Cement production decomposing furnace temperature control system parameter optimization method and medium

Abstract

The invention discloses a parameter optimization method for a temperature control system of a cement production decomposing furnace, and the method comprises the steps: employing an MPC controller to control the temperature, carrying out the real-time adjustment and optimization of the optimal parameter of the MPC controller through a reinforcement learning algorithm, and carrying out the execution of the optimal parameter through the MPC controller, and carrying out the control of the temperature. Relevant control parameters of MV, CV and DV of the MPC are set in real time by using a reinforcement learning algorithm, and a given value of the coal feeding amount of the decomposition furnace is obtained through the MPC controller, so that the material temperature of the sintering decomposition furnace is in a stable state, the temperature of the decomposition furnace is stable, and the production parameters are constant so as to improve the quality of cement.

Description

Technical Field

[0001] This invention relates to the field of cement production control, and in particular to a method and medium for optimizing parameters of a temperature control system for a cement production decomposition furnace. Background Art

[0002] Model predictive control (MPC) algorithms are applied to automatic control systems for cement production calcination models because they can take into account input and output constraints, perform rolling time-domain optimization, and provide optimal (local) solutions.

[0003] Traditional MPC control parameter tuning typically employs manual or semi-automatic methods. However, the parameters obtained through this method have a certain degree of uncertainty under different operating conditions and unknown disturbances. Therefore, a parameter tuning method that does not rely on human experience, possesses intelligent thinking, and can adaptively select the optimal parameters according to environmental changes is needed. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and medium for optimizing the parameters of the temperature control system of a cement production decomposition furnace. This method can automatically adjust and optimize the parameters of the MPC based on real-time production line data, thereby providing a foundation for the MPC controller to achieve stable temperature control.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a method for optimizing parameters of a temperature control system for a cement production decomposition furnace, which uses an MPC controller to control the temperature, and uses a reinforcement learning algorithm to adjust and optimize the optimal parameters of the MPC controller in real time, and the MPC controller executes the optimal parameters to control the temperature.

[0006] The steps involved in adjusting the optimal parameters of the optimization MPC controller in real time using reinforcement learning algorithms are as follows:

[0007] S1. Obtain real-time production line data;

[0008] S2. Obtain the state-space model;

[0009] The optimal parameters for S3 and MPC are obtained through parameter optimization.

[0010] In step S1, the real-time acquired cement production control line data includes MV data, CV data, and DV data; where MV data refers to the data corresponding to the operating variables of MPC, CV data refers to the controlled variable data of MPC, and DV data refers to the disturbance variable data of MPC.

[0011] MV data includes the setpoint for coal feed rate in the decomposer; dv includes filtering of the raw material silo Roots blower outlet pressure and CKK feed rate and speed setpoints; cv includes the decomposer outlet temperature B, C6 material temperature, and preheater outlet CO.

[0012] Step S2, obtaining the state-space model, includes establishing a model transformed from the first-order transfer function into the state space; the first-order time-delay transfer system function is as follows:

[0013]

[0014] Among them, K p T is the gain coefficient. p T is the time constant. d The time delay coefficient is obtained through system identification.

[0015] The system function is transformed into state-space equation form as follows:

[0016]

[0017] Δcv=C*x p

[0018] Where x is the initial system state vector x p A is the predicted system state vector; A is the system state transition matrix; mv is the input matrix, corresponding to the given values ​​of the opening and closing degrees of the air valve and gas valve in the preheating section of the heating furnace; B is the input transition matrix; C is the output transition matrix; cv is the output matrix, corresponding to the temperature in the preheating section of the heating furnace; Δcv is the output difference.

[0019] Step S3 includes: using the DDPG algorithm under the Actor-Critic framework with continuous input and output space, adjusting the parameters of MPC in real time, and then inputting them into the MPC controller to achieve precise control of the target.

[0020] In step S3, during the oxygen-enriched combustion control of the heating furnace, the deep learning agent receives the raw material Kurox blower outlet pressure and CKK feed rate and speed setpoints generated by the interaction between the surrounding environment and the agent. By setting the desired C6B material temperature, and through the MPC control algorithm of the upper DDPG, the final output of the decomposition furnace coal feed rate setpoint is achieved, thus realizing constant material temperature control.

[0021] The deep reinforcement learning algorithm DDPG used in step S3 includes: the agent in the DDPG algorithm consists of an Actor network and a Critic network. The Actor network, as the action decision unit, outputs MPC parameters to adjust the working state of the system. The Critic network evaluates the action based on the operation results of the upper control system and updates the Actor network and Critic network according to the evaluation results.

[0022] During the optimization process, the state space is defined as a vector consisting of environmental data such as the given value of coal feed rate in the decomposer, the outlet pressure of the Roots blower in the raw material silo, the given value of feed rate and speed of CKK, the outlet temperature of the decomposer B, the material temperature of C6B, and the CO content at the outlet of the preheater; the action space is defined as the MPC control parameters.

[0023] A computer storage medium storing a computer program that, when executed by a processor, implements the method.

[0024] The advantages of this invention are: by using reinforcement learning algorithms to tune the relevant control parameters of the MPC's manipulated variable (MV), controlled variable (CV), and disturbance variable (DV) in real time, the MPC controller obtains the given value of the coal feed rate to the decomposition furnace, so that the temperature of the calcined material in the decomposition furnace is in a stable state, thus stabilizing the decomposition furnace temperature and keeping the production parameters constant to improve the quality of cement. Attached Figure Description

[0025] The following is a brief explanation of the contents of each of the accompanying drawings and the markings in the drawings:

[0026] Figure 1 This is a flowchart illustrating the optimization method of the present invention;

[0027] Figure 2 This is a schematic diagram illustrating the framework principle of the Actor network of this invention;

[0028] Figure 3 This is a schematic diagram illustrating the framework principle of the Critic network of this invention. DETAILED DESCRIPTION

[0029] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and the description of the preferred embodiments.

[0030] This invention proposes a strategy for MPC controller parameters using a genetic algorithm (GA). The strategy utilizes reinforcement learning to tune the relevant control parameters of the MPC's manipulated variable (MV), controlled variable (CV), and disturbance variable (DV) in real time. The MPC controller obtains the given value of the coal feed rate to the decomposer, thereby ensuring that the temperature of the C6 material in the decomposer remains stable.

[0031] like Figure 1 As shown, an MPC controller is used to control the temperature. The optimal parameters of the MPC controller are adjusted and optimized in real time through a reinforcement learning algorithm, and the MPC controller executes the optimal parameters to control the temperature.

[0032] The steps involved in adjusting the optimal parameters of the optimization MPC controller in real time using reinforcement learning algorithms are as follows:

[0033] S1. Obtain real-time production line data;

[0034] S2. Obtain the state-space model;

[0035] The optimal parameters for S3 and MPC are obtained through parameter optimization.

[0036] In step S1, real-time data of the cement production control line is acquired, including MV, OV, and DV data; where MV data refers to the data corresponding to the operating variables of MPC; OV data refers to the controlled variable data of MPC; and DV data refers to the disturbance variable data of MPC.

[0037] MV data includes the setpoint for the coal feed rate to the precalciner; dv includes filtering of the raw material silo Roots blower outlet pressure and CKK feed rate and speed setpoints; cv includes the precalciner outlet temperature B, C6 material temperature, and preheater outlet CO. The CKK (CONCHKawasaki Kiln system) is a waste treatment technology that utilizes a new dry-process kiln system in the cement industry to treat municipal solid waste, effectively treating and utilizing the waste gas, ash, and leachate generated from waste incineration.

[0038] The decomposition furnace has two outlet temperatures, A (left) and B (right). Outlet temperature B refers to one of the temperatures. C6 material refers to the material of the sixth stage of the six-stage preheater system. Preheater outlet CO refers to carbon monoxide at the preheater outlet.

[0039] Step S2, obtaining the state-space model, includes establishing a model transformed from the first-order transfer function into the state space; the first-order time-delay transfer system function is as follows:

[0040]

[0041] Among them, K p T is the gain coefficient. p T is the time constant. d The time delay coefficient is obtained through system identification. The above system function is transformed into state-space equation form as follows:

[0042]

[0043] Δcv=C*x p

[0044] Where x is the initial system state vector x p A is the predicted system state vector; A is the system state transition matrix; mv is the input matrix, corresponding to the given values ​​of the opening and closing degrees of the air valve and gas valve in the preheating section of the heating furnace; B is the input transition matrix; C is the output transition matrix; cv is the output matrix, corresponding to the temperature in the preheating section of the heating furnace; Δcv is the output difference.

[0045] The MPC parameter optimization framework based on reinforcement learning: After system identification and with the transfer function determined, the MPC control parameters are optimized. The selection of MPC control parameters has a significant impact on the control effect of the system, but their values ​​are difficult to determine. The adjustment process is not only time-consuming and laborious, but also cannot adapt to various different working conditions.

[0046] This paper analyzes the principles of MPC control theory and DDPG deep reinforcement learning algorithm. Considering that the input and output spaces of the heating furnace control problem are continuous, the DDPG algorithm under the Actor-Critic (AC) framework with continuous input and output spaces is used. That is, the DDPG deep reinforcement learning algorithm is selected to adjust the important parameters of MPC in real time, and then input them into the MPC controller to achieve precise control of the target.

[0047] In the oxygen-enriched combustion control of the heating furnace, the deep learning agent receives the raw material Kurox blower outlet pressure and CKK feed rate and speed setpoints generated by the interaction between the surrounding environment and the agent. By setting the desired C6B material temperature, and through the MPC control algorithm of the upper DDPG, the final output of the decomposition furnace coal feed rate setpoint is achieved, thus realizing constant material temperature control.

[0048] The Deep Deterministic Policy Gradient (DDPG) algorithm, as a model-free DRL method, can output continuous actions. The agent in the DDPG algorithm consists of an Actor network and a Critic network. The Actor network, acting as the action decision unit, outputs MPC parameters to adjust the system's operating state. The Critic network evaluates the action based on the results of the upper-level control system's operation, updating both the Actor and Critic networks accordingly. During continuous system operation, the DDPG network continuously improves its mechanism through interaction with the environment, constantly optimizing the control performance of the upper-level control system by combining MPC control parameters, ultimately achieving the optimal parameter combination for system operation.

[0049] In the system, the state space is defined as a vector comprising environmental data such as the given value of the coal feed rate to the decomposer, the outlet pressure of the Roots blower in the raw material silo, the given values ​​of the feed rate and speed of CKK, the outlet temperature B of the decomposer, the material temperature C6B, and the CO content at the preheater outlet. C6B material refers to the material temperature on the right side of the sixth stage of the six-stage preheater system.

[0050] The action space is defined by MPC control parameters, as shown in the table below:

[0051]

[0052]

[0053]

[0054]

[0055] ①Actor Network

[0056] In this framework, the Actor network outputs the optimal control parameters for MPC based on the system's state information. The network output is a 12-dimensional MPC parameter vector, using a Transformer structure, as shown below. Figure 2 As shown.

[0057] ②Critic Network

[0058] The basic process of the Critic network involves the agent interacting with environmental information, perceiving the corresponding environmental state information, and then outputting MPC parameters after training with the Actor network. At the next time step, the environmental state information changes, and the agent receives a reward. Based on the accumulated reward value, the Critic network updates its parameters. The Actor network then updates its weights in the same direction as the Critic network, and this process is repeated until the Actor network performs well. Compared to the Actor network, the Critic network structure includes additional MPC parameters as input, as detailed below. Figure 3 As shown.

[0059] ③ Reward function

[0060] For a heating furnace control system, the reward function guides the agent to make correct control behaviors and directly determines whether the agent can learn the optimal MPC parameters.

[0061] The positive incentive for actions adopts a tiered reward design, defining the initial reward function as 0:

[0062] (1) Temperature fluctuation at the outlet of the decomposition furnace

[0063] When the fluctuation range of the decomposition furnace outlet temperature B does not exceed 0.5%, set the reward to +20. When the fluctuation range of the gas flow rate is between 0.5% and 1%, set the reward to +10.

[0064] (2) C6B material temperature rolling prediction

[0065] The rolling prediction cost function for C6B material temperature is as follows. Both the cost function and the evaluation index adopt the rolling time domain RMSE. The cost function and the RMSE are defined as follows:

[0066]

[0067] Among them, y i y represents the C6B material temperature prediction vector obtained by solving mv in the i-th reciprocal step of MPC. o This represents the measured target vector of C6B material temperature. n represents the length of the reverse-ordered MPC calculation iterations. When RMSE ≤ 0.1, the reward is set to +20.

[0068] (3) Other situations

[0069] In other cases, the reward is -20.

[0070] After obtaining the optimal control parameters, MPC closed-loop testing is performed to ultimately determine the quality of the control parameters.

[0071] After the parameters of this invention are optimized, the MPC control effect is improved compared with the previous scheme, such as the decrease in the temperature fluctuation of C6B material and the outlet temperature of the decomposition furnace.

[0072] Based on a production line example, the above solution is implemented as follows:

[0073] (1) Obtain actual production line operation data from the actual control system

[0074] Taking the cement production data from 2024-02-02 00:00:01 to 2024-02-02 23:59:59 as an example, obtain the data for the following fields.

[0075]

[0076]

[0077] (2) Obtain the coal feeding control system model of the precalciner.

[0078] The model uses a first-order transfer function state-space equation, primarily representing the coal feeding loop of the preheater in a calcination model. Taking the raw material Kurroot blower outlet pressure and CKK feed rate / speed as disturbances, this model serves as an example. After identification, the model matrix is ​​as follows:

[0079]

[0080]

[0081] This model is a 3-input, 3-output state-space matrix model. The inputs include one mv and two dv, and the outputs include three cv.

[0082] mv includes the given value of coal feed rate to the precalciner;

[0083] dv includes filtering of the outlet pressure of the raw material silo Roots blower and filtering of the CKK feed rate and speed setpoint;

[0084] cv includes the decomposition furnace outlet temperature B, C6 material temperature, and preheater outlet CO.

[0085] (3) MPC Online Optimization Framework

[0086] Optimization is achieved through an optimization framework.

[0087] (4) Obtain the control parameters and use the obtained parameters in the actual control system for control.

[0088] Obviously, the specific implementation of this invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution of this invention are within the protection scope of this invention.

Claims

1. A method for optimizing parameters of a temperature control system for a cement production decomposition furnace, characterized in that: The temperature is controlled by an MPC controller. The optimal parameters of the MPC controller are adjusted and optimized in real time through a reinforcement learning algorithm, and the MPC controller executes the optimal parameters to control the temperature.

2. The method for optimizing parameters of a cement production decomposition furnace temperature control system as described in claim 1, characterized in that: The steps involved in adjusting the optimal parameters of the optimization MPC controller in real time using reinforcement learning algorithms are as follows: S1. Obtain real-time data from the cement production control line; S2. Obtain the state-space model; The optimal parameters for S3 and MPC are obtained through parameter optimization.

3. The method for optimizing parameters of a cement production decomposition furnace temperature control system as described in claim 2, characterized in that: In step S1, the real-time acquired cement production control line data includes MV data, CV data, and DV data; where MV data refers to the data corresponding to the operating variables of MPC, CV data refers to the controlled variable data of MPC, and DV data refers to the disturbance variable data of MPC.

4. The method for optimizing parameters of a cement production decomposition furnace temperature control system as described in claim 3, characterized in that: MV data includes the setpoint for the coal feed rate of the decomposer; DV data includes the filter for the outlet pressure of the Roots blower in the raw material silo and the filter for the setpoint of the feed rate and speed of CKK; CV data includes the outlet temperature of the decomposer (B), the material temperature of C6, and the outlet temperature of the preheater (CO).

5. The method for optimizing parameters of a cement production decomposition furnace temperature control system as described in claim 2, characterized in that: Step S2, obtaining the state-space model, includes establishing a model that transforms the first-order transfer function into the state space; the first-order time-delay transfer function is as follows: Among them, K p T is the gain coefficient. p T is the time constant. d The time delay coefficient is obtained through system identification. The system function is transformed into state-space equation form as follows: Δcv=C*x p Where x is the initial system state vector x p A is the predicted system state vector; A is the system state transition matrix; mv is the input matrix, corresponding to the given values ​​of the opening and closing degrees of the air valve and gas valve in the preheating section of the heating furnace; B is the input transition matrix; C is the output transition matrix; cv is the output matrix, corresponding to the temperature in the preheating section of the heating furnace; Δcv is the output difference.

6. The method for optimizing parameters of a cement production decomposition furnace temperature control system as described in any one of claims 2-5, characterized in that: Step S3 includes: using the DDPG algorithm under the Actor-Critic framework with continuous input and output space, adjusting the parameters of MPC in real time, and then inputting them into the MPC controller to achieve precise control of the target.

7. The method for optimizing parameters of a cement production decomposition furnace temperature control system as described in claim 6, characterized in that: In the oxygen-enriched combustion control of the heating furnace, the deep learning agent receives the raw material Kurox blower outlet pressure and CKK feed rate and speed setpoints generated by the interaction between the surrounding environment and the agent. By setting the desired C6B material temperature, and through the MPC control algorithm of the upper DDPG, the final output of the decomposition furnace coal feed rate setpoint is achieved, thus realizing constant material temperature control.

8. The method for optimizing parameters of a cement production decomposition furnace temperature control system as described in claim 7, characterized in that: The deep reinforcement learning algorithm DDPG used in step S3 includes: the agent in the DDPG algorithm consists of an Actor network and a Critic network. The Actor network, as the action decision unit, outputs MPC parameters to adjust the working state of the system. The Critic network evaluates the action based on the operation results of the upper control system and updates the Actor network and Critic network according to the evaluation results.

9. The method for optimizing parameters of a cement production decomposition furnace temperature control system as described in claim 8, characterized in that: During the optimization process, the state space is defined as a vector consisting of environmental data such as the given value of coal feed rate in the decomposer, the outlet pressure of the Roots blower in the raw material silo, the given value of feed rate and speed of CKK, the outlet temperature of the decomposer B, the material temperature of C6B, and the CO content at the outlet of the preheater; the action space is defined as the MPC control parameters.

10. A computer storage medium, characterized in that: The computer storage medium stores a computer program, which, when executed by a processor, implements the method as described in any one of claims 1-9.

Images