Alumina concentration soft measurement method based on IGWO-DBN algorithm

Abstract

The present application relates to the technical field of alumina concentration measurement, and particularly relates to an alumina concentration soft measurement method based on an IGWO-DBN algorithm. The steps are as follows: S1: a soft measurement model is constructed according to a deep belief network to perform prediction; and S2: the three network parameters of the number of hidden layer nodes, the back-tuning learning rate and the back-iteration number of the deep belief network are optimized through an improved grey wolf algorithm. The alumina concentration soft measurement method based on the IGWO-DBN algorithm can effectively learn the relationship between the alumina concentration and the related input conditions, accurately predict the alumina concentration, and realize the effect of predicting the alumina concentration multiple times through once training of the algorithm model by using the recent electrolytic aluminum industrial production data, thereby effectively reducing the industrial production cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of alumina concentration measurement, and particularly relates to an alumina concentration soft measurement method based on an IGWO-DBN algorithm. BACKGROUND

[0002] The aluminum electrolysis production process is a typical complex industrial process, and the alumina concentration is one of the key parameters affecting the electrolysis efficiency and product quality in this process. It has been proven that when the alumina concentration is maintained within the optimal range of 1.5% to 2.5%, the current efficiency of aluminum electrolysis reaches the highest. However, once the alumina concentration exceeds this suitable range, its adverse effects will be significantly manifested. In order to obtain higher current efficiency, it is necessary to strictly control the alumina concentration in the cell to maintain it within a relatively narrow range, both to ensure that the concentration is not too low and to prevent the concentration from being too high to cause anode effect. Therefore, accurate prediction and control of the alumina concentration is an important prerequisite for achieving efficient, stable and energy-saving operation of the aluminum electrolysis production process. In addition, the alumina concentration also has an impact on the solubility of carbon dioxide in the electrolyte. Lower alumina content can reduce the solubility of carbon dioxide in the electrolyte, reduce the total area of anode bubbles, and thus reduce the secondary oxidation loss of aluminum liquid, which has significant significance for improving the quality of aluminum and reducing production costs.

[0003] In summary, the monitoring of the alumina concentration in the aluminum electrolysis production process is important because it helps to maintain the stability of the electrolytic cell, improve the current efficiency, reduce energy consumption, reduce production costs, and improve the quality of aluminum. However, due to the complex and uncertain characteristics of the aluminum electrolysis production process occurring in the high-temperature, corrosive and strong magnetic aluminum electrolytic cell. Traditional physical measurement methods often cannot meet the needs of real-time and online monitoring, therefore, the monitoring and control of the alumina concentration in the aluminum electrolysis production process is particularly critical, and it is essential to establish a soft measurement model of the alumina concentration.

[0004] Therefore, a kind of alumina concentration soft measurement method based on IGWO-DBN algorithm is designed to provide another technical solution to the above technical problems. SUMMARY

[0005] Therefore, it is necessary to provide an alumina concentration soft measurement method based on IGWO-DBN algorithm to solve the technical problems raised in the background art.

[0006] In order to solve the above technical problems, the present application adopts the following technical solutions:

[0007] An alumina concentration soft measurement method based on IGWO-DBN algorithm, the steps are as follows:

[0008] S1: constructing a soft measurement model according to a deep belief network to perform prediction;

[0009] S2: optimizing three network parameters of the deep belief network, including the number of hidden layer nodes, the learning rate of back propagation and the number of iterations of back propagation, by improving the grey wolf algorithm.

[0010] As a preferred embodiment of the IGWO-DBN algorithm-based soft measurement method for the concentration of aluminum oxide, the input data is standardized before the concentration of aluminum oxide is predicted, and the steps are as follows:

[0011] S1: performing preprocessing on the original data before constructing a soft measurement prediction algorithm model;

[0012] S2: performing EMD decomposition on the data of the first five time points of the concentration of aluminum oxide to obtain IMF s and a trend term;

[0013] S3: extracting features from each IMF of the first five time points, and combining the features and the features of the trend term into a feature vector for use as input of the prediction model;

[0014] S4: merging the feature vector and the original prediction input physical quantity to realize 8-dimensional input and 1-dimensional output of the concentration of aluminum oxide.

[0015] As a preferred embodiment of the IGWO-DBN algorithm-based soft measurement method for the concentration of aluminum oxide, the feature vector and the original prediction input physical quantity are merged, and the steps are as follows:

[0016] (1) combining the extracted related features F k into a feature vector f(t), and the feature vector f(t-1) of the previous time point is expressed as follows:

[0017] f(t-1) = |F1(t-1), F2(t-1), …, F k (t-1), F RES (t-1) |;

[0018] Wherein, t-1 represents the previous time point.

[0019] (2) merging the feature vectors of the first five time points and the original physical quantities of the prediction of the concentration of aluminum oxide, including two-level voltage V(t), anode lead current I(t) and pole distance R(t), to obtain a merged input vector X(t), and the expression is as follows:

[0020] X(t) = |V(t), I(t), R(t), f(t-5), f(t-4), f(t-3), f(t-2), f(t-1) |;

[0021] Wherein, f(t-5), f(t-4), f(t-3), f(t-2), f(t-1) are the feature vectors of the previous five time points;

[0022] (3) Using the merged input vector X(t) as input data to train the DBN model to predict the alumina concentration, expressed as follows:

[0023]

[0024] As a preferred embodiment of the alumina concentration soft measurement method based on the IGWO-DBN algorithm provided by the application, in the S2 step, the three network parameters of the number of hidden layer nodes of the deep belief network, the backpropagation learning rate and the backpropagation iteration number are optimized, and the steps are as follows:

[0025] A. In the initial stage, the number of hidden layer nodes H, the backpropagation learning rate a and the backpropagation iteration number β are randomly initialized;

[0026] B. In the motion stage, for each grey wolf individual ω, the experiences of neighbors and random individuals are learned, and the weighted average value of each dimension d is calculated;

[0027] C. In the selection and update stage, the current optimal solution X * is selected, and the fitness values of the new individual and the current optimal solution are compared.

[0028] As a preferred embodiment of the alumina concentration soft measurement method based on the IGWO-DBN algorithm provided by the application, in the A step, N grey wolf individuals are randomly distributed in the preset search space, and the expression is as follows:

[0029] X ij =l j +rand j [0,1]*(u j -l j ),i∈[1,N],j∈[1,D];

[0030] Wherein, l j , u j are the lower limit and upper limit of the search space of each grey wolf individual i in the jth dimension; rand j [0,1] is a random number in the interval [0,1], which is used for the random distribution of the grey wolf individual in the jth dimension;

[0031] In the B step, in the motion stage, for each grey wolf individual ω, the experiences of neighbors and random individuals are learned, and the weighted average value of each dimension d is calculated, and the expression is as follows:

[0032]

[0033] in, It is the neighbor n of individual ω j Summation, w n is a neighbor n j weight, w r R is the weight of individual r; random(P) is the set of individuals randomly selected from population P; It represents the neighbor n at the i-th iteration. j Position or state in the d-th dimension It represents the position or state of individual ω in the d-th dimension at the i-th iteration. It represents the position or state of a random individual r in the d-th dimension at the i-th iteration; It is the new position of individual ω in dimension d at the (i+1)th iteration; fitness(n j ), fitness(r), and fitness(ω) are the values ​​of the neighboring individuals n. j The fitness values ​​of reference individuals r and ω;

[0034] In step C, the selection and update phase selects the current optimal solution X. * The fitness values ​​of the new individual and the current optimal solution are compared, as shown in the following expression:

[0035] f(X ω,i+1 ) <f(X * );

[0036] If the fitness value of the new individual is less than the fitness value of the current optimal solution, then update the current optimal solution.

[0037] As a preferred embodiment of the soft measurement method for alumina concentration based on the IGWO-DBN algorithm provided by the present invention, a deep belief network is composed of several layers of restricted Boltzmann machines and one layer of BP. The deep belief network is trained by pre-training and fine-tuning.

[0038] As a preferred embodiment of the alumina concentration soft measurement method based on the IGWO-DBN algorithm provided by the present invention, the deep belief network is trained by pre-training and fine-tuning, and the steps are as follows:

[0039] (1) The alumina concentration prediction data is first input into the network structure by the input layer to generate a vector V;

[0040] (2) H is obtained by passing the weight value W to the hidden layer;

[0041] (3) Use the three network parameters optimized by the improved gray wolf algorithm—the number of hidden layer nodes, the reverse fine-tuning learning rate, and the number of reverse iterations—to train each layer of the RBM network separately in unsupervised training to ensure that the feature vectors are mapped to different feature spaces.

[0042] (4) The BP network receives the output feature vector of the RBM as its input feature vector, and then propagates the error information to each layer of RBM from top to bottom, updating the weights of each hidden layer and output layer, thereby realizing the fine-tuning of the DBN network.

[0043] (5) Obtain the weights and biases in the network for the final DBN model to output the predicted alumina concentration value based on the input data.

[0044] As a preferred embodiment of the alumina concentration soft measurement method based on the IGWO-DBN algorithm provided by the present invention, the performance of this algorithm model is described by calculating the coefficient of determination R2, root mean square error RMSE, squared absolute error MAE, and the goodness of fit of the predicted images on the validation set. The model structure or parameters are finely adjusted according to the validation results. The expression is as follows:

[0045]

[0046] Among them, y i This is the actual value. Here are the predicted values, Var is the variance, and R is the predicted value. 2 The larger the value, the better the model fits the data; the smaller the difference between the predicted and actual values, the smaller the MAE value, and the better the prediction performance; the RMSE value is used to measure the prediction accuracy of the prediction model on continuous data, and the smaller the value, the higher the prediction accuracy.

[0047] It is clear without a doubt that the technical solution described above in this application can solve the technical problem that this application aims to address.

[0048] Meanwhile, through the above technical solutions, the present invention has at least the following beneficial effects:

[0049] 1. The present invention provides a soft measurement method for alumina concentration based on the IGWO-DBN algorithm, which can effectively learn the relationship between alumina concentration and its related input conditions, achieve accurate prediction of alumina concentration, and achieve the effect of multiple predictions of alumina concentration by training the algorithm model with recent electrolytic aluminum industrial production data, thereby effectively reducing industrial production costs.

[0050] 2. This invention's improved Grey Wolf Algorithm (IGWO) based on a dimensional learning-based hunting search strategy optimizes key parameters in the Restricted Boltzmann Machine (RBM), such as the number of nodes in the hidden layer, the learning rate during backpropagation fine-tuning, and the number of iterations. This optimization strategy effectively solves the local optimum problem that may occur during parameter initialization in the traditional Grey Wolf Algorithm, thereby achieving a more comprehensive improvement in the performance of the DBN network and effectively enhancing the algorithm's prediction accuracy.

[0051] 3. The dynamic soft measurement algorithm of the present invention uses recently collected electrolytic aluminum industrial production data to train a highly accurate algorithm model, which has the ability to predict alumina concentration multiple times after a single training, significantly optimizing the prediction process and thus effectively reducing the overall cost of industrial production. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a schematic diagram of the overall framework of the dynamic soft measurement algorithm of the present invention;

[0054] Figure 2 This is a flowchart of the IGWO-DBN algorithm model of the present invention;

[0055] Figure 3 This is a schematic diagram of the DBN network structure of the present invention;

[0056] Figure 4 This is a schematic diagram of the RBM structure of the present invention;

[0057] Figure 5 This is a graph showing the predicted alumina concentration in the validation set of this invention.

[0058] Figure 6 This is a comparison chart of predicted and actual alumina concentration values ​​using different methods of the present invention;

[0059] Figure 7 Convergence curves for different methods of this invention;

[0060] Figure 8 This is a schematic diagram of the interface design for predicting alumina concentration using the IGWO-DBN model of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0062] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0063] It should be noted that, unless otherwise specified, the embodiments and features and technical solutions in the embodiments of the present invention can be combined with each other.

[0064] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0065] Reference Figures 1-8 A soft measurement method for alumina concentration based on the IGWO-DBN algorithm.

[0066] 1. Principle of Dynamic Soft Measurement Prediction Algorithm

[0067] Aluminum electrolysis is a complex production process characterized by multiple variables, nonlinearity, and time lag. Due to the extreme environmental influences during production, a large number of parameters generated are highly complex and variable, making it impossible to continuously measure the alumina concentration in real time. Therefore, the controllable input of this prediction algorithm is the combined input vector of the feature vectors of the previous five time points and the original physical quantities for alumina concentration prediction: two-stage voltage, anode rod current, and electrode distance. The system output is the predicted alumina concentration for the next time point, ultimately achieving an 8-dimensional input and 1-dimensional output for alumina concentration prediction.

[0068] Due to the complexity of the numerous parameters generated during aluminum electrolysis, a soft measurement model is constructed using the classic Deep Belief Networks (DBN) for prediction. An improved Grey Wolf Optimizer (IGWO) based on the DLH search strategy is used to optimize three network parameters: the number of hidden layer nodes, the backpropagation learning rate, and the number of backpropagation iterations. These optimized parameters are then used to construct the DBN network. The input layer of the DBN receives various relevant parameters from the aluminum electrolysis process. The hidden layers extract key features of the predicted alumina concentration signal through dynamic mode decomposition (EMD) and nonlinear transformation, transforming them into input conditions. The output layer outputs the prediction result, i.e., the alumina concentration.

[0069] 2. Dynamic Soft Measurement Prediction Algorithm

[0070] Before predicting alumina concentration, the input data needs to be standardized.

[0071] Before constructing the soft measurement prediction algorithm model, the raw data was preprocessed, including data cleaning, normalization, and feature selection. For feature selection, the Empirical Mode Decomposition (EMD) method was employed to improve model training performance. EMD decomposition was performed on the first five time-stamped data points of alumina concentration to obtain the IMF. s The system extracts features from each IMF at the first five time points and combines these features with the features of the trend term into a feature vector. This feature vector is then used as the input to the prediction model. Finally, the feature vector and the original prediction input physical quantities are merged to achieve an 8-dimensional input and 1-dimensional output for alumina concentration prediction.

[0072] The above process can be described using mathematical expressions as follows:

[0073] Let X (t) This is the original time series, where t = 1, 2, 3, ..., N represents the alumina concentration value.

[0074] 1) EMD decomposition: For X (t) Perform EMD decomposition to obtain K IMFs. s IMF1, IMF2, ..., IMF k And a trend term RES,X (t) =IMF1(t) + IMF2(t) + ... + IMF k (t)+RES(t). Then from each IMF k Extract the standard deviation σ from (t) k Feature F k ,F k ={σ k}

[0075] 2) Constructing feature vectors and merging prediction input conditions: The extracted relevant features F... k If these are combined to form an eigenvector f(t), then the eigenvector f(t-1) of the previous time step is f(t-1), f(t-1) = |F1(t-1),F2(t-1),…,F… k (t-1),F RES(t-1)|, where t-1 represents the previous time step. Let X(t) be the combined input vector containing the feature vectors f(t-5), f(t-4), f(t-3), f(t-2), f(t-1) of the previous five time steps, and the original physical quantities for predicting alumina concentration: two-stage voltage V(t), anode conductor current I(t), and electrode distance R(t), X(t) = V(t), I(t), R(t), f(t-5), f(t-4), f(t-3), f(t-2), f(t-1)|. Finally, the combined input vector X(t) is used as input data to train the DBN model (M represents the prediction model) to predict the alumina concentration.

[0076] The detailed design of the dynamic soft measurement algorithm mainly consists of the following two steps:

[0077] (1) Improved Gray Wolf Algorithm (IGWO) Optimization of Deep Belief Network Related Parameters Algorithm Design

[0078] The Improved Grey Wolf Algorithm (IGWO) based on dimensional learning-based hunting search strategy optimizes the number of hidden layer nodes, back-tuning learning rate, and back-iteration number parameters of the deep belief network. It mainly consists of three stages: initialization stage, movement stage, and selection and update stage.

[0079] In the initial stage, the number of hidden layer nodes H is randomly initialized, the learning rate α is fine-tuned in reverse, and the number of reverse iterations β is adjusted. The gray wolf population P is initialized within a given range [I...]. j ,u j Within the space, N individual gray wolves are randomly distributed within the preset search space using formula (1), where each individual gray wolf ω represents a parameter combination (H, α, β).

[0080] X ij =l j +rand j [0,1]*(u j -l j ), i∈[1,N],j∈[1,D] (1)

[0081] Among them, l j u j These are the lower and upper bounds of the search space for each individual gray wolf i in the j-th dimension, respectively; rand j [0,1] is a random number within the interval [0,1], used for the random distribution of individual gray wolves in the j-th dimension;

[0082] During the movement phase, for each individual gray wolf ω, the experience of its neighbors and random individuals is learned, and a weighted average is calculated for each dimension d according to the DLH search strategy formulas (2) and (3):

[0083]

[0084] in, It is the neighbor n of individual ω j Summation, w n is a neighbor n j weight, w r R is the weight of individual r; random(P) is the set of individuals randomly selected from population P; It represents the neighbor n at the i-th iteration. j Position or state in the d-th dimension It represents the position or state of individual ω in the d-th dimension at the i-th iteration. It represents the position or state of a random individual r in the d-th dimension at the i-th iteration; It is the new position of individual ω in dimension d at the (i+1)th iteration; fitness(n j ), fitness(r), and fitness(ω) are the values ​​of the neighboring individuals n. j The fitness values ​​of reference individuals r and ω;

[0085] During the selection and update phase, the current optimal solution X is selected. * Compare the fitness values ​​of the new individual and the current optimal solution, f(X) ω,i+1 ) <f(X * If the fitness value of a new individual is less than that of the current optimal solution, the current optimal solution is updated. This process is repeated until the predetermined number of iterations is reached or the termination condition is met. In each iteration, all gray wolves update their positions according to the DLH search strategy, simulating the interaction and competition among gray wolves to discover better parameter combinations and improve network performance. Through multiple iterations, the optimal parameter combination—the number of hidden layer nodes, the back-adjustment learning rate, and the number of back-adjustment iterations—is finally found and applied to the DBN network to optimize network performance.

[0086] (2) Design of Deep Belief Network (DBN) for Predicting Alumina Concentration

[0087] DBN network structure as follows Figure 3 As shown, this is a deep neural network consisting of several layers of Restricted Boltzmann Machines (RBMs) and one layer of Backpropagation (BP). The RBM model is as follows... Figure 4 As shown, the undirected probability graph contains one observable variable (v) and a single-layer hidden variable (h). The RBM has only two layers of neurons and is a bipartite graph. Units in the two layers are interconnected, but there are no connections between any units within a layer. The RBM is an energy-based model where any two connected neurons have a weight W representing the connection weight. ij Wij Let v represent the weight between observable variable unit i and hidden variable unit j, and let v and h represent the observation layer and hidden layer, respectively. Then, the joint allotment energy connecting the observation layer variable v and the hidden layer variable h, which are determined by the weights and biases, is as follows:

[0088]

[0089] Among them, a i It is the bias of the observable layer cell, b j It is the bias of the hidden layer unit, W ij As the connection weights between observable layer units and hidden layer units, the probability distributions of the observable and hidden layers can be obtained based on the energy function:

[0090]

[0091] Where Z is the normalization constant of the partition function, which is the sum of the energies under all parameters. This function accumulates all possible combinations of observable and hidden vectors. For each neuron in an RBM layer, there are only two states: 0 or 1. Given the states of each neuron in any layer, the state probabilities of the observable layer neurons and the hidden layer neurons can be obtained as follows:

[0092]

[0093] The DBN training process consists of pre-training and fine-tuning. Alumina concentration prediction data is first input into the network structure through the input layer, generating a vector V, which is then passed to the hidden layers via weights W to obtain H. Next, each layer of the RBM network is trained independently in unsupervised mode using three network parameters optimized by the improved Grey Wolf algorithm: the number of hidden layer nodes, the backpropagation learning rate, and the number of backpropagation iterations. This ensures that the feature vectors map to different feature spaces. Finally, the BP network receives the output feature vector of the RBM as its input feature vector and then propagates the error information back to each layer of the RBM from top to bottom, updating the weights of each hidden and output layer, thus achieving fine-tuning of the DBN network and performing supervised training. The forward and backpropagation steps are repeated until the predetermined number of iterations is reached or the termination condition is met, ultimately obtaining the weights and biases in the network for the final DBN model to output the predicted alumina concentration value based on the input data.

[0094] Finally, the performance of the model is evaluated using a validation set, as shown in Equation (10-12) and Figure 5 As shown, the performance of this algorithm model is described by calculating the coefficient of determination R2, root mean square error RMSE, and squared absolute error MAE, as well as the goodness of fit of the predicted images on the validation set. The model structure or parameters are then finely adjusted based on the validation results.

[0095]

[0096] Among them, y i This is the actual value. Here are the predicted values, Var is the variance, and R is the predicted value. 2 A larger value (closer to 1) indicates a better fit of the model to the data. Generally, the smaller the difference between the predicted and actual values, the smaller the MAE value, the better the prediction performance and the higher the prediction accuracy. The RMSE value is used to measure the prediction accuracy of the prediction model on continuous data. The smaller the value, the higher the prediction accuracy.

[0097] By comprehensively improving the Grey Wolf algorithm, optimizing the parameter design of the deep belief network, and designing the deep belief network for predicting alumina concentration, a soft measurement method for alumina concentration based on the IGWO-DBN algorithm is obtained. This method can effectively learn the relationship between alumina concentration and its relevant input conditions, achieving accurate prediction of alumina concentration. By training the algorithm model using recent electrolytic aluminum industrial production data, it achieves the effect of multiple alumina concentration predictions with a single training iteration, effectively reducing industrial production costs.

[0098] 3) Comparison of simulation results of IGWO-DBN algorithm with other algorithm models

[0099] Depend on Figure 6 It can be seen that after optimization with different algorithms, the prediction accuracy of each combined model was improved to varying degrees. Comparing different optimization algorithms, the DBN model optimized using the improved Grey Wolf optimization algorithm outperforms the other two models in overall performance, exhibiting smaller fluctuations between the predicted and actual alumina concentration values, higher prediction accuracy, and greater system stability. In terms of convergence, the IGWO-DBN model also significantly outperforms the other two models. The IGWO-DBN model achieved convergence after only six iterations, and its fitness function value is also the lowest. Compared to the other two models, it converges faster and has a lower final convergence value. Figure 7 As shown.

[0100] Table 1 presents the evaluation metrics for the four models. Each DBN combination model optimized using different algorithms shows smaller errors than the standard DBN model. Among PSO-DBN, SSA-DBN, and IGWO-DBN, IGWO-DBN exhibits the best fit. Furthermore, the MAE and RMSE values ​​of the IGWO-DBN model are slightly lower than the other models, confirming the model's effectiveness and accuracy.

[0101] Table 1 Comparison of prediction results optimized by different algorithms

[0102]

[0103] 3. Visual interface for dynamic soft measurement prediction algorithm

[0104] refer to Figure 8 Interface Design: First, set the learning rate, number of pre-training iterations, and number of pre-training groups, import the training data, and perform data standardization and partitioning. Then, after parameter optimization using the IGWO algorithm, the number of hidden layer nodes, number of backpropagation iterations, and backpropagation learning rate of the RBM are automatically set to the algorithm parameter area, eliminating the need for manual input. Next, a DBN prediction model is constructed and trained using the training set data. Finally, the model's prediction accuracy is judged by the image fit and related error values ​​such as R2, RMSE, and MAE in the alumina concentration prediction result display image of the training set. After successful model construction, import actual working condition data and perform predictions, simultaneously saving and exporting the prediction results to an Excel file, achieving the final effect of multiple predictions through a single training iteration.

[0105] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An alumina concentration soft measurement method based on an IGWO-DBN algorithm, characterized in that, The steps are as follows: S1: according to the deep belief network, a soft measurement model is constructed for prediction; S2: the number of hidden layer nodes, the backpropagation learning rate and the number of backpropagation iterations of the deep belief network are optimized by improving the grey wolf algorithm; Before predicting the concentration of alumina, the input data is standardized, and the steps are as follows: S1: before constructing the soft measurement prediction algorithm model, the original data is preprocessed; S2: the first five time data of the alumina concentration value is decomposed by EMD to obtain IMF and trend term; S3: features are extracted from each IMF of the first five time, and the features and the features of the trend term are combined into a feature vector, which is used as the input of the prediction model; S4: the feature vector and the original prediction input physical quantity are combined to realize the prediction of the 8-dimensional input and 1-dimensional output of the alumina concentration; The feature vector and the original prediction input physical quantity are combined, and the steps are as follows: (1) The extracted relevant features F k are combined into a feature vector f(t), and the feature vector f(t-1) of the previous time is expressed as follows: f(t-1) = |F1(t-1), F2(t-1),..., F k (t-1), F REs (t-1) |; Wherein, t-1 represents the previous time; (2) the feature vector of the first five time and the original physical quantity of the alumina concentration prediction two-stage voltage V(t), anode guide rod current I(t), and polar distance R(t) are combined to obtain the combined input vector X(t), and the expression is as follows: X(t) = |V(t), I(t), R(t), f(t-5), f(t-4), f(t-3), f(t-2), f(t-1)| Wherein, f(t-5), f(t-4), f(t-3), f(t-2), f(t-1) are the feature vectors of the first five time; (3) the combined input vector X(t) is used as the input data to train the DBN model to predict the alumina concentration, and the expression is as follows: In step S2, the number of hidden layer nodes, the backpropagation learning rate and the number of backpropagation iterations of the deep belief network are optimized, and the steps are as follows: A. In the initial stage, the number of hidden layer nodes H, the backpropagation learning rate a, and the number of backpropagation iterations b are randomly initialized; B. In the motion stage, for each grey wolf individual ω, the experience of neighbors and random individuals is learned, and the weighted average value of each dimension d is calculated; C. Selection and update phase selects the current best solution X * , compare the fitness value of the new individual and the current best solution; In step A, N grey wolf individuals are randomly distributed in the preset search space, and the expression is as follows: X ij = l j + rand j [0, 1] * (u j - l j ), i ∈ [1, N], j ∈ [1, D]; where, l j , u j are the lower and upper bounds of the search space for each grey wolf individual i in the jth dimension; rand j [0,1] is a random number in the interval [0,1] for the random distribution of the grey wolf individual in the jth dimension; In step B, in the motion stage, for each grey wolf individual ω, the experience of neighbors and random individuals is learned, and the weighted average value of each dimension d is calculated, and the expression is as follows: wherein is the neighbor n j of the individual ω n is the weight w j of the neighbor n r is the weight of the individual r; random(P) is a set of randomly selected individuals in the population P; is the position or state of the neighbor n j in the d-th dimension at the i-th iteration, is the position or state of the individual ω in the d-th dimension at the i-th iteration, is the position or state of the random individual r in the d-th dimension at the i-th iteration; is the new position of the individual ω in the d-th dimension at the i+1-th iteration; fitness(n j ), fitness(r), fitness(ω) are the fitness values of the neighbor individual n j , the reference individual r and the individual ω, respectively; In the C step, the selection and update stage selects the current optimal solution X * , and compares the fitness value of the new individual and the current optimal solution, as shown in the following expression: f(X ω，i+1 )<f(X * ) If the fitness value of the new individual is less than the fitness value of the current optimal solution, the current optimal solution is updated.

2. The IGWO-DBN algorithm-based soft measurement method for alumina concentration according to claim 1, characterized in that, The deep belief network is composed of several layers of restricted Boltzmann machines and one layer of BP, and the training of the deep belief network is composed of pre-training and fine-tuning. 3.The method of claim 2, wherein the method is based on an IGWO-DBN algorithm. The training of the deep belief network is composed of pre-training and fine-tuning, and the steps are as follows: (1) the alumina concentration prediction data is first input into the network structure to generate a vector V; (2) through the weight value W, it is transmitted to the hidden layer to obtain H; (3) the number of hidden layer nodes, the backpropagation learning rate and the number of backpropagation iterations of the deep belief network optimized by the improved grey wolf algorithm are used to train each RBM network independently and unsupervisedly, and the feature vector is mapped to different feature spaces; (4) The output feature vector of the RBM is received by the BP network as its input feature vector, and then the error information is propagated from top to bottom to each layer of the RBM, and the weights of each hidden layer and output layer are updated to realize fine tuning of the DBN network; (5) The weights and biases in the network are obtained for the final DBN model to output the predicted value of the alumina concentration of the input data. 4.The method of claim 1, wherein the method is based on an IGWO-DBN algorithm. The performance of the algorithm model is described by calculating the determination coefficient R2, the root mean square error RMSE and the mean absolute error MAE, and the fitting degree of the predicted image of the verification set, and the model structure or parameters are adjusted according to the verification result, and the expression is as follows: wherein y i is the actual value, is the predicted value, Var is the variance, R 2 is the correlation coefficient, MAE is the mean absolute error, RMSE is the root mean square error. The smaller the MAE value, the better the fitting effect of the model on the data; the smaller the difference between the predicted value and the actual value, the smaller the MAE value, and the better the prediction performance; the smaller the RMSE value, the higher the prediction accuracy.

Images