A prediction model for cement raw material aging reaction system

Abstract

The present application discloses a prediction model for a cement raw material aging reaction system, and the prediction model is trained according to the following steps: obtaining known raw material parameters, reaction conditions and clinker parameters as basic data; setting possible raw material aging reaction equations and possible products, and obtaining data on the participation ratio of each reaction; performing standardization and normalization preprocessing on the basic data and the data on the participation ratio of each reaction; building a multi-scale residual convolutional neural network model to learn the basic data and the data on the participation ratio of each reaction, and constructing a deep network structure; compiling the trained model, observing and optimizing the model performance, and using the prediction model to be applied to the design and development of high-performance iron-aluminum cement clinker, to perform performance prediction and material design on iron-aluminum cement, to screen raw materials and optimize the ratio, and to reduce the consumption of manpower, material resources and time.

Description

Technical Field

[0001] The present application relates to the technical field of cement clinker production, and in particular to a prediction model for a cement raw material aging reaction system. Background Art

[0002] Iron-aluminum cement clinker uses limestone, high-iron bauxite, gypsum, etc. as raw materials, which are sintered in a rotary kiln to form clinker. Its mineral phase reaction system is a five-element system of CaO-SiO2-Al2O3-Fe2O3-SO3. The main mineral components of the clinker are calcium sulfoaluminate, dicalcium silicate, calcium aluminoferrite, etc. Adding appropriate amounts of gypsum and limestone to the clinker and grinding them together can produce iron-aluminum cement with properties such as fast setting, early strength, and slight expansion. The performance of iron-aluminum cement is closely related to the mineral composition of its clinker.

[0003] However, the existing cement clinker design and development process has the following defects: Since there is no accurate theoretical calculation method for the CaO-SiO2-Al2O3-Fe2O3-SO3 five-component reaction system and the principle of mineral phase coexistence, the current design and development of iron-aluminum cement clinker mainly relies on the "trial and error method", which consumes a lot of manpower, material resources and time. Summary of the invention

[0004] One purpose of the present application is to provide a cement raw material aging reaction system prediction model that can predict the performance, design materials, screen raw materials and optimize the proportion of iron-aluminum cement.

[0005] To achieve the above objectives, the technical solution adopted in this application is: a prediction model for cement raw material aging reaction system, and the prediction model is trained according to the following steps:

[0006] S100, obtaining known raw material parameters, reaction conditions and clinker parameters as basic data;

[0007] S200, setting possible raw material cooking reaction equations and possible products, and obtaining the participation ratio data of each reaction;

[0008] S300, standardize and normalize the basic data and the data of each reaction participation ratio;

[0009] S400, build a multi-scale residual convolutional neural network model to learn basic data and the proportion of participation in each reaction, and build a deep network structure;

[0010] S500, compile the trained model, observe and optimize the model performance.

[0011] In some embodiments, the raw material cooking reaction equation and products include:

[0012]

[0013]

[0014]

[0015]

[0016]

[0017] in, a , b , c is the proportion of Fe2O3 participating in the relevant reaction, a + b + c ≤1.

[0018] In some embodiments, before the prediction model learns the basic data and the data of each reaction participation ratio, the basic data and the data of each reaction participation ratio are arranged in the following format:

[0019] X=[x1x2……x q ], x1, x2...x q ∈R N×1

[0020] Y=[y1y2……y p ],y1,y2……y p ∈R N×1

[0021] Among them, x1, x2, ... x q Represent the raw material parameters, reaction conditions and the proportion of each reaction in each group of cement raw materials, totaling q variables, y1, y2...y p They represent the mineral phase percentage of clinker, with a total of p variables.

[0022] In some embodiments, the TensorFlow framework is used to standardize and normalize the arranged basic data and the data on the participation ratio of each reaction, so that data with different characteristics have similar scales, and then the preprocessed data is converted into a tensor format to comply with the data structure operated by the TensorFlow framework.

[0023] In some embodiments, the preprocessed data is divided into a training set, a validation set, and a test set according to m:n:1-mn, where 0.8<m<1, n<0.2.

[0024] In some embodiments, the construction process of the multi-scale residual convolutional neural network model is: construct w residual blocks, set up convolution layers, batch normalization layers and activation layers, define multi-scale residual blocks to combine convolution kernels of different sizes in the convolution layer to capture multi-scale features, and use the residual blocks and multi-scale residual blocks defined above to construct a multi-scale residual convolutional neural network model.

[0025] In some embodiments, after constructing the multi-scale residual convolutional neural network model, the multi-scale residual convolutional neural network model is compiled, and the loss function, optimizer and evaluation index optimization model are specified.

[0026] In some embodiments, the loss function uses sparse_categorical_crossentropy, the optimizer uses Bayesian optimization, the evaluation indicators select mean absolute error and mean square error, and the test set data is used to draw a scatter plot of actual values ​​and predicted values ​​for evaluation. If the model performs poorly, Bayesian optimization is used to adjust parameters.

[0027] In some embodiments, developing model computing software and embedding the prediction model into the model computing software comprises:

[0028] K100, layout the software front-end interface and install the software front-end in the smart terminal;

[0029] K200, develops a database, embeds the prediction model into the database as a module, and sets up the developed database on the server as the software backend to connect to the front end;

[0030] K300, input basic data information in the software front-end interface, and the database determines the scope and format of the basic data information;

[0031] K400, if the judgment is no, the user will be prompted with a data error in the software front-end interface;

[0032] K500, if the judgment is yes, the basic data information is uploaded to the database, and the database returns the prediction result and outputs it in the software front-end interface.

[0033] In some embodiments, the database includes a data upload module, a data query module and a data prediction module. The data upload module conducts human-computer interaction through a data upload interface to upload and manage basic data. The data query module conducts human-computer interaction through a data query interface to query basic data information of cement clinker through raw material parameters and reaction conditions, or to query existing raw material parameters and reaction condition information through clinker parameters. The data prediction module includes a prediction model. The data prediction module conducts human-computer interaction through a data prediction interface to predict the mineral content of known clinker.

[0034] Compared with the prior art, the beneficial effects of this application are:

[0035] 1. The cement raw material clinkering reaction system prediction model of the present application can predict the mineral composition of the corresponding cement clinker according to specific / known raw materials (raw materials) and their proportions. By comparing with the expected value, the raw material proportion can be further adjusted and the reaction process can be optimized to make the predicted value approach the expected value continuously. This process solves the problem of uncertainty in the preparation of products in the design and development of traditional iron-aluminum cement clinkers, and is beneficial to the performance prediction and material design of iron-aluminum cement.

[0036] 2. The cement raw meal aging reaction system prediction model of the present application can target the established target clinker mineral composition. The iron-aluminum cement raw meal aging reaction system prediction model software developed according to the prediction model continuously inputs different raw material parameters and reaction conditions to achieve the purpose of screening raw materials and optimizing the ratio. In the process, a large amount of manpower, material resources and time loss can be avoided, thereby improving efficiency.

[0037] 3. This application puts the prediction model of cement raw meal maturation reaction system into practical use and embeds it into the database, so that the prediction model can be dynamically updated according to the changes in cement raw meal parameters and clinker parameters in the database, thereby making the output results of the prediction model of iron-aluminum cement raw meal maturation reaction system more accurate and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a structural schematic diagram of a prediction model for an iron-aluminum cement raw material aging reaction system according to a preferred embodiment of the present application.

[0039] Figure 2 It is a schematic diagram of the Transformer Encoder structure according to a preferred embodiment of the present application.

[0040] Figure 3 A scatter plot of actual values ​​and predicted values ​​is drawn using test set data according to a preferred embodiment of the present application. DETAILED DESCRIPTION

[0041] Below, the present application is further described in conjunction with specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0042] In the description of the present application, it should be noted that directional words, such as the terms "center", "lateral", "longitudinal", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", etc., indicating directions and positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of narrating the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and cannot be understood as limiting the specific scope of protection of the present application.

[0043] It should be noted that the terms "first", "second", etc. in the description and claims of the present application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0044] The terms "including" and "having" and any variations thereof in the specification and claims of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0045] The present application is further described below with reference to the accompanying drawings:

[0046] like Figures 1 to 3 As shown, the present application provides a prediction model for cement raw meal aging reaction system, which is mainly suitable for predicting the reaction process and products of the CaO-SiO2-Al2O3-Fe2O3-SO3 quinary system to obtain the mineral composition of the raw meal, or for the current clinker composition, the prediction model can be used to continuously regulate and optimize the raw meal ratio.

[0047] Among them, Figure 1 As shown, the prediction model is trained according to the following steps:

[0048] S100, obtaining known raw material parameters, reaction conditions and clinker parameters as basic data;

[0049] S200, setting possible raw material cooking reaction equations and possible products, and obtaining the participation ratio data of each reaction;

[0050] S300, standardize and normalize the basic data and the data of each reaction participation ratio;

[0051] S400, build a multi-scale residual convolutional neural network model to learn basic data and the proportion of participation in each reaction, and build a deep network structure;

[0052] S500, compile the trained model, observe and optimize the model performance.

[0053] This prediction model can combine experimental data and computational data, and through machine learning, deeply explore the CaO-SiO2-Al2O3-Fe2O3-SO3 five-component reaction system and the principle of mineral phase coexistence. Finally, based on the neural network algorithm, a prediction model for the iron-aluminum cement raw material aging reaction system is established, and it is applied to the design and development of high-performance iron-aluminum cement clinker, thereby reducing the difficulty of design and development, saving resources, and reducing the consumption of manpower, material resources and time.

[0054] In some embodiments, the raw material parameters mainly include: the percentages of main chemical components such as CaO, SiO2, Al2O3, Fe2O3, SO3, MgO, TiO2, and the percentages of trace elements such as P2O5, Cl, Cr2O3, and ZrO2.

[0055] In some embodiments, the reaction conditions mainly include firing temperature and firing time.

[0056] In some embodiments, the clinker parameters mainly include , , , the percentage of minerals such as iron and aluminum solid solution phases.

[0057] In some embodiments, the raw material cooking reaction equation and products include:

[0058]

[0059]

[0060]

[0061]

[0062]

[0063] in, a , b , c is the proportion of Fe2O3 participating in the relevant reaction, a + b + c ≤1, it can be understood that when Fe2O3 is completely reacted, a + b + c =1.

[0064] It is worth noting that the raw material cooking reaction equations and products listed above are only the more common raw material cooking reaction equations and products in the prior art. In actual application, the prediction breadth of the prediction model can be further improved by improving and adding less common raw material cooking reaction equations and trace (rare) products according to needs.

[0065] In some embodiments, before the prediction model learns the basic data and the data of each reaction participation ratio, the basic data and the data of each reaction participation ratio are arranged in the following format:

[0066] X=[x1x2……x q ], x1, x2...x q ∈R N×1

[0067] Y=[y1y2……y p ],y1,y2……y p ∈R N×1

[0068] Among them, x1, x2, ... x q Represent the raw material parameters, reaction conditions and the proportion of each reaction in each group of cement raw materials, totaling q variables, y1, y2...y p They represent the mineral phase percentage of clinker, with a total of p variables.

[0069] The meaning of the arranged equation is that X consists of q matrices x (i.e. x1x2…x q ), each matrix x is a single-column matrix with n rows, and the same is true for Y.

[0070] In some embodiments, since a large data distribution range will affect the prediction results of the prediction model, in order to ensure the accuracy and rationality of the prediction results, this embodiment uses the TensorFlow framework to standardize and normalize the arranged basic data and the data of each reaction participation ratio, so that data with different characteristics have similar scales, and then converts the preprocessed data into a tensor format to comply with the TensorFlow framework operation data structure.

[0071] In some embodiments, in order to subsequently verify and test the model, the present embodiment further needs to divide the preprocessed data into a training set, a validation set, and a test set according to m:n:1-mn, where 0.8<m<1, n<0.2.

[0072] Among them, the data in the training set is used to be put into the neural network for learning, the data in the validation set is used to be put into the neural network for verification, and the data in the test set is used to be put into the neural network for actual operation processing.

[0073] In some embodiments, the construction process of the multi-scale residual convolutional neural network model is: constructing w residual blocks, and setting up convolution layers, batch normalization layers and activation layers, defining multi-scale residual blocks to combine convolution kernels of different sizes in the convolution layer to capture multi-scale features, and using the residual blocks and multi-scale residual blocks defined above to construct a multi-scale residual convolutional neural network model, which can facilitate model training, combine convolution kernels of different scales to capture multi-scale features in input data, and use residual connections to alleviate the gradient vanishing problem of deep networks.

[0074] In some embodiments, the size of the convolution kernel may be 3×3, 5×5, 7×7, etc.

[0075] In some embodiments, the number of residual blocks w is generally greater than 10. It is understandable that the greater the number of w, the stronger and more complete the computing power of the model, but correspondingly, the higher the demand for computing power.

[0076] like Figure 2 In the illustrated embodiment, the construction process of the multi-scale residual convolutional neural network model can be implemented using a Transformer Encoder.

[0077] In some embodiments, after constructing the multi-scale residual convolutional neural network model, the multi-scale residual convolutional neural network model is compiled, and the loss function, optimizer and evaluation index optimization model are specified.

[0078] The loss function, optimizer and evaluation index optimization model work together and are three parameters defined under model.compile. These three parameters respectively determine the optimization of the learning rate of the training model algorithm, the calculation of loss and the evaluation of model performance. The loss function (loss) is used to minimize the data loss in the training process and improve the prediction accuracy of the model. The selection of its parameters determines how to calculate the loss. The main function of the optimizer is to make the gradient descend faster and better during the gradient descent, so as to find the minimum value of the objective function as soon as possible. It can be understood as how to find the optimal "loss" (loss) as quickly as possible. The evaluation index optimization model is used to evaluate the accuracy of the model's prediction. During the model training process, it is necessary to record the prediction accuracy of the model on the training set and the validation set, and then draw a curve of the accuracy rate changing with the number of training times.

[0079] In some embodiments, the loss function adopts sparse_categorical_crossentropy, which is used for multi-classification problems when the labels are integers, and has the advantage of high computational efficiency.

[0080] In some embodiments, the mean absolute error (MAE) and the mean square error (MSE) are selected as evaluation indicators, and a scatter plot of actual values ​​and predicted values ​​is drawn using the test set data for evaluation.

[0081] In some embodiments, the optimizer uses Bayesian optimization. Based on the evaluation results of the mean absolute error and the mean square error, if the model performs poorly, Bayesian optimization can be used to adjust parameters.

[0082] like Figure 3 In the illustrated embodiment, when a scatter plot of actual values ​​and predicted values ​​is drawn using the test set data for evaluation, ideally all points should be close to the red line (ie, the actual value y = predicted value x shown in the figure).

[0083] In some embodiments, developing model computing software and embedding the prediction model into the model computing software comprises:

[0084] K100, layout the software front-end interface and install the software front-end in the smart terminal;

[0085] Among them, the software front-end interface usually includes a raw material parameter and reaction condition upload interface and a clinker parameter prediction interface.

[0086] K200, develop a database to form a database of raw material parameters, reaction conditions and clinker parameters, embed the prediction model into the database as a module, and set up the developed database on the server as the software backend to connect to the front end;

[0087] In some embodiments, the database backend includes a data upload module, a data query module and a data prediction module. The data upload module conducts human-computer interaction through a data upload interface to upload and manage basic data. The data query module conducts human-computer interaction through a data query interface to query basic data information of cement clinker through raw material parameters and reaction conditions, or to query existing raw material parameters and reaction condition information through clinker parameters. The data prediction module includes a prediction model. The data prediction module conducts human-computer interaction through a data prediction interface to predict the mineral content of known clinker.

[0088] In some embodiments, the database backend is hosted on a solid-state server, which can improve the efficiency of receiving, transmitting, and processing data.

[0089] An example of a table of data upload to the database and its corresponding format is shown in Table 1.

[0090] Table 1 Correspondence between known raw material chemical composition and corresponding clinker mineral content

[0091]

[0092] K300, input basic data information in the software front-end interface, and the database determines the scope and format of the basic data information;

[0093] K400, if the judgment is no, the user will be prompted with a data error in the software front-end interface;

[0094] K500, if the judgment is yes, the basic data information is uploaded to the database, and the database returns the prediction result and outputs it in the software front-end interface.

[0095] Specifically, the model calculation software developed according to the prediction model of this application, the process of applying it and verifying the calculation prediction results is as follows:

[0096] The production data of cement enterprises, including basic data such as raw material parameters, reaction conditions and clinker parameters, are collected and input into the database. By training the prediction model, a prediction model for the raw material clinker reaction system of iron-aluminum cement is obtained.

[0097] Using limestone, low-grade bauxite and gypsum as raw materials, the main chemical components of the raw materials are shown in Table 2.

[0098] Table 2 Main chemical components of raw materials (wt%)

[0099]

[0100] The designed raw material ratio is limestone: bauxite: gypsum = 58:33:9. The relevant parameters are calculated as shown in Table 3. Enter the parameters in the upload interface of the model calculation software.

[0101] Table 3 Raw material parameters and reaction conditions

[0102]

[0103] The cement clinker performance prediction and mineral phase design model outputs the prediction results of cement clinker parameters, as shown in Table 4.

[0104] Table 4 Prediction results of clinker mineral content data

[0105]

[0106] To verify the prediction results, the raw materials are mixed and stirred evenly according to the above raw material ratio, i.e., limestone: bauxite: gypsum = 58:33:9, and 8%~15% water is added to mix together, and pressed into round cakes under a pressure of 800kN~1000kN; put into an electric furnace, heated at 10℃ / min, and the final firing temperature is 1300℃. The insulation time is 40min. After firing, it is taken out immediately, quickly cooled with an electric fan, and then ground into powder and sieved to obtain iron-aluminum cement clinker.

[0107] The mineral composition of cement clinker was obtained by XRD quantitative analysis using the Rietveld method (see Table 5).

[0108] Table 5 Clinker mineral content test data

[0109]

[0110] By comparing Table 4 and Table 5 above, the accuracy of the prediction results can be calculated, and the model prediction error value is shown in Table 6.

[0111] Table 6 Error values ​​between prediction results and test data

[0112]

[0113] The above describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited by the above embodiments, and the above embodiments and the specification only describe the principles of the present application. The present application may have various changes and improvements without departing from the spirit and scope of the present application, and these changes and improvements fall within the scope of the present application for which protection is sought. The scope of protection claimed by the present application is defined by the attached claims and their equivalents.

Claims

1. A prediction model for cement raw material aging reaction system, characterized in that: The prediction model is mainly suitable for predicting the products of the CaO-SiO2-Al2O3-Fe2O3-SO3 quinary system. The prediction model is trained according to the following steps: S100, obtaining known raw material parameters, reaction conditions and clinker parameters as basic data, wherein the raw material parameters are the percentage of chemical composition of cement raw material, the reaction conditions include firing temperature and firing time, and the clinker parameters are the percentage of mineral content of clinker; S200, set the raw material cooking reaction equation as follows: in, a , b , c is the proportion of Fe2O3 participating in the relevant reaction, a + b + c ≤1, when Fe2O3 is completely reacted, a + b + c =1, Obtain the participation rate data of each reaction; S300, arrange the basic data and the data of participation ratio of each reaction in the following format: X=[x1x2……x q ],x1,x2......x q ∈R N×1 Y=[y1y2……y p ],y1,y2……and p ∈R N×1 Among them, x1, x2, ... x q Represent the raw material parameters, reaction conditions and the proportion of each reaction in each group of cement raw materials, totaling q variables, y1, y2...y p Respectively represent the percentage of mineral phase of clinker, a total of p variables, The TensorFlow framework is used to standardize and normalize the arranged basic data and the data on the participation ratio of each reaction. Divide the preprocessed data into a training set, a validation set and a test set according to m:n:1-mn, where 0.8<m<1, n<0.2, the data in the training set is used to be put into the neural network for learning, the data in the validation set is used to be put into the neural network for verification, and the data in the test set is used to be put into the neural network for actual operation processing; S400, building a multi-scale residual convolutional neural network model to learn the training set and construct a deep network structure; S500, compile the trained model, observe and optimize the model performance.

2. A cement raw material aging reaction system prediction model according to claim 1, characterized in that: The construction process of the multi-scale residual convolutional neural network model is as follows: construct w residual blocks, set up convolutional layers, batch normalization layers and activation layers, define multi-scale residual blocks to combine convolution kernels of different sizes in the convolutional layer to capture multi-scale features, and use the residual blocks and multi-scale residual blocks defined above to construct a multi-scale residual convolutional neural network model.

3. A cement raw material aging reaction system prediction model according to claim 1, characterized in that: After building the multi-scale residual convolutional neural network model, compile the multi-scale residual convolutional neural network model and specify the loss function, optimizer, and evaluation index to optimize the model.

4. A cement raw material aging reaction system prediction model as claimed in claim 3, characterized in that: The loss function uses sparse_categorical_crossentropy, the optimizer uses Bayesian optimization, the evaluation indicators select mean absolute error and mean square error, and the test set data is used to draw a scatter plot of actual values ​​and predicted values ​​for evaluation. If the model performs poorly, Bayesian optimization is used to adjust the parameters.

5. The prediction model for cement raw material aging reaction system according to claim 1, characterized in that: Develop model calculation software and embed the prediction model into the model calculation software. The steps include: K100, layout the software front-end interface and install the software front-end in the smart terminal; K200, develops a database, embeds the prediction model into the database as a module, and sets up the developed database on the server as the software backend to connect to the front end; K300, input basic data information in the software front-end interface, and the database determines the scope and format of the basic data information; K400, if the judgment is no, the user will be prompted with a data error in the software front-end interface; K500, if the judgment is yes, the basic data information is uploaded to the database, and the database returns the prediction result and outputs it in the software front-end interface.

6. A cement raw material aging reaction system prediction model according to claim 5, characterized in that: The database includes a data upload module, a data query module and a data prediction module. The data upload module conducts human-computer dialogue through the data upload interface to upload and manage basic data. The data query module conducts human-computer dialogue through the data query interface to query basic data information of cement clinker through raw material parameters and reaction conditions, or to query existing raw material parameters and reaction condition information through clinker parameters. The data prediction module includes a prediction model. The data prediction module conducts human-computer dialogue through the data prediction interface to predict the mineral content of known clinker.

Images