A multi-objective optimization method for copper flash smelting process

Abstract

The present invention provides a multi-objective optimization method for copper flash smelting process, comprising the following steps: constructing an estimation model using a modeling method; determining the optimization interval of each operating variable based on historical data analysis; constructing a corresponding normalized optimization objective function based on the characteristics of each indicator variable; weighting and summing the constructed multiple normalized objective functions to construct a comprehensive optimization objective function; using a genetic algorithm to search for an optimization value within the value range of each operating variable, that is, the operating variable value with the smallest comprehensive optimization objective function value; judging whether the optimization effect meets expectations and whether the optimization process meets the end condition, if so, outputting the optimization result of the operating parameter and ending, otherwise entering weight feedback adjustment. The method is easy to implement on a computer, and can optimize the process index and energy efficiency index of copper flash smelting without changing the current equipment and process of copper flash smelting.

Description

Technical Field

[0001] The invention relates to the technical field of smelting, and in particular to a multi-objective optimization method for a copper flash smelting process. Background Art

[0002] Smelting is a key process in pyrometallurgical copper smelting, and flash smelting has become the mainstream process of copper smelting due to its advantages such as good environmental protection and high production efficiency. In the process of copper flash smelting, copper concentrate powder is sprayed into the flash furnace reaction tower at a high speed, mixed with air (or oxygen) and undergoes a violent chemical reaction, thereby producing copper matte and slag. The indicators for evaluating the operation effect of this process can usually be divided into two categories: process indicators and energy efficiency indicators: process indicators are indicators that reflect the characteristics of the main products of the process, including copper matte grade, copper matte temperature, iron-silicon ratio in slag, etc.; and energy efficiency indicators are indicators that reflect the energy consumption status of the process, including comprehensive energy consumption per unit product (copper matte). Compared with other smelting processes (such as molten pool smelting), flash smelting has more operating variables, which makes flash smelting more "flexible" and adaptable than other processes, and also makes the optimization of operating variables in the flash smelting process more important.

[0003] The optimization approaches of the flash smelting process can be divided into two categories: mechanism-driven and data-driven. The former focuses on analyzing the process operation mechanism, while the latter mainly analyzes the accumulated process data. The mechanism-driven optimization method has the advantage of clear cause-effect relationships, but for industrial processes with extremely complex operating mechanisms such as flash smelting, mechanism analysis often has to ignore many "minor" factors, which will affect the optimization effect of the model; the data-driven optimization method can obtain the influence of "minor" factors ignored by the mechanism model from the data, so it can explore better optimization solutions than the former.

[0004] At present, many scholars have applied data-driven methods to the optimization of operating variables in complex industrial processes such as flash smelting. In the article "Data-driven Optimization and Application of Copper Flash Smelting Process Operation Mode", Gui Weihua et al. proposed a data-driven optimization control framework for the operation mode of copper flash smelting process, and discussed the prediction and optimization methods of process indicators such as matte temperature, matte grade, and iron-silicon ratio in slag based on data; Peng Xiaobo et al. proposed the use of dynamic TS recursive fuzzy neural network to predict the process indicators of flash smelting process in the article "Intelligent Optimization System of Copper Flash Smelting Process Operation Mode Based on Chaotic Genetic Algorithm", and used chaotic genetic algorithm to seek the operation mode optimization scheme to improve these process indicators; Wang Jinliang et al. proposed the optimization goal of minimizing energy consumption cost in the article "Control Optimization of Copper Flash Smelting Process Based on Genetic Algorithm", and explored the optimization control method of key operation parameters of copper flash smelting process by using neural network and genetic algorithm; Zhang Qi et al. proposed the goal of reducing energy consumption, cost and CO2 emission in the article "Research and Application of Multi-objective Optimization Model of Blast Furnace Ironmaking Process", and studied the multi-objective optimization method of blast furnace ironmaking process. The research work of scholars has effectively improved the control level of flash smelting process and confirmed the feasibility and effect of applying data-driven methods to multi-objective optimization of flash smelting process. However, the optimization of flash smelting process is a multi-objective optimization problem. There is no consensus on how to design a solution to deal with multiple objectives and determine the weight coefficients of multiple optimization objectives, which brings uncertainty to the implementation effect of the method.

[0005] In summary, there is an urgent need for a multi-objective optimization method for copper flash smelting process with reliable optimization results. Summary of the invention

[0006] The object of the present invention is to provide a multi-objective optimization method for copper flash smelting process with simple optimization process and reliable optimization results. The specific technical scheme is as follows:

[0007] A multi-objective optimization method for a copper flash smelting process comprises the following steps:

[0008] S1: Collect historical data, including data of conditional variables, operating variables and indicator variables, to obtain a data set D for building an estimation model; use a modeling method and the data in the data set D to build a prediction model for the indicator variables;

[0009] S2: Determine the operating variable y to be optimized based on the historical data in step S1 i The optimal interval [y imin ,y imax ]; where i is greater than or equal to 1 and less than or equal to Ny, Ny represents the operation variable y in the input data set D i The number of imin andimax are the lower and upper limits of the optimization interval respectively;

[0010] S3: According to the relationship between the value of the indicator variable and the degree of optimization it represents, the indicator variable is divided into monotonic indicator variables z k and the non-monotonic indicator variable z l , and construct a monotonic indicator variable z based on the prediction model k and the non-monotonic indicator variable z l The corresponding normalized optimization objective function O j ;

[0011] S4: weighted summing the multiple normalized optimization objective functions constructed in step S3, thereby constructing a comprehensive optimization objective function based on the prediction model;

[0012] S5: Apply the genetic algorithm to manipulate the variable y in step S2 i The optimal interval [y imin ,y imax ] to search for the operating variable value that minimizes the function value of the comprehensive optimization objective function;

[0013] S6: Determine whether the operation variable value in step S5 makes the values ​​of each normalized optimization objective function item in the comprehensive optimization function less than or equal to 1. If so, go to step S7; otherwise, further determine condition 1 and condition 2;

[0014] Condition 1: The number of weight feedback corrections Nc exceeds the upper limit of the number of weight feedback corrections Nc mac , Nc is the number of weight feedback corrections, and its initial value is zero, Nc mac The upper limit of the number of weight feedback corrections;

[0015] Condition 2: There exists a delta smaller than 1 1 The normalized optimization objective function value, δ 1 is a range value, which is greater than 0 and less than 1;

[0016] If the two conditions are not met at the same time, go to step S7;

[0017] If both conditions are met, increase O j Normalized optimization objective function O > 1 j The corresponding weight coefficient λ j , while reducing O j ≤1-δ 1 The weight coefficient λ corresponding to the normalized optimization objective function value j , weight coefficient λ j Satisfied when modified Nz represents the number of indicator variables; then, the weight feedback correction times Nc is increased by 1, that is, Nc=Nc+1; then, go to step S5;

[0018] S7: End the optimization process, output the current optimization calculation results of each operating parameter as the optimization recommendation value, and output the corresponding comprehensive optimization objective function value and each normalized optimization objective function value.

[0019] Preferably, the monotonic indicator variable z in step S3 k The corresponding normalized optimization objective function expression is formula 1):

[0020]

[0021] Among them, z kmin is the actual optimal value of the monotonic indicator variable, that is, the actual minimum value; z komax is the expected optimization upper limit;

[0022] The non-monotonic indicator variable z l The corresponding normalized optimization objective function expression is formula 2):

[0023]

[0024] Among them, z lo 、z lomax and z lomin They are the non-monotonic indicator variables z l The expected optimization value, expected optimization upper limit and expected optimization lower limit of .

[0025] Preferably, the data set D in step S3 is expressed as formula 3):

[0026]

[0027] Where i is the data number, x, y and z represent the time period averages of the conditional variable, the operational variable and the indicator variable respectively. express The average value of the previous period; Nx, Ny and Nz represent the number of conditional variables, the number of operating variables and the number of indicator variables respectively; Nd is the total number of data in the data set D;

[0028] For the indicator variable to be predicted, select the candidate input variable set V = (z p1 ,z p2 ,…,z pNz ,x1,x2,…,x Nx ,y1,y2,…,y Ny ,z1,z2,…,z Nz); the set V consists of candidate input variables φ corresponding one-to-one to the elements in the set D k Composition; select the alternative input variable φ according to the selection conditions k As the input variable of the prediction model, the prediction model of the indicator variable is constructed using the modeling method and the data in the data set D, which is recorded as formula 4):

[0029]

[0030] Where f is the prediction model constructed using the modeling method; is the output variable of the prediction model, Ni is the number of input variables; φ j is the input variable of the prediction model, j is greater than or equal to 1 and less than or equal to Ni.

[0031] Preferably, the selection condition is: keep other input variables unchanged, compare the candidate input variable φ k The prediction error of the prediction model for the data in the data set D before and after joining; if the error level is significantly reduced after joining, it means that the above conditions are met, that is, the variable should be selected as an input variable; otherwise, it means that the above conditions are not met, that is, the variable should not be selected as an input variable.

[0032] Preferably, in step S2, y imin and imax The calculation method expression is as follows:

[0033]

[0034]

[0035] min(y i ) and max(y i ) are the operation variables y in the historical data set D i The minimum and maximum values ​​of ; δ is the expansion coefficient, and its value is greater than or equal to 0.05 and less than or equal to 0.15.

[0036] Preferably, the comprehensive optimization objective function expression is: The weight coefficient λ j satisfy:

[0037] Preferably, the comprehensive optimization objective function expression is: Among them, the weight coefficient λ j satisfy:

[0038] Preferably, the expression of the condition 1 is formula a: Nc<Nc max 9); The expression of condition 2 is formula b: The Ncmax Greater than or equal to 5 and less than or equal to 20; δ 1 Greater than 0 and less than 1.

[0039] Preferably, the δ in step S6 1 Greater than 0.2 and less than 0.5.

[0040] Preferably, a computer-readable storage medium is provided, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the multi-objective optimization method described above are implemented.

[0041] The application of the technical solution of the present invention has the following beneficial effects:

[0042] The present invention establishes an estimation model through historical data, and then constructs multiple normalized optimization functions through real-time data of indicator variables, and then obtains a comprehensive optimization objective function according to the weighted sum of multiple normalized optimization objective functions. The technical solution optimizes the operating parameters in the existing flash smelting process, and the obtained optimization calculation results are reliable, and the optimization process of multiple objectives is simplified to a single-objective optimization problem. Its mechanism is: based on historical data (including conditional variable data, operating variable data and indicator variable data), a mathematical estimation model that reflects the indicator variables and their key influencing factors (conditional variables and operating variables) is constructed, and then the optimization interval of the operating variables is determined as the range of operating variable values ​​for solving the comprehensive optimization objective function using a genetic algorithm; then an optimization problem is established, and the operating variables that meet the conditions are directly output, and the operating variables that do not meet the conditions are corrected until they meet the conditions and are output. Ensure that the output optimization calculation results are reliable.

[0043] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0045] Figure 1 It is a schematic diagram of the flow chart of the preferred embodiment 1 of the present invention. DETAILED DESCRIPTION

[0046] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0047] Embodiment 1:

[0048] The implementation object is a flash smelting furnace of a domestic copper smelting enterprise, see Figure 1 , a multi-objective optimization method for a copper flash smelting process, comprising the following steps:

[0049] S1: Collect historical data and obtain 5564 sets of historical data including conditional variables, operational variables and indicator variables to form the modeling data set, which is recorded as:

[0050]

[0051] In the formula, the superscript i is the data number, corresponding to a statistical period of a certain hour span (for example, 12:00-13:00 on a certain day); the meaning of each variable is shown in Table 1 below.

[0052] Table 1 Description of variables in the modeling dataset

[0053]

[0054] Note: In Table 1, the current statistical period refers to the statistical period corresponding to number i, and the previous statistical period refers to the statistical period before this statistical period. For example, when number i corresponds to 12:00-13:00 on a certain day, the current period refers to 12:00-13:00 on that day, and the previous statistical period refers to 11:00-12:00 on that day.

[0055] S2: Perform statistical analysis on the operation variable data in the data set D constructed in step S1 to obtain its average value, minimum value and maximum value, and then calculate the lower limit and upper limit of the optimization interval of each operation variable according to formula 5) and 6). In this implementation case, the expansion coefficient δ is taken as 0.1, and the relevant statistics and calculation results are shown in Table 2.

[0056] Table 2 Key characteristics of smelting process operating variables

[0057]

[0058] S3: This implementation case uses ε-insensitive support vector machine (ε-SVM) and data in data set D to construct prediction models for four indicator variables respectively, and the method in claim 3 is used to select input variables for each model. The input variables and output variables of each indicator prediction model are shown in Table 3, and the parameter settings and modeling effects of each model support vector machine are shown in Table 4.

[0059] Table 3 Input / output variables of indicator prediction model

[0060]

[0061] Table 4 Meta-parameter settings and modeling effects of process / energy efficiency index estimation model

[0062]

[0063] Based on the prediction of each indicator variable by the prediction model, the corresponding normalized optimization objective function is constructed according to the physical meaning and numerical distribution characteristics of each indicator variable. The specific method is as follows:

[0064] According to the relationship between the numerical value of the indicator variable and the degree of optimization represented by it, the indicator variables (including process indicators and energy efficiency indicators) of the copper flash smelting process can be divided into two categories: monotonic indicators and non-monotonic indicators. The numerical value of the monotonic indicator is monotonically related to the degree of optimization represented by it, that is, the smaller the numerical value of these indicators is, the better (or worse) the process optimization effect is, mainly including the comprehensive energy consumption per unit product z4, etc.; the numerical value of the non-monotonic indicator is not monotonically related to the degree of optimization represented by it, mainly including the copper matte grade z1, copper matte temperature z2, iron-silicon ratio in slag z3, etc., to construct the standardized optimization objective function of each indicator variable. Among them, for the non-monotonic indicator variables z1, z2 and z3, formula 2) is used to construct; for the monotonic indicator variable z4, formula 1) is used to construct. The parameters involved in formulas 1) and 2) are shown in Table 5.

[0065] Table 5 Expected values ​​and actual characteristic values ​​of indicator variables

[0066]

[0067] S4: weighted sum of the multiple normalized optimization objective functions constructed in step S3 to construct a comprehensive optimization objective function. In this implementation case, according to actual engineering requirements, the initial values ​​of the weight coefficients are set to Of course, in some other specific embodiments, the user can also set the weight coefficients to λ1=λ2=λ3=λ4 and λ1+λ2+λ3+λ4=1 based on the principle of equal action.

[0068] S5: Apply the genetic algorithm to manipulate the variable y in step S2 i The optimal interval [y imin ,y imax ] to search for the operating variable value that minimizes the function value of the comprehensive optimization objective function. In this optimization case, the key settings of the genetic algorithm are shown in Table 6.

[0069] Table 6 Key settings of genetic algorithm

[0070]

[0071]

[0072] S6: Determine whether the values ​​of the operation variables in step S5 make the values ​​of the function items of each normalized optimization objective in the comprehensive optimization function less than or equal to 1. If so, go to step S7; otherwise, further determine condition 1 and condition 2;

[0073] Condition 1: The number of weight feedback corrections Nc exceeds the upper limit of the number of weight feedback corrections Nc mac , Nc is the number of weight feedback corrections, and its initial value is zero, Nc mac is the upper limit of the number of weight feedback corrections (20 in this implementation case);

[0074] Condition 2: There exists a delta smaller than 1 1 The normalized optimization objective function value, δ 1 is a range value, which is greater than 0 and less than 1. In this implementation case, δ 1 =0.2;

[0075] If the two conditions are not met at the same time, go to step S7;

[0076] If both conditions are met, increase O j >1 comprehensive optimization objective function O j The corresponding weight coefficient λ j , while reducing O j ≤1-δ 1 The weight coefficient λ corresponding to the normalized optimization objective function value j , weight coefficient λ j Satisfied when modified Then, the weight feedback correction times Nc is increased by 1, that is, Nc=Nc+1; then, go to step S5;

[0077] S7: End the optimization process, output the current optimization calculation results of each operating parameter as the optimization recommendation value, and output the corresponding comprehensive optimization objective function value and each normalized optimization objective function value.

[0078] The relevant statistics of the smelting process indicator variables in one month before and one month after the implementation of the case are shown in Table 7. The data in Table 7 show that the patented method can significantly improve each indicator variable (the mean is closer to the expected optimization value and the standard deviation is smaller).

[0079] Table 7 Comparison of indicators before and after case implementation

[0080]

[0081] In this embodiment, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the multi-objective optimization method described above are implemented.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-objective optimization method for copper flash smelting process, characterized in that: The steps include: S1: Collect historical data, including data of conditional variables, operating variables and indicator variables, to obtain a data set D for building an estimation model; use a modeling method and the data in the data set D to build a prediction model for the indicator variables; S2: Determine the operating variable y to be optimized based on the historical data in step S1 i The optimal interval [y imin ,y imax ]; where i is greater than or equal to 1 and less than or equal to Ny, Ny represents the operation variable y in the input data set D i The number of imin and imax are the lower and upper limits of the optimization interval respectively; S3: According to the relationship between the value of the indicator variable and the degree of optimization it represents, the indicator variable is divided into monotonic indicator variables z k and the non-monotonic indicator variable z l , and construct a monotonic indicator variable z based on the prediction model k and the non-monotonic indicator variable z l The corresponding normalized optimization objective function O j ; S4: weighted summing the multiple normalized optimization objective functions constructed in step S3, thereby constructing a comprehensive optimization objective function based on the prediction model; S5: Apply the genetic algorithm to manipulate the variable y in step S2 i The optimal interval [y imin ,y imax ] to search for the operating variable value that minimizes the function value of the comprehensive optimization objective function; S6: Determine whether the operation variable value in step S5 makes the values ​​of each normalized optimization objective function item in the comprehensive optimization function less than or equal to 1. If so, go to step S7; otherwise, further determine condition 1 and condition 2; Condition 1: The number of weight feedback corrections Nc exceeds the upper limit of the number of weight feedback corrections Nc mac , Nc is the number of weight feedback corrections, and its initial value is zero, Nc mac The upper limit of the number of weight feedback corrections; Condition 2: There exists a delta smaller than 1 1 The normalized optimization objective function value, δ 1 is a range value, which is greater than 0 and less than 1; If the two conditions are not met at the same time, go to step S7; If both conditions are met, increase O j Normalized optimization objective function O > 1 j The corresponding weight coefficient λ j , while reducing O j ≤1-δ 1 The weight coefficient λ corresponding to the normalized optimization objective function value j , weight coefficient λ j Satisfied when modified Nz represents the number of indicator variables; then, the weight feedback correction times Nc is increased by 1, that is, Nc=Nc+1; then, go to step S5; S7: End the optimization process, output the current optimization calculation results of each operating parameter as the optimization recommendation value, and output the corresponding comprehensive optimization objective function value and each normalized optimization objective function value.

2. A multi-objective optimization method according to claim 1, characterized in that: In step S3, the monotonic indicator variable z k The corresponding normalized optimization objective function expression is formula 1): Among them, z kmin is the actual optimal value of the monotonic indicator variable, that is, the actual minimum value; z komax is the expected optimization upper limit; The non-monotonic indicator variable z l The corresponding normalized optimization objective function expression is formula 2): Among them, z lo 、z lomax and z lomin They are the non-monotonic indicator variables z l The expected optimization value, expected optimization upper limit and expected optimization lower limit of .

3. A multi-objective optimization method according to claim 1, characterized in that: The data set D in step S1 is expressed as formula 3): Where i is the data number, x, y and z represent the time period averages of the conditional variable, the operational variable and the indicator variable respectively. express The average value of the previous period; Nx, Ny and Nz represent the number of conditional variables, the number of operating variables and the number of indicator variables respectively; Nd is the total number of data in the data set D; For the indicator variable to be predicted, select the candidate input variable set V = (z p1 ,z p2 ,…,z pNz ,x1,x2,…,x Nx ,y1,y2,…,y Ny ,z1,z2,…,z Nz ); the set V consists of candidate input variables φ corresponding one-to-one to the elements in the set D k composition; Select the candidate input variable φ according to the selection conditions k As the input variable of the prediction model, the prediction model of the indicator variable is constructed using the modeling method and the data in the data set D, which is recorded as formula 4): Where f is the prediction model constructed using the modeling method; is the output variable of the prediction model, Ni is the number of input variables; φ j is the input variable of the prediction model, j is greater than or equal to 1 and less than or equal to Ni.

4. A multi-objective optimization method according to claim 3, characterized in that: The selection condition is: keep other input variables unchanged, compare the alternative input variable φ k The prediction error of the prediction model for the data in the data set D before and after joining; if the error level is significantly reduced after joining, it means that the above conditions are met, that is, the variable should be selected as an input variable, otherwise, it means that the above conditions are not met, that is, the variable should not be selected as an input variable.

5. A multi-objective optimization method according to claim 1, characterized in that: In step S2, y imin and imax The calculation method expression is as follows: min(y i ) and max(y i ) are the operation variables y in the historical data set D i The minimum and maximum values ​​of ; δ is the expansion coefficient, and its value is greater than or equal to 0.05 and less than or equal to 0.

15.

6. A multi-objective optimization method according to claim 1, characterized in that: The comprehensive optimization objective function expression is: The weight coefficient λ j satisfy:

7. A multi-objective optimization method according to claim 1, characterized in that: The comprehensive optimization objective function expression is: Among them, the weight coefficient λ j satisfy:

8. A multi-objective optimization method according to claim 1, characterized in that: The expression of the condition 1 is formula a: Nc<Nc max 9); The expression of condition 2 is formula b: O j ≤1-δ 1 10); said Nc max Greater than or equal to 5 and less than or equal to 20; δ 1 Greater than 0 and less than 1.

9. A multi-objective optimization method according to claim 8, characterized in that: The δ in step S6 1 Greater than 0.2 and less than 0.

5.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the multi-objective optimization method according to any one of claims 1 to 9 are implemented.

Images