A method of charging a flash furnace

Abstract

The application provides a method for batching a flash furnace, which comprises: generating a plurality of batching schemes; inputting each batching scheme into a digital model of the flash furnace for simulation calculation to obtain a corresponding heat balance parameter of each batching scheme; calculating a corresponding comprehensive loss value of each batching scheme to update a mean vector and a covariance matrix in a covariance matrix self-adaptive evolution algorithm; judging whether a difference between a predicted value and a target value of the heat balance parameter of each batching scheme exceeds an allowable error; if there is a batching scheme that does not exceed the allowable error, outputting the batching scheme and ending; and if all the batching schemes exceed the allowable error, repeatedly executing the above steps according to the updated mean vector and covariance matrix. The batching scheme is generated by an algorithm, and simulation calculation is performed to predict the heat balance parameter of each batching scheme, and the batching scheme is evaluated by judging the heat balance parameter, so that the optimal batching scheme is obtained.

Description

Technical Field

[0001] This invention relates to the field of metallurgical automation and intelligent manufacturing technology, and in particular to a batching method for a flash furnace. Background Technology

[0002] In copper smelting, the flash furnace, as the core reaction equipment, has a decisive impact on the efficiency, energy consumption, product quality, and environmental indicators of the entire smelting process. Batching is a crucial upstream step in flash furnace operation, directly affecting the furnace's heat balance, flue gas generation, slag shape control, and the enrichment behavior of harmful elements (such as arsenic, antimony, bismuth, and nickel). Currently, traditional flash furnace batching mainly relies on manual experience combined with simple spreadsheet calculations, which presents the following problems.

[0003] Firstly, the furnace condition is unstable: because the control of heat balance (such as reaction heat and cooling load) and flue gas balance (such as SO2 concentration and flue gas volume) depends on experience judgment, it is easily affected by raw material fluctuations, which can lead to nodule formation, slag buildup or abnormal heat load in the furnace, affecting the operating rate.

[0004] Secondly, inaccurate element balance control: excessive moisture content affects combustion efficiency; imbalance in slag-iron-silicon ratio affects slag shape control; excessive levels of impurities such as arsenic, antimony, bismuth, and nickel can easily lead to a decline in anode plate quality or environmental risks.

[0005] Third, the material inventory management is crude: insufficient cold material inventory or full batching warehouses lead to supply interruptions; excessive occupation of concentrate warehouses affects the allocation of other batches of raw materials, and there is a lack of dynamic coordination mechanism.

[0006] Fourth, knowledge transfer is difficult: excellent ingredient formulation plans lack a systematic storage and reuse mechanism, new employees are slow to get started, and experience is seriously lost.

[0007] Although some companies have tried to introduce automated batching systems, most of them are still based on fixed algorithms or static rules, lacking dynamic adaptability and failing to meet the needs of modern copper smelting for efficient, stable and intelligent production. Summary of the Invention

[0008] In view of the above-mentioned defects in the prior art, the present invention provides a batching method for a flash furnace to solve the technical problems of unintelligent, unreasonable and unstable batching schemes in the prior art.

[0009] To achieve the above and other related objectives, this invention provides a batching method for a flash furnace, comprising: generating multiple batching schemes using a covariance matrix adaptive evolutionary algorithm; inputting each batching scheme and preset material composition data into a flash furnace digital model for simulation calculation to obtain predicted values ​​of process indicators corresponding to each batching scheme, wherein the process indicators include heat balance parameters; calculating a comprehensive loss value corresponding to each batching scheme based on the predicted values ​​of the process indicators and a predefined overall objective function; updating the mean vector and covariance matrix in the covariance matrix adaptive evolutionary algorithm based on the comprehensive loss value corresponding to each batching scheme; determining whether the difference between the predicted value and the target value of the heat balance parameter of each batching scheme exceeds the allowable error: if there is a batching scheme that does not exceed the allowable error, outputting the batching scheme and ending the process; if all batching schemes exceed the allowable error, repeating the above steps based on the updated mean vector and covariance matrix.

[0010] In one embodiment of the present invention, multiple ingredient schemes are generated using an adaptive evolutionary algorithm based on the covariance matrix, including: constructing a Gaussian distribution x~N(m k C k ), where x is a single batching scheme, m k C k The mean vector and the covariance matrix are respectively used in the k-th iteration. Multiple ingredient schemes are generated by sampling from the Gaussian distribution in each iteration.

[0011] In one embodiment of the present invention, before the step of generating multiple batching schemes using the covariance matrix adaptive evolution algorithm, the method further includes: calculating the similarity between the current production task of the flash furnace and the saved historical cases, and finding the most similar historical case; using the batching scheme of the most similar historical case as the initial mean vector in the covariance matrix adaptive evolution algorithm.

[0012] In one embodiment of the present invention, the comprehensive loss value corresponding to each batching scheme is calculated based on the predicted values ​​of the process indicators and the predefined overall objective function, including: calculating the target deviation loss based on the predicted and target values ​​of the heat balance parameters and a first weighting coefficient; calculating the constraint violation loss based on the predicted values ​​of the chemical balance parameters in the process indicators, the constraints, and a second weighting coefficient; calculating the material level feasibility penalty loss based on the batching scheme, the initial silo level, the minimum safe material level, and the optimization cycle; and obtaining the comprehensive loss value based on the target deviation loss, the constraint violation loss, and the material level feasibility penalty loss.

[0013] In one embodiment of the present invention, the feasibility penalty loss for material level is calculated based on the batching scheme, the initial silo level, the minimum safe level, and the optimization cycle. This includes: obtaining the usage amount of each material based on the batching scheme and the optimization cycle; obtaining the silo overuse loss based on the usage amount of each material, the initial silo level, and the silo overuse penalty coefficient; obtaining the low-level usage loss based on the initial silo level, the minimum safe level, the indicator function, and the low-level usage penalty coefficient; and obtaining the feasibility penalty loss for material level based on the silo overuse loss and the low-level usage loss.

[0014] In one embodiment of the present invention, before the step of calculating the comprehensive loss value corresponding to each batching scheme, the method includes: calculating the similarity between the current production task of the flash furnace and the saved historical cases, and finding the most similar historical case; adjusting the weight coefficients in the overall objective function based on the similarity of the most similar historical case.

[0015] In one embodiment of the present invention, adjusting the weight coefficients in the overall objective function based on the similarity of the most similar historical cases includes: decreasing the weight coefficients in the overall objective function when the similarity of the most similar historical cases is greater than or equal to a first threshold; maintaining the weight coefficients in the overall objective function unchanged when the similarity of the most similar historical cases is greater than or equal to a second threshold and less than the first threshold; and increasing the weight coefficients in the overall objective function when the similarity of the most similar historical cases is less than the second threshold.

[0016] In one embodiment of the present invention, the weight coefficients in the total objective function after reduction or increase are calculated based on the similarity of the most similar historical cases, a preset adjustment coefficient, and the standard value of the weight coefficients in the total objective function.

[0017] In one embodiment of the present invention, updating the mean vector and covariance matrix in the covariance matrix adaptive evolution algorithm according to the comprehensive loss value corresponding to each ingredient scheme includes: selecting several ingredient schemes with low comprehensive loss values ​​in the current iteration according to the comprehensive loss value corresponding to each ingredient scheme; updating the mean vector in the covariance matrix adaptive evolution algorithm using the weighted average of the several ingredient schemes with low comprehensive loss values; and calculating the updated covariance matrix according to the covariance matrix before the update, the preset covariance matrix update step size, and the direction vector of the search direction of excellent individuals.

[0018] In one embodiment of the present invention, before the step of generating multiple batching schemes using the covariance matrix adaptive evolution algorithm, the method further includes: acquiring heterogeneous multi-source data from the actual production process of the flash furnace and preprocessing it; wherein, the heterogeneous multi-source data includes preset parameters and real-time acquired parameters; the preset parameters include target values ​​of thermal balance parameters, constraints of chemical balance parameters, and silo composition data; the real-time acquired parameters include the current silo level; the preprocessing includes one or more of the following: data interpolation, abnormal data filtering, normalization, and data time alignment.

[0019] The beneficial effects of this invention are as follows: This invention proposes a batching method for a flash furnace. This method generates batching schemes through a covariance matrix adaptive evolutionary algorithm and performs simulation calculations using a digital model to predict the heat balance parameters of each batching scheme. The batching schemes are evaluated by judging the heat balance parameters. At the same time, the comprehensive loss value of each batching scheme is calculated through a predefined overall objective function to update the covariance matrix adaptive evolutionary algorithm. This ensures that when there is no batching scheme that meets the requirements, a more suitable batching scheme can be found through algorithm iteration. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The accompanying drawings are incorporated in and constitute a part of this specification, illustrating embodiments consistent with this application, and are used together with the description to explain the principles of this application. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0021] Figure 1 This is a general flowchart of a batching method provided in an embodiment of the present invention;

[0022] Figure 2 A flowchart illustrating the process of obtaining the initial mean vector according to an embodiment of the present invention;

[0023] Figure 3 A flowchart illustrating the calculation of the comprehensive loss value provided in an embodiment of the present invention;

[0024] Figure 4 A flowchart for calculating material level feasibility penalty loss provided in an embodiment of the present invention;

[0025] Figure 5 A flowchart illustrating the introduction of adjustable weighting coefficients is provided in one embodiment of the present invention;

[0026] Figure 6 A flowchart illustrating the adjustment of weighting coefficients according to an embodiment of the present invention;

[0027] Figure 7 This is a flowchart illustrating the update process of the mean vector and covariance matrix according to an embodiment of the present invention. Detailed Implementation

[0028] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. It should be noted that, unless otherwise specified, the following embodiments and features can be combined with each other. In addition to the specific methods, equipment, and materials used in the embodiments, based on the knowledge of the prior art and the description of the present invention by those skilled in the art, any prior art methods, equipment, and materials similar to or equivalent to the methods, equipment, and materials in the embodiments of the present invention can be used to implement the present invention.

[0029] It should be understood that the terminology used in the embodiments of this invention is for describing specific implementations and not for limiting the scope of protection of this invention. Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art.

[0030] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In some embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0031] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functions, and operations that may be implemented in the methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0032] Please see Figure 1 , Figure 1An embodiment of the present invention provides a batching method for a flash furnace, comprising steps S101 to S105.

[0033] Step S101: Generate multiple ingredient plans using the Covariance Matrix Adaptive Evolutionary Algorithm (CMA-ES). The Covariance Matrix Adaptive Evolutionary Algorithm (CMA-ES) is an adaptive search algorithm that adaptively adjusts the search distribution during optimization, thereby efficiently exploring and utilizing the solution space of the ingredient plans. This algorithm can generate multiple ingredient plans to be selected.

[0034] In a specific embodiment of the present invention, step S101 includes: constructing a Gaussian distribution x~N(m k C k ), where x is a single batching scheme, m k C k These are the mean vector and covariance matrix at the k-th iteration, respectively, with mean vector m. k This refers to the ingredient combination scheme that is currently considered most likely to succeed. In each iteration, multiple ingredient combinations are generated by sampling from the Gaussian distribution. For example, 10 to 50 different ingredient combinations {x1,x2,x3,…} can be sampled in each iteration.

[0035] The covariance matrix adaptive evolution algorithm updates the mean vector and covariance matrix during iteration, and these two parameters also need to be set with an initial value.

[0036] The initial mean vector (i.e. the initial value of the mean vector) can be, for example, a batching scheme specified by an experienced engineer, a batching scheme currently being implemented in the flash furnace system, or a batching scheme used in the case most similar to the current production conditions found from historical cases.

[0037] Please see Figure 2 In a specific embodiment of the present invention, steps S201 and S202 are included before step S101.

[0038] Step S201: Calculate the similarity between the current production task of the flash furnace and the saved historical cases, and find the most similar historical case.

[0039] For each production process, feature vectors can be extracted based on some parameter settings of the current production task. The corresponding feature vectors can also be extracted from historical cases (including historical production tasks and their optimal ingredient plans) in the same way. Then, the similarity between the two can be calculated using the following formula:

[0040] ,

[0041] In the formula, the superscript i represents the number of the historical case. That is, the similarity between the current production task and the i-th historical case.

[0042] Step S202: Use the ingredient plan of the most similar historical case as the initial mean vector in the covariance matrix adaptive evolution algorithm. For the current production task, there is a similarity between it and each historical case. Find the historical case with the highest similarity and use its corresponding ingredient plan as the initial mean vector.

[0043] The initial covariance matrix (i.e. the initial value of the covariance matrix) is usually set as an identity matrix or a diagonal matrix, indicating that there is no directional preference at the beginning.

[0044] Step S102: Input each batching scheme and the preset material composition data into the flash furnace digital model for simulation calculation to obtain the predicted values ​​of the process indicators corresponding to each batching scheme, including heat balance parameters. Here, a digital model corresponding to a real flash furnace is constructed, and the predicted values ​​of the process indicators corresponding to each batching scheme are obtained through simulation calculation.

[0045] In this embodiment, the preset material composition data can be obtained from the LIMS system, which is a laboratory information management system that stores and manages all high-quality, high-reliability material composition data. Its data is characterized by being very accurate, but the update frequency is low (because testing takes time).

[0046] In this embodiment, the process parameters generally include chemical balance parameters in addition to thermal balance parameters. Thermal balance parameters typically include flux ratio, process oxygen, oxygen coefficient, and amount of flue dust used; chemical balance parameters include matte grade, slag-iron-silicon ratio, immersion slag temperature, lead-zinc content, arsenic-antimony-bismuth-nickel content, magnesium-aluminum content, and concentrate moisture content.

[0047] Step S103: Based on the predicted values ​​of the process indicators and the predefined overall objective function, calculate the comprehensive loss value corresponding to each batching scheme. This step transforms the complex multi-objective optimization problem into a quantifiable single-objective minimization problem. By calculating the comprehensive loss value, the system can clearly and objectively compare the advantages and disadvantages of different batching schemes. This provides a clear search direction for subsequent optimization algorithms, enabling them to efficiently select the feasible scheme that is optimal in terms of technology, safety, and economy, rather than a solution that only satisfies a single condition.

[0048] Please see Figure 3 In a specific embodiment of the present invention, step S103 includes steps S301 to S304.

[0049] Step S301: Calculate the target deviation loss based on the predicted and target values ​​of the heat balance parameters and the first weighting coefficient. The target deviation loss L1 can be calculated, for example, using the following formula:

[0050] ,

[0051] In the formula, M is the total number of thermal equilibrium parameters, and w i f is the first weighting coefficient for the i-th thermal equilibrium parameter. i (x) represents the predicted value of the i-th heat balance parameter corresponding to the ingredient scheme x. Let be the target value of the i-th thermal equilibrium parameter.

[0052] Step S302: Based on the predicted values ​​of the chemical equilibrium parameters in the process indicators, the constraints, and the second weighting coefficient, calculate the constraint violation loss. The constraint violation loss L2 can be calculated, for example, using the following formula:

[0053] ,

[0054] In the formula, N is the total number of chemical equilibrium parameters, and λ j The second weighting coefficient for the j-th chemical equilibrium parameter is... and Let $\frac{j}{j}$ be the constraints for the $j$-th chemical equilibrium parameter, corresponding to the minimum and maximum values, respectively. In this formula, when the predicted value of the chemical equilibrium parameter is between the minimum and maximum values, the last two terms in $\frac{max}{j} are both negative, and this loss is 0. When the predicted value of the chemical equilibrium parameter is lower than the minimum or higher than the maximum value, one of the last two terms in $\frac{max}{j} will be positive, and this loss will not be 0. The reason for squaring the result of $\frac{max}{j}$ in this formula is to amplify the penalty for serious violation of the constraints.

[0055] Step S303: Calculate the material level feasibility penalty loss based on the batching plan, initial silo level, minimum safe level, and optimization cycle.

[0056] Please see Figure 4 In a specific embodiment of the present invention, step S303 includes steps S401 to S404.

[0057] Step S401: Based on the batching plan and optimization cycle, obtain the usage of each material. Generally, the calculated batching plan is the material ratio. For example, if there are three materials, the batching plan is {20%, 30%, 50%}. The usage per unit time of each material can be calculated based on the ratio and the target total output. The optimization cycle represents the duration for which a batching plan needs to run stably. It is generally determined by the process, but can also be set manually, such as 4 hours, 8 hours, or one day. Given the batching plan and optimization cycle, the total usage of each material within the optimization cycle can be calculated.

[0058] Step S402: Based on the usage of each material, the initial silo level, and the silo over-use penalty coefficient, obtain the silo over-use loss. The silo over-use loss can be calculated, for example, using the following formula:

[0059] ,

[0060] In the formula, L 3i This represents the over-utilization loss of the silo for the i-th material in the batching scheme x. x is the penalty coefficient for overuse of the silo. i T represents the unit time usage of the i-th material in the batching scheme x. horizon To optimize the cycle, Q i Let be the initial silo level for the i-th material.

[0061] Step S403: Based on the initial silo level, minimum safe level, indication function, and low level usage penalty coefficient, obtain the low level usage loss. The low level usage loss can be calculated, for example, using the following formula:

[0062] ,

[0063] In the formula, I is an indicator function; when the i-th material is used in the plan, I = 1; when the i-th material is not used in the plan, I = 0. 4i This represents the low-level material usage loss of the i-th material in the batching scheme x. A penalty coefficient is applied to low material levels. This represents the minimum safe level for the i-th material.

[0064] Step S404: Based on the overuse loss and low-level usage loss of the silo, obtain the level feasibility penalty loss. The level feasibility penalty loss L5 can be calculated, for example, using the following formula:

[0065] ,

[0066] In the formula, n is the total number of materials.

[0067] Step S304: Based on the target deviation loss, constraint violation loss, and material level feasibility penalty loss, obtain the comprehensive loss value, which can be expressed by the formula: L(x) = L1 + L2 + L5. That is, the comprehensive loss value is equal to the sum of these three losses.

[0068] Please see Figure 5 In a specific embodiment of the present invention, the step of calculating the comprehensive loss value corresponding to each ingredient scheme includes steps S501 and S502. Step S501 is the same as the previous step S201; that is, when performing the aforementioned step S201, in addition to continuing to perform step S202, step S502 can also be performed based on the most similar historical case.

[0069] Step S502: Adjust the weight coefficients in the overall objective function based on the similarity of the most similar historical cases.

[0070] Please see Figure 6 In a specific embodiment of the present invention, step S502 includes steps S601 to S603.

[0071] Step S601: When the similarity of the most similar historical cases is greater than or equal to the first threshold, decrease the weight coefficient in the overall objective function. When there are highly similar historical cases, we can choose to trust historical experience and allow slight constraint violations, thus decreasing the weight coefficient in the overall objective function.

[0072] Step S602: When the similarity of the most similar historical case is greater than or equal to the second threshold and less than the first threshold, the weight coefficients in the overall objective function remain unchanged. If the similarity of historical cases is neither too high nor too low, the weight coefficients are neither strengthened nor weakened.

[0073] Step S603: When the similarity of the most similar historical case is less than the second threshold, increase the weight coefficient in the overall objective function. When the similarity is low, it indicates that the historical case differs significantly from the current production task, and the constraints must be strictly followed; therefore, the weight coefficient in the overall objective function is increased.

[0074] In a specific embodiment of the present invention, the weight coefficients in the overall objective function, after being reduced or increased, are calculated based on the similarity of the most similar historical case, a preset adjustment coefficient, and the standard value of the weight coefficients in the overall objective function. The following detailed description is provided through specific embodiments, wherein the similarity of the most similar historical case is denoted as Sim. max The first threshold is set to 0.85, and the second threshold is set to 0.7.

[0075] In a specific embodiment of the present invention, when Sim max When the value is ≥0.85, the first and second weighting coefficients are adjusted according to the following formula:

[0076] , .

[0077] When 0.7≤Sim max When <0.85, the first and second weighting coefficients remain unchanged, that is:

[0078] , .

[0079] When Sim max If the value is less than 0.7, the first and second weighting coefficients will be adjusted according to the following formula:

[0080] ,

[0081] .

[0082] When adjustable weighting coefficients are used, w is no longer used in the specific calculation formulas for the target deviation loss L1 and constraint violation loss L2. i and λ j Instead, it was changed to and That is, using the adjusted first weighting coefficient To calculate the target deviation loss L1, the adjusted second weighting coefficient is used. To calculate the constraint violation loss L2.

[0083] Step S104: Update the mean vector and covariance matrix in the covariance matrix adaptive evolution algorithm according to the comprehensive loss value corresponding to each ingredient scheme.

[0084] Please see Figure 7 In a specific embodiment of the present invention, step S104 includes steps S701 to S703.

[0085] Step S701: Based on the comprehensive loss value corresponding to each ingredient scheme, select several ingredient schemes with relatively low comprehensive loss values ​​in this iteration. After calculating the comprehensive loss value of each ingredient scheme, they can be sorted in ascending order, and then the ingredient schemes with relatively low comprehensive loss values ​​can be selected. When selecting, ingredient schemes with comprehensive loss values ​​less than a preset loss threshold can be selected; alternatively, a preset number of ingredient schemes with relatively low comprehensive loss values ​​can be selected.

[0086] Step S702: Update the mean vector in the adaptive evolutionary algorithm for the covariance matrix using the weighted average of several ingredient schemes with low overall loss values. After selecting several ingredient schemes, their overall loss values ​​can be normalized and used as weights, and then a weighted average can be performed to obtain an averaged ingredient scheme, which will be used as the mean vector for the next iteration.

[0087] Step S703: Based on the covariance matrix C before the update... k The updated covariance matrix is ​​calculated using the preset covariance matrix update step size c and the direction vector v of the search direction for excellent individuals. This step can be expressed by the formula:

[0088] C k+1 =(1-c)C k +c vv T ,

[0089] In the formula, C k+1 This is the updated covariance matrix, and the direction vector v of the search direction for excellent individuals is the updated covariance matrix C. k The evolutionary path.

[0090] Step S105: Determine whether the difference between the predicted and target values ​​of the heat balance parameters for each batching scheme exceeds the allowable error. If there is a batching scheme that does not exceed the allowable error, output that batching scheme and end the process. If all batching schemes exceed the allowable error, repeat the above steps based on the updated mean vector and covariance matrix. This step is mainly used to determine whether each batching scheme meets the heat balance condition. If no batching scheme meets the heat balance condition, the entire step needs to be repeated to continue searching for other batching schemes until a batching scheme that meets the heat balance condition is found.

[0091] Understandably, when multiple batching schemes exist that do not exceed the allowable error, the one with the smallest overall loss value can be selected as the optimal batching scheme for output. Furthermore, the current production task and its corresponding optimal batching scheme can be saved as historical cases for later use.

[0092] Understandably, steps S103 and S104 are mainly for updating the mean vector and covariance matrix in the covariance matrix adaptive evolution algorithm, so as to regenerate a batch of ingredient plans. In actual execution, the judgment in step S105 can be executed first, that is, to determine whether there is an ingredient plan that meets the requirements. If there is, the ingredient plan is directly output and the process ends, without having to execute steps S103 and S104 again.

[0093] In a specific embodiment of the present invention, the following steps are included before step S101: acquiring heterogeneous multi-source data from the actual production process of the flash furnace and preprocessing it. The heterogeneous multi-source data includes preset parameters and real-time acquired parameters; the preset parameters include target values ​​for thermal balance parameters, constraints for chemical balance parameters, and silo composition data; the real-time acquired parameters include the current silo level; the preprocessing includes one or more of the following: data interpolation, abnormal data filtering, normalization, and data time alignment.

[0094] Understandably, during the execution of steps S101 to S105, some preset parameters and real-time acquisition parameters are involved. In actual execution, preset parameters can be written in a configuration file, while real-time acquisition parameters need to be obtained from other systems. Therefore, in this embodiment, before executing step S101, heterogeneous multi-source data is acquired once and preprocessed in a centralized manner to facilitate subsequent processing.

[0095] The current material level data in the silos can be obtained from the DCS (Distributed Control System); the material composition data in each silo can be obtained from the LIMS (Laboratory Information Management System); the target values ​​of the thermal balance parameters and the constraints of the chemical balance parameters (including: matte grade, slag-iron-silicon ratio, immersion slag temperature, lead-zinc content, arsenic-antimony-bismuth-nickel content, magnesium-aluminum content, concentrate moisture content, flux rate, process oxygen, oxygen coefficient, and flue dust dosage, etc.) can be obtained from the SCADA (Supervisory Control and Data Acquisition) system. This is just an example; the actual acquisition of these data can be configured according to the specific site conditions.

[0096] Regarding data preprocessing, there are several scenarios: (1) Since some data needs to be collected in real time, the collection frequency may vary, which leads to the problem of missing data at a certain moment. In this case, linear interpolation can be used to fill in the missing data, or the nearest neighbor alignment strategy can be used to align the data in time. (2) During the data preprocessing process, there may be some abnormal data. The IQR quartile method can be used to filter and remove these abnormal data. (3) For the problem of inconsistent data units that may exist during the data preprocessing process, normalization can be used for data processing. In addition, other preprocessing methods can be used to process the collected data according to the requirements.

[0097] It should be noted that the steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they contain the same logical relationship, they are all within the scope of protection of this application. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, but without changing the core design of the algorithm and process, are also within the scope of protection of this patent.

[0098] This invention has many innovations: (1) It uses Gaussian distribution sampling and covariance matrix adaptive update mechanism to efficiently search for the optimal ratio that meets the thermal balance condition in the flash furnace batching space; (2) It introduces a penalty term related to the material inventory status in multi-objective optimization to prevent the optimization scheme from exceeding the actual supply capacity and ensure the feasibility of the scheme; (3) Based on cosine similarity matching historical cases, it extracts the "experience preference vector" as the basis for weight adjustment, and dynamically adjusts the constraint weight in the objective function according to the similarity to achieve flexible optimization guided by experience.

[0099] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for batching ingredients for a flash furnace, characterized in that, include: Multiple ingredient plans are generated using an adaptive evolutionary algorithm based on the covariance matrix. Each batching scheme and the preset material composition data are input into the flash furnace digital model for simulation calculation to obtain the predicted value of the process index corresponding to each batching scheme, including the heat balance parameter. Based on the predicted values ​​of the process indicators and the predefined overall objective function, calculate the comprehensive loss value corresponding to each batching scheme; Based on the comprehensive loss value corresponding to each ingredient scheme, update the mean vector and covariance matrix in the covariance matrix adaptive evolution algorithm; Determine whether the difference between the predicted and target values ​​of the heat balance parameters for each batching scheme exceeds the allowable error: If there is a batching scheme that does not exceed the allowable error, output the batching scheme and end; If all ingredient combinations exceed the allowable error, repeat the above steps based on the updated mean vector and covariance matrix.

2. The batching method for a flash furnace according to claim 1, characterized in that, Multiple ingredient formulations are generated using an adaptive evolutionary algorithm based on the covariance matrix, including: Construct a Gaussian distribution x ~ N(m) k C k ), where x is a single batching scheme, m k C k The mean vector and the covariance matrix are respectively used in the k-th iteration. Multiple ingredient schemes are generated by sampling from the Gaussian distribution in each iteration.

3. The batching method for a flash furnace according to claim 2, characterized in that, Before the step of generating multiple ingredient plans using the covariance matrix adaptive evolution algorithm, the following is also included: Calculate the similarity between the current production task of the flash furnace and the saved historical cases, and find the most similar historical case; The ingredient scheme of the most similar historical case is used as the initial mean vector in the covariance matrix adaptive evolution algorithm.

4. The batching method for a flash furnace according to claim 1, characterized in that, Based on the predicted values ​​of the process indicators and the predefined overall objective function, calculate the comprehensive loss value corresponding to each batching scheme, including: The target deviation loss is calculated based on the predicted and target values ​​of the thermal balance parameters and the first weighting coefficient. The constraint violation loss is calculated based on the predicted values ​​of the chemical equilibrium parameters in the process indicators, the constraints, and the second weighting coefficient. Based on the batching plan, initial silo level, minimum safe level, and optimization cycle, the feasibility penalty loss for the material level is calculated. The comprehensive loss value is obtained based on the target deviation loss, the constraint violation loss, and the material level feasibility penalty loss.

5. The batching method for a flash furnace according to claim 4, characterized in that, Based on the batching plan, initial silo level, minimum safe level, and optimization cycle, the feasibility penalty loss for the material level is calculated, including: Based on the ingredient ratio and the optimization cycle, the amount of each material is obtained; The overuse loss of the silo is obtained based on the usage of each material, the initial silo level, and the silo overuse penalty coefficient. The low-level usage loss is obtained based on the initial silo level, the minimum safe level, the indication function, and the low-level usage penalty coefficient. The material level feasibility penalty loss is obtained based on the overuse loss of the silo and the low material level usage loss.

6. The batching method for a flash furnace according to claim 4, characterized in that, Before the step of calculating the overall loss value corresponding to each ingredient formulation, the following steps are included: Calculate the similarity between the current production task of the flash furnace and the saved historical cases, and find the most similar historical case; The weight coefficients in the overall objective function are adjusted based on the similarity of the most similar historical cases.

7. The batching method for a flash furnace according to claim 6, characterized in that, Based on the similarity of the most similar historical cases, the weight coefficients in the overall objective function are adjusted, including: When the similarity of the most similar historical cases is greater than or equal to the first threshold, the weight coefficient in the overall objective function is reduced. When the similarity of the most similar historical cases is greater than or equal to the second threshold and less than the first threshold, the weight coefficients in the overall objective function remain unchanged. When the similarity of the most similar historical cases is less than the second threshold, the weight coefficient in the overall objective function is increased.

8. The batching method for a flash furnace according to claim 7, characterized in that, Based on the similarity of the most similar historical cases, the preset adjustment coefficient, and the standard value of the weight coefficient in the overall objective function, the weight coefficient in the overall objective function after reduction or increase is calculated.

9. The batching method for a flash furnace according to claim 1, characterized in that, Based on the comprehensive loss value corresponding to each ingredient scheme, the mean vector and covariance matrix in the covariance matrix adaptive evolution algorithm are updated, including: Based on the overall loss value corresponding to each ingredient scheme, select several ingredient schemes with relatively low overall loss values ​​in this iteration; The mean vector in the adaptive evolution algorithm of the covariance matrix is ​​updated by using the weighted average of several ingredient schemes with low overall loss values. The updated covariance matrix is ​​calculated based on the covariance matrix before the update, the preset covariance matrix update step size, and the direction vector of the search direction of excellent individuals.

10. The batching method for a flash furnace according to claim 1, characterized in that, Before the step of generating multiple ingredient plans using the covariance matrix adaptive evolution algorithm, the following is also included: Acquire heterogeneous multi-source data from the actual production process of the flash furnace and preprocess it; The heterogeneous multi-source data includes preset parameters and real-time acquisition parameters; the preset parameters include target values ​​for thermal balance parameters, constraints for chemical balance parameters, and silo composition data; the real-time acquisition parameters include the current silo level; the preprocessing includes one or more of the following: data interpolation, abnormal data filtering, normalization, and data time alignment.

Images