Method for predicting compressive strength of anode carbon block

Abstract

The invention relates to the technical field of data prediction, in particular to an anode carbon block compressive strength prediction method, which comprises the steps of S1, acquiring original core dynamic feature data of an anode carbon block at each time stage, and constructing a multi-dimensional feature data set; s2, unifying the multi-dimensional features through time sequence reconstruction and outputting a first data set; dividing the first data set according to roasting and cooling process stages; s3, performing non-uniform sampling transformation processing on the multi-dimensional data of each process stage, re-sampling each processed stage, and splicing according to a time sequence to obtain a second data set; the second data set is divided into a prediction set and a training set; s4, inputting the training set into an anode carbon block compressive strength prediction model to carry out training and data fitting on the model; and S5, inputting the prediction set into the trained anode carbon block compressive strength prediction model to predict the physical correlation and result of the compressive strength of the anode carbon block, and outputting the compressive strength value of the anode carbon block, thereby realizing the quality control of the anode carbon block in the industry.

Description

Technical Field

[0001] The invention belongs to the technical field of data prediction, and in particular relates to a method for predicting the compressive strength of an anode carbon block. Background Art

[0002] In the electrolytic aluminum production process, anode carbon blocks are key consumables of electrolytic cells, and their performance has a significant impact on indicators such as current efficiency, electrolysis life, energy consumption and environmental emissions. Among them, compressive strength, as an important indicator to measure the structural integrity and mechanical properties of anode carbon blocks, is directly related to their stability in use under high temperature and high load environments. Therefore, accurate prediction and intelligent evaluation of the compressive strength of anode carbon blocks is of great significance for improving product consistency, reducing defect rates and optimizing production costs.

[0003] For the prediction of the compressive strength of anode carbon blocks, neural networks such as LSTM and BP are currently mainly used to build prediction models. Overfitting is prone to occur during the training process, especially when the training data is limited or the noise is large. This reduces the generalization ability of the model and cannot adapt well to changes in different batches, different production lines or different raw materials. Neural networks such as LSTM and BP mainly process data of a single mode. For the prediction of the compressive strength of anode carbon blocks, in addition to production parameters, the influence of multiple information such as environmental factors should also be considered; secondly, in addition to the use of materials, the factors that are difficult to control that determine the compressive strength of anode carbon blocks are mainly concentrated in the roasting and cooling process stages. There are currently few prediction methods for these two stages.

[0004] The Ant Nest Optimization Algorithm (ANA) constructs a mathematical optimization model by simulating the behavior of ants in finding the optimal placement when building nests. The optimization mathematical model only includes the search strategy and constructs a position change rate model by simulating three different methods. Updating the individual positions of the agents and using the ANA algorithm to optimize the parameters of the LSTM network of the anode carbon block compressive strength prediction model can improve the optimization speed of the LSTM network parameters. However, the search strategy lacks control over the optimization accuracy, which makes it easy to fall into the local optimum during the optimization process, affecting the parameter accuracy and resulting in unsatisfactory prediction results. Summary of the invention

[0005] The present invention proposes a method for predicting the compressive strength of an anode carbon block, which is suitable for the quality control and process optimization scenarios of anode carbon blocks in the electrolytic aluminum industry. The method of the present invention is an anode carbon block compressive strength prediction method that is oriented to the actual process flow, integrates multi-source process data and has time series modeling capabilities, so as to achieve higher prediction accuracy, better generalization ability and stronger real-time response characteristics, thereby providing an intelligent quality assessment method for electrolytic aluminum anode manufacturing, and improving the process control level and product stability.

[0006] The present invention proposes a method for predicting the compressive strength of an anode carbon block, which makes full use of the multi-dimensional characteristic parameters collected by the anode carbon block during the roasting and cooling process stages, optimizes the traditional LSTM prediction network through an improved ant nesting optimization algorithm (LANA algorithm), and constructs an anode carbon block compressive strength prediction model based on LANA-LSTM to improve the physical correlation and discrimination ability of compressive strength prediction. The specific steps are as follows: S1. Collect the original core dynamic feature data of anode carbon blocks at each time stage and construct a multidimensional feature data set; S2, unifying the above multi-dimensional features through time series reconstruction and outputting a first data set; dividing the first data set according to the roasting and cooling process stages; S3, performing non-uniform sampling transformation processing on the multidimensional data of each process stage, resampling each processed stage, and splicing them in time sequence to obtain a second data set; the second data set is divided into a prediction set and a training set; S4, the training set is input into the anode carbon block compressive strength prediction model to train the model and perform data fitting, the anode carbon block compressive strength prediction model is based on the enhanced LSTM prediction model, and the improved ant nesting optimization algorithm is used to optimize the LSTM unit number L and the learning rate factor α of the traditional LSTM prediction network; the improved ant nesting optimization algorithm includes: introducing a historical path memory mechanism and a dynamic memory guidance factor based on cognitive information increment, and constructing a nonlinear spiral search mechanism based on anti-motion angle and rotation disturbance to improve the position update strategy of the ant nesting optimization algorithm; S5. The prediction set is input into the trained anode carbon block compressive strength prediction model to predict the physical correlation and result of the anode carbon block compressive strength, and output the anode carbon block compressive strength value.

[0007] Furthermore, the strength of the anode carbon block is not only affected by a single process stage, but is the result of the coupled evolution of the entire process from forming to roasting to cooling. The original process data of the anode carbon block at each time stage is collected as input data, including the baking temperature, oxygen content, furnace pressure value in the baking stage, and the cooling temperature value and total cooling time in the cooling stage, as well as the volume density of the carbon block; the output data is the compressive strength of the anode carbon block; the process data (i.e., simple time training data) that is flattened by time lacks structure. The present invention arranges the multidimensional feature data from small to large in time to construct a first data set, identifies the process stage boundaries according to the time series, divides the prediction data set based on the time series into roasting and cooling process stage data sets, and then performs non-uniform sampling transformation on each stage according to its multidimensional features, resamples each processed stage, and splices them in sequence to obtain the final data set; specifically: S21, for the kth process stage, at each time point At , construct the physical disturbance intensity index. The higher the value of the physical disturbance intensity index, the more significant the impact of this time period on the evolution of the compressive strength of the anode carbon block. The mathematical model is: ; in, is the physical disturbance intensity index at the tth time point, is the weight at the tth time point, is the baking temperature at the tth time point, is the cooling temperature of the anode carbon block at the tth time point, is the furnace pressure value at the tth time point, is the oxygen content at the tth time point; S22, perform non-uniform sampling transformation, resample the data of each process stage, the sampling probability is proportional to the PD intensity, and construct a resampling probability sequence; ; in, is the resampling probability of each time point t in the kth stage, ranging from [0,1], and its denominator is the sum of PDs of all time points in the current stage; S23, for each stage, the resampling probability Time point greater than 0.5 The multidimensional data of the time point is retained, otherwise the time point is discarded The multidimensional data of Q time points are retained in each stage, then: ; in, For the second data set, is the Q group data set in the baking stage, is the Q group data set in the cooling stage; S24. Construct a second data set, combining Q groups of 3D data sets of the baking stage, Q groups of 3D data sets of the cooling stage and the output data into a second data set with 7-dimensional features, wherein the second data set includes Q groups of data, each group of data includes the input baking temperature, oxygen content, furnace pressure value, cooling temperature value and total cooling time in the cooling stage, and carbon block volume density; the output real anode carbon block compressive strength value data, a total of 7-dimensional data.

[0008] Furthermore, the physical disturbance intensity index is constructed using four most critical features that affect the compressive strength of the anode carbon block. The non-uniform sampling transformation processing and resampling processing convert the original process data of the anode carbon block at each time stage from the original equally spaced sampling form into a structure-enhanced expression format, in which the important disturbance time periods in each stage are weighted and retained, and the non-critical change areas are eliminated. The second data set in the present invention is a complete data set with high correlation and no missing data problems, which improves the information density of the input data and ensures the consistency of the input structure of the anode carbon block compressive strength prediction model, which facilitates the unified processing of the subsequent anode carbon block compressive strength prediction model.

[0009] Furthermore, the second data set is divided into a test set and a training set according to a certain ratio, and the proportion of the training set is higher than that of the test set; wherein, the training set is used for anode carbon block compressive strength prediction model training and parameter fitting, and the parameters include the number of LSTM units L and the learning rate factor α. The data in the training set covers the entire process of roasting and cooling process stages, ensuring the adaptability of the anode carbon block compressive strength prediction model in different process stages.

[0010] Furthermore, the test set is used to independently evaluate the generalization ability of the prediction model after the model training is completed. The test set and the training set do not overlap in the time period numbers, and representatively cover the process characteristics changes of the anode carbon blocks in the forming, roasting and cooling stages. During the test, the multi-dimensional process data sequence in the test set is fed as input into the LSTM compressive strength prediction model based on LANA optimization. The predicted value output by the model is compared with the actual compressive strength value in the test set, and then the prediction error index is calculated. When the error reaches the target value, the final anode carbon block compressive strength prediction value is output.

[0011] Furthermore, in the search strategy of the standard ant nesting optimization algorithm, the random movement of the ant population is simulated. However, ants in nature do not just move randomly, but tend to walk along the routes they have walked before and the locations they have successfully visited. Based on this, the standard ant nesting optimization algorithm is improved, and a new mathematical model for updating the position of the agent individual (ant individual) is constructed. Specifically, a memory path set is introduced into each agent individual, and the memory path set is the best location visited by the agent individual in the last n times. ; in, is the memory path set of the i-th agent, is the first location visited by the i-th agent individual, is the location that the nth agent has visited; Secondly, define cognitive information increments, and construct driving memory guidance factors through cognitive information increments. The mathematical model is: ; in, To increase cognitive information, the cognitive metric between the memory set and the current agent individual position is calculated. The mathematical model is: ; is the memory difference value of the iter-th iteration, defined as the mean of the cognitive information increments of N agents in the previous iteration, and λ is the parameter that controls the response intensity; Finally, the driving memory guidance factor is introduced to guide the agent to explore the area around the solution with the smallest fitness value in the iteration process again, and to construct an improved mathematical model of the search strategy. The mathematical model of the improved search strategy is: ; in, is the updated position of the i-th agent, is the position of the i-th agent in the current iteration, It is the best position of the agent individual in the current iteration population.

[0012] Furthermore, through adaptive memory guidance and information increment mechanism, the LANA algorithm can search for the optimal parameter area more accurately, avoid over-exploration of invalid areas, and improve the global search efficiency. In the anode carbon block compressive strength prediction model, this means that the optimization process can find the appropriate LSTM unit number L and learning rate factor α configuration more quickly, reducing time and computing costs; at the same time, the memory path guidance mechanism helps the model make full use of historical training experience and avoids the overfitting problem of the model during training; finally, by dynamically adjusting the memory guidance factor, the information increment mechanism enhances the learning ability of the model, and the LANA algorithm can adaptively adjust its search strategy, which enables the model to adjust and adapt more flexibly when facing different types of data; for the anode carbon block compressive strength prediction model, this means that it can better cope with data fluctuations in different process stages and at different time points, and improve the overall prediction performance.

[0013] Furthermore, the search strategy of the standard ant nesting optimization algorithm is weighted by the ant colony. The decision is made based on the current and historical positions and the change in the fitness calculation tendency, so that the position of the proxy individual shrinks along a straight line toward the target; the present invention proposes a nonlinear spiral migration search strategy driven by track disturbance, which geometrically maps the search process of the standard ant nesting optimization algorithm into a set of nonlinear ant circle tracks, and each proxy individual will not simply shrink along a straight line toward the target, but will advance in a roundabout way by simulating the "ant colony forming a vortex circle around the queen ant", so that the proxy individual "spiral centripetal motion" around the global optimal position, thereby enhancing the search diversity and improving the global escape ability; wherein, in the ant nesting optimization algorithm, the position of the ant individual is recorded as the proxy individual position, and the position of the queen ant, i.e., the core position, is recorded as the global optimal position; the mathematical model is: ; in, is the updated position of the i-th agent, is the position of the i-th agent in the current iteration, is the dynamic scaling factor for the iterth iteration, is the disturbance angle of the ith agent at the iterth iteration, is the two-dimensional rotation matrix generated in the current search plane, used for the spiral perturbation direction, is the unit disturbance direction, and the mathematical model is: ; Among them, the disturbance angle is not purely random, but historical tendency and population synergy are introduced as disturbance weights. The mathematical model is: ; in, is the basic disturbance angle; is the fitness value of the position of the i-th agent in the current iteration, is the fitness value of the position of the i-th agent individual at the iter-1th iteration, rand is the phase offset unique to the individual, which is used to increase the heterogeneity of the solution and takes a random number between 0 and 1.

[0014] Furthermore, the multi-dimensional search strategy, dynamic weight factor adjustment and historical experience re-exploration mechanism introduced by the orbital perturbation-driven nonlinear spiral migration search strategy can accelerate the convergence process, shorten the model training time, and find the appropriate parameter settings more quickly in different search spaces, thereby improving the model training efficiency.

[0015] Furthermore, the improved ant nesting optimization algorithm is used to optimize the number of LSTM units L and the learning rate factor α of the traditional LSTM prediction network. Through the adaptive adjustment of the number of LSTM units L and the learning rate factor α, the LSTM prediction network can process the multi-dimensional feature data of the anode carbon block at each time stage more accurately. The improved ant nesting optimization algorithm is used to optimize the number of LSTM units L and the learning rate factor α. It is necessary to map the position of the proxy individual (ant individual) of the LANA algorithm with the number of LSTM units L and the learning rate factor α value. Each proxy individual position It is in the form of a two-dimensional space vector, where the two dimensions correspond to the number of LSTM units L and the learning rate factor α. The mathematical model is: ; Update the number of LSTM units L and the learning rate factor α by updating the position of the agent individual, and calculate the current prediction result through the fitness function. The fitness function is a feedback model. The improved ant nesting optimization algorithm updates the position of the agent individual according to the fitness function to guide the total number of iterations. At this time, the position of the agent individual corresponding to the minimum fitness value is resolved to be the optimal number of LSTM units L and the learning rate factor α.

[0016] Furthermore, the improved ant nesting optimization algorithm is used to optimize the number of LSTM units L and the learning rate factor α of the traditional LSTM prediction network. The specific steps are as follows: S41. Determine the maximum population size N, the maximum number of iterations Tmax, the problem dimension, and the position range [lb, ub] of the improved ant nesting optimization algorithm; define the position of each agent as ; S42, initializing the proxy individual position, using the parameters corresponding to each proxy individual position to train the LSTM model to predict the compressive strength of the anode carbon block, and calculating the fitness value corresponding to each current individual position through the fitness function, the smaller the fitness value, the better the proxy individual position; S43, record the n positions with the smallest fitness values ​​in the history of each agent individual to build a memory path set ; S44, define cognitive information increment, construct a driving memory guidance factor through cognitive information increment, guide the agent individual to explore the area around the solution with the minimum fitness value in the iteration process again through the driving memory guidance factor; update the agent individual position through the improved search strategy; S45. Update the position of individual agents using a nonlinear spiral migration search strategy driven by orbital perturbations; S46, calculate the fitness value corresponding to the updated proxy individual position through the fitness function, retain the individual position corresponding to the minimum fitness value of each proxy individual, and when the proxy individual finds a better position during this iteration, the best position of this individual is insert In the collection; S47. Whether the current number of iterations iter reaches the maximum number of iterations Tmax, if so, output the LSTM unit number L and learning rate factor α parameter combination of the current global optimal agent individual; otherwise, return to execute S43 to update the memory path set and continue to optimize the LSTM prediction network.

[0017] Furthermore, the LSTM network of the anode carbon block compressive strength prediction model is reconstructed by using a combination of LSTM unit number L and learning rate factor α parameters, and the training set part of the second data set is input into the LSTM network training model of the reconstructed anode carbon block compressive strength prediction model, the training set includes an input data part and an output data part, wherein the output data is the anode carbon block compressive strength value, and the error between the predicted anode carbon block compressive strength value output during the training process and the actual anode carbon block compressive strength value is evaluated; after the training is completed, the test set is input to test the prediction accuracy of the anode carbon block compressive strength prediction model.

[0018] Compared with the existing methods and technologies, the method proposed in the present invention has the following beneficial effects: the method of the present invention adaptively and collaboratively optimizes the structural parameters of the traditional LSTM network by introducing non-uniform sampling transformation processing and resampling data processing tests and improved ant nesting optimization algorithms, which not only significantly improves the accuracy of the prediction of the compressive strength of the anode carbon block, but also improves the robustness and generalization ability of the model when facing different process conditions and raw material fluctuations; compared with the existing LSTM method with fixed structure or empirical setting, the present invention can more fully mine the key timing information of the roasting and cooling stages, realize structured, high-density, and physically relevant input data expression, and combine nonlinear search strategies to improve the model optimization efficiency, showing stronger real-time, reliability and promotion value in industrial applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a schematic diagram of the process of predicting the compressive strength of anode carbon blocks; Figure 2 constructing a flow diagram for the second data set; Figure 3 Flowchart for optimizing the number of LSTM units L and learning rate factor α of the traditional LSTM prediction network using the improved ant nesting optimization algorithm; Figure 4 This is a comparison chart of fitness value changes during model training; Figure 5 This is a comparison chart of the changes in the optimization process of the number of LSTM units L; Figure 6 This is a comparison chart of the changes in the optimization process of the learning rate factor α; Figure 7 This is the training model result graph; Figure 8 This is a comparison chart of the predicted results of compressive strength of anode carbon blocks. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0021] Based on this, an embodiment of the present invention provides a method for predicting the compressive strength of an anode carbon block; Figure 1 The flow chart of the method for predicting the compressive strength of anode carbon blocks is as follows: Figure 1 As shown, the method for predicting the compressive strength of anode carbon blocks according to an embodiment of the present invention comprises the following steps: S1, collecting original core dynamic feature data of anode carbon blocks at each time stage, and constructing a multi-dimensional feature data set.

[0022] Specifically, the collected original process data of the anode carbon blocks at each time stage are used as input data, including the baking temperature, oxygen content, furnace pressure value in the baking stage, the cooling temperature value and total cooling time in the cooling stage, and the volume density of the carbon blocks; the output data is the compressive strength of the anode carbon blocks; the characteristic data of 100 time points in the baking stage and the cooling stage are collected to construct a multidimensional feature data set as the anode carbon block compressive strength prediction data set, which is organized into an .xlsx table.

[0023] Further, S2, unify the above multi-dimensional features through time series reconstruction and output a first data set; divide the first data set according to the roasting and cooling process stages.

[0024] Furthermore, S3, a non-uniform sampling transformation process is performed on the multidimensional data of each process stage, each processed stage is resampled, and the second data set is obtained by splicing in time sequence; the second data set is divided into a prediction set and a training set.

[0025] Specifically, the multidimensional feature data are arranged in ascending order of time to construct a first data set, the process stage boundary is identified according to the time sequence, the prediction data set based on the time sequence is divided into roasting and cooling process stage data sets, and then each stage is subjected to non-uniform sampling transformation according to its multidimensional features, each processed stage is resampled, and the final data set is obtained by splicing in sequence; in the embodiment, k is the stage, the maximum value of k is 2, and the time point t=1,2,...,100; the specific implementation steps are as follows Figure 2 Shown are: S21, for the kth process stage, at each time point At , construct the physical disturbance intensity index. The higher the value of the physical disturbance intensity index, the more significant the impact of this time period on the evolution of the compressive strength of the anode carbon block. The mathematical model is: ; in, is the physical disturbance intensity index at the tth time point, is the weight at the tth time point, is the baking temperature value at the tth time point, is the cooling temperature of the anode carbon block at the tth time point, is the furnace pressure value at the tth time point, is the oxygen content value at the tth time point; S22, perform non-uniform sampling transformation, resample the data of each process stage, the sampling probability is proportional to the PD intensity, and construct a resampling probability sequence; ; in, is the resampling probability of each time point t in the kth stage, ranging from [0,1], and its denominator is the sum of PDs of all time points in the current stage; S23, for each stage, the resampling probability Time point greater than 0.5 The multidimensional data of the time point is retained, otherwise the time point is discarded The multidimensional data of Q time points are retained in each stage, then: ; in, For the second data set, is the Q group data set in the baking stage, is the Q group data set in the cooling stage; S24. Construct a second data set, combine Q groups of 3D data sets of the baking stage, Q groups of 3D data sets of the cooling stage and the output data into a second data set with 7-dimensional features, wherein the second data set includes Q groups of data, each group of data includes the input baking temperature, oxygen content, furnace pressure value, cooling temperature value and total cooling time in the cooling stage, and carbon block volume density; the output real anode carbon block compressive strength value data, a total of 7-dimensional data; specifically, Q reads the number of groups of the anode carbon block compressive strength prediction data set, that is, Q=100.

[0026] Specifically, the data of the second data set is divided into a training set and a test set according to a ratio of 8 to 2, which are 80 groups and 20 groups respectively; the last column of the data set is set as the output value, that is, the compressive strength of the anode carbon block; the above implementation process is used as data analysis, and the Matlab code is as follows: res = xlsread('Anode carbon block compressive strength prediction data set.xlsx'); num_size = 0.8; #Ratio of training set to data set; num_train_s = round(num_size * num_samples); #Number of training set samples; outdim = 1; #The last column is output; num_samples = size(res, 1); #Number of samples; res =construct_sample_dataset(alpha_t, T, Pt, rho_t, gamma_t, phase_label); #Construct the second data set; f_ = size(res, 2) - outdim; #Calculate the input feature dimension.

[0027] Furthermore, the non-uniform sampling transformation process resamples each processed stage and constructs the function code of the second data set as follows: function final_dataset = construct_sample_dataset(alpha_t, T, Pt,rho_t, gamma_t, phase_label) #enter: #alpha_t is the weight (physical intensity weight at a time point); #T is the baking / cooling temperature; #Pt is the furnace pressure; #rho_t is the volume density; #gamma_t is the oxygen content; #phase_label is the phase label (e.g. 1=baking, 2=cooling); #Output: #final_dataset is a two-dimensional structure matrix; #Code execution; PD_t = alpha_t .* (T + Pt + rho_t + gamma_t); PD_total = accumarray(phase_label, PD_t); #Total PD of each phase; pi_t = PD_t . / PD_total(phase_label); #The weight of each time point; valid_idx = false(length(pi_t), 1); phases = unique(phase_label); for k = 1:length(phases) phase = phases(k); idx = (phase_label == phase); max_pi = max(pi_t(idx)); valid_idx(idx) = pi_t(idx)>= 0.5 * max_pi; end; #Filter the retained data points; T_q = T(valid_idx); Pt_q = Pt(valid_idx); rho_q = rho_t(valid_idx); gamma_q = gamma_t(valid_idx); PD_q = PD_t(valid_idx); pi_q = pi_t(valid_idx); phase_q = phase_label(valid_idx); #Splice to form the second data set, each line is a 7-dimensional vector; res = [T_q, Pt_q, rho_q, gamma_q, PD_q, pi_q, phase_q]; end.

[0028] Specifically, the mapminmax function is used to normalize the input data (P_train) and target data (T_train) of the training set, scaling them to the range of [0, 1], and then ps_input and ps_output save the normalized parameters, i.e., the minimum and maximum values; the same normalization method is used for the test set data (P_test and T_test), and the normalized data is converted from a column-first matrix to a row-first storage format through a loop, i.e., each group of samples is stored as a row separately; specifically: for the training set, each column (one sample) of p_train and t_train is converted to a row of vp_train and vt_train; for the test set, each column of p_test and t_test is converted to a row of vp_test and vt_test; the above implementation process is used as data processing, and the Matlab code is as follows: [p_train, ps_input] = mapminmax(P_train, 0, 1); p_test = mapminmax('apply', P_test, ps_input); [t_train, ps_output] = mapminmax(T_train, 0, 1); t_test = mapminmax('apply', T_test, ps_output); #Format conversion; for i = 1 : 80 vp_train{i, 1} = p_train(:, i); vt_train{i, 1} = t_train(:, i); end for i = 1 : 20 vp_test{i, 1} = p_test(:, i); vt_test{i, 1} = t_test(:, i); end.

[0029] Furthermore, S4, the training set is input into the anode carbon block compressive strength prediction model to train and fit the model with data. The anode carbon block compressive strength prediction model is based on the enhanced LSTM prediction model, and the improved ant nesting optimization algorithm is used to optimize the LSTM unit number L and the learning rate factor α of the traditional LSTM prediction network; the improved ant nesting optimization algorithm includes: introducing a historical path memory mechanism and a dynamic memory guidance factor based on cognitive information increment, and constructing a nonlinear spiral search mechanism based on anti-motion angle and rotational disturbance to improve the position update strategy of the ant nesting optimization algorithm.

[0030] Specifically, the standard ant nesting optimization algorithm is improved to construct a new mathematical model for updating the position of the proxy individual (ant individual). The specific implementation steps are: introducing a memory path set in each proxy individual, the memory path set is the best position visited by the proxy individual in the last n times; ; in, is the memory path set of the i-th agent, is the first location visited by the i-th agent individual, is the location that the nth agent has visited; Secondly, the cognitive information increment is defined, and the driving memory guidance factor is constructed through the cognitive information increment. The mathematical model of the driving memory guidance factor is: ; in, To increase cognitive information, the cognitive metric between the memory set and the current agent individual position is calculated. The mathematical model is: ; is the memory difference value of the iter-th iteration, defined as the mean of the cognitive information increments of N agent individuals in the previous iteration, λ is a parameter for controlling the response intensity, and is set to 0.5 in the implementation of the present invention; Finally, the driving memory guidance factor is introduced to guide the agent to explore the area around the solution with the smallest fitness value in the iteration process again, and to construct an improved mathematical model of the search strategy. The mathematical model of the improved search strategy is: ; in, is the updated position of the i-th agent, is the position of the i-th agent in the current iteration, It is the best position of the agent individual in the current iteration population.

[0031] Specifically, a nonlinear spiral migration search strategy driven by orbital perturbations is proposed. The search process of the standard ant nesting optimization algorithm is geometrically mapped into a set of nonlinear ant circle orbits. Each agent individual will not simply shrink toward the target along a straight line, but will advance in a roundabout way by simulating the "vortex circle formed by the ant colony around the queen ant". This makes the agent individual "spiral centripetal motion" around the global optimal position, thereby enhancing the search diversity and improving the global escape ability. In the ant nesting optimization algorithm, the position of the ant individual is recorded as the agent individual position, and the position of the queen ant, i.e., the core position, is recorded as the global optimal position. The implementation mathematical model is: ; in, is the updated position of the i-th agent, is the position of the i-th agent in the current iteration, is the dynamic scaling factor for the iterth iteration, according to Implementation, is the disturbance angle of the ith agent at the iterth iteration, is the two-dimensional rotation matrix generated in the current search plane, used for the spiral perturbation direction, is the unit disturbance direction, and the mathematical model is: ; Among them, the disturbance angle is not purely random, but historical tendency and population synergy are introduced as disturbance weights. The mathematical model is: ; in, is the basic disturbance angle, with an initial value of 1; is the fitness value of the position of the i-th agent in the current iteration, is the fitness value of the position of the i-th agent individual at the iter-1th iteration, rand is the phase offset unique to the individual, which is used to increase the heterogeneity of the solution and takes a random number between 0 and 1.

[0032] Furthermore, the above method is used to construct the optimization method of the present invention, which is recorded as an improved ant nesting optimization algorithm; an LSTM prediction model is constructed, including an input layer, an LSTM layer, a Relu activation layer and a regression layer initialization; the maximum number of training times is set to 1000 times, the gradient threshold is set to 1, the initial learning rate is set to 0 and the initial number of LSTM units is set to 0, the learning rate is adjusted after 850 training times, and the regularization parameter is set to 0.5.

[0033] Specifically, the prediction set LSTM prediction model is used to start training the anode carbon block compressive strength prediction model based on the LSTM prediction network. At the same time, the improved ant nesting optimization algorithm is used to optimize the LSTM unit number L and the learning rate factor α of the traditional LSTM prediction network. The maximum number of iterations is set to 60, the maximum population size is 20, the problem dimension is 2, the position range lb=[1 1e-6], ub=[40 2e-3], and the proxy individual (ant individual) position of the LANA algorithm is mapped to the LSTM unit number L and the learning rate factor α value. The LSTM unit number L and the learning rate factor α value are updated by updating the proxy individual position. The specific implementation steps are as follows: Figure 3 Shown are: S41. Determine the maximum population size N, the maximum number of iterations Tmax, the problem dimension, and the position range [lb, ub] of the improved ant nesting optimization algorithm; define the position of each agent as ; S42, initializing the proxy individual position, using the parameters corresponding to each proxy individual position to train the LSTM model to predict the compressive strength of the anode carbon block, and calculating the fitness value corresponding to each current individual position through the fitness function, the smaller the fitness value, the better the proxy individual position; S43, record the n positions with the smallest fitness values ​​in the history of each agent individual to build a memory path set ; S44, define cognitive information increment, construct a driving memory guidance factor through cognitive information increment, guide the agent individual to explore the area around the solution with the minimum fitness value in the iteration process again through the driving memory guidance factor; update the agent individual position through the improved search strategy; S45. Update the position of individual agents using a nonlinear spiral migration search strategy driven by orbital perturbations; S46, calculate the fitness value corresponding to the updated proxy individual position through the fitness function, retain the individual position corresponding to the minimum fitness value of each proxy individual, and when the proxy individual finds a better position during this iteration, the best position of this individual is insert In the collection; S47. Whether the current number of iterations iter reaches the maximum number of iterations Tmax, if so, output the LSTM unit number L and learning rate factor α parameter combination of the current global optimal agent individual; otherwise, return to execute S43 to update the memory path set and continue to optimize the LSTM prediction network.

[0034] Specifically, the LSTM network of the anode carbon block compressive strength prediction model is reconstructed using the optimal LSTM unit number L and learning rate factor α parameter combination obtained by optimization, and the training set part of the second data set is input into the LSTM network training model of the reconstructed anode carbon block compressive strength prediction model, the training set includes an input data part and an output data part, wherein the output data is the anode carbon block compressive strength value, and the root mean square error RMSE is used to evaluate the error between the predicted anode carbon block compressive strength value output during the training process and the actual anode carbon block compressive strength value; after the training is completed, the test set is input to test the prediction accuracy of the anode carbon block compressive strength prediction model.

[0035] Specifically, the root mean square error RMSE is used as the fitness function, and the mathematical model is: ;in, is the true value of the compressive strength of the anode carbon block of the i-th group of samples, is the predicted value of the compressive strength of the anode carbon block of the i-th group of samples.

[0036] Specifically, Figure 4 This is a comparison chart of the fitness value changes during the model training process. It can be seen from the figure that the method strategy of the present invention can obtain a lower fitness value within the 43rd iteration, which indicates that the LANA algorithm has a more efficient local optimal solution exploration capability, so that the anode carbon block compressive strength prediction model can find a near-optimal solution from the initialization state more quickly; while converging quickly, the method strategy of the present invention has achieved a more obvious optimization in fitness value. Compared with the traditional ANA algorithm strategy (dashed line), the method strategy of the present invention better reduces the deviation, thereby providing a more accurate optimization solution for the subsequent LSTM model.

[0037] Specifically, Figure 5 and Figure 6 This is a comparison chart of the changes in the number of LSTM units L and the learning rate factor α during the optimization process; Figure 4 and Figure 5 It can be seen that the optimization trends of the number of LSTM units L and the learning rate factor α are the same as the fitness value. The optimization strategy of the method of the present invention reaches stability at 43 iterations, and the fitness value reaches the minimum. The corresponding number of LSTM units L is 13 and the learning rate factor α is 0.002. The standard ANA algorithm strategy reaches stability at 31 iterations, and the fitness value is relatively large. The corresponding number of LSTM units L is 8 and the learning rate factor α is 0.00089.

[0038] Specifically, Figure 7This is the training model result graph. From the analysis of the training results in the graph, the LSTM model based on LANA optimization has more stable prediction results than the existing method; the predicted values ​​of the LANA-LSTM method are closer to the actual measured values ​​at most data points, showing higher accuracy; the predicted values ​​of the existing method fluctuate greatly, especially in some intervals, the predicted values ​​deviate far from the actual measured values; in contrast, the predicted values ​​of the LANA-LSTM method fluctuate less, reflecting better robustness and stability; this shows that the optimization of the LANA algorithm can effectively reduce the prediction error and improve the prediction stability of the model; by comparing the errors of different prediction methods, it can be found that the LANA-LSTM model can effectively reduce the error in most cases, especially at some extreme points, the error between the predicted value and the actual measured value is the smallest. In contrast, the predicted value error of the existing method is larger, indicating that the performance of the traditional LSTM model in predicting the compressive strength of anode carbon blocks is not as good as the prediction method proposed in the present invention.

[0039] Specifically, Figure 8 This is a comparison chart of the prediction results of the compressive strength of anode carbon blocks. The method of the present invention fits the true value curve more closely at most sample points, and the error is significantly smaller than that of the existing method. For example, in samples No. 1, 5, 10, 13, 17, and 20, the LANA-LSTM prediction results basically coincide with the actual values, indicating that the method has a high fitting ability; the existing method has severe deviations in predictions in many places, such as samples No. 6, 12, and 16, whose prediction results deviate greatly from the actual values, while the method of the present invention is stable at these points, and the prediction curve is smoother and less volatile, reflecting good robustness; This shows that the prediction model proposed in the present invention shows higher prediction accuracy and stability in the prediction verification of the compressive strength of anode carbon blocks, can more accurately reflect the trend of changes in real physical strength, and is suitable for applications in industrial scenarios that require high prediction reliability.

Claims

1. A method for predicting the compressive strength of an anode carbon block, characterized in that: include: S1. Collect the original core dynamic feature data of anode carbon blocks at each time stage and construct a multidimensional feature data set; S2, unifying the above multi-dimensional features through time series reconstruction and outputting a first data set; Divide the time series-based prediction data set according to the roasting and cooling process stages; S3, performing non-uniform sampling transformation processing on the multidimensional data of each process stage, resampling each processed stage, and splicing them in time sequence to obtain a second data set; the second data set is divided into a prediction set and a training set; S4, the training set is input into the anode carbon block compressive strength prediction model to train the model and perform data fitting, wherein the anode carbon block compressive strength prediction model is based on the enhanced LSTM prediction model, and the improved ant nesting optimization algorithm is used to optimize the LSTM unit number L and the learning rate factor α of the traditional LSTM prediction network; The improved ant nesting optimization algorithm includes: introducing a historical path memory mechanism and a dynamic memory guidance factor based on cognitive information increment, and constructing a nonlinear spiral search mechanism based on anti-dynamic angle and rotation disturbance to improve the position update strategy of the ant nesting optimization algorithm; S5. The prediction set is input into the trained anode carbon block compressive strength prediction model to predict the physical correlation and result of the anode carbon block compressive strength, and output the anode carbon block compressive strength value.

2. The method for predicting the compressive strength of an anode carbon block according to claim 1, characterized in that: The non-uniform sampling transformation of the first data set is processed and the second data set is obtained by resampling. The specific method is as follows: S21. For the kth process stage, at each time point At , construct the physical disturbance intensity index. The higher the value of the physical disturbance intensity index, the more significant the impact of this time period on the evolution of the compressive strength of the anode carbon block. The mathematical model is: ； in, is the physical disturbance intensity index at the tth time point, is the weight at the tth time point, is the baking temperature at the tth time point, is the cooling temperature of the anode carbon block at the tth time point, is the furnace pressure value at the tth time point, is the oxygen content at the tth time point; S22, perform non-uniform sampling transformation, resample the data of each process stage, the sampling probability is proportional to the PD intensity, and construct a resampling probability sequence; ； in, is the resampling probability of each time point t in the kth stage, ranging from [0,1], and its denominator is the sum of PDs of all time points in the current stage; S23, for each stage, the resampling probability Time point greater than 0.5 The multidimensional data of the time point is retained, otherwise the time point is discarded The multidimensional data of Q time points are retained in each stage, then: ； in, For the second data set, is the Q group data set in the baking stage, is the Q group data set in the cooling stage; S24. Construct a second data set, combining Q groups of 3D data sets of the baking stage, Q groups of 3D data sets of the cooling stage and the output data into a second data set with 7-dimensional features, wherein the second data set includes Q groups of data, each group of data includes the input baking temperature, oxygen content, furnace pressure value, cooling temperature value and total cooling time in the cooling stage, and carbon block volume density; the output real anode carbon block compressive strength value data, a total of 7-dimensional data.

3. The method for predicting the compressive strength of an anode carbon block according to claim 2, characterized in that: The historical path memory mechanism based on cognitive information increment and the dynamic memory guiding factor improve the ant nesting optimization algorithm. The specific method is as follows: introducing a memory path set in each agent individual, and the memory path set is the best location visited by the agent individual in the last n times; ； in, is the memory path set of the i-th agent, is the first location visited by the i-th agent individual, is the location that the nth agent has visited; Secondly, define cognitive information increments, and construct driving memory guidance factors through cognitive information increments. The mathematical model is: ； in, To increase cognitive information, the cognitive metric between the memory set and the current agent individual position is calculated. The mathematical model is: ; is the memory difference value of the iter-th iteration, defined as the mean of the cognitive information increments of N agent individuals in the previous iteration, and λ is the parameter that controls the response intensity; The driving memory guiding factor guides the agent to explore the area around the solution with the smallest fitness value in the iteration process again, and constructs an improved search strategy mathematical model. The improved search strategy mathematical model is: ； in, is the updated position of the i-th agent, is the position of the i-th agent in the current iteration, It is the best position of the agent individual in the current iteration population.

4. The method for predicting the compressive strength of an anode carbon block according to claim 3, characterized in that: The nonlinear spiral search mechanism improves the position update strategy of the ant nesting optimization algorithm, specifically: the search process of the standard ant nesting optimization algorithm is geometrically mapped into a set of nonlinear ant circle tracks, and the circuitous advancement is performed by simulating the "vortex circle formed by the ant colony around the queen ant", so that the proxy individual "spiral centripetal motion" around the global optimal position is enhanced to enhance the search diversity and improve the global escape ability; wherein, in the ant nesting optimization algorithm, the position of the ant individual is recorded as the proxy individual position, and the position of the queen ant, i.e., the core position, is recorded as the global optimal position; the mathematical model is: ； in, is the dynamic scaling factor for the iterth iteration, is the disturbance angle of the i-th agent in the iter-th iteration, is the two-dimensional rotation matrix generated in the current search plane, used for the spiral perturbation direction, is the unit disturbance direction, and the mathematical model is: ; Among them, the disturbance angle is not purely random, but historical tendency and population synergy are introduced as disturbance weights. The mathematical model is: ； in, is the basic disturbance angle; is the fitness value of the position of the i-th agent in the current iteration, is the fitness value of the position of the i-th agent individual at the iter-1th iteration, rand is the phase offset unique to the individual, which is used to increase the heterogeneity of the solution and takes a random number between 0 and 1.

5. A method for predicting the compressive strength of an anode carbon block according to any one of claims 1 to 4, characterized in that: The improved ant nesting optimization algorithm is used to optimize the number of LSTM units L and the learning rate factor α of the traditional LSTM prediction network. The specific steps are as follows: S41. Determine the maximum population size N, the maximum number of iterations Tmax, the problem dimension, and the position range [lb, ub] of the improved ant nesting optimization algorithm; define the position of each agent as ; S42, initializing the proxy individual position, using the parameters corresponding to each proxy individual position to train the LSTM model to predict the compressive strength of the anode carbon block, and calculating the fitness value corresponding to each current individual position through the fitness function, the smaller the fitness value, the better the proxy individual position; S43, record the n positions with the smallest fitness values ​​in the history of each agent individual to build a memory path set ; S44, define cognitive information increment, construct driving memory guidance factor through cognitive information increment, guide agent individual to explore the area around the solution with the minimum fitness value in the iteration process again through driving memory guidance factor; update agent individual position through improved search strategy, S45, using the nonlinear spiral migration search strategy driven by orbital perturbations to update the position of individual agents; S46, calculate the fitness value corresponding to the updated proxy individual position through the fitness function, retain the individual position corresponding to the minimum fitness value of each proxy individual, and when the proxy individual finds a better position during this iteration, the best position of this individual is insert In the collection; S47. Whether the current number of iterations iter reaches the maximum number of iterations Tmax, if so, output the parameter combination of the number of hidden layer neuron nodes L and the learning rate factor α of the current global optimal agent individual; otherwise, return to execute S43 to update the memory path set and continue to optimize the LSTM prediction network.

Images