Real-time intelligent prediction method and system for cobalt ion concentration in zinc hydrometallurgy purification process

Abstract

The invention discloses a real-time intelligent prediction method and system for cobalt ion concentration in a zinc hydrometallurgy purification process, and the method comprises the steps: S1, collecting a cobalt ion concentration data set generated in the zinc hydrometallurgy purification process in a preset time period, and carrying out the preprocessing of the cobalt ion concentration data set, and obtaining an updated cobalt ion concentration data set; s2, decomposing the updated cobalt ion concentration data set by using an ITD decomposition algorithm, wherein a decomposed cobalt ion concentration sub-sequence comprises an inherent rotation component and a residual component; s3, performing complexity analysis on the decomposed cobalt ion concentration subsequences by adopting approximate entropy, and classifying the decomposed cobalt ion concentration subsequences based on an entropy calculation result; s4, predicting the classification results by using a machine learning algorithm to obtain prediction results; and S5, reconstructing the prediction result to obtain a cobalt ion concentration prediction result, and displaying the cobalt ion concentration prediction result.

Description

Technical Field

[0001] The present invention relates to the fields of metallurgical engineering, environmental engineering and chemical engineering, and in particular to a real-time intelligent prediction method and system for cobalt ion concentration in a hydrometallurgical zinc purification process. Background Art

[0002] The purification process is a crucial step in hydrometallurgy zinc refining, its primary purpose being to remove various impurity ions from the neutral supernatant to provide qualified new solution for electrolysis. Cobalt ions are one of the primary impurities in the neutral supernatant and are also among the most difficult to remove. Traditional methods for detecting cobalt ion concentration in the second purification stage are inaccurate, resulting in excessive zinc powder input or "burned plate" phenomena in the later stages, and offline detection has significant lag, leading to long response times. The present invention combines the advantages of three methods: data decomposition algorithm, fuzzy entropy, and machine learning algorithm to predict cobalt ion concentration data in real time. The prediction results are then fed back to a Win64 programmable controller, achieving precise control over the zinc powder input, effectively avoiding the lag caused by manual detection and the inaccuracy of traditional prediction models. This invention solves the problem of metal impurity ion contamination in the hydrometallurgical zinc refining process in the metallurgical industry, improves the purity of metallic zinc, recovers valuable metal elements, reduces the accumulation of metallurgical solid waste, and brings significant economic and environmental benefits to the country, contributing to the sustainable development of enterprises. Summary of the Invention

[0003] In order to solve the above problems, the present invention proposes a real-time intelligent prediction method and system for cobalt ion concentration in a hydrometallurgical zinc purification process, the method comprising the following steps:

[0004] Step S1, collecting a cobalt ion concentration dataset generated during a hydrozinc purification process within a preset time period, and preprocessing the cobalt ion concentration dataset to obtain an updated cobalt ion concentration dataset;

[0005] Step S2, using an ITD decomposition algorithm to decompose the updated cobalt ion concentration data set to obtain a cobalt ion concentration subsequence, wherein the decomposed cobalt ion concentration subsequence includes an intrinsic rotation component and a residual component;

[0006] Step S3, performing complexity analysis on the decomposed cobalt ion concentration subsequences using approximate entropy, and classifying the decomposed cobalt ion concentration subsequences based on entropy calculation results;

[0007] Step S4: Use a machine learning algorithm to predict the classification results respectively to obtain prediction results;

[0008] Step S5: reconstruct the prediction result to obtain a cobalt ion concentration prediction result and display it.

[0009] Optionally, in step S1, the preprocessing process specifically includes:

[0010] Interpolating missing values ​​of the cobalt ion concentration data using linear interpolation to obtain first processed data;

[0011] Performing outlier detection on the first processed data using a box plot method to obtain second processed data;

[0012] The second processed data was standardized using the Standard-Scaler method to obtain an updated cobalt ion concentration data set.

[0013] Optionally, in step S2, the process of decomposing the updated cobalt ion concentration dataset using an ITD decomposition algorithm to obtain a cobalt ion concentration subsequence specifically includes:

[0014] Calculate the baseline component using the cobalt ion concentration data set and the acquisition time corresponding to the data set;

[0015] calculating an intrinsic rotation component based on the baseline component;

[0016] A decomposed cobalt ion concentration subsequence is obtained based on the intrinsic rotation component and the residual component.

[0017] Optionally, the contents of calculating the baseline component specifically include:

[0018]

[0019] Among them, τ∈(τ k ,τ k+1 ), which is X k and L k The extreme value interval, λ is the gain control parameter of the inherent PRC component amplitude, and 0<λ<1; L is the baseline extraction factor; L t is the baseline component, X t is the cobalt ion concentration dataset, L k is the kth baseline component, L k+1 is the k+1th baseline component, L k+2 is the k+2th baseline component, τ k The upper limit of the extreme value interval.

[0020] Optionally, the intrinsic rotation component H is calculated based on the baseline component t The content specifically includes:

[0021] H t =(1-L)X t =X t -L t .

[0022] Optionally, in step S3, the process of performing complexity analysis on the decomposed cobalt ion concentration subsequence using approximate entropy includes:

[0023] S31, based on X t Generate a set of m-dimensional vectors X i :

[0024] S32, based on X t Two adjacent variables X in i and X j The ratio of the distance d to the total number of vectors is recorded as

[0025] S33. Increase the dimension to m+1, repeat S31-S32, and get the ratio of the distance of the dimension vector to the total number of vectors. and the entropy value Φ under the m-dimensional pattern m+1 (r);

[0026] S34, based on the and the Φ m+1 (r) Calculate approximate entropy;

[0027] S35. Divide the cobalt ion concentration subsequence into a complex sequence and a non-complex sequence based on the value of the approximate entropy.

[0028] Optional,

[0029]

[0030] Among them, r is the threshold, r>0, N-m+1 is X i and X j The distance d is greater than the total number of vectors.

[0031] The present invention provides a real-time intelligent prediction system for cobalt ion concentration in a hydrometallurgical zinc purification process, the system comprising:

[0032] an acquisition and preprocessing module, configured to acquire a cobalt ion concentration dataset generated during a hydrometallurgical zinc purification process within a preset time period, and preprocess the cobalt ion concentration dataset to obtain an updated cobalt ion concentration dataset;

[0033] a sequence decomposition module, configured to decompose the updated cobalt ion concentration dataset using an ITD decomposition algorithm to obtain a cobalt ion concentration subsequence, wherein the decomposed cobalt ion concentration subsequence includes an intrinsic rotation component and a residual component;

[0034] a complexity analysis module, configured to perform complexity analysis on the decomposed cobalt ion concentration subsequences using approximate entropy, and classify the decomposed cobalt ion concentration subsequences based on entropy calculation results;

[0035] The classification prediction module is used to use the machine learning algorithm to predict the classification results and obtain the prediction results;

[0036] The concentration prediction module is used to reconstruct the prediction result to obtain the cobalt ion concentration prediction result and display it.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] (1) The present invention addresses the difficulty in processing the nonlinear and non-stationary cobalt ion concentration dataset of the hydrometallurgical zinc purification process. The ITD decomposition algorithm is used to denoise the dataset, capture the dynamic characteristics of the dataset, and obtain a series of interpretable cobalt ion concentration subsequences.

[0039] (2) The present invention analyzes the decomposed cobalt ion concentration sequence using approximate entropy to obtain the ApEn value, which is classified into a complex cobalt ion concentration sequence and a non-complex cobalt ion concentration sequence according to the value, so as to facilitate subsequent accurate prediction;

[0040] (3) The present invention uses two machine learning algorithms, LSSVM and BiGRU, to predict the cobalt ion concentration subsequences respectively, making full use of the advantages of each algorithm in data prediction to obtain more accurate and efficient cobalt ion concentration prediction results;

[0041] (4) The present invention provides a concise and easier-to-operate method for detecting cobalt ion concentration, thereby simplifying the detection process, reducing the difficulty of operation, and enabling more operators to conveniently control;

[0042] (5) The present invention inputs the prediction results of cobalt ion concentration by the ITD-LSSVM-BiGRU hybrid prediction framework into the Win64 programmable controller, and the input amount of zinc powder can be set according to the needs or the experience of the technicians. During the wet zinc smelting process, the amount of zinc powder can be reduced by about 12 kg in each purification process. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 The cobalt ion concentration subsequence and corresponding spectrum of the cobalt ion concentration data decomposed by ITD in an embodiment of the present invention are shown in FIG. Figure 1 (a) is a schematic diagram of the ITD decomposition results of the cobalt ion concentration dataset. Figure 1 (b) is a schematic diagram of the spectrum results corresponding to the cobalt ion concentration subsequence;

[0045] Figure 2 Schematic diagram of the results of complexity analysis of the cobalt ion concentration subsequence after decomposition using approximate entropy according to an embodiment of the present invention;

[0046] Figure 3 This is a flow chart of the combined prediction method based on ITD-LSSVM-BiGRU according to an embodiment of the present invention;

[0047] Figure 4 This is a comparison chart of evaluation indicators for cobalt ion concentration prediction using various single models and combined prediction methods according to an embodiment of the present invention, wherein: Figure 4 (a) is a schematic diagram showing the comparison of the error evaluation index results of various prediction methods for cobalt ion concentration prediction. Figure 4 (b) is a schematic diagram showing the comparison of the fitting results of various prediction methods for cobalt ion concentration prediction;

[0048] Figure 5 A comparison chart of the actual values ​​and predicted values ​​of the prediction model proposed in the embodiment of the present invention and other prediction models;

[0049] Figure 6 This is a diagram of the method steps of a method for real-time intelligent prediction of cobalt ion concentration in a hydrometallurgical zinc purification process according to an embodiment of the present invention;

[0050] Figure 7 Schematic diagram of a combined prediction device for cobalt ion concentration in a zinc hydrometallurgy purification process according to an embodiment of the present invention;

[0051] Description of reference numerals:

[0052] 1. Purification tank for the first stage; 2. Purification tank for the second stage; 3. Win64 system for collecting cobalt ion concentration data sets; 4. Programmable controller; 5. Zinc powder silo; 6. Conveyor belt. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] Example 1

[0055] A real-time intelligent prediction method for cobalt ion concentration in the hydrometallurgical zinc purification process, such as Figure 6 As shown, the method includes:

[0056] Step S1: collecting a cobalt ion concentration dataset generated during the hydrozinc smelting purification process within a preset time period, and preprocessing the cobalt ion concentration dataset to obtain an updated cobalt ion concentration dataset.

[0057] Linear interpolation method was used to interpolate missing values ​​in the cobalt ion concentration dataset, and box plot method was used to detect and eliminate outliers in the cobalt ion concentration dataset, thus obtaining a complete cobalt ion concentration dataset.

[0058] The preprocessing process specifically includes: using linear interpolation to interpolate missing values ​​of the cobalt ion concentration dataset to obtain first processed data; using a box plot method to detect outliers on the first processed data to obtain second processed data; and using a Standard-Scaler method to standardize the second processed data to obtain an updated cobalt ion concentration dataset.

[0059] The box plot method is used to process the outliers in the cobalt ion concentration dataset obtained from the first purification process of hydrometallurgy zinc smelting:

[0060]

[0061] Where: Q l is located at the 25% position of the entire cobalt ion concentration data set; Q p is located at the 50% position of the entire cobalt ion concentration data set

[0062] The Standard-Scaler method is used to preprocess the original cobalt ion concentration dataset, as shown below:

[0063]

[0064] Where x i is the cobalt ion concentration data set, μ is the mean of all cobalt ion concentration data sets, σ ​​is the standard deviation of all cobalt ion concentration data sets, y i Input values ​​for the cobalt ion concentration dataset after normalization for standard deviation.

[0065] The method is based on the device, such as Figure 7 As shown, the device includes:

[0066] Purification tank 1, purification tank 2, programmable controller 4, zinc powder bin 5, and conveyor belt 6. Cobalt ion concentration data 3 is collected from purification tank 1 using a Win64 system. The cobalt ion concentration data 3 is preprocessed and input into the programmable controller. The result is placed in zinc powder bin 5 and then returned to purification tank 2 via conveyor belt 6.

[0067] Step S2: Decompose the updated cobalt ion concentration data set using an ITD decomposition algorithm to obtain a cobalt ion concentration subsequence, wherein the decomposed cobalt ion concentration subsequence includes an intrinsic rotation component and a residual component.

[0068] The process of decomposing the cobalt ion concentration subsequence using the ITD decomposition algorithm specifically includes:

[0069] Select the original cobalt ion concentration sample set x k The corresponding time t k , k is the number of extreme values. The extraction factor L of the data baseline is as follows:

[0070]

[0071] Among them, τ∈(τ k ,τ k+1 ), which is X k and L k The extreme value interval, λ is the gain control parameter of the inherent PRC component amplitude, and 0<λ<1; L is the baseline extraction factor; L t is the baseline component, X t is the cobalt ion concentration dataset, L k is the kth baseline component, L k+1 is the k+1th baseline component, L k+2 is the k+2th baseline component, τ k The upper limit of the extreme value interval.

[0072] calculating an intrinsic rotation component based on the baseline component;

[0073] Calculate the intrinsic rotation component H based on the baseline component t The content specifically includes:

[0074] H t =(1-L)X t =X t -L t .

[0075] A decomposed cobalt ion concentration subsequence is obtained based on the intrinsic rotation component and the residual component.

[0076] The calculated baseline component L t Assuming it is the original sampling data, repeat the above steps and decompose the sampling data multiple times until the baseline component L t When it becomes a monotonic signal, the decomposition stops and the operation process is shown as follows:

[0077]

[0078] Where: H is the rotation component extraction operator; is the PRC component of the k+1th layer; is the baseline component of the k+1th layer; is the monotonic trend component after decomposition. The ITD algorithm can capture the dynamic characteristics of the data set, such as changes in frequency and amplitude, for nonlinear and non-stationary cobalt ion concentration data sets that are difficult to handle with traditional methods.

[0079] In addition to using the ITD algorithm to decompose the cobalt ion concentration dataset, EMD, EEMD, CEEMD and ICEEMDA can also be used to decompose it to remove the noise of the nonlinear cobalt ion concentration dataset and improve the data quality.

[0080] Step S3: performing complexity analysis on the decomposed cobalt ion concentration subsequences using approximate entropy, and classifying the decomposed cobalt ion concentration subsequences based on entropy calculation results.

[0081] The process of performing complexity analysis on the decomposed cobalt ion concentration subsequence using approximate entropy includes:

[0082] S31, based on X t Generate a set of m-dimensional vectors X i :

[0083] X i =[x i , x i+1 ,...,x i+m-1 ]

[0084] Where m is the pattern dimension; I = 1, 2, ..., n-M+1.

[0085] S32, based on X t Two adjacent variables X in i and X j The ratio of the distance d to the total number of vectors is recorded as

[0086] S33. Increase the dimension to m+1, repeat S31-S32, and get the ratio of the distance of the dimension vector to the total number of vectors. and the entropy value Φ under the m-dimensional pattern m+1 (r);

[0087] X i and X j The distance d is greater than the total number of vectors N-m+1, and we get Find its average value for all i, denoted as Φ m+1 (r), then

[0088] in,

[0089]

[0090] Among them, r is the threshold, r>0, N-m+1 is X i and X j The distance d is greater than the total number of vectors.

[0091] S34, based on the and the Φ m+1 (r) Calculate approximate entropy;

[0092] In actual engineering applications, N is a finite value. In this case, the sequence approximate entropy can be calculated as follows:

[0093] ApEn(m, r, N) = Φ m (r)-Φ m+1 (r)

[0094] When the value of the subsequence ApEn is larger, it indicates that the cobalt ion concentration subsequence is a complex sequence; otherwise, it indicates that the cobalt ion concentration subsequence is a non-complex sequence.

[0095] S35. Divide the cobalt ion concentration subsequence into a complex sequence and a non-complex sequence based on the value of the approximate entropy.

[0096] Step S4: Use a machine learning algorithm to predict the classification results and obtain prediction results. LSSVM is used to predict complex sequences. LSSVM algorithm takes advantage of its ability to process complex sequences to obtain prediction results for the cobalt ion concentration of the complex sequence.

[0097] The principle and process of the machine learning algorithm LSSVM are as follows:

[0098] For nonlinear complex data sets, LSSVM can map the nonlinear problem in the original input space to a high-dimensional feature space, making it a linearly separable problem in the high-dimensional space. The training sample set is defined as: S = {(X k ,Y k ), k=1,2,…,N}, where X k is the input sample, Y k To output the sample, the optimized objective function is

[0099]

[0100] Where: k is the error variable; w * is the weight vector; b k is the deviation; C is the adjustable regularization parameter; x is a linearly differentiable nonlinear high-dimensional mapping.

[0101] The linear constraints are obtained based on the KKT condition and Mercer's theorem, and are solved using least squares. The final LSSVM prediction model can be expressed as:

[0102]

[0103] Where: z is the support vector machine; σ 2 is the kernel function parameter; kernel function K(zk , z i ) is the radial basis kernel function RBF; α k is the Lagrange factor; b is a constant.

[0104] Compared to traditional SVM methods, LSSVM significantly improves computational efficiency when processing nonlinear and complex data sets by solving systems of linear equations. LSSVM, through clever algorithmic design, can deeply explore the patterns hidden in the data, accurately grasp the nonlinear relationships within the data, and effectively fit the existing data during model training, resulting in relatively accurate predictions of cobalt ion concentrations.

[0105] Next, for the nonlinear and non-complex cobalt ion concentration subsequence, the BiGRU method combines its own forward and reverse GRU network to process the data sequence from the beginning to the end one by one to extract the hidden information of the data sequence. The algorithm steps are as follows:

[0106]

[0107] Where: A t is the weight of forward propagation; B t is the weight of back propagation; c t is the bias vector corresponding to the hidden layer state, h t is the hidden layer state; is the forward hidden layer input, is the reverse hidden layer output, x t is the cobalt ion concentration subsequence.

[0108] The BiGRU algorithm is used to predict the nonlinear and non-complex cobalt ion concentration subsequence after decomposition, which can effectively integrate the inherent regular characteristics of the data, have high interpretability of the characteristics, and facilitate improving the expression ability of the model and the prediction results.

[0109] In order to test the performance of the ITD-LSSVM-BiGRU combined prediction model, the root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage (MAPE), and coefficient of determination (R 2 ) to evaluate the model prediction results. The smaller the RMSE, MAE and MAPE values ​​are, the better the prediction effect is. 2 The value range is [0,1], R 2 The closer it is to 1, the higher the prediction accuracy. The specific calculation expression is as follows:

[0110]

[0111] Where: v i is the actual value of the i-th cobalt ion concentration data set, is the predicted value of the i-th cobalt ion concentration data set, is the average value of the cobalt ion concentration data set, and N represents the total number of cobalt ion concentration data sets.

[0112] Step S5: Reconstruct the predicted results to obtain and display predicted cobalt ion concentration results. Import the predicted cobalt ion concentration results into a Win64 programmable controller to achieve online, real-time control of zinc powder dosage. Combining artificial intelligence with the field of metallurgical engineering, intelligent methods are used to improve production efficiency and achieve precise control and optimization of the production process.

[0113] The real-time intelligent prediction method adopted is based on the ITD-LSSVM-BiGRU combined prediction framework. The combined prediction framework integrates the advantages of multiple models to obtain more accurate prediction results, increase robustness and improve interpretability.

[0114] All of the above-mentioned real-time intelligent prediction methods for cobalt ion concentration in the hydrometallurgical zinc purification process can be integrated into a Win64 system programmable controller to perform real-time control of the zinc powder input amount.

[0115] The present invention can not only predict the cobalt ion concentration in the hydrometallurgical zinc smelting purification process, but can also be applied to the prediction of other metal ions in pyrometallurgical zinc smelting and electrolysis processes.

[0116] Example 2

[0117] The acquisition and preprocessing module is used to acquire a cobalt ion concentration dataset generated during the hydrometallurgical zinc smelting purification process within a preset time period, and preprocess the cobalt ion concentration dataset to obtain a cobalt ion concentration subsequence.

[0118] Linear interpolation method was used to interpolate missing values ​​in the cobalt ion concentration dataset, and box plot method was used to detect and eliminate outliers in the cobalt ion concentration dataset, thus obtaining a complete cobalt ion concentration dataset.

[0119] The preprocessing process specifically includes: using linear interpolation to interpolate missing values ​​of the cobalt ion concentration dataset to obtain first processed data; using a box plot method to detect outliers on the first processed data to obtain second processed data; and using a Standard-Scaler method to standardize the second processed data to obtain an updated cobalt ion concentration dataset.

[0120] The Standard-Scaler method is used to preprocess the original cobalt ion concentration dataset, as shown below:

[0121]

[0122] Where, X tis the cobalt ion concentration data set, μ is the mean of all cobalt ion concentration data sets, σ ​​is the standard deviation of all cobalt ion concentration data sets, Y t Input values ​​for the cobalt ion concentration dataset after normalization for standard deviation.

[0123] The sequence decomposition module is used to decompose the updated cobalt ion concentration sub-dataset using an ITD decomposition algorithm to obtain a cobalt ion concentration sub-sequence, wherein the decomposed cobalt ion concentration sub-sequence includes an intrinsic rotation component and a residual component.

[0124] The process of decomposing the cobalt ion concentration subsequence using the ITD decomposition algorithm specifically includes:

[0125] Select the original cobalt ion concentration sample set x k The corresponding time t k , k is the number of extreme values. The extraction factor L of the data baseline is as follows:

[0126]

[0127] Among them, τ∈(τ k ,τ k+1 ), which is X k and L k The extreme value interval, λ is the gain control parameter of the inherent PRC component amplitude, and 0<λ<1; L is the baseline extraction factor; L t is the baseline component, X t is the cobalt ion concentration dataset, L k is the kth baseline component, L k+1 is the k+1th baseline component, L k+2 is the k+2th baseline component, τ k The upper limit of the extreme value interval.

[0128] calculating an intrinsic rotation component based on the baseline component;

[0129] Calculate the intrinsic rotation component H based on the baseline component t The content specifically includes:

[0130] H t =(1-L)X t =X t -L t .

[0131] A decomposed cobalt ion concentration subsequence is obtained based on the intrinsic rotation component and the residual component.

[0132] The calculated baseline component L t Assuming it is the original sampling data, repeat the above steps and decompose the sampling data multiple times until the baseline component L tWhen it becomes a monotonic signal, the decomposition stops and the operation process is shown as follows:

[0133]

[0134] Where: H is the rotation component extraction operator; is the PRC component of the k+1th layer; is the baseline component of the k+1th layer; is the monotonic trend component after decomposition.

[0135] a complexity analysis module, configured to perform complexity analysis on the decomposed cobalt ion concentration subsequences using approximate entropy, and classify the decomposed cobalt ion concentration subsequences based on entropy calculation results;

[0136] The process of performing complexity analysis on the decomposed cobalt ion concentration subsequence using approximate entropy includes:

[0137] Based on X t Generate a set of m-dimensional vectors X i :

[0138] X i =[x i , x i+1 ,...,x i+m-1 ]

[0139] Where m is the pattern dimension; i = 1, 2, ..., N-m+1.

[0140] Based on X t Two adjacent variables X in i and X j The ratio of the distance d to the total number of vectors is recorded as

[0141] Increase the dimension to m+1, repeat S31-S32, and get the ratio of the distance of the dimension vector to the total number of vectors and the entropy value Φ under the m-dimensional pattern m+1 (r);

[0142] X i and X j The distance d is greater than the total number of vectors N-m+1, and we get Find its average value for all i, denoted as Φ m+1 (r), then

[0143] in,

[0144]

[0145] Among them, r is the threshold, r>0, N-m+1 is X i and Xj The distance d is greater than the total number of vectors.

[0146] Based on the and the Φ m+1 (r) Calculate approximate entropy;

[0147] In actual engineering applications, N is a finite value. In this case, the sequence approximate entropy can be calculated as follows:

[0148] ApEn(m, r, N) = Φ m (r)-Φ m+1 (r)

[0149] When the value of the subsequence ApEn is larger, it indicates that the cobalt ion concentration subsequence is a complex sequence; otherwise, it indicates that the cobalt ion concentration subsequence is a non-complex sequence.

[0150] The cobalt ion concentration subsequences are divided into complex sequences and non-complex sequences based on the value of the approximate entropy.

[0151] The classification prediction module is used to use machine learning algorithms to predict the classification results separately and obtain prediction results.

[0152] like Figure 3 As shown, the principle and process of the machine learning algorithm LSSVM are as follows:

[0153] For nonlinear complex data sets, LSSVM can map the nonlinear problem in the original input space to a high-dimensional feature space, making it a linearly separable problem in the high-dimensional space. The training sample set is defined as: S = {(X k ,Y k ), k=1,2,…,N}, where X k is the input sample, Y k To output the sample, the optimized objective function is

[0154]

[0155] Where: k is the error variable; w * is the weight vector; b k is the deviation; C is the adjustable regularization parameter.

[0156] Then introduce the Lagrangian factor α k , linear constraints are obtained according to KKT conditions and Mercer theorem, and solved by least squares. The final constructed LSSVM prediction model can be expressed as:

[0157]

[0158] Where: z is the support vector machine; σ2 is the kernel function parameter; kernel function K(z k , z i ) is the radial basis kernel function RBF.

[0159] Compared to traditional SVM methods, LSSVM significantly improves computational efficiency when processing nonlinear and complex data sets by solving systems of linear equations. LSSVM, through clever algorithmic design, can deeply explore the patterns hidden in the data, accurately grasp the nonlinear relationships within the data, and effectively fit the existing data during model training, resulting in relatively accurate predictions of cobalt ion concentrations.

[0160] Next, for the nonlinear and non-complex cobalt ion concentration subsequence, the BiGRU method combines its own forward and reverse GRU network to process the data sequence from the beginning to the end one by one to extract the hidden information of the data sequence. The algorithm steps are as follows:

[0161]

[0162] Where: A t is the weight of forward propagation; B t is the weight of back propagation; c t is the bias vector corresponding to the hidden layer state.

[0163] The BiGRU algorithm is used to predict the nonlinear and non-complex cobalt ion concentration subsequence after decomposition, which can effectively integrate the inherent regular characteristics of the data, have high interpretability of the characteristics, and facilitate improving the expression ability of the model and the prediction results.

[0164] In order to test the performance of the ITD-LSSVM-BiGRU combined prediction model, the root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage (MAPE), and coefficient of determination (R 2 ) to evaluate the model prediction results. The smaller the RMSE, MAE and MAPE values ​​are, the better the prediction effect is. 2 The value range is [0,1], R 2 The closer it is to 1, the higher the prediction accuracy. The specific calculation expression is as follows:

[0165]

[0166] Where: v i is the actual value of the i-th cobalt ion concentration data set, is the predicted value of the i-th cobalt ion concentration data set, is the average value of the cobalt ion concentration data set, and N represents the total number of cobalt ion concentration data sets.

[0167] The concentration prediction module is used to reconstruct the prediction result to obtain the cobalt ion concentration prediction result and display it.

[0168] Example 3

[0169] The present invention collects metal ion concentration data of the first stage of hydrometallurgical zinc purification from a factory in my country. First, a feasibility analysis is performed on the problems of missing, incomplete, and non-uniform data formats in the collected original metal ion concentration data. The missing values ​​and outliers in the data set are preprocessed by linear interpolation and box detection methods to obtain a complete metal ion concentration data set. The data set is then standardized using the Standard-Scaler method. Then, descriptive statistical analysis and preprocessing are performed on the original metal ion concentration data set, and it is found that the data exhibits nonlinear and non-stationary characteristics. The original metal ion concentration data set is decomposed using the ITD decomposition algorithm to decompose the nonlinear and non-stationary original cobalt ion concentration data set into several gradually stable PRC components. The PRC components, i.e., cobalt ion concentration subsequences, obtained by the decomposition algorithm are subjected to similarity analysis using the approximate entropy method to better explore the characteristics of the subsequences. The cobalt ion concentration subsequences with ApEn values ​​exceeding 0.5 are classified as complex subsequences, and vice versa. Secondly, the LSSVM machine learning algorithm is used to predict the complex cobalt ion concentration subsequence, and the BiGRU machine learning algorithm is used to predict the non-complex cobalt ion concentration sequence; finally, the results predicted by the two machine learning algorithms are reconstructed to obtain the final cobalt ion concentration prediction result; in order to verify whether the combined prediction method proposed in this invention meets the application standard, the root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAP E) and determination coefficient (R 2 ) four evaluation indicators for performance analysis. The smaller the values ​​of RMSE, MAE and MAPE, the higher the prediction results; 2 The value range is [0,1], R 2 The closer it is to 1, the higher the prediction accuracy.

[0170] Example 4

[0171] The present invention collects a cobalt ion concentration dataset of the first stage of hydrometallurgical zinc purification from a factory in my country, preprocesses the cobalt ion concentration dataset using the median interpolation method and the Standard-Scaler method, and obtains 240 sets of complete cobalt ion concentration datasets; Figure 1 The ITD algorithm is used to decompose the original metal ion concentration data set, and the nonlinear and non-stationary original cobalt ion concentration data set is decomposed into five PRC components and one residual component R; Figure 2The six cobalt ion concentration subsequences after decomposition are analyzed for complexity by the approximate entropy method. The ApEn values ​​can be used to determine that if the ApEn values ​​of PRC1 to PRC3 are greater than 0.5, they are classified as complex cobalt ion concentration subsequences, and the rest are non-complex cobalt ion concentration subsequences.

[0172] Example 5

[0173] like Figures 4-5 As shown in the figure, it is a visualization graph obtained by comparing the prediction results of the combined prediction method proposed in the present invention with those of other prediction models. The RMSE of the ITD-LSSVM-BiGRU combined prediction model proposed in the present invention is reduced by 23.9%, MAE is reduced by 14.6%, and MAPE is reduced by 29.7% compared with the other 7 prediction models. 2 This is 3.6% higher than the other seven prediction models. This shows that the error between the true value and the predicted value of the combined model proposed in this invention is relatively small, and the fit between the predicted value and the actual value is very high. The predicted results of cobalt ion concentration are integrated into a Win64 programmable controller. Based on the input value set by the skilled workers' experience, the zinc powder input amount is controlled in real time, thereby controlling the cobalt ion concentration at the outlet of the second purification stage of hydrometallurgy.

[0174] Example 6

[0175] The ITD-LSSVM-BiGRU combined prediction method proposed in this invention is applied to the field of blast furnace smelting. Since the temperature, slag composition, smelting time and other factors in the iron smelting process also affect the silicon content, which directly leads to the non-stationary silicon content of molten iron, the combined prediction method is used to predict the historical data of the silicon content of molten iron in a smelter. Compared with the prediction results of other prediction models such as SVM and GRU, the RMSE is reduced by 54.3% compared with other models, the MAE is reduced by 21.0% compared with other models, and the MAPE is reduced by 19.5% compared with other models. 2 The result is 62% higher than other models. The error between the actual value and the predicted value of silicon content in molten iron is small, indicating that the combined prediction model proposed by the present invention is also more accurate in predicting silicon content in molten iron, with better effect and high stability.

[0176] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A real-time intelligent prediction method for cobalt ion concentration in a hydrometallurgical zinc purification process, characterized in that: The method comprises: Step S1, collecting a cobalt ion concentration dataset generated during a hydrozinc purification process within a preset time period, and preprocessing the cobalt ion concentration dataset to obtain an updated cobalt ion concentration dataset; Step S2, using an ITD decomposition algorithm to decompose the updated cobalt ion concentration data set to obtain a cobalt ion concentration subsequence, wherein the decomposed cobalt ion concentration subsequence includes an intrinsic rotation component and a residual component; Step S3, performing complexity analysis on the decomposed cobalt ion concentration subsequences using approximate entropy, and classifying the decomposed cobalt ion concentration subsequences based on entropy calculation results; Step S4: Use a machine learning algorithm to predict the classification results respectively to obtain prediction results; Step S5: reconstruct the prediction result to obtain a cobalt ion concentration prediction result and display it.

2. The real-time intelligent prediction method for cobalt ion concentration in the hydrometallurgical zinc purification process according to claim 1, characterized in that: In step S1, the pre-processing process specifically includes: Interpolating missing values ​​of the cobalt ion concentration data using linear interpolation to obtain first processed data; Performing outlier detection on the first processed data using a box plot method to obtain second processed data; The second processed data was standardized using the Standard-Scaler method to obtain an updated cobalt ion concentration data set.

3. The real-time intelligent prediction method for cobalt ion concentration in the hydrometallurgical zinc purification process according to claim 1, characterized in that: In step S2, the process of decomposing the updated cobalt ion concentration dataset using the ITD decomposition algorithm to obtain a cobalt ion concentration subsequence specifically includes: Calculate the baseline component using the cobalt ion concentration data set and the acquisition time corresponding to the data set; calculating an intrinsic rotation component based on the baseline component; A decomposed cobalt ion concentration subsequence is obtained based on the intrinsic rotation component and the residual component.

4. The method for real-time intelligent prediction of cobalt ion concentration in a zinc hydrometallurgy purification process according to claim 3, characterized in that: The contents of calculating the baseline component specifically include: Among them, τ∈(τ k ,τ k+1 ), which is X k and L k The extreme value interval, λ is the gain control parameter of the inherent PRC component amplitude, and 0<λ<1; L is the baseline extraction factor; L t is the baseline component, X t is the cobalt ion concentration dataset, L k is the kth baseline component, L k+1 is the k+1th baseline component, L k+2 is the k+2th baseline component, τ k The upper limit of the extreme value interval.

5. The real-time intelligent prediction method for cobalt ion concentration in the hydrometallurgical zinc purification process according to claim 4, characterized in that: Calculate the intrinsic rotation component H based on the baseline component t The content specifically includes: H t =(1-L)X t =X t -L t 。 6. The method for real-time intelligent prediction of cobalt ion concentration in a zinc hydrometallurgy purification process according to claim 1, characterized in that: In step S3, the process of performing complexity analysis on the decomposed cobalt ion concentration subsequence using approximate entropy includes: S31, based on X t Generate a set of m-dimensional vectors X i : S32, based on X t Two adjacent variables X in i and X j The ratio of the distance d to the total number of vectors is recorded as S33. Increase the dimension to m+1, repeat S31-S32, and get the ratio of the distance of the dimension vector to the total number of vectors. and the entropy value Φ under the m-dimensional pattern m+1 (r); S34, based on the and the Φ m+1 (r) Calculate approximate entropy; S35. Divide the cobalt ion concentration subsequence into a complex sequence and a non-complex sequence based on the value of the approximate entropy.

7. The method for real-time intelligent prediction of cobalt ion concentration in a zinc hydrometallurgy purification process according to claim 6, characterized in that: Among them, r is the threshold, r>0, N-m+1 is X i and X j The distance d is greater than the total number of vectors.

8. A real-time intelligent prediction system for cobalt ion concentration in a hydrometallurgical zinc purification process, the system being used to implement the method according to any one of claims 1 to 7, characterized in that: The system includes: an acquisition and preprocessing module, configured to acquire a cobalt ion concentration dataset generated during a hydrometallurgical zinc purification process within a preset time period, and preprocess the cobalt ion concentration dataset to obtain an updated cobalt ion concentration dataset; a sequence decomposition module, configured to decompose the updated cobalt ion concentration dataset using an ITD decomposition algorithm to obtain a cobalt ion concentration subsequence, wherein the decomposed cobalt ion concentration subsequence includes an intrinsic rotation component and a residual component; a complexity analysis module, configured to perform complexity analysis on the decomposed cobalt ion concentration subsequences using approximate entropy, and classify the decomposed cobalt ion concentration subsequences based on entropy calculation results; The classification prediction module is used to use the machine learning algorithm to predict the classification results and obtain the prediction results; The concentration prediction module is used to reconstruct the prediction result to obtain the cobalt ion concentration prediction result and display it.

Images