Multivariable predictive control formaldehyde aqueous solution optimization control method

Abstract

The invention discloses a multi-variable predictive control formaldehyde aqueous solution optimization control method. The method comprises the following steps: step 1, firstly, carrying out data acquisition on a methanol aqueous solution system; step 2, communicating with an operator, and determining a target that the methanol rectification section and the formaldehyde production section need to be optimized; step 3, carrying out step test on the data; 4, modeling is carried out on the branch loops; 5, constructing a model matrix, and obtaining a convergence model; and step 6, putting into use, and adjusting model parameters according to time optimization. By adopting the multivariable predictive control method, the stable operation of the methanol rectification and formaldehyde aqueous solution production system can be better realized, the automation level of the system can be greatly improved, and the safe, stable and high-quality operation of the formaldehyde aqueous solution production system is ensured.

Description

Technical Field

[0001] The invention relates to an optimization control method for a formaldehyde aqueous solution product by multivariable predictive control, belonging to the technical field of chemical production control, and specifically to the control of two sections of methanol distillation and formaldehyde production in the production of formaldehyde aqueous solution. Background Art

[0002] Formaldehyde is one of the important carbon-1 chemicals. It is industrially produced by oxidation with methanol as raw material and is widely used in synthetic resins, coatings, synthetic pesticides and medicines. Shanxi Jinmei Tianyuan Chemical Co., Ltd. has an annual production of 25,000 tons of formaldehyde project. It uses refined methanol as raw material and electrolytic silver as catalyst. Refined methanol undergoes oxidation and dehydrogenation reactions at a high temperature of 620-690°C to generate formaldehyde gas, which is then continuously absorbed in the first, second and third absorption towers with stainless steel structured packing by soft water in countercurrent to obtain qualified formaldehyde aqueous solution products. Due to the lack of oxygen during the reaction and the high reaction temperature, the dehydrogenation and oxidation reactions occur simultaneously: CH3OH CH2O+H2-85.270kJ / mol CH3OH+1 / 2O2=== CH2O+H2O+156.557kJ / mol H2+O2=== H2O+214.827kJ / mol Reaction 1 starts at about 200°C, and reaction 2 proceeds at about 480°C, which is a reversible reaction. The progress of reaction 3 can make reaction 2 continue to proceed to the right, and by the heating effect of the ingredient steam introduced into the raw materials, the excess heat is removed from the reaction system to allow the reaction to proceed normally.

[0003] The raw material for synthesizing formaldehyde is methanol. Methanol distillation is an important processing step in formaldehyde production. Methanol distillation is an operation process that separates the mixture by obtaining purer components after multiple partial vaporizations and multiple partial condensations based on the different volatilities of different components in the same liquid mixture at the same temperature. As a physical separation device composed of a typical industrial distillation tower system, there is not only a coupling problem between the front and rear towers, but also heat integration between the towers. That is, the distillation process is a complex control process with multiple variables and strong coupling. At the same time, the influence of various disturbance factors is also obvious, which makes it impossible for the single-input and single-output control system of conventional PID to comprehensively consider and coordinate the control of multiple related variables, and it is difficult to meet the overall control requirements required in the production process of the device. For key process parameters such as the top tower temperature, reflux temperature, reflux volume, tower bottom liquid level and other important parameter indicators, there is a lack of attention, and more manual intervention by operators is required.

[0004] In order to overcome the various defects existing in the existing formaldehyde aqueous solution production process, improve the device's self-control rate, improve the device's anti-interference ability, and enhance the stability of process parameters. Multivariable predictive control technology is used to establish a multivariable predictive control mathematical model for the process based on historical operation data and actual operation experience, and the mathematical relationship between the controlled variables, interference variables, and operating variables is explored. The experience of excellent operating engineers is solidified, and on this basis, the "edge" optimization operation of the main process parameters is achieved, so that the device can be operated in the safest and most economical state, achieving energy saving and consumption reduction, increasing production capacity, improving the level of device intelligence, and reducing the labor intensity of operators. Summary of the invention

[0005] The purpose of the invention of this project is to overcome the shortcomings of existing control technology and provide a formaldehyde aqueous solution optimization control method based on multivariable predictive control.

[0006] The purpose of the present invention is achieved through the following technical solutions A method for optimizing and controlling a formaldehyde aqueous solution based on multivariable predictive control comprises the following steps:

[0007] Step 1: Add a multivariable predictive control server to the DCS system of the formaldehyde water device to collect all variables of the methanol distillation and formaldehyde production sections. This server is the subsequent multivariable predictive control server, and the corresponding operation interface is added to the original DCS interface.

[0008] Step 2: Communicate with the operators to determine the targets that need to be optimized in the methanol distillation and formaldehyde production sections (pre-tower top pressure, pre-tower reflux flow, reflux tank level, pressurized tower reflux tank level, pressurized tower top temperature, atmospheric tower top temperature, atmospheric tower reflux tank level, new atmospheric tower reflux tank level, new atmospheric tower kettle temperature, new atmospheric tower reflux flow, absorption tower bottom temperature, desalted water flow, evaporator level, condensate storage tank level), and determine the operating variables and interference variables corresponding to the controlled variables based on the process flow and operator experience.

[0009] The operating variables are selected according to the following principles: Step 3: Perform a step test on the data, and try to perform the step test continuously; try to move only one variable at a time during the test. If more than two variables need to be adjusted at the same time, then these variables should be unrelated; valve and variable saturation should be avoided during the test. During the test, the signal-to-noise ratio should be no less than 6, and should be greater than 10 as much as possible. The signal-to-noise ratio refers to the ratio of the change amplitude of the controlled variable caused by the test to the normal noise amplitude of the controlled variable. If the signal-to-noise ratio is too small, the step test will have little significance for the modeling effect.

[0010] Modeling 4: Modeling. It is divided into real-time data modeling and historical data modeling. Generally, historical data modeling (data during step test) is used to model each corresponding loop and select appropriate interference variables. For liquid level and pressure integral variables, the integral dynamic curve is relatively simple, a straight line. Temperature and other parameters are displayed as the gain effect of the model. According to the comparison between the predicted value and the actual value of the model, the better the overlap of the two curves, the better the quality of the model. When comparing, it is important to see whether the change trends of the predicted and actual data are consistent. If the trend of the change is consistent, but the amplitude of the change is slightly different, this situation is not serious. Because the controller will compensate for the deviation in each operating cycle after it is put into use, the difference in amplitude can be eliminated.

[0011] Step 5: Establish a model matrix and restrict the corresponding operating variables and controlled variables. Select the steady-state time of the model. For example, the steady-state time of the reflux flow is 12 minutes, while the steady-state time of the top temperature is 60 minutes, so 60 minutes should be selected as the steady-state time of the model matrix. Select the control frequency of the model, which is generally to run the controller once per minute, which can be adjusted according to specific working conditions. After all models are trained, perform a gain matrix analysis of the large matrix to check whether the positions of the model matrix variables are appropriate to prevent the calculation time from being too long during optimization, or even the model from failing to converge. For the model matrix of methanol distillation and formaldehyde production, the steady-state time is selected to be 60 minutes, and the control frequency is once per minute. The specific dynamic matrix algorithm is divided into three stages: control, prediction, and correction.

[0012] (1) Control: In the case of step response data at N sampling points of the system starting from time 0, the output prediction Y(k) of the system at time k for the future time p can be calculated by the dynamic matrix A composed of the system output prediction initial value Y0(k) at time k and M continuous input increment sequences △U(k) and step response data. The calculation expression is as follows: Y(k)=Y0(k)+A△U(k)

[0013] Where A is the dynamic matrix of DMC, P is the rolling optimization time domain length, and M is the control time domain length.

[0014] (2) Prediction: In the actual system, at each moment k, the controller will determine a sequence of M control increments from that moment on for predictive control. It is obtained by using the optimal control solution method. Each time the rolling optimization is performed, it can be regarded as the optimal solution to the problem of determining the future control increments based on the current known information. The basic idea of ​​optimal control is to determine the objective function and solve it according to the constraints. In simple applications of DMC, the solution is generally obtained under unconstrained conditions, and the objective function of the DMC control increment solution is usually defined as controlling the future M control increments through the optimization index so that the future P output prediction values ​​are as close to the expected trajectory as possible. w(k+i),i= 0…P; W p (k)=[w(k+1) w(k+2)…w(k+P)] T Objective function of incremental solution of DMC control

[0015] The objective function is defined as: The sum of the squares of the Q norm plus The sum of the squares of the R norms of Q and R is the selected positive definite matrix; Q=diag[Q1,Q2…Q p ] is the error weight matrix, and each element of Q i is the weight coefficient of the squared error of the system output at the i-th moment, R=diag[R1,R2… R p ] is the control weight matrix, and each element R j It is the weight coefficient of the square value of the control increment at the jth moment, which is introduced to reduce the fluctuation of the control effect.

[0016] To find the optimal extreme value is to differentiate the optimization variable so that it is equal to 0. Find the partial differential and get

[0017] (3) Correction During operation, multivariable predictive control collects real-time information to calibrate the prediction model, and then performs new optimization. At time k, the calculated optimal control increment △U(k) is applied to the model, and the model output prediction for the next N time periods can be obtained, which is recorded as y N1 (k), take the predicted output value y1(k+1|k) at the next moment and compare it with the actual output value y(k+1) of the controlled variable at the next moment to get the predicted error value e(k+1): e(k+1)= y(k+1)-y1(k+1|k) The error e(k+1) is weighted to obtain a weighted error sequence to correct the prediction model. The calculation expression is as follows: cor (k+1)=Y N1 (k)+he(k+1) where y cor (k+1) is the corrected model prediction value at the next moment, and h is the N-dimensional error weighted sequence. cor (k+1) Introduce the shift matrix S to get the initial value of the next prediction, y N0 (k+1) is used as the initial value of the model prediction at the next moment, and the calculation expression is as follows:N0 (k+1)=S y cor (k+1) The S matrix is

[0018] The entire control calculation process is to measure the actual variable and calculate the error, model prediction correction, shift setting at this moment of the prediction initial value, calculate the control increment, and finally output the calculated prediction value.

[0019] Step 6: Establish two model controllers for methanol distillation and formaldehyde production sections; optimize the model in real time according to the model prediction value. Modify and adjust the prediction model according to the actual effect. For integral variables, adjust the integral rate of the controlled variable, the speed at which the controlled variable is pushed to the set value. The larger the integral rate, the faster it approaches; adjust the rotation factor to correct the proportion of the prediction deviation. Adjust the dynamic deviation factor and steady-state deviation factor for non-integral variables. The operating parameters include (maximum step of single change, damping coefficient, priority control level between variables, etc.) and optimization to improve control accuracy and system stability.

[0020] The working process and principle of the present invention are: Multivariable predictive control (MPC) is a control method that can be used to optimize and improve complex industrial processes. The following is an overview of the working process and principle of MPC in the formaldehyde aqueous solution project: Data collection and modeling: First, historical data on variables related to the formaldehyde aqueous solution production process needs to be collected. These variables may include temperature, pressure, flow rate, chemical concentration, etc. The collected data is used to build a mathematical model.

[0021] Multivariate Modeling: Using the collected data, a multivariate model was developed to describe the formaldehyde aqueous solution production process. This model is able to relate the input variables to the output variables to describe the dynamic behavior of the system.

[0022] Controller Design: Based on the established multivariable model, a predictive controller is designed. The MPC controller is able to generate the best control strategy by predicting the future system response. The controller will consider the current state of the system and the desired goal and use the model to make predictions to calculate the optimal control action.

[0023] Optimization and constraints: When designing a controller, you also need to consider the performance and operational limitations of the system. You can set some optimization goals and use these goals as constraints for the optimization problem.

[0024] Real-time execution: During real-time operation, the MPC controller will make predictions based on the current system status and calculate the optimal control strategy. The output of the controller will be sent to the actuator as a control instruction to adjust the operating variables in the formaldehyde aqueous solution production process.

[0025] Feedback and Iteration: The MPC controller is a model predictive controller that performs feedback and iteration based on real-time system responses. By continuously collecting and updating data from the actual operation process, the model can be corrected and improved to improve the performance and adaptability of the controller.

[0026] In general, multivariable predictive control can achieve precise control and optimization of formaldehyde aqueous solution production and improve the efficiency and quality of the production process by establishing mathematical models, designing controllers, optimizing and constraining conditions, and performing and feedback iterations in real time. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 This is the matrix model diagram of the formaldehyde production section Figure 2 This is the picture of the liquid level of the atmospheric tower reflux tank before it is put into use Figure 3 This is the picture of the liquid level of the atmospheric tower reflux tank after it was put into use Figure 4 The bottom temperature of the absorption tower before commissioning Figure 5 The bottom temperature of the absorption tower after commissioning DETAILED DESCRIPTION

[0027] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0028] A method for optimizing and controlling a formaldehyde aqueous solution based on multivariable predictive control comprises the following steps: Step 1: Add a multivariable predictive control server to the DCS system of the formaldehyde water device to collect all variables of the methanol distillation and formaldehyde production sections. This server is the subsequent multivariable predictive control server, and the corresponding operation interface is added to the original DCS interface.

[0029] Step 2: Communicate with the operators to determine the targets that need to be optimized in the methanol distillation and formaldehyde production sections (pre-tower top pressure, pre-tower reflux flow, reflux tank level, pressurized tower reflux tank level, pressurized tower top temperature, atmospheric tower top temperature, atmospheric tower reflux tank level, new atmospheric tower reflux tank level, new atmospheric tower kettle temperature, new atmospheric tower reflux flow, absorption tower bottom temperature, desalted water flow, evaporator level, condensate storage tank level), and determine the operating variables and interference variables corresponding to the controlled variables based on the process flow and operator experience.

[0030] The operating variables are selected according to the following principles: Step 3: Perform a step test on the data, and try to perform the step test continuously; try to move only one variable at a time during the test. If more than two variables need to be adjusted at the same time, then these variables should be unrelated; valve and variable saturation should be avoided during the test. During the test, the signal-to-noise ratio should be no less than 6, and should be greater than 10 as much as possible. The signal-to-noise ratio refers to the ratio of the change amplitude of the controlled variable caused by the test to the normal noise amplitude of the controlled variable. If the signal-to-noise ratio is too small, the step test will have little significance for the modeling effect.

[0031] Modeling 4: Modeling. It is divided into real-time data modeling and historical data modeling. Generally, historical data modeling (data during step test) is used to model each corresponding loop and select appropriate interference variables. For liquid level and pressure integral variables, the integral dynamic curve is relatively simple, a straight line. Temperature and other parameters are displayed as the gain effect of the model. According to the comparison between the predicted value and the actual value of the model, the better the overlap of the two curves, the better the quality of the model. When comparing, it is important to see whether the change trends of the predicted and actual data are consistent. If the trend of the change is consistent, but the amplitude of the change is slightly different, this situation is not serious. Because the controller will compensate for the deviation in each operating cycle after it is put into use, the difference in amplitude can be eliminated.

[0032] Step 5: Establish a model matrix and restrict the corresponding operating variables and controlled variables. Select the steady-state time of the model. For example, the steady-state time of the reflux flow is 12 minutes, while the steady-state time of the top temperature is 60 minutes, so 60 minutes should be selected as the steady-state time of the model matrix. Select the control frequency of the model, which is generally to run the controller once per minute, which can be adjusted according to specific working conditions. After all models are trained, perform a gain matrix analysis of the large matrix to check whether the positions of the model matrix variables are appropriate to prevent the calculation time from being too long during optimization, or even the model from failing to converge. For the model matrix of methanol distillation and formaldehyde production, the steady-state time is selected to be 60 minutes, and the control frequency is once per minute. The specific dynamic matrix algorithm is divided into three stages: control, prediction, and correction.

[0033] (1) Control: In the case of step response data at N sampling points of the system starting from time 0, the output prediction Y(k) of the system at time k for the future time p can be calculated by the dynamic matrix A composed of the system output prediction initial value Y0(k) at time k and M continuous input increment sequences △U(k) and step response data. The calculation expression is as follows: Y(k)=Y0(k)+A△U(k)

[0034] Where A is the dynamic matrix of DMC, P is the rolling optimization time domain length, and M is the control time domain length.

[0035] (2) Prediction: In the actual system, at each moment k, the controller will determine the M control increment sequence from that moment on for predictive control. It is obtained by the solution method of optimal control. In each rolling optimization, it can be regarded as the optimal solution problem of determining the future control increment based on the current known information. The basic idea of ​​optimal control is to determine the objective function and solve it according to the constraints. In simple applications of DMC, it is generally solved under unconstrained conditions, and the objective function of DMC control increment solution is usually defined as controlling the future M control increments through optimization indicators so that the future P output prediction values ​​are as close as possible to the expected trajectory w(k+i), i= 0…P; W p (k)=[w(k+1) w(k+2)…w(k+P)] T Objective function of incremental solution of DMC control

[0036] The objective function is defined as: The sum of the squares of the Q norm plus The sum of the squares of the R norms of Q and R is the selected positive definite matrix; Q=diag[Q1,Q2…Q p ] is the error weight matrix, and each element of Q i is the weight coefficient of the squared error of the system output at the i-th moment, R=diag[R1,R2… R p ] is the control weight matrix, and each element R j It is the weight coefficient of the square value of the control increment at the jth moment, which is introduced to reduce the fluctuation of the control effect.

[0037] To find the optimal extreme value is to differentiate the optimization variable so that it is equal to 0. Find the partial differential and get

[0038] (3) Correction During operation, multivariable predictive control collects real-time information to calibrate the prediction model, and then performs new optimization. At time k, the calculated optimal control increment △U(k) is applied to the model, and the model output prediction for the next N time periods can be obtained, which is recorded as y N1 (k), take the predicted output value y1(k+1|k) at the next moment and compare it with the actual output value y(k+1) of the controlled variable at the next moment to get the predicted error value e(k+1): e(k+1)= y(k+1)-y1(k+1|k) The error e(k+1) is weighted to obtain a weighted error sequence to correct the prediction model. The calculation expression is as follows: cor (k+1)=Y N1 (k)+he(k+1) where y cor (k+1) is the corrected model prediction value at the next moment, and h is the N-dimensional error weighted sequence. cor (k+1) Introduce the shift matrix S to get the initial value of the next prediction, y N0 (k+1) is used as the initial value of the model prediction at the next moment, and the calculation expression is as follows: N0 (k+1)=S y cor (k+1) The S matrix is

[0039] The entire control calculation process is to measure the actual variable and calculate the error, model prediction correction, shift setting at this moment of the prediction initial value, calculate the control increment, and finally output the calculated prediction value.

[0040] Step 6: Establish two model controllers for methanol distillation and formaldehyde production sections; optimize the model in real time according to the model prediction value. Modify and adjust the prediction model according to the actual effect. For integral variables, adjust the integral rate of the controlled variable, the speed at which the controlled variable is pushed to the set value. The larger the integral rate, the faster it approaches; adjust the rotation factor to correct the proportion of the prediction deviation. Adjust the dynamic deviation factor and steady-state deviation factor for non-integral variables. The operating parameters include (maximum step of single change, damping coefficient, priority control level between variables, etc.) and optimization to improve control accuracy and system stability.

[0041] The beneficial effects of the present invention are: more stable working conditions, reduced unit consumption of refined methanol, and energy saving and consumption reduction.

[0042] Figure 1 It represents the matrix model of the formaldehyde production section, where each curve in the matrix is ​​a 60-minute model curve. The upper right corner is the corresponding gain coefficient, which is divided into positive and negative gains, representing the influence of the operating variable and interference variable on the controlled variable. A blank indicates no relationship.

[0043] Figure 2 and Figure 3 They are comparisons of the liquid level in the reflux tank of the atmospheric tower before and after the multivariable predictive control system was put into use. It can be clearly seen that the liquid level in the reflux tank of the atmospheric tower is very stable after it was put into use. If the liquid level is stable, the overall heat balance of the tower is achieved, and the gas and liquid distribution of the products in the tower is uniform, thus ensuring the product separation effect.

[0044] Figure 4 and Figure 5 They are respectively a comparison of the bottom temperature of the absorption tower before and after the multivariable predictive control system is put into use. The temperature directly indicates that the absorption temperature will be kept stable to the optimal absorption temperature at the current stage, thereby improving and ensuring product quality.

[0045] Multivariable predictive control solves complex industrial control problems that are difficult to solve with conventional PID control, and achieves good control effects. The fluctuation range of key loop standards is significantly reduced, the standard deviation of the controlled variables is reduced, the systems within each tower are stabilized, and the entire formaldehyde aqueous solution production system is coordinated and optimized to ensure safe and stable production of the system equipment. While reducing the labor of operators, the automation and intelligence level of the device is greatly improved.

Claims

1. A multivariable predictive control method for optimizing formaldehyde aqueous solution, comprising the following steps: Step 1: Add a multivariable predictive control server to the DCS system of the formaldehyde water device to collect all variables of the methanol distillation and formaldehyde production sections. This server is the subsequent multivariable predictive control server, and the corresponding operation interface is added to the original DCS interface. Step 2: Communicate with the operator to determine the goals that need to be optimized in the methanol distillation and formaldehyde production sections, and determine the operating variables and interference variables corresponding to the controlled variables based on the process flow and operator experience. The operating variables are selected according to the following principles: Step 3: Perform step tests on the data, and try to perform step tests continuously; try to move only one variable at a time during the test. If more than two variables need to be adjusted at the same time, these variables should be unrelated; valve and variable saturation should be avoided during the test. During the test, the signal-to-noise ratio should be no less than 6, and try to be greater than 10. The signal-to-noise ratio refers to the ratio of the change amplitude of the controlled variable caused by the test to the normal noise amplitude of the controlled variable. If the signal-to-noise ratio is too small, the step test has little significance for the modeling effect. Modeling 4: Modeling. It is divided into real-time data modeling and historical data modeling. Generally, historical data modeling (data during step tests) is used to model each corresponding loop and select appropriate interference variables. For liquid level and pressure integral variables, the integral dynamic curve is relatively simple, a straight line. Temperature and other displays are the gain effects of the model. According to the comparison between the predicted value and the actual value of the model, the better the overlap of the two curves, the better the quality of the model. When comparing, it is important to see whether the change trends of the predicted and actual data are consistent. If the trends of the changes match, but the magnitude of the changes is slightly different, this situation is not serious. Because after the controller is put into use, the deviation will be compensated in each operation cycle, which can eliminate the difference in magnitude. Step 5: Establish a model matrix and restrict the corresponding operating variables and controlled variables. Select the steady-state time of the model. For example, the steady-state time of the reflux flow is 12 minutes, and the steady-state time of the top temperature is 60 minutes, so 60 minutes should be selected as the steady-state time of the model matrix. Select the control frequency of the model, generally run the controller once per minute, which can be adjusted according to specific working conditions. After all models are trained, perform a gain matrix analysis of the large matrix to check whether the positions of the variables in each matrix of the model are appropriate to prevent the calculation time from being too long during optimization, or even the model from failing to converge. Step 6: Establish two model controllers for the methanol distillation and formaldehyde production sections; optimize the model in real time according to the model prediction value. Modify and adjust the prediction model according to the actual effect. For integral variables, adjust the integral rate of the controlled variable, the speed at which the controlled variable is pushed to the set value. The larger the integral rate, the faster it approaches; Adjust the rotation factor to correct the proportion of the predicted deviation. Adjust the dynamic deviation factor and steady-state deviation factor for non-integral variables. The operating parameters include (maximum step of single change, damping coefficient, priority control level between variables, etc.) and optimization to improve control accuracy and system stability.

2. The method according to claim 1, wherein the controlled variables determined in the described step 2 are the pre-tower top pressure, the pre-tower reflux flow, the reflux tank level, the pressure tower reflux tank level, the pressure tower top temperature, the atmospheric tower top temperature, the atmospheric tower reflux tank level, the new atmospheric tower reflux tank level, the new atmospheric tower kettle temperature, the new atmospheric tower reflux flow, the absorption tower bottom temperature, the desalted water flow, the evaporator level, and the condensed water storage tank level.

3. The method according to claim 1, wherein in the model matrix of methanol distillation and formaldehyde production in step 5, the steady-state time is selected as 60 minutes and the control frequency is once per minute.

Images