An Industrial Kiln Control System and Method Based on Online Simulation
Patent Information
- Application Number
- CN202610930084.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-25
- Publication Date
- 2026-09-15
Smart Images
Figure CN122755873A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial kiln control technology, and more specifically, to an industrial kiln control system and method based on online simulation. Background Technology
[0002] Tunnel kilns are core thermal equipment in industries such as ceramics and refractory materials, and their firing process involves complex heat transfer, mass transfer, and chemical reactions. Traditional industrial kiln control relies heavily on operator experience, maintaining the kiln's temperature distribution by setting fixed burner power and cart speed. However, due to factors such as batch fluctuations in billets, changes in kiln car heat storage, thermocouple aging and drift, and differences in fuel calorific value, the actual firing temperature field often deviates from the preset curve, leading to over-firing, under-firing, or cracking of products, severely impacting product quality and energy efficiency. Existing control systems mostly employ PID or simple cascade regulation, which struggles to cope with the strong nonlinearity, large hysteresis, and multivariable coupling characteristics within the kiln.
[0003] In recent years, model-based predictive control and simulation technologies have been gradually applied to industrial kilns, but existing solutions still have significant shortcomings. On the one hand, offline simulation models typically use fixed physical property parameters (such as thermal conductivity, specific heat capacity, and enthalpy change), which cannot adapt to the time-varying drift of the kiln's thermal characteristics during long-term operation, leading to a decrease in model prediction accuracy over time. On the other hand, the three-dimensional temperature field inside the kiln is difficult to acquire online in real time, and most systems rely only on a limited number of thermocouple measuring points for approximate control, ignoring spatial temperature gradient information and failing to accurately determine the risk of local overheating or underfiring. Furthermore, existing model predictive control systems mostly adopt a single-layer structure, which cannot simultaneously address slow-varying operating conditions and suppress rapid time-varying disturbances, and lacks a dynamic compensation mechanism for simulation residuals.
[0004] Therefore, there is an urgent need for an industrial kiln control system and method that can automatically correct simulation model parameters online, reconstruct the three-dimensional temperature field of the entire kiln in real time, and have dual-layer optimization control capabilities, in order to solve the technical problems of model mismatch, invisible temperature field, and untimely control response in the existing technology. Summary of the Invention
[0005] To overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide an industrial kiln control system and method based on online simulation.
[0006] To achieve the above objectives, the present invention provides the following technical solution: An industrial kiln control system based on online simulation includes the following modules: The model self-calibration module is used to establish a lumped parameter thermal balance model for the temperature zone, collect on-site data of the kiln and determine the sintering stage, and use unscented Kalman filtering with embedded adaptive covariance matching algorithm to correct the equivalent thermal conductivity, equivalent specific heat capacity and reaction enthalpy change parameters online. It is recursively calculated every 1 minute and updates the simulation model parameters after convergence. The three-dimensional reconstruction compensation module is used to quickly calculate the background field of the three-dimensional temperature field of the whole kiln by using the intrinsic orthogonal decomposition reduced-order model. It combines the measured temperature points and reconstructs the optimal estimated three-dimensional temperature field by fusion of Kalman filter data. The difference between the reconstructed field and the simulation field is decomposed into the gradual trend deviation and the transient disturbance deviation. The gradual trend deviation is converted into a property correction amount and fed back to the model self-calibration module. The transient disturbance deviation is converted into a temperature compensation value for the feedforward of the lower-level controller. The dual-layer optimization control module consists of an upper-layer slow-cycle controller and a lower-layer fast-cycle controller. They exchange set trajectories and compensation values through shared memory. The upper-layer controller uses a 10-minute cycle to solve for the optimal temperature setting trajectory with total gas consumption and the temperature bandwidth of the combustion zone as optimization objectives. The lower-layer controller uses a 30-second cycle to linearize the simulation model and add transient disturbance compensation. It constructs a model to predict the control problem, solves the control increment, and executes it. At the same time, it monitors the temperature difference between the upper and lower layers and switches to protection mode when the limit is exceeded.
[0007] Specifically, the model self-calibration module performs the following operations: The tunnel kiln is divided into 30 temperature zones along its length. Each temperature zone is regarded as a fully mixed-flow reactor. An energy conservation equation is established that includes burner heat release, flue gas convection, kiln wall heat dissipation, and reaction enthalpy change. The temperature of 28 thermocouples, the mass flow rate of natural gas and combustion air of each group of burners, the kiln car's forward displacement, the humidity and oxygen content of the flue gas are collected at a frequency of 1 second via industrial Ethernet, and then filtered and outlier is removed. Based on the temperature gradient, humidity decrease rate, and oxygen content, the free water evaporation stage, organic matter oxidation stage, and crystal transformation stage are determined. The sintering stage identifier and reaction progress index are output by looking up a table. The parameters are estimated using the equivalent thermal conductivity, equivalent specific heat capacity, and reaction enthalpy change as state variables, and the covariance matrix trace is monitored to determine convergence.
[0008] Specifically, the model self-calibration module embeds an adaptive covariance matching algorithm in the measurement update stage of the unscented Kalman filter: Let the measurement innovation in the k-th filtering cycle be the difference between the measured temperature vector and the predicted temperature vector, and let the theoretical covariance matrix of the innovation be the sum of the measurement prediction covariance and the observation noise covariance of the previous cycle. The observation noise covariance matrix is updated online using exponential weighted recursion based on the current information, with a forgetting factor of 0.05. The initial matrix is a diagonal matrix of the temperature measurement noise variance. After each update, the observation noise covariance matrix is decomposed into eigenvalues, and values less than 1 are decomposed into eigenvalues. The eigenvalues are forced to be Reconstruct the matrix to ensure positive definiteness.
[0009] Specifically, the three-dimensional reconstruction compensation module performs the following operations: The intrinsic orthogonal decomposition-Galogen reduced-order model of the flow field and temperature field in the kiln is pre-generated. The first 20 basis functions are obtained by performing singular value decomposition on the steady-state simulation results of 23 different working conditions. The Navier-Stokes equations and energy equations are projected to obtain a set of ordinary differential equations about the modal coefficients. The online calculation time is less than 0.5 seconds. The temperature of 30 measuring points is acquired in real time. The current burner power and fan frequency are input into the reduced-order model to obtain the three-dimensional predicted temperature field of the whole kiln as the background field. The 30 measured temperatures are used as observations. Data fusion is performed by ensemble Kalman filtering. The state vector consists of 20 modal coefficients and 12 burner power adjustment factors. The ensemble number is set to 50.
[0010] Specifically, the 3D reconstruction compensation module performs adaptive covariance dilation and spatial localization in the ensemble Kalman filter: Calculate the observed innovation vector and its outer product, and obtain the inflation factor by the ratio of the vector to the trace of the theoretical innovation covariance. The inflation factor is limited to not less than 1.0. The prediction error covariance is then inflated and used. A compactly supported localized matrix is constructed using a fifth-order piecewise rational function. Its elements are related to the physical distance between grid points. When the distance exceeds the width of the kiln cross section of 2.4 meters, the weights are reset to zero. The expanded prediction error covariance is multiplied element-wise with the localized matrix to obtain the localized covariance used for measurement updates.
[0011] Specifically, the upper-level controller in the dual-layer optimization control module performs the following operations: Triggered every 10 minutes, the simulation model parameters after self-calibration and the current temperature status of the whole kiln are received, and the updated temperature zone heat balance model is used as the prediction model; the firing zone temperature is 1150±5℃, the preheating zone heating rate is ≤15℃ / min, and the cooling zone cooling rate is ≤30℃ / min as hard constraints. The optimized performance index is the weighted sum of total gas consumption and temperature difference within the combustion zone over a future push cycle, with the weighting coefficient adjustable according to either the quality-first mode or the energy-first mode. The optimal set trajectory and burner power benchmark are solved using a sequential quadratic programming algorithm and written into a shared memory set trajectory queue.
[0012] Specifically, the lower-level controller in the dual-layer optimization control module continuously monitors the temperature gradients above and below each cross-section calculated from the optimal estimated three-dimensional temperature field. When the temperature difference between the upper and lower sections of any cross-section exceeds 80°C, a protection mode is triggered. The objective function of the model predictive control problem is switched to a single minimization of the temperature gradient, the power constraint of the corresponding burner is tightened, the prediction time domain is reduced, and the solution is solved and executed quickly; until the temperature difference between the upper and lower parts recovers to below 60°C, the objective function is switched back to the standard model predictive control objective.
[0013] An industrial kiln control method based on online simulation includes the following steps: Step 1: Collect data on the temperature of each thermocouple, the flow rate of the burner gas, the flow rate of the combustion air, the displacement of the kiln car, the humidity of the flue gas, and the oxygen content of each part of the tunnel kiln at a frequency of 1 second via industrial Ethernet. Perform moving median filtering and Raida criterion to remove outliers for each measuring point, and fill missing values with the nearest neighbor valid value. Step 2: Establish a lumped parameter thermal balance model for each temperature zone. Determine the current sintering stage based on the temperature gradient, humidity decrease rate, and oxygen content, and obtain the reaction progress index by looking up a table. Use unscented Kalman filtering and embed an adaptive covariance matching algorithm to correct the equivalent thermal conductivity, equivalent specific heat capacity, and reaction enthalpy change parameters online. It is recursively calculated every 1 minute, and the simulation model parameters are updated after convergence. Step 3: Quickly calculate the background three-dimensional temperature field of the entire kiln using the pre-constructed intrinsic orthogonal decomposition reduced-order model. Combined with 30 measured temperature points, reconstruct the optimal estimated three-dimensional temperature field by fusing Kalman filter data. Subtract the simulated field from the reconstructed field to obtain the deviation field. Separate the slowly varying trend deviation and transient disturbance deviation through empirical orthogonal function decomposition and low-pass filtering. Convert the slowly varying trend deviation into a property correction amount and feed it back to Step 2. Convert the transient disturbance deviation into a temperature compensation value for use by the lower-level controller feedforward. Step 4: The upper-level controller uses a 10-minute cycle to solve for the optimal temperature setpoint trajectory and burner power baseline, with total gas consumption and combustion zone temperature bandwidth as optimization objectives; the lower-level controller uses a 30-second cycle to linearize the simulation model and add transient disturbance compensation, construct the model predictive control problem, solve the control increment, and execute it; at the same time, it monitors the temperature difference between the upper and lower sections, and switches to protection mode when the temperature difference exceeds 80℃, minimizing the temperature gradient until the temperature difference recovers to below 60℃.
[0014] The technical effects and advantages of this invention are as follows: This invention establishes a lumped-parameter thermal balance model for different temperature zones and employs an unscented Kalman filter (UKF) embedded with an adaptive covariance matching algorithm to correct the equivalent thermal conductivity, equivalent specific heat capacity, and reaction enthalpy change parameters online. This effectively solves the problem of decreased prediction accuracy caused by the time-varying drift of kiln thermal characteristics in traditional offline simulation models. Simultaneously, it utilizes an intrinsic orthogonal decomposition (POD) reduced-order model to quickly calculate the background three-dimensional temperature field of the entire kiln. Combined with finite measured temperature points, it reconstructs the optimal estimated three-dimensional temperature field through ensemble Kalman filter (EnKF) data fusion. The deviation field is decomposed into slowly varying trend deviations and transient disturbance deviations, which are fed back to the model self-correction module and fed forward to the lower-level controller, respectively. This significantly improves the spatial resolution of the temperature field and the dynamic compensation capability of the simulation residuals.
[0015] This invention employs a two-layer hierarchical rolling time-domain optimization control strategy. The upper layer optimizes total gas consumption and firing zone temperature bandwidth with a 10-minute cycle to solve for the optimal set trajectory. The lower layer linearizes the model with a 30-second cycle and incorporates transient disturbance compensation, constructing a model predictive control (MPC) problem to quickly solve for control increments, thus balancing global optimization under slowly varying operating conditions with rapid suppression of fast time-varying disturbances. Simultaneously, the lower-layer controller continuously monitors the temperature difference between the upper and lower sections, automatically switching to protection mode when limits are exceeded, effectively preventing thermal shock cracking of the products. Compared with existing technologies, this invention achieves self-calibration of kiln model parameters, online reconstruction of the three-dimensional temperature field, and two-layer synergy of control strategies, significantly improving the stability of the firing process, product qualification rate, and energy utilization efficiency. Attached Figure Description
[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] like Figure 1 As shown, the following modules represent an industrial kiln control system and method based on online simulation: The model self-calibration module establishes a temperature-zone thermal equilibrium model, collects data such as thermocouple temperature and gas flow rate, and uses unscented Kalman filtering (UKF) and adaptive covariance matching algorithms to correct the equivalent thermal conductivity, specific heat capacity, and reaction enthalpy change parameters online, updating the model every minute; the process is as follows: An online simulation model based on lumped parameter heat balance in different temperature zones is established. The tunnel kiln is divided into 30 temperature zones along its length, each zone is considered a fully mixed-flow reactor, and its energy conservation equation is: In the formula: For the first Average temperature inside the kiln in the temperature zone The coordinates are for the kiln length direction. For the specific heat capacity of flue gas, For the first Average temperature inside the kiln in the temperature zone For the first Heat dissipation area of the kiln wall in the temperature zone For ambient temperature, , This represents the equivalent mass and equivalent specific heat capacity of the material and gas within this temperature range. The heat dissipation power of the burner, For flue gas flow rate, The heat dissipation coefficient of the kiln wall. and The enthalpy change and reaction rate of the green body reaction. The speed at which the cart is pushed.
[0019] The simulation model runs on an industrial computer backend and receives online parameter updates; the specific execution steps are as follows: The temperature of 28 thermocouples in the tunnel kiln, the mass flow rate of natural gas and combustion air for each group of burners, the pulse count of the kiln car propulsion displacement sensor, and the humidity and oxygen content of the exhaust gas were collected at a frequency of 1 second via industrial Ethernet. A moving median filter with a window length of 5 was applied to each measuring point, and instantaneous jump values were removed using the Laida criterion. Missing values were filled with the nearest neighbor valid value.
[0020] The distance the billet has been advanced is calculated by integrating the kiln car's displacement. Dividing this distance by the kiln car's speed yields the equivalent firing time. The average temperature gradient of the three thermocouples at the tail of the preheating zone is calculated in real time. and humidity sensor absolute moisture content reduction rate .when and When, it is determined to be the free water evaporation section; when Drop to 3-8℃ / m and After a significant reduction, the combined oxygen content rises to over 18%, indicating the organic oxidation stage; when the temperature reaches 950-1100℃ and humidity changes approach zero, it is identified as the crystal transformation stage, etc. The database stores the TG-DSC characteristic temperature table for this billet, and the identifier of the current sintering stage is output using a lookup table method. and reaction process index (0-1).
[0021] Based on the current sintering stage Retrieve the equivalent thermal conductivity from the prior parameter library Equivalent specific heat capacity and enthalpy change of reaction The initial Gaussian distribution. Construct with For the nonlinear state-space model of the state, the state equations employ random walks. ;in For at any time The state vector, This is process noise.
[0022] The observation equation is obtained by discretizing the energy conservation equation mentioned above. Specifically, it calculates the temperature change rate of each temperature zone using measured burner power, flow rate, and adjacent temperature zones combined with current parameter estimates, and sums these values to obtain the predicted temperature for the next time step. This summation is then combined with the measured thermocouple temperature to form the observation vector. An unscented Kalman filter (UKF) is employed, with three parameters as the state dimension. A proportionally symmetric sampling strategy is selected, and the process noise covariance matrix is... The observation noise covariance is taken as twice the diagonal matrix of the temperature measurement noise variance.
[0023] Meanwhile, to overcome the problem of slow time-varying statistical characteristics of measurement noise caused by thermocouple aging, dust accumulation, and electromagnetic interference during long-term operation, an adaptive covariance matching algorithm is embedded in the measurement update stage of the unscented Kalman filter. Let the measurement information in the k-th filtering cycle be denoted as... ,in For the measurement information in the k-th filtering period, This is a multidimensional measured vector composed of the thermocouple temperatures collected in each temperature zone. This is a multidimensional predicted temperature vector calculated from state-predicted values using observation equations. Within the UKF framework, the theoretical covariance matrix of the innovation is... , It is the measurement prediction covariance obtained from Sigma point propagation. This is the observation noise covariance matrix used in the previous period. It is then updated online using exponentially weighted recursion with the current information. : Among them, the forgetting factor Take 0.05, for Transpose. Initial matrix. The original temperature measurement noise variance was set to twice the diagonal matrix. To maintain... The positive definiteness, after each update, for Perform eigenvalue decomposition to find values less than 10 ... The eigenvalues are forced to be Reconstruct the matrix. Adaptively obtain... The measurement update is incorporated into the next filtering cycle, enabling the filter to automatically adapt to the real-time noise level and preventing a decrease in parameter identification accuracy due to underestimation or overestimation of noise.
[0024] It iterates every minute, outputting the posterior mean and covariance matrix of the parameters.
[0025] Monitor the trace of the covariance matrix output by UKF When 5 consecutive filtering cycles Convergence is determined when the temperature drops below 2% of the initial trace and ceases to decrease monotonically. The posterior mean is then written into the physical property parameter dictionary of the online simulation model to update the corresponding temperature range. , and The updated parameter set is then set as a model update flag. This updated parameter set is pushed to the 3D field reconstruction module and the two-layer predictive control module via shared memory, triggering a relinearization of the offline simulation model for use by the lower-level controller.
[0026] The 3D reconstruction and compensation module utilizes a reduced-order POD model to quickly calculate the background 3D temperature field of the entire kiln. Combining this with data from 30 measured temperature points, it reconstructs the optimal estimated 3D temperature field using ensemble Kalman filtering (EnKF). The reconstructed field is subtracted from the simulated field to obtain the deviation field. The slowly varying trend deviation is separated and fed back to the model self-correction module, while transient disturbance deviations are used as feedforward compensation for the lower-level controller. The process is as follows: The intrinsic orthogonal decomposition (POD)-Galogen reduced-order model (ROM) of the flow and temperature fields inside the kiln was pre-generated. The construction method involved using computational fluid dynamics (CFD) software to perform steady-state simulations of the tunnel kiln under 23 different burner power and trolley speed conditions. Snapshots of the temperature and velocity fields across 1.5 million grid nodes of the entire kiln were extracted and used to construct a snapshot matrix. Singular value decomposition is performed to obtain the POD basis functions. (Taking the first 20 orders, capturing 99% of the energy). Then, the Navier-Stokes equations and the energy equations are... Projection to obtain information about the modal coefficients The system of ordinary differential equations is the ROM. The online calculation of this ROM takes less than 0.5 seconds. The steps are detailed below: The system acquires real-time temperature values from 30 measurement points, including 28 thermocouples and 2 flue gas thermocouples. Simultaneously, it reads the simulation temperature values at the corresponding spatial nodes output by the online simulation model and collects the power percentage of each burner, the frequency of the exhaust fan inverter, and the speed of the combustion fan as boundary conditions.
[0027] Input the current burner power, fan frequency, etc. into the ROM, and obtain the predicted temperature field and velocity field on the three-dimensional grid of the whole kiln through the fourth-order Runge-Kutta method integral projection equation, which serve as the background field.
[0028] Thirty measured temperatures were used as observations, with the ROM prediction field as the background, and ensemble Kalman filtering (EnKF) was employed for data fusion. The ensemble number was set to 50, and the state vector was derived from the ROM mode coefficients. It consists of 20 elements and burner power adjustment factors (12 elements), and is generated by perturbing the initial conditions and power. The observation operator is a matrix that extracts the temperature of the measurement point by interpolation from the three-dimensional temperature field.
[0029] In each assimilation cycle, the prediction error covariance matrix is obtained from the ensemble member forecasts. Then, adaptive covariance dilation is first applied. The observation innovation vector for that period is then calculated. , The measured temperature vectors collected from 30 thermocouples. For the observation operator, Let be the mean of the modal coefficients from the ensemble forecast. Let the observation error covariance matrix be... Then the theoretical information covariance is The actual information covariance can be approximated as... The inflation factor is obtained by comparing the ratio of the two traces. In the formula The covariance inflation factor, The trace of the matrix, To observe the innovation vector, For the observation operator, For the prediction error covariance matrix, for Transpose; Restrict The forecast error covariance is inflated to This is to ensure that the ensemble dispersion can cover the actual forecast error.
[0030] Subsequently, to eliminate long-range spurious correlations caused by a finite number of sets, spatial localization is introduced. A compactly supported localization matrix is constructed using fifth-order piecewise rational functions. Its elements are only related to the physical distance between grid points. When the distance exceeds the influence radius (in this embodiment, the kiln cross-section width is taken as 2.4 meters), the weights are reset to zero. The localized covariance is obtained by performing a Schur product (element-by-element multiplication) on the expanded prediction error covariance and the localization matrix. ;in This is the localized prediction error covariance matrix. For compactly supported localized matrices, This is the Schur product (element-by-element multiplication). This is the expanded prediction error covariance matrix.
[0031] by The original prediction error covariance is replaced with a standard EnKF measurement update, the Kalman gain is calculated and the members of each set are corrected, and finally the optimal estimated modal coefficients are obtained, and the optimal estimated three-dimensional temperature field is reconstructed.
[0032] The optimal estimated 3D temperature field is subtracted grid-by-grid from the synchronous 3D simulation temperature field output by the online simulation model to obtain the 3D spatiotemporal deviation field. The bias field is then decomposed using Empirical Orthogonal Function (EOF) decomposition to extract the first 10 spatial modes and their corresponding time coefficients. For each time coefficient Perform first-order digital low-pass filtering (cutoff frequency) (i.e., corresponding to a period of 10 minutes or more), to obtain the gradual variation coefficient. The remaining high-frequency part is The deviation field reconstructed from the gradually varying coefficients is the gradually varying trend deviation. The transient disturbance deviation is reconstructed from the high-frequency coefficients. .
[0033] Gradual trend deviation Through a pre-calibrated sensitivity matrix Converted to equivalent thermophysical property corrections. The sensitivity matrix is obtained offline using the perturbation method: for each temperature range in the simulation model. and Apply a small perturbation, calculate the resulting change in the three-dimensional temperature field, and construct the Jacobian matrix. Solve for the result. ( (as a pseudo-inverse), to obtain each temperature zone and The additional correction terms are written to the correction queue in shared memory and used as the pre-bias of the augmented observations during the next self-calibration; where The deviation field represents a gradually changing trend. The transient disturbance deviation... Based on the influence radius of each burner, the temperature compensation values for each temperature zone are decomposed into four directions: front, back, left, and right. The compensation scale is directly written into the shared memory for use by the lower-level controller feedforward.
[0034] The dual-layer optimization control module has two layers. The upper layer (10-minute cycle) optimizes the total gas consumption and the temperature bandwidth of the firing zone to find the optimal temperature setpoint trajectory. The lower layer (30-second cycle) linearizes the model and incorporates transient disturbance compensation, constructs an MPC problem to solve for the control increment, and simultaneously monitors the temperature difference between the upper and lower layers. If the temperature exceeds the limit, it switches to a protection mode to prevent product cracking. The process is as follows: The system is divided into a slow-cycle upper layer and a fast-cycle lower layer, and the trajectory and compensation amount are set by exchanging shared memory.
[0035] The upper-level controller triggers every 10 minutes. It receives the self-calibrated simulation model parameters and the current overall kiln temperature status, using the updated temperature-zone thermal balance model as the prediction model. The firing process requirements for the finished product are used as a hard constraint: the firing zone temperature. Preheating rate Cooling rate of cooling zone Optimize performance metrics as follows: In the formula To optimize the objective function value at the upper level, For the first The sum of the heat output power of all burners. For the first The highest temperature within the firing zone. For the first The lowest temperature within the firing zone; that is, the temperature during the next trolley cycle (approximately 90 minutes). The weighted sum of total fuel consumption within the firing zone and the temperature range (temperature bandwidth) within the firing zone, with weighting coefficients... Adjustable between quality-priority and energy-priority modes. The optimal trajectory is solved using a sequential quadratic programming (SQP) algorithm. The burner power reference is written to the shared memory setting trajectory queue.
[0036] The lower-level fast-cycle controller triggers with a 30-second cycle. First, it reads the current temperatures of all thermocouples and then uses the self-calibrated simulation model to linearize it near the operating point: the energy conservation equation is currently... and input A Taylor expansion is performed at (gas flow rate, damper opening) to obtain a linear discrete state-space model: ,in This is the state vector for the next time step. Here is the state transition matrix. For the input matrix, The control increment at the current time step; the state vector at the current time step. , To account for the deviation from the set value of the upper layer, For the first The deviation between the temperature zone temperature and the upper-level set value. For the first Deviation in the rate of temperature change within the temperature zone. The amount of transient disturbance compensation. As a measurable disturbance term Add the output equation: This forms a linear time-varying prediction model with bias compensation, in which... For the output vector, For the output matrix, The perturbation input matrix is... This is a measurable disturbance term.
[0037] Each fast cycle constructs an MPC problem, predicting the time domain... Step (10 minutes), control the time domain Step. The objective function is: ;in To optimize the objective function value at the lower level, To predict the time domain, The weighted Euclidean norm is given by Q. The weighted Euclidean norm is given by matrix R. To predict the output at time k+i at time k, To control the time domain, n is the total number of temperature zones. The deviation between the temperature of the (k+i)th temperature zone and the set value of the upper layer is given. The term represents the deviation between the temperature of the j-th temperature zone at step k+i and the set value from the upper layer; the last term is the temperature gradient penalty term. The gradient penalty coefficient is used to prevent local temperature overshoot. Constraints include actuator limit (gas valve opening 0-100%), upper limit of control increment (no more than 5% opening per step), and soft temperature constraints at various points in the firing zone. , As slack variables, add to the objective function Penalties are imposed to soften constraints when infeasibility is not guaranteed. The problem is transformed into a standard quadratic programming problem. The penalty coefficient for slack variables, These are soft-constraint slack variables.
[0038] The optimal control increment sequence is obtained by solving the quadratic programming problem online using the interior-point method. , only The current gas flow rate and damper opening settings are superimposed on the actuator via an analog output module. After execution, the temperature of each measuring point in the next cycle is immediately collected, and the status is fed back to the upper and lower layers, forming a closed-loop rolling optimization.
[0039] The lower-level controller continuously monitors the temperature gradients above and below each cross-section, calculated from the optimally estimated three-dimensional temperature field. When the temperature difference above and below any cross-section... When the threshold is set according to the thermal stress fracture limit of the billet, the protection mode is immediately triggered: in the MPC problem, the objective function is switched to a single minimization of the temperature gradient, and the power constraint of the corresponding burner is tightened. At the same time, the prediction time domain is reduced, the solution is solved and executed quickly until the gradient recovers to below 60°C, and then the standard MPC objective is switched back to prevent thermal shock cracking of the product.
[0040] The steps of an industrial kiln control method based on online simulation are as follows: Step 1: Collect field data such as thermocouple temperature, burner gas flow rate, combustion air flow rate, kiln car movement displacement, flue gas humidity, and oxygen content of each measuring point at a frequency of 1 second via industrial Ethernet. For each measuring point, perform moving median filtering and Raida criterion to remove outliers. Missing values are filled with nearest neighbor valid values, providing reliable input for subsequent model correction and state estimation.
[0041] Step 2: Establish a lumped parameter thermal balance model for each temperature zone. Determine the current sintering stage based on the temperature gradient, humidity decrease rate, and oxygen content, and obtain the reaction progress index by looking up a table. Employ unscented Kalman filtering (UKF) with an embedded adaptive covariance matching algorithm to correct the equivalent thermal conductivity, equivalent specific heat capacity, and reaction enthalpy change parameters online. The model is recursively updated every minute and updated upon convergence.
[0042] Step 3: Utilize the pre-built POD reduced-order model to quickly calculate the background three-dimensional temperature field of the entire kiln. Combined with 30 measured temperature points, reconstruct the optimal estimated three-dimensional temperature field through ensemble Kalman filtering (EnKF) data fusion. Subtract the simulated field from the reconstructed field to obtain the deviation field. Separate the slowly varying trend deviation and feed it back to the model self-correction module. Transient disturbance deviations are used as feedforward compensation for the lower-level controller.
[0043] Step 4: The upper-level controller, with a 10-minute cycle, uses total gas consumption and the temperature bandwidth of the firing zone as optimization objectives to solve for the optimal temperature setting trajectory and burner power baseline. The lower-level controller, with a 30-second cycle, linearizes the model and adds transient disturbance compensation, constructs an MPC problem to solve for the control increment and executes it; simultaneously, it monitors the temperature difference between the upper and lower sections, switching to protection mode when the limit is exceeded to prevent thermal shock cracking of the product.
[0044] The above formulas are all dimensionless calculations. Dimensionless calculations can be performed using various methods such as standardization, which will not be elaborated here. The formulas are derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas can be set by those skilled in the art according to the actual situation.
[0045] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, ATA hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state ATA hard disk.
[0046] It should be understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.
[0047] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0048] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.
[0049] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.
[0050] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.
[0051] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable ATA hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0052] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. An industrial kiln control system based on online simulation, characterized in that, Includes the following modules: The model self-calibration module is used to establish a lumped parameter thermal balance model for the temperature zone, collect on-site data of the kiln and determine the sintering stage, and use unscented Kalman filtering with embedded adaptive covariance matching algorithm to correct the equivalent thermal conductivity, equivalent specific heat capacity and reaction enthalpy change parameters online. It is recursively calculated every 1 minute and updates the simulation model parameters after convergence. The three-dimensional reconstruction compensation module is used to quickly calculate the background field of the three-dimensional temperature field of the whole kiln by using the intrinsic orthogonal decomposition reduced-order model. It combines the measured temperature points and reconstructs the optimal estimated three-dimensional temperature field by fusion of Kalman filter data. The difference between the reconstructed field and the simulation field is decomposed into the gradual trend deviation and the transient disturbance deviation. The gradual trend deviation is converted into a property correction amount and fed back to the model self-calibration module. The transient disturbance deviation is converted into a temperature compensation value for the feedforward of the lower-level controller. The dual-layer optimization control module consists of an upper-layer slow-cycle controller and a lower-layer fast-cycle controller. They exchange set trajectories and compensation values through shared memory. The upper-layer controller uses a 10-minute cycle to solve for the optimal temperature setting trajectory with total gas consumption and the temperature bandwidth of the combustion zone as optimization objectives. The lower-layer controller uses a 30-second cycle to linearize the simulation model and add transient disturbance compensation. It constructs a model to predict the control problem, solves the control increment, and executes it. At the same time, it monitors the temperature difference between the upper and lower layers and switches to protection mode when the limit is exceeded.
2. The industrial kiln control system based on online simulation according to claim 1, characterized in that, The model self-calibration module specifically performs the following operations: The tunnel kiln is divided into 30 temperature zones along its length. Each temperature zone is regarded as a fully mixed-flow reactor. An energy conservation equation is established that includes burner heat release, flue gas convection, kiln wall heat dissipation, and reaction enthalpy change. The temperature of 28 thermocouples, the mass flow rate of natural gas and combustion air of each group of burners, the kiln car's forward displacement, the humidity and oxygen content of the flue gas are collected at a frequency of 1 second via industrial Ethernet, and then filtered and outlier is removed. Based on the temperature gradient, humidity decrease rate, and oxygen content, the free water evaporation stage, organic matter oxidation stage, and crystal transformation stage are determined. The sintering stage identifier and reaction progress index are output by looking up a table. The parameters are estimated using the equivalent thermal conductivity, equivalent specific heat capacity, and reaction enthalpy change as state variables, and the covariance matrix trace is monitored to determine convergence.
3. The industrial kiln control system based on online simulation according to claim 2, characterized in that, The model self-calibration module embeds an adaptive covariance matching algorithm in the measurement update stage of the unscented Kalman filter: Let the measurement innovation in the k-th filtering cycle be the difference between the measured temperature vector and the predicted temperature vector, and let the theoretical covariance matrix of the innovation be the sum of the measurement prediction covariance and the observation noise covariance of the previous cycle. The observation noise covariance matrix is updated online using exponential weighted recursion based on the current information, with a forgetting factor of 0.
05. The initial matrix is a diagonal matrix of the temperature measurement noise variance. After each update, the observation noise covariance matrix is decomposed into eigenvalues, and values less than 1 are decomposed into eigenvalues. The eigenvalues are forced to be Reconstruct the matrix to ensure positive definiteness.
4. The industrial kiln control system based on online simulation according to claim 1, characterized in that, The three-dimensional reconstruction compensation module performs the following operations: The intrinsic orthogonal decomposition-Galogen reduced-order model of the flow field and temperature field in the kiln is pre-generated. The first 20 basis functions are obtained by performing singular value decomposition on the steady-state simulation results of 23 different working conditions. The Navier-Stokes equations and energy equations are projected to obtain a set of ordinary differential equations about the modal coefficients. The online calculation time is less than 0.5 seconds. The temperature of 30 measuring points is acquired in real time. The current burner power and fan frequency are input into the reduced-order model to obtain the three-dimensional predicted temperature field of the whole kiln as the background field. The 30 measured temperatures are used as observations. Data fusion is performed by ensemble Kalman filtering. The state vector consists of 20 modal coefficients and 12 burner power adjustment factors. The ensemble number is set to 50.
5. The industrial kiln control system based on online simulation according to claim 4, characterized in that, The 3D reconstruction compensation module performs adaptive covariance dilation and spatial localization in the ensemble Kalman filter: Calculate the observed innovation vector and its outer product, and obtain the inflation factor by the ratio of the vector to the trace of the theoretical innovation covariance. The inflation factor is limited to not less than 1.
0. The prediction error covariance is then inflated and used. A compactly supported localized matrix is constructed using a fifth-order piecewise rational function. Its elements are related to the physical distance between grid points. When the distance exceeds the width of the kiln cross section of 2.4 meters, the weights are reset to zero. The expanded prediction error covariance is multiplied element-wise with the localized matrix to obtain the localized covariance used for measurement updates.
6. The industrial kiln control system based on online simulation according to claim 1, characterized in that, The upper-level controller in the dual-layer optimization control module performs the following operations: Triggered every 10 minutes, the simulation model parameters after self-calibration and the current temperature status of the whole kiln are received, and the updated temperature zone heat balance model is used as the prediction model; the firing zone temperature is 1150±5℃, the preheating zone heating rate is ≤15℃ / min, and the cooling zone cooling rate is ≤30℃ / min as hard constraints. The optimized performance index is the weighted sum of total gas consumption and temperature difference within the combustion zone over a future push cycle, with the weighting coefficient adjustable according to either the quality-first mode or the energy-first mode. The optimal set trajectory and burner power benchmark are solved using a sequential quadratic programming algorithm and written into a shared memory set trajectory queue.
7. The industrial kiln control system based on online simulation according to claim 6, characterized in that, The lower-level controller in the dual-layer optimization control module continuously monitors the temperature gradients above and below each cross-section calculated from the optimal estimated three-dimensional temperature field. When the temperature difference between the upper and lower sections of any cross-section exceeds 80°C, a protection mode is triggered. The objective function of the model predictive control problem is switched to a single minimization of the temperature gradient, the power constraint of the corresponding burner is tightened, the prediction time domain is reduced, and the solution is solved and executed quickly; until the temperature difference between the upper and lower parts recovers to below 60°C, the objective function is switched back to the standard model predictive control objective.
8. A method for an industrial kiln control system based on online simulation as described in any one of claims 1-7, characterized in that, Includes the following steps: Step 1: Collect data on the temperature of each thermocouple, the flow rate of the burner gas, the flow rate of the combustion air, the displacement of the kiln car, the humidity of the flue gas, and the oxygen content of each part of the tunnel kiln at a frequency of 1 second via industrial Ethernet. Perform moving median filtering and Raida criterion to remove outliers for each measuring point, and fill missing values with the nearest neighbor valid value. Step 2: Establish a lumped parameter thermal balance model for each temperature zone. Determine the current sintering stage based on the temperature gradient, humidity decrease rate, and oxygen content, and obtain the reaction progress index by looking up a table. Use unscented Kalman filtering and embed an adaptive covariance matching algorithm to correct the equivalent thermal conductivity, equivalent specific heat capacity, and reaction enthalpy change parameters online. It is recursively calculated every 1 minute, and the simulation model parameters are updated after convergence. Step 3: Quickly calculate the background three-dimensional temperature field of the entire kiln using the pre-constructed intrinsic orthogonal decomposition reduced-order model. Combined with 30 measured temperature points, reconstruct the optimal estimated three-dimensional temperature field by fusing Kalman filter data. Subtract the simulated field from the reconstructed field to obtain the deviation field. Separate the slowly varying trend deviation and transient disturbance deviation through empirical orthogonal function decomposition and low-pass filtering. Convert the slowly varying trend deviation into a property correction amount and feed it back to Step 2. Convert the transient disturbance deviation into a temperature compensation value for use by the lower-level controller feedforward. Step 4: The upper-level controller uses a 10-minute cycle to solve for the optimal temperature setpoint trajectory and burner power baseline, with total gas consumption and combustion zone temperature bandwidth as optimization objectives; the lower-level controller uses a 30-second cycle to linearize the simulation model and add transient disturbance compensation, construct the model predictive control problem, solve the control increment, and execute it; at the same time, it monitors the temperature difference between the upper and lower sections, and switches to protection mode when the temperature difference exceeds 80℃, minimizing the temperature gradient until the temperature difference recovers to below 60℃.