Tin smelting process soft measurement method based on MVMD and improved BLS

Abstract

The invention provides a tin smelting process soft measurement method based on MVMD and improved BLS, and belongs to the technical field of new information processing, and the method comprises the steps: obtaining tin smelting process data, and carrying out the normalization preprocessing of the tin smelting process data, and obtaining the normalized tin smelting process data; inputting the normalized tin smelting process data into a multivariate variable mode decomposition (MVMD) model, and extracting predefined K multivariate intrinsic mode function sets from multivariate data; inputting the K multivariate intrinsic mode function sets into an improved generalized regression neural network BLS model for training to obtain a trained soft measurement prediction model; inputting new tin smelting process data into the soft measurement prediction model for prediction to obtain a prediction value; performing comprehensive evaluation on the performance of the soft measurement prediction model according to multiple evaluation indexes of R2, RMSE and MAE; and carrying out reverse normalization processing on the prediction value to obtain a final prediction result of soft measurement in the tin smelting process.

Description

Technical Field

[0001] The present invention relates to the technical field of information processing, and particularly to a soft sensing method for the tin smelting process based on MVMD and improved BLS. Background Art

[0002] During the tin smelting process, it is difficult to measure multiple key parameters in real time, which severely restricts the monitoring and optimization of the production process. Traditional methods cannot accurately capture the complex non-linear dynamic characteristics in tin smelting, resulting in insufficient parameter prediction accuracy. At the same time, the tin smelting process involves multiple interrelated parameters, which have complex time-frequency characteristics and non-linear relationships, and it is difficult for a single data processing method to comprehensively reflect these characteristics. In addition, the tin smelting environment is complex and changeable, and various factors interact with each other. How to maintain the robustness and adaptability of the model under dynamically changing conditions is also a major challenge. These problems make it difficult to establish an accurate soft sensing prediction model, affecting the real-time prediction of key parameters. If these parameters cannot be obtained in a timely and accurate manner, it will directly affect the control decisions of the smelting process, and further affect the product quality and production efficiency. How to establish a soft sensing prediction model that can adapt to dynamic changes, has high precision and robustness by comprehensively considering the time-frequency characteristics and non-linear relationships among multiple parameters in the complex tin smelting environment has become the core technical problem to be solved urgently. Summary of the Invention

[0003] The present invention provides a soft sensing method for the tin smelting process based on MVMD and improved BLS, mainly including:

[0004] Obtain the tin smelting process data, and perform normalization preprocessing on the tin smelting process data to obtain the normalized tin smelting process data; input the normalized tin smelting process data into the Multivariate Variable Mode Decomposition (MVMD) model, and extract a predefined set of K multivariate intrinsic mode functions from the multivariate data; input the set of K multivariate intrinsic mode functions into the improved Generalized Regression Neural Network (BLS) model for training to obtain a trained soft sensing prediction model; input new tin smelting process data into the soft sensing prediction model for prediction to obtain a predicted value; comprehensively evaluate the performance of the soft sensing prediction model according to multiple evaluation indexes such as R 2 , RMSE and MAE; perform denormalization processing on the predicted value to obtain the final prediction result of the soft sensing of the tin smelting process.

[0005] Further, the normalization preprocessing of the tin smelting process data includes: obtaining the initial data set X from the DCS system, denoted as where each row is the data generated at each time point, the number of variables is D, and the above data set is the data taken at m time points; perform normalization processing on the initial data set X, and the formula for its normalization processing is:

[0006] Among them, and are the minimum and maximum values of the initial dataset X, respectively.

[0007] Furthermore, the input of the normalized tin smelting process data into the MVMD model includes: constructing a multi-constrained variational optimization problem with the goal of minimizing the sum of the bandwidths of the multivariate intrinsic mode function set; constructing an augmented Lagrangian function to remove the constraints of the multivariate variational optimization problem; solving the variational optimization problem in the constraints by the alternating direction method of multipliers; after multiple iterations, outputting the decomposition result of MVMD.

[0008] Furthermore, the input of the set of K multivariate intrinsic mode functions into the improved BLS model for training includes: combining the set of K multivariate intrinsic mode functions with the observable data to form a data sample Y, and using it as the input for training the improved BLS model; mapping the data sample Y into h groups of feature nodes through the BLS model, and integrating the h groups of feature nodes to obtain h groups of feature parts, that is Substituting the h groups of feature parts into the formula Constructing d groups of enhanced nodes; constructing a feature interaction layer H c ,

[0009] H c =σ(W·[F n ,E d +b), where σ is a non-linear activation function, W is a weight matrix, and b is a bias term; the feature matrix A composed of all nodes of BLS, A = [F n E d H c , and combining with the BLS network model to obtain the output result.

[0010] Furthermore, the input of the new tin smelting process data into the soft sensor prediction model for prediction includes: after training is completed, saving the trained BLS model parameters, including the weights of feature nodes, the weights of enhanced nodes, and the weights of the output layer, etc.; when new data arrives, first perform the same MVMD preprocessing on these new data, and then input the obtained set of multivariate intrinsic mode functions into the trained BLS model for prediction to obtain the predicted value.

[0011] Furthermore, the overall performance of the model is evaluated by comprehensively comparing indicators such as R 2 , RMSE, and MAE, including: using the R 2 indicator to measure the fitting degree of the model to the data, R 2The closer it is to 1, the stronger the interpretability of the model; the RMSE index is used to measure the difference between the predicted value and the true value, and the smaller the RMSE, the higher the prediction accuracy; the MAE index is used to measure the average absolute difference between the predicted value and the true value, and the smaller the MAE, the smaller the prediction error.

[0012] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0013] The present invention discloses a method for predicting key parameters in the tin smelting process. Aiming at the problem that multiple key parameters in the tin smelting process are difficult to measure in real time, a soft-sensing prediction model based on multivariate variational mode decomposition and improved generalized regression neural network is proposed. The present invention first obtains and normalizes multivariate time series data, and then uses the multivariate variational mode decomposition method for time-frequency domain decomposition to obtain multiple multivariate intrinsic mode functions. These functions are used as features and input into an improved generalized regression neural network model with a feature interaction layer, and a soft-sensing prediction model for the tin smelting process is obtained through training. The present invention also includes a model performance evaluation and iterative optimization mechanism to ensure the accuracy of the prediction results. Finally, the predicted values of the key parameters are obtained through anti-normalization processing, providing an effective means for real-time monitoring and control of the tin smelting process, and improving production efficiency and product quality. Brief Description of the Drawings

[0014] Figure 1 It is a flowchart of a soft-sensing method for the tin smelting process based on MVMD and improved BLS of the present invention. Detailed Embodiments

[0015] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] To facilitate the understanding of the technical solutions of the present invention, the technical terms involved in the present invention will be described first as follows:

[0017] Multivariate variable mode decomposition MVMD model: In the tin smelting process, accurately monitoring and controlling key parameters is crucial for improving product quality and production efficiency. The MVMD model can be used to process multi-channel sensor data and separate different frequency components, which helps to monitor and control the smelting process. However, modal aliasing may occur in MVMD, that is, there is frequency overlap between different modes, resulting in inaccurate decomposition results.

[0018] Improved Generalized Regression Neural Network BLS Model: The improved GRNN-BLS model can be used to establish a non-linear mapping between input variables and output performance indicators to achieve soft sensing. However, its performance depends on the quality and quantity of training data. When the data is insufficient or noisy, the model effect may decline.

[0019] Such as Figure 1 , in this embodiment, a soft sensing method for the tin smelting process based on MVMD and improved BLS specifically may include:

[0020] S101, Obtain the tin smelting process data, and perform normalization preprocessing on the tin smelting process data to obtain the normalized tin smelting process data.

[0021] The specific steps include:

[0022] First, obtain the initial data X from the DCS system, denoted as where each row is the data generated at each time point, the number of variables is D, and the above data set is the data collected at m time points;

[0023] Secondly, perform normalization processing on the initial data set X, and the formula for its normalization processing is:

[0024] where and are the minimum and maximum values of the initial data set X respectively.

[0025] S102, Input the normalized tin smelting process data into the Multivariate Variable Mode Decomposition MVMD model to extract a predefined set of K multivariate intrinsic mode functions from the multivariate data.

[0026] Inputting the normalized tin smelting process data into the MVMD model specifically includes:

[0027] 1) Construct a multi-constrained variational optimization problem, with the goal of minimizing the sum of the bandwidths of the set of multivariate intrinsic mode functions: {u k,c} is the set of IMFs; { ω k} is the set of central frequencies of {u k,c}; u k,c is the k-th IMF of the c-th channel; ω k is 's central frequency; is the partial derivative operation with respect to time; Adopt the Hilbert transform operator for the analytical representation of the vector signal u k,c (t); x c (t) is the input data of the c-th channel.

[0028] 2) Construct the augmented Lagrangian function to remove the constraints of the multivariate variational optimization problem. The augmented Lagrangian function contains two penalty terms, where the quadratic term is used to ensure the reconstruction accuracy, and the Lagrange multiplier term is used to ensure strict satisfaction of the constraint conditions:

[0029]

[0030] 3) Solve the variational optimization problem in the constrained \(L(\{\mathbf{u} k,c ,\{\omega k \},\lambda c \})\) by the alternating direction method of multipliers. \(\{\mathbf{u} k,c \}\) is the set of IMFs; { \omega k} is the set of central frequencies of \(\{\mathbf{u} k,c \}\); \(\mathbf{u} k,c \) is the \(k\)th IMF of the \(c\)th channel; \(\omega k is 's central frequency, and \(\lambda c is the representative Lagrange multiplier of the \(c\)th channel.. Among them, the mode update formula The update equation of the central frequency is: and are the Fourier transforms of \(x(t)\), \(\lambda(t)\) and \(u(t)\) respectively.

[0031] 4) After multiple iterations, output the decomposition result of MVMD

[0032] S103. Input the set of \(K\) multivariate intrinsic mode function sets into the improved generalized regression neural network BLS model for training to obtain a trained soft sensor prediction model.

[0033] The specific steps are as follows:

[0034] 1) Model training: Combine the set of \(K\) multivariate intrinsic mode function sets with the observable data to form a data sample \(Y\), and use it as the input of the improved BLS model for training.

[0035] 2) Feature node construction: Map the data sample into \(h\) groups of feature nodes through the BLS model, and integrate the \(h\) groups of feature nodes to obtain \(h\) groups of feature parts \(F h , that is where \(F i is the \(i\)th group of feature nodes.

[0036] 3) Enhanced node construction: Substitute \(F n into the formula Among them, E d is the d - group enhanced nodes; Ej is the j - th group of enhanced nodes. Among them, W fi , β fi , W ej , β ej are random parameter matrices; φ i is a linear function, and ξ i is a non - linear function.

[0037] 4) Construct the feature interaction layer H c : H c =σ(W·[F n , E d +b), where σ is a non - linear activation function, W is the weight matrix, and b is the bias term.

[0038] 5) The feature matrix A composed of all nodes of BLS is: A = [F n E d H c ; Combining with the BLS network model Y = AW, the output result is obtained, where the link weight W=(A T A + λI) -1 A T Y, where λ is the regularization coefficient, I is the identity matrix, and Y is the sample label.

[0039] S104, Input the new tin smelting process data into the soft - sensing prediction model for prediction to obtain the predicted value.

[0040] After training is completed, save the trained BLS model parameters, including the feature node weights, enhanced node weights, and output layer weights; when new data arrives, first perform the same MVMD pre - processing on this new data, and then input the obtained set of multivariate intrinsic mode functions into the trained BLS model for prediction to obtain the predicted value

[0041] S105, Comprehensively evaluate the performance of the soft - sensing prediction model according to multiple evaluation indexes such as the coefficient of determination R 2 , root - mean - square error RMSE, and mean absolute error MAE.

[0042] If the model performs well in multiple indexes, it is considered that the model has high prediction accuracy and stability. R2 is used to measure the fitting degree of the model to the data. The closer R2 is to 1, the stronger the explanatory ability of the model; RMSE is used to measure the difference between the predicted value and the true value. The smaller RMSE is, the higher the prediction accuracy; MAE is used to measure the average absolute difference between the predicted value and the true value. The smaller MAE is, the smaller the prediction error. is the predicted value, y i is the true value; n is the number of samples.

[0043] S106. Perform inverse normalization processing on the predicted value to obtain the final prediction result of the soft measurement in the tin smelting process.

[0044] Perform inverse normalization processing on the output of the model: assign the predicted value to the model output represents the output of the model; represents the result of inverse normalization, that is, the final predicted value.

[0045] In summary, the traditional VMD method is only used to decompose univariate time series and cannot handle multivariate data. Introducing MVMD can handle multivariate time series, realize the time-frequency synchronous analysis of multi-dimensional data, and improve the processing ability of complex data. The technical solution of the present invention combines MVMD with the improved BLS for the first time and applies it to the tin smelting process. Combining the feature interaction mechanism enhances the feature extraction ability of BLS. It has both the feature extraction ability of BLS and the training speed of BLS, improves the prediction accuracy of BLS, and thus can more accurately predict the key parameters in the tin smelting process, improving production efficiency and product quality.

[0046] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present application without departing from the spirit and scope of the embodiments of the present application. In this way, if these modifications and variations of the embodiments of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these changes and modifications.

Claims

1. A soft measurement method for the tin smelting process based on MVMD and improved BLS, characterized in that, Including: Obtain the tin smelting process data, and perform normalization preprocessing on the tin smelting process data to obtain the normalized tin smelting process data; Input the normalized tin smelting process data into the Multivariate Variable Mode Decomposition (MVMD) model, and extract a predefined set of K multivariate intrinsic mode functions from the multivariate data; Input the set of K multivariate intrinsic mode functions into the improved Generalized Regression Neural Network (BLS) model for training to obtain a trained soft sensor prediction model; Input new tin smelting process data into the soft sensor prediction model for prediction to obtain a predicted value; According to the coefficient of determination R 2 , multiple evaluation indexes such as root mean square error RMSE and mean absolute error MAE are used to comprehensively evaluate the performance of the soft sensor prediction model; Perform denormalization processing on the predicted value to obtain the final prediction result of the soft sensor for the tin smelting process.

2. The method according to claim 1, characterized in that, The normalization preprocessing of the tin smelting process data includes: Obtain the initial data X from the DCS system; Denoted as Wherein, each row is the data generated at each time point, the number of variables is D, and the above dataset is the data collected at m time points; Perform normalization processing on the initial data set X, and the formula for the normalization processing is: Among them, and are the minimum value and the maximum value of the initial data set X, respectively.

3. The method according to claim 1, characterized in that The input of the normalized tin smelting process data into the MVMD model includes: Construct a variational optimization problem with multiple constraints: , and through the variational optimization method, minimize the total bandwidth of the multivariate IMFs under the satisfaction of multiple constraints; Construct an augmented Lagrangian function and remove the constraints of the multivariate variational optimization problem; Solving the variational optimization problem in L({u k,c ,{ω k},λ c}) by the alternating direction multiplier method, where {u k,c} is the set of IMFs; { ω k} is the set of center frequencies of {u k,c}; u k,c is the k-th IMF of the c-th channel; ω k is the center frequency, and λ c is the representative Lagrange multiplier of the c-th channel; After multiple iterations, output the decomposition result of MVMD.

4. The method according to claim 1, characterized in that, The input of the set of K multivariate intrinsic mode functions into the improved BLS model for training includes: Combine the set of K multivariate intrinsic mode functions with the observable data into a data sample Y, and use it as the input for training the improved BLS model; The data sample Y is mapped into h groups of feature nodes through the BLS model, and the h groups of feature nodes are integrated to obtain h groups of feature parts, that is Substitute h groups of feature parts into the formula Construct d groups of enhanced nodes; Constructive Feature Interaction Layer H c , H c = σ(W · [F n , E d + b), where σ is a non-linear activation function, W is a weight matrix, and b is a bias term; The feature matrix A composed of all BLS nodes, A = [F n E d H c , combined with the BLS network model, to obtain the output result.

5. The method according to claim 1, wherein The input of the new tin smelting process data into the soft sensor prediction model for prediction includes: After the training is completed, save the parameters of the trained BLS model, including the weights of the feature nodes, the weights of the enhancement nodes, and the weights of the output layer; When new data arrives, first perform the same MVMD preprocessing on these new data, and then input the obtained set of multivariate intrinsic mode functions into the trained BLS model for prediction to obtain a predicted value.

6. The method according to claim 1, wherein By comprehensively comparing R 2 , RMSE, MAE and other indicators, the overall performance of the model is evaluated, including: Using R 2 to measure how well the model fits the data, the closer R 2 is to 1, the stronger the explanatory power of the model; Use the RMSE index to measure the difference between the predicted value and the true value. The smaller the RMSE, the higher the prediction accuracy; Use the MAE index to measure the average absolute difference between the predicted value and the true value. The smaller the MAE, the smaller the prediction error.

Images