A mechanism and data fusion-based prediction control method for dry quenching system

Abstract

The present application belongs to the field of intelligent control and information technology, relates to the calculation of the burning loss rate of the dry quenching system, the control of the carbon monoxide (CO) content and the online real-time optimization of the air introduction amount, and is a predictive control method of the dry quenching system based on mechanism and data fusion. The present application uses the existing historical data in the industrial field, combines with the mechanism formula of the dry quenching furnace to construct a mechanism and data fusion dry quenching furnace model, uses the model as a predictive model of model predictive control (MPC) to optimize the control of the dry quenching furnace, and ensures the rapidity and robustness of the control. The present application solves the problem of the high burning loss rate of the dry quenching system during the operation process, causes economic loss and energy waste, achieves good economic benefits, and has great application value in realizing the energy saving and emission reduction of the dry quenching furnace.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control and information technology, and involves the calculation of burn-off rate of dry quenching system, control of carbon monoxide (CO) content and online real-time optimization of air introduction volume. It is a predictive control method for dry quenching system based on mechanism and data fusion. Background Technology

[0002] Dry quenching systems are characterized by multiple variables, variable constraints, and strong nonlinearity. During operation, due to the control of circulating gas composition, the tightness of the dry quenching system, and the influence of pre-storage section pressure control, high burn-off rates are a common problem. This leads to increased coke ash content, decreased quality and yield, resulting in energy waste and economic losses. Therefore, it is necessary to design an optimized control method for dry quenching systems to ensure safe and stable operation while reducing the burn-off rate. Current control methods for dry quenching systems mainly focus on regulating furnace temperature, material level, coke discharge temperature, and air-material ratio. However, research on methods for optimizing the burn-off rate control is limited. Therefore, a control method is urgently needed in industrial settings to effectively regulate the burn-off rate of dry quenching systems.

[0003] Currently, the optimization control of dry quenching furnaces mostly relies on manual control, supplemented by intelligent control algorithms. (Fu Lijun, Bai Xiaoguang, Lu Jianwen, et al. A method to reduce the burn-off rate of dry quenched coke [P]. China. 202210705573) A method to reduce the burn-off rate by relying on the comprehensive control of circulating gas and air introduction by human intervention was designed; (Zhu Zhixiong, Li Ning. Control of burn-off rate of dry quenched coke [J]. Zhejiang Metallurgy, 2009(2):5.) Combining the relationship between coke burn-off rate and steam output and coke powder output, and the relationship between steam output and coke powder output, an effective control method for the burn-off rate of dry quenched coke was designed; (Jia M, Chi H. The research of temperature control system for cycle gas of CDQexport [C] / / International Conference on Electronic & Mechanical Engineering & Information Technology. IEEE, 2011.) Based on the fuzzy PID temperature control system of ARM, a control method for the circulating air temperature at the outlet of dry quenched coke was designed; (Jian G, Chen X. CDQ System Designing and Dual-LoopPID Tuning Method for Air Steam Temperature [C] / / International Symposium on Distributed (Computing & Applications to Business. IEEE Computer Society, 2013.) A dual-loop PID control method was designed to control the main steam temperature of the dry quenching system; (Lu Min, Zou Jun, Huang Bin. Adaptive control system for CO in dry quenching based on BP neural network [J]. China Metallurgy, 2018, 28(11):4.) Based on BP neural network, a new adaptive control method for CO in dry quenching was designed, which controlled the CO concentration within an effective and reasonable range and achieved good control results.

[0004] The dry quenching system, characterized by multiple variables, variable constraints, and strong nonlinearity, presents significant challenges in controller design. Current control research for dry quenching systems typically employs a method primarily based on manual control, supplemented by intelligent control algorithms. While this approach has yielded some economic benefits, existing methods result in a coke loss rate of approximately 1.2%, leading to increased coke ash content, decreased quality and yield, and increased energy losses and costs for the enterprise. Therefore, an optimized control method is urgently needed to address these issues. Summary of the Invention

[0005] The technical problem this invention aims to solve is the excessive burn-off rate during the operation of dry quenching systems, which leads to economic losses and energy waste. To address this issue, data from a factory site is analyzed, and a dry quenching furnace model integrating mechanism and data is constructed using the dry quenching furnace mechanism formula. This model is then used as a predictive model for the MPC controller to optimize the control of the dry quenching furnace, calculating the burn-off rate, CO content, and air intake in real time. This achieves rapid, safe, and effective control of the dry quenching furnace, ensuring both speed and robustness. This invention allows for accurate regulation of the dry quenching furnace when factors such as the air-to-material ratio change, providing decision support for on-site personnel in optimization, control, and scheduling.

[0006] The technical solution of this invention is as follows:

[0007] A predictive control method for a dry quenching system based on mechanism and data fusion, the specific steps of which are as follows:

[0008] 1. Data acquisition and preprocessing: Read historical data from the real-time database on site, and perform noise reduction and anomaly processing on it;

[0009] 2. Dry quenching furnace model establishment: Based on the obtained actual historical data and the dry quenching furnace mechanism formula model, a model combining the dry quenching furnace mechanism data is constructed using the subspace identification method;

[0010] 3. Coke loss rate calculation: The real-time coke loss rate is calculated using the coke loss rate calculation formula for dry quenching furnace based on actual data such as the air introduction rate of the dry quenching furnace.

[0011] 4. MPC Controller Construction: Based on the dry quenching furnace burnout rate calculation model, an appropriate loss function is selected as the optimization objective, and a corresponding MPC controller is constructed.

[0012] 5. Real-time optimization and control of coke loss rate: The above-designed controller is used to optimize the control of the dry quenching system.

[0013] The effects and benefits of this invention are:

[0014] This invention, when optimizing the control of a dry quenching system in an industrial setting, utilizes a subspace identification method, employing the dry quenching mechanism formula and actual field data to establish a dry quenching model. By transforming the model's equivalent constraints, the computational load of planning and control is reduced, improving control effectiveness. Model predictive control adds multiple constraints during the control process, ensuring the safe and stable operation of the dry quenching system and reducing the system's burn-off rate while meeting actual production requirements. This provides technical support for optimizing the control of dry quenching systems in industrial settings. This invention solves the problem of economic losses and energy waste caused by excessive burn-off rates during the operation of dry quenching systems, achieving significant economic benefits and demonstrating great application value in achieving energy conservation and emission reduction in dry quenching furnaces. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the implementation of a predictive control method for a dry quenching system based on mechanism and data fusion, as described in this invention.

[0016] Figure 2(a) is the MPC control block diagram of the dry quenching furnace.

[0017] Figure 2(b) is a comparison chart of the calculated burn rate.

[0018] Figure 2(c) is a comparison chart of CO calculation results.

[0019] Figure 2(d) is a comparison chart of the calculated air import volume. Detailed Implementation

[0020] To better understand the technical solution of this invention, the implementation method of this invention is described in detail with reference to the accompanying drawings, taking the optimization calculation of coke loss rate in a dry quenching system as an example. This invention utilizes actual on-site operating data and relevant mechanism formulas of the dry quenching furnace to establish a simulation model of the dry quenching furnace, and uses this model as the prediction model for the MPC controller to optimize the control of the air introduction volume of the dry quenching furnace under different production schedules, thereby reducing the coke loss rate of the dry quenching system and reducing production costs. Figure 1 The method flow shown in the invention, and the specific implementation steps of the invention are as follows:

[0021] Step 1: Data Acquisition and Preprocessing

[0022] Historical data such as air intake, circulating air flow, coke discharge volume, and coke discharge temperature are read from a real-time database at the industrial site and then subjected to noise reduction and anomaly processing.

[0023] (1.1) Noise Treatment:

[0024] The wavelet denoising method is used to reduce noise in the collected data. The collected data is represented as:

[0025] d i =f i +εz i , i = 1, ..., N (1)

[0026] Where d i f represents the i-th data point collected. i Let z represent the i-th noise-free data point, ε represent the noise level, and z i Let N be the i-th independent and identically distributed Gaussian white noise, and let N represent the data volume. Perform wavelet decomposition on the collected data:

[0027] W0d=W0f+εW0z (2)

[0028] Where d represents the collected data, f represents noise-free data, and W0 is the coefficient. The wavelet coefficient η is taken. tN (w,t N )for:

[0029]

[0030] in w represents the input signal; further, the denoised signal f is obtained. * :

[0031]

[0032] (1.2) Exception handling

[0033] When there are missing data, the following actions are taken: if the missing data is less than 1 hour, the average of the data before and after the missing data is selected as the missing data segment; if the missing data is more than 1 hour, the missing data segment is deleted directly.

[0034] Step 2: Establishing the dry quenching furnace model

[0035] For dry quenching systems, a subspace identification method is used to model the dry quenching system, considering the following system:

[0036]

[0037] Where A∈R n×n , B∈R n×m , C∈R l×n , D∈R l×m u k ∈R m For the system input at time k, y k ∈R l For the system output at time k, x k ∈R n To construct the system state value at time k, the following Hankel matrix is ​​constructed: (Inputs include the system's coke loading rate, air intake rate, coke discharge rate, and circulating air flow rate; outputs include CO content, boiler temperature, and coke discharge temperature.)

[0038]

[0039] Define the generalized observable matrix Γ i =(C CA CA) 2 … CA i-1 ) T Define a lower triangular matrix:

[0040]

[0041] The dry quenching furnace system then satisfies:

[0042]

[0043] Where Δ i =(A i-1 BA i-2 B … AB B), V· represents the set of output perturbations, X i Let i represent the set of states from i to i+j-1, defined as follows:

[0044]

[0045]

[0046] In the formula, · / * indicates that the line space of · is projected onto the line space of *. A matrix of size li×mi is further obtained as follows:

[0047]

[0048] Pick:

[0049]

[0050] Perform singular value decomposition on it:

[0051]

[0052] Where U1, U2, and V represent orthogonal matrices, and ∑1 is a diagonal matrix, we further obtain:

[0053]

[0054] in

[0055]

[0056] The answers are A, B, C, and D.

[0057] Step 3: Calculation of burn rate

[0058] The main reason for burn-out in the dry quenching furnace system is the combination of oxygen from the air and carbon from the furnace. The oxygen content entering and leaving the system primarily comes from the air. The burn-out rate is calculated using the following formula:

[0059]

[0060] In the formula, η represents the burn-off rate, and G 放散 This indicates the amount of flue gas discharged from the system after the fan; CO represents the carbon monoxide content; CO2 represents the carbon dioxide content; C mol V represents the mass of carbon per mole, taken as 12 g / mol. molThis represents the molar volume of the gas, with a value of 22.4 L / mol.

[0061] Step 4: MPC Controller Construction

[0062] From steps 2 and 3, we can see that there is a non-linear relationship between the burn-off rate and the air introduction rate:

[0063] η=f(G 空导 (17)

[0064] In the formula, f(·) represents the nonlinear mapping relationship between the air intake and the coking loss rate, and G 空导 This represents the amount of air introduced. To optimize the coke loss rate of the dry quenching system, the objective function is defined as follows:

[0065]

[0066] In the formula, M represents the control time domain, P represents the prediction time domain, Q represents the output error weighting coefficient, R represents the control quantity change weighting coefficient, and the output η(k) represents the burn-off rate of the dry quenching system at the current moment. r (k) represents the burn-off rate setpoint, and u(k) represents the air intake of the dry quenching system. To meet the actual production needs of dry quenching, the following boundary constraints also need to be satisfied:

[0067]

[0068] Among them, u min u max y min y max η min η max These represent the minimum and maximum values ​​of air intake, CO content, and calculated burn-off rate, respectively. At time k, the model output η(k|k),…,u(k+M-1|k) is predicted based on the future control sequence u(k|k),…,η(k+P-1|k) and the prediction model, where M≤P. The model output is then compensated based on the deviation between the model output and the actual output at time k-1 to reduce the negative impact of model mismatch on the control effect.

[0069] The MPC control block diagram of the dry quenching furnace in this embodiment is shown in Figure 2(a). Specifically, at each sampling time, based on the obtained current burn-off rate measurement information, a finite-time open-loop optimization problem is solved online to obtain the air introduction amount acting on the controlled object. At the next sampling time, the above process is repeated, the optimization problem is refreshed, and the solution is recalculated.

[0070] The comparison of the burn rate calculation results in this embodiment is shown in Figure 2(b). It can be seen from the figure that the present invention can effectively reduce the burn rate of the system.

[0071] The comparison of CO calculation results in this embodiment is shown in Figure 2(c). As can be seen from the figure, the present invention can maintain the CO content within the safe production line and ensure the safe and stable operation of the system.

[0072] The comparison of the air import volume calculation results in this embodiment is shown in Figure 2(d). It can be seen from the figure that the present invention can respond quickly to various situations that occur on site, ensuring the safe and stable operation of the dry quenching system.

[0073] Step 5: Real-time optimization and control of foam loss rate

[0074] For the optimization problem, formulas (18) and (19) can be simplified as follows:

[0075]

[0076]

[0077] In the formula, U represents the amount of air introduced at time P in the future, and g u (·) denotes an inequality constraint, h v (·) denotes inequality constraints, and u and v represent the number of inequality constraints and equality constraints, respectively; it is placed at the iteration point U k Expanding and simplifying, we get:

[0078]

[0079]

[0080] in Represent the Hamiltonian operator; let

[0081] S=UU k (twenty two)

[0082] The optimization problem described above then becomes:

[0083]

[0084]

[0085] Further orders:

[0086]

[0087]

[0088]

[0089]

[0090] B eq =[h1(Uk ),…,h u (U k )] T

[0091] A = [g1(U k ),…,g v (U k )] T

[0092] Then (24) can be transformed into the general form of a quadratic programming problem:

[0093]

[0094]

[0095] The optimization problem (25) is solved online in real time to obtain the real-time optimal air intake, and the burn-off rate of the dry quenching system is optimized and controlled in real time.

[0096] Table 1 presents the optimized control adjustment results of this embodiment:

[0097] Table 1 Results of optimized burn-off rate control in this invention

[0098]

[0099] This invention enables optimized control of the dry quenching system, effectively solving the problem of excessive burn-off rate and energy waste during dry quenching operation. It effectively reduces the burn-off rate and saves costs, providing technical support for dry quenching operation control in industrial settings.

Claims

1. A predictive control method for a dry quenching system based on mechanism and data fusion, characterized in that, The specific steps are as follows: Step 1: Data Acquisition and Preprocessing Historical data, including air intake, circulating air flow, coke discharge volume, and coke discharge temperature, are read from a real-time database at the industrial site and then subjected to noise reduction and anomaly processing. (1.1) Noise control: The collected data is denoised using wavelet denoising; the collected data is represented as follows: ， (1) in This represents the i-th data point collected. This represents the i-th noise-free data point. Indicates noise level, Let i be the i-th independent and identically distributed Gaussian white noise. Indicate the data volume; perform wavelet decomposition on the collected data: (2) in This indicates that data has been collected. Indicates noise-free data. The coefficients are taken; wavelet coefficients are used. for: (3) in , This represents the input signal; further, a denoised signal is obtained. : (4) (1.2) Exception handling When there are missing data, the following actions are taken: if the missing data is less than 1 hour, the average of the data before and after the missing data is selected as the missing data segment; if the missing data is more than 1 hour, the data segment is deleted directly. Step 2: Establishing the dry quenching furnace model For dry quenching systems, a subspace identification method is used to model the dry quenching system, considering the following system: (5) in , , , , For the system input at time k, The output at time k of the system. To construct the system state value at time k, the system's coke loading rate, air intake rate, coke discharge rate, and circulating air flow rate are used as inputs, and CO content, boiler temperature, and coke discharge temperature are used as system outputs. ， (6) Define the generalized observable matrix Define a lower triangular matrix: (7) The dry quenching furnace system then satisfies: (8) in , This represents the set of output perturbations. express arrive The set of states, defined as: (9) (10) In the formula Indicates to make The line space in Projected onto the row space, Size is The matrix, further, yields: (11) Pick: (12) Perform singular value decomposition on it: (13) in , , Represents an orthogonal matrix. As a diagonal matrix, we further obtain: (14) in (15) The answers are A, B, C, and D. Step 3: Calculation of burn rate The main reason for burn-out in the dry quenching furnace system is the combination of oxygen from the air and carbon within the system. The oxygen content entering and exiting the system primarily comes from the air. The burn-out rate is calculated using the following formula: (16) In the formula Indicates the burn-off rate. This indicates the amount of flue gas discharged from outside the system after the fan. Indicates carbon monoxide content. Indicates carbon dioxide content, This indicates the mass of carbon per mole. Represents the molar volume of a gas; Step 4: MPC Controller Construction From steps 2 and 3, we can see that there is a non-linear relationship between the burn-off rate and the air introduction rate: (17) In the formula This represents the nonlinear mapping relationship between air intake and coke loss rate. This represents the amount of air introduced; to optimize the coke loss rate of the dry quenching system, the objective function is defined as follows: (18) In the formula, M represents the control time domain, P represents the prediction time domain, Q represents the output error weighting coefficient, R represents the control quantity change weighting coefficient, and the output... This indicates the burn-off rate of the dry quenching system at the current moment. Set the burn-off rate value. This refers to the air intake volume of the dry quenching system. To meet the actual production needs of dry quenching, the following boundary constraints also need to be satisfied: (19) in, , , , , , These represent the minimum and maximum values ​​of air intake, CO content, and calculated burn-off rate, respectively; at time k, based on the future control sequence... And the predictive model predicts the output of the future time. , And based on The deviation between the model output and the actual output at any given time is compensated by the model prediction output to reduce the negative impact of model mismatch on the control effect; Step 5: Real-time optimization and control of foam loss rate For the optimization problem, formulas (18) and (19) can be simplified as follows: (20) In the formula, U represents the amount of air introduced at time P in the future. To represent inequality constraints, Let u and v represent the number of inequality constraints and the number of equality constraints, respectively; then, at the iteration point... Expanding and simplifying, we get: (21) in Let Hamiltonian operator, let (22) The optimization problem described above then becomes: (23) Further orders: (24) Then (24) can be transformed into the general form of a quadratic programming problem: (25) The optimization problem (25) is solved online in real time to obtain the real-time optimal air intake, and the burn-off rate of the dry quenching system is optimized and controlled in real time.

Images