings. Figure 1 -Appendix Figure 2 The application is described in detail to enable those skilled in the art to more fully understand the purpose, features and effects of the application.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. If there is a conflict between the definitions of terms in this disclosure and that understood by those of ordinary skill in the art, the definitions provided in this disclosure control.

[0044] The present application provides an industrial multi-objective hierarchical optimization method, device, electronic equipment and storage medium, which solves the technical problems of low efficiency and poor accuracy in complex industrial production environment.

[0045] Embodiment one

[0046] As a specific embodiment of the present application, the embodiment provides an industrial multi-objective hierarchical optimization method, which refers to Figure 1 , the specific steps are as follows:

[0047] S100, obtaining initial real-time production data in an industrial production process, preprocessing the initial real-time production data to obtain target real-time production data;

[0048] S200, performing feature extraction on the target real-time production data to obtain decision variables, and determining a target function and a constraint condition based on the decision variables;

[0049] S300, using an industrial multi-objective optimization algorithm based on dynamic multi-dimensional hierarchical and adaptive neighborhood information feedback, combining hardware accelerated parallel computing, solving the target function to realize industrial multi-objective hierarchical optimization.

[0050] The industrial multi-objective hierarchical optimization method of the embodiment realizes the combination of dynamic hierarchical mechanism and adaptive neighborhood information feedback mechanism by introducing dynamic multi-dimensional hierarchical and adaptive neighborhood information feedback, so that the method can adaptively adjust the algorithm structure when dealing with complex multi-objective optimization problems, thereby providing flexibility and adjustability. This flexibility makes the method particularly suitable for scenarios that require continuous adjustment of optimization strategies in industrial production processes, such as steel rolling.

[0051] In the following, the embodiment takes the production scenario of steel rolling as an example to explain the above technical solutions in detail.

[0052] Specifically, in S100, the obtained initial real-time production data in the industrial production process includes temperature, pressure, speed, energy consumption, production progress, and equipment state data in the steel rolling production process. The initial real-time production data can be obtained in real time by sensors, PLCs (Programmable Logic Controllers), SCADAs (Supervisory Control And Data Acquisition systems) and other devices located on the steel rolling production line.

[0053] The steel rolling production process generally includes billet preparation, heating, rolling, finishing, which involves various equipment, such as a heating furnace for heating the billet, a rough rolling mill for preliminary thinning, a finishing mill, a laminar cooling system, a coiling machine, etc. For the heating furnace, the production data involved include heating temperature, heating time, furnace pressure, heating rate, etc. For the rough rolling mill, the rolling diameter, rolling speed, rolling force, etc. are involved. In the cooling stage, the cooling speed, cooling water pressure / flow, and cooling termination temperature are involved. Various real-time production data in the steel rolling production process are acquired.

[0054] Further, the initial real-time production data of the steel rolling production is preprocessed to obtain target real-time production data, including data cleaning, denoising, missing value filling, standardization and normalization of the initial real-time production data to obtain the target real-time production data.

[0055] In this embodiment, through data cleaning, incomplete or abnormal records are removed, such as error data caused by sensor failure or invalid data caused by inconsistent acquisition time. Through denoising, the real signal or effective information is extracted from the mixed data, so that the data is cleaner and more reliable, thereby facilitating subsequent analysis, processing and application. Optionally, a filtering algorithm is used for denoising. Through missing value filling, the data integrity is improved. Optionally, an interpolation method or a statistical method based on historical production data called from an existing database is used to fill the missing data, wherein the interpolation method can use linear interpolation or nearest neighbor interpolation. Through standardization and normalization, all production data are adjusted to the same order of magnitude. Optionally, Z-score standardization method and Min-Max normalization method are used to scale the data value to the interval [0, 1], so as to eliminate the scale difference between different variables. The specific method for preprocessing the initial real-time production data given in this embodiment is a method already known in the prior art, and therefore will not be described here.

[0056] Specifically, in S200, feature extraction is performed on the target real-time production data, and the decision variables obtained include temperature, pressure, speed, energy consumption, and time in the steel rolling production process. The temperature refers to the temperature that needs to be controlled in the steel rolling production, the pressure refers to the pressure applied in the steel rolling production, the speed refers to the running speed in the steel rolling production, the energy consumption refers to the energy consumption in the steel rolling production, and the time refers to the life cycle time of the steel rolling production.

[0057] Optionally, for the target real-time production data in the steel rolling process, the most relevant features to the target optimization are extracted by using machine learning or statistical methods and existing feature engineering technologies such as principal component analysis, and the most relevant features are taken as the decision variables of the optimization problem.

[0058] In another possible embodiment, the principal component analysis method is used to extract the most relevant features to the target optimization as the decision variables, which are respectively: rolling temperature, rolling pressure, rolling speed, energy consumption, equipment load.

[0059] In the present embodiment, the decision variables are used as the basis for determining the objective function and the constraint conditions of the steel rolling production optimization. Preferably, the decision variables are initialized before determining the objective function and the constraint conditions of the steel rolling production optimization. Through the initialization, the dimensional problem is solved, and the scale difference between different variables is eliminated.

[0060] In the initialization, the population is first initialized , that is, all possible solutions of the objective function are set to form a population, wherein the population is composed of individuals, and each individual is an arbitrary solution composed of the decision variables, and is represented by , the th individual, is the number of decision variables; is the th decision variable of the th individual.

[0061] For each individual, the objective function and the constraint conditions are determined, the objective function is the target to be optimized, including different optimization targets, such as energy efficiency, material strength, production speed, etc. related to the steel rolling production, and the objective function needs to be maximized under the premise of maintaining the production process requirements;

[0062]

[0063] In the formula, is the th objective function, which is determined by different combinations of the decision variables , and is determined according to the actual production requirements of the steel rolling, such as minimization of energy efficiency, maximization of product strength, etc.

[0064] In a possible embodiment, the optimization targets are:

[0065] maximization of rolling strength;

[0066] minimization of unit energy consumption;

[0067] optimal control of process tempo (speed);

[0068] minimization of heat loss;

[0069] optimization of temperature uniformity.

[0070] Each objective function needs to be associated with a corresponding steel rolling process constraint The constraints are jointly acting, such as temperature, pressure range, and in this embodiment, inequality constraints are adopted:

[0071]

[0072] In the formula, is the first process constraint, =1, 2, …, m, m is the number of inequality constraints. For example, in an example, the finishing mill rated rolling force is 30 MN, which determines that the maximum rolling force of the process cannot exceed 30 MN.

[0073] In another possible embodiment, the specific industrial constraints are:

[0074] The pressure interval is 150-350 MPa;

[0075] The temperature control accuracy is ±15℃;

[0076] The rolling speed cannot exceed the upper limit of 20 m / s;

[0077] The equipment load cannot exceed 90%.

[0078] The constraints ensure that the solution meets the process requirements of actual steel rolling production in the optimization process.

[0079] Specifically, in S300, in order to adapt to high-dimensional objectives, diversified constraints, and dynamically changing production environments, multi-objective hierarchical optimization is performed through an industrial multi-objective optimization algorithm based on dynamic multi-dimensional hierarchical and adaptive neighborhood information feedback. At the same time, combined with hardware acceleration parallel computing, the calculation efficiency is improved.

[0080] In order to handle the conflict of objective functions in the multi-objective optimization of steel rolling production, the dynamic multi-dimensional hierarchical mechanism is adopted in this embodiment. Unlike the pre-set static hierarchical structure in traditional algorithms, the dynamic multi-dimensional hierarchical mechanism adopts a dynamic hierarchical strategy, which adjusts the hierarchical structure of the objective function in each generation in real time according to the distribution of solutions in the objective space and actual optimization requirements. In each generation, the objective space is divided based on the weighted sum of the objective functions, and each objective space is divided into several objective subspaces. Wherein, each generation refers to one complete cycle of generating a new group of individuals from the current population and completing complete evolution operations (such as evaluation, selection, etc.).

[0081] Preferably, the division formula of the objective subspace is:

[0082]

[0083] In the formula, indicates the first Each objective subspace is a high-dimensional space composed of the weighted sum of the objective functions. This high-dimensional space contains all solutions, i.e., individuals, that satisfy the specific objective constraints. This can be viewed as, under the current optimization conditions, ensuring that the objective weighted combination in the solution space does not exceed a threshold. The set of all solutions; These are the dynamic weights of each objective function, representing the importance of different objective functions in the optimization process. They are determined based on specific needs, with a reference value of [value missing]. And the sum of these values ​​is 1. It is the first The threshold of each objective subspace is used to limit the upper limit of the weighted objective function value. It is dynamically adjusted in each generation of optimization to control the range of the objective and ensure that the weighted sum of the objective function of each solution does not exceed the threshold. It is determined based on engineering requirements and equipment specifications. It is the space of decision variables, one A real space of 3D represents the space of decision variables.

[0084] The dynamic multidimensional hierarchical mechanism avoids the limitations of a fixed hierarchical structure on optimization by adjusting the partitioning of the target space in real time. In each generation, the target subspace is dynamically divided according to the distribution of solutions in the target space and the optimization requirements, making the optimization of each subspace more targeted and avoiding redundant calculations and ineffective searches in the global search.

[0085] Furthermore, to improve the global search capability during the optimization process and avoid the pitfalls of local optima, an adaptive neighborhood information feedback mechanism is employed after determining the target subspace. This mechanism calculates the target subspace... medium-sized individuals with neighboring individuals The search process is carried out based on the neighborhood distance between the target individuals, wherein the neighborhood individuals are taken from the target subspace. The neighborhood distance metric considers not only the difference in target values ​​in the target space, but also the difference in variables in the decision space.

[0086] Preferably, the neighborhood distance calculation formula is as follows:

[0087]

[0088] In the formula, Individual with neighboring individuals The neighborhood distance between two individuals measures the similarity between them in the target space and the decision space, and is mainly used to determine the search direction and the selection in the evolutionary process. It is the number of individuals in the neighborhood; It is an index adjustment factor that controls the weighting of differences in the objective function and is determined according to specific needs; It is the sensitivity of controlling the spatial distance to the target; It is the first The weights of each decision variable reflect the importance of that decision variable in calculating the neighborhood distance; It controls the degree to which the differences in decision variables contribute to the neighborhood distance; It adjusts the influence of decision space differences on neighborhood distance calculation, and is used to adjust the sensitivity of the decision space; It is a weighting factor used to adjust the influence of the combined effect of the objective function and the decision space difference. These are non-linear weighting coefficients used to adjust the influence of differences between objective function values; It is a nonlinear influence coefficient that controls the differences in the decision space, used to strengthen or weaken the contribution of the differences between solutions in the decision space to distance calculation.

[0089] The aforementioned neighborhood distance calculation takes into account the distance difference between the target space and the decision space, which can effectively feed back neighborhood information and enable individual search to develop in a better direction.

[0090] Driven by the aforementioned adaptive neighborhood information feedback mechanism, the update formula for an individual is:

[0091]

[0092] In the formula, It is the current moment. No. Decision variables for each individual; It is the updated version of the first Decision variables for each individual; It is the learning rate or step size factor, which controls the step size in the update process and determines the relative impact of feedback between the target space and the decision space on the update.

[0093] By employing an adaptive neighborhood information feedback mechanism, the distance between an individual and its neighbors is continuously calculated during the search process, and this information guides the search direction, further enhancing the global search capability. This effectively avoids getting trapped in local optima during the search process and improves the quality of the optimized solution.

[0094] Furthermore, based on the adaptive neighborhood information feedback mechanism, a hyperplane Pareto front sampling strategy is adopted. The hyperplane Pareto front sampling strategy can generate an efficient Pareto front in the target space. By cutting the solution space, the redundancy caused by full target space computation is avoided.

[0095] The updated individual is input into the objective function and constraint condition, and the evaluation of the objective function and constraint condition of the updated individual is realized. Preferably, in order to improve the calculation efficiency of the evaluation process of the objective function and constraint condition of the individual in each generation, a hardware-accelerated parallel structure is used to realize the parallel execution of multiple individual evaluation processes, thereby significantly reducing the iteration time consumption and improving the overall operation efficiency of the optimization algorithm.

[0096] Further, in each generation evolution, the optimal individual is screened based on the existing Pareto dominance principle, the screening process is selected by calculating the crowding degree, and the diversity of the solution set is ensured. Optionally, the crowding degree is obtained by using the existing calculation method, such as the k-neighborhood density estimation (k-NN) method.

[0097] Preferably, the screened optimal individual will constitute the elite solution set of the current generation, as the parent population of the next generation individual generation, participating in the existing crossover, mutation and other evolutionary operations, to obtain the individual of the next generation, constituting the candidate solution set of the next generation, continuing the above process. And when the termination condition (such as the maximum number of iterations and the target convergence threshold preset according to the expert experience method, etc.) is met, the current optimal non-dominated solution set (Pareto frontier) is extracted from the elite solution set of all generations, which is the final optimal solution set obtained in this round of optimization. The Pareto dominance principle is a prior art means in the art, which will not be described here.

[0098] The industrial multi-objective multi-layer optimization method of the embodiment is based on the industrial multi-objective optimization algorithm with dynamic multi-dimensional hierarchical and adaptive neighborhood information feedback. By introducing dynamic multi-dimensional hierarchical mechanism, adaptive neighborhood information feedback mechanism and hardware-accelerated parallel computing, the target weight, target subspace threshold, neighborhood information, etc. can be dynamically adjusted according to the real-time production data of steel rolling as the steel rolling production environment and process conditions change, ensuring the real-time adaptability of the algorithm, and improving the adaptability, calculation efficiency and optimization precision.

[0099] The industrial multi-objective multi-layer optimization method of the embodiment greatly improves the calculation speed of the objective function and constraint condition evaluation of the individual by introducing hardware-accelerated parallel computing, such as using FPGA acceleration card. In traditional optimization algorithms, the objective function evaluation is often the bottleneck of calculation, because the objective function and constraint condition of a large number of individuals need to be calculated in each generation of optimization process. Through the parallel computing capability of FPGA, the objective function and constraint condition of multiple individuals can be calculated simultaneously in each generation evolution process, thereby significantly shortening the calculation time and improving the overall calculation efficiency of the algorithm. Especially when the population size is large or the objective function calculation is complex, this acceleration effect is particularly significant.

[0100] Embodiment two

[0101] As another specific embodiment of the present application, the embodiment provides an industrial multi-objective hierarchical optimization device, referring to Figure 2 , comprising a real-time production data acquisition module 10, a target function determination module 20, and an industrial multi-objective hierarchical optimization module 30.

[0102] The real-time production data acquisition module 10 is configured to acquire initial real-time production data in an industrial production process, pre-process the initial real-time production data, and obtain target real-time production data.

[0103] The target function determination module 20 is configured to extract features from the target real-time production data, obtain decision variables, and determine a target function and a constraint condition based on the decision variables.

[0104] The industrial multi-objective hierarchical optimization module 30 is configured to use an industrial multi-objective optimization algorithm based on dynamic multi-dimensional hierarchical optimization and adaptive neighborhood information feedback, combine hardware acceleration parallel computing, solve the target function, and realize industrial multi-objective hierarchical optimization.

[0105] Embodiment three

[0106] As another specific embodiment of the present application, the embodiment provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to realize the steps of the industrial multi-objective hierarchical optimization method of the embodiment one:

[0107] S100, acquiring initial real-time production data in an industrial production process, pre-processing the initial real-time production data, and obtaining target real-time production data;

[0108] S200, extracting features from the target real-time production data, obtaining decision variables, and determining a target function and a constraint condition based on the decision variables;

[0109] S300, using an industrial multi-objective optimization algorithm based on dynamic multi-dimensional hierarchical optimization and adaptive neighborhood information feedback, combining hardware acceleration parallel computing, solving the target function, and realizing industrial multi-objective hierarchical optimization.

[0110] Embodiment four

[0111] As another specific embodiment of the present application, the embodiment provides a storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to realize the steps of the industrial multi-objective hierarchical optimization method of the embodiment one:

[0112] S100, acquiring initial real-time production data in an industrial production process, pre-processing the initial real-time production data, and obtaining target real-time production data;

[0113] S200, feature extraction is performed on the target real-time production data to obtain a decision variable, and a target function and a constraint condition are determined based on the decision variable;

[0114] S300, an industrial multi-objective optimization algorithm based on dynamic multi-dimensional layering and adaptive neighborhood information feedback is used to solve the target function in combination with hardware acceleration parallel computing, so as to realize industrial multi-objective layering optimization.

[0115] The above is only a preferred embodiment of the present application, and is not intended to limit the present application in any other form, and any modification or equivalent change made according to the technical essence of the present application still falls within the scope of the present application.

Claims

1. An industrial multi-objective hierarchical optimization method, characterized in that, The method comprises: acquiring initial real-time production data in an industrial production process, pre-processing the initial real-time production data to obtain target real-time production data, wherein the initial real-time production data comprises temperature, pressure, speed, production progress, and equipment state data; performing feature extraction on the target real-time production data to obtain decision variables, and determining a target function and a constraint condition based on the decision variables; solving the target function by using an industrial multi-objective optimization algorithm based on dynamic multi-dimensional layering and adaptive neighborhood information feedback in combination with hardware-accelerated parallel computing, to realize industrial multi-objective layered optimization; wherein the industrial multi-objective optimization algorithm based on dynamic multi-dimensional layering and adaptive neighborhood information feedback comprises a dynamic multi-dimensional layering mechanism and an adaptive neighborhood information feedback mechanism, the dynamic multi-dimensional layering mechanism adopts a dynamic layering strategy, adjusts the hierarchical structure of the target function in each generation in real time according to the distribution of solutions in the target space and actual optimization requirements, and divides the target space based on the weighted sum of the target function in each generation, each target space being divided into a plurality of target subspaces; the adaptive neighborhood information feedback mechanism calculates the neighborhood distance between an individual and neighborhood individuals in a target subspace, and realizes a search process based on the neighborhood distance, the neighborhood individuals being taken from the target subspace.

2. The industrial multi-objective hierarchical optimization method of claim 1, wherein, The pre-processing of the initial real-time production data comprises data cleaning, denoising, missing value filling, standardization, and normalization of the initial real-time production data.

3. The industrial multi-objective hierarchical optimization method of claim 1, wherein, Before determining the objective function and constraints based on the aforementioned decision variables, a population is formed by assembling all possible solutions to the objective function, and the population is initialized. The population consists of Individual composition, using Indicates the first Individual, It is the number of decision variables; It is the first The first individual One decision variable.

4. The industrial multi-objective layering optimization method of claim 3, wherein, The target function is: wherein is the first objective function determined by different combinations of decision variables ; The constraint condition is: wherein is the mth process constraint, = 1, 2, …, m, m is the number of inequality constraints.​ 5. The industrial multi-objective hierarchical optimization method according to any one of claims 1-4, characterized in that, The division formula of the target subspace is: In the formula, Indicates the first The objective subspace is a high-dimensional space composed of the weighted sum of the objective functions. This high-dimensional space contains all solutions that satisfy the specific objective constraints. To ensure that, under the current optimization conditions, the objective weighted combination in the solution space does not exceed a threshold. The set of all solutions; These are the dynamic weights of each objective function, representing the importance of different objective functions in the optimization process. A reference value is... And the sum of these values ​​is 1. It is the first The threshold of each objective subspace is used to limit the upper limit of the weighted objective function value. It is dynamically adjusted in each generation of optimization to control the range of the objective and ensure that the weighted sum of the objective function of each solution does not exceed the threshold. It is the space of decision variables.

6. The industrial multi-objective hierarchical optimization method of claim 5, wherein, The calculation formula of the neighborhood distance is: wherein, is the individual is the neighborhood distance between the individual and the neighborhood individual, measuring the similarity of two individuals in the objective space and the decision space; is the number of neighborhood individuals; is the exponential adjustment factor, controlling the weighting degree of the objective function difference; is the sensitivity control of the objective space distance; is the weight of the th decision variable, reflecting the importance of the decision variable in calculating the neighborhood distance; is the contribution degree of the decision variable difference to the neighborhood distance; is the adjustment of the influence of the decision space difference in the neighborhood distance calculation, used to adjust the sensitivity of the decision space; is the weight factor, used to adjust the influence of the objective function and the decision space difference comprehensive part; is the nonlinear weighting coefficient, used to adjust the influence of the difference between the objective function values; is the nonlinear influence coefficient of the decision space difference, used to strengthen or weaken the contribution of the difference between the solutions in the decision space to the distance calculation.

7. The industrial multi-objective layering optimization method of claim 6, wherein, The update formula of the individual is: wherein is the current time The decision variable of the i-th individual; is the updated decision variable of the i-th individual; is the updated decision variable of the i-th individual; is the learning rate or step size factor, controlling the pace of the update process, determining the relative influence of the feedback from the objective space and the decision space in the update.

8. An industrial multi-objective hierarchical optimization apparatus, characterized by, The device comprises a real-time production data acquisition module (10), a target function determination module (20), and an industrial multi-objective layered optimization module (30); The real-time production data acquisition module (10) is configured to acquire initial real-time production data in an industrial production process, pre-process the initial real-time production data to obtain target real-time production data, wherein the initial real-time production data comprises temperature, pressure, speed, production progress, and equipment state data; The target function determination module (20) is configured to perform feature extraction on the target real-time production data to obtain decision variables, and determine a target function and a constraint condition based on the decision variables; The industrial multi-objective layered optimization module (30) is configured to solve the target function by using an industrial multi-objective optimization algorithm based on dynamic multi-dimensional layering and adaptive neighborhood information feedback in combination with hardware-accelerated parallel computing, to realize industrial multi-objective layered optimization; The industrial multi-objective optimization algorithm based on dynamic multi-dimensional hierarchical layering and adaptive neighborhood information feedback comprises a dynamic multi-dimensional hierarchical layering mechanism and an adaptive neighborhood information feedback mechanism.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program comprises instructions that, when executed by the processor, cause the electronic device to perform the method of any one of claims 1-8. The processor executes the computer program to implement the steps of the method of any one of claims 1-7.

10. A storage medium having stored thereon a computer program, characterized in that The computer program, when executed by the processor, implements the steps of the method of any one of claims 1-7.

Images