A constrained multi-objective dynamic optimization method and device based on operation indicators

Abstract

The present application proposes a constrained multi-objective dynamic optimization method and device based on operating indicators, which belongs to the field of intelligent optimization technology. Through a pre-trained random weight neural network model, the relationship between operating indicators, working conditions and multiple comprehensive production indicators is obtained; a dynamic constrained multi-objective optimization problem model is established; when the working conditions are different from the preset working conditions, in an initial population with N solutions, the K-means clustering algorithm is used to solve the dynamic constrained multi-objective optimization problem model, the working conditions are dynamically clustered with the preset working conditions, and the dynamically clustered population is output; a non-dominated sorting genetic algorithm is used to sort and select the dynamically clustered population, and the selected result is used as the optimal operating indicator under the working conditions. The present application automatically responds dynamically, quickly obtains a set of optimal operating indicators under the current environment, and improves the production efficiency of the mineral processing process under a dynamic environment.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent optimization, and in particular relates to a constrained multi-objective dynamic optimization method and device based on operating indicators. Background Art

[0002] In the face of global competition and environmental protection requirements, the problem of operating index optimization has become increasingly important in industrial production. The ultimate goal of this problem is to determine the optimal set value of industrial operating indicators based on production constraints. For production equipment, dynamic environments such as changes in raw material properties and fluctuations in production conditions will cause fluctuations in the set value of the optimal operating variable in the production process. Failure to adjust the set value of the operating variable in a timely manner will reduce product quality, the economic benefits of the enterprise, and even endanger the production safety of the enterprise.

[0003] At present, only some large-scale mineral processing enterprises in China have realized digital, automated and intelligent control, and most small and medium-sized mineral processing industrial enterprises still use manual control methods, such as operators simply adjusting the relevant indicators of the equipment based on their own work experience. However, the external environment and internal interference of the actual mineral processing process are dynamically uncertain. When the external dynamic environment and internal interference appear, the optimal setting values ​​of the operating indicators in the production process will also undergo complex changes. Usually, it is difficult for operators to grasp the correct adjustment direction and amplitude of the decision variables, and it often takes multiple adjustments to make the operating indicators meet the comprehensive production indicators, which reduces the economic benefits of the mineral processing process and may also cause production safety accidents.

[0004] The existing optimization methods for mineral processing process operating indicators do not simultaneously consider multiple factors such as users' multi-objective demands for comprehensive production indicators, internal and external dynamic interference in the mineral processing process, and production indicator constraints. It is difficult to quickly obtain the optimal operating indicators under current operating conditions after production conditions change. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present application proposes a constrained multi-objective dynamic optimization method and device based on operating indicators.

[0006] In the first aspect, the present application proposes a constrained multi-objective dynamic optimization method based on operating indicators, comprising:

[0007] Step S1: obtaining the operating indicators, working conditions and multiple comprehensive production indicators of each process in the actual industrial production of mineral processing;

[0008] Step S2: according to the operation index, the working condition and the multiple comprehensive production indexes, the relationship between the operation index, the working condition and the multiple comprehensive production indexes is obtained through a pre-trained random weight neural network model;

[0009] Step S3: taking multiple comprehensive production indicators as optimization targets and operation indicators as decision variables, and establishing a dynamic constraint multi-objective optimization problem model according to the relationship between the operation indicators, working conditions and multiple comprehensive production indicators;

[0010] Step S4: generating an initial population with N solutions according to the preset upper and lower limits of the operating indicators;

[0011] Step S5: determining whether the current operating condition is the same as the operating condition of the last iterative optimization;

[0012] Step S6: When the current working condition is different from the working condition of the last iterative optimization, the K-means clustering algorithm is used to solve the dynamic constrained multi-objective optimization problem model in the initial population with N solutions, and the population after dynamic clustering is output, and the process goes to step S7 with the population after dynamic clustering as the current population; when the current working condition is the same as the working condition of the last iterative optimization, go to step S7;

[0013] Step S7: Perform crossover mutation on the current population to generate a sub-population, merge the current population with the sub-population, and use a non-dominated sorting genetic algorithm to sort and select the merged population to obtain a selected result;

[0014] Step S8: When the iteration stop condition is met, the selected result is used as the optimal operating indicator under the operating conditions. When the iteration stop condition is not met, the operating conditions corresponding to the selected result are used as the current operating conditions, and the selected result is used as the initial population with N solutions, and return to step S5 to continue iteration.

[0015] The K-means clustering algorithm is used to solve the dynamic constraint multi-objective optimization problem model and output the population after dynamic clustering, including:

[0016] Step S6.1: In the initial population with N solutions, a new population is generated by a random initialization method;

[0017] Step S6.2: Under the current working conditions, calculate the optimization target value of the new population;

[0018] Step S6.3: Calculate the optimization target value of each new population with a penalty term according to the optimization target value of the new population;

[0019] Step S6.4: using the K-means clustering algorithm to divide the initial population with N solutions into k sub-populations;

[0020] Step S6.5: Divide each subpopulation into a dominated solution set and a non-dominated solution set;

[0021] Step S6.6: Move the centroid of the dominated solution set in each subpopulation toward the centroid of the non-dominated solution set, and use the moved population as the population after dynamic clustering.

[0022] According to the optimization target value of the new population, the optimization target value of each new population with a penalty term is calculated, and the calculation formula is as follows:

[0023]

[0024]

[0025] in, is the first optimization target value of the new population after adding the penalty term, is the second optimization target value of the new population after adding the penalty term, Q1' is the first optimization target value of the new population after normalization, Q2' is the second optimization target value of the new population after normalization, cv' is the normalized constraint violation value, r f is the proportion of feasible solutions in the population.

[0026] The centroid of the dominated solution set in each subpopulation is moved to the centroid of the non-dominated solution set, and the calculation formula is as follows:

[0027] m=c D -c N

[0028]

[0029] Among them, c D is the centroid of the dominating solution set, c N is the centroid of the non-dominated solution, m is the distance vector between the centroid of the dominated solution set and the centroid of the non-dominated solution set, is the i-th dominant solution, and the dominant solution is an element in the dominant solution set.

[0030] The method of performing crossover mutation on the current population to generate a sub-population, merging the current population with the sub-population, and using a non-dominated sorting genetic algorithm to sort and select the merged population to obtain a selected result includes:

[0031] Calculate the optimization objective value of each initial population with a penalty term, wherein the penalty term is constructed according to the optimization objective value of the initial population, the constraint violation value, and the proportion of feasible solutions in the population;

[0032] According to the optimization target value of the initial population with penalty term, a non-dominated sorting genetic algorithm is used to sort and select the initial population with N solutions to obtain a parent population;

[0033] Performing simulated binary crossover and polynomial mutation on the parent population to form a daughter population;

[0034] Calculate the optimization target value of each offspring population with a penalty term, wherein the penalty term is constructed according to the optimization target value of the offspring population, the constraint violation value, and the proportion of feasible solutions in the offspring population;

[0035] According to the optimization target value of each offspring population with penalty items, a non-dominated sorting genetic algorithm is used to sort and select the parent population and the offspring population to obtain a new generation population, which is used as the result after selection.

[0036] In the second aspect, the present application proposes a constrained multi-objective dynamic optimization device based on operating indicators, including: a data acquisition module, a relationship calculation module, a model building module, a population generation module, a working condition judgment module, a dynamic response module, a population optimization module and an iterative jump module;

[0037] The data acquisition module is connected to the relationship calculation module, the relationship calculation module is connected to the model building module, the model building module is connected to the population generation module, the working condition judgment module is connected to the dynamic response module and the population optimization module respectively, the population optimization module is connected to the iteration jump module, the iteration jump module is connected to the working condition judgment module, and the dynamic response module is connected to the population optimization module;

[0038] Data acquisition module, used to obtain the operating indicators, working conditions and multiple comprehensive production indicators of each process in the actual industrial production of mineral processing;

[0039] A relationship calculation module, used to obtain the relationship between the operating indicators, the operating conditions and the multiple comprehensive production indicators through a pre-trained random weight neural network model according to the operating indicators, the operating conditions and the multiple comprehensive production indicators;

[0040] A model building module is used to establish a dynamic constraint multi-objective optimization problem model based on the relationship between the operating indicators, working conditions and the multiple comprehensive production indicators, taking the multiple comprehensive production indicators as optimization targets and the operating indicators as decision variables;

[0041] A population generation module is used to generate an initial population with N solutions according to the upper and lower limits of the preset operation indicators;

[0042] The working condition judgment module is used to judge whether the current working condition is the same as the working condition of the previous iterative optimization;

[0043] A dynamic response module is used to solve the dynamic constraint multi-objective optimization problem model using a K-means clustering algorithm in an initial population with N solutions when the current working condition is different from the working condition of the last iterative optimization, output the population after dynamic clustering, and transfer the population optimization module with the population after dynamic clustering as the current population; when the current working condition is the same as the working condition of the last iterative optimization, transfer to the population optimization module;

[0044] The population optimization module is used to perform crossover mutation on the current population to generate sub-populations, merge the current population with the sub-populations, and use a non-dominated sorting genetic algorithm to sort and select the merged population to obtain the selected result;

[0045] The iterative jump module is used to use the selected result as the optimal operating indicator under the working conditions when the iteration stop condition is met. When the iteration stop condition is not met, the working condition corresponding to the selected result is used as the current working condition, and the selected result is used as the initial population with N solutions, and returns to the working condition judgment module to continue iteration.

[0046] In a third aspect, the present application proposes an electronic device comprising: one or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the constrained multi-objective dynamic optimization method based on operating indicators.

[0047] In a fourth aspect, the present application proposes a computer-readable storage medium storing executable instructions, which, when executed, enable a processor to execute the constrained multi-objective dynamic optimization method based on operating indicators.

[0048] In a fifth aspect, the present application proposes a computer program product, including a computer program or instructions, which, when executed by a processor, implements the constrained multi-objective dynamic optimization method based on operating indicators.

[0049] Beneficial effects:

[0050] This application proposes a constrained multi-objective dynamic optimization method and device based on operating indicators. After the working conditions change, this application will automatically respond dynamically, improve the optimization efficiency, and quickly obtain a set of optimal operating indicators under the current environment. This application significantly improves the production efficiency of the mineral processing process under a dynamic environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of a constrained multi-objective dynamic optimization method based on operating indicators according to an embodiment of the present application;

[0052] Figure 2A schematic diagram of a multi-objective dynamic optimization method based on operating indicators according to an embodiment of the present application;

[0053] Figure 3 It is a schematic diagram of the mineral processing process of the prior art;

[0054] Figure 4 A schematic diagram of a random weight neural network according to an embodiment of the present application;

[0055] Figure 5 A schematic diagram of dynamic clustering of the K-means clustering algorithm according to an embodiment of the present application;

[0056] Figure 6 This is a principle block diagram of a constrained multi-objective dynamic optimization device based on operating indicators according to an embodiment of the present application. DETAILED DESCRIPTION

[0057] The ore dressing process plays an important role in the mining development process. It can remove impurities in the raw ore after a series of complex chemical and physical changes, and enrich the useful ore. Figure 3 As shown in the figure, the mineral processing process needs to go through the processes of raw ore screening, vertical furnace roasting, grinding, magnetic separation, concentrate and tailings treatment, which involves multiple production devices such as ball mills and vertical furnaces. Each process in the mineral processing process has corresponding operating indicators, and the set values ​​of these indicators determine the final comprehensive production indicators. At the same time, factors such as raw ore composition and equipment operating capacity are directly related to the output, energy consumption and resource utilization of iron ore, which in turn affect the production efficiency of the enterprise. Therefore, it is necessary to optimize the operating indicators in combination with the properties of raw materials, working conditions, etc., in order to improve the production and operation capabilities of mineral processing enterprises, reduce dependence on iron ore imports, and thus reduce economic pressure.

[0058] In general, the optimization of operating indicators in the mineral processing process mainly faces the following challenges:

[0059] 1. The actual mineral processing process is often in a complex and dynamic environment due to the influence of internal disturbances or external environment. For example, the adjustment and supply of raw ore will cause the change of raw ore grade, and the change of production capacity caused by equipment failure and wear and tear. Therefore, how to develop response technology in combination with dynamic environment is a difficult problem.

[0060] 2. The decision variables and targets for optimizing the operating indicators of the mineral processing process are subject to production constraints. Due to the limitation of equipment processing capacity and the user's requirements for comprehensive production indicators, the operating indicators, comprehensive production indicators and average comprehensive production indicators under different working conditions must be within a certain range. At present, enterprises use manual experience-based adjustment methods that lack the support of real-time data, making it difficult to comprehensively weigh the above constraints, and thus difficult to optimize process indicators based on constraints, dynamic environments, etc.

[0061] Therefore, it is of great practical application value to study how to timely and effectively optimize the operation indicators of each process in the mineral processing process under a dynamic environment to improve product quality and output, thereby improving the economic benefits of mineral processing enterprises.

[0062] In order to solve the problem that the existing technology cannot quickly track the optimal setting value of the decision variable under dynamic environments such as working condition changes, this application proposes a constrained multi-objective dynamic optimization method and device based on operating indicators. On the one hand, considering that the change of the optimal solution set is irregular and difficult to predict after the working conditions change, this application uses the K-MEANS clustering method (K-Means clustering algorithm) to promote the solution to move in multiple optimization directions; on the other hand, the user hopes that the comprehensive production indicators are within a certain range. This application uses an adaptive penalty function method to handle constraints, which can avoid premature convergence of the population and obtain more feasible solutions close to the Pareto frontier. The above method can determine the operating indicators of the relevant processes according to the current production status and the properties of the original ore, thereby improving the concentrate output and concentrate grade.

[0063] Embodiment 1:

[0064] This embodiment proposes a constrained multi-objective dynamic optimization method based on operating indicators, such as Figure 1 , Figure 2 As shown, the following steps are included:

[0065] Step S1: obtaining the operating indicators, working conditions and multiple comprehensive production indicators of each process in the actual industrial production of mineral processing;

[0066] Step S2: according to the operation index, the working condition and the multiple comprehensive production indexes, the relationship between the operation index, the working condition and the multiple comprehensive production indexes is obtained through a pre-trained random weight neural network model;

[0067] Step S3: taking multiple comprehensive production indicators as optimization targets and operation indicators as decision variables, and establishing a dynamic constraint multi-objective optimization problem model according to the relationship between the operation indicators, working conditions and multiple comprehensive production indicators;

[0068] Step S4: generating an initial population with N solutions according to the preset upper and lower limits of the operating indicators;

[0069] Step S5: determining whether the current operating condition is the same as the operating condition of the last iterative optimization;

[0070] Step S6: When the current working condition is different from the working condition of the last iterative optimization, the K-means clustering algorithm is used to solve the dynamic constrained multi-objective optimization problem model in the initial population with N solutions, and the population after dynamic clustering is output, and the process goes to step S7 with the population after dynamic clustering as the current population; when the current working condition is the same as the working condition of the last iterative optimization, go to step S7;

[0071] Step S7: Perform crossover mutation on the current population to generate a sub-population, merge the current population with the sub-population, and use a non-dominated sorting genetic algorithm to sort and select the merged population to obtain a selected result;

[0072] Step S8: When the iteration stop condition is met, the selected result is used as the optimal operating indicator under the operating conditions. When the iteration stop condition is not met, the operating conditions corresponding to the selected result are used as the current operating conditions, and the selected result is used as the initial population with N solutions, and return to step S5 to continue iteration.

[0073] In this embodiment, the mineral processing process includes a series of serial and parallel processes, and each process has its own production task. Figure 3 As shown in the figure, the process mainly includes ore screening, vertical furnace roasting, grinding, magnetic separation, concentrate and tailings treatment.

[0074] The mined raw ore first enters the screening device to complete the screening process, through which the raw ore is divided into lump ore larger than 15mm and powder ore smaller than 15mm.

[0075] Because the magnetism of powder ore is stronger and that of lump ore is weaker, the lump ore needs to be roasted in a vertical furnace to be reduced to ore with stronger magnetism, that is, Fe2O3 with weaker magnetism is reduced to Fe3O4 with stronger magnetism. The roasted ore is separated into waste rock and useful ore by magnetic separation through a magnetic pulley. The operating index of the vertical furnace roasting process is the recovery rate of the magnetic separator, which is affected by the waste rock rate.

[0076] Then the roasted lump ore and powder ore are subjected to weak magnetic grinding and strong magnetic grinding respectively. Grinding is the process of grinding lump ore into smaller particle sizes, which is conducive to the full separation of useful minerals and useless minerals in the future. The grinding process consists of two ball mill circuits. The raw ore to be ground is mixed with a certain amount of water and then ground through two circuits. The fine-grained slurry that meets the requirements enters the next process. The operating indicators of the grinding process are weak magnetic particle size and strong magnetic particle size. This process is affected by the processing capacity of the weak magnetic grinding ball mill, the running time of the weak magnetic grinding ball mill, the processing capacity of the strong magnetic grinding ball mill and the running time of the strong magnetic grinding ball mill.

[0077] The magnetic separation process is to separate the ore pulp into concentrate and tailings. The ore pulp produced by grinding enters the sorting box and then enters the magnetic field area; according to the principle that the strength of the magnetic field is different in magnitude, under various forces in the magnetic field, the ore particles with stronger magnetism are gathered together, while the ore particles with weaker magnetism are washed away in the magnetic separation process and become tailings. The operating indicators of the magnetic separation process are weak concentrate grade, strong concentrate grade, weak tail grade, and strong tail grade. This process is affected by the strong magnetic mill grade and the weak magnetic mill grade.

[0078] Finally, the concentrate and tailings after the magnetic separation process are dehydrated separately.

[0079] The operating indicators in the mineral processing process are affected by the properties of the original ore (waste rock rate, strong magnetic grinding grade and weak magnetic grinding grade) and the operating conditions of the equipment (weak magnetic grinding ball mill processing capacity, weak magnetic grinding ball mill operating time, strong magnetic grinding ball mill processing capacity and strong magnetic grinding ball mill operating time), which are collectively referred to as operating conditions. In the actual production process, the operating conditions change irregularly.

[0080] The embodiment mainly solves the problem of optimizing the operating index of each process in the above-mentioned mineral processing process under the dynamic change of working conditions, so as to obtain the expected comprehensive production index, and then send the optimized operating index to the operation control layer (used to control each process of the above-mentioned mineral processing process), and the operation control layer completes the adjustment of the process such as the vertical furnace, grinding, and magnetic separation according to the optimized index.

[0081] This embodiment mainly solves the optimization of the performance indicators generated in the above-mentioned mineral processing process, and then sends the results to the operation control layer, so that the comprehensive production indicators meet the expected target values. The performance indicators involved in the mineral processing process include the operation indicators of each process, boundary constraints and comprehensive production indicators. Among them, the operation indicators are seven indicators: magnetic separator recovery rate (α), strong magnetic grinding particle size (D1), weak magnetic grinding particle size (D2), strong magnetic concentrate grade (β1), weak magnetic concentrate grade (β2), strong magnetic tailings grade (β3), and weak magnetic tailings grade (β4); the working conditions are seven indicators: weak magnetic mill grade (ρ1), strong magnetic mill grade (ρ2), strong magnetic ball mill processing capacity (A1), weak magnetic ball mill processing capacity (A2), waste rock grade (ρ3), strong magnetic ball mill operation time (B1), and weak magnetic ball mill operation time (B2); the comprehensive production indicators are comprehensive concentrate output (Q1) and comprehensive concentrate grade (Q2).

[0082] In step S1, the operation index, working condition and multiple comprehensive production indexes of each process in the actual industrial production of mineral processing are obtained; wherein, the operation index includes: magnetic separator recovery rate α, strong magnetic particle size D1, weak magnetic particle size D2, strong fine product grade β1, weak fine product grade β2, strong tail grade β3, weak tail grade β4; the working condition includes: weak magnetic mill grade ρ1, strong magnetic mill grade ρ2, strong magnetic ball mill processing capacity A1, weak magnetic ball mill processing capacity A2, waste rock grade ρ3, strong magnetic ball mill operation time B1, weak magnetic ball mill operation time B2; multiple comprehensive production indexes include two comprehensive production indexes, including: comprehensive concentrate output Q1 and comprehensive concentrate grade Q2, which constitute the multi-objective optimization target {Q1, Q2}. The optimization target of the method of this embodiment is to maximize the comprehensive concentrate output and comprehensive concentrate grade within the target interval of the comprehensive production index.

[0083] In step S2, according to the operating indicators, operating conditions and multiple comprehensive production indicators, the relationship between the operating indicators, the operating conditions and the multiple comprehensive production indicators is obtained through a pre-trained random weight neural network model;

[0084] The relationship between operation indicators, working conditions and comprehensive concentrate production and grade is established through the pre-trained random weight neural network model. The collected data set includes 575 sets of production data in different process periods. According to the actual production situation, the production data is divided into operation indicator data set and working condition data set. The operation indicators, working conditions, constraints and target variables of all production periods are consistent. The operation indicators include magnetic separation tube recovery rate, strong magnetic particle size, weak magnetic particle size, strong fine grade, weak fine grade, strong tail grade, weak tail grade. The operating conditions include weak magnetic mill grade, strong magnetic mill grade, strong magnetic ball mill processing capacity, weak magnetic ball mill processing capacity, waste rock grade, strong magnetic ball mill operation time, weak magnetic ball mill. The operation time constraints include: comprehensive concentrate grade constraint, comprehensive concentrate production constraint, average concentrate grade constraint and average concentrate production constraint. The target variables include comprehensive concentrate production and comprehensive concentrate grade. Among them, the collected historical data set, including 575 sets of production data in different process periods, is used to train the random weight neural network model to obtain a pre-trained random weight neural network model. The training process is as follows:

[0085] Step S2.1: Determine the input and output of the random weight neural network model;

[0086] The inputs of the random weighted neural network model are: magnetic separation tube recovery rate, strong magnetic particle size, weak magnetic particle size, strong fine product grade, weak fine product grade, strong tail grade, weak tail grade; weak magnetic mill grade, strong magnetic mill grade, strong magnetic ball mill processing capacity, weak magnetic ball mill processing capacity, waste rock grade, strong magnetic ball mill operating time and weak magnetic ball mill operating time, a total of fourteen inputs.

[0087] The input of the random weight neural network model is X = [X1, X2] = [O, C] = [α, D1, D2, β1, β2, β3, β4, ρ1, ρ2, ρ3, A1, A2, B1, B2], and the output of the random weight neural network model is the optimization target, which is the comprehensive concentrate output Q1 and the comprehensive concentrate grade Q2. O is the operating index, and C is the operating condition;

[0088] Among them, X1=O=[α,D1,D2,β1,β2,β3,β4] is the operating index, and X2=C=[ρ1,ρ2,ρ3,A1,A2,B1,B2] is the operating condition.

[0089] Step 2.2 uses a random weight neural network to establish a model between input and output. The random weight neural network structure is shown in the figure below: Figure 4 As shown, the mathematical expression is as follows:

[0090]

[0091] Where ψ=[ψ1,ψ2,...,ψ P ] is the output weight, P is the number of hidden layer nodes, X is the input of the random weight neural network model, W p represents the weight vector between the pth hidden layer neuron and the input, b p represents the threshold of the pth hidden layer neuron, and G is the hidden layer activation function, as shown below:

[0092]

[0093] Among them, z∈[-∞,+∞], a=1.716, b=0.666.

[0094] Step 2.3 sets the parameters of the random weight neural network.

[0095] Among them, the number of hidden layer nodes P = 80, the dimension of input data d = 14, and the input weight vector W p and hidden layer node threshold b p It is randomly generated.

[0096] Step 2.4 calculates the output weights of the random weight neural network.

[0097] This embodiment uses the least squares method to obtain the output weights as:

[0098] ψ=(H T H) -1 H T T

[0099] Among them, H is the hidden layer output matrix, expressed as:

[0100]

[0101] Where T is the target matrix, expressed as:

[0102] T=[y1,y2,…y S ]

[0103] Wherein, S is the number of samples collected. In this embodiment, the historical operation indicators, working conditions and multiple comprehensive production indicators of each process are used as samples. X is the input of the random weight neural network model, and y is the output of the random weight neural network model.

[0104] Step 2.5 After the random weight neural network calculation, the relationship between the operating index, working conditions and multiple comprehensive production indicators is obtained, which is expressed as:

[0105] Q1=f1(X1)=f1(O,C)

[0106] Q2=f2(X2)=f2(O,C)

[0107] Among them, X1 and O represent operating indicators, X2 and C represent operating conditions, Q1 represents the comprehensive concentrate output, and Q2 represents the comprehensive concentrate grade.

[0108] In step S3, a dynamic constrained multi-objective optimization problem model is established based on the relationship between the operating indicators, working conditions and the multiple comprehensive production indicators, taking the multiple comprehensive production indicators as optimization targets and the operating indicators as decision variables;

[0109] In this embodiment, the dynamic constraint multi-objective optimization problem model established takes the comprehensive concentrate output Q1 and the comprehensive concentrate grade Q2 as the optimization targets, and the operation index is the decision variable. Considering the change of working conditions over time, the dynamic constraint multi-objective optimization problem model is expressed as follows during the mineral processing process:

[0110] minF(X,t)=(-Q1(t),-Q2(t))=(-f1(O(t),C(t)),-f2(O(t),C(t)))

[0111] Among them, O(t)=[α(t),D1(t),D2(t),β1(t),β2(t),β3(t),β4(t)]

[0112] C(t)=[ρ1(t),ρ2(t),ρ3(t),A1(t),A2(t),B1(t),B2(t)]

[0113] Constraints:

[0114]

[0115]

[0116]

[0117]

[0118]

[0119]

[0120]

[0121]

[0122] in, and Represent the maximum and minimum values ​​of the i-th operating indicator respectively. and Represent the maximum and minimum values ​​of the i-th working condition respectively. and Respectively represent the minimum and maximum values ​​of concentrate production, and Respectively represent the minimum and maximum values ​​of concentrate grade, and They represent the average output and grade of concentrate obtained under the first k environments respectively, k represents the number of working conditions experienced in the entire optimization process, and t represents the number of working conditions experienced in the entire optimization process for the current working condition.

[0123] In step S4, an initial population with N solutions is generated according to the upper and lower limits of the preset operation index;

[0124] In this embodiment, it is necessary to determine the constraint boundary and operating conditions, wherein the constraint boundary is the upper and lower limits of the operating index:

[0125] Step 4.1 Determine the constraint boundaries of the operating indicators:

[0126] According to the data collected on site and the user's requirements for comprehensive production indicators, the constraint boundaries of the operation indicators and comprehensive production indicators are set as shown in Table 1.

[0127] Table 1 Constraint boundary setting table:

[0128] α(％) <![CDATA[D1(%)]]> <![CDATA[D2(%)]]> <![CDATA[β1(%)]]> <![CDATA[β2(%)]]> <![CDATA[β3(%)]]> <![CDATA[β4(%)]]> <![CDATA[Q1(t)]]> <![CDATA[Q2(%)]]> Minimum 81.53 68.29 48.62 54.15 46.09 15.93 17.97 8300 51 Maximum 84.80 84.88 84.03 57.83 53.38 20.18 23.15 9200 53

[0129] Step 4.2 Set the working conditions:

[0130] The seven working environment settings are shown in Table 2.

[0131] Table 2 Working environment setting table:

[0132] <![CDATA[ρ1(%)]]> <![CDATA[ρ2(%)]]> <![CDATA[A1(t / h)]]> <![CDATA[A2(t / h)]]> <![CDATA[B1(h)]]> <![CDATA[B2(h)]]> <![CDATA[ρ3(%)]]> Condition 1 44.10 32.37 73.16 60.06 83.14 96.00 13.22 Condition 2 41.43 35.34 75.85 56.78 94.00 93.00 14.63 Condition 3 43.38 36.73 61.49 72.08 96.00 96.00 13.00 Condition 4 41.93 35.56 82.18 66.89 95.00 96.00 12.65 Condition 5 42.33 33.56 62.12 47.34 78.00 79.00 11.33 Condition 6 41.68 35.04 62.76 63.68 73.00 96.00 15.00 Condition 7 42.07 35.26 97.95 83.48 81.00 80.00 16.21

[0133] In step S5, it is determined whether the current operating condition is the same as the operating condition of the previous iterative optimization;

[0134] In this embodiment, if the current working condition is the same as the working condition of the last iterative optimization, it will go directly to step S8 for static optimization. If the current working condition is different from the working condition of the last iterative optimization, then execute step S6 to first perform the dynamic response step, and then perform the static optimization step. It can be understood that: the first iterative optimization does not need to determine whether it is consistent with the working condition of the last iteration. In the actual production process, the first iteration is to set the working condition in the optimization method according to the on-site working conditions. In this embodiment, a simulation experiment is carried out to set seven working conditions, and it is pre-set how many times the working condition changes every optimization iteration.

[0135] In step S6, when the current operating condition is different from the operating condition of the last iterative optimization, the dynamic constrained multi-objective optimization problem model is solved using a K-means clustering algorithm in an initial population with N solutions, and a population after dynamic clustering is output. The population after dynamic clustering is used as the current population and the process proceeds to step S8, including:

[0136] Step S6.1: In the initial population with N solutions, a new population is generated by a random initialization method;

[0137] Step S6.2: Under the current working conditions, calculate the optimization target value of the new population;

[0138] Step S6.3: Calculate the optimization target value of each new population with a penalty term according to the optimization target value of the new population;

[0139] Step S6.4: using the K-means clustering algorithm to divide the initial population with N solutions into k sub-populations;

[0140] Step S6.5: Divide each subpopulation into a dominated solution set and a non-dominated solution set;

[0141] In this embodiment, the definition of dominance is as follows: for x1, x2∈Ω, if f m (x1,t)≥f m If (x2,t)(m=1,...,n) holds, and at least one of them is a strict inequality, then x2 is said to dominate x1. Definition of a dominated solution: If x*∈Ω, and there is no other x∈Ω such that f m (x*,t)≥f mIf (x,t)(m=1,...,n) holds, and at least one of them is a strict inequality, then x* is called a Pareto optimal solution or a non-dominated solution. The set of all dominated solutions is called the dominated set, and similarly, the set of all non-dominated solutions is called the non-dominated solution set.

[0142] Step S6.6: Move the centroid of the dominated solution set in each subpopulation toward the centroid of the non-dominated solution set, and use the moved population as the population after dynamic clustering.

[0143] In this embodiment, the proposed dynamic constraint multi-objective optimization decision method for the mineral processing process based on K-MEANS clustering is used to solve the established dynamic constraint multi-objective optimization problem model, and obtain the frontier of the Pareto optimal solution under different production environments, so as to maximize the concentrate output and grade. The flow chart of the dynamic constraint multi-objective optimization decision method for the mineral processing process based on K-MEANS clustering is shown in FIG. Figure 5 The details are as follows:

[0144] In step S6.1, a new population is generated from an initial population with N solutions by a random initialization method;

[0145] In this embodiment, the population size P is set to N = 100, and the maximum number of iterations FE in each dynamic environment is 8, 10, and 12 respectively. The crossover probability and distribution index of the simulated binary crossover are set to 0.9 and 5 respectively, and the mutation probability and distribution index of the polynomial mutation are set to 0.1 and 20 respectively. The decision variable is the operation index, expressed as X1 = [x1, x2, ..., x7] = [α, D1, D2, β1, β2, β3, β4]. The following calculation formula is used to generate the initial operation index:

[0146]

[0147] Among them, rand is a random number between [0,1]. and are the minimum and maximum values ​​of the jth operating index, respectively. Repeat N times to generate the initial operating index, and then obtain an initial population with N solutions. Set the number of iterations g = 1.

[0148] In step S6.2, under the operating conditions, the optimization target value of the new population is calculated;

[0149] In this embodiment, it should be noted that: determine the current operating conditions. If the current operating conditions are the same as those in the last iterative optimization, there is no need for dynamic response, and then go to step S7; if the current operating conditions change, go to step S5 and set the number of iterations g = 1. Specifically, in step S5, it is necessary to dynamically respond to changes in the current operating conditions. A dynamic response strategy based on an adaptive penalty function method and clustering is used to balance diversity and convergence by adaptively adjusting the coefficient of the penalty term: so that when the proportion of feasible solutions is large, the population is more inclined to move in a direction with better convergence; when the proportion of feasible solutions is small, the population is more inclined to move to the feasible domain. At the same time, the clustering method avoids the movement of the population in a single direction, thereby maintaining the diversity of the population.

[0150] In this embodiment, the current working condition changes. First, a new solution of 20%N is generated by a random initialization method as a new population, and the target value of the new population under the current working condition is calculated.

[0151] In step S6.3, according to the optimization target value of the new population, the optimization target value of each new population with a penalty term is calculated;

[0152] Arrange all constraints into g k (X)<0, g k (X) is the kth constraint. The constraint violation (cv) value is calculated by the following formula:

[0153]

[0154] When cv(X) is 0, it means that X is a feasible solution. K is the index of the constraint and g is the number of constraints.

[0155] The optimization target value of each new population with penalty term is calculated as follows:

[0156]

[0157]

[0158] in, is the first optimization target value of the new population after adding the penalty term. In this embodiment, the first optimization target value is the comprehensive concentrate output of the new population after adding the penalty term. is the second optimization target value of the new population after adding the penalty term. In this embodiment, the second optimization target value is the comprehensive concentrate grade of the new population after adding the penalty term. Q1' is the first optimization target value of the normalized new population. Q2' is the second optimization target value of the normalized new population. cv' is the normalized constraint violation value. r fis the proportion of feasible solutions in the new population.

[0159] In steps S6.4 to S6.6, the K-means clustering algorithm is used to divide the initial population with N solutions into k sub-populations; each sub-population is divided into a dominated solution set and a non-dominated solution set; and the centroid of the dominated solution set in each sub-population is moved toward the centroid of the non-dominated solution set.

[0160] In this embodiment, according to and The population is divided into sub-populations using the K-NEANS clustering method. The user can adjust the set value of the number of sub-populations based on the population size and the need for diversity. and As the target value, non-dominated sorting is performed for each sub-population. At this time, each sub-population is divided into two sets: dominated solutions and non-dominated solutions. Then, the movement of dominated solutions to non-dominated solutions in each sub-population is realized by the following calculation formula.

[0161] The centroid of the dominated solution set in each subpopulation is moved to the centroid of the non-dominated solution set, and the calculation formula is as follows:

[0162] m=c D -c N

[0163]

[0164] Among them, c D is the centroid of the dominating solution set, c N is the centroid of the non-dominated solution, m is the distance vector between the centroid of the dominated solution set and the centroid of the non-dominated solution set, is the i-th dominant solution, and the dominant solution is an element in the dominant solution set.

[0165] By using the adaptive penalty function method to handle constraints, infeasible solutions can be effectively handled and premature convergence of the population can be avoided.

[0166] In step S7, the current population is subjected to crossover mutation to generate a sub-population, and the current population is merged with the sub-population, and the merged population is sorted and selected using a non-dominated sorting genetic algorithm to obtain a selected result, including:

[0167] Step S7.1: Calculate the optimization target value of each initial population with a penalty term, wherein the penalty term is constructed according to the optimization target value of the initial population, the constraint violation value, and the proportion of feasible solutions in the population, and the calculation formula is as follows:

[0168] F j '(X)=F j '(X)+penalty

[0169] penalty = a × (cv' × (1-F j (X)))+b

[0170]

[0171]

[0172] Among them, F j (X) is the jth optimization objective value of the solution X, cv' is the normalized constraint violation value, r f is the proportion of feasible solutions in the population.

[0173] Step S7.2: According to the optimization target value of the initial population with penalty term, a non-dominated sorting genetic algorithm is used to sort and select the initial population with N solutions to obtain a parent population;

[0174] In this embodiment, the population P is evaluated according to the optimization target value of the initial population with penalty term. The population is sorted and selected using the NSGA-II (Non-dominated Sorting Genetic Algorithms-Ⅱ, non-dominated sorting genetic algorithm II) method, and in this embodiment, N / 2 individuals are selected as the parent population.

[0175] Step S7.3: performing simulated binary crossover and polynomial mutation on the parent population to form a child population;

[0176] In this embodiment, a progeny population OP with a size of N=100 is formed by simulating binary crossover and polynomial mutation.

[0177] The above simulated binary crossover calculation formula is as follows:

[0178]

[0179]

[0180] Among them, c 1 , c 2 is the parent individual S 1 , S 2 Two offspring are generated. r is a random number in the range [0,1]. The distribution index η = 5.

[0181] The above polynomial mutation calculation formula is as follows:

[0182]

[0183]

[0184] in, Is After mutation, r is a random number in the range [0,1]. The distribution index λ = 20. Δ is the disturbance term.

[0185] Step S7.4: Calculate the optimization target value of each offspring population with a penalty term, wherein the penalty term is constructed according to the optimization target value of the offspring population, the constraint violation value, and the proportion of feasible solutions in the offspring population;

[0186] Step S7.5: According to the optimization target value of each offspring population with a penalty item, a non-dominated sorting genetic algorithm is used to sort and select the parent population and the offspring population to obtain a new generation population, and the new generation population is used as the optimal operating indicator under the operating conditions.

[0187] In this embodiment, the populations P and OP are evaluated according to the optimization target value of each child population with a penalty term. The populations P and OP are sorted and selected using the NSGA-II method, and N solutions are saved as the new generation population. The number of iterations g = g+1.

[0188] Finally, the optimal operating indicators are sent to each process through the industrial control system.

[0189] In this embodiment, the optimized magnetic separator recovery rate is used as the control input of the vertical furnace roasting process, the strong magnetic particle size and the weak magnetic particle size are used as the control input of the grinding process, and the strong fine grade, the weak fine grade, the strong magnetic grade and the weak magnetic grade are used as the control input of the magnetic separation process.

[0190] In the above embodiment, a random weight neural network is used to establish the relationship between the operating indicators and operating conditions and the comprehensive production indicators, a constraint processing method is used to process infeasible solutions, and the K-MEANS clustering method and penalty function method are used to respond to changes in operating conditions to improve the optimization efficiency of the algorithm.

[0191] In order to verify that the proposed constrained multi-objective dynamic optimization method based on operation indicators has an advanced level, the hypervolume (HV) index is used as the evaluation index. The larger the HV, the better the optimization effect of the algorithm. The calculation formula is as follows:

[0192]

[0193] Where VOL is the Lebesgue measure used to measure volume, and Z is the set of feasible non-dominated solutions obtained by optimization. i represents the hypervolume consisting of the i-th solution in the set Z and the reference point, where the reference point is set to (-8200, -51).

[0194] The method of this embodiment is compared with the results obtained by other dynamic constraint optimization algorithms on the optimization problem of operating indicators in the mineral processing process. Seven working conditions are selected in the experiment, and the change frequency of each working condition is set to 8 generations, 10 generations or 12 generations. At the same time, in order to reduce the impact of static optimization, the algorithm is run 50 times under the first working condition. The experimental results are shown in Table 3.

[0195] Table 3 Experimental results comparison table:

[0196]

[0197]

[0198] It can be seen from Table 3 that the method of this embodiment has an advanced level in solving the constrained multi-objective optimization problem of the operating indicators of the ore dressing process under a dynamic environment. The numbers in Table 3 are the hypervolume (HV) values, an evaluation index of an optimization algorithm, which is described in detail above. The larger the HV, the better the optimization effect of the algorithm.

[0199] Embodiment 2:

[0200] This embodiment proposes a constrained multi-objective dynamic optimization device based on operating indicators, such as Figure 6 As shown, it includes: a data acquisition module, a relationship calculation module, a model building module, a population generation module, a working condition judgment module, a dynamic response module, a population optimization module and an iterative jump module;

[0201] The data acquisition module is connected to the relationship calculation module, the relationship calculation module is connected to the model building module, the model building module is connected to the population generation module, the working condition judgment module is connected to the dynamic response module and the population optimization module respectively, the population optimization module is connected to the iteration jump module, the iteration jump module is connected to the working condition judgment module, and the dynamic response module is connected to the population optimization module;

[0202] Data acquisition module, used to obtain the operating indicators, working conditions and multiple comprehensive production indicators of each process in the actual industrial production of mineral processing;

[0203] A relationship calculation module, used to obtain the relationship between the operating indicators, the operating conditions and the multiple comprehensive production indicators through a pre-trained random weight neural network model according to the operating indicators, the operating conditions and the multiple comprehensive production indicators;

[0204] A model building module is used to establish a dynamic constraint multi-objective optimization problem model based on the relationship between the operating indicators, working conditions and the multiple comprehensive production indicators, taking the multiple comprehensive production indicators as optimization targets and the operating indicators as decision variables;

[0205] A population generation module is used to generate an initial population with N solutions according to the upper and lower limits of the preset operation indicators;

[0206] The working condition judgment module is used to judge whether the current working condition is the same as the working condition of the previous iterative optimization;

[0207] A dynamic response module is used to solve the dynamic constraint multi-objective optimization problem model using a K-means clustering algorithm in an initial population with N solutions when the current working condition is different from the working condition of the previous iterative optimization, output the population after dynamic clustering, and transfer the population optimization module with the population after dynamic clustering as the current population;

[0208] Static jump module, used to jump to the population optimization module when the current working condition is the same as the working condition of the previous iterative optimization;

[0209] The population optimization module is used to perform crossover mutation on the current population to generate sub-populations, merge the current population with the sub-populations, and use a non-dominated sorting genetic algorithm to sort and select the merged population to obtain the selected result;

[0210] The iterative jump module is used to use the selected result as the optimal operating indicator under the working conditions when the iteration stop condition is met. When the iteration stop condition is not met, the working condition corresponding to the selected result is used as the current working condition, and the selected result is used as the initial population with N solutions, and returns to the working condition judgment module to continue iteration.

[0211] Embodiment 3:

[0212] This embodiment proposes an electronic device, including: one or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the constrained multi-objective dynamic optimization method based on operating indicators.

[0213] The electronic device may be a mobile phone, a computer or a tablet computer, etc., including a memory and a processor, wherein a computer program is stored on the memory, and when the computer program is executed by the processor, the multi-objective dynamic optimization method based on the operating index as described in the embodiment is implemented. It is understood that the electronic device may also include an input / output (I / O) interface and a communication component.

[0214] The processor is used to execute all or part of the steps in the multi-objective dynamic optimization method based on operating indicators as described in the above embodiment. The memory is used to store various types of data, which may include instructions of any application or method in the electronic device, as well as data related to the application.

[0215] The processor can be an application specific integrated circuit (ASIC), a digital signal processor (DSP), a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller, a microprocessor or other electronic components, and is used to execute the constrained multi-objective dynamic optimization method based on operating indicators described in the above embodiments.

[0216] Embodiment 4:

[0217] This embodiment provides a computer-readable storage medium storing executable instructions. When the instructions are executed and implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0218] The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the constrained multi-objective dynamic optimization method based on operating indicators described in each embodiment of the present application.

[0219] The aforementioned storage media include: flash memory, hard disk, multimedia card, card-type memory (for example, SD (Secure Digital Memory Card) or DX (Memory Data Register, MDR abbreviation, memory data register) memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, CD, server, APP (Application, abbreviation of application software) application store and other media that can store program verification codes, on which computer programs are stored. When the computer program is executed by the processor, the various steps of the above-mentioned constrained multi-objective dynamic optimization method based on operating indicators can be implemented.

[0220] Embodiment 5:

[0221] This embodiment proposes a computer program product, including a computer program or instructions, which, when executed by a processor, implements the constrained multi-objective dynamic optimization method based on operating indicators.

[0222] Based on such understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a computer program product.

[0223] The various embodiments in the present application are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0224] The protection scope of the present application is not limited to the above-mentioned embodiments. Obviously, those skilled in the art can make various changes and modifications to the present disclosure without departing from the scope and spirit of the present disclosure. If these changes and modifications fall within the scope of the claims of the present disclosure and their equivalents, the intention of the present disclosure also includes these changes and modifications.

Claims

1. A constrained multi-objective dynamic optimization method based on operating indicators, characterized in that: include: Step S1: obtaining the operating indicators, working conditions and multiple comprehensive production indicators of each process in the actual industrial production of mineral processing; Step S2: according to the operation index, the working condition and the multiple comprehensive production indexes, the relationship between the operation index, the working condition and the multiple comprehensive production indexes is obtained through a pre-trained random weight neural network model; Step S3: taking multiple comprehensive production indicators as optimization targets and operation indicators as decision variables, and establishing a dynamic constraint multi-objective optimization problem model according to the relationship between the operation indicators, working conditions and multiple comprehensive production indicators; Step S4: generating an initial population with N solutions according to the preset upper and lower limits of the operating indicators; Step S5: determining whether the current operating condition is the same as the operating condition of the last iterative optimization; Step S6: When the current working condition is different from the working condition of the last iterative optimization, the K-means clustering algorithm is used to solve the dynamic constrained multi-objective optimization problem model in the initial population with N solutions, and the population after dynamic clustering is output, and the process goes to step S7 with the population after dynamic clustering as the current population; when the current working condition is the same as the working condition of the last iterative optimization, go to step S7; Step S7: Perform crossover mutation on the current population to generate a sub-population, merge the current population with the sub-population, and use a non-dominated sorting genetic algorithm to sort and select the merged population to obtain a selected result; Step S8: If the iteration stop condition is met, the selected result is used as the optimal operating index under the operating condition; if the iteration stop condition is not met, the operating condition corresponding to the selected result is used as the current operating condition, and the selected result is used as the initial population with N solutions, and the process returns to step S5 to continue iteration; The K-means clustering algorithm is used to solve the dynamic constraint multi-objective optimization problem model and output the population after dynamic clustering, including: Step S6.1: In the initial population with N solutions, a new population is generated by a random initialization method; Step S6.2: Under the current working conditions, calculate the optimization target value of the new population; Step S6.3: Calculate the optimization target value of each new population with a penalty term according to the optimization target value of the new population; Step S6.4: using the K-means clustering algorithm to divide the initial population with N solutions into k sub-populations; Step S6.5: Divide each subpopulation into a dominated solution set and a non-dominated solution set; Step S6.6: Move the centroid of the dominated solution set in each subpopulation toward the centroid of the non-dominated solution set, and use the moved population as the population after dynamic clustering; The centroid of the dominated solution set in each subpopulation is moved to the centroid of the non-dominated solution set, and the calculation formula is as follows: m=c D -c N Among them, c D is the centroid of the dominating solution set, c N is the centroid of the non-dominated solution, m is the distance vector between the centroid of the dominated solution set and the centroid of the non-dominated solution set, is the ith dominant solution, the dominant solution is an element in the dominant solution set, is the i-th dominant solution after the move.

2. The constrained multi-objective dynamic optimization method based on operating indicators according to claim 1 is characterized in that: According to the optimization target value of the new population, the optimization target value of each new population with a penalty term is calculated, and the calculation formula is as follows: in, is the first optimization target value of the new population after adding the penalty term, is the second optimization target value of the new population after adding the penalty term, Q1' is the first optimization target value of the new population after normalization, Q2' is the second optimization target value of the new population after normalization, cv' is the normalized constraint violation value, r f is the proportion of feasible solutions in the population.

3. The constrained multi-objective dynamic optimization method based on operating indicators according to claim 1 is characterized in that: The method of performing crossover mutation on the current population to generate a sub-population, merging the current population with the sub-population, and using a non-dominated sorting genetic algorithm to sort and select the merged population to obtain a selected result includes: Calculate the optimization objective value of each initial population with a penalty term, wherein the penalty term is constructed according to the optimization objective value of the initial population, the constraint violation value, and the proportion of feasible solutions in the population; According to the optimization target value of the initial population with penalty term, a non-dominated sorting genetic algorithm is used to sort and select the initial population with N solutions to obtain a parent population; Performing simulated binary crossover and polynomial mutation on the parent population to form a daughter population; Calculate the optimization target value of each offspring population with a penalty term, wherein the penalty term is constructed according to the optimization target value of the offspring population, the constraint violation value, and the proportion of feasible solutions in the offspring population; According to the optimization target value of each offspring population with penalty items, a non-dominated sorting genetic algorithm is used to sort and select the parent population and the offspring population to obtain a new generation population, which is used as the result after selection.

4. A constrained multi-objective dynamic optimization device based on operating indicators, characterized in that: include: Data acquisition module, relationship calculation module, model building module, population generation module, working condition judgment module, dynamic response module, population optimization module and iterative jump module; The data acquisition module is connected to the relationship calculation module, the relationship calculation module is connected to the model building module, the model building module is connected to the population generation module, the working condition judgment module is connected to the dynamic response module and the population optimization module respectively, the population optimization module is connected to the iteration jump module, the iteration jump module is connected to the working condition judgment module, and the dynamic response module is connected to the population optimization module; Data acquisition module, used to obtain the operating indicators, working conditions and multiple comprehensive production indicators of each process in the actual industrial production of mineral processing; A relationship calculation module, used to obtain the relationship between the operating indicators, the operating conditions and the multiple comprehensive production indicators through a pre-trained random weight neural network model according to the operating indicators, the operating conditions and the multiple comprehensive production indicators; A model building module is used to establish a dynamic constraint multi-objective optimization problem model based on the relationship between the operating indicators, working conditions and the multiple comprehensive production indicators, taking the multiple comprehensive production indicators as optimization targets and the operating indicators as decision variables; A population generation module is used to generate an initial population with N solutions according to the upper and lower limits of the preset operation indicators; The working condition judgment module is used to judge whether the current working condition is the same as the working condition of the previous iterative optimization; A dynamic response module is used to solve the dynamic constraint multi-objective optimization problem model using a K-means clustering algorithm in an initial population with N solutions when the current working condition is different from the working condition of the last iterative optimization, output the population after dynamic clustering, and transfer the population optimization module with the population after dynamic clustering as the current population; when the current working condition is the same as the working condition of the last iterative optimization, transfer to the population optimization module; The population optimization module is used to perform crossover mutation on the current population to generate sub-populations, merge the current population with the sub-populations, and use a non-dominated sorting genetic algorithm to sort and select the merged population to obtain the selected result; An iterative jump module is used to use the selected result as the optimal operating indicator under the working condition when the iteration stop condition is met, and use the working condition corresponding to the selected result as the current working condition when the iteration stop condition is not met, and use the selected result as the initial population with N solutions to return to the working condition judgment module to continue iteration; The K-means clustering algorithm is used to solve the dynamic constraint multi-objective optimization problem model and output the population after dynamic clustering, including: Step S6.1: In the initial population with N solutions, a new population is generated by a random initialization method; Step S6.2: Under the current working conditions, calculate the optimization target value of the new population; Step S6.3: Calculate the optimization target value of each new population with a penalty term according to the optimization target value of the new population; Step S6.4: using the K-means clustering algorithm to divide the initial population with N solutions into k sub-populations; Step S6.5: Divide each subpopulation into a dominated solution set and a non-dominated solution set; Step S6.6: Move the centroid of the dominated solution set in each subpopulation toward the centroid of the non-dominated solution set, and use the moved population as the population after dynamic clustering; The centroid of the dominated solution set in each subpopulation is moved to the centroid of the non-dominated solution set, and the calculation formula is as follows: m=c D -c N Among them, c D is the centroid of the dominating solution set, c N is the centroid of the non-dominated solution, m is the distance vector between the centroid of the dominated solution set and the centroid of the non-dominated solution set, is the ith dominant solution, the dominant solution is an element in the dominant solution set, is the i-th dominant solution after the move.

5. An electronic device, characterized in that: include: One or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the constrained multi-objective dynamic optimization method based on operating indicators as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that: It stores executable instructions, which, when executed, enable the processor to execute the constrained multi-objective dynamic optimization method based on operating indicators as described in any one of claims 1 to 3.

7. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the constrained multi-objective dynamic optimization method based on operating indicators described in any one of claims 1 to 3 is implemented.

Images