Copper smelting optimization method and device, electronic equipment and storage medium

Abstract

This invention provides a copper smelting optimization method, apparatus, electronic device, and storage medium, relating to the field of copper smelting technology. The method includes: constructing an initial dataset, which is obtained by combining sample decision parameters with multiple copper smelting indices; inputting the initial dataset into a corresponding copper smelting surrogate model, outputting target predicted values ​​characterizing the prediction results of the copper smelting indices, wherein the copper smelting surrogate model is determined for each copper smelting index based on a deep Gaussian process and Gaussian noise; constructing multiple merged prediction groups based on the target predicted values ​​corresponding to each copper smelting index, determining a target solution set, and, when a new merged prediction group is added to the target solution set, determining the optimal solution of the merged prediction group for copper smelting optimization and its corresponding optimal solution for decision parameters based on the expected supervolume improvement. This invention can achieve copper smelting optimization under multiple competing objectives and obtain the optimal combination of process parameters in real time.

Description

Technical Field

[0001] This invention relates to the field of copper smelting technology, and in particular to a copper smelting optimization method, apparatus, electronic device, and storage medium. Background Technology

[0002] Copper smelting is a complex industrial process characterized by numerous processes, a long workflow, and inter-process coupling. During copper smelting, the composition of raw materials, equipment operating conditions, and industrial parameters all affect the process's operation. Therefore, optimizing the comprehensive production indicators in copper smelting is a key research issue.

[0003] Copper smelting optimization involves finding a set of process parameters, such as raw material composition, production conditions, and operating parameters, to optimize multiple production indicators, such as maximizing copper output while minimizing waste gas emissions.

[0004] In actual production, decisions are mainly made by technicians based on their long-accumulated experience and relevant process knowledge. However, manual decision-making is highly arbitrary and has low accuracy, making it impossible to guarantee the optimization of production indicators during copper smelting. At the same time, because the copper smelting process involves complex physicochemical reactions, and the relationship between production indicators and decision variables is nonlinear, uncertain, and has complex mechanisms, it is difficult to establish analytical mathematical models for decision optimization. When using evolutionary algorithms for copper smelting optimization, this method is mostly used for optimizing single production indicators, and it requires generating a large number of samples, which is computationally complex and time-consuming, failing to meet real-time requirements. Summary of the Invention

[0005] This invention provides a copper smelting optimization method, apparatus, electronic device, and storage medium to address the shortcomings of existing technologies in terms of low fit to copper smelting problems and low real-time performance, thereby achieving copper smelting optimization under multiple competing objectives and obtaining the optimal combination of process parameters in real time.

[0006] This invention provides an optimized method for copper smelting, comprising:

[0007] An initial dataset is constructed, which is obtained by combining sample decision parameters with multiple copper smelting indicators.

[0008] The initial dataset is input into the corresponding copper smelting surrogate model, and the target predicted value is output to characterize the prediction results of copper smelting indicators. The copper smelting surrogate model is determined for each copper smelting indicator based on a deep Gaussian process and Gaussian noise.

[0009] Multiple merging prediction groups are constructed based on the target predicted values ​​corresponding to each of the copper smelting indices. A target solution set is determined. When a new merging prediction group is added to the target solution set, the optimal solution of the merging prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters are determined based on the expected hypervolume improvement. The target solution set is determined based on the Pareto front approximation set of the current iteration period. The Pareto front approximation set of the current iteration period is obtained by adding the optimal solution of decision parameters that maximizes the expected hypervolume improvement determined in the previous iteration period to the Pareto front approximation set of the previous iteration period. The expected hypervolume improvement is used to characterize the expected improvement of the new merging prediction group on the current copper smelting proxy model.

[0010] According to the copper smelting optimization method provided by the present invention, the step of determining the optimal solution of the combined prediction group and its corresponding optimal solution of decision parameters for copper smelting optimization based on the expected overvolume improvement when a new combined prediction group is added to the target solution set includes:

[0011] Based on the target solution set, the first hypervolume index of the current copper smelting proxy model is determined. The first hypervolume index is the hypervolume value composed of the merged prediction group and the reference prediction group in the target solution set.

[0012] With the addition of a new merged prediction group to the target solution set, a second hypervolume index for the new copper smelting proxy model is determined.

[0013] The amount of hypervolume improvement is determined based on the difference between the second hypervolume index and the first hypervolume index.

[0014] The desired hypervolume improvement is determined based on the hypervolume improvement amount and the multivariate probability density corresponding to the new merging prediction group.

[0015] Based on the expected supervolume improvement, the optimal solution of the combined prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters are determined. The optimal solution of the combined prediction group is used to characterize that each copper smelting index corresponding to the optimal solution of the combined prediction group is better than each copper smelting index corresponding to the combined prediction group other than the optimal solution of the combined prediction group.

[0016] According to the copper smelting optimization method provided by the present invention, determining the optimal solution of the merged prediction group and its corresponding optimal solution of decision parameters for copper smelting optimization based on the expected overvolume improvement includes:

[0017] Based on the first hypervolume index, determine the multi-objective optimization evaluation index;

[0018] When the multi-objective optimization evaluation index converges, the maximum value of the expected hypervolume improvement and the corresponding merging prediction group are determined, the merging prediction group is determined as the optimal solution of the merging prediction group, and the decision parameters corresponding to the optimal solution of the merging prediction group are determined as the optimal solution of the decision parameters.

[0019] According to the copper smelting optimization method provided by the present invention, the step of determining the optimal solution of the merged prediction group and its corresponding optimal solution of decision parameters for copper smelting optimization based on the expected overvolume improvement further includes:

[0020] If the multi-objective optimization evaluation index does not converge, determine the maximum value of the expected hypervolume improvement and the corresponding merging prediction group;

[0021] Determine the decision parameters corresponding to the merged forecast group, and determine the copper smelting indicators corresponding to the decision parameters;

[0022] Add the decision parameters and the copper smelting indicators to the initial dataset to construct an updated dataset;

[0023] The updated dataset is then input into the copper smelting agent model for training until the multi-objective optimization evaluation index converges, thereby determining the optimal solution of the merged prediction group and its corresponding optimal solution for decision parameters.

[0024] According to the copper smelting optimization method provided by the present invention, the deep Gaussian process includes an L-layer Gaussian process;

[0025] The copper smelting agency model is obtained based on the following steps:

[0026] Construct an L-layer Gaussian process, where the output of the previous layer Gaussian process is the input of the next layer Gaussian process, and the input of the first layer Gaussian process is the sample decision parameters;

[0027] Based on the output of the Lth layer Gaussian process and the sum of Gaussian noise, the copper smelting proxy model corresponding to the copper smelting index is determined.

[0028] According to the copper smelting optimization method provided by the present invention, the construction of the initial dataset includes:

[0029] Obtain sample sampling parameters;

[0030] The sample sampling parameters are simulated to determine multiple copper smelting indicators corresponding to the sample sampling parameters;

[0031] The sample sampling parameters are combined with each of the copper smelting indices to obtain the data pairs corresponding to the copper smelting indices;

[0032] Based on each of the data pairs, construct the initial dataset.

[0033] The present invention also provides a copper smelting optimization apparatus, comprising:

[0034] The construction module is used to construct an initial dataset, which is obtained by combining sample decision parameters with multiple copper smelting indicators.

[0035] The output module is used to input the initial dataset into the corresponding copper smelting proxy model and output the target predicted value used to characterize the prediction results of copper smelting indicators. The copper smelting proxy model is determined for each copper smelting indicator based on a deep Gaussian process and Gaussian noise.

[0036] The determination module is used to determine a target solution set based on multiple merged prediction groups constructed according to the target predicted values ​​corresponding to each of the copper smelting indicators, and, when a new merged prediction group is added to the target solution set, to determine the optimal solution of the merged prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters based on the expected hypervolume improvement. The target solution set is determined based on the Pareto front approximation set of the current iteration period. The Pareto front approximation set of the current iteration period is obtained by adding the optimal solution of decision parameters that maximizes the expected hypervolume improvement determined in the previous iteration period to the Pareto front approximation set of the previous iteration period. The expected hypervolume improvement is used to characterize the expected improvement of the new merged prediction group on the current copper smelting proxy model.

[0037] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the copper smelting optimization method as described above.

[0038] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the copper smelting optimization method as described above.

[0039] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the copper smelting optimization method as described above.

[0040] The copper smelting optimization method, apparatus, electronic device, and storage medium provided by this invention address the non-stationary and noisy problems in copper smelting. They construct copper smelting surrogate models corresponding to various copper smelting indicators using a deep Gaussian process and Gaussian noise for effective fitting. Based on the target predicted values ​​output by multiple copper smelting surrogate models, and using the expected hypervolume improvement as a multi-objective acquisition function, they achieve joint optimization of multiple copper smelting indicators. This determines the optimal solutions for multiple copper smelting indicators and the corresponding optimal solutions for decision parameters. The optimal combination of decision parameters can be quickly found through a finite number of iterations and model updates, improving the efficiency of copper smelting optimization and meeting real-time requirements. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0042] Figure 1 This is one of the flowcharts of the copper smelting optimization method provided by the present invention;

[0043] Figure 2 This is the second schematic diagram of the copper smelting optimization method provided by the present invention;

[0044] Figure 3 This is a schematic diagram of the experimental results of the copper smelting optimization method provided by the present invention;

[0045] Figure 4 This is a schematic diagram of the copper smelting simulation provided by the present invention;

[0046] Figure 5 This is a schematic diagram of the structure of the copper smelting optimization device provided by the present invention;

[0047] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0049] Because copper smelting involves complex physicochemical reactions, and production indicators and decision parameters exhibit nonlinearity, uncertainty, and complex mechanisms, it is difficult to establish analytical data models for decision optimization. Evolutionary algorithms, on the other hand, have low optimization efficiency and cannot meet real-time requirements. To address these issues, this invention provides a copper smelting optimization method. Figure 1 This is one of the flowcharts illustrating the optimized copper smelting method provided by the present invention, such as... Figure 1 As shown, the method includes:

[0050] Step 110: Construct an initial dataset, which is obtained by combining sample decision parameters with multiple copper smelting indices.

[0051] Optionally, the sample decision parameters can be a multidimensional array composed of multiple types of data. For example, the sample decision parameters can be a ten-dimensional array [air injection rate, copper concentrate injection rate, hot oil injection rate, electric furnace unit coke injection rate, electric furnace unit air injection rate, boiling water injection rate, C in hot oil]. 10 [H8 content, dryer temperature, moisture content after drying, waste heat boiler pressure], each decision sub-parameter has its own value range; the above copper smelting indicators can be copper yield or waste gas yield; the initial dataset can be multiple data pairs composed of the above ten-dimensional array and copper yield, and multiple data pairs composed of the above ten-dimensional array and waste gas yield.

[0052] Step 120: Input the initial dataset into the corresponding copper smelting surrogate model and output the target predicted value used to characterize the prediction results of copper smelting indicators. The copper smelting surrogate model is determined based on deep Gaussian process and Gaussian noise for each copper smelting indicator.

[0053] Specifically, sudden changes in operating conditions during copper smelting can lead to changes in the chemical reaction state. For example, altering the amount of material injected can affect the intensity of the chemical reaction during copper smelting, thus influencing the final copper smelting result. Consequently, copper smelting indicators exhibit different characteristics under different smelting conditions, indicating a non-stationary problem in the copper smelting process. Therefore, in this embodiment of the invention, a copper smelting proxy model is constructed by combining a deep Gaussian process with Gaussian noise for different copper smelting indicators to better fit the copper smelting process. For example, a copper yield proxy model is constructed with copper yield as the objective function, and a waste gas yield proxy model is constructed with waste gas yield as the objective function.

[0054] Step 130: Based on the target predicted values ​​corresponding to each copper smelting index, construct multiple merged prediction groups, determine the target solution set, and when a new merged prediction group is added to the target solution set, determine the optimal solution of the merged prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters based on the expected hypervolume improvement. The target solution set is determined based on the Pareto front approximation set of the current iteration period. The Pareto front approximation set of the current iteration period is obtained by adding the optimal solution of decision parameters that maximizes the expected hypervolume improvement determined in the previous iteration period to the Pareto front approximation set of the previous iteration period. The expected hypervolume improvement is used to characterize the expected improvement of the new merged prediction group on the current copper smelting proxy model.

[0055] Specifically, in the copper smelting process, multiple copper smelting indicators are in competition, requiring the search for an optimal solution to balance these indicators, such as finding the optimal solution to balance copper yield and waste gas yield. In this embodiment of the invention, the desired supervolume improvement is used as a multi-objective acquisition function to jointly optimize multiple copper smelting indicators, thereby obtaining the optimal solution for the decision parameters corresponding to the optimal solution for balancing the optimal solutions for each copper smelting indicator.

[0056] Optionally, Figure 2 This is the second schematic diagram of the copper smelting optimization method provided by the present invention, as shown below. Figure 2 As shown, when a new merged prediction group is added to the target solution set, determining the optimal solution of the merged prediction group and its corresponding optimal solution of decision parameters for copper smelting optimization based on the expected overvolume improvement includes:

[0057] Based on the target solution set, the first hypervolume index of the current copper smelting proxy model is determined. The first hypervolume index is the hypervolume value composed of the merged prediction group and the reference prediction group in the target solution set.

[0058] With the addition of a new merged prediction group to the target solution set, a second hypervolume index for the new copper smelting proxy model is determined.

[0059] The amount of hypervolume improvement is determined based on the difference between the second hypervolume index and the first hypervolume index.

[0060] The desired hypervolume improvement is determined based on the hypervolume improvement amount and the multivariate probability density corresponding to the new merging prediction group.

[0061] Based on the expected supervolume improvement, the optimal solution of the combined prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters are determined. The optimal solution of the combined prediction group is used to characterize that each copper smelting index corresponding to the optimal solution of the combined prediction group is better than each copper smelting index corresponding to the combined prediction group other than the optimal solution of the combined prediction group.

[0062] Specifically, based on the consideration of balancing various copper smelting indicators, in this embodiment of the invention, the target predicted values ​​output by each copper smelting proxy model are combined into a merging prediction group, and a target solution set is determined based on multiple merging prediction groups. That is, the target solution set includes multiple optimal points that balance various copper smelting proxy models. By calculating the difference between the second hypervolume index after adding a new optimal point to the target solution set and the first hypervolume index of the target solution set, the hypervolume improvement brought about by adding a new optimal point to the target solution set is determined. The expected hypervolume improvement is calculated by integrating the hypervolume improvement with the multivariate probability density of the new optimal point, thereby determining the expected improvement of the current copper smelting proxy model by the new optimal point, so as to achieve joint optimization of various copper smelting indicators.

[0063] Optionally, the target solution set for the current iteration cycle is obtained based on the following steps:

[0064] (1) Determine the first Pareto front approximation set of the previous iteration period as: [Decision parameter 1, decision parameter 2, ..., decision parameter n], and the first objective solution set of the previous iteration period as: [Merged prediction group 1, merged prediction group 2, ..., merged prediction group n];

[0065] (2) If the previous iteration cycle has not converged, the merged prediction group that maximizes the expected supervolume improvement and its corresponding decision parameter k1 are obtained. The decision parameter k1 is added to the first Pareto front approximation set, and the second Pareto front approximation set is obtained as: [decision parameter 1, decision parameter 2, ..., decision parameter n, decision parameter k1]. After training each copper smelting proxy model again in the second iteration cycle, the corresponding second objective solution set is obtained as: [merged prediction group 1', merged prediction group 2', ..., merged prediction group n', merged prediction group k1].

[0066] (3) Add a new optimal point y to the second objective solution set in the current iteration cycle, and the resulting third objective solution set is: [Merge prediction group 1', merge prediction group 2', ..., merge prediction group n', merge prediction group k1, merge prediction group y].

[0067] Optionally, the Pareto front approximation set of the initial iteration period is composed of the decision parameters corresponding to several better combined prediction groups obtained by balancing multiple copper smelting indices from multiple combined prediction groups determined in the initial iteration period.

[0068] Optionally, the hypervolume (HV) index refers to the m-dimensional Lebesgue measure of the dominance subspace with the reference point as the upper limit. Specifically, the first hypervolume index refers to the size of the hypervolume formed by the merged prediction groups in the target solution set, while the second hypervolume index refers to the size of the hypervolume formed by the original merged prediction groups and the new points after adding a new point to the target solution set. In this invention, the hypervolume index measures the size of the dominance subspace of each copper smelting index. The hypervolume index HV(Y) is shown in equation (1), which is:

[0069]

[0070] Where, λ m Let m denote the Wieleberg measure, m represent the number of copper smelting indices, < denote the Pareto dominance order, ∧ denotes AND, Y represents the current objective solution set, and y represents any point in the objective solution set Y. i represents the i-th copper smelting index, r represents the reference point, which can be set based on experience, and z represents any point in the dominance subspace of the corresponding copper smelting index.

[0071] Optionally, based on the above hypervolume indices, the first hypervolume index is determined as HV(Y), and the second hypervolume index is determined as HV(Y∪{y′}), where y′ represents the new merged prediction group. Then, the hypervolume improvement (HVI) HVI(Y,y′) is shown in equation (2), which is:

[0072] HVI(Y,y′)=HV(Y∪{y′})-HV(Y)

[0073] Optionally, the target predicted value includes the mean and variance of the index predicted values ​​output by each copper smelting proxy model. A merged prediction group is constructed from the mean of each copper smelting index, and a target solution set is constructed using this merged prediction group. The first hypervolume index and the second hypervolume index are then calculated, and the hypervolume improvement amount is calculated. When calculating the expected hypervolume improvement amount, considering the uncertainty of the predictions of each copper smelting proxy model, the multivariate probability density corresponding to the new point is determined by the mean and variance of the index predicted values. The expected hypervolume improvement amount is then determined by the multivariate probability density and the hypervolume improvement amount, ensuring the reliability of the expected hypervolume improvement amount.

[0074] Optionally, based on the multivariate probability density and hypervolume improvement of the new merging prediction group, the expected hypervolume improvement (EHVI) is determined. The expected hypervolume improvement EHVI(μ,∑,Y,y′) is shown in Equation (3), which is:

[0075]

[0076] Where μ represents the mean vector composed of the predicted means of multiple indicators, and ∑ represents the oblique variance matrix composed of the predicted variances of multiple indicators. Represents m-dimensional real numbers, PDF μ,∑ (y′) represents the multivariate probability density corresponding to the new point, as shown in equation (4), which is:

[0077]

[0078] For example, taking the joint optimization of copper yield and waste gas yield as an example, the optimization objective can be determined as the optimal solution that maximizes copper yield and minimizes waste gas yield. The current objective solution set contains multiple two-dimensional optimal points composed of the average copper yield and the average waste gas yield output by each copper smelting proxy model. Among the multiple two-dimensional optimal points, the copper yield value of each optimal point is inverted to determine the minimum value of copper yield and waste gas yield in the current objective solution set, and then the first hypervolume index is determined according to equation (1). After that, a new two-dimensional optimal point is added to the current objective solution set to determine the second hypervolume index, and then the hypervolume improvement amount is determined. Based on the average copper yield, the variance of copper yield, the average waste gas yield, and the variance of waste gas yield, the multivariate probability density of the new two-dimensional optimal point is determined, and then the expected hypervolume improvement amount is determined.

[0079] It should be noted that the optimization problem in the above example can be to simultaneously optimize copper production and environmental protection, that is, to maximize copper yield and minimize waste gas yield when the decision parameters are at their optimal values, as shown in equation (5). Equation (5) is:

[0080]

[0081] Where Q(x) represents the copper yield, C(x) represents the waste gas yield, x represents the sample decision parameter, and n represents the number of sample decision parameters. i [x] represents the i-th decision sub-parameter in the sample decision parameters. i,min ,x i,max ] represents the range of values ​​for the i-th decision sub-parameter.

[0082] Optionally, determining the optimal solution of the merged prediction group and its corresponding optimal solution of decision parameters for copper smelting optimization based on the expected supervolume improvement includes:

[0083] Based on the first hypervolume index, determine the multi-objective optimization evaluation index;

[0084] When the multi-objective optimization evaluation index converges, the maximum value of the expected hypervolume improvement and the corresponding merging prediction group are determined, the merging prediction group is determined as the optimal solution of the merging prediction group, and the decision parameters corresponding to the optimal solution of the merging prediction group are determined as the optimal solution of the decision parameters.

[0085] Specifically, the decision parameters and the combined prediction group corresponding to the maximum expected hypervolume improvement are determined, and the multi-objective optimization evaluation index is determined based on the first hypervolume index. When the multi-objective optimization evaluation index converges, it indicates that the fit between each copper smelting proxy model and the real copper smelting process has reached the optimal level, that is, the target prediction value output by each copper smelting proxy model is close to the actual value. At this time, the determined decision parameters are taken as the optimal solution of the decision parameters, and copper smelting can be carried out based on the optimal solution of the decision parameters.

[0086] Optionally, determining the optimal solution of the merged prediction group and its corresponding optimal solution of decision parameters for copper smelting optimization based on the expected supervolume improvement further includes:

[0087] If the multi-objective optimization evaluation index does not converge, determine the maximum value of the expected hypervolume improvement and the corresponding merging prediction group;

[0088] Determine the decision parameters corresponding to the merged forecast group, and determine the copper smelting indicators corresponding to the decision parameters;

[0089] Add the decision parameters and the copper smelting indicators to the initial dataset to construct an updated dataset;

[0090] The updated dataset is then input into the copper smelting agent model for training until the multi-objective optimization evaluation index converges, thereby determining the optimal solution of the merged prediction group and its corresponding optimal solution for decision parameters.

[0091] Specifically, if the multi-objective optimization evaluation index fails to converge, it indicates that the copper smelting surrogate models have not yet been trained. The decision parameters and their corresponding copper smelting indices are then identified as new points, and these new points are added to the initial dataset to obtain an updated dataset. This updated dataset is then input into each copper smelting surrogate model to continue training, and the expected hypervolume improvement is recalculated. Multi-objective joint optimization is then performed again until the multi-objective optimization evaluation index reaches a convergent state, thereby determining the optimal solution for copper smelting.

[0092] Optionally, Figure 3 This is a schematic diagram of the experimental results of the copper smelting optimization method provided by the present invention, as shown in the figure. Figure 3 As shown, the multi-objective optimization evaluation index can be the logarithmic hypervolume, that is, the logarithmic value of the first hypervolume index is used as the multi-objective optimization evaluation index. Figure 3In this invention, the DGMBO scheme lags slightly behind the existing GMBO method in the early stages of iteration. As the number of new points added gradually increases, i.e., with the increase of the number of iterations, after about 20 iterations, the logarithmic hypervolume value of the scheme of this invention is higher than that of the existing GMBO method, and the logarithmic hypervolume value increases faster. With the increase of the number of iterations, the logarithmic hypervolume converges earlier than the existing method. That is, the copper smelting proxy model in this invention approaches the real copper smelting process faster, significantly improving the efficiency of copper smelting optimization. Moreover, the logarithmic hypervolume value of the convergence state of this invention is higher than that of the existing method, i.e., the hypervolume index value of the convergence state of this invention is larger, indicating that the convergence and diversity of the scheme of this invention are superior to those of the existing scheme.

[0093] Optionally, the deep Gaussian process includes an L-layer Gaussian process;

[0094] The copper smelting agency model is obtained based on the following steps:

[0095] Construct an L-layer Gaussian process, where the output of the previous layer Gaussian process is the input of the next layer Gaussian process, and the input of the first layer Gaussian process is the sample decision parameters;

[0096] Based on the output of the Lth layer Gaussian process and the sum of Gaussian noise, the copper smelting proxy model corresponding to the copper smelting index is determined.

[0097] Specifically, in order to address the instability and noise inherent in the copper smelting process, this embodiment of the invention employs a nested structure of multi-level Gaussian models to construct a deep Gaussian process. The output value of the deep Gaussian model is then combined with Gaussian noise to fit the copper smelting process using the noise-driven deep Gaussian process, thereby constructing a copper smelting proxy model.

[0098] Optionally, the deep Gaussian process includes L layers of Gaussian processes, and is a nested structure, that is, the output of the upper layer Gaussian process is the input of the lower layer Gaussian process. The copper smelting proxy model is shown in Equation (5), which is:

[0099] y = f L (f L-1 (…(f2(f1(x)))))+δ

[0100] Wherein, the i-th layer Gaussian process is f i (x), where x represents the input sample decision parameters, and δ represents Gaussian noise, and the Gaussian noise δ~N(0,σ) 2 ), where y represents the target predicted value output by the copper smelting proxy model.

[0101] Optionally, during the training of the copper smelting proxy model, the dataset is input into the copper smelting proxy model, and the minimum value of the loss function is obtained using the gradient descent method. When the loss function reaches its minimum value, the training of the copper smelting proxy model in this iteration cycle is completed, and the predicted values ​​of the indicators, the mean and variance of the predicted values ​​of the indicators corresponding to the decision parameters of multiple samples are obtained. The mean and variance of the predicted values ​​of the indicators are then output as the target predicted values.

[0102] Optionally, to optimize the parameters of the copper smelting proxy model, it is necessary to calculate the marginal likelihood of all samples. This calculation involves inverting multiple covariance matrices, resulting in a large computational load. Therefore, in this embodiment of the invention, minimizing the KL divergence between the variational posterior and the true posterior reduces the computational load. The copper smelting proxy model is trained by maximizing the lower bound of the marginal likelihood. The loss function is shown in equation (6), which is:

[0103]

[0104] Among them, g (j) Let y represent the true value of the j-th sample. (j) Let p(g) represent the predicted value of the j-th sample. (j) |y (j) ) indicates that in y (j) In the event of g (j) The probability of occurrence, q(U) i ) represents the variational distribution, and passes through the variational distribution q(U i To approximate variable U l The posterior distribution, U i =f i (Z i Z i U represents the induced variable input in the i-th Gaussian process. i This indicates that the induced variable Z is added to the input of the i-th level Gaussian process. i The output value after that, p(U) l ) represents U i The probability distribution, Used to optimize prediction results Used to optimize the posterior distribution.

[0105] Optionally, the Adam optimizer can be used to calculate the gradient descent of the above loss function, thereby enabling the training of the copper smelting proxy model.

[0106] Optionally, constructing the initial dataset includes:

[0107] Obtain sample sampling parameters;

[0108] The sample sampling parameters are simulated to determine multiple copper smelting indicators corresponding to the sample sampling parameters;

[0109] The sample sampling parameters are combined with each of the copper smelting indices to obtain the data pairs corresponding to the copper smelting indices;

[0110] Based on each of the data pairs, construct the initial dataset.

[0111] Specifically, in order to supervise the training of the copper smelting proxy model and enable the model to better fit the copper smelting process, in this embodiment of the invention, after collecting sample decision parameters, the sample decision parameters are input into the simulation system for simulation to obtain the copper smelting index corresponding to the sample decision parameters. The copper smelting index is used as the true value of the index. The sample decision parameters and the copper smelting index obtained by simulation are combined into a data pair, and multiple data pairs constitute the initial dataset.

[0112] Optionally, Figure 4 This is a schematic diagram of the copper smelting simulation provided by the present invention, as shown below. Figure 4 As shown, the copper smelting simulation system includes a batching silo, dryer, elevator, dry ore bin, flash furnace, electric furnace, and waste heat boiler. Powdered concentrate is transported from the batching silo to the dryer for deep dehydration. After dehydration, it is lifted by the elevator to the dry ore bin for further drying, where it is thoroughly mixed with air and then sprayed at high speed from the top into the high-temperature flash furnace reaction tower. The dried powdered concentrate is in a suspended state within the reaction tower, and the decomposition, oxidation, and melting of sulfides are essentially completed within 2-3 seconds. The molten sulfides and oxides fall into the sedimentation tank at the bottom of the flash furnace and collect there, continuing the final formation process of matte and slag, and undergoing sedimentation and separation. The matte is discharged through the bottom of the flash furnace, thus yielding the copper yield. The slag is depleted in the electric furnace before being discharged. The high-temperature suspended solids recover waste heat through the waste heat boiler and are discharged as exhaust gas after electrostatic precipitator treatment, thus yielding the exhaust gas yield.

[0113] Optionally, the high speed range can be 60-70m / s, and the high temperature range can be 1450-1550℃.

[0114] The copper smelting optimization method provided by this invention addresses the non-stationary and noisy problems in copper smelting. It constructs copper smelting surrogate models corresponding to various copper smelting indicators through a deep Gaussian process and Gaussian noise for effective fitting. Based on the target prediction values ​​output by multiple copper smelting surrogate models, the expected hypervolume improvement is used as the acquisition function for multiple objectives to achieve joint optimization of multiple copper smelting indicators. This determines the optimal solution for multiple copper smelting indicators and the optimal solution for the corresponding decision parameters. The optimal combination of decision parameters can be quickly found through a finite number of iterations and model updates, improving the efficiency of copper smelting optimization and meeting real-time requirements.

[0115] The copper smelting optimization apparatus provided by the present invention is described below. The copper smelting optimization apparatus described below can be referred to in correspondence with the copper smelting optimization method described above.

[0116] Figure 5 This is a schematic diagram of the copper smelting optimization device provided by the present invention, as shown below. Figure 5 As shown, the copper smelting optimization device 500 includes: a construction module 501, an output module 502, and a determination module 503, wherein:

[0117] Module 501 is used to construct an initial dataset, which is obtained by combining sample decision parameters with multiple copper smelting indicators.

[0118] The output module 502 is used to input the initial dataset into the corresponding copper smelting proxy model and output the target predicted value used to characterize the prediction result of the copper smelting index. The copper smelting proxy model is determined for each copper smelting index based on a deep Gaussian process and Gaussian noise.

[0119] The determination module 503 is used to determine a target solution set based on multiple merged prediction groups constructed according to the target predicted values ​​corresponding to each of the copper smelting indicators, and, when a new merged prediction group is added to the target solution set, to determine the optimal solution of the merged prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters based on the expected hypervolume improvement. The target solution set is determined based on the Pareto front approximation set of the current iteration period. The Pareto front approximation set of the current iteration period is obtained by adding the optimal solution of decision parameters that maximizes the expected hypervolume improvement determined in the previous iteration period to the Pareto front approximation set of the previous iteration period. The expected hypervolume improvement is used to characterize the expected improvement of the new merged prediction group on the current copper smelting proxy model.

[0120] The copper smelting optimization device provided by this invention addresses the non-stationary and noisy problems in copper smelting. It constructs copper smelting surrogate models corresponding to various copper smelting indicators through a deep Gaussian process and Gaussian noise for effective fitting. Based on the target predicted values ​​output by multiple copper smelting surrogate models, the expected hypervolume improvement is used as the acquisition function for multiple objectives to achieve joint optimization of multiple copper smelting indicators. This determines the optimal solution for multiple copper smelting indicators and the optimal solution for the corresponding decision parameters. The optimal combination of decision parameters can be quickly found through a finite number of iterations and model updates, improving the efficiency of copper smelting optimization and meeting real-time requirements.

[0121] Optionally, module 503 is specifically used for:

[0122] When a new merged prediction group is added to the target solution set, the optimal solution of the merged prediction group for copper smelting optimization and its corresponding optimal solution for decision parameters are determined based on the expected overvolume improvement, including:

[0123] Based on the target solution set, the first hypervolume index of the current copper smelting proxy model is determined. The first hypervolume index is the hypervolume value composed of the merged prediction group and the reference prediction group in the target solution set.

[0124] With the addition of a new merged prediction group to the target solution set, a second hypervolume index for the new copper smelting proxy model is determined.

[0125] The amount of hypervolume improvement is determined based on the difference between the second hypervolume index and the first hypervolume index.

[0126] The desired hypervolume improvement is determined based on the hypervolume improvement amount and the multivariate probability density corresponding to the new merging prediction group.

[0127] Based on the expected supervolume improvement, the optimal solution of the combined prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters are determined. The optimal solution of the combined prediction group is used to characterize that each copper smelting index corresponding to the optimal solution of the combined prediction group is better than each copper smelting index corresponding to the combined prediction group other than the optimal solution of the combined prediction group.

[0128] Optionally, module 503 is specifically used for:

[0129] The process of determining the optimal solution of the merged prediction group and its corresponding optimal solution for decision parameters for copper smelting optimization based on the expected overvolume improvement includes:

[0130] Based on the first hypervolume index, determine the multi-objective optimization evaluation index;

[0131] When the multi-objective optimization evaluation index converges, the maximum value of the expected supervolume improvement and the corresponding maximum merged prediction group are determined. The maximum merged prediction group is determined as the optimal solution of the merged prediction group, and the decision parameters corresponding to the optimal solution of the merged prediction group are determined as the optimal solution of the decision parameters.

[0132] Optionally, module 503 is specifically used for:

[0133] The step of determining the optimal solution of the merged prediction group and its corresponding optimal solution of decision parameters for copper smelting optimization based on the expected overvolume improvement also includes:

[0134] If the multi-objective optimization evaluation index does not converge, determine the maximum value of the expected hypervolume improvement and the corresponding merging prediction group;

[0135] Determine the decision parameters corresponding to the merged forecast group, and determine the copper smelting indicators corresponding to the decision parameters;

[0136] Add the decision parameters and the copper smelting indicators to the initial dataset to construct an updated dataset;

[0137] The updated dataset is then input into the copper smelting agent model for training until the multi-objective optimization evaluation index converges, thereby determining the optimal solution of the merged prediction group and its corresponding optimal solution for decision parameters.

[0138] Optionally, the output module 502 is specifically used for:

[0139] The deep Gaussian process includes an L-layer Gaussian process;

[0140] The copper smelting agency model is obtained based on the following steps:

[0141] Construct an L-layer Gaussian process, where the output of the previous layer Gaussian process is the input of the next layer Gaussian process, and the input of the first layer Gaussian process is the sample decision parameters;

[0142] Based on the output of the Lth layer Gaussian process and the sum of Gaussian noise, the copper smelting proxy model corresponding to the copper smelting index is determined.

[0143] Optionally, module 501 is constructed specifically for:

[0144] The construction of the initial dataset includes:

[0145] Obtain sample sampling parameters;

[0146] The sample sampling parameters are simulated to determine multiple copper smelting indicators corresponding to the sample sampling parameters;

[0147] The sample sampling parameters are combined with each of the copper smelting indices to obtain the data pairs corresponding to the copper smelting indices;

[0148] Based on each of the data pairs, construct the initial dataset.

[0149] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6 As shown, the electronic device 600 may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a copper smelting optimization method, which includes:

[0150] An initial dataset is constructed, which is obtained by combining sample decision parameters with multiple copper smelting indicators.

[0151] The initial dataset is input into the corresponding copper smelting surrogate model, and the target predicted value is output to characterize the prediction results of copper smelting indicators. The copper smelting surrogate model is determined for each copper smelting indicator based on a deep Gaussian process and Gaussian noise.

[0152] Multiple merging prediction groups are constructed based on the target predicted values ​​corresponding to each of the copper smelting indices. A target solution set is determined. When a new merging prediction group is added to the target solution set, the optimal solution of the merging prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters are determined based on the expected hypervolume improvement. The target solution set is determined based on the Pareto front approximation set of the current iteration period. The Pareto front approximation set of the current iteration period is obtained by adding the optimal solution of decision parameters that maximizes the expected hypervolume improvement determined in the previous iteration period to the Pareto front approximation set of the previous iteration period. The expected hypervolume improvement is used to characterize the expected improvement of the new merging prediction group on the current copper smelting proxy model.

[0153] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0154] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the copper smelting optimization method provided by the above methods, the method comprising:

[0155] An initial dataset is constructed, which is obtained by combining sample decision parameters with multiple copper smelting indicators.

[0156] The initial dataset is input into the corresponding copper smelting surrogate model, and the target predicted value is output to characterize the prediction results of copper smelting indicators. The copper smelting surrogate model is determined for each copper smelting indicator based on a deep Gaussian process and Gaussian noise.

[0157] Multiple merging prediction groups are constructed based on the target predicted values ​​corresponding to each of the copper smelting indices. A target solution set is determined. When a new merging prediction group is added to the target solution set, the optimal solution of the merging prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters are determined based on the expected hypervolume improvement. The target solution set is determined based on the Pareto front approximation set of the current iteration period. The Pareto front approximation set of the current iteration period is obtained by adding the optimal solution of decision parameters that maximizes the expected hypervolume improvement determined in the previous iteration period to the Pareto front approximation set of the previous iteration period. The expected hypervolume improvement is used to characterize the expected improvement of the new merging prediction group on the current copper smelting proxy model.

[0158] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the copper smelting optimization method provided by the methods described above, the method comprising:

[0159] An initial dataset is constructed, which is obtained by combining sample decision parameters with multiple copper smelting indicators.

[0160] The initial dataset is input into the corresponding copper smelting surrogate model, and the target predicted value is output to characterize the prediction results of copper smelting indicators. The copper smelting surrogate model is determined for each copper smelting indicator based on a deep Gaussian process and Gaussian noise.

[0161] Multiple merging prediction groups are constructed based on the target predicted values ​​corresponding to each of the copper smelting indices. A target solution set is determined. When a new merging prediction group is added to the target solution set, the optimal solution of the merging prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters are determined based on the expected hypervolume improvement. The target solution set is determined based on the Pareto front approximation set of the current iteration period. The Pareto front approximation set of the current iteration period is obtained by adding the optimal solution of decision parameters that maximizes the expected hypervolume improvement determined in the previous iteration period to the Pareto front approximation set of the previous iteration period. The expected hypervolume improvement is used to characterize the expected improvement of the new merging prediction group on the current copper smelting proxy model.

[0162] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0163] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An optimized method for copper smelting, characterized in that, include: An initial dataset is constructed, which is obtained by combining sample decision parameters with multiple copper smelting indicators. The initial dataset is input into the corresponding copper smelting surrogate model, and the target predicted value is output to characterize the prediction results of copper smelting indicators. The copper smelting surrogate model is determined for each copper smelting indicator based on a deep Gaussian process and Gaussian noise. Multiple merging prediction groups are constructed based on the target predicted values ​​corresponding to each copper smelting index. A target solution set is determined. When a new merging prediction group is added to the target solution set, the optimal solution of the merging prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters are determined based on the expected supervolume improvement. The target solution set is determined based on the Pareto front approximation set of the current iteration period. The Pareto front approximation set of the current iteration period is obtained by adding the optimal solution of decision parameters that maximizes the expected supervolume improvement determined in the previous iteration period to the Pareto front approximation set of the previous iteration period. The expected supervolume improvement is used to characterize the expected improvement of the new merging prediction group on the current copper smelting proxy model. When a new merged prediction group is added to the target solution set, the optimal solution of the merged prediction group for copper smelting optimization and its corresponding optimal solution for decision parameters are determined based on the expected overvolume improvement, including: Based on the target solution set, the first hypervolume index of the current copper smelting proxy model is determined. The first hypervolume index is the hypervolume value composed of the merged prediction group and the reference prediction group in the target solution set. With the addition of a new merged prediction group to the target solution set, a second hypervolume index for the new copper smelting proxy model is determined. The amount of hypervolume improvement is determined based on the difference between the second hypervolume index and the first hypervolume index. The desired hypervolume improvement is determined based on the hypervolume improvement amount and the multivariate probability density corresponding to the new merging prediction group. Based on the expected supervolume improvement, the optimal solution of the combined prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters are determined. The optimal solution of the combined prediction group is used to characterize that each copper smelting index corresponding to the optimal solution of the combined prediction group is better than each copper smelting index corresponding to the combined prediction group other than the optimal solution of the combined prediction group.

2. The copper smelting optimization method according to claim 1, characterized in that, The process of determining the optimal solution of the merged prediction group and its corresponding optimal solution for decision parameters for copper smelting optimization based on the expected overvolume improvement includes: Based on the first hypervolume index, determine the multi-objective optimization evaluation index; When the multi-objective optimization evaluation index converges, the maximum value of the expected hypervolume improvement and the corresponding merging prediction group are determined, the merging prediction group is determined as the optimal solution of the merging prediction group, and the decision parameters corresponding to the optimal solution of the merging prediction group are determined as the optimal solution of the decision parameters.

3. The copper smelting optimization method according to claim 2, characterized in that, The step of determining the optimal solution of the merged prediction group and its corresponding optimal solution of decision parameters for copper smelting optimization based on the expected overvolume improvement also includes: If the multi-objective optimization evaluation index does not converge, determine the maximum value of the expected hypervolume improvement and the corresponding merging prediction group; Determine the decision parameters corresponding to the merged forecast group, and determine the copper smelting indicators corresponding to the decision parameters; Add the decision parameters and the copper smelting indicators to the initial dataset to construct an updated dataset; The updated dataset is then input into the copper smelting agent model for training until the multi-objective optimization evaluation index converges, thereby determining the optimal solution of the merged prediction group and its corresponding optimal solution for decision parameters.

4. The copper smelting optimization method according to any one of claims 1 to 3, characterized in that, The deep Gaussian process includes an L-layer Gaussian process; The copper smelting agency model is obtained based on the following steps: Construct an L-layer Gaussian process, where the output of the previous layer Gaussian process is the input of the next layer Gaussian process, and the input of the first layer Gaussian process is the sample decision parameters; Based on the output of the Lth layer Gaussian process and the sum of Gaussian noise, the copper smelting proxy model corresponding to the copper smelting index is determined.

5. The copper smelting optimization method according to claim 4, characterized in that, The construction of the initial dataset includes: Obtain sample sampling parameters; The sample sampling parameters are simulated to determine multiple copper smelting indicators corresponding to the sample sampling parameters; The sample sampling parameters are combined with each of the copper smelting indices to obtain the data pairs corresponding to the copper smelting indices; Based on each of the data pairs, construct the initial dataset.

6. An optimized copper smelting apparatus, characterized in that, include: The construction module is used to construct an initial dataset, which is obtained by combining sample decision parameters with multiple copper smelting indicators. The output module is used to input the initial dataset into the corresponding copper smelting proxy model and output the target predicted value used to characterize the prediction results of copper smelting indicators. The copper smelting proxy model is determined for each copper smelting indicator based on a deep Gaussian process and Gaussian noise. The determination module is used to determine the target solution set based on multiple merged prediction groups constructed according to the target predicted values ​​corresponding to each of the copper smelting indicators, and, when a new merged prediction group is added to the target solution set, to determine the optimal solution of the merged prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters based on the expected supervolume improvement. The target solution set is determined based on the Pareto front approximation set of the current iteration period. The Pareto front approximation set of the current iteration period is obtained by adding the optimal solution of decision parameters that maximizes the expected supervolume improvement determined in the previous iteration period to the Pareto front approximation set of the previous iteration period. The expected supervolume improvement is used to characterize the expected improvement of the new merged prediction group on the current copper smelting proxy model. The determining module is specifically used for: determining a first hypervolume index of the current copper smelting proxy model based on the target solution set, wherein the first hypervolume index is the hypervolume value composed of the merged prediction group and the reference prediction group in the target solution set; determining a second hypervolume index of the new copper smelting proxy model when a new merged prediction group is added to the target solution set; determining a hypervolume improvement amount based on the difference between the second hypervolume index and the first hypervolume index; determining a desired hypervolume improvement amount based on the hypervolume improvement amount and the multivariate probability density corresponding to the new merged prediction group; and determining the optimal solution of the merged prediction group for copper smelting optimization and its corresponding optimal solution of decision parameters based on the desired hypervolume improvement amount, wherein the optimal solution of the merged prediction group is used to characterize that each copper smelting index corresponding to the optimal solution of the merged prediction group is better than each copper smelting index corresponding to the merged prediction group other than the optimal solution of the merged prediction group.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the copper smelting optimization method as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the copper smelting optimization method as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the copper smelting optimization method as described in any one of claims 1 to 5.

Images