Multi-stage preheater working condition prediction method based on intrinsic orthogonal decomposition and neural network

Abstract

The invention discloses a multi-stage preheater working condition prediction method based on intrinsic orthogonal decomposition and a neural network, and belongs to the technical field of multi-stage preheater digital twinning. According to the method, through the intrinsic orthogonal decomposition and back propagation neural network technology, a reduced-order model for rapidly predicting the smooth working condition of a multi-stage preheater is established; and a filtering correction algorithm is coupled to correct a prediction result through a sensor measured value, so that the efficiency of flow field solving calculation is greatly improved, the time cost is reduced, meanwhile, the method has considerable generalization ability, can be actually deployed in a cement factory, and is convenient for field personnel to determine the flow field condition of the preheater according to the flow field condition of the preheater. And working condition parameters are adjusted and optimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-stage preheater digital twin, and specifically, relates to a multi-stage preheater operating condition prediction method based on intrinsic orthogonal decomposition and neural network. Background Art

[0002] CFD (Computational Fluid Dynamics) simulation is currently widely used in modern engineering design and scientific research, such as aerospace, automotive design, biomedicine, environmental engineering and other fields. Through high-precision numerical simulation, CFD can help researchers deeply understand the complex behavior of fluids, optimize product design, reduce the number of experiments, and thus accelerate the research and development process. However, due to the complexity of fluid dynamics itself and the huge number of grids divided to achieve simulation accuracy, CFD simulation requires the help of high-performance computers to complete. This not only means that the investment cost of hardware equipment is high, but also the long calculation time also significantly increases the overall cost of the project. Especially when dealing with large-scale, high-precision and complex turbulent characteristics simulation tasks such as multi-stage preheaters, the CFD simulation calculation time may be as long as several days or even weeks, and it is impossible to quickly calculate and respond to the complex and changeable working conditions of the actual project.

[0003] In order to improve the calculation speed of numerical simulation and realize the real-time prediction of the whole flow field data, it is necessary to reduce the order of the multi-stage preheater simulation model and establish a digital twin model with sufficient accuracy and computational efficiency. The POD (Properorthogonal decomposition) method can capture the characteristics of the flow field and greatly reduce the complexity of the flow field. By combining POD and RBF (radial basis function), Kriging and other proxy models, the prediction of steady and unsteady flow fields can be realized, but these proxy models are difficult to handle multiple input and output parameters and have low modeling efficiency; the method of directly establishing a reduced-order model based on data-driven deep learning of flow field data has the disadvantages of large consumption of computing resources, complex modeling and poor generalization ability. How to use the characteristics obtained by POD analysis to accurately predict and represent the flow field under specified conditions is a key technical issue in realizing the reduction of the multi-stage preheater model and digital twin. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-stage preheater operating condition prediction method based on intrinsic orthogonal decomposition and neural network, which can achieve real-time and rapid response of multi-stage preheater operating condition data, greatly improve the calculation speed and reduce time cost.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] The multi-stage preheater operating condition prediction method based on intrinsic orthogonal decomposition and neural network includes the following steps:

[0007] SS1. Numerical simulation: Use the Euler multiphase flow model to simulate the gas-solid two-phase flow behavior of the multi-stage preheater, and compare it with engineering data to verify the accuracy;

[0008] SS2. Obtain sample space: Consider the flow field law of the preheater under different operating parameters as the sample space;

[0009] The operating parameters include the inclination angle of the spreading box, the system wind speed and the particle size distribution;

[0010] SS3, eigenvalue decomposition: Through the POD orthogonal decomposition method, based on the sample variance maximization theory, all the basic mode flow fields and corresponding eigenvalues ​​of the multi-stage preheater in the entire working space are obtained;

[0011] SS4. Training neural network: 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;

[0012] SS5, model deployment and flow field prediction;

[0013] SS6, sensor deployment and data correction: By performing sensor detection at local points in the preheater, local observation data is obtained, and the predicted data is updated throughout the field based on the filtering correction algorithm.

[0014] Beneficial effects of the present invention:

[0015] 1. The present invention establishes a reduced-order model for quickly predicting the smooth working condition of a multi-stage preheater through intrinsic orthogonal decomposition and back-propagation neural network technology, and couples the filtering correction algorithm to correct the prediction results through the actual measured values ​​of the sensor. This not only greatly improves the efficiency of flow field solution calculation and reduces the time cost, but also has considerable generalization ability. It can be actually deployed in cement plants, which is convenient for on-site personnel to adjust and optimize the working parameters according to the flow field conditions of the preheater.

[0016] 2. The present invention can achieve fast prediction and millisecond-level response: Different from the traditional CFD simulation which requires the calculation of a large number of mathematical and physical equations, the POD-BPNN model greatly improves the efficiency of flow field calculation by reducing the order of the model, and can achieve millisecond-level working condition parameter input - flow field data output;

[0017] 3. The present invention can achieve accurate prediction under complex working parameters: POD can map high-dimensional flow field data to low-dimensional orthogonal basis mode space, effectively remove redundant features of data, thereby effectively improving the generalization ability of the model and having the ability to accurately predict under complex input parameters;

[0018] 4. The training and prediction of the model of the present invention are completely data-driven, with no migration cost. At the same time, the model calculation cost is low, which is convenient for deployment on the industrial control platform, providing the necessary conditions for system-level digital twins;

[0019] 5. The present invention corrects the algorithm based on the measured data: through the filtering correction algorithm, the full-field prediction data is updated with local data, so that the prediction data is closer to the real physical field. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention will be further described below in conjunction with the accompanying drawings.

[0021] Figure 1 It is a simulation cloud diagram of a multi-stage preheater;

[0022] Figure 2 This is the BPNN neural network architecture diagram;

[0023] Figure 3 It is the flow chart of POD-BPNN preheater flow field prediction;

[0024] Figure 4 It is a prediction value filtering correction algorithm. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not 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.

[0026] The core of the present invention is to use neural network as a proxy model to build an approximate model from operating variables to flow field data, and realize the establishment of a data-driven multi-stage preheater reduced-order model. Through this method, we only need to use CFD software to simulate the steady-state flow field of the multi-stage preheater under dozens of different operating conditions, and then perform intrinsic orthogonal decomposition of the flow field data to obtain all the fundamental modal flow fields and corresponding eigenvalues ​​of the multi-stage preheater in the entire operating space, and establish a multi-layer neural network model from operating parameters to fundamental modal coefficients through the back propagation neural network (BPNN) to achieve rapid prediction of the flow field.

[0027] Specifically, Figure 3 As shown, the multi-stage preheater operating condition prediction method based on intrinsic orthogonal decomposition and neural network of the present invention comprises the following steps:

[0028] SS1, multi-stage preheater CFD simulation

[0029] like Figure 1 As shown in the figure, the cyclone separator is an important gas-solid separation equipment in the cement industry, in which the airflow is turbulent and exhibits strong non-uniformity and anisotropy, resulting in an extremely complex flow process. The multi-stage preheater is a series of multiple cyclones, and the upper cyclone and the lower cyclone are connected by a weight valve, presenting a special boundary that only transmits materials but not gas, and the simulation difficulty is relatively higher. This method adopts the Euler multiphase flow model in FLUENT and considers the particle motion behavior as a fluid. The key point of this method is to describe the viscosity and pressure of the solid phase motion, which is generally modeled by introducing the Kinetic Theory of Granular Flow (KTGF) theory. This method only requires a set of computational grids, each of which is approximately several or dozens of times the particle size, which can greatly reduce the amount of calculation, better predict the gas-solid flow in larger-scale industrial equipment, and reflect its internal flow field laws.

[0030] SS2. Set multi-parameter input to obtain sample space

[0031] By using the Design point function in ANSYS, the 6-stage preheater simulation models with different input parameters were submitted in batches to explore the effects of different parameters such as the inclination angle of the spreading box, system wind speed, and particle size on the flow field of the preheater system.

[0032] SS3, Proper orthogonal decomposition of sample data

[0033] Intrinsic orthogonal decomposition or principal component analysis is a method of statistical analysis of vector data. It can reduce the order of high-dimensional flow field data and map it to a low-dimensional orthogonal basis mode space, thereby analyzing the main characteristics of the flow field and its corresponding basis mode coefficients. In essence, it maximizes the sample variance in each dimension after reducing the flow field to a low dimension.

[0034] 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.

[0035]

[0036] Formula (1) is to obtain the mean of the sample, and formula (2) is to subtract the mean from the original data. Data standardization is conducive to eliminating the differences between sample data features and avoiding the influence of individual discrete values ​​on the analysis results.

[0037] From this we can find the covariance matrix of the standardized data:

[0038]

[0039] 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 m The selection of m value is determined according to the proportion of variance values ​​of different fundamental modes to the total variance value, ensuring that the characteristic components contained in the fundamental mode account for more than 95% of the entire sample space.

[0040] Then the original sample data can be approximately expressed as:

[0041] X=U λ ξ (4)

[0042] In this way, 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.

[0043] SS4. Constructing a neural network (BPNN) model

[0044] The reduced-order eigenvalue matrix corresponding to the preheater flow field under different input parameters cannot be estimated artificially. The back propagation neural network with multiple input parameters and multiple output results has the characteristics of strong nonlinear mapping ability, self-learning weights and good generalization ability, and is suitable as a proxy model to train the mapping relationship between input parameters and POD basis coefficient matrix.

[0045] The sample data obtained from CFD simulation of multi-stage preheater is relatively small (generally dozens or hundreds of groups, which are small sample data). The shallow neural network has sufficient nonlinear mapping and generalization capabilities. The BPNN model with two hidden layers can be selected to take into account stability and feature description capabilities. The neural network structure is as follows: Figure 2 As shown:

[0046] The neural network establishes a mapping from r-dimensional input to m-dimensional output, where the input is x = [x 1 ,x 2 ,…,x k ] T , k corresponds to the number of input parameters of the training sample, the number of neurons in the hidden layer are p and q respectively, and the output is y = [y 1 ,y 2 ,…,y m ] T , m corresponds to the number of POD basic modes after dimensionality reduction. 1 ,ω2 ,ω 3 are the weight matrices between layers, and their sizes are p×k, qxp, and m×q respectively.

[0047] The BPNN model updates the weight matrix by back-propagating the error, and uses different activation functions to increase the nonlinear mapping ability and stability of the model, thereby completing the training of the model. The specific training steps are as follows:

[0048] 1) Divide the samples into training set and validation set

[0049] 2) Use normal distribution to initialize the weight matrix to avoid gradient vanishing or gradient exploding in the early stage of training;

[0050] 3) Calculate the input and output of the first hidden layer, and use the "Swish" activation function to increase the nonlinear expression ability of the model;

[0051] 4) Calculate the input and output of the second hidden layer, and use the "Leaky Relu" activation function to accelerate training convergence while avoiding the problems of gradient vanishing and neuron saturation;

[0052] 5) Calculate the reverse error of each layer and update the weight matrix;

[0053] 6) Calculate the prediction error of the validation set to evaluate the appropriate number of training rounds until the model has sufficient computational accuracy and generalization ability.

[0054] SS5, model deployment and flow field prediction

[0055] 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. j , 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;

[0056] SS6, sensor deployment and prediction data correction

[0057] The actual production conditions in factories vary greatly, and it is inevitable that the predicted data will deviate from the actual values ​​on site. Therefore, a filtering algorithm is used to optimally estimate the system state by inputting and outputting observed data and predicted data, making the predicted data closer to the real physical field.

[0058] like Figure 4As shown, this method adopts a filtering algorithm based on distance correction: Assume that n sensors are arranged in the factory, 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:

[0059]

[0060] Among them, D i For data point x k The distance from each sensor position, R is the maximum radius of the cyclone of each stage of the preheater. The algorithm systematically considers the correction of the predicted data by all measurement points, and can effectively improve the accuracy of the flow field prediction value while retaining the real characteristics of the flow field, and fit the actual working conditions on site.

[0061] After calculating the corrected predicted value, on-site personnel can evaluate the overall situation of the preheater in real time, and adjust process parameters in a targeted manner based on experience to optimize resource utilization.

[0062] The present invention simulates the flow field of a multi-stage preheater through CFD technology, obtains a large amount of sample data, and then combines the intrinsic orthogonal decomposition method with a multi-layer back propagation neural network to achieve rapid prediction of the flow field of the multi-stage preheater, solving the problems of high cost and long time of CFD simulation calculation. At the same time, the coupling filter correction algorithm is used to connect the predicted data with the actual sensor detection value, solving the problem of errors between the simulation prediction value and the on-site detection value, and realizing the actual deployment of the reduced-order model of the multi-stage preheater in the factory and guiding the optimized production.

[0063] The above contents are merely examples and explanations of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the invention or exceed the scope defined by the claims, they shall all fall within the protection scope of the present invention.

Claims

1. A multi-stage preheater operating condition prediction method based on intrinsic orthogonal decomposition and neural network, characterized in that: The steps include: SS1. Numerical simulation: simulate the gas-solid two-phase flow behavior of the multi-stage preheater and compare with engineering data to verify the accuracy; SS2. Obtain sample space: Consider the flow field law of the preheater under different operating parameters as the sample space; The operating parameters include the inclination angle of the spreading box, the system wind speed and the particle size distribution; SS3, eigenvalue decomposition: Through the POD orthogonal decomposition method, based on the sample variance maximization theory, all the basic mode flow fields and corresponding eigenvalues ​​of the multi-stage preheater in the entire working space are obtained; SS4. Training neural network: 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; SS5, model deployment and flow field prediction; SS6, sensor deployment and data correction: By performing sensor detection at local points in the preheater, local observation data is obtained, and the predicted data is updated throughout the field based on the filtering correction algorithm.

2. The multi-stage preheater operating condition prediction method based on intrinsic orthogonal decomposition and neural network according to claim 1 is characterized in that: Step SS3 includes: Standardize the flow field data samples calculated by simulation; The covariance matrix of the standardized data can be found: To find the sample mean, In order to subtract the mean from the original data, the eigenvalues ​​of the nXn-order covariance matrix are solved, and 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 m ; The selection of m value is determined according to the proportion of variance values ​​of different fundamental modes to the total variance value, ensuring that the characteristic components contained in the fundamental 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.

3. The multi-stage preheater operating condition prediction method based on intrinsic orthogonal decomposition and neural network according to claim 1 is characterized in that: The neural network in step SS4 uses a BPNN model with two hidden layers to establish a mapping from r-dimensional input to m-dimensional output, where the input is x=[x1, x2,…, x k ] T , k corresponds to the number of input parameters of the training sample, the number of neurons in the hidden layer are p and q respectively, and the output is y = [y1, y2, ..., y m ] T , m corresponds to the number of POD basic modes after dimensionality reduction; ω 1 ,ω 2 ,ω 3 are the weight matrices between layers; their sizes are p×k, qxp, and m×q respectively.

4. The multi-stage preheater operating condition prediction method based on intrinsic orthogonal decomposition and neural network according to claim 1 is characterized in that: The specific training steps of the neural network in step SS4 are as follows: Divide the samples into training set and validation set; Initialize the weight matrix using normal distribution; Calculate the input and output of the first hidden layer, and use the "Swish" activation function to increase the nonlinear expression ability of the model; Calculate the input and output of the second hidden layer, and use the "Leaky Relu" activation function to accelerate training convergence; Calculate the reverse error of each layer and update the weight matrix; Calculate the prediction error of the validation set to evaluate the appropriate number of training rounds until the model has sufficient computational accuracy and generalization ability.

5. The multi-stage preheater operating condition prediction method based on intrinsic orthogonal decomposition and neural network according to claim 1 is characterized in that: In step SS5, for a new set of plant actual parameter combinations Z j , the fundamental mode eigenvalue coefficient is predicted by BPNN model Then, based on POD, reverse reconstruction is performed to obtain the predicted flow field value.

6. The method for predicting the operating conditions of a multi-stage preheater based on intrinsic orthogonal decomposition and neural network according to claim 1 is characterized in that: In step SS6, a distance-based correction filtering algorithm is used: Assume that n sensors are arranged in the factory, 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 location, R is the maximum radius of the cyclone at each stage of the preheater.

Images