A high-gain sliding mode variable structure methanol distillation temperature control method based on multi-classification voting strategy

Abstract

The application provides a high-gain sliding mode variable structure methanol rectification temperature control method of a multi-classification voting strategy, belongs to the field of chemical reaction rectification, and comprises the following steps: constructing a methanol rectification tower data model related to temperature and a sliding mode control rate; in step 2, based on the methanol rectification tower data model and observation values, the model is decomposed into a linear part and a nonlinear part, a high-gain observer is designed, an observer error with an actual value is calculated, and the error is used as a sliding mode control input; in step 3, a sliding mode surface is designed, and a sliding mode control rate is calculated; in step 4, for different working conditions, a suitable sliding mode control law is selected through a voting strategy; and in step 5, model simulation is carried out, and the performance index of the control system is verified. The application can solve the problems of manual adjustment of control parameters and diversified index control required by different working conditions, has strong robustness, and in the strategy space search process, the generation of a local optimal solution is avoided, and the calculation complexity is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of chemical reaction rectification, and particularly relates to a high-gain sliding mode variable structure methanol rectification temperature control method based on multi-classification voting strategy. BACKGROUND

[0002] The production conditions of the methanol rectification process are complex and changeable, and the temperature control requirements tend to be diverse. The traditional control method mainly relies on the experience of operators, and the switching and transition between different working conditions are realized manually. The accuracy is easily affected by the quality of personnel and limited by the processing capacity. In recent years, many scholars use data-driven and machine learning techniques to realize intelligent control of complex chemical production processes.

[0003] The rectification process usually involves different working conditions, and the different working conditions of the rectification temperature control process refer to different parameter control indexes given by the process operating parameter control due to product demand. To continuously adjust the temperature control of the entire rectification process, the control parameters need to be changed to obtain the expected effect. Since the temperature control of the entire process is very strict, the temperature control will directly affect the final product of methanol. Rapid adjustment of control parameters through algorithms will improve the efficiency of the entire process.

[0004] The three-tower rectification process is usually used in the crude methanol rectification process. The crude methanol with a concentration of about 93% in the crude methanol tank enters the pre-distillation tower through steam condensate water, and the lower pre-heating tower is reboiled to make the hot steam superheat and then enter the condenser at the top of the pre-distillation tower. Part of the material in the pre-distillation tower condenser enters the pre-distillation reflux tank and then enters the pre-distillation tower again through the reflux pump. The remaining crude methanol tower kettle liquid enters the pressurized tower through the pre-methanol pump. The internal tower kettle temperature of the pressurized tower is controlled at 130°C, and the tower top temperature is controlled at 121°C. Due to the different temperatures, the methanol at the top of the tower enters the atmospheric tower reboiler, and the steam is condensed into liquid and flows into the reflux tank of the pressurized tower. Part of the liquid in the reflux tank enters the pressurized tower again through the pressurized reflux pump. The other part of the product after passing through the condenser is cooled to 40°C, and finally forms high-precision refined methanol.

[0005] The rectification tower is a complex nonlinear system. Generally, the rectification tower has dozens of tower plates, and the model order is relatively high, which is not convenient for theoretical analysis and calculation. A practical and simplified model that can reflect the static and dynamic characteristics of the full-order model of the rectification tower is required. The temperature control of methanol rectification is usually selected on the tower plate that is most sensitive to temperature changes, i.e., the tower plate with the largest slope. Since the temperature change will directly affect the change of the rectification component concentration, rapid adjustment of the temperature has important significance for the entire rectification process. SUMMARY

[0006] The application provides a high-gain sliding mode variable structure methanol rectification temperature control method based on a multi-classification voting strategy, which solves the problems of manual adjustment of control parameters and diversified index control under different working conditions.

[0007] A high-gain sliding mode variable structure methanol rectification temperature control method based on a multi-classification voting strategy comprises the following steps:

[0008] Step 1: constructing a methanol rectification tower data model related to temperature and sliding mode control rate;

[0009] Step 2: based on the methanol rectification tower data model and observation value, decomposing the model into a linear part and a nonlinear part, and designing a high-gain observer;

[0010] Step 3: designing a sliding mode surface, calculating a sliding mode control rate, calculating the error between the observer and the actual value as a sliding mode control input; Step 4: for different working conditions, selecting a suitable sliding mode control law through a voting strategy;

[0011] Step 5: performing model simulation and verifying the performance index of the control system.

[0012] Preferably, the model assumes the following conditions:

[0013] 1) assuming that the flow rates of reflux and steam in the rectification tower remain unchanged, and there is no leakage of reflux and steam, and the condenser can completely condense;

[0014] 2) assuming that the liquid amount of each tray is constant, and the steam retention amount can be ignored;

[0015] 3) assuming that the efficiency of the tray is constant, the gas-liquid phase is in equilibrium, and the gas phase is an ideal gas;

[0016] 4) assuming that there is no heat loss, and the gas-liquid phase of the tray and the reboiler is completely mixed.

[0017] Further, the methanol rectification tower data model is constructed based on a dynamic material balance equation and a static material balance equation, and the dynamic balance equation is:

[0018]

[0019]

[0020]

[0021]

[0022]

[0023]

[0024]

[0025] The static balance equation is:

[0026] V = L + D

[0027] Where t is time, dt is differential of time, L is reflux flow, M d is the residence amount of the top condenser, M s is the residence amount of the reboiler, V is steam flow, X d is the liquid phase methanol concentration of the top condenser, X i is the liquid phase methanol concentration of the tray i (i = 1, 2, …, 6), X s is the liquid phase methanol concentration of the reboiler, Y i is the gas phase methanol concentration of the tray i (i = 1, 2, …, 6), Y s is the gas phase methanol concentration of the reboiler, D is the methanol product flow collected after the top condensation;

[0028] The relationship between the gas phase methanol concentration Y and the liquid phase methanol concentration X in the rectification tower can be expressed by a constant relative volatility:

[0029]

[0030]

[0031] Where a c is the volatility coefficient.

[0032] The mathematical model of the methanol rectification tower is preferably:

[0033]

[0034] Where a is the top temperature change rate coefficient, b is the top temperature coefficient, c is the sliding mode control rate coefficient, a, b and c are all constants, represents the derivative of the top temperature change rate at t, represents the top temperature change rate at t, T(t) represents the top temperature at t, u(t) represents the sliding mode control rate, and d(t) represents the disturbance term.

[0035] Further, the high gain observer formula is:

[0036]

[0037] Where represents the observer temperature change rate, represents the actual system temperature second-order derivative, h1 represents the high gain coefficient, y represents the actual system temperature, represents the observer temperature, represents the second order derivative of the observer temperature, x1 represents the actual system temperature, h2 represents a high gain coefficient,

[0038] wherein y=x1=T(t),

[0039] Further, the sliding mode surface formula is:

[0040]

[0041] wherein s represents the sliding mode surface, a represents the sliding mode coefficient, a>0, e represents the error, represents the error derivative,

[0042] Then:

[0043]

[0044] wherein, represents the sliding mode surface, represents the error, represents the error derivative, (according to the sliding mode surface formula corresponding sliding mode surface is designed)

[0045] e=T d -T, T d is a set temperature, is a constant, T represents a temperature, represents the set temperature rate of change, represents the actual system temperature rate of change, represents the observer temperature, represents the observer temperature rate of change,

[0046] Therefore The mathematical model of methanol distillation temperature is as follows:

[0047]

[0048] The sliding mode control law is:

[0049]

[0050] wherein, represents the set temperature rate of change per unit time, η represents a to-be-determined parameter under different working conditions, and is a constant greater than 0.

[0051] Preferably, 1<η<3, so as to balance the stability and sensitivity of the model.

[0052] Preferably, the different working condition demand indexes are the control signal ut, the recovery time rt, and the cumulative error Acce.

[0053] Further, the step 4 specifically includes:

[0054] Step 4.1: change the numerical value of the control variable, and obtain the corresponding control index [ut, rt, Acce] data set by using the simulation system;

[0055] Step 4.2: for different control variables, establish optimization problems by using three different indexes, and obtain decision functions after optimization to complete voting on different input samples;

[0056] Step 4.3: the voting strategy is used to obtain the membership degree P(ut), P(rt), P(Acce) of the temperature control signal corresponding to the three indexes, and finally the control variable η corresponding to the determined sliding mode control rate u(t) under the current working condition index can be obtained.

[0057] Further, the membership degree satisfies the following equation:

[0058] η = η ut P(ut) + η rt P(rt) + η Acce P(Acce)

[0059] Wherein, η ut represents the optimal control variable value of ut, η rt represents the optimal control variable value of rt, and η Acce represents the optimal control variable value of Acce, and the η value corresponding to the optimal control index is solved.

[0060] The application provides a high-gain sliding mode variable structure methanol rectification temperature control method based on a multi-classification voting strategy, which solves the problems of manual adjustment of control parameters and diversified index control under different working conditions.

[0061] 1、The application firstly changes the numerical value of the control variable η continuously, obtains the corresponding control index ut, rt, Acce data set by simulation, and then fixes the training label of the given working condition index, calculates the membership degree of η corresponding to the three indexes by using the SVM multi-classification voting strategy, and finally obtains the control law under the current working condition index. Under the determined control law, the temperature signal and the observation value of the temperature change rate are obtained by using the high-gain observer, and then fed back to the sliding mode control law, so that the sliding mode variable structure control has strong robustness when subjected to external disturbance and perturbation. Through this data-driven way to obtain the membership degree relationship between the control index and the control parameter under different working conditions, not only the control performance can be more accurate, but also the local optimal solution can be avoided in the strategy space search process, and the calculation complexity is reduced.

[0062] 2、The application decomposes the nonlinear dynamic into linear part and nonlinear part by designing high gain observer, and selects the gain of the observer, so that the linear part dominates the nonlinear part. By selecting a large enough observer gain, the observation error can be made arbitrarily small. By using high gain, the uncertainty of the system is suppressed, and the fast convergence of the observed state is realized.

[0063] 3、The application introduces the voting strategy of the support vector machine multi-classification algorithm under the multi-working condition of the production process, which can quickly adjust the parameters of the sliding mode control rate under the determined working condition. The method is a new type of machine learning method developed based on the statistical learning theory, which mainly constructs the optimal classification hyperplane according to the principle of structural risk minimization. When the inter-class interval is maximized, the probability of correct classification of the sample is greater. The purpose is to find a hyperplane to separate two different sets. When the data is linearly inseparable, the kernel function is used to map the low-dimensional space inseparable data to the high-dimensional space, so that it is linearly distinguishable. Therefore, the SVM multi-classification can reasonably classify the nonlinear indexes in the temperature control system of the methanol rectification tower. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 The application is a step flow chart;

[0065] Figure 2 The application provides a high gain sliding mode variable structure methanol rectification reaction rectification tower temperature control flow diagram of a multi-classification voting strategy;

[0066] Figure 3 The application provides a high gain sliding mode variable structure rectification tower plate control block diagram of a multi-classification voting strategy;

[0067] Figure 4 The application is a classification result of the data support vector machine multi-classification voting strategy of working condition one, working condition two and working condition three;

[0068] Figure 5 The application is a temperature control curve simulated by matlab of working condition one, working condition two and working condition three. DETAILED DESCRIPTION

[0069] The application will be further described in detail below in combination with the drawings and specific implementation methods.

[0070] The technical solutions in the embodiments of the present application will be clearly and completely described below with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of the present application.

[0071] A high-gain sliding mode variable structure methanol rectification temperature control method of a multi-classification voting strategy, as shown in the formula (1), comprises the following steps: Figure 1

[0072] Step 1: Establishing a mathematical model of a methanol rectification tower

[0073] Before the model is established, the following assumptions are needed: 1) It is assumed that the reflux and steam flow in the rectification tower remain unchanged, and there is no leakage of reflux and steam, and the condenser can be completely condensed; 2) It is assumed that the liquid amount of each tray is constant, and the steam retention amount can be ignored; 3) It is assumed that the efficiency of the tray is constant, the gas-liquid phase is balanced, and the gas phase is an ideal gas; 4) It is assumed that there is no heat loss, and the gas-liquid phase of the tray and the reboiler is completely mixed.

[0074] Therefore, the dynamic material balance equations of the reboiler, the overhead condenser and the trays (the number of trays is preset to 6) 1 to 6 of the rectification tower can be obtained:

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082] The static material balance equation is V=L+D

[0083] Wherein t is time, dt is the differential of time, dX / dt is the change value of concentration per unit time, L is reflux flow, M d is the retention liquid amount of the overhead condenser, M s is the retention liquid amount of the reboiler, V is steam flow, X d is the liquid phase methanol concentration of the overhead condenser, X i is the liquid phase methanol concentration of the tray i (i=1, 2, …, 6), and X s ​is the liquid phase methanol concentration of the reboiler, Y i is the gas phase methanol concentration of the tray i (i = 1, 2, …, 6), Y s is the gas phase methanol concentration of the reboiler, D is the methanol product flow rate collected after condensation at the top of the column.

[0084] The reflux ratio: R = L / D, where, given α c is the volatility coefficient, then the relationship between the gas phase methanol concentration Y and the liquid phase methanol concentration X in the rectification column can be represented by a constant relative volatility:

[0085]

[0086]

[0087] According to the above formula, the mathematical model of the methanol rectification column is obtained, and according to the temperature-concentration model obtained from previous experimental research, the corresponding methanol concentration X at the top of the column can be estimated from the measured top temperature T, which is very useful for the prediction and control of the methanol concentration in the rectification column, specifically: In the simulation of the rectification column, this formula is often used to calculate the temperature of the tray from the known methanol concentration, and is used for the simulation and optimization of the temperature feedback control strategy of the rectification column.

[0088] Assuming that there is a certain methanol concentration at the top of the rectification column, the theoretical methanol top temperature value T d is calculated, and according to the input and output data of the actual operation of the pressurized column in the methanol rectification column, the approximate mathematical model of the system is obtained as follows:

[0089]

[0090] Here d(t) represents the disturbance term, and the state equation of the methanol rectification column top temperature is as follows:

[0091]

[0092] where a is the top temperature change rate coefficient, b is the top temperature coefficient, and c is the sliding mode control rate coefficient, a, b, and c are all constants,

[0093] Step 2: Design of high gain observer, considering the form of state equation, the observer is designed as follows:

[0094]

[0095] where, represents the observer temperature change rate, represents the actual system temperature second-order derivative, h1 represents the high gain coefficient, y represents the actual system temperature, represents the observer temperature, The second derivative of the observer temperature is represented by x1, the actual system temperature is represented by h2, and the high gain coefficient is represented by h2,

[0096] Where y=x1=T(t), for convenience of writing, the disturbance term is combined with the control term into the control term. From the above two formulas, we can get:

[0097]

[0098] Where, That is

[0099]

[0100] The temperature control system often has strong nonlinearity, which means that the complexity of the system state is much higher than that of a general linear system. The purpose of high gain observer design is to decompose the nonlinear dynamic into linear and nonlinear parts, and to select the gain of the observer so that the linear part dominates the nonlinear part. By selecting a large enough observer gain, the observation error can be made arbitrarily small. By using high gain, the uncertainty of the system is suppressed, and the state of the observation is quickly converged.

[0101] Step 3: Design of sliding mode control rate, for the above model, design the sliding surface

[0102] Where s represents the sliding surface, a represents the sliding coefficient, a>0, and e represents the error, The derivative of the error is represented by e,

[0103] Then Where, The sliding surface is represented by s, The error is represented by e, The derivative of the error is represented by e, (according to the above sliding surface formula Design the corresponding sliding surface)

[0104] Where e=T d -T, In this system, T d is the set temperature, which is a constant, and T represents the temperature, The set temperature rate is represented by T The actual system temperature rate is represented by T The observer temperature is represented by T The observer temperature rate is represented by T

[0105] Therefore The mathematical model of methanol distillation temperature is as follows:

[0106]

[0107] The sliding mode control law is taken as:

[0108]

[0109] wherein, represents the setting temperature unit time rate of change, η represents the to-be-determined parameter under different working conditions, η > 0, respectively, the derivative of the temperature rate of change and the temperature rate of change, for the convenience of notation, let T = T(t).

[0110] Wherein, the system stability condition needs the following theoretical support:

[0111] In order to make the system asymptotically stable, the matrix A needs to be a Hurwitz matrix. That is:

[0112] If and only if the eigenvalue of A is negative, the following formula is established:

[0113]

[0114] wherein σ0respectively represents the stability coefficient and the exponential decay rate, both are normal numbers, ||||| represents the norm, t0represents the initial value time, represents the system state value at time t, represents the initial state value.

[0115] The characteristic equation of the system is then

[0116] s 2 +(h1+a)s+h2=0→s 2 +2ps+p 2 =0,

[0117] s represents the system state variable, I represents the unit matrix, p represents the system damping parameter, a represents the tower top temperature rate of change coefficient, and by comparison, we can get:

[0118]

[0119] And the design of p needs to meet h1=2p-a>0.

[0120] For the high-gain sliding mode control loop, take Lyapunov function

[0121]

[0122]

[0123] We get:

[0124]

[0125] where, represents error derivative, represents error second derivative, represents observer error derivative, represents observer and setpoint error derivative, represents sliding surface designed from error of observer and setpoint, represents sliding surface designed from error of observer and actual system, so

[0126]

[0127] where, represents Lyapunov function for sliding surface, when system is stable, always holds, original equation, so η>1.

[0128] For the whole closed loop system, take Lyapunov function

[0129]

[0130]

[0131] where, V s represents Lyapunov function for sliding surface, V sys represents Lyapunov function for open loop system, represents derivative of closed loop system Lyapunov function, represents derivative of Lyapunov function for sliding surface,

[0132] Because A is Hurwitz matrix, the above equation converges. And

[0133]

[0134] where, η>1, χ(·) is K type function,

[0135] For V:[0,∞)∈R, the solution of inequality equation is

[0136]

[0137] where, α is any constant,

[0138] Thus it can be proved that the solution of equation

[0139]

[0140] is

[0141] And because V(t) ≥ 0, t→∞, V(t) = 0.

[0142] Where, system stability requires η > 1, therefore, when designing the value of control variable η, it is necessary to set its range reasonably, since when η value is greater than 3, the output of the system is less sensitive to the recovery time and cumulative error, and the control signal will far exceed the output limit of the temperature controller, therefore, the control parameter selected by the present application is between 1 and 3.

[0143] Step 4: For different working conditions, the appropriate control law is selected by voting strategy, which is implemented as follows:

[0144] Step 4.1: Change the value of control variable η, and use simulink simulation system to obtain the corresponding control index [ut, rt, Acce] data set; wherein, [ut, rt, Acce] data set corresponds to the control signal of methanol distillation temperature, recovery time, and cumulative error respectively; in the methanol distillation temperature control system, the temperature control in the rectifying tower usually adopts pneumatic control valve to control the steam flow, the control signal ut of the valve is generally a voltage signal; the control index rt refers to the time required for the temperature in the rectifying tower to recover to steady state after being disturbed; the control index Acce refers to the cumulative change value of the temperature in the recovery time, these indexes are all influencing factors of methanol distillation products, and the recovery time rt and the cumulative error Acce are very crucial to the reaction concentration. The temperature control in the rectifying tower usually adopts pneumatic control valve to control the steam flow, and the valve control signal ut also needs to be considered as a key index.

[0145] Step 4.2: For different control variables, establish optimization problems using three different indexes, obtain decision function after optimization, and complete voting for different input samples; here, the support vector machine multi-class voting strategy is introduced under the condition of multiple working conditions in the production process, which can quickly adjust the parameters of the sliding mode control rate under certain working conditions. Through the voting strategy in the multi-class process, the respective membership degrees P(ut), P(rt), P(Acce) of the different working condition demand indexes ut, rt, Acce on the control variable η are established. The SVM multi-class studied in the present application selects these three indexes as classification objects, the principle is based on the voting strategy in the SVM multi-class process, and the respective membership degrees P(ut), P(rt), P(Acce) of the different working condition demand indexes (control signal ut, recovery time rt, cumulative error Acce) on the control variable η are established. The value of the required control variable η is determined through this membership degree.

[0146] The value of the required control variable is determined through this membership degree η = η ut P(ut) + η rt P(rt) + η AcceP(Acce), where, η ut represents the optimal control variable value of ut, η rt represents the optimal control variable value of rt, η Acce represents the optimal control variable value of Acce, η * is the corresponding η value when the control index reaches the optimum. The voting strategy used in the present application is the pairwise classification method in SVM multi-classification. By using the voting strategy, the problem of improper selection of the control variable η in the sliding mode control law in the case of insufficient human experience can be avoided. The voting strategy can replace human selection because the one-against-one method of the SVM pairwise classification method selects a pair of each two classes, then performs SVM binary classification. If there are n class sample data sets, n(n-1) / 2 binary classifiers need to be constructed, two sample points are used to form two problem training sets T(i,j), and then all the decision functions D i,j (data) of the two problems are obtained, and it is determined which class the data belongs to.

[0147] Step 4.3: The membership degrees P(ut), P(rt), and P(Acce) of the temperature control signal corresponding to the three indexes are obtained by using the results of the voting strategy, and finally the control variable η under the current working condition index can be obtained, and the determined sliding mode control rate u(t) is determined;

[0148] 1) The voting strategy needs to be implemented according to the decision function, and the steps and principles of the one-against-one method of the SVM multi-classification pairwise classification method are as follows. A pair of each two classes is selected between each class, and then SVM binary classification is performed. If there are n class sample data sets, n(n-1) / 2 binary classifiers need to be constructed, two sample points are used to form two problem training sets T(i,j), T(i,j) represents a labeled data set [i,j] (where i, j includes a label column and a data column, which is a control variable [L i , η i ], [L j , η j ]), and all the decision functions D i,j (data) of the two problems are obtained, and it is determined which class the data belongs to. i , L j , L i , L j represents two classes of binary classification, and each binary SVM has a decision function for new data data new has a vote or prediction. If the decision function D new of data i,j is calculated, it is determined which class the data belongs to.(data) > 0, then L i The vote count for a given category increases by one; otherwise, L... j The number of votes for each category increases by one;

[0149] 2) Combine all binary classification results; the category with the most votes is the one corresponding to the data. new The prediction.

[0150] 3) If a tie occurs, simply select the category with the smaller index as the data. new The classification.

[0151] For L i Class and L j Classifier SVM i,j The following optimization problem is obtained by solving it.

[0152]

[0153] st

[0154]

[0155] Where C is the penalty factor, usually taken as 1, ζ is the slack variable, b is the classification threshold, and L... i,j Using sample multipliers, we obtain n(n-1) / 2 decision functions after solving.

[0156] D i,j (data)=(L i,j ) T φ(data m )+b i,j

[0157] Among them, b i,j The classification threshold is represented by the kernel function φ(x), which can be φ(x) = x. Clearly, D... i,j (data) = -D j,i (data) is established.

[0158] For the input sample data data, its L can be calculated. i The final vote count in the class:

[0159]

[0160] In the formula, the sign function sign(*) is defined as follows:

[0161]

[0162] Therefore, β represents a parameter that can be used to classify the sample data according to the voting strategy:

[0163]

[0164] According to the voting value obtained by training, each control index ut, rt, Acce is divided into an optimal parameter set and a non-optimal parameter set under the corresponding control variable η, and the membership P(ut), P(rt), P(Acce) is calculated by the following formula.

[0165]

[0166] P(·) represents the membership, the control variable η is the control variable in the sliding mode control law, which determines the adjustment trajectory of the control law, and the variable structure control itself is a nonlinear control method that designs the switching surface according to the desired dynamic characteristics, so that the control action exhibits discontinuity. Unlike other control strategies, variable structure control changes the control action according to the current state during the dynamic change of the system, so that the structure of the system changes. The whole process of sliding mode variable structure control consists of two processes: the approaching motion of the arbitrary initial state to the switching surface and the sliding mode motion on the switching surface. However, when dealing with different working conditions, the adjustment trajectory of the control law often differs greatly, and since only the approximate range of the parameters in the control law can be derived in theory, the operator needs to set the initial value based on experience, but the effect of variable structure control depends on the accuracy of the model, and in the face of complex multi-working conditions, it is difficult to ensure that the entire control system meets the production conditions by setting the initial value artificially.

[0167] The introduction of the voting strategy in the support vector machine multi-classification algorithm under the condition of multiple working conditions in the production process can quickly adjust the parameters of the sliding mode control rate under certain working conditions. This method is a new type of machine learning method based on statistical learning theory, which mainly constructs an optimal classification hyperplane according to the principle of structural risk minimization. When the inter-class interval is maximized, the probability of correct classification of samples is greater. The purpose is to find a hyperplane that separates two different sets. When the data is linearly inseparable, use the kernel function to map the low-dimensional space of the data that cannot be separated to the high-dimensional space, so that it is linearly distinguishable. Thus, SVM multi-classification can make reasonable classification of the nonlinear indexes in the temperature control system of the methanol rectification tower.

[0168] In this study, due to the presence of disturbances in the system, sliding mode control is employed to improve the system's robustness by continuously adjusting the sliding mode control rate. For the temperature control system of a distillation tray, the recovery time is a crucial factor for controlling the concentration of the material on the tray. While conventional variable-structure sliding mode control requires measuring the rate of temperature change, a sliding mode observer is a type of dynamic system that estimates state variables based on measured values ​​of external variables (input and output variables). Sliding mode control methods based on state observers do not require measuring the rate of temperature change and exhibit good performance in practical engineering applications. Therefore, this invention employs a sliding mode control method based on a high-gain observer to address the problem. Temperature control systems often exhibit strong nonlinearity, meaning their system state complexity is far greater than that of typical linear systems. The design purpose of the high-gain observer is to decompose the nonlinear dynamics into linear and nonlinear components and select the observer gain so that the linear component dominates the nonlinear component. By selecting a sufficiently large observer gain, the observation error can be made arbitrarily small. By using high gain, the system's uncertainties are suppressed, achieving rapid convergence of the observed state.

[0169] Step 5: System Performance Analysis. Under the control law described above, the control action is continuously varied according to the current operating conditions, causing changes in the system structure. Simulations are then performed using MATLAB / Simulink tools to verify the performance indicators of the control system.

[0170] Application examples

[0171] Figure 2 and Figure 3 This invention provides a schematic diagram of the process and control logic for temperature control in a high-gain sliding mode variable structure methanol reactive distillation column using a multi-class voting strategy, as illustrated in the embodiments of the present invention. The steps of this method are given by the following simulation example:

[0172] Assuming there is a certain concentration of methanol at the top of the distillation column, the mathematical relationship between temperature and concentration during methanol distillation, based on relevant data, is as follows:

[0173]

[0174] The theoretical methanol tower top temperature T was calculated. d The temperature is set to 121℃, and based on the input and output data obtained from the actual operation of the methanol distillation column, the state space transformation of the transfer function is as follows:

[0175]

[0176] Where d(t) = 20τ(t) - 20τ(t-5), and τ(t) is the step signal.

[0177] The parameters h1=30.803 and h2=400 of the designed high-gain observer satisfy the stability of the observer system. η=13 is taken to change to satisfy the stability of the feedback loop system, and the working condition indexes (control signal ut, recovery time rt, and cumulative error Acce) are given as follows:

[0178] 1) Working condition one: ut<10; rt<1.3; Acce<90;

[0179] 2) Working condition two: ut<12; rt<1.3; Acce<80;

[0180] 3) Working condition three: ut<7; rt<2; Acce<120;

[0181] Three groups of data sets with a dimension of 1000 are obtained by using Simulink, and different membership degrees corresponding to different indexes are obtained by using SVM multi-classification.

[0182] The specific steps are as follows:

[0183] 1) The control index and control variable data set is obtained by changing the parameter setting of the controller.

[0184] 2) The classifier model is obtained by using the machine classification algorithm SVM multi-classification training, and the membership degree of the control variable η to different control indexes is obtained.

[0185] According to Table 1, the test data set is obtained to test the classification result, as shown in Figure 4 , which are respectively the classification results of the data support vector machine multi-classification voting strategy of working conditions one, two and three.

[0186] 3) The model and membership degree obtained by training are used as standards, and for a given control requirement, the corresponding control parameter value is selected by using the formula η=η ut P(ut)+η rt P(rt)+η Acce P(Acce), wherein η * is the η value corresponding to the optimal control index,

[0187]

[0188] Finally, the control parameters obtained by classification are brought into the control strategy, and the temperature control curve can be obtained, and the simulation results obtained by using matlab are as follows, as shown in Figure 5 , which are respectively the temperature control curves of working conditions one, two and three, wherein the protrusions in the figure are the temperature fluctuations caused by the given disturbance signal considering the actual system disturbed by the external disturbance.

[0189] The variable structure control is a nonlinear control method which designs the switching surface according to the expected dynamic characteristics, and makes the control action show discontinuity. Unlike other control strategies, the variable structure control changes the control action purposefully according to the current state in the process of system dynamic change, so that the structure of the system changes. The whole process of the sliding mode variable structure control consists of two processes: the approaching motion s→0 of the arbitrary initial state to the switching surface and the sliding mode motion on the switching surface. However, when dealing with different working conditions, the adjustment trajectories of the control law often differ greatly. Since only the approximate range of the parameters in the control law can be derived in theory, the operator needs to set the initial value by combining the experience method, but the effect of the variable structure control depends on the accuracy of the model, and in the face of complex multi-working conditions, it is difficult to ensure that the entire control system meets the production conditions by setting the initial value artificially. The high-gain sliding mode variable structure control based on the voting strategy proposed in the present application is found through simulation verification that the voting strategy can well utilize the historical data to obtain suitable control variables under different working conditions.

[0190] It is apparent to those skilled in the art that the present application is not limited to the details of the foregoing exemplary embodiments, and that the present application can be implemented in other particular forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered in all respects as illustrative and not restrictive, the scope of the present application being defined by the appended claims rather than by the foregoing description, and it is intended that all changes which come within the meaning and range of equivalency of the claims are resolutely intended to be embraced therein. Any reference signs in the claims should not be construed as limiting the claims to the figures in which the reference signs are used.

[0191] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application, and any slight modification, equivalent replacement and improvement made according to the technical essence of the present application to the above embodiments shall be included in the protection scope of the technical scheme of the present application.

Claims

1. A high-gain sliding mode variable structure methanol distillation temperature control method of a multi-classification voting strategy, characterized in that, Comprising the following steps: Step 1: Constructing the methanol distillation column data model about temperature and sliding mode control rate; Step 2: Based on the methanol distillation column data model and observation value, decomposing the model into linear part and nonlinear part, designing high gain observer, calculating observer error with actual value as sliding mode control input; Step 3: Designing sliding mode surface, calculating sliding mode control rate; Step 4: For different working conditions, selecting appropriate sliding mode control law through voting strategy; Step 5: Carrying out model simulation, verifying the performance index of control system; Different working condition demand index is control signal ut, recovery time rt and cumulative error Acce; Step 4 specific steps include: Step 4.1: Changing the numerical value of control variable, obtaining corresponding control index [ut, rt, Acce] data set by using simulation system; Step 4.2: For different control variables, establishing optimization problem by using three different indexes, obtaining decision function after optimization, and completing voting on different input samples; Step 4.3: Using the voting strategy result to obtain the membership degree P(ut), P(rt), P(Acce) of temperature control signal corresponding to the three indexes, and finally obtaining the control variable η under the current working condition index, corresponding to the determined sliding mode control rate u(t).

2. The high gain sliding mode variable structure methanol distillation temperature control method of a multi-classification voting strategy according to claim 1, characterized in that, Model assumption conditions are: 1) Assuming that the reflux and steam flow in the distillation column remain unchanged, and there is no leakage of reflux and leakage of steam, The condenser is completely condensed; 2) Assuming that the liquid amount of each tray is constant, and the steam retention amount can be ignored; 3) Assuming that the efficiency of the tray is constant, the gas-liquid phase is in equilibrium, and the gas phase is ideal gas; 4) Assuming that there is no heat loss, and the gas-liquid phase in the tray and reboiler is completely mixed.

3. The high-gain sliding mode variable structure methanol distillation temperature control method of a multi-classification voting strategy according to claim 2, characterized in that, The methanol distillation column data model is constructed based on dynamic material balance equation and static material balance equation, and the dynamic balance equation is: The static balance equation is: V = L + D where t is time, dt is the differential of time, L is the reflux flow, M d is the amount of liquid holdup in the overhead condenser s is the amount of liquid holdup in the reboiler, V is the vapor flow, X d is the liquid phase methanol concentration in the overhead condenser i is the liquid phase methanol concentration in tray i, where i = 1, 2,..., 6, X s is the liquid phase methanol concentration in the reboiler, Y i is the vapor phase methanol concentration in tray i, Y s is the vapor phase methanol concentration in the reboiler, D is the flow of methanol product collected after the overhead condensation; The relationship between the gas phase methanol concentration Y and the liquid phase methanol concentration X in the distillation column is represented by a constant relative volatility: wherein a c is the volatility factor.

4. The high-gain sliding mode variable structure methanol distillation temperature control method of a multi-classification voting strategy according to claim 3, characterized in that, The mathematical model of methanol distillation column is: wherein a is a tower top temperature change rate coefficient, b is a tower top temperature coefficient, and c is a sliding mode control rate coefficient, a, b, and c are all constants, denotes a derivative of a tower top temperature change rate at time t, denotes a tower top temperature change rate at time t, T(t) denotes a tower top temperature at time t, u(t) denotes a sliding mode control rate, and d(t) denotes a disturbance term.

5. The high-gain sliding mode variable structure methanol distillation temperature control method of a multi-classification voting strategy according to claim 4, characterized in that, The high gain observer formula is: wherein represents the rate of change of the observer temperature, represents the second derivative of the actual system temperature, hi represents a high gain coefficient, and y represents the actual system temperature, represents the observer temperature, represents the second derivative of the observer temperature, xi represents the actual system temperature, and h2 represents a high gain coefficient, wherein y = xi = T(t).

6. The high gain sliding mode variable structure methanol distillation temperature control method of a multi-classification voting strategy according to claim 1, characterized in that, The sliding mode surface formula is: Wherein, s represents the sliding mode surface, a represents the sliding mode coefficient, a > 0, e represents the error, e represents the error derivative, Then: Wherein, s represents the sliding mode surface, e represents the error, e represents the error derivative, e = T d -T, T d T is a set temperature, T is a constant, T represents a temperature, T_d represents a set temperature change rate, T represents an actual system temperature change rate, T represents an observer temperature, T represents an observer temperature change rate, Therefore The mathematical model for methanol distillation temperature is as follows: The sliding mode control law is: wherein, represents the set temperature unit time change rate, η represents the to-be-determined parameter under different working conditions, and is a constant greater than 0.

7. The high-gain sliding mode variable structure methanol distillation temperature control method of a multi-classification voting strategy according to claim 6, characterized in that: 1 < η < 3, in order to balance the stability and sensitivity of the model.

8. The high gain sliding mode variable structure methanol distillation temperature control method of a multi-classification voting strategy according to claim 7, characterized in that: The membership degree satisfies the following equation: η = η ut P(ut) + η rt P(rt) + η Acce P(Acce) Wherein, η ut The control variable value of ut optimal, η rt The control variable value of rt optimal, η Acce The control variable value of Acce optimal, η

Images