Cement clinker free lime soft-sensing method based on adversarial causal learning

Abstract

The application relates to the technical field of soft measurement of cement clinker production processes, and discloses a cement clinker free calcium oxide soft measurement method based on adversarial causal learning, which comprises the following steps: (1) data acquisition and integration; (2) data selection; (3) modeling training; and (4) model prediction; the method is a cement clinker f-CaO soft measurement method based on adversarial causal learning, the prediction effect of key variables in the cement clinker production process is improved, and the method has universality and generality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of soft measurement of cement clinker production process, and particularly relates to a cement clinker free calcium oxide soft measurement method based on adversarial causal learning. BACKGROUND

[0002] At present, the new dry process cement manufacturing method is widely used in main cement production enterprises in China, and the production technical process can be generally divided into three links: raw material preparation, clinker calcination and cement preparation. The content of free calcium oxide (f-CaO) in cement clinker is one of the main indexes for measuring the quality of cement. When the content of f-CaO in cement is too high, due to the slow hydration speed, after the cement hardens and gradually hydrates to form a certain strength, it will cause uneven expansion of the cement volume, reduce the strength, and even cause cracks and collapse, thereby reducing the stability of the cement. When the content of f-CaO in cement clinker is between 0.5% and 1.5%, the cement has good stability and strength, and at this time, the cement can be identified as a qualified product. Therefore, real-time online measurement of the content of f-CaO in clinker to adjust the related process parameters in the cement production process is the key to ensuring the quality of cement. The commonly used f-CaO content detection method is to periodically take samples manually after the clinker is formed and cooled, and then conduct laboratory offline testing. The manual offline measurement method has a long measurement period, lacks real-time performance, has a large hysteresis, and increases labor costs.

[0003] With the data-driven modeling method showing great advantages in industrial processes, and in order to overcome the difficulty of analyzing key product quality, more and more people use data-driven modeling methods to estimate key product quality from other online measurable variables. The cement clinker production process has strong nonlinearity between product quality and operating variables, so the data-driven modeling method based on the cement clinker production process is mainly divided into statistical methods, shallow machine learning methods and deep learning methods. Among them, the statistical methods mainly include principal component analysis (PCA), partial least squares (PLS). Shallow machine learning methods refer to traditional machine learning models other than deep learning, including shallow artificial neural networks (ANN), support vector machines (SVM), Gaussian process models (GPM). Deep learning mainly includes various deep neural networks (DNN) with more than three layers, which have been proven to be able to approximate any nonlinear system and are widely used in process modeling.

[0004] Since its inception, Graph Neural Networks (GNNs) have developed rapidly, demonstrating satisfactory learning capabilities in many fields. Furthermore, GNN-based methods can handle non-grid data, thus capturing the relationships between features. This software proposes a soft measurement method for f-CaO in cement clinker based on Adversarial Causal Graph Learning (AGCL). AGCL takes all features from the cement clinker production process as input to predict key product quality. Moreover, during the training phase, the causal matrix between features is optimized along with other parameters. In this way, the network can learn and determine the causal relationships between features based on given training data, thereby providing more accurate prediction results. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a soft measurement method for free calcium oxide in cement clinker based on adversarial causal learning.

[0006] The technical solution adopted by this invention to solve its technical problem is:

[0007] A soft measurement method for free calcium oxide in cement clinker based on adversarial causal learning includes the following steps:

[0008] (1) Data acquisition and integration: Collect sensor data from the cement clinker production process, and classify, collect and integrate the data.

[0009] (2) Data selection: Select data according to the maximum mutual information criterion (MMI) and remove redundant and useless data.

[0010] (3) Modeling and training: Construct the algorithm model AGCL for the f-CaO soft measurement method of cement clinker based on graph network. The input of the model is the selected and retained data. Then, the AGCL model is used to learn and train the integrated data.

[0011] (4) Model prediction: The trained AGCL model is used to predict the f-CaO concentration in the cement clinker production process, and the prediction results are given.

[0012] Furthermore, the process of step (1) is as follows:

[0013] Step 1.1: Collect sensor data for the cement clinker production process, which includes 74 variables.

[0014] Step 1.2: Divide the data into training set, validation set and test set according to a 3:1:1 ratio.

[0015] Step 1.3: Due to the significant differences in data among different feature variables, it is necessary to standardize the data to facilitate model processing and calculation. The specific formula is shown below:

[0016]

[0017] Where X is the standardized data, X' is the unstandardized original data, μ is the mean of the data, and β is the standard deviation of the data.

[0018] Furthermore, the process of step (2) is as follows:

[0019] Step 2.1: Calculate the maximum mutual information coefficient of each variable with respect to the target variable Y as follows:

[0020]

[0021] Where B = N 0.55～0.6 N is the size of the sample set. The calculated mutual information is substituted into the formula to determine the maximum mutual information.

[0022] Step 2.2: Sort the variables according to the maximum mutual information, select the top 15 variables, and complete the variable selection process.

[0023] Furthermore, the process of step (3) is as follows:

[0024] Step 3.1: AGCL is a cement clinker f-CaO concentration prediction model built based on relevant algorithms. AGCL is mainly constructed using a self-attention mechanism (SAM) and a graph convolutional network (GCN). In SAM, data X undergoes different linear transformations to obtain query Q and key K, which can be described by the following formula:

[0025]

[0026] Among them W q and W k b q and b k Let A and B represent the trainable matrix and trainable bias, respectively, and let A be the adjacency matrix representing the causal relationship between the variables.

[0027] GCN is an algorithm that performs convolution on data in non-Euclidean space. GCN can be represented as:

[0028]

[0029] Where X is the standardized data, W is the weight matrix, σ(.) is the activation function, A represents the input adjacency matrix, and D represents the degree matrix of adjacency matrix A.

[0030] Step 3.2: First, use SAM to generate matrix A representing causal relationships; then, construct an autoencoder structure based on GCN, where both the decoder and encoder are single-layer GCNs; input the data X and matrix A into the autoencoder. The autoencoder's task is to reduce the mutual information coefficient between the encoded data and f-CaO, while the SAM's task is to increase the mutual information coefficient between the two. Through alternating training and iterative updates until convergence, matrix A is obtained. Figure 2 This is an alternating training process of autoencoders and SAMs, used to discover causal relationships between variables.

[0031] Step 3.3: Input the data X and matrix A into the three-layer GCN to achieve the prediction of the f-CaO concentration of cement clinker. Figure 1 The framework for AGCL.

[0032] Furthermore, the process of step (4) is as follows:

[0033] Step 4.1: Input the test data into the trained AGCL model. Use the root mean square error (RMSE) as the evaluation metric, calculated as follows:

[0034]

[0035] Among them, y' i Represents real data, y i This represents the model's output, and n represents the number of samples in the test set. Generally, the smaller the RMSE, the closer the model's predictions are to the true values, meaning the model's prediction performance is better.

[0036] The beneficial effects of this invention are mainly reflected in the following: This invention proposes a soft measurement method for free calcium oxide in cement clinker based on adversarial causal learning, which improves the prediction accuracy of key variables in the cement clinker production process. This method utilizes a self-attention mechanism and a graph convolutional network to extract the causal relationships of variables, increasing the generalization ability of the model. This method can accurately measure key variables in different cement clinker production processes. Attached Figure Description

[0037] Figure 1 This is a model framework diagram of the present invention;

[0038] Figure 2 This is a flowchart of the causal discovery process of this invention. Detailed Implementation

[0039] The present invention will now be further described with reference to the accompanying drawings.

[0040] Reference Figure 1 A soft measurement method for free calcium oxide in cement clinker based on adversarial causal learning, the method comprising the following steps:

[0041] (1) Obtain the cement clinker production process dataset:

[0042] Step 1.1: Collect sensor data on the clinker production process of a large cement plant in China. A total of 1844 samples were obtained, with each sample containing 74 variables.

[0043] Step 1.2: Divide the data into three parts using a sliding time window with a window size of 4 and a step size of 1: 1107×74×4, 369×74×4, and 369×74×4, which are the training set, validation set, and test set, respectively.

[0044] Step 1.3: Process each data point according to the standardized formula.

[0045] (2) Select data for the cement clinker production process dataset. The process is as follows:

[0046] Step 2.1: Calculate the mutual information value of each variable according to the mutual information formula.

[0047] Step 2.2: Calculate the maximum mutual information value based on the mutual information value.

[0048] Step 2.3: Sort the variables according to the maximum mutual information value and retain the top 15 variables: raw material temperature, first stage cylinder outlet temperature, first stage cylinder outlet pressure, sixth stage cylinder negative pressure, magnesium oxide content at the mill outlet, iron oxide in the kiln, magnesium oxide content in the kiln, aluminum ratio in the kiln, silica content in the clinker, alumina content in the clinker, iron oxide content in the clinker, calcium oxide content in the clinker, magnesium oxide content in the clinker, aluminum ratio in the clinker, and liquid phase content in the clinker.

[0049] (3) Train the AGCL model as follows:

[0050] Step 3.1: Input the training set into the model to obtain the model's first prediction.

[0051] Step 3.2: Adjust the model parameters based on the difference between the predicted and output values ​​to reduce the difference between the predicted and actual values.

[0052] Step 3.3: Input the validation set into the model and obtain the model's output results for the validation set.

[0053] Step 3.4: Based on the model's output on the validation set, reset and tune the model's hyperparameters.

[0054] Step 3.5: Repeat steps 3.1 to 3.4 until the difference between the model's predicted value and the true value is within the allowable error range.

[0055] (4) Test the model using test data, as follows:

[0056] Step 4.1: Input the training data into the trained model and PLS, SVR, and LSTM to obtain the prediction values ​​of AGCL, PLS, SVR, and LSTM on the test set, respectively.

[0057] Step 4.2: Calculate the RMSE between the predicted and actual values ​​on the test set to evaluate the performance of AGCL, PLS, SVR, and LSTM. The evaluation results are shown in the table below:

[0058] Table 1 Comparison of RMSE for AGCL, PLS, SVR, and LSTM

[0059]

[0060] The comparison results above show that the present invention has a better predictive effect on key variables in the cement clinker production process than the traditional prediction model.

[0061] The method of this invention adopts a graph network-based f-CaO soft measurement method for cement clinker, which improves the prediction effect of key variables in the cement clinker production process and has universality and versatility.

[0062] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms stated in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A soft measurement method for free calcium oxide in cement clinker based on adversarial causal learning, characterized in that, Includes the following steps: 1) Data acquisition and integration: Collect sensor data from the cement clinker production process, and classify, collect and integrate the data; The specific process of step 1) is as follows: 1.1): Collect sensor data during the cement clinker production process; 1.2): Divide the data into training, validation, and test sets according to the proportions; 1.3): Due to the significant differences in data among different feature variables, it is necessary to standardize the data to facilitate model processing and calculation. The specific formula is shown below: Where X is the standardized data, X' is the unstandardized original data, μ is the mean of the data, and β is the standard deviation of the data; 2) Data selection and sorting: Data is selected based on the Maximum Mutual Information (MMI) criterion, and redundant and useless data is removed; The specific process of step 2) is as follows: 2.1): Calculate the maximum mutual information coefficient of each variable with respect to the target variable Y: Where B = N 0.55～0.6 N is the size of the sample set, and X is the standardized data; the calculated mutual information is substituted into the formula to determine the maximum mutual information; 2.2): Sort the variables according to the maximum mutual information, select the top k variables, and complete the variable selection process; 3) Modeling and Training: Construct the AGCL algorithm model for the f-CaO soft sensing method of cement clinker based on graph networks. The input of the model is the selected and retained data. Then, the AGCL model is used to learn and train on the integrated data. The specific process of step 3) is as follows: 3.1): AGCL is a model for predicting f-CaO concentration in cement clinker, constructed based on the self-attention mechanism SAM and the graph convolutional network GCN. In SAM, input data X undergoes different linear transformations to obtain query Q and key K, described by the following formula: Among them W q and W k b q and b k Let A and k represent the trainable matrix and trainable bias, respectively. Let A be the adjacency matrix representing the causal relationship between the variables, and k represent the number of variables. GCN is an algorithm that performs convolution on data in non-Euclidean space. GCN is represented as: Where X is the standardized data, W is the weight matrix, σ(.) is the activation function, A represents the input adjacency matrix, and D represents the degree matrix of adjacency matrix A; 3.2): First, SAM is used to generate matrix A representing causal relationships; then, an autoencoder structure is constructed based on GCN, where both the decoder and encoder are single-layer GCNs; the data X and matrix A are input into the autoencoder; the task of the autoencoder is to reduce the mutual information coefficient between the encoded data and f-CaO, while the task of SAM is to increase the mutual information coefficient between the two; through alternating training and continuous updates until convergence, matrix A is obtained. 3.3): Input the data X and matrix A into the three-layer GCN to achieve the prediction of f-CaO concentration in cement clinker; 4) Model prediction: The trained AGCL model is used to predict the f-CaO concentration in the cement clinker production process, and the prediction results are given.

2. The soft measurement method for free calcium oxide in cement clinker based on adversarial causal learning as described in claim 1, characterized in that, The specific process of step 4) is as follows: 4.1): Input the test data into the trained AGCL model; use the mean squared error (RMSE) as the evaluation metric, calculated using the following formula: Where y'i represents the real data, yi represents the model output, n represents the number of samples in the test set, and the smaller the RMSE, the closer the model's prediction is to the real value, that is, the better the model's prediction performance.

Images