Decomposition furnace dynamic process adjustment method fusing sensor data and numerical prediction

Abstract

The invention discloses a decomposing furnace dynamic process adjustment method fusing sensor data and numerical prediction, relates to the technical field of decomposing furnace dynamic process adjustment, and intelligently and dynamically adjusts process parameters of a decomposing furnace through combination of an online monitoring technology, a CFD simulation technology and a digital twinning technology. Stable operation of the decomposing furnace is guaranteed, material distribution uniformity, pulverized coal combustion efficiency and production efficiency of the decomposing furnace are improved, energy is saved, consumption is reduced, and the process control level is optimized; the invention provides a set of complete dynamic process adjustment method for the decomposing furnace, the advantage of accurate simulation of fluid-solid movement and complex chemical reaction in the decomposing furnace by a CFD fluid simulation technology and the advantage of rapid response of a digital twinning technology are integrated, and online monitoring sensor data are accessed for correction; rapid calculation of the flow field of the decomposition furnace and intelligent dynamic adjustment of process parameters are realized.

Description

Technical Field

[0001] The invention belongs to the technical field of dynamic process adjustment of a decomposition furnace, and in particular relates to a dynamic process adjustment method of a decomposition furnace by integrating sensor data and numerical prediction. Background Art

[0002] The decomposition furnace, that is, the decomposition technology outside the kiln, has been widely used in the cement production industry since it appeared in the 1970s as a technology that can greatly reduce the length and heat consumption of the rotary kiln and significantly increase the output. The preheating and pre-decomposition rotary kiln calcination system composed of a suspended preheater and a decomposition furnace is different from the suspended preheater kiln, also known as the new preheater kiln, referred to as the NSP kiln. The NSP kiln has the advantages of high output per unit volume, large single-machine production capacity, long service life of the kiln lining bricks, small equipment footprint, low equipment cost, and low environmental pollution. The principle is that the fuel and raw material powder in the decomposition furnace are both in a suspended state, combustion and heat exchange are carried out simultaneously, and the contact surface between the gas and solid phases is thousands of times larger than that in the rotary kiln. Therefore, the combustion, heat exchange and reaction efficiency in the decomposition furnace have significant advantages. However, in the actual production process, due to the limitations of sensor monitoring capabilities, complex and changeable working conditions, and factors such as the decomposition furnace, it is easy to cause problems such as crusting and stacking of decomposition furnaces, local temperature abnormalities, and large fluctuations in clinker quality, which brings challenges to the company's process adjustment and quality management. At present, the management of decomposition furnaces mainly relies on manual control and regular shutdown and maintenance. This method is relatively inefficient and difficult to accurately and timely detect potential problems of the equipment. At the same time, it relies on manual experience and lacks quantitative indicators to reflect internal laws. Therefore, how to achieve dynamic adjustment and optimization of decomposition furnace process parameters to improve the operating efficiency and reliability of the equipment and increase product output and quality has become a technical problem that needs to be solved urgently in this field.

[0003] The patent with publication number CN113589693A discloses a predictive control method for the outlet temperature of a decomposition furnace in a cement burning system based on domain optimization, which includes: a decomposition furnace outlet temperature prediction model constructed based on a long-short time network; a neighborhood intervention model constructed based on a long-short time network; a local optimization objective function solution method using neighborhood information; a decomposition furnace outlet temperature control method based on a predictive control framework; the present invention can utilize neighborhood information to continuously optimize the objective function of real-time control, and the predictive control method ensures the stable operation of the working conditions.

[0004] However, by using CFD simulation technology to establish a finite element model of the decomposition furnace and conduct simulation analysis, the overall operating conditions of the equipment can be predicted in order to optimize the operating parameters and improve the production efficiency and operating stability of the decomposition furnace. In actual engineering practice, due to the large size and complex operating conditions of the decomposition furnace equipment itself, numerical simulation has the problem of untimely calculation. During the operation of the equipment, real-time simulation cannot be performed, and it is difficult to quickly provide optimization suggestions for operating parameters under the current operating conditions. It can only be used for design optimization in the equipment development stage or troubleshooting in the problem diagnosis stage, and cannot effectively guide production.

[0005] Moreover, due to the cost of sensor layout and monitoring capability limitations, there is currently no means to monitor the operating conditions of the entire flow field of the decomposition furnace, making it difficult for on-site personnel to detect possible problems with the equipment based on limited data; while numerical simulation technology can predict the overall operating conditions of the equipment, its slow calculation time and high computing resource requirements make it difficult to adjust production parameters in a timely manner according to real-time operating conditions, and it lacks the ability to predict the operating conditions after adjusting the parameters; based on this, a solution is provided. Summary of the invention

[0006] 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 dynamic process adjustment method for a decomposition furnace by fusing sensor data with numerical prediction.

[0007] A dynamic process adjustment method for a decomposition furnace by integrating sensor data and numerical prediction, the method specifically comprises the following steps: Step 1: Obtain the flow field data in the decomposition furnace as sample data of the decomposition furnace; Step 2: Perform intrinsic orthogonal decomposition on the sample data. The specific method is as follows: Standardize the sample data of the calciner calculated by simulation; A BPNN model with two hidden layers is selected to construct a BPNN model, and a mapping from r-dimensional input to m-dimensional output is established for the neural network; Divide the samples into training set and validation set, and train the BPNN model; Step 3: Use a distance-based correction filtering algorithm to perform a prediction data correction step based on the monitoring data on the data predicted by the BPNN model.

[0008] Compared with the prior art, the present invention has the following beneficial effects: The present invention combines online monitoring technology, CFD simulation technology and digital twin technology to intelligently and dynamically adjust the process parameters of the decomposition furnace, ensure the stable operation of the decomposition furnace and improve its material distribution uniformity, coal powder combustion efficiency and production efficiency, save energy and reduce consumption, and optimize the process control level; the present invention provides a complete set of dynamic process adjustment methods for decomposition furnaces, which integrates the advantages of CFD fluid simulation technology in accurately simulating fluid-solid motion and complex chemical reactions in the decomposition furnace and the advantages of digital twin technology in rapid response, and accesses the sensor data of online monitoring for correction, thereby realizing rapid calculation of the decomposition furnace flow field and intelligent dynamic adjustment of process parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 This is a technical flow chart of the dynamic process adjustment of the decomposition furnace of the present invention; Figure 2 This is a simulated cloud diagram of the decomposition furnace of the present invention; Figure 3 This is a diagram of the BPNN neural network architecture of the present invention. DETAILED DESCRIPTION

[0010] 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.

[0011] See also Figure 1 The present application provides a method for dynamic process adjustment of a decomposition furnace by integrating sensor data and numerical prediction, and the method specifically comprises the following steps: Step 1: CFD simulation of the calciner is the basis of this technology; At present, the number of particles in industrial-grade decomposition furnaces has reached the order of 1e16. It is difficult to meet the current computing resources for direct simulation of the particles. The present invention considers the movement behavior of the particles as a fluid. The key point is to describe the viscosity and pressure of the solid phase movement. It 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. In recent years, it has been widely used to predict gas-solid flow in larger-scale industrial equipment, and can obtain accurate particle concentration distribution field and velocity field.

[0012] There are also complex chemical reactions in the decomposition furnace, such as coal powder combustion, raw material decomposition, NOx generation / reduction, etc. In order to speed up the calculation, the Euler-Euler multiphase flow combustion model is further used to obtain the temperature field, gas component field, raw material component distribution field, etc. more accurately. According to the actual adjustable process parameters, such as the coal feeding amount of the decomposition furnace, the tertiary air volume, the inlet oxygen concentration, etc., the decomposition furnace model established by Euler multiphase flow is used for multi-parameter simulation calculation. In order to avoid the exponential growth of the number of samples with the increase of parameters, the orthogonal experimental design is used to balance the sample space accuracy and calculation time. The CFD simulation results of the decomposition furnace can be seen Figure 2 .

[0013] The digital twin model based on POD-BPNN training can realize instant response calculation of specified process parameters.

[0014] This method uses proper orthogonal decomposition (POD) and back-propagation neural network (BPNN) to establish a reduced-order digital twin model for rapid prediction of the overall operating conditions of the decomposition furnace, achieving millisecond-level input-output response, greatly improving the calculation speed and reducing time cost.

[0015] The method includes the following two steps: 2.1 Proper orthogonal decomposition of sample data Proper orthogonal decomposition is a method of statistically processing high-dimensional data. It can analyze the components of high-dimensional flow field data, reduce its order and map it to a low-dimensional orthogonal basis mode space, and use the main characteristics of the flow field and its corresponding basis mode coefficients to approximate the actual situation, thereby reducing the amount of sample data. In essence, it maximizes the sample variance of the flow field in each low-dimensional dimension.

[0016] The first step is to standardize the calciner sample (flow field data) calculated by simulation. Assume that the original data is ,in , sample data is an n-dimensional vector (n is equal to the number of grids divided during CFD calculations), and r is the number of samples.

[0017] (1) (2) Formula (1) is used to obtain the mean of the sample, and formula (2) is used to subtract the mean from the original data. Data standardization is helpful to eliminate the differences between sample data features and avoid the influence of individual discrete values ​​on the analysis results. Normal standardization can also be used.

[0018] The covariance matrix of the standardized data can then be found: (3) Through this The eigenvalues ​​of the covariance matrix of order m can be solved, and the first m eigenvalues ​​can be written as , the corresponding basic mode eigenvector can be written as The selection of m value is determined according to the ratio of the variance value 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.

[0019] Then the original sample data can be approximately expressed as: (4) 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.

[0020] 2.2 Building BPNN Model The reduced-order eigenvalue matrix corresponding to the decomposition furnace 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.

[0021] Considering that the CFD simulation result data of the calciner obtained by multi-parameter simulation analysis is relatively small (generally dozens or hundreds of groups, which are small sample data), the shallow neural network has sufficient nonlinear mapping ability and generalization ability. Therefore, the BPNN model with two hidden layers is selected to take into account stability and feature description ability. The model architecture is shown in Figure 3 .

[0022] The neural network establishes a mapping from r-dimensional input to m-dimensional output, where the input is x=[x1, x2,…, xk]T, k corresponds to the number of input parameters of the training sample, the number of neurons in the hidden layer is p and q respectively, and the output is y=[y1, y2,…,ym]T, m corresponds to the number of POD basis modes after dimensionality reduction. ω1, ω2, ω3 are the weight matrices between layers, and their sizes are p×k, qxp and m×q respectively.

[0023] 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: Divide the samples into training set and validation set Use normal distribution to initialize the weight matrix to avoid gradient vanishing or gradient exploding in the early stages of training; 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; 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; 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.

[0024] 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. , the fundamental mode eigenvalue coefficient can be quickly predicted through the BPNN model , and then perform reverse reconstruction based on POD to obtain the predicted flow field prediction value , achieving millisecond-level input and output.

[0025] 3. Prediction data correction algorithm based on monitoring data 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.

[0026] This method uses a distance-based correction filtering algorithm: Assume that n sensors are arranged in the factory, and the measured value of each sensor is (i=1,2,3…n), the predicted flow field value at the same position is (i=1,2,3…n), then for any data point of the predicted flow field , and its corrected value is: (5) in, For data points The distance from each sensor position; μ is the correction coefficient, which is related to the radius of the decomposition furnace and is fitted according to the specific working conditions of different decomposition furnaces. This algorithm systematically considers the correction of the predicted data at 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 real working conditions on site.

[0027] 4. Control strategy reinforcement learning system based on historical decision-making Through the corrected decomposition furnace flow field data, on-site staff can conveniently and intuitively observe the temperature field, pressure field, component field and other comprehensive information in the decomposition furnace, so as to comprehensively evaluate the key indicators such as the operation status and production efficiency of the decomposition furnace, and perform optimization control. In order to learn this decision-making process, improve the intelligence of the process adjustment system, reduce the dependence on the experience of on-site staff, and discover the inherent laws of process parameter adjustment, this method introduces a reinforcement learning system based on historical decision-making, records the on-site process parameter debugging results, continues to learn, and intelligently gives parameter adjustment suggestions based on the working conditions to assist on-site personnel in their work. The main steps are as follows: 4.1 Build a deep neural network (a BPNN neural network with 3-4 hidden layers is selected according to the number of adjustable process parameters, which will not be described here) to analyze the historical sensor monitoring values , parameter adjustment strategy , Adjustment effect evaluation , i=1,2,3…,n-1) for training, and based on the current sensor observations , giving the optimal parameter adjustment strategy ; 4.2 On-site personnel according to the strategy Based on engineering experience, a real adjustment plan for the process parameters of the calciner is given (Also responsible for clearing abnormal strategies to ensure equipment safety), update process parameters and wait for the equipment status to stabilize; 4.3 Based on the flow field and component field conditions of the decomposition furnace before and after the process parameter adjustment, the effect evaluation of this adjustment is given , and recorded in the repository.

[0028] 4.4 Train the neural network again, update the weight parameters, and repeat the above steps.

[0029] Through reinforcement learning of accumulated long-term historical decisions, the deep neural network will discover the potential impact of process parameters on the operating conditions of the decomposition furnace and provide adjustment suggestions with reference value, which can significantly provide convenient and effective reference opinions for on-site personnel, ensure that the decomposition furnace operates under optimal conditions, and improve production efficiency and product quality.

[0030] 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 method for dynamic process adjustment of a decomposition furnace by integrating sensor data and numerical prediction, characterized in that: The method specifically comprises the following steps: Step 1: Obtain the flow field data in the decomposition furnace as sample data of the decomposition furnace; Step 2: Perform intrinsic orthogonal decomposition on the sample data. The specific method is as follows: Standardize the sample data of the calciner calculated by simulation; A BPNN model with two hidden layers is selected to construct a BPNN model, and a mapping from r-dimensional input to m-dimensional output is established for the neural network; Divide the samples into training set and validation set, and train the BPNN model; Step 3: Use a distance-based correction filtering algorithm to perform a prediction data correction step on the prediction value of the BPNN model based on the monitoring data.

2. The method for dynamic process adjustment of a decomposition furnace by fusing sensor data and numerical prediction according to claim 1 is characterized in that: The specific method of standardizing the sample data of the calciner obtained by simulation is as follows: Suppose the original sample data is x i , where i = 1, 2, 3, ... r, sample data x i is an n-dimensional vector, where n is equal to the number of grids divided during CFD calculation, and r is the number of samples; Formula (1) is to obtain the mean of the sample, and formula (2) subtracts the mean from the original data; The covariance matrix of the standardized data can then be found: By solving the eigenvalue 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; Then the original sample data can be approximately expressed as: X=U λ ξ (4).

3. The method for dynamic process adjustment of a decomposition furnace by fusing sensor data and numerical prediction according to claim 2 is characterized in that: In the process of building the BPNN model, The input is x = [x1, x2, ..., xk]T, k corresponds to the number of input parameters of the training sample, and the number of neurons in the hidden layer is p and q respectively; The output is y = [y1, y2, ..., ym]T, where m corresponds to the number of POD basis modes after dimensionality reduction; The overall structure is an input layer, two densely connected hidden layers and an output layer. The layers are connected by weight matrices ω1, ω2, and ω3, whose sizes are p×k, qxp, and m×q, respectively.

4. The method for dynamic process adjustment of a decomposition furnace by fusing sensor data and numerical prediction according to claim 3 is characterized in that: The specific steps for training the BPNN model are: Use normal distribution to initialize the weight matrix to avoid gradient vanishing or gradient exploding in the early stages of training; 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; 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; Calculate the reverse error of each layer and update the weight matrix; The prediction error of the validation set is calculated to evaluate the reasonable number of training rounds until the model has sufficient computational accuracy and generalization ability, so that the average absolute error of the training set does not exceed 2%, and the average absolute error of the test set in the sample space does not exceed 3%.

5. The method for dynamic process adjustment of a decomposition furnace by fusing sensor data and numerical prediction according to claim 4 is characterized in that: The specific correction method for the forecast data is as follows: Assume that there are n sensors 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 position; μ is a correction coefficient that is positively correlated with the radius of the decomposition furnace and is obtained by fitting the sensor data of different decomposition furnaces through the nonlinear least squares method.

6. The method for dynamic process adjustment of a decomposition furnace by fusing sensor data and numerical prediction according to claim 5 is characterized in that: After completing step 3 of correcting the model's prediction data, the following steps are required: A reinforcement learning system based on historical decisions is introduced to conduct continuous learning and intelligently provide parameter adjustment suggestions based on working conditions. The specific methods are as follows: According to the number of adjustable process parameters, a BPNN neural network with 3-4 hidden layers is selected to calculate the historical sensor monitoring value O(T i ), parameter adjustment strategy A(T i )、Adjustment effect evaluation Reward(T i ), i = 1, 2, 3…, n-1 for training, and based on the current sensor observation value O(T n ), giving the optimal parameter adjustment strategy A′(T n ); The administrator uses the policy A′(T n ) gives the actual adjustment scheme of the decomposition furnace process parameters A(T n ), and is also responsible for clearing abnormal strategies to ensure equipment safety, updating process parameters and waiting for the equipment status to stabilize; According to the flow field and composition field of the decomposition furnace before and after the process parameter adjustment, the effect evaluation Reward (T n ) and recorded in the repository; Train the neural network again, update the weight parameters, and repeat the above steps.

Images