Rotary kiln working condition prediction method of coupling simulation and deep learning neural network

Abstract

The invention discloses a rotary kiln working condition prediction method coupling simulation and a deep learning neural network, relates to the technical field of rotary kiln working condition prediction, and develops an efficient prediction model by simulating a rotary kiln to obtain sample data and combining a POD method and a neural network, thereby realizing rapid prediction of a flow field and a temperature field of the rotary kiln. According to the method, large expenses of traditional CFD simulation on computing resources and time are remarkably reduced, the limitation of a previous simulation technology on the aspect of real-time performance is solved, prediction data are adjusted through a correction algorithm, the practicability and reliability of the model are improved, rotary kiln order reduction and digital twinborn model deployment in an industrial site is achieved, and the method is suitable for large-scale industrial production. Guidance is provided for the production process in real time, and the production efficiency and the quality are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of rotary kiln operating condition prediction, and specifically relates to a rotary kiln operating condition prediction method of coupling simulation and deep learning neural network. Background Art

[0002] The rotary kiln is a continuously running rotating cylinder thermal equipment. Its working principle is to use the inclination and slow rotation of the kiln body to make the materials roll forward in the kiln, and at the same time exchange heat with the hot gas in the kiln, so that the internal materials can complete a series of heat treatment processes such as heating, decomposition, and calcination under high temperature conditions. In this process, the rotation of the rotary kiln not only promotes the full contact between the material and the hot gas, but also contributes to the physical and chemical reactions inside the material, such as water evaporation, organic matter decomposition, and mineral reorganization, thereby achieving efficient and uniform heat treatment effects. It is an indispensable key equipment in cement production, metallurgy, chemical industry and other industries.

[0003] However, in the actual production process, the rotary kiln is a key equipment. Unstable control level may cause "under-burning" or "over-burning" of clinker during calcination, and cause problems such as rotary kiln ringing and local temperature abnormality, which affect the normal flow of materials and heat exchange efficiency, affect the quality of clinker, and bring challenges to quality management and process adjustment. In addition, the internal temperature of the rotary kiln is extremely high, with the highest temperature reaching over 1800°C, and the gas contains a large amount of dust. It is difficult to directly measure the material flow and temperature distribution at high temperature, which increases the difficulty of precise control. It relies on the operator's manual experience and lacks quantitative indicators to reflect internal laws. There are problems of subjectivity and response lag.

[0004] Computational fluid dynamics (CFD) simulation technology plays a vital role in modern engineering design and scientific research. By using CFD technology to establish a rotary kiln simulation model and conduct simulation analysis, the overall working conditions of the equipment can be predicted in order to optimize the operating parameters and improve the production efficiency of the rotary kiln. However, when dealing with large-scale, high-precision simulation tasks involving flow and heat transfer such as rotary kilns, the computational cost required for CFD simulation is high, which limits its application in actual engineering problems. At the same time, due to the complex working conditions, it is difficult to achieve real-time simulation and cannot quickly provide optimization suggestions or decision support. At present, CFD simulation is mainly used in the scientific research and development stage or in the process of problem diagnosis, but it is difficult to achieve real-time calculation to guide production practice.

[0005] Based on the CFD simulation model, a large amount of data can be collected to analyze and monitor the operating status of the rotary kiln. However, due to the high computational complexity of the CFD model, in order to improve the calculation speed of the numerical simulation and realize the real-time prediction of the full flow field data, the rotary kiln simulation model must be reduced in order to establish a digital twin model based on the rotary kiln simulation model that takes into account both accuracy and computational efficiency. How to use the features obtained by POD intrinsic orthogonal decomposition to accurately predict and represent the temperature field under specified working conditions is a key technical issue in realizing the reduction of the rotary kiln model and digital twin, and obtain the optimal setting value through real-time prediction to ensure that the rotary kiln is always in the best state.

[0006] So how to combine CFD simulation with deep learning to quickly obtain the operating parameters of the rotary kiln, provide optimization suggestions or decision support, and make intelligent and dynamic adjustments to the rotary kiln to ensure the stable operation of the rotary kiln and improve the pulverized coal combustion efficiency and production efficiency, thereby saving energy and reducing consumption? This is a problem, based on which a solution is provided. Summary of the invention

[0007] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a rotary kiln operating condition prediction method of coupling simulation and deep learning neural network.

[0008] A rotary kiln operating condition prediction method based on coupled simulation and deep learning neural network, the method specifically comprises the following steps:

[0009] Step 1: The numerical simulation module is based on the Euler two-phase flow basic model, and couples the heat transfer model with the chemical reaction model to establish a transient simulation model of the rotary kiln, and verifies the results and adjusts the model; determines the boundary and operating parameters of the rotary kiln, and simulates dozens of different operating conditions through CFD software to obtain sample data;

[0010] Step 2: Use intrinsic orthogonal decomposition to process the simulation sample data, so as to extract the fundamental modes and key eigenvalues ​​of the rotary kiln under various working conditions; establish a multi-layer neural network model from working condition parameters to fundamental mode coefficients through back propagation neural network, and train a multi-input and output mapping model between the working condition parameter space and the fundamental mode eigenvalue coefficients of the flow field to achieve a better fitting effect;

[0011] Step 3: After configuring a small amount of open source Python libraries, the model is deployed on the factory industrial control equipment. The BPNN model is used to quickly predict the flow field and temperature field of the rotary kiln. Sensor detection is performed at local points in the rotary kiln to obtain local observation data. The data from the local observation points is combined with the full-field prediction data. The prediction data is updated for the entire field using a filtering correction algorithm. The accuracy of the flow field prediction value is effectively improved while retaining the true characteristics of the flow field, making the prediction data closer to the data of the real physical field.

[0012] Compared with the prior art, the present invention has the following beneficial effects:

[0013] The present invention simulates the rotary kiln to obtain sample data, combines the POD method with the neural network, and develops an efficient prediction model, thereby realizing the rapid prediction of the flow field and temperature field of the rotary kiln, significantly reducing the huge overhead of traditional CFD simulation in computing resources and time, solving the limitations of previous simulation technology in real-time, and adjusting the predicted data through the correction algorithm, thereby improving the practicability and reliability of the model, realizing the deployment of the rotary kiln reduction and digital twin model to the industrial site, providing real-time guidance for the production process, and ensuring the improvement of production efficiency and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a flow chart of the prediction method of the present invention;

[0015] Figure 2 The neural network structure of the BPNN neural network of the present invention;

[0016] Figure 3 These are the specific steps for model training in the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] See also Figure 1-Figure 3The core of this application is to build an approximate model from working condition variables to flow field data based on neural network as a proxy model, and realize the establishment of a data-driven rotary kiln reduced-order model and digital twin method. This process only requires the use of CFD software to simulate dozens of different working conditions, and then use the intrinsic orthogonal decomposition to process the simulation sample data, so as to extract the basic mode and its key eigenvalues ​​of the rotary kiln under various working conditions, and establish a multi-layer neural network model from working condition parameters to basic mode coefficients through the back propagation neural network (Back Propagation based Neura l Network, BPNN), so as to achieve rapid prediction of the rotary kiln flow field, solve the problem of high cost and long time of CFD simulation calculation, solve the problem of high cost and long time of traditional CFD simulation calculation, and also calibrate the difference between simulation prediction data and actual sensor data by combining the correction algorithm, ensure the accuracy of the model, realize the actual deployment of reduced-order model and digital twin model in industrial site, and provide real-time guidance and support for optimizing production processes.

[0019] The present application provides a rotary kiln operating condition prediction method using coupled simulation and deep learning neural network, which specifically includes the following steps:

[0020] S1: CFD simulation of rotary kiln;

[0021] The numerical simulation module is based on the basic Euler two-phase flow model, and couples the heat transfer model with the chemical reaction model to establish a transient simulation model of the rotary kiln, and performs simulation result verification and model adjustment.

[0022] The Euler two-fluid model has three phases. The first phase is the gas phase, which is the main phase. It contains various gas components: O2, CO2, N2, NO, CO, H2O(g), HCN, CH4, C2H4, C2H6, H2, a total of 11 gas components; the second phase is coal powder solid particles, which contain four components: fixed carbon, volatile matter, ash, and H2O(l); the third phase is the burning material particles, which contain six components: CaCO3, CAO, C2S, C3S, C3A, and C4AF.

[0023] For rotary kiln simulation, the first is the simulation of coal powder combustion, which determines the overall temperature field of the rotary kiln; further, the multiphase flow transmission and heat transfer simulation studies the movement law and heat exchange law of the combustion material particles inside the rotary kiln; the simulation model of combustion material calcination studies the evolution law of the internal mineral composition of the combustion material under the material transmission / heat transfer / mass transfer behavior. Finally, during the calcination of the combustion material, the particles will gradually grow, which in turn affects the movement law and heat exchange law of the combustion material particles. The above parts are interrelated and influence each other. The simulation model fully considers the structural characteristics, material characteristics and other factors of the rotary kiln, and better predicts the combustion, flow and heat exchange conditions in the rotary kiln, reflecting its velocity field, temperature field, three-phase gas-solid component field data and other internal laws. At the same time, the simulation results are verified based on the actual production data of the plant. By comparing the simulation results with the actual production data, the simulation model is adjusted to verify the accuracy of the simulation parameters and model construction to improve the accuracy and reliability of the model;

[0024] Euler-Euler two-phase flow continuity equation

[0025]

[0026] Where: α q is the volume fraction of the q phase, is the q-phase velocity, is the mass flow rate transferred from phase p to phase q, is the mass flow rate transferred from phase q to phase p, S q is the source term.

[0027] The momentum control equation is:

[0028]

[0029] Where:

[0030]

[0031] where μ q and λ q The q phase is the shear viscosity and the volume viscosity, F q is the external force, F lift,q It is lift. is the turbulent dissipation force, Virtual mass force, is the interaction force between the phases, is the phase speed.

[0032] The energy governing equation is:

[0033]

[0034] where k eff,qis the thermal conductivity, S q is the source term, Q pq is the heat exchange between phase p and phase q, h q is the enthalpy of phase q, p op is the reference atmospheric pressure and p is the gauge pressure.

[0035] S2: Set multi-parameter input to obtain sample space;

[0036] According to the actual adjustable process parameters, such as coal feed rate, primary air volume and temperature, secondary air volume and temperature, etc., the established rotary kiln simulation model is used to carry out multi-parameter simulation calculations. In order to avoid the exponential growth of sample number with the increase of parameters, orthogonal experimental design is adopted to balance the sample space accuracy and calculation time to obtain different sample spaces and data.

[0037] S3: eigen-orthogonal decomposition of sample data;

[0038] The simulation sample data is processed by using the intrinsic orthogonal decomposition method based on the sample variance maximization theory, so as to extract the fundamental modes and key eigenvalues ​​of the rotary kiln under various working conditions.

[0039] Intrinsic orthogonal decomposition is a method of vector data statistical analysis that can reduce the order of high-dimensional flow field data and map it to a low-dimensional orthogonal fundamental modal space, thereby analyzing the main characteristics of the flow field and its corresponding fundamental modal coefficients. First, the sample (flow field data) calculated by simulation needs to be standardized. Let the original data be x i , where i = 1, 2, 3, ..., r, sample data x i is an n-dimensional vector (n is related to the number of grids divided during CFD calculations), and r is the number of samples.

[0040]

[0041]

[0042] Data standardization helps to eliminate the differences between sample data features and avoid the influence of individual discrete values ​​on the analysis results. From this, the covariance matrix of standardized data can be obtained:

[0043]

[0044] By solving the eigenvalues ​​of the nXn-order covariance matrix, the first m-order eigenvalues ​​can be recorded as λ 1 , 2 , …, λ m , the corresponding fundamental mode eigenvector can be recorded as ξ 1 , 2 ,…,ξ mThe selection of m value is determined by the ratio of the variance value of different basic modes to the total variance value, ensuring that the characteristic components contained in the basic mode account for more than 95% of the entire sample space. Then the original sample data can be approximately expressed as X = U λ ξ, so that the fundamental mode coefficient matrix and a small number of POD fundamental modes can be used to represent most of the information of the original sample.

[0045] S4: training BPNN neural network;

[0046] Build a neural network and train a multi-input-output mapping model between the working condition parameter space and the basic modal eigenvalue coefficients of the flow field. The specific training method is as follows:

[0047] 1) Divide the samples into training set and validation set for training and result verification;

[0048] 2) Use orthogonal initialization of weight matrix. Weight initialization affects the training effect and convergence speed of neural network. By initializing the weight matrix as an orthogonal matrix, the information transmission is maintained and the gradient vanishing problem in training is reduced. The weight matrix W is orthogonal, that is, W T W = I;

[0049] 3) Calculate the input and output of the first hidden layer, and use the "Sigmoid" activation function to increase the nonlinear expression ability of the model, smooth the gradient, and avoid jumping output values;

[0050] 4) Calculate the input and output of the second hidden layer, and use the "Leaky Re lu" activation function to accelerate training convergence, which solves the problem of gradient disappearance during back propagation, that is, if a negative number is input, the gradient is completely zero, and also avoids the problem of neuron saturation;

[0051] 5) Calculate the reverse error of each layer and update the weight matrix;

[0052] 6) Calculate the prediction error of the validation set by comparing the predicted value with the validation value and using the average relative error: Among them A i Indicates the verification value, B i Represent the predicted value, ensure that the average relative error of the predicted set to the validation set in the sample space does not exceed 3%, evaluate the reasonable training rounds, and ensure that the model has sufficient calculation accuracy;

[0053] S5: Model deployment and flow field prediction;

[0054] The reduced-order model generated by this method is relatively small in size, fast in calculation, low in memory usage, and does not rely on foreign CFD software environments. It can be deployed on factory industrial control equipment after configuring a small amount of open source Python libraries. For a new set of factory actual parameter combinations Zj , the fundamental mode eigenvalue coefficient can be quickly predicted through the BPNN model Then, based on POD, reverse reconstruction is performed to obtain the predicted flow field value. Achieve millisecond-level input and output;

[0055] S6: sensor deployment and prediction data correction;

[0056] In the actual production process of the factory, due to the interweaving of various complex factors, there is inevitably a deviation between the predicted data and the actual situation on site. By performing sensor detection at local points of the rotary kiln, local observation data is obtained, and the data of the local observation point is combined with the full-field prediction data. The prediction data is updated in the whole field using a distance-based correction and a Gaussian function-based filtering correction algorithm. Assuming that n sensors are arranged, the actual measured value of each sensor is T i (i=1,2,3…n), the predicted flow field value at the same position is Then for any data point x in the predicted flow field k , and its corrected value is:

[0057]

[0058] Among them, d i For data point x k The distance from each sensor position; σ is the correction coefficient, which is related to the radius of the rotary kiln and is obtained by fitting the sensor data of different rotary kilns through the nonlinear least squares method. This algorithm systematically considers the correction of the predicted data by all sensor measurement points, and can effectively improve the accuracy of the flow field prediction value on the basis of retaining the real characteristics of the flow field, making the predicted data closer to the data of the real physical field. After the measured data is corrected, the predicted flow field and temperature field data of the rotary kiln are obtained, and the internal flow field of the rotary kiln is mastered. These data are used as an important basis for optimizing decision support and adjusting operating conditions to ensure that the rotary kiln operates under the optimal conditions.

[0059] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A rotary kiln operating condition prediction method based on coupled simulation and deep learning neural network, characterized in that: The method specifically comprises the following steps: Step 1: The numerical simulation module is based on the Euler two-phase flow basic model, and couples the heat transfer model with the chemical reaction model to establish a transient simulation model of the rotary kiln, and verifies the results and adjusts the model; Determine the boundary and operating parameters of the rotary kiln, simulate dozens of different operating conditions through CFD software, and obtain sample data; Step 2: Use intrinsic orthogonal decomposition to process batches of simulation sample data, so as to extract the fundamental modes and key eigenvalues ​​of the rotary kiln under various working conditions; Build and train the BPNN back propagation neural network, establish a multi-layer neural network model from operating parameters to fundamental modal eigenvalue coefficients, and train a multi-input and output mapping model between the operating parameter space and the fundamental modal eigenvalue coefficients of the flow field; Step 3: After configuring a small amount of open source Python libraries, the model is deployed on the factory industrial control equipment. The BPNN model is used to quickly predict the flow field and temperature field of the rotary kiln. Sensor detection is performed at local points in the rotary kiln to obtain local observation data. The data from the local observation points is combined with the full-field prediction data, and the prediction data is updated throughout the field using a filtering correction algorithm.

2. The rotary kiln operating condition prediction method of coupled simulation and deep learning neural network according to claim 1 is characterized in that: The numerical simulation module is based on the basic Euler two-phase flow model and couples the heat transfer model with the chemical reaction model to establish a transient simulation model of the rotary kiln. The Euler two-fluid model has three phases, the first phase is the main phase; the first phase is the gas phase, which contains various gas components: O2, CO2, N2, NO, CO, H2O(g), HCN, CH4, C2H4, C2H6, H2, a total of 11 gas components; the second phase is coal powder solid particles, which contain four components: fixed carbon, volatile matter, ash, and H2O(l); the third phase is the burning material particles, which contain six components: CaCO3, CAO, C2S, C3S, C3A, and C4AF.

3. The rotary kiln operating condition prediction method of coupling simulation and deep learning neural network according to claim 1 is characterized in that: The rotary kiln simulation also includes the following sub-models: The pulverized coal combustion simulation model is used to simulate the entire process of pulverized coal combustion and determine the overall temperature field of the rotary kiln; Multiphase flow transmission and heat transfer simulation model, used to simulate the movement law and heat exchange law of burning material particles inside the rotary kiln; Clinker calcination simulation model is used to simulate the evolution of mineral components inside the burning material under material transmission, heat transfer and mass transfer behavior; The particle growth model is used to simulate the influence of particle growth on the movement law and heat exchange law of the calcination material particles during the calcination process.

4. The rotary kiln operating condition prediction method of coupling simulation and deep learning neural network according to claim 1 is characterized in that: The specific method of training the multi-input-output mapping model is: 1) Divide the samples into training set and validation set for training and result verification; 2) Use orthogonal initialization of weight matrix. Weight initialization affects the training effect and convergence speed of neural network. By initializing the weight matrix as an orthogonal matrix, the information transmission is maintained and the gradient vanishing problem in training is reduced. The weight matrix W is orthogonal, that is, W T W = I; 3) Calculate the input and output of the first hidden layer, and use the "Sigmoid" activation function to increase the nonlinear expression ability of the model, smooth the gradient, and avoid jumping output values; 4) Calculate the input and output of the second hidden layer, and use the "Leaky Relu" activation function to accelerate training convergence; 5) Calculate the reverse error of each layer and update the weight matrix; 6) Calculate the prediction error of the validation set by comparing the predicted value with the validation value and using the average relative error: Among them A i Indicates the verification value, B i Represents the predicted value, ensuring that the average relative error of the prediction set to the validation set in the sample space does not exceed 3%, and evaluating reasonable training rounds until the model has sufficient computational accuracy.

5. The rotary kiln operating condition prediction method of coupled simulation and deep learning neural network according to claim 1 is characterized in that: For a new set of factory actual parameter combinations Z j , the fundamental mode eigenvalue coefficient can be quickly predicted through the BPNN model Then, based on POD, reverse reconstruction can be performed to obtain the predicted flow field value.

6. The rotary kiln operating condition prediction method of coupled simulation and deep learning neural network according to claim 1 is characterized in that: The predicted data is corrected based on the data from the observation points where the factory sensors are deployed. The specific method is as follows: By performing sensor detection at local points of the rotary kiln, local observation data is obtained, the data of the local observation points are combined with the full-field prediction data, and the prediction data is updated in the full field using distance-based correction and Gaussian function-based filtering correction algorithms; Assume that n sensors are deployed and the measured value of each sensor is T i (i=1,2,3…n), the predicted flow field value at the same position is Then for any data point x in the predicted flow field k , and its corrected value is: Among them, d i For data point x k The distance from each sensor position; σ is the correction coefficient, which is related to the radius of the rotary kiln and is obtained by fitting the sensor data of different rotary kilns through the nonlinear least squares method.

Images