New energy battery coating post-drying multivariable model predictive control system

Abstract

The application discloses a new energy battery coating post-drying multivariable model predictive control system APC The control system performs real-time optimization control and rolling update according to the calling period. Specifically, the control system performs prediction and optimization calculation based on a model according to the set values of the return air valve opening, the exhaust fan frequency and the external circulating fan frequency of the last calling period, the NMP concentration measurement value at the current time, the oven pressure measurement value, the target value of the NMP concentration and the measurable external disturbance change data, and outputs the optimal sequence of the set values of the return air valve opening, the exhaust fan frequency and the external circulating fan frequency of the next period. Finally, the output values of the first time of all input variables in the optimal sequence are written back to the related actuators of the drying system, and the closed-loop control is completed. The application realizes real-time online optimization of the post-coating drying process based on the predicted dynamic optimization control under the premise of guaranteeing the stability of the system and the non-exceeding of the toxic solvent NMP concentration, and realizes energy consumption reduction while guaranteeing the collaborative and stable operation of the multiple ovens.

Description

Technical Field

[0001] This invention belongs to the field of drying control after battery coating, and particularly relates to a multivariate model predictive control system for drying after coating of new energy batteries. Background Technology

[0002] The post-coating drying process is a crucial step in new energy battery production. It involves drying the wet film after coating the battery. In the drying chamber, circulating high-temperature hot air is used to dry the coating. Because NMP solvent is used in the battery cathode coating, this solvent vaporizes during the drying process, producing a weakly toxic gas. When the gas concentration reaches a certain level, there is a risk of explosion. Therefore, the battery drying process must strictly control the NMP gas concentration within safe limits. The battery coating drying system employs a cascaded design of multiple ovens. The coated electrode sheets pass uniformly through multiple ovens at a set speed. The temperature and NMP concentration vary in different ovens. The quality of the electrode sheets leaving the drying process is strongly coupled with the drying process in each oven. Therefore, the battery coating drying system is a typical complex control system characterized by multivariables, nonlinearity, strong coupling, and a certain time delay.

[0003] The battery coating and drying system needs to ensure that the temperature and pressure in each drying chamber remain stable to meet the drying control objectives, while avoiding excessive NMP concentration. Therefore, the design of this control system is a typical multi-objective control problem. However, in current practical operations, most systems employ multiple independent single-loop control designs, with some loops remaining fixed at a setpoint for extended periods. This makes it difficult to adjust in real time according to changes in the actual temperature, pressure, and NMP concentration within the drying chamber. Consequently, in order to prevent NMP from exceeding the limit, the system often operates in overshoot control mode, which, while reducing NMP concentration, leads to energy waste.

[0004] Furthermore, different specifications of battery electrode sheets correspond to different dynamic response gains and time delays in the battery coating and drying system, resulting in different amounts of NMP gas evaporation in the oven. Consequently, the temperature and pressure changes in the oven also differ. Different electrode sheets have different optimal operating points. During product switching and standby to production switching processes, there is a lack of real-time controllers. Existing manual operations often involve significantly increasing the circulating air volume to drastically reduce the NMP concentration during switching in order to ensure that the NMP concentration meets the standard. This method of large ventilation and fast circulation leads to excessive low-temperature return air being heated before entering the drying chamber, consuming more electrical energy and causing a large amount of energy waste.

[0005] In summary, existing battery coating and drying systems lack effective real-time adjustment and control methods, resulting in unstable battery drying quality and energy waste. Summary of the Invention

[0006] To address the problems existing in the prior art, the purpose of this invention is to provide a multivariate model predictive control system for drying new energy batteries after coating, which achieves online real-time optimization of the drying process while ensuring the stability of the drying system.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A multivariate model predictive control system for drying new energy batteries after coating is provided. The drying system includes m drying ovens and n sections of main air supply and return ducts. Each drying oven includes an electric return air valve, an internal circulation fan, an electric heating element, and an exhaust fan. The return air volume is controlled by adjusting the frequency of the m internal circulation fans and the opening of the m return air valves. The temperature inside the drying oven is controlled by adjusting the power of the m electric heating elements. The pressure and NMP concentration changes inside the drying oven are compensated by adjusting the frequency of the m exhaust fans. The NMP concentration in each drying oven section and the total NMP concentration are controlled by adjusting the frequency of the n external circulation fans. Wherein, n is less than m.

[0009] The control system is based on T APC The system performs real-time optimization control and rolling updates for each call cycle. Specifically, the control system performs model-based prediction and optimization processing based on the set values ​​of the return air valve opening, exhaust fan frequency, and external circulation fan frequency of the previous call cycle, as well as the current measured values ​​of NMP concentration, oven pressure, and target NMP concentration, and measurable external disturbance change data. The optimized output is the optimal sequence of set values ​​for the return air valve opening, exhaust fan frequency, and external circulation fan frequency for the next cycle. Finally, the output values ​​of all input variables in the optimal sequence at the first moment are written back to the relevant actuators of the drying system to complete the closed-loop control.

[0010] The APC optimization control module defines the input variable U and the output variable Y. The mathematical relationship of the system's multiple inputs and multiple outputs is expressed as follows:

[0011] Y 2m+1 =G (2m+1)×(3m+n) U 3m+n (1)

[0012] In the formula, G is a (2m+1)×(3m+n) model matrix, where each element is a 1×1 transfer function, and U is a 3m+n column vector containing m return air valve openings u1, u2, ..., u... m m internal circulation fan frequencies u m+1 u m+2 ... u 2m m exhaust fan frequencies u 2m+1 u 2m+2 ... u 3m and n external circulation fan frequencies u3m+1 u 3m+2 ... u 3m+n Y is a column vector with 2m+1 rows, including m sections of the oven containing NMP concentrations y1, y2, ... y. m m-section oven cavity pressure y m+1 y m+2 ... y 2m And the total NMP concentration y of one drying system 2m+1 Based on the above description, this multivariate control problem can be specifically expressed as:

[0013]

[0014]

[0015] st,

[0016] y min ≤y(k+l|k)≤y max l = 1, 2, ..., H cy (5)

[0017] u min ≤u(k+r-1|k)≤u max r = 1, 2, ..., H cu (6)

[0018] Δu(k+i|k)=u(k+i)-u(k+i-1) (7)

[0019] |Δu(k+i-1|k)|≤Δu max i = 1, 2, ..., H u (8)

[0020] Wherein, the symbol (*k) represents the k+* future periods based on the k-th control period, where H u Known as the control step size, it represents the input sequence. Time span; H y Known as the prediction step size, it represents the predicted output trajectory. Time span;

[0021]

[0022] Represents {k,k+1,k+H} starting from the current k-th period. y The predicted sequence of the future output variables at time {k,k+1,k+Hu} and the input variables at time {k,k+1,k+Hu};

[0023] [y min ,y max] represents the upper and lower bounds of the output variable; [u min ,u max ] represents the upper and lower bounds of the input variable; Δu max The maximum rate of change; where, Indicates weighted Q i The 2-norm is calculated; Q1, Q2, and Q3 correspond to the weight matrices of the three summation terms in the objective function J1, respectively, and Q1 is H. y line H y Column numerical matrices, Q2 and Q3 are H u line H u Column numerical matrix; where, These are the target values ​​of output variables that users manually set or check based on different batches;

[0024] The control system is optimized based on the above model G(z) and objective function J1. The optimization control of each APC cycle is completed by solving the input sequence that minimizes the objective function value that satisfies the constraints through quadratic programming. Among them, the first term on the right side of the equation (4) expresses the minimization of the error between the predicted trajectory and the control target, the second term expresses the minimization of the error between the control variable U and its predetermined target, and the third term expresses the minimization of the change index of the control variable U each time.

[0025] The specific implementation steps of the control system are as follows:

[0026] The first step is to determine the target value Y based on the objective function, constraints, current input / output variable values, and adjustable weight values ​​Qi (i = 1, 2, 3). TGT Solving the quadratic programming problem yields the optimal input sequence that satisfies the constraints.

[0027] The second step is to obtain the optimal input sequence. The first element The controller of the drying system sends the data to form an online closed-loop control, which then sends the remaining input sequence elements... throw away;

[0028] Thirdly, at the next sampling time, the system will obtain a new output value y. k+1 Then, using the updated (k+1 time) y k+1 Repeat the above two steps to obtain the optimal system input for the next step.

[0029] By repeating the above process iteratively, the optimal input setpoint after rolling optimization in each APC cycle can be obtained. It is applied to control systems to complete the application of multivariable model predictive controllers.

[0030] The control system performs dimensionality reduction on the input variables of the multivariable model: the internal circulation fan frequency u... m+1 u m+2 ... u 2m It is separated from U and does not participate in APC optimization;

[0031] If the internal circulation fan frequency is not selected for optimization control, the internal circulation fan frequency will only be set to the initial frequency value after batch start-up and shutdown. The initial frequency value can be manually adjusted by the operator, and after reaching a stable state, it can be adjusted to a new fixed value. Alternatively, conditional judgment logic can be programmed to solidify the manual operation into the logic program, but it will not participate in APC control.

[0032] The control system performs dimensionality reduction processing on the external circulation fan frequency in the input variables of the multivariable model matrix, and adopts a linkage design for multiple independent external circulation fan frequencies: selecting the i-th external circulation fan frequency u 2m+i As a benchmark, the frequencies of other external circulation fans are adjusted accordingly, and the adjustment methods are as follows:

[0033] u 2m+j =a j u 2m+i +b j , j=1,2,…,n (10)

[0034] Where, when j = i, a j =1,b j =0, corresponding to the original value is not modified.

[0035] The control system compares the static gain value of the transfer function of each row of the multivariable model matrix. When the static value of some functions is much smaller than the maximum transfer function gain, a simplified approximation is performed: when the model gain of the opening of the i-th return air valve on the NMP concentration and pressure of the j-th (j≠i) oven is one-fifth or even less than that of the oven in this section, "0" is used to replace the less influential model g. i,j (i = 1, 2, ..., m, j ≠ i, j ≠ i ± 1; or i = m + 1, m + 2, ..., 2m, j ≠ im, j ≠ im ± 1); Similarly, when the model gain of the i-th exhaust fan on the NMP concentration and pressure of the j-th (j ≠ i) oven is one-fifth or even less than that of the ovens in this section, g is replaced with "0". i,j (i=m+1, m+2,…,2m, j≠im, j≠im±1).

[0036] The control system performs dimensionality reduction on the input variables of the multivariable model: it links the linear relationship function mapping of the outlet exhaust fan frequency and the inlet return air valve opening of the same section in the multivariable model's input variables, thereby merging the control design of two variables into a single variable control design; its adjustment method is as follows:

[0037] u j =c j u i +d j , i=1,2,…,m, j=m+1,m+2,…,2m (13)

[0038] In the formula, u j and u i These represent the frequency of the exhaust fan and the opening degree of the return air valve in the same section, respectively. j and d i These are the follow-up adjustment parameters for the exhaust fan of the i-th oven, where the follow-up adjustment parameters are manually set according to the air volume balance requirements.

[0039] After adopting the above scheme, this invention, based on the input u and output y of the multivariate control matrix, controls the NMP concentration and pressure in the oven using T. APC It performs real-time optimization control and rolling updates for the operating cycle, and predicts future trends in real time through rolling optimization strategies. It achieves dynamic optimization while ensuring system stability. This not only ensures system stability, but also enables real-time online optimization of multiple variables. It solves the problems of severe mutual coupling and interference between multiple single-loop controllers operating independently in traditional systems, or the need for frequent manual intervention.

[0040] Furthermore, the dimensionality reduction and approximation of the multivariate control matrix in this invention aims to reduce its complexity and improve operational efficiency. Regardless of whether the model matrix is ​​dimensionality reduced, the prediction-based optimization rolling update strategy ensures the temperature gradient of the circulating air in each oven section, thus preventing any impact on the final battery electrode drying quality. Simultaneously, it suppresses NMP concentration fluctuations, improving the stability of the intelligent optimization operation of the drying process after battery coating, and reducing drying energy consumption while ensuring that the NMP concentration does not exceed the standard. Attached Figure Description

[0041] Figure 1 Schematic diagram of battery coating and drying system;

[0042] Figure 2 This is a schematic diagram of the control system of the present invention. Detailed Implementation

[0043] like Figure 1As shown, the battery coating and drying system includes multiple oven sections, air supply ducts, and return air ducts. Each oven section includes an electric return air valve, an internal circulation fan, an electric heating element, and an exhaust fan. The NMP-containing gas discharged from the multiple oven sections enters the air supply duct through the exhaust ducts, and then the high-temperature, high-NMP-concentration gas is sent to a cooling absorption device. Through heat exchange and cooling, the NMP is liquefied and precipitated. The cooled, low-NMP-concentration circulating air is then blown back into the oven through the return air duct and multiple external circulation fans. During the return air process, a return air valve at the inlet of each oven section controls the return air volume of each section. Simultaneously, to ensure temperature stability within the oven, the circulating return air needs to be preheated through a heat exchanger before entering the oven, continuously replenishing the heat and air consumed by the exhaust air.

[0044] The battery coating and drying system is designed with m drying ovens and n main supply and return air ducts. To ensure temperature and airflow balance within the ovens, the main control loops of the drying system include: adjusting the return air volume by controlling the frequency of m internal circulation fans and the opening of m return air valves; adjusting the oven temperature by controlling the power of m electric heating elements; regulating the NMP concentration by controlling the frequency of n external circulation fans; and compensating for the pressure changes within the oven caused by the actions of other actuators by adjusting the frequency of m exhaust fans, thus maintaining stable operating conditions within the oven. Here, n can be equal to or less than m. In practical applications, several nearby drying ovens often share a single external circulation duct and external circulation fan to reduce costs.

[0045] For each oven section, the controlled variables (CV) include the oven internal temperature, oven internal pressure, and NMP concentration at the oven exhaust outlet. The manipulated variables (MV) include the exhaust fan frequency, return air valve opening, internal circulation fan frequency, electric heating pack power, and external circulation fan frequency. The disturbance variables (DV) include the coating machine speed and the temperature of the returned circulating air after cooling. These variables typically vary depending on the type of battery electrode being dried and the production schedule. Consequently, the oven internal temperature and the minimum external circulation fan frequency will also vary depending on the coating process.

[0046] The main energy consumption in the drying process includes: the motor energy consumption of the exhaust fan and internal circulation fan in each drying chamber, the heating energy of the electric heating pack, and the motor energy consumption of the external circulation fan on the return air duct. The energy balance of the entire process can be characterized by the change in heat per unit time. The electrodes entering the drying chamber are at a relatively low temperature, and the temperature of the electrodes leaving the drying chamber is the final temperature of the drying chamber. At the same time, the airflow entering the exhaust duct loses heat due to the cooling and separation of liquid NMP. Finally, the heating pack raises the temperature to compensate for some of the heat, allowing the temperature inside the drying chamber to reach equilibrium.

[0047] Reference Figure 2As shown, this invention designs a multivariate model predictive control system for drying new energy batteries after coating. Based on the set values ​​of the return air valve opening, exhaust fan frequency, and external circulation fan frequency from the previous call cycle, as well as the current measured NMP concentration, current measured oven pressure, target NMP concentration, and measurable external disturbance data, it predicts different trends and corresponding optimization targets within a certain time interval. Finally, it dynamically outputs the optimal sequence of the set values ​​for the return air valve opening, exhaust fan frequency, and external circulation fan frequency for the next call cycle. This optimal sequence is then written back to the relevant actuators of the drying system, ultimately completing closed-loop control. This achieves real-time adjustment of the drying system, ensuring system stability (NMP concentration, oven temperature, and pressure) while simultaneously achieving significant energy savings. The control system uses T... APC Real-time optimization control and rolling updates are performed for the call cycle. Rolling updates refer to obtaining a new future optimal sequence for each call cycle, using only the output value of the first moment of the optimal sequence, and discarding the values ​​of other future moments.

[0048] This invention describes the input-output relationship of battery drying control using a multivariable model matrix expression. Taking a drying system consisting of m drying ovens as an example, this invention defines the input variable U and the output variable Y, where U is a column vector with 3m+n rows and Y is a column vector with 2m+1 rows. Considering the spatial transfer function relationship between U and Y, it is as follows:

[0049] Y 2m+1 =G (2m+1)×(3m+n) U 3m+n (1)

[0050] In the formula, G is a (2m+1)×(3m+n) model matrix, where each element is a 1×1 transfer function. The 3m+n input variables include m return air valve openings u1, u2, ..., u... m m internal circulation fan frequencies u m+1 u m+2 ... u 2m m exhaust fan frequencies u 2m+1 u 2m+2 ... u 3m and n external circulation fan frequencies u 3m+1 u 3m+2 ... u 3m+n The 2m+1 output variables include the NMP concentrations y1, y2, ... y in m oven sections. m m-section oven cavity pressure y m+1 y m+2 ... y 2m And the total NMP concentration y of a drying system 2m+1The values ​​of each input and output variable are acquired using existing conventional technologies, such as PLC or DCS acquisition of the opening degree of the return air valve's associated sensor, the frequency of the fan's associated inverter, the NMP concentration in the oven measured by the NMP concentration meter, and the chamber pressure measured by the pressure gauge. This multivariate control problem is specifically expressed as:

[0051]

[0052] In the formula, g i,j This represents the mathematical relationship between the j-th manipulated variable and the i-th controlled variable, i.e., the transfer function model.

[0053] The control system of this invention utilizes a model predictive control (APC) algorithm to control the NMP concentration and pressure in the oven according to the input u and output y of the multivariable model matrix, with T... APC Real-time optimization control and rolling updates are performed for the operating cycle. The control problem is transformed into an optimization problem with constraints, and future trends are predicted in real time through a rolling optimization strategy, while dynamic corrections are made to achieve dynamic optimization while ensuring system stability.

[0054] The control system transforms the continuous-time transfer function model into a discrete-time mathematical model.

[0055] y(k)=G(z)u(k) (3)

[0056] Substituting the values ​​into the model predictive control algorithm, the objective optimization function is constructed as follows:

[0057]

[0058] st,

[0059] y min ≤y(k+l|k)≤y max l = 1, 2, ..., H cy (5)

[0060] u min ≤u(k+r-1|k)≤u max r = 1, 2, ..., H cu (6)

[0061] Δu(k+i|k)=u(k+i)-u(k+i-1) (7)

[0062] |Δu(k+i-1|k)|≤Δu max i = 1, 2, ..., H u (8)

[0063] Wherein, input sequence Output predicted sequence The symbol (*k) represents the predicted input / output variables generated in the k-th control cycle; G(z) is the discrete-time system model expression; [y min ,y max ] represents the upper and lower bounds of the output variable (controlled variable); [u min ,u max ] represents the upper and lower bounds of the input variable (execution quantity); Δu max The maximum rate of change; where, This indicates a weighted 2-norm calculation, where Q1, Q2, and Q3 correspond to the weight matrices of the three summation terms in the objective function J1 (Formula 4), and Q1 is H. y line H y Column numerical matrices, Q2 and Q3 are H u line H u Column numerical matrix. It is the target value of the output variable that the user manually sets or sets according to different batches or by looking up the recipe. It can be a constant or a given time series.

[0064] It is easy to see from the above mathematical expressions that the APC algorithm can not only effectively solve the constrained input-output optimization problem, but also cleverly transform the control problem into an optimization problem.

[0065] The optimization problem formula (4) is a typical quadratic regression (QP) problem with constraint adjustment. However, unlike general quadratic regression, this invention incorporates a rolling optimization strategy. Therefore, the control system of this invention performs the following optimization control based on the above optimization model:

[0066] Solving optimization problems essentially involves finding a future input sequence that satisfies the constraints on u. The state-space model represents G(z), and the actual output value y of the system at the current time is given. k To predict the future H of the system y System output trajectory within time Then predict the trajectory Input sequence And input the difference at each step. Substitute the values ​​into the objective function to obtain its value; continuously adjust... The objective function is obtained, and finally, the optimal sequence of U corresponding to the minimum objective function value d is found among all possible sequences of U. However, in practical applications, considering computational efficiency, for objective functions conforming to quadratic normal form, a quadratic programming (QP) algorithm can be used for fast solution.

[0067] The implementation steps of the control system are as follows:

[0068] The first step is to determine the target value Y based on the objective function, constraints, current input / output variable values, and adjustable weight values ​​Qi (i = 1, 2, 3). TGT Solving the quadratic programming problem yields the optimal input sequence that satisfies the constraints.

[0069] The second step is to obtain the optimal input sequence. The first element The controller of the drying system sends the data to form an online closed-loop control, which then sends the remaining input sequence elements... throw away;

[0070] Thirdly, at the next sampling time, the system will obtain a new output value y. k+1 Then, using the updated (k+1 time) y k+1 Repeat the above two steps to obtain the optimal system input for the next step.

[0071] By repeating the above process iteratively, the optimal input setpoint after rolling optimization in each APC cycle can be obtained. It is applied to control variables to complete the application of multivariable model predictive controller.

[0072] For situations where multiple control variables affect the same controlled variable, this invention performs dimensionality reduction on the multivariate model matrix, simplifies the model, and improves the operational efficiency and stability of controller optimization.

[0073] The dimensionality reduction strategy includes two aspects. Firstly, considering the relatively slow adjustment cycle of the internal circulation fan, this invention will reduce the internal circulation fan frequency u... m+1 u m+2 ... u 2m Separated from U in control problem (2), it is only adjusted by the operator during batch start-up and shutdown, and runs at the set value after stabilization. The original (3m+n) input variables are reduced to (2m+n) input variables. The reduced 2m+n input variables include m return air valve openings u1, u2, ..., u m m exhaust fan frequencies u m+1 u m+2 ... u 2m and n external circulation fan frequencies u 2m+1 u 2m+2 ... u 2m+n Therefore, the multivariate control problem (2) is redefined as:

[0074]

[0075] Furthermore, the model can be further reduced in dimensionality. Based on the design and function of the external circulation fan (ensuring sufficient circulating airflow without directly affecting pressure changes within each oven), and considering that the external circulation fan ensures optimized temperature and airflow relationships in different sections of the drying process, a follow-up calculation is adopted. The frequency of the external circulation fan in a key section is selected as the benchmark, such as the frequency of the external circulation fan at the highest oven temperature (or the main drying section). Other external circulation fan frequencies are followed using different adjustment parameters, with the frequency u of the i-th external circulation fan being used as the reference. 2m+i As a benchmark, the frequencies of other external circulation fans are adjusted accordingly, and the adjustment methods are as follows:

[0076] u 2m+j =a j u 2m+i +b j , j=1,2,…,n (10)

[0077] Applying formula (10) can further reduce the dimensionality of the multivariable control problem (9), retaining only the n external circulation fan frequencies u. 2m+1 u 2m+2 ... u 2m+n A reference frequency This multivariate control problem can be further transformed into:

[0078]

[0079] By reducing the dimensionality of the model matrix, the complexity of the multivariable system matrix is ​​reduced to some extent, the operating efficiency is improved, and the temperature gradient of the circulating air is guaranteed, thus not affecting the final drying quality of the battery electrode sheets.

[0080] For battery coating and drying systems where coupling between non-current and non-adjacent oven sections is weak, and when the model coupling of actuators of the same type for the same output variable is low, a decoupling form that ignores off-diagonal models is proposed. The static gain value of the transfer function for each row is compared, and when some static values ​​are much smaller than the maximum transfer function gain, they are simplified and approximated to 0. This approximation method effectively decouples the actuators between different oven sections, significantly improving the operating efficiency of APC control. The structure of the APC controller using this simplified strategy is the same as that of multiple independent single-loop controllers.

[0081] Since the opening of the return air valve has a relatively small impact on the NMP concentration and pressure of ovens outside this section and those outside the adjacent section (i.e., the corresponding model gain is one-fifth or even less than that of the oven in this section), and the settling time is also very short, decoupling control can be considered to simplify the controller by replacing the less influential model g with "0". i,j(i = 1, 2, ..., m, j ≠ i, j ≠ i ± 1; or i = m + 1, m + 2, ..., 2m, j ≠ im, j ≠ im ± 1); Similarly, since the external exhaust fan has a relatively small impact on the NMP concentration and pressure of ovens outside this section and outside adjacent sections, g is also replaced by "0". i,j (i = m+1, m+2, ..., 2m, j ≠ im, j ≠ im ± 1), thus we can obtain the new multivariable model matrix G after decoupling under specific operating conditions of the system:

[0082]

[0083] If the oven system does not have particularly high pressure requirements, and the system pressure sensor is unreliable or missing, the exhaust fan in the same section can be controlled with a set of defined adjustment parameters following the valve. This ensures the dynamic balance and stable operation of the pressure inside the oven, guaranteeing the feasibility of the control strategy. The adjustment method is as follows:

[0084] u j =c j u i +d j , i=1,2,…,m, j=m+1,m+2,…,2m (13)

[0085] In the formula, u j and u i c represents the frequency of the exhaust fan and the opening degree of the return air valve in the same section, respectively. j and d i These are the adjustment parameters for the exhaust fan of the i-th oven.

[0086] In summary, this invention, based on the input u and output y of a multivariate control matrix, controls the NMP concentration and pressure within the oven using T... APC It performs real-time optimization control and rolling updates for the operating cycle, and predicts future trends in real time through rolling optimization strategies, while making dynamic corrections. Under the premise of ensuring system stability, it achieves dynamic optimization, which not only ensures system stability, but also enables real-time online optimization of multiple variables. It solves the problem of existing control systems having multiple APC controllers operating independently or being forced to operate partially manually.

[0087] Furthermore, this invention reduces the dimensionality of the multivariate control matrix, thereby reducing its complexity and improving operational efficiency. It also ensures the circulating air temperature gradient and controls the fluctuation of NMP concentration. Under the premise of ensuring that the NMP concentration does not exceed the standard, it improves the stability of the intelligent optimization operation of the drying process after battery coating and reduces energy consumption.

[0088] The above description is merely an embodiment of the present invention and does not constitute any limitation on the technical scope of the present invention. Therefore, any minor modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A multivariate model predictive control system for drying after coating of new energy batteries, applied to a battery coating and drying system, wherein the drying system includes m drying ovens and n sections of main air supply duct and main air return duct, wherein each drying oven includes an electric return air valve, an internal circulation fan, an electric heating element, and an exhaust fan; characterized in that: By adjusting m The frequency of the internal circulation fan and m The return air volume is controlled by adjusting the opening degree of the return air valve; m The temperature inside the oven is controlled by adjusting the power of each electric heating element; m The frequency of the exhaust fan is used to compensate for changes in pressure and NMP concentration inside the oven; by adjusting the frequency of the exhaust fan... n The frequency of the external circulation fan is used to control the NMP concentration in each section of the oven and the total NMP concentration; where n is less than m. The control system T APC The system performs real-time optimization control and rolling updates for each call cycle. Specifically, the control system performs model-based prediction and optimization processing based on the set values ​​of the return air valve opening, exhaust fan frequency, and external circulation fan frequency of the previous call cycle, as well as the current NMP concentration measurement value, oven pressure measurement value, target NMP concentration value, and external disturbance change data. The system optimizes and outputs the optimal sequence of set values ​​for the return air valve opening, exhaust fan frequency, and external circulation fan frequency for the next cycle. Finally, the system writes back the output values ​​of all input variables in the optimal sequence at the first moment to the relevant actuators of the drying system to complete the closed-loop control. The control system defines input variables. U and output variables Y The mathematical relationship of the system's multiple inputs and multiple outputs is expressed as follows: （1） In the formula, G Yes (2) m+ 1)×(3 m+n) The model matrix, where each element is a 1×1 transfer function. U It is 3 m +n The column vector of the row, including m Each return air valve opening u 1. u 2, ... u m , m Internal circulation fan frequency u m+1 , u m+2 ... u 2m , m Each exhaust fan frequency u 2m+1 , u 2m+2 ... u 3m ,as well as n Frequency of external circulation fan u 3m+1 , u 3m+2 ... u 3m+n ; Y It is 2 m+ A column vector with 1 row, including m Drying oven NMP concentration y 1. y 2, ... y m , m Adjust the oven cavity pressure , ... and the total NMP concentration of one drying system Based on the above description, this multivariate control problem can be specifically expressed as: (2) In the formula, Indicates the first j The control variable affects the first... i The mathematical relationship expression between the controlled variables, i.e., the corresponding continuous-time transfer function model; The control system is transformed into a discrete-time system mathematical model based on the continuous-time transfer function model described above. （3） The control system defines a constrained quadratic programming objective function J1, which includes input and output variables for predicting future times, as follows: （4） （5） （6） （7） （8） Among them, symbols Indicates the first k The future k+* periods are based on a control period, where... Known as the control step size, it represents the input sequence. Time span; Known as the prediction step size, it represents the predicted output trajectory. Time span; 、 Represents {k,k+1,k+H} starting from the current k-th period. y The predicted sequence of the future output variables at time {k,k+1,k+Hu} and the input variables at time {k,k+1,k+Hu}; This defines the upper and lower bounds of the output variable; This defines the upper and lower bounds of the input variables. The maximum rate of change; where, Indicates weighted operation Calculation of the 2-norm; Q 1. Q 2. Q 3 correspond to the objective function respectively J The weight matrices of the three summation terms in section 1, Q 1 is H y OK H y Column numerical matrix, Q 2 and Q 3 is H u OK H u Column numerical matrix; where, These are the target values ​​of output variables that users manually set or check based on different batches; The control system is optimized based on the above model G(z) and objective function J1. The optimization control of each APC cycle is completed by solving the input sequence that minimizes the objective function value that satisfies the constraints through quadratic programming. Among them, the first term on the right side of the equation (4) expresses the minimization of the error between the predicted trajectory and the control target, the second term expresses the minimization of the error between the control variable U and its predetermined target, and the third term expresses the minimization of the change index of the control variable U each time.

2. The multivariate model prediction and control system for drying new energy batteries after coating, as described in claim 1, is characterized in that: The specific implementation steps of the control system are as follows: The first step, based on the objective function and constraints, the current values ​​of the input / output variables, and the adjustable weight values ​​Qi (i=1,2,3), output the target value. Solving the quadratic programming problem yields the optimal input sequence that satisfies the constraints. ; The second step is to obtain the optimal input sequence. The first element The controller of the drying system sends the data to form an online closed-loop control, which then sends the remaining input sequence elements... throw away; Thirdly, at the next sampling time, the system will obtain a new output value. Then utilize k Updated at +1 time ( , Repeat the above two steps to obtain the optimal system input for the next step. ; By repeating the above process iteratively, the optimal input setpoint after rolling optimization in each APC cycle can be obtained. It is applied to the control system to complete the application of multivariable model predictive controller.

3. The multivariate model prediction and control system for drying new energy batteries after coating, as described in claim 1, is characterized in that: The control system performs dimensionality reduction on the input variables of the multivariable model: the frequency of the internal circulation fan... u m+1 , u m+2 ... u 2m from U It is separated from the process and does not participate in APC optimization; If the internal circulation fan frequency is not selected for optimization control, the internal circulation fan frequency will only be set to the initial frequency value after batch start-up and shutdown. The initial frequency value can be manually adjusted by the operator, and after reaching a stable state, it can be adjusted to a new fixed value. Alternatively, conditional judgment logic can be programmed to solidify the manual operation into the logic program, but it will not participate in APC control.

4. The multivariate model prediction and control system for drying new energy batteries after coating, as described in claim 1 or 3, is characterized in that: The control system performs dimensionality reduction processing on the external circulation fan frequency in the input variables of the multivariable model matrix, and adopts a linkage design for multiple independent external circulation fan frequencies: selecting the first... i Frequency of external circulation fan As a benchmark, the frequencies of other external circulation fans are adjusted accordingly, and the adjustment methods are as follows: （10） When j=i, aj=1, bj=0, and the original values ​​are not modified.

5. The multivariate model predictive control system for drying new energy batteries after coating, as described in claim 1 or 3, is characterized in that: The control system compares the static gain value of the transfer function of each row of the multivariable model matrix, and when the opening degree of the i-th return air valve is relative to the j-th ( j ≠ i When the model gain for NMP concentration and pressure in the oven is one-fifth or even less than that of the oven in this section, "0" is used to replace the model whose influence is less than the set threshold. ( i =1, 2, ..., m , j ≠ i , j ≠ i ±1; or i = m +1, m +2, ..., 2 m , j ≠ i - m , j ≠ i - m ±1); similarly, when the i-th exhaust fan is applied to the j-th ( j ≠ i When the model gain for NMP concentration and pressure in the oven is one-fifth or even less than that of the oven in this section, it is replaced with "0". ( i = m +1, m +2, ..., 2 m , j ≠ i - m , j ≠ i - m ±1).

6. The multivariate model predictive control system for drying new energy batteries after coating, as described in claim 1 or 3, is characterized in that: The control system performs dimensionality reduction on the input variables of the multivariable model: it links the linear relationship function mapping of the outlet exhaust fan frequency and the inlet return air valve opening of the same section in the multivariable model's input variables, thereby merging the control design of two variables into a single variable control design; its adjustment method is as follows: （13） In the formula, and These represent the frequency of the exhaust fan and the opening degree of the return air valve in the same section, respectively. and For the first i The following adjustment parameters of the oven exhaust fan are set manually according to the air volume balance requirements.

7. The multivariate model prediction and control system for drying new energy batteries after coating, as described in claim 4, is characterized in that: The control system performs dimensionality reduction on the input variables of the multivariable model: it links the linear relationship function mapping of the outlet exhaust fan frequency and inlet return air valve opening of the same section in the multivariable model's input variables, thereby merging the design of two variables into a single variable design; the adjustment method is as follows: （13） In the formula, and These represent the frequency of the exhaust fan and the opening degree of the return air valve in the same section, respectively. and For the first i The following adjustment parameters of the oven exhaust fan are set manually according to the air volume balance requirements.

8. The multivariate model predictive control system for drying new energy batteries after coating, as described in claim 5, is characterized in that: The control system performs dimensionality reduction on the input variables of the multivariable model: it links the linear relationship function mapping of the outlet exhaust fan frequency and inlet return air valve opening of the same section in the multivariable model's input variables, thereby merging the design of two variables into a single variable design; the adjustment method is as follows: （13） In the formula, and These represent the frequency of the exhaust fan and the opening degree of the return air valve in the same section, respectively. and For the first i The following adjustment parameters of the oven exhaust fan are set manually according to the air volume balance requirements.

Images