A method for constructing and verifying environmental variables for cement clinker calcination

Abstract

The present invention belongs to the technical field of cement clinker calcination, and discloses a method for constructing and verifying cement clinker calcination environmental variables. According to cement plant data, a clinker calcination data pool is constructed; according to the data in the data pool, cement clinker calcination environmental variables and environmental feature data extraction based on ICA are constructed to obtain clinker calcination environmental variables; the main variables and environmental variables of the cement clinker calcination environment are used to respectively construct prediction models of the safety and quality of the cement clinker calcination process, and the prediction results of these models are analyzed to verify the validity of the environmental variables. The present invention extracts environmental variables from the cement clinker calcination environment, which can not only reflect the changing trend of the environment, but also remain stable in a short time, and is easy to better complete the tasks such as prediction and simulation of the clinker calcination system, which is conducive to realizing the intelligentization of cement clinker calcination.

Description

Technical Field

[0001] The invention belongs to the technical field of cement clinker calcination, and in particular relates to a method for constructing and verifying cement clinker calcination environmental variables. Background Art

[0002] In recent years, the industrial Internet platform can use intelligent technical means to reduce energy consumption and pollution emissions, thereby reducing corporate costs and enhancing corporate competitiveness. With the development of intelligent technology in the cement industry, future cement plants will be more intelligent and automated. Through the analysis of data in the industrial process, scientific and efficient control can be made in the production process to improve product quality and reduce energy consumption and pollution emissions. In cement industrial production, energy consumption costs account for a very high proportion of the entire production cost, and in the total energy consumption of the cement production process, the heat consumption of the clinker calcination process accounts for about 70-80%. The cement clinker calcination process is a key component of the cement production process, and the energy consumption in the entire process depends to a large extent on the heat consumption of the clinker calcination process, and there is a coupling relationship between the heat consumption of the cement clinker calcination process and the quality of the clinker. Therefore, combining the heat consumption simulation of the clinker calcination process with the prediction of clinker quality can provide a basis for the equipment parameter regulation of the cement clinker calcination process, which is conducive to the improvement of heat consumption in the entire calcination process and the control of cement clinker quality.

[0003] The energy consumption in the process of cement clinker calcination is mainly heat consumption (coal consumption), which can be measured by weighing. However, the quality of clinker in the future cannot be directly measured during the clinker calcination process. The acquisition of clinker quality information often requires manual testing, which has a large lag. There is a lack of quality prediction information to control the production equipment during the calcination process. Therefore, the simulation of heat consumption and the prediction of clinker quality in clinker calcination are extremely important for the scheduling and operation of the calcination process. However, the calcination process of cement clinker is a complex process composed of multiple production processes. The various processes affect each other. There are also complex coupling relationships between the various variables in the calcination process, and there are lags and uncertainties, which makes it difficult to establish an accurate prediction model for the cement clinker calcination process.

[0004] In response to the above problems, some scholars at home and abroad have used different algorithms to study and establish prediction models for key parameters in the process of cement clinker calcination: In the field of energy consumption in cement production, some studies have provided excellent algorithms and model processing methods. For example, Zhao et al. first used the principal component analysis method and the improved multivariate nonlinear regression algorithm to establish a cement power consumption prediction model. Although this algorithm can achieve certain results, it is prone to overfitting when there are many noise points. On this basis, in order to overcome the limitations of a single model, Zhao et al. established a hybrid model of least squares support vector regression (LSSVR) and an integrated empirical mode decomposition algorithm (EMD) to predict the power consumption demand of cement mills. This combined model can further improve the accuracy and generalization ability of the overall model. In the field of cement production prediction, studies based on different equipment parameters provide different ideas. For example, Yang et al. used a combination of BP neural network and Elman neural network to predict the temperature in the cement rotary kiln, which provided a basis for controlling the rotary kiln working conditions in the cement clinker calcination process. Minchala et al. used a combination of artificial neural network and adaptive neural fuzzy inference system to predict the raw material fineness of cement. In addition, Pani et al. also used a combination of principal component analysis and artificial neural network to predict the raw material fineness of cement. These accurate predictions of different key parameters can provide a guarantee for the subsequent simulation of clinker calcination environment.

[0005] In the field of cement production quality, free CaO content and clinker three-rate value are important indicators for evaluating cement product quality, but these indicators are difficult to measure directly during the production process. In this regard, some scholars use extreme learning machines, support vector machines, and deep belief networks to predict the free CaO content in cement clinker. In addition, some scholars use integrated algorithms to conduct research in this area. These studies on clinker quality can provide guarantees for subsequent clinker quality control.

[0006] Through the above analysis, although the existing technology has achieved good results, it still stays in the laboratory or at a certain point in time, and cannot grasp that clinker calcination is a process. The existing problems and defects are: the calcination process of cement clinker is a complex process composed of multiple production processes. The various processes affect each other, and there are also complex coupling relationships between the various variables in the calcination process, and there are lags and uncertainties, which makes it difficult to establish an accurate prediction model for the cement clinker calcination process. So, how can we obtain the characteristics of the clinker calcination process.

[0007] The difficulty in solving the above problems and defects is as follows: the equipment for clinker calcination is connected from the first-stage cyclone to the grate cooler, and the logistics, fuel flow and airflow run through it, with complex flow fields such as swirl and reflux; the physical and chemical reactions in the cement calcination process are complex, the combustion of coal powder releases heat, the chemical reaction of materials absorbs heat, the heat exchange between logistics and airflow, and the changes in raw materials from different sources cause changes in the calcination process; the clinker calcination process involves many variables, and the variables are coupled, nonlinear, and lagging; clinker calcination is a continuous process with real-time and continuity, etc. Different working conditions will affect the calcination process, and clinker calcination has inertia.

[0008] The significance of solving the above problems and defects is: if the environment during the clinker calcination process can be described and extracted, the main characteristics of the clinker calcination process can be described through this variable, some simulation or prediction tasks can be completed, and a good foundation can be laid for the intelligent control of the clinker calcination process. Summary of the invention

[0009] In view of the problems existing in the prior art, the present invention provides a method for constructing and verifying cement clinker calcination environmental variables. The present invention applies machine learning methods to the cement clinker calcination process, extracts environmental variables from the cement clinker calcination environment, and can be used to build a prediction model for the cement clinker calcination process, providing an effective means for realizing the intelligentization of clinker calcination.

[0010] The present invention is implemented by a method for constructing and verifying environmental variables for cement clinker calcination, the method comprising:

[0011] Step 1: Build a clinker calcination data pool based on cement plant data;

[0012] Step 2: Based on the data in the data pool, cement clinker calcination environmental variables and ICA-based environmental feature data extraction are constructed to obtain clinker calcination environmental variables;

[0013] Step three, use the variables of the cement clinker calcination environment and the environmental variables to respectively construct the safety prediction model and the quality prediction model of the cement clinker calcination process, and use the prediction results of the safety prediction model and the quality prediction model to verify the validity of the environmental variables; among them, 4 learning algorithms are used to construct the clinker calcination process safety prediction model and the clinker calcination process quality model, that is, modeling, and then these two prediction models are used to verify the validity of the environmental variables.

[0014] Furthermore, in the step 1, the clinker calcination data pool includes: various equipment parameters and various material properties, which together constitute multiple variables of the entire cement clinker calcination environment, and these variables are interconnected and influence each other.

[0015] Furthermore, the clinker calcination environment variable in step 2 is a variable or a matrix, which reflects the changes in the entire cement clinker calcination environment by its own changes. It is a manifestation of the common characteristics of the main variables in this process, and because the data in the cement clinker calcination process is periodic, it remains basically stable in a short period of time without causing huge fluctuations.

[0016] Furthermore, the specific process of extracting environmental feature data based on ICA in step 2 is as follows:

[0017] 1) Organize the original data into a matrix X with n rows and m columns by column;

[0018] 2) Zero-mean each row of X (representing a feature), that is, subtract the mean of this row;

[0019] 3) Perform whitening preprocessing on the data;

[0020] 4) Set the value of the parameter learning rate α;

[0021] 5) In the i-th solution W, where the initial W is a random matrix with the sum of each row being 1;

[0022] 6) Based on the W and formula in the previous step Solve for the independent component at the i-th moment

[0023] 7) Repeat 4) and 5) until the independent components at all moments The solution is complete;

[0024] 8) Combine the independent components obtained at each moment to get the final result

[0025] Furthermore, in step 3, the main variables and environmental variables of the cement clinker calcination environment are used to construct prediction models for the safety and quality of the cement clinker calcination process, and the prediction results of these models are analyzed to verify the validity of the environmental variables, which specifically include:

[0026] Selection of clinker calcination heat consumption model based on machine learning, model parameter selection based on grid searcher, division of training set, test set and validation set; prediction model construction, comparison and analysis of prediction results.

[0027] Furthermore, the selection process of the clinker calcination heat consumption model based on machine learning is:

[0028] Since the data distribution of each variable in the cement clinker calcination environment is different, the number of noise points contained in the data of each variable is also different;

[0029] Four regression models were constructed for the same input and output data, and the best model was selected through the evaluation index of the prediction model;

[0030] The four regression models are divided into: ridge regression, support vector regression, decision tree regression, and random forest regression.

[0031] Further, the specific process of model parameter selection based on the grid searcher is:

[0032] The main parameters of Ridge are: alpha, fit_intercept, max_iter, solver. Alpha is of float type, which represents the regularization coefficient. The larger the value, the stronger the regularization. fit_intercept is of bool type, which indicates whether to calculate the intercept. max_iter is of int type, which indicates the maximum number of iterations of the solver. solver indicates the type of interpreter, which includes auto, svd, cholesky, sparse_cg, lsqr, and sag.

[0033] The main parameters of SVR are: C, kernel; C is of float type, which reflects the penalty coefficient of the algorithm on the sample data; kernel represents the kernel function used by the model, which has 5 types: linear, poly, rbf, sigmoid, and precomputed;

[0034] The main parameters of Decision-Tree are: max_depth, min_samples_split, min_samples_leaf, max_features; max_depth indicates the maximum depth of the decision tree; min_samples_split indicates the minimum number of samples required for internal node redivision; min_samples_leaf indicates the number of minimum samples in the leaf node group; max_features indicates the maximum number of features considered during division;

[0035] The main parameters of Random-forest are: n_estimators, max_features, min_sample_leaf; n_estimators represents the number of decision trees; max_features represents the maximum number of features; min_sample_leaf represents the minimum number of samples in a leaf node.

[0036] Furthermore, the specific process of dividing the training set, test set, and validation set is as follows:

[0037] Training data refers to the data used by the model during training. It can be divided into training set and validation set. Model parameters are adjusted based on the performance of the model on the validation set. The test set is used to estimate the generalization ability of the model in actual use.

[0038] Each variable in the cement clinker calcination environment has strong continuity. When dividing the data set for these variables, the continuity of each set is ensured;

[0039] The specific process of constructing the prediction model is as follows:

[0040] The training set and test set are divided according to the following input and output data, and prediction models are constructed based on four regression algorithms. The GridSearchCV method is then used to adjust the parameters of these models to obtain the optimal parameters of each model, and finally a prediction model with the optimal parameters is obtained.

[0041] Another object of the present invention is to provide a storage medium for receiving a user input program, wherein the stored computer program enables an electronic device to execute the method for constructing and verifying the cement clinker calcining environment variable, comprising the following steps:

[0042] Step 1: Build a clinker calcination data pool based on cement plant data;

[0043] Step 2: Based on the data in the data pool, cement clinker calcination environmental variables and ICA-based environmental feature data extraction are constructed to obtain clinker calcination environmental variables;

[0044] Step three, the main variables and environmental variables of cement clinker calcination environment are used to construct prediction models for the safety and quality of cement clinker calcination process, and the prediction results of these models are analyzed to verify the validity of environmental variables.

[0045] Another object of the present invention is to provide a computer program product stored on a computer-readable medium, including a computer-readable program, which, when executed on an electronic device, provides a user input interface to implement the method for constructing and verifying the cement clinker calcining environmental variables.

[0046] Combining all the above technical solutions, the advantages and positive effects of the present invention are as follows:

[0047] (1) Through simple configuration, it can be applied to different cement clinker manufacturers with good adaptability; the obtained cement clinker calcination environmental variables can describe the current characteristics of the current clinker calcination and can be used for tasks such as real-time monitoring of clinker calcination processes; with the clinker calcination environmental variables, the characteristics of the current calcination can be comprehensively reflected without considering the complexity, coupling, and sustainability of cement clinker calcination data. Direct effect + overall effect

[0048] (2) The data, models used and parameter determination in this method are given. The prediction model given is a model that has been screened and has good results in prediction simulation of cement clinker calcination heat consumption or clinker quality. In addition, the cement clinker calcination process is real-time, continuous and changing, and the selection of the above algorithm is beneficial to actual production.

[0049] (3) By constructing and verifying the cement clinker calcination environmental variables, a new idea can be provided for realizing the intelligent production of cement clinker, and it can also provide an idea for the intelligentization of similar production processes.

[0050] The present invention extracts environmental variables from the cement clinker calcining environment, which can not only reflect the changing trend of the environment, but also remain stable in a short time, and is easy to better complete the prediction and simulation of the clinker calcining system and other tasks, which is conducive to realizing the intelligentization of cement clinker calcining. The present invention extracts environmental variables from the cement clinker calcining environment for simulating the cement clinker calcining environment, and can further carry out heat consumption simulation research. At the same time, the present invention can be applied to the simulation of the cement clinker calcining environment, so as to construct a more stable and accurate heat consumption simulation model of the clinker calcining environment, thereby reducing the heat consumption of the entire calcining process under the premise of ensuring the quality of the clinker. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flow chart of a method for constructing and verifying cement clinker calcination environmental variables provided in an embodiment of the present invention.

[0052] Figure 2 It is a schematic diagram of the idea of ​​constructing cement clinker calcining environment variables provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0054] In view of the problems existing in the prior art, the present invention provides a method for constructing and verifying environmental variables for cement clinker calcination. The present invention is described in detail below in conjunction with the accompanying drawings.

[0055] The method for constructing and verifying cement clinker calcination environmental variables provided by the present invention can also be implemented by other steps by ordinary technicians in the industry. Figure 1 The method for constructing and verifying cement clinker calcination environmental variables provided by the present invention is only a specific embodiment.

[0056] like Figure 1 As shown, the method for constructing and verifying cement clinker calcination environmental variables provided by the embodiment of the present invention includes:

[0057] S101: Construct a clinker calcination data pool based on cement plant data.

[0058] S102: According to the data in the data pool, cement clinker calcination environmental variables and ICA-based environmental feature data extraction are constructed to obtain clinker calcination environmental variables.

[0059] S103: Construct prediction models for the safety and quality of the cement clinker calcination process based on the main variables and environmental variables of the cement clinker calcination environment, and verify the validity of the environmental variables by analyzing the prediction results of these models.

[0060] In S101 provided in the embodiment of the present invention, the clinker calcination data pool includes: various equipment parameters and various material properties, which together constitute multiple variables of the entire cement clinker calcination environment, and these variables are interconnected and influence each other.

[0061] In S102 provided in the embodiment of the present invention, the clinker calcination environment variable is a variable or a matrix, which reflects the changes in the entire cement clinker calcination environment by its own changes. It is a manifestation of the common characteristics of the main variables of this process, and because the data in the cement clinker calcination process is periodic, it can remain basically stable in a short time without causing huge fluctuations.

[0062] In S102 provided in the embodiment of the present invention, the specific process of extracting environmental feature data based on ICA is as follows:

[0063] 1) Organize the original data into a matrix X with n rows and m columns by column;

[0064] 2) Zero-mean each row of X (representing a feature), that is, subtract the mean of this row;

[0065] 3) Perform whitening preprocessing on the data;

[0066] 4) Set the value of the parameter learning rate α;

[0067] 5) In the i-th solution W, where the initial W is a random matrix with the sum of each row being 1;

[0068] 6) Based on the W and formula in the previous step Solve for the independent component at the i-th moment

[0069] 7) Repeat 4) and 5) until the independent components at all moments The solution is complete;

[0070] 8) Combine the independent components obtained at each moment to get the final result

[0071] In S103 provided by the embodiment of the present invention, the prediction models of the safety and quality of the cement clinker calcining process are respectively constructed based on the main variables and environmental variables of the cement clinker calcining environment, and the prediction results of these models are analyzed to verify the validity of the environmental variables, which specifically includes:

[0072] Selection of clinker calcination heat consumption model based on machine learning, model parameter selection based on grid searcher, division of training set, test set and validation set; prediction model construction, comparison and analysis of prediction results.

[0073] The selection process of the clinker calcination heat consumption model based on machine learning is:

[0074] Since the data distribution of each variable in the cement clinker calcination environment is different, the number of noise points contained in the data of each variable is also different;

[0075] Four regression models were constructed for the same input and output data, and the best model was selected through the evaluation index of the prediction model;

[0076] The four regression models are divided into: ridge regression, support vector regression, decision tree regression, and random forest regression.

[0077] The specific process of model parameter selection based on the grid searcher is as follows:

[0078] The main parameters of Ridge are: alpha, fit_intercept, max_iter, solver. Alpha is of float type, which represents the regularization coefficient. The larger the value, the stronger the regularization. fit_intercept is of bool type, which indicates whether to calculate the intercept. max_iter is of int type, which indicates the maximum number of iterations of the solver. solver indicates the type of interpreter, which includes auto, svd, cholesky, sparse_cg, lsqr, and sag.

[0079] The main parameters of SVR are: C, kernel; C is of float type, which reflects the penalty coefficient of the algorithm on the sample data; kernel represents the kernel function used by the model, which has 5 types: linear, poly, rbf, sigmoid, and precomputed;

[0080] The main parameters of Decision-Tree are: max_depth, min_samples_split, min_samples_leaf, max_features; max_depth indicates the maximum depth of the decision tree; min_samples_split indicates the minimum number of samples required for internal node redivision; min_samples_leaf indicates the number of minimum samples in the leaf node group; max_features indicates the maximum number of features considered during division;

[0081] The main parameters of Random-forest are: n_estimators, max_features, min_sample_leaf. n_estimators indicates the number of decision trees; max_features indicates the maximum number of features; min_sample_leaf indicates the minimum number of samples in a leaf node.

[0082] The specific process of dividing the training set, test set, and validation set is as follows:

[0083] Training data refers to the data used by the model during training. It can be divided into training set and validation set. Model parameters are adjusted based on the performance of the model on the validation set. The test set is used to estimate the generalization ability of the model in actual use.

[0084] Each variable in the cement clinker calcination environment has strong continuity. When dividing the data set for these variables, the continuity of each set is ensured.

[0085] The specific process of constructing the prediction model is as follows:

[0086] The training set and test set are divided according to the following input and output data, and prediction models are constructed based on four regression algorithms. The GridSearchCV method is then used to adjust the parameters of these models to obtain the optimal parameters of each model, and finally a prediction model with the optimal parameters is obtained.

[0087] The technical solution of the present invention is described in detail below in conjunction with specific embodiments.

[0088] The cement clinker calcining process is a production process involving multiple links and multiple equipments. Various equipment parameters and various material properties together constitute multiple variables of the entire cement clinker calcining environment. These variables are interrelated and affect each other. Therefore, simulating the cement clinker calcining environment is the basis of task simulation research of the clinker calcining process.

[0089] 1. Introduction of cement clinker calcination environmental variables

[0090] The equipment of cement clinker calcining process mainly consists of five-stage cyclone (preheater), decomposition furnace, rotary kiln, grate cooler and fan. The cement clinker calcining process is the process of raw material becoming clinker after pre-decomposition, chemical reaction, calcination, cooling and other stages. At the same time, this process is also a process of gas heat exchange, in which the material path (material flow route) and the gas path (gas flow route) are in opposite directions, that is, the cement raw material and the counter-flowing high-temperature gas flow form a mixture in the equipment and exchange heat. Next, the cement clinker calcining environment will be analyzed according to the cement production process.

[0091] During the calcination of cement clinker, each link has a certain impact on the quality of cement clinker. First, the cement raw material enters the preheater, and each stage of the cyclone needs to preheat the cement raw material. Then, the cement raw material is pre-decomposed in the decomposition furnace, which is the main place for gas heat exchange. After that, the cement raw material enters the rotary kiln. The rotary kiln requires a higher temperature to make the raw material undergo a series of complex physical and chemical reactions. The heat mainly comes from the coal feeding amount of the kiln head scale and the kiln tail scale. In addition to the temperature requirements, this process also requires the rotary kiln to rotate continuously to ensure the uniform calcination of the material. Finally, the calcined clinker is cooled and transported out through the grate cooler. In addition to the flow of materials, the whole process also includes the flow of gas, including the primary and tertiary winds blown in by the fan. Since the cement plant has been repeating the above process, and the cement plant is mostly in safe production, the whole process will not have a large fluctuation. Therefore, all the data generated by this process must have a certain cycle, but this cycle is difficult to determine. If this property can be used, it will produce a good effect.

[0092] In summary, cement clinker calcination is a complex and continuous process, and the main variables in it are all related to the final clinker quality, but at the same time, these variables are also inevitably related to the heat consumption indicator variables. If these variables are directly used to simulate tasks such as heat consumption, the accuracy of the simulation results will be affected. For example, when the heat consumption indicator variables are manually adjusted, the calcination environment will also change, making it difficult to accurately predict the safety and quality of cement clinker calcination. Since the safety of production and the quality of clinker cannot be guaranteed, such a simulation is a failure.

[0093] Therefore, the cement clinker environmental variable is introduced. This variable needs to contain most of the effective information of the cement clinker calcination environment, and it must ensure that it can maintain its own stability under short-term heat consumption and other task simulations, that is, when the heat consumption indicator variable changes in a short time, the environmental variable can remain stable. The clinker calcination environmental variable can be a variable or a matrix. It can use its own changes to reflect the changes in the entire cement clinker calcination environment. It is a manifestation of the common characteristics of the main variables of this process. And because the data in the cement clinker calcination process is periodic, it can remain basically stable in a short time without huge fluctuations. This is the reason for extracting the cement clinker calcination environmental variable.

[0094] 2 Environmental variable extraction based on independent component analysis

[0095] Independent Component Correlation Algorithm (ICA) is often used to extract independent components from data, and can often discover potential essential changes from the original data. Therefore, the ICA method is used to extract environmental variables from the cement clinker calcination environment. The fixed point algorithm (Fast ICA) is widely used in the field of data processing due to its fast convergence speed and good separation effect. The algorithm is based on the fixed point recursive algorithm. It can be applied to any type of data, and it can always highlight its superior performance when analyzing high-dimensional data. Therefore, based on the Fast ICA algorithm, environmental variables are extracted from the main variables of the cement clinker calcination environment.

[0096] Table 1 Environment variable extraction algorithm based on Fast ICA

[0097]

[0098] The time complexity of the algorithm is T(n)=O(n·m), where n is the number of rows of input data and m is the number of columns of input data. The environmental variable extraction algorithm based on Fast ICA is shown in Table 1. The environmental variable extracted by the algorithm is a multidimensional matrix, each column of which is independent and contains most of the effective information of the cement clinker calcination environment.

[0099] 3. Verification of the validity of environment variables

[0100] Environmental variables were extracted through the above experiments. Although in theory environmental variables represent the common characteristics of cement clinker calcination environment, whether the changes in environmental variables can reflect the changing trends of these parameters still needs to be proved by experiments. Therefore, the main variables and environmental variables of cement clinker calcination environment are used to construct prediction models for the safety and quality of cement clinker calcination process, and the prediction results of these models are analyzed to verify the effectiveness of environmental variables.

[0101] 3.1 Selection of clinker calcination heat consumption model based on machine learning

[0102] Various variables in the cement clinker calcination environment change continuously, so the prediction of safety and quality indicator variables adopts the following regression type prediction:

[0103] (1) Ridge Regression (Ridge): The advantage of Ridge is that the regression coefficient is more significant, and it often has greater practical value in studies with a large number of pathological data; its disadvantage is that the predicted R2 score is generally lower.

[0104] (2) Support Vector Regression (SVR). The advantages of SVR are that the algorithm is simple, it has good robustness, and the prediction results are easy to explain. Its disadvantages are that it is very sensitive to the choice of parameters and kernel functions, and it is difficult to process large-scale data.

[0105] (3) Decision Tree Regression (Decision-Tree). The advantages of Decision-Tree are fast training and prediction speed, good fitting effect for nonlinear relationships in data sets, and easy-to-interpret results. Its disadvantages are low prediction accuracy, susceptibility to outliers or noise, and even possible overfitting.

[0106] (4) Random Forest Regression (Random-Forest). The advantage of Random-Forest is that it has stronger resistance to overfitting and can reduce the risk of overfitting by averaging decision trees. Its disadvantages are that it is more complex, has higher computational cost, and is also easily affected by noise.

[0107] Since the data distribution of each variable in the cement clinker calcination environment is different, and the number of noise points contained in each variable data is also different, it is difficult to select the model based on the data distribution. Therefore, the above four regression models are constructed for the same input and output data, and the best model is selected through the evaluation index of the prediction model.

[0108] 3.2 Model parameter selection based on grid searcher

[0109] Usually, various regression models of machine learning have different model parameters, and the change of parameters will seriously affect the accuracy and generalization ability of the model. For example, the main parameters in the SVR model include the penalty coefficient C, which reflects the degree of penalty for illegal sample data. When C is too small, the training error of the model will increase; when C is too large, the generalization ability of the model will deteriorate.

[0110] GridSearchCV is a model parameter adjustment tool that can automatically adjust parameters and find the optimal results and parameters. In addition, GridSerchCV can also set the evaluation criteria of the model and perform cross-validation. Therefore, the GridSearchCV tool is used to select parameters for each regression model. Among them, the parameters of each regression algorithm are different. The main parameters of the above four regression models are as follows:

[0111] (1) Ridge main parameters: alpha, fit_intercept, max_iter, solver. alpha is of float type, which represents the regularization coefficient. The larger the value, the stronger the regularization. fit_intercept is of bool type, which indicates whether to calculate the intercept. max_iter is of int type, which indicates the maximum number of iterations of the solver. solver indicates the type of interpreter. There are 6 types: auto, svd, cholesky, sparse_cg, lsqr, and sag.

[0112] (2) SVR main parameters: C, kernel. C is of float type, which reflects the penalty coefficient of the algorithm on the sample data; kernel represents the kernel function used by the model, which has five types: linear, poly, rbf, sigmoid, and precomputed.

[0113] (3) Decision-Tree main parameters: max_depth, min_samples_split, min_samples_leaf, max_features. max_depth indicates the maximum depth of the decision tree; min_samples_split indicates the minimum number of samples required for internal node re-division; min_samples_leaf indicates the minimum number of samples in the leaf node group; max_features indicates the maximum number of features considered during division.

[0114] (4) Random-forest main parameters: n_estimators, max_features, min_sample_leaf. n_estimators indicates the number of decision trees. Generally speaking, the more decision trees there are, the better the generalization ability of the model will be, but the accuracy may decrease; max_features indicates the maximum number of features; min_sample_leaf indicates the minimum number of samples in a leaf node.

[0115] 3.3 Training set, test set, and validation set division

[0116] Usually, training data refers to the data used by the model during the training process. It can be divided into training set and validation set. The model parameters can be adjusted based on the performance of the model on the validation set. The test set is used to estimate the generalization ability of the model in actual use. Each variable in the cement clinker calcination environment has a strong continuity. When dividing the data set for these variables, the continuity of each set should be guaranteed. Therefore, in order to screen out the globally optimal parameters and taking into account the size of the data set, the entire data set is divided into 3:1:1 continuous parts, of which the training set accounts for 3 parts, the validation set accounts for 1 part, and the test set accounts for 1 part, and predictions are made for different test sets separately, that is, 5 predictions are made.

[0117] 3.4 Prediction model construction

[0118] The training set and test set are divided according to the following input and output data, and prediction models are constructed based on four regression algorithms. The GridSearchCV method is then used to adjust the parameters of these models to obtain the optimal parameters of each model, and finally a prediction model with the optimal parameters is obtained.

[0119] (1) Prediction model with cement clinker calcination data as input

[0120] Input variable 1: heat consumption indicator variable (kiln head scale, kiln tail scale), cement clinker calcination data (35 variables determined based on input data).

[0121] Output variables: safety (clinker quantity), quality (clinker KH rate value, clinker SM rate value, clinker IM rate value).

[0122] (2) Prediction model with cement clinker environmental variables as input

[0123] Input variable 2: heat consumption indicator variable (kiln head scale, kiln tail scale), cement clinker calcination environment variable.

[0124] Output variables: safety (clinker quantity), quality (clinker KH rate value, clinker SM rate value, clinker IM rate value).

[0125] 3.5 Comparison and analysis of prediction results

[0126] Using input variable 1 (clinker calcination data) and input variable 2 (clinker calcination environmental variables), the scores of these models under two different input variables are compared. The specific scores (average of 5 times) are shown in Table 2. Table 2 Model output scores using input variable 1 (clinker calcination data) and input variable 2 (clinker calcination environmental variables).

[0127] Table 2 Scoring

[0128]

[0129] From the above prediction result comparison table, it can be seen that in the prediction experiment of clinker quantity, input variable 1 performs better, and its MSE and R2 evaluation indicators are better; in the prediction experiment of clinker KH rate value, input variable 2 performs better, and its MSE and R2 evaluation indicators are better, but the difference in MSE between the two input variables is not large; in the prediction experiment of clinker SM rate value, input variable 2 performs better, and its MSE and R2 evaluation indicators are better, and input variable 2 shows an absolute advantage at this time; in the prediction experiment of clinker IM rate value, input variable 2 performs better, and its MSE and R2 evaluation indicators are better, but the difference in MSE between the two input variables is not large.

[0130] In summary, the two input variables show their own advantages in predicting the safety and quality of clinker calcination. Input variable 2 is more advantageous in predicting the three rates of clinker, while input variable 1 is more advantageous in predicting the amount of clinker. However, the model evaluation indicators of the two input variables in predicting the amount of clinker are not much different, and this gap is acceptable. Therefore, it is believed that the environmental variables extracted in the previous experiment are effective.

[0131] The implementation of the present invention is carried out in a unified experimental environment, and the implementation of all algorithms and models is all based on Python 3.7 and its extension package, including data analysis tools such as numpy and pandas, matplotlib for drawing, and sklearn is mainly used in the machine learning environment. The specific experimental environment configuration is shown in Table 3.

[0132] Table 3 Software and hardware environment for the implementation of the present invention

[0133]

[0134] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for constructing and verifying environmental variables for cement clinker calcination, characterized in that: The construction and verification method of the cement clinker calcination environmental variables includes: Step 1: Use cement plant data to build a clinker calcination data pool; Step 2: Use the data in the data pool to construct cement clinker calcination environmental variables and extract environmental feature data based on ICA to obtain clinker calcination environmental variables; Step 3: Use the variables of cement clinker calcination environment and environmental variables to respectively construct a safety prediction model and a quality prediction model for cement clinker calcination process, and use the prediction results of the safety prediction model and the quality prediction model to verify the validity of the environmental variables; The specific process of extracting environmental feature data based on ICA in step 2 is as follows: 1) Organize the original data into columns OK Column Matrix ; 2) Zero-mean each row of , minus the mean of this row; 3) Perform whitening preprocessing on the data; 4) Set parameter learning rate The value of 5) In the It's a solution , where the initial is a random matrix whose sum of each row is 1; 6) According to the previous step and formula Solve Independent component of time ; 7) Repeat 4) and 5) until all independent components at all moments are , the solution is completed; 8) Combine the independent components obtained at each moment to get the final result .

2. The method for constructing and verifying cement clinker calcination environmental variables according to claim 1, characterized in that: The clinker calcination data pool in step 1 includes: multiple variables that constitute the entire cement clinker calcination environment using equipment parameters and material properties.

3. The method for constructing and verifying cement clinker calcination environmental variables according to claim 1, characterized in that: In step three, the safety prediction model and quality prediction model of cement clinker calcination process are constructed by selecting the clinker calcination heat consumption model based on machine learning, selecting the model parameters based on the grid searcher, dividing the training set, test set, and validation set, and constructing the prediction model.

4. The method for constructing and verifying cement clinker calcination environmental variables as claimed in claim 3, characterized in that: The selection of the clinker calcination heat consumption model based on machine learning includes: Since the data distribution of each variable in the cement clinker calcination environment is different, the number of noise points contained in the data of each variable is also different; For the same input and output data, four regression models were constructed and the best model was selected through the evaluation index of the prediction model; The four regression models are: ridge regression, support vector regression, decision tree regression, and random forest regression.

5. The method for constructing and verifying cement clinker calcination environmental variables as claimed in claim 3, characterized in that: The specific process of model parameter selection based on the grid searcher is as follows: Ridge parameters: alpha, fit_intercept, max_iter, solver, alpha is of float type, which represents the regularization coefficient. The larger the value, the stronger the regularization; fit_intercept is of bool type, indicating whether to calculate the intercept; max_iter is of int type, indicating the maximum number of iterations of the solver; solver indicates the type of interpreter, including auto, svd, cholesky, sparse_cg, lsqr, sag; SVR parameters: C, kernel; C is of float type, which reflects the penalty coefficient of the algorithm on the sample data; kernel represents the kernel function used by the model, including linear, poly, rbf, sigmoid, and precomputed; Decision-Tree parameters: max_depth, min_samples_split, min_samples_leaf, max_features; max_depth indicates the maximum depth of the decision tree; min_samples_split indicates the minimum number of samples required for internal node redivision; min_samples_leaf indicates the number of minimum samples in the leaf node group; max_features indicates the maximum number of features considered during division; Random-forest parameters: n_estimators, max_features, min_sample_leaf; n_estimators represents the number of decision trees; max_features represents the maximum number of features; min_sample_leaf represents the minimum number of samples in a leaf node.

6. The method for constructing and verifying cement clinker calcination environmental variables as claimed in claim 3, characterized in that: The specific process of dividing the training set, test set, and validation set is as follows: The data used in the training process is divided into a training set and a validation set. The model parameters are adjusted based on the performance of the model on the validation set; the test set is used to estimate the generalization ability of the model in actual use.

7. The method for constructing and verifying cement clinker calcination environmental variables as claimed in claim 3, characterized in that: The specific process of constructing the prediction model is as follows: a training set and a test set are divided according to the input and output data, and a prediction model is constructed based on four regression algorithms. Then, the GridSearchCV method is used to adjust the parameters of the prediction model to obtain the optimal parameters of each model, and a prediction model with the optimal parameters is obtained.

8. A storage medium for receiving user input program, wherein the stored computer program enables an electronic device to execute the method for constructing and verifying cement clinker calcining environmental variables as described in any one of claims 1 to 7, comprising the following steps: Step 1: Build a clinker calcination data pool based on cement plant data; Step 2: Based on the data in the data pool, cement clinker calcination environmental variables and ICA-based environmental feature data extraction are constructed to obtain clinker calcination environmental variables; Step three, the main variables and environmental variables of cement clinker calcination environment are used to construct prediction models for the safety and quality of cement clinker calcination process, and the prediction results of these models are analyzed to verify the validity of environmental variables.

9. A computer program product stored on a computer-readable medium, comprising a computer-readable program, which, when executed on an electronic device, provides a user input interface to implement the method for constructing and verifying cement clinker calcining environmental variables as described in any one of claims 1 to 7.

Images