A temperature adjustment method and system for a decarburizer drying apparatus

Abstract

The present application belongs to the field of temperature regulation, and particularly relates to a temperature regulation method and system for a carburant drying device, to solve the technical problem of insufficient prediction accuracy and poor reliability of the existing regulation method, the carburant drying device temperature regulation method comprising the following steps: S1, a Gaussian mixture model is used to establish a joint probability distribution model of process disturbance and measurement noise; S2, a probability density function of a state trajectory in a future prediction time domain is obtained; S3, a probability constraint characterized by a conditional value at risk is set; S4, a first element of an optimal heater power sequence is used as a heater power control amount of a current control period, and a first element of an optimal working mode sequence is used as a working mode of the current control period, which are applied to the device together. Under the premise of ensuring high safety standards, the heater power and the working mode are optimized cooperatively, and the energy consumption of the device is reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of temperature regulation, and particularly relates to a temperature regulation method and system for a recarburizer drying device. BACKGROUND

[0002] Recarburizer is an indispensable additive in the steelmaking and casting process, and the quality of recarburizer affects the quality of metallurgical products and production cost. The drying of recarburizer is a link in the production process, which aims to remove excess moisture in the material through precise temperature control to avoid the burning loss of effective components due to excessively high temperature.

[0003] Industrial sites generally use PID controllers for temperature regulation. Although the implementation of the PID controller is simple, the PID controller based on the assumption of a linear model is difficult to match the changes in characteristics of the drying process at different stages (such as rapid heating, constant temperature drying, and temperature reduction and preservation). When facing working condition switching or external disturbances, the PID control often leads to a large overshoot or fluctuation of the temperature, which not only affects the uniformity of product quality, but also may cause problems such as high energy consumption or safety hazards.

[0004] Model predictive control (MPC) can explicitly handle multivariable constraint problems in the system through online rolling optimization, and shows superiority in handling systems with time delay and constraints. However, there are also some deficiencies: on the one hand, the standard MPC uses a single linear or nonlinear model, which is difficult to comprehensively and accurately describe the dynamic behavior of the equipment at different working modes, resulting in insufficient prediction accuracy and limiting the improvement of control performance. On the other hand, the drying process is inevitably affected by process disturbances such as batch differences of raw materials, changes in environmental temperature and humidity, and sensor measurement noise. These random uncertainties will reduce the prediction reliability of the model, leading to violation of key safety constraints in actual application of control decisions.

[0005] Therefore, how to construct a prediction model that can adapt to multi-modal switching and realize the multi-objective collaborative optimization of control performance and economic cost is a technical problem to be solved in the field of accurate temperature regulation of recarburizer drying equipment. SUMMARY

[0006] The application provides a recarburizer drying device temperature regulation method and system to solve the technical problems of insufficient prediction accuracy and poor reliability of existing regulation methods.

[0007] In a first aspect, the application provides a recarburizer drying device temperature regulation method, comprising the following steps:

[0008] S1, obtaining state parameters of the device in the current control period, including the furnace temperature and the material humidity; dividing the drying process of the device into multiple discrete working modes, including fast heating, stable drying and holding; based on historical operation data, establishing a joint probability distribution model of process disturbance and measurement noise using a Gaussian mixture model;

[0009] S2, constructing a switching nonlinear state space prediction model corresponding to the working mode; using a sparse polynomial chaos expansion method, propagating the uncertainty of process disturbance and measurement noise described by the joint probability distribution model in the switching nonlinear state space prediction model to obtain the probability density function of the state trajectory in the future prediction time domain;

[0010] S3, constructing a dictionary type multi-objective optimization function, the first priority target of which is to minimize the sum of the heater power and the working mode switching cost, and the second priority target of which is to minimize the deviation of the predicted temperature and the set temperature trajectory; setting a probability constraint characterized by a conditional value at risk to limit the conditional expected value of the device temperature exceeding the preset safety interval in the prediction time domain to be less than a set risk threshold;

[0011] S4, in each control period, solving the multi-objective optimization problem satisfying the probability constraint on line to obtain the optimal heater power sequence and the optimal working mode sequence, taking the first element of the optimal heater power sequence as the heater power control amount of the current control period, and taking the first element of the optimal working mode sequence as the working mode of the current control period, and applying them to the device.

[0012] Further, in S1, the process of establishing the joint probability distribution model of process disturbance and measurement noise using the Gaussian mixture model includes the following steps:

[0013] Collecting the heater power, furnace temperature and material humidity sequences in the historical operation data, calculating the predicted value using the nominal model of the switching nonlinear state space prediction model, and defining the difference between the historical measured values of the furnace temperature and the material humidity and the predicted values as the residual sequence;

[0014] Taking the residual sequence as the training sample, using the expectation maximization algorithm to iteratively estimate the parameters of the Gaussian mixture model, determining the optimal number of Gaussian components through the Bayesian information criterion, and obtaining the weight, mean and covariance matrix of each Gaussian component.

[0015] Further, after obtaining the analog signals of the furnace temperature and the material humidity, they are converted into digital signals through a data acquisition card.

[0016] Further, in S2, the process of obtaining the probability density function of the state trajectory in the future prediction time domain includes the following steps:

[0017] determine the random variables of process disturbance and measurement noise based on the joint probability distribution model, construct the standard orthogonal polynomial basis functions;

[0018] select the subset of the standard orthogonal polynomial basis functions which have significant influence on the state variables by using the sparse regression method, and use the subset for sparse expansion of the state variables in the switched nonlinear state space prediction model;

[0019] project the switched nonlinear state space prediction model onto the space spanned by the selected sparse basis functions to obtain a set of reduced-dimension deterministic differential equations for solving the expansion coefficients;

[0020] obtain the expansion coefficients of the state variables at each time in the future prediction time domain by solving the reduced-dimension deterministic differential equations, and reconstruct the probability density function of the state variables based on the expansion coefficients.

[0021] Further, the weighted L1 norm minimization algorithm is selected as the sparse regression method.

[0022] Further, the switched nonlinear state space prediction model is composed of a set of subsystem models and a switching rule for determining which subsystem model is activated at any time; each subsystem model is described in the form of a nonlinear state space, and a dedicated nonlinear state space subsystem model is established for each operating mode: ;

[0023] wherein, is the state vector of the device at the next time , such as the furnace temperature and the material humidity; is the state vector at the current time , such as the furnace temperature and the material humidity; is the control input at the current time , such as the heater power; is the currently activated operating mode; is a nonlinear function.

[0024] Further, in S3, the process of constructing the dictionary-type multi-objective optimization function includes the following steps:

[0025] set the first priority target as the cumulative sum of the control cost in the prediction time domain, and the cumulative sum is in the form of: ;

[0026] wherein, is the heater power in the kth control period, is the heater power in the (k-1)th control period, is a binary variable indicating whether the operating mode is switched, , and are preset non-negative weight coefficients;

[0027] the second priority target is the sum of squared errors between the predicted furnace temperature and the set temperature in the prediction time domain, in the form of:

[0028] wherein, is the expected value of the predicted furnace temperature calculated by the polynomial chaos expansion method, is the set temperature of the furnace.

[0029] Further, the process of setting the probability constraint characterized by the conditional value at risk includes the following steps:

[0030] setting the upper limit and the lower limit of the safety range of the furnace temperature, defining the loss function L(T) related to the temperature T, the value of the loss function L(T) being zero when the temperature is within the safety range , and monotonically increasing with the amount of exceeding when the temperature exceeds the safety range;

[0031] under the confidence level , setting the conditional value at risk CVaR constraint as wherein is a preset risk threshold, is the conditional value at risk of the loss under the confidence level .

[0032] In the optimization solution, auxiliary variables are introduced to convert the CVaR constraint into a set of deterministic algebraic inequalities, and the expected values involved in the algebraic inequalities are accurately calculated by the Gaussian quadrature rule compatible with the polynomial chaos expansion.

[0033] Further, the process of solving the multi-objective optimization problem satisfying the probability constraint online includes the following steps:

[0034] converting the dictionary multi-objective optimization problem into a mixed integer nonlinear programming problem, wherein the working mode sequence is an integer decision variable, and the heater power sequence is a continuous decision variable;

[0035] by introducing a slack variable to convert the non-smooth function in the probability constraint into a smooth algebraic inequality constraint, so that the converted problem is suitable for gradient-based nonlinear programming solvers;

[0036] ​The mixed algorithm is used for solving, wherein the mixed algorithm combines branch and bound method and sequential quadratic programming; at each node of the branch and bound method, a relaxed nonlinear programming subproblem is solved through the sequential quadratic programming until an integer and continuous optimal solution satisfying a preset convergence condition is obtained.

[0037] In a second aspect, the present application provides a recarburizer drying equipment temperature regulation system, comprising a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the recarburizer drying equipment temperature regulation method described above is realized.

[0038] The beneficial effects are: the present application can combine the complex characteristics of the whole drying process, calculate the influence of various uncertainties on future temperature, and improve the reliability and accuracy of the prediction by constructing a switching nonlinear state space prediction model matched with the actual operation stage of the equipment, combining the Gaussian mixture model and the sparse polynomial chaos expansion method. The probability constraint represented by the conditional value at risk can more strictly manage the key safety event of temperature overrun, not only control the probability of its occurrence, but also limit its severity once it occurs, thereby enhancing the safety of equipment operation. By solving the dictionary multi-objective optimization problem considering economy and control performance, the collaborative optimization of the heater power and the working mode is realized under the premise of ensuring high safety standards, thereby reducing the energy consumption of the equipment. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 The flowchart of the recarburizer drying equipment temperature regulation method;

[0040] Figure 2 The structure block diagram of the recarburizer drying equipment temperature regulation system. DETAILED DESCRIPTION

[0041] The embodiment of the recarburizer drying equipment temperature regulation method provided by the present application:

[0042] As shown in Figure 1 A recarburizer drying equipment temperature regulation method comprises the following steps:

[0043] S1, obtaining the state parameters of the equipment in the current control period, including the furnace temperature and the material humidity; dividing the drying process of the equipment into multiple discrete working modes, including rapid heating, stable drying and heat preservation; based on the historical operation data, a joint probability distribution model of process disturbance and measurement noise is established by using a Gaussian mixture model.

[0044] Specifically, an S-type thermocouple is deployed inside the furnace of the carbonizer drying equipment to convert furnace temperature data into analog signals; an infrared humidity sensor is installed at the discharge port to convert the surface humidity data of the material into analog signals. These analog signals are then converted into digital signals by a data acquisition card and transmitted to the host computer as real-time input for the control algorithm.

[0045] Based on production experience, the entire drying process is divided into three working modes: Working mode one is rapid heating, which aims to quickly raise the furnace temperature from the ambient temperature to near the target drying temperature using maximum power; Working mode two is stable drying, which aims to maintain the furnace temperature within a small range near the set value to stably remove moisture from the material; Working mode three is heat preservation, which aims to reduce power at the end of the drying process and use the residual heat of the equipment for low-temperature heat preservation to prevent the carbon components of the material from burning off and to ensure uniform discharge temperature.

[0046] Historical data of the carbon raiser drying equipment under different operating conditions were collected over a long period. The residual sequence between the model predictions and actual measurements was calculated. The residual sequence represents the combined impact of process disturbances and measurement noise. The expectation-maximization algorithm (EM algorithm) was used to fit the residual sequence, and a Gaussian mixture model composed of multiple Gaussian components was trained. The Gaussian mixture model can non-Gaussian and asymmetrically represent the probability distribution characteristics of uncertainty.

[0047] In an optional embodiment, the process of establishing a joint probability distribution model of process disturbance and measurement noise using a Gaussian mixture model includes the following steps:

[0048] Collect heater power, furnace temperature and material humidity sequences from historical operating data, calculate predicted values ​​using the nominal model of the switching nonlinear state-space prediction model, and define the difference between historical measured values ​​and predicted values ​​of furnace temperature and material humidity as residual sequences.

[0049] Using the residual sequence as training samples, the parameters of the Gaussian mixture model are iteratively estimated using the expectation-maximization algorithm. The optimal number of Gaussian components is determined by the Bayesian information criterion, and the weights, mean, and covariance matrices of each Gaussian component are obtained.

[0050] For example, suppose the collected historical operating data contains 1000 time points, where the heater power fluctuates between 50 and 100 kW, and the furnace temperature varies between 800 and 850°C. The temperature residual sequence obtained after prediction by the nominal model exhibits various fluctuation patterns. The temperature residual sequence is used as training samples, and after calculation using the Bayesian information criterion, it is determined that using three Gaussian components can optimally fit the residual distribution, obtaining the weights, mean, and covariance matrices of these three Gaussian components.

[0051] S2. Construct a switching nonlinear state-space prediction model corresponding to the working mode. Using the sparse polynomial chaotic expansion method, the uncertainty of process disturbance and measurement noise described by the joint probability distribution model is propagated in the switching nonlinear state-space prediction model to obtain the probability density function of the state trajectory in the future prediction time domain.

[0052] Specifically, for the three operating modes of rapid heating, stable drying, and heat preservation, nonlinear state-space models based on the fundamental equations of heat conduction and heat convection are established respectively. The process perturbation variables described by the Gaussian mixture model are expressed as a series sum of a set of orthogonal polynomial basis functions using the sparse polynomial chaotic expansion method. Substituting the series sum into the switching nonlinear state-space prediction model, the coefficients of the orthogonal polynomial basis function expansion are solved forward recursively in the prediction time domain using the Galerkin projection method. Based on these coefficients, the probability density functions of future furnace temperature, material humidity, and other state variables in the probability space are reconstructed.

[0053] In one embodiment, the switching nonlinear state-space prediction model is a mathematical model describing a hybrid system, possessing both continuous dynamic behavior and discrete event switching characteristics. The switching nonlinear state-space prediction model consists of a set of subsystem models and a switching rule that determines which subsystem model is activated at any given time. Each subsystem model is described in nonlinear state-space form; specifically, a dedicated nonlinear state-space subsystem model is established for each operating mode. ;

[0054] in, For the carbon raiser drying equipment at the next moment The state vector, such as furnace temperature and material humidity; For the current moment The state vector, such as furnace temperature and material humidity; For the current moment Control inputs, such as heater power; The currently active working mode; It is a nonlinear function.

[0055] A sparse polynomial is a function composed of a large number of basis functions, in which only a very small number of basis functions have non-zero coefficients. A sparse polynomial retains only the terms that significantly affect the result, discarding the vast majority of terms with zero coefficients. The sparse polynomial chaotic expansion method represents the random variables such as the state and parameters in a carbonizer drying device as a series of orthogonal polynomials about a basic random source; it uses only a few polynomial basis functions that significantly affect the output, thus achieving an accurate approximation of the probability distribution of the random variables.

[0056] In an optional embodiment, the process of obtaining the probability density function of the future predicted state trajectory in the time domain includes the following steps:

[0057] Based on the random variables of process disturbance and measurement noise determined by the joint probability distribution model, standard orthogonal polynomial basis functions are constructed.

[0058] Using the sparse regression method, a subset that has a significant impact on the state variables is selected from the standard orthogonal polynomial basis functions. This subset is used to sparsely expand and switch the state variables in the nonlinear state-space prediction model.

[0059] By projecting the switching nonlinear state-space prediction model onto the space spanned by the selected sparse basis functions, a set of dimension-reduced deterministic differential equations is obtained for solving the expansion coefficients.

[0060] By solving a system of dimension-reduced deterministic differential equations, the expansion coefficients of the state variables at each moment in the future prediction time domain are obtained, and the probability density function of the state variables is reconstructed based on these expansion coefficients.

[0061] Here, the weighted L1 norm minimization algorithm is chosen as the sparse regression method.

[0062] Specifically, Hermitian polynomial basis functions are used as standard orthogonal polynomial basis functions. For random perturbations described by Gaussian mixture models, corresponding Hermitian polynomial basis functions are constructed, generating a total of hundreds of higher-order polynomial terms.

[0063] The weighted L1 norm minimization algorithm is used to filter and automatically identify the 10 polynomial basis functions that have the greatest impact on the furnace temperature in the next 30 steps from hundreds of higher-order polynomial terms. The 10 polynomial basis functions form a sparse subset.

[0064] The original high-dimensional stochastic differential equations are transformed into a system of low-dimensional ordinary differential equations containing only 10 basis function expansion coefficients by Galerkin projection. Solving the system of low-dimensional ordinary differential equations yields an accurate analytical description of the future temperature distribution.

[0065] S3. Construct a dictionary-based multi-objective optimization function. Its first priority objective is to minimize the sum of heater power and operating mode switching cost, and its second priority objective is to minimize the deviation between the predicted temperature and the set temperature trajectory. Set a probability constraint represented by conditional risk value to limit the expected value of the equipment temperature exceeding the preset safe range to less than the set risk threshold within the prediction time domain.

[0066] Based on the probability density function of the obtained future temperature, the Value at Risk (VaR) at a high confidence level (e.g., 99%) is calculated, which is essentially the temperature quantile. The conditional expected value when the temperature exceeds the VaR value is calculated; this value is the Conditional Value at Risk (CVaR). The requirement that the CVaR value must be less than a preset limit safe temperature (e.g., 850°C) is used as a hard constraint in the optimization problem.

[0067] In this approach, the dictionary-based multi-objective optimization function sets a strict priority order for multiple optimization objectives, optimizing each objective sequentially from highest to lowest priority. When optimizing lower-priority objectives, it is crucial to ensure that the optimization results of higher-priority objectives that have already been achieved are not compromised.

[0068] In an optional embodiment, the process of constructing a dictionary-based multi-objective optimization function includes the following steps:

[0069] The first priority goal Set as the cumulative sum of control costs within the prediction time domain, with the cumulative sum in the following form: ;

[0070] in, The heater power in the k-th control cycle. The heater power in the (k-1)th control cycle. A binary variable indicating whether the operating mode has switched. , and These are preset non-negative weighting coefficients;

[0071] Second priority target Set as the sum of the squared errors between the predicted furnace temperature and the set temperature within the prediction time domain. The sum of squared errors is in the form of: ;

[0072] in, This is the expected value of the predicted furnace temperature calculated using the polynomial chaotic expansion method. Set the temperature for the furnace.

[0073] The controller aims to minimize control costs during optimization. The optimization strategy involves finding a set of heater power operation sequences that minimizes controller energy consumption, ensures the smoothest power fluctuations, and reduces the number of operating mode transitions. After identifying multiple feasible operating schemes to achieve this economic goal, the controller selects the scheme that best brings the desired furnace temperature close to the furnace set temperature. This priority-based optimization strategy ensures that temperature control accuracy is maximized while meeting economic requirements, achieving a balance between economic benefits and control performance.

[0074] In an optional embodiment, the process of setting probabilistic constraints represented by conditional value at risk includes the following steps:

[0075] Set a safe upper limit for furnace temperature and safety lower limit Define a loss function L(T) related to temperature T, when the temperature is within the safe range [ Within the safe range, the loss function value L(T) is zero. When the temperature exceeds the safe range, the loss function value increases monotonically with the amount of excess.

[0076] At confidence level Below, the conditional value at risk (CVaR) constraint is set as follows: ,in As a preset risk threshold, To at the confidence level Down, loss Conditional risk value;

[0077] In the optimization solution, auxiliary variables are introduced to transform the CVaR constraints into a set of deterministic algebraic inequalities. The expected values ​​involved in the algebraic inequalities are accurately calculated using the Gaussian quadrature rule, which is compatible with polynomial chaotic expansion.

[0078] For example, the safe operating range for furnace temperature is set to 810°C to 840°C, and a confidence level is specified. The risk threshold is 95%. The value is 2. When making decisions, the controller must ensure that, in the worst-case scenario of the worst 5% of all possible operating conditions, the loss corresponding to the average furnace temperature exceeding the safe range does not exceed a risk value equivalent to a 2°C deviation. This constraint is calculated directly using the coefficients of the polynomial chaotic expansion through the Gaussian quadrature method, ensuring both the strictness of the safety constraint and the efficiency of the calculation.

[0079] S4. In each control cycle, solve the multi-objective optimization problem that satisfies the probability constraint online to obtain the optimal heater power sequence and the optimal operating mode sequence. Use the first element of the optimal heater power sequence as the heater power control quantity for the current control cycle and the first element of the optimal operating mode sequence as the operating mode for the current control cycle, and apply them together to the equipment.

[0080] In each control cycle, the dictionary-based multi-objective function and probabilistic constraints are integrated into a nonlinear, non-convex stochastic optimization problem. Efficient numerical optimization algorithms, such as the interior-point method or sequential quadratic programming, are used to solve this stochastic optimization problem, yielding a set of optimal heater power sequences and a set of optimal operating mode sequences for the future prediction time domain. Based on the rolling time domain concept of model predictive control, only the first value in the optimal heater power sequence is output to the PID controller or the conduction angle of the thyristor is directly controlled. The first value of the optimal operating mode sequence is used as the command to switch the system's control logic, completing the control for the current cycle. At the next sampling time, all the above steps are repeated.

[0081] In an optional embodiment, the process of solving a multi-objective optimization problem satisfying probabilistic constraints online includes the following steps:

[0082] The dictionary-based multi-objective optimization problem is transformed into a mixed-integer nonlinear programming problem, where the working mode sequence is an integer decision variable and the heater power sequence is a continuous decision variable.

[0083] By introducing slack variables, the non-smooth functions in the probabilistic constraints are transformed into smooth algebraic inequality constraints, making the transformed problem suitable for gradient-based nonlinear programming solvers.

[0084] A hybrid algorithm is used to solve the problem, which combines the branch and bound method with sequential quadratic programming (SQP). At each node of the branch and bound method, the relaxed nonlinear programming subproblem is solved by SQP until an integer and continuous optimal solution that satisfies the preset convergence condition is obtained.

[0085] The entire optimization problem is constructed as a mixed-integer nonlinear programming model, where decision variables include the continuous heater power values ​​for the next 20 steps and binary integer variables determining which prediction model to use at each step. To facilitate solving, the maximum function in the conditional risk-value constraint is transformed into a smooth form by introducing a set of auxiliary variables and inequalities. The nonlinear programming solver employs a branch-and-bound framework to systematically explore different combinations of operating modes. Under each combination, a sequential quadratic programming algorithm is used to quickly calculate the optimal heater power sequence, yielding the optimal control strategy that balances economy, performance, and extreme risk avoidance within seconds.

[0086] An embodiment of the temperature control system for the carbon raiser drying equipment provided by the present invention:

[0087] like Figure 2 As shown, the temperature control system of the carbon raiser drying equipment includes a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned temperature control method for the carbon raiser drying equipment is implemented.

[0088] The temperature control system of the carbon raiser drying equipment also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0089] In addition, in the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

Claims

1. A method for temperature regulation in a carbon raiser drying device, characterized in that, Includes the following steps: S1, acquire the status parameters of the equipment in the current control cycle, including furnace temperature and material humidity; divide the drying process of the equipment into multiple discrete working modes, including rapid heating, stable drying and heat preservation; Based on historical operating data, a joint probability distribution model of process disturbance and measurement noise is established using a Gaussian mixture model. S2, construct a switching nonlinear state-space prediction model corresponding to the working mode; using the sparse polynomial chaotic expansion method, propagate the uncertainty of process disturbance and measurement noise described by the joint probability distribution model in the switching nonlinear state-space prediction model to obtain the probability density function of the state trajectory in the future prediction time domain, including: constructing standard orthogonal polynomial basis functions based on the random variables of process disturbance and measurement noise determined by the joint probability distribution model. The sparse regression method is used to select a subset of state variables that have a significant impact from the standard orthogonal polynomial basis functions. This subset is used to sparsely expand the state variables in the switching nonlinear state-space prediction model. The switching nonlinear state-space prediction model is then projected onto the space spanned by the selected sparse basis functions, resulting in a set of dimension-reduced deterministic differential equations for solving the expansion coefficients. By solving the dimension-reduced deterministic differential equations, the expansion coefficients of the state variables at each time step in the future prediction time domain are obtained. Based on these expansion coefficients, the probability density function of the state variables is reconstructed. S3, construct a dictionary-based multi-objective optimization function. Its first priority objective is to minimize the sum of heater power and working mode switching cost, and its second priority objective is to minimize the deviation between the predicted temperature and the set temperature trajectory. Set a probability constraint represented by conditional risk value to limit the expected value of the equipment temperature exceeding the preset safe range to less than the set risk threshold within the prediction time domain. S4. In each control cycle, solve the multi-objective optimization problem that satisfies the probability constraint online to obtain the optimal heater power sequence and the optimal operating mode sequence. Use the first element of the optimal heater power sequence as the heater power control quantity for the current control cycle and the first element of the optimal operating mode sequence as the operating mode for the current control cycle, and apply them together to the equipment.

2. The temperature adjustment method for the carbon raiser drying equipment according to claim 1, characterized in that, In S1, the process of establishing the joint probability distribution model of process disturbance and measurement noise using a Gaussian mixture model includes the following steps: Collect heater power, furnace temperature and material humidity sequences from historical operating data, calculate predicted values ​​using the nominal model of the switching nonlinear state-space prediction model, and define the difference between historical measured values ​​and predicted values ​​of furnace temperature and material humidity as residual sequences. Using the residual sequence as training samples, the parameters of the Gaussian mixture model are iteratively estimated using the expectation-maximization algorithm. The optimal number of Gaussian components is determined by the Bayesian information criterion, and the weights, mean, and covariance matrices of each Gaussian component are obtained.

3. The temperature adjustment method for the carbon raiser drying equipment according to claim 1, characterized in that, After acquiring analog signals of furnace temperature and material humidity, the data acquisition card converts them into digital signals.

4. The temperature adjustment method for the carbon raiser drying equipment according to claim 1, characterized in that, We selected the weighted L1 norm minimization algorithm as the sparse regression method.

5. The temperature adjustment method for the carbon raiser drying equipment according to claim 1, characterized in that, The switching nonlinear state-space prediction model consists of a set of subsystem models and a switching rule that determines which subsystem model is activated at any given time. Each subsystem model is described in nonlinear state-space form, and a dedicated nonlinear state-space subsystem model is established for each operating mode. ; in, For the equipment in the next control cycle The state vector, For the current control cycle The state vector, For the current control cycle The control input, The currently active working mode, It is a nonlinear function; the state vector includes furnace temperature and material humidity, and the control input includes heater power.

6. The temperature adjustment method for the carbon raiser drying equipment according to claim 1, characterized in that, In S3, the process of constructing a dictionary-based multi-objective optimization function includes the following steps: The first priority goal Set as the cumulative sum of control costs within the prediction time domain, with the cumulative sum in the following form: ; in, The heater power in the k-th control cycle. The heater power in the (k-1)th control cycle. A binary variable indicating whether the operating mode has switched. , and These are preset non-negative weighting coefficients; Second priority target Set as the sum of the squared errors between the predicted furnace temperature and the set temperature within the prediction time domain. The sum of squared errors is in the form of: ; in, This is the expected value of the predicted furnace temperature calculated using the polynomial chaotic expansion method. Set the temperature for the furnace.

7. The temperature adjustment method for the carbon raiser drying equipment according to claim 6, characterized in that, The process of setting probabilistic constraints represented by conditional value at risk includes the following steps: Set a safe upper limit for furnace temperature and safety lower limit Define a loss function L(T) related to temperature T, when the temperature is within the safe range. When the temperature is within the safe range, the loss function value L(T) is zero. When the temperature exceeds the safe range, the loss function value increases monotonically with the amount of excess. At confidence level Below, the conditional value at risk (CVaR) constraint is set as follows: ,in As a preset risk threshold, To at the confidence level Down, loss Conditional risk value.

8. The temperature regulation method for the carbon raiser drying equipment according to any one of claims 1-7, characterized in that, The process of solving multi-objective optimization problems that satisfy probabilistic constraints online includes the following steps: The dictionary-based multi-objective optimization problem is transformed into a mixed-integer nonlinear programming problem, where the working mode sequence is an integer decision variable and the heater power sequence is a continuous decision variable. By introducing slack variables, the non-smooth functions in the probabilistic constraints are transformed into smooth algebraic inequality constraints, making the transformed problem suitable for gradient-based nonlinear programming solvers. A hybrid algorithm is used to solve the problem, which combines the branch and bound method with sequential quadratic programming. At each node of the branch and bound method, the relaxed nonlinear programming subproblem is solved by sequential quadratic programming until an integer and continuous optimal solution that satisfies the preset convergence condition is obtained.

9. A temperature control system for a carbon raiser drying equipment, characterized in that, It includes a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the temperature adjustment method of the carbonizer drying equipment according to any one of claims 1-8 is implemented.

Images