A mould level control method based on improved Smith prediction compensation

Abstract

The application discloses a kind of crystallizer liquid level control methods based on improved Smith prediction compensation, by setting Smith prediction compensation link in liquid level control system, the pre-estimation and compensation to control process dynamic response are established, so that the leading reaction of the time delay time in system is advanced to liquid level regulator, the advance action of liquid level regulator is realized, including steps:1: according to the signal of the liquid level control system of the time delay time of bar impact is determined;2: with the form of decoupling of set response and disturbance response, the structure setting of Smith predictor is completed, the Smith predictor structure with double controller of tracking controller and disturbance controller is formed3: respectively complete the determination of each parameter in tracking controller and disturbance controller, and according to the determination, the Smith-based prediction compensation is completed.The application can realize the determination of time delay time and the elimination control of system steady-state error.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of steelmaking continuous casting, and particularly relates to a mold liquid level control method based on improved Smith prediction compensation. BACKGROUND

[0002] Stable control of the mold liquid level is crucial for continuous casting safety production and product quality assurance. The liquid steel is injected into the mold through the immersion nozzle by the adjustment of the stopper mechanism in the tundish. The liquid steel forms a casting billet with a certain shell thickness under the action of the mold and cold water. The continuous casting machine pulls out the gradually solidified casting billet from the mold at the set speed. In the continuous casting production process, the mold liquid level is required to be controlled within a certain range. If the liquid level fluctuates too much, it is easy to cause slag inclusion and cracks on the surface of the casting billet, resulting in quality loss, even leakage, and causing serious production accidents. Therefore, great attention is paid to the high-precision control method of the continuous casting mold liquid level. The mold liquid level control adjustment, like most industrial processes, has a time delay problem, which will lead to poor control quality of the mold liquid level, and the Smith prediction control is an effective method to solve the pure time delay process object. However, the application of the classic Smith prediction control in the mold liquid level control system has two difficult problems to solve, 1. The mold liquid level control system is a time-varying lag system, and the lag time varies with different working conditions and steel grades, which is difficult to accurately obtain; 2. Since the mold object is an integral proportional element, the steady-state error of the disturbance is not 0.

[0003] For problem 1, the lag characteristics of the liquid level can be accurately identified in real time according to the data of the mold liquid level adjustment system, which is used to improve the monitoring of the liquid level flow characteristics and improve the control effect. The conventional system identification method is mainly divided into non-parametric model identification method and parametric model identification method. The non-parametric model identification method assumes that the system is linear, but does not need to determine the specific structure of the model in advance, and the parameter model of the system is obtained from the response characteristics according to the continuous time input and output. This method can be applied to any complex system, and is commonly used in engineering. This method mainly includes step response method, frequency method and correlation method; the parametric model identification method must first assume the model structure, and determine the model parameters by minimizing the error function of the model output and the real output of the system. According to different basic principles, this method can be divided into least square method, gradient correction method and maximum likelihood method.

[0004] Due to the many factors affecting the mold liquid level, mutual coupling and signal aliasing, it is difficult to identify the lag time of the mold adjustment system from the existing time domain signals by using the conventional system identification method.

[0005] For problem 2, the prior art has methods such as identifying the object model and the time delay size through the collected step response data, and then setting a prediction function to correct and compensate the accumulated error caused by process disturbance. However, in actual continuous casting production process, it is not possible to set a step response for the mold level control system, which will cause a large impact on the level control system and may cause serious production accidents. Moreover, the method of using a prediction function to control to reduce the steady-state error of disturbance is difficult to select the prediction function, and it is difficult to ensure long-term stable effect.

[0006] How to identify the pure time delay of the mold level control process in real time during production under Smith prediction compensation and ensure that the disturbance steady-state error does not occur has become a problem to be solved in the existing mold level control based on improved Smith prediction compensation.

[0007] The invention application with application number CN201310647691X discloses a "servo system identification method based on relay position feedback time domain characteristics", the implementation steps of which are: setting the parameters of the relay module and the experimental parameters; performing a relay position feedback experiment and recording the corresponding actual displacement oscillation curve; determining the oscillation amplitude and oscillation period; determining the servo system model and its parameters; accurately calculating the gain and time constant of the controlled servo system.

[0008] The invention application with application number CN2019111061938 discloses a "linear quadratic optimal dynamic feedforward-feedback PID control system based on closed-loop identification model and control method thereof", which includes sampling field closed-loop operation data, identifying an ARX model, converting it into an observable canonical form discrete-time state space model, obtaining a measurable disturbance amount feedforward control model; designing a quadratic performance index; minimizing the performance index to obtain a linear state feedback control matrix; calculating the state quantity estimate of the measurable disturbance amount, and then obtaining the feedforward control law of the linear optimal quadratic dynamic feedforward controller; combining the linear quadratic optimal dynamic feedforward controller with the PID controller to design a feedforward-feedback PID control system.

[0009] The invention application with application number CN2022105346458 discloses a "time delay integral process prediction function control method based on Smith prediction correction", which first models according to the real-time collected step response data, finds out the basic characteristics of the object, then performs new error correction and compensation on the prediction function control based on Smith prediction correction, and finally implements the obtained optimal control law on the controlled integral process. SUMMARY

[0010] To solve the above problems, the present application provides a mold level control method based on improved Smith prediction compensation, and the technical scheme is as follows:

[0011] The crystallizer liquid level control method based on improved Smith prediction compensation, by setting Smith prediction compensation link in the liquid level control system, establishing the pre-estimation and compensation of the dynamic response of the control process, so as to advance the reaction of the time delay time in the system to the liquid level regulator, realize the advance action of the liquid level regulator, characterized in that, comprising the following steps:

[0012] S1: According to the ram signal, the time delay time of the liquid level control system is determined;

[0013] S2: The structure of Smith predictor is set in the form of decoupling of set response and disturbance response, forming a Smith predictor structure with tracking controller and disturbance controller double controller;

[0014] S3: The determination of each parameter in the tracking controller and the disturbance controller is completed respectively, and the Smith-based prediction compensation is completed according to the determination.

[0015] Further,

[0016] Step S1 is specifically:

[0017] S11: According to the set sampling time, the actual stopper opening signal and the measured liquid level signal are periodically obtained to form respective time sequence signals;

[0018] S12: The time sequence signal of the actual stopper opening and the time sequence signal of the measured liquid level are respectively processed, and the stopper opening fluctuation signal and the liquid level fluctuation signal are respectively obtained;

[0019] S13: The stopper opening fluctuation signal and the liquid level fluctuation signal are subjected to time-frequency conversion, and the stopper opening phase and the liquid level phase under the set frequency of the stopper signal are determined respectively according to the set frequency of the stopper signal; and the actual phase difference of the stopper signal is determined accordingly;

[0020] S14: According to the set frequency of the stopper signal, the theoretical lag phase of the stopper signal is determined in combination with the structural characteristics of the crystallizer object;

[0021] S15: According to the actual phase difference of the stopper signal determined in step S13 and the theoretical lag phase of the stopper signal determined in step S14, the determination of the time delay time is completed.

[0022] Further,

[0023] Step S14 is specifically: according to the proportional and differential link structure of the crystallizer object, in combination with the set frequency of the stopper signal, the theoretical lag phase of the stopper signal is determined as follows:

[0024]

[0025] In the formula,

[0026] Theoretical lag phase of the ram signal, unit: degree;

[0027] f0: Set frequency of the ram signal, unit: Hz;

[0028] L: Time lag time, unit: S.

[0029] Further,

[0030] The disturbance controller in step S2 is set according to the basic control structure characterized by proportional control for steady-state error and integral control for reducing steady-state error.

[0031] Further,

[0032] The specific structure of the disturbance controller is as follows:

[0033]

[0034] In the formula,

[0035] k1, k2, k3: Optimizable parameters.

[0036] Further,

[0037] The determination of each parameter in the disturbance controller in step S3 is completed on the premise that the steady-state error is zero.

[0038] The crystallizer liquid level control method based on improved Smith prediction compensation disclosed in the present application characterizes the phase lag of the ram signal with model lag, and accordingly, the time lag time is determined through the ram signal; and by setting the Smith predictor as a double controller structure form with tracking controller and disturbance controller, the optimization of the tracking controller and the disturbance controller is formed respectively, so that the control of the disturbance response steady-state error is realized on the premise that the set response and the disturbance response do not interfere with each other, so as to overcome the deterioration of the system disturbance response characteristics caused by the introduction of Smith prediction compensation. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 The Smith prediction compensation step of the present application is shown in the figure;

[0040] Figure 2 The time lag time determination step in the present application is shown in the figure;

[0041] Figure 3 The block diagram of the control system based on improved Smith prediction compensation of the present application is shown in the figure;

[0042] Figure 4This is a block diagram of a conventional Smith-predicted compensation crystallizer level control system, as described in the principle and setup process of this invention.

[0043] Figure 5 This is a flowchart illustrating the lag time identification process in the explanation of the principle and setup process of the present invention.

[0044] Figure 6 This is a simulation model of the improved Smith predictive compensation control system in this embodiment of the invention.

[0045] Figure 7 This is a schematic diagram of the time series of actual opening degree and measured liquid level in an embodiment of the present invention;

[0046] Figure 8 The simulation model of the stopper opening and the phase spectrum of the measuring liquid level impeller component in this embodiment of the invention;

[0047] Figure 9 This is a schematic diagram of the dynamic lag time identification results in an embodiment of the present invention;

[0048] Figure 10 This is a schematic diagram showing the comparison results of the measured liquid level following the set liquid level under different control conditions in an embodiment of the present invention;

[0049] Figure 11 This is a schematic diagram showing the comparison results of measuring liquid level following the set liquid level under different control conditions in an embodiment of the present invention. Detailed Implementation

[0050] The following is a further detailed description of a crystallizer level control method based on improved Smith prediction compensation, with reference to the accompanying drawings and specific embodiments.

[0051] Explanation of principles and setup process:

[0052] like Figure 1 The method shown is a crystallizer level control method based on improved Smith prediction compensation, which, through... Figure 4 The control system shown is an improvement upon the conventional Smith predictive compensation crystallizer level control system. The block diagram of the improved control system is shown below. Figure 3 As shown, the improved system allows for complete decoupling of the setpoint response and the disturbance response. Furthermore, by setting specific structural parameters for H(s), it addresses the issue of steady-state errors in traditional Smith predictive control, where the integral element of the controlled object model cannot eliminate disturbances. Simultaneously, by utilizing real data from the actual production process of the crystallizer level control, such as the stopper opening and measured liquid level, the pure delay time of the crystallizer level control system is identified online based on the punch signal of this technical solution.

[0053] In the crystallizer level control process, there is a pure lag process in the controlled process object, which leads to poor control effect due to the untimely feedback. Smith predictor compensation is an effective means to solve the above problems. By predicting the output results in advance through the mathematical model of the object process and the lag time, and adjusting the system input in time according to the predicted results, in the ideal case, the system regulation performance is the same as that of the non-lag process, only the output is delayed in time. The original PI control system is improved by Smith predictor, and the Smith predictor compensation crystallizer level control system is obtained, as shown in Figure 1 .

[0054] Where s is the Laplace transform operator, Y(s) is the Laplace transform of the measured level, R(s) is the Laplace transform of the set level, K(s) is the set of PI controller and actuator Ga(s), Gp(s) includes flow coefficient and crystallizer object, e -Ls is a pure lag function, and e -L′s is the identified lag function.

[0055] The closed-loop transfer function of the original crystallizer level PI control system is:

[0056]

[0057] The closed-loop transfer function of the Smith predictor compensation crystallizer level control system is:

[0058]

[0059] When L' = L, it can be seen that the closed-loop transfer function is:

[0060]

[0061] The pure lag link of the original PI crystallizer level control system exists in the closed-loop characteristic equation, which will adversely affect the stability and dynamic performance of the system. Through the Smith predictor compensation of the crystallizer level control system, the pure lag link is not included in the characteristic equation of the system, but is transferred outside the closed-loop control loop, so it will no longer have adverse effects on the system. The introduction of Smith predictor well compensates for the system with pure lag, which can effectively improve the stability and dynamic performance of the system.

[0062] However, the introduction of Smith predictor compensation leads to poor disturbance response characteristics, Figure 4 According to Mason's formula, the transfer function of the system from the process disturbance D(s) to the measured level Y(s) is:

[0063] When PI control, the disturbance response is:

[0064]

[0065] The disturbance response of Smith's estimated compensation is:

[0066]

[0067] When L' = L, the disturbance response is:

[0068]

[0069] The transient response of the disturbance depends on the object model and whether the poles of the object are close to the imaginary axis. The crystallizer object process is a proportional-integral model. If there is a constant process disturbance, the influence of the disturbance can exist for a long time in this case. According to the above equation, it is obvious that the Smith's estimated compensation crystallizer level control has poor anti-interference ability. According to the final value theorem, the final value theorem of the disturbance response is further verified:

[0070]

[0071]

[0072] When D(s) is a unit step disturbance, a constant steady-state error exists in the disturbance response This has an adverse effect on the accurate crystallizer level control.

[0073] By changing the structure of the Smith's estimated compensator, the decoupling of the setpoint response and the disturbance response is achieved, so that the setpoint response and the disturbance response can be optimized separately. And through the stopper opening degree and the measured liquid level data, the lag time is identified, and the time lag of the improved Smith's estimated compensator is provided with real-time parameter update. The Smith's estimated compensation crystallizer level control system set by the technology is as shown in Figure 3 The designed control system module includes a PI controller, a liquid level actuator, a flow coefficient, a dynamic lag time, a crystallizer object, an improved Smith's estimated compensator, and a lag time identification module.

[0074] The setpoint response is:

[0075]

[0076] The disturbance response is:

[0077]

[0078] When L' = L,

[0079]

[0080]

[0081] Since the control object Gp(s) is unchangeable, but the set response can optimize K(s) individually, a better following effect on the set liquid level is achieved, and the disturbance response is not affected; the disturbance response can optimize H(s) individually, a better inhibition effect on the disturbance is achieved, and the set response is not affected.

[0082] The optimization method of K(s) can be determined by the conventional PID parameter tuning method in engineering, including: critical proportion method, reaction curve method, attenuation method, and extended critical proportion degree tuning method. The selection of H(s) is more complex. First, the steady-state error of the disturbance response cannot appear, and the overshoot is as small as possible.

[0083] H(s) selects a specific structure, and the structure form is as follows:

[0084]

[0085] Where k1, k2, k3, and L are the system lag time.

[0086] At this time, the disturbance response is:

[0087]

[0088] According to the final value theorem, the final stable liquid level output of the unit step disturbance is:

[0089]

[0090] To ensure that the steady-state error is 0, let:

[0091] k4=k2+k3*L

[0092] That is, it can be ensured that:

[0093]

[0094] By selecting different k1, k2, and k3, different control effects on the disturbance response can be obtained, which can be used as tuning parameters.

[0095] The present technology not only improves the structure of Smith predictor control, but also eliminates the steady-state error of the disturbance through a specific structure, and realizes lag time identification and online adjustment of Smith predictor time lag parameters according to the stopper opening degree and measured liquid level data under different working conditions and steel grades. The technical scheme of real-time lag time identification of the present technology is as follows: Figure 5As shown, the process is as follows: 1) Collect discrete actual stopper opening data and measured liquid level data acquired by the PLC in S7-400 according to the scanning cycle; 2) Preprocess the acquired actual stopper opening and measured liquid level data; 3) Perform Fast Fourier Transform (FFT) and phase angle conversion on the preprocessed actual opening and measured liquid level data in step 2 to obtain the phase of each frequency of the actual opening and measured liquid level; 4) Find the phase difference corresponding to the punch frequency in the actual opening and measured liquid level using the phase obtained in step 3; 5) Calculate the theoretical lag phase of the punch signal from the actual opening to the measured liquid level based on the phase frequency characteristics of the process model from stopper opening to measured liquid level; 6) Calculate the initial lag time L' based on the actual phase difference of the punch signal from the actual opening to the liquid level calculated in step 4 and the theoretical lag phase calculated in step 5 based on the phase frequency characteristics of the process model; 7) Update the time delay parameters in the parameter model of the Smith predictor. The specific process of lag time identification is as follows (can be combined with...). Figure 2 (To understand):

[0096] (1) Collect discrete actual stopper opening data and measured liquid level data collected by PLC in S7-400 according to the scanning cycle. Each time, the actual stopper opening and measured liquid level time series data with the current nearest time span of TN are collected, and the data update window is 1s.

[0097] (2) The actual stopper opening data and measured liquid level data are preprocessed. By subtracting the average value, the fluctuation period signal is obtained from the actual stopper opening and measured liquid level. The processing method is as follows:

[0098] Liquid level fluctuation signal y(n) = Measured liquid level signal - Average value of measured liquid level signal Opening fluctuation signal sn(n) = Actual opening signal - Average value of actual opening signal

[0099] The liquid level fluctuation signal y(n) and the opening fluctuation signal sn(n) mentioned above are discrete-time signals, both with a data length of N and a sampling period of Ts.

[0100] (3) Perform Fast Fourier Transform (FFT) on the liquid level fluctuation signal y(n) and the actual opening signal sn(n) obtained in step 2 to obtain a complex sequence in the frequency domain. The complex sequence is then used for angle calculations in the four quadrants to obtain the phase of the final sampling frequency interval between the liquid level fluctuation signal and the actual opening signal. The specific process is as follows:

[0101] Performing Discrete Fourier Transform (DFT) on the liquid level fluctuation signal y(n) and the opening fluctuation signal sn(n) yields the Discrete Fourier Transform sequence:

[0102]

[0103] In the above formula:

[0104]

[0105] The basic idea of FFT algorithm is to combine some terms in DFT operation and decompose DFT of a long sequence into DFT of short sequences, which is not a new transformation, but a fast algorithm for the simplified operation of discrete Fourier transform, so the frequency domain sequence of the transformation is the same as the above.

[0106] Through Euler's formula, it can be known that:

[0107]

[0108] Through the above process, it can be known that the complex sequence ultimately obtained through FFT fast Fourier transform is:

[0109]

[0110] The angle operation in the four quadrants of the complex sequence is performed, and the phase of the complex sequence is:

[0111]

[0112]

[0113] There is a corresponding relationship between the discrete k and the frequency f, and the relationship is as follows:

[0114]

[0115]

[0116] In the formula, fs is the sampling frequency of the actual opening signal and the measured liquid level, and through the above process, the phases corresponding to the frequency of the actual opening and the measured liquid level at each sampling interval, i.e. ∠Y(f) and ∠SN(f) can be obtained. When k=0, the corresponding frequency f=0, and ∠SN(f)=0 and ∠Y(f)=0, obviously the direct current component has no phase.

[0117] (4) According to the frequency f of the set anti-blocking ram signal o , find the corresponding ram component in the fast Fourier transform FFT results of the actual opening signal and the measured liquid level signal, and according to step 3, respectively obtain the phases ∠SN(f o ) = ph1 and ∠Y(f o ) = ph2 of the actual opening signal and the measured liquid level signal at the frequency, and calculate the phase difference Δph therefrom, and the calculation method is as follows:

[0118]

[0119] (5) According to step 4, the actual phase difference of the tundish rod signal Δph is obtained, and the process model of the actual opening degree of the tundish rod to the liquid level data process is known as:

[0120]

[0121] Where Gc is the flow coefficient, A is the cross-sectional area of the crystallizer, s is the Laplace differential operator, and Gp(s) is the crystallizer object. It is known that the crystallizer object is similar to a container and can be approximated as an integral proportional link. The lag time L needs to be identified.

[0122] The process model of the actual opening degree to the measured liquid level data process can obtain the theoretical phase frequency characteristic as:

[0123]

[0124] The frequency of the tundish rod signal is known as f o According to the theoretical phase frequency characteristic, the theoretical lag phase of the tundish rod signal from the actual opening degree to the measured liquid level can be obtained as:

[0125]

[0126] (6) The actual tundish rod opening action is basically to offset the up and down fluctuations caused by the process disturbance D(s) to the measured liquid level. Due to the existence of the process disturbance D(s), the results of time domain or frequency domain identification using the actual tundish rod opening degree and the measured liquid level are not good. The process disturbance D(s) mainly includes low-frequency disturbances such as bulging and standing waves, as well as some high-frequency noise. But the tundish rod signal is different: 1) The tundish rod signal has a higher frequency, and the energy of the high-frequency noise is relatively dispersed, with a small amplitude near the tundish rod frequency, so the external disturbance has a small effect on the phase of the tundish rod signal; 2) The tundish rod signal is actively acted upon by the tundish rod according to a specific set frequency and amplitude, which is different from the passive action of the tundish rod caused by the feedback of the liquid level fluctuations. The above two points are the main reasons why the tundish rod signal can be used for lag time identification alone.

[0127] The lag time L' is calculated according to the actual phase difference of the tundish rod signal Δph obtained in step 4 and the theoretical lag phase of the tundish rod signal obtained in step 5, and the expression is:

[0128]

[0129] Simplifying the expression gives the lag time calculation expression:

[0130]

[0131] L' is the identified lag time, Δph is the actual phase difference, f o is the frequency of the tundish rod signal, and L minL is the minimum time lag time max L is the maximum time lag time.

[0132] (7) The lag time L' calculated according to step 6 is used to update the Smith predictor parameters online, i.e. the lag time L' is used to improve the stability and dynamic characteristics of the original crystallizer level control system through the Smith predictor.

[0133] Embodiment

[0134] 1. Specific identification method of time lag parameter

[0135] The simulation model of the crystallizer level control system of the embodiment is shown in Fig. 1. Figure 6

[0136] a) The level set value is 64 from 0 to 250 s, and is 84 from 251 to 1000 s.

[0137] b) The PI controller expression is:

[0138]

[0139] c) The level actuator expression is:

[0140]

[0141] d) The anti-blocking rod signal is set to a sine periodic signal with a frequency of 1 Hz and an amplitude of 1.

[0142] e) The process disturbance d(t) includes a standing wave signal, a bulging signal, and a Gaussian white noise. The standing wave signal is set to a frequency of 0.6 Hz and an amplitude of 0.2, and the bulging signal is set to a frequency of 0.055 Hz and an amplitude of 0.2.

[0143] f) The controlled object Gp(s) includes the flow coefficient Gc of the stopper opening degree to flow, and the crystallizer process object expression is:

[0144]

[0145] wherein the flow coefficient Gc = 0.35, and the crystallizer cross-sectional area A = 0.264.

[0146] g) The dynamic lag time e -Ls wherein L varies with time t, and the process is as follows:

[0147] L = L0 + sin(2*pi*f1*t)

[0148] ​L0 is the time delay under normal conditions, L0 = 0.5; f1 is the time delay variation frequency, f1 = 0.001 Hz; t is time; the time delay time range is L ∈ [0, 1].

[0149] h) a specific structure H(s) with the following structure parameters:

[0150]

[0151] Let a set of parameters k1 = 4, k2 = 3, k3 = 0.6, k4 = k2 + k3 * L', L' is the time delay identification result. The final H(s) is:

[0152]

[0153] i) Time delay identification process

[0154] According to step 1, take the simulated actual stopper opening degree and measured liquid level data 200s data, as shown in Figure 7 .

[0155] 1) After data preprocessing by step 2, the periodic fluctuation signals of actual opening degree and measured liquid level are obtained.

[0156] 2) The periodic fluctuation signals of stopper opening degree and measured liquid level obtained in step 2 are subjected to fast Fourier transform (FFT) according to the process of step 3, and angle operation is performed in four quadrants to obtain the FFT phase spectrum, wherein the phase spectrum of the stopper opening degree and the measured liquid level is shown in Figure 8 .

[0157] 3) According to Figure 6 , the phase of the stopper opening degree signal component is ph1 = 264.7°, and the phase of the measured liquid level signal component is ph2 = -6.531°.

[0158] 4) According to the calculation of step 3, the phase of the stopper opening degree signal component ph1 and the phase of the measured liquid level signal component ph2 are calculated, and the actual phase difference of the stopper signal is calculated according to the phase difference expression of step 4:

[0159] Δph = ph2 - ph1 = -271.231°

[0160] 5) According to the process model of the actual opening degree to the measured liquid level of the above simulation model, the theoretical time delay phase of the stopper signal from the actual opening degree to the measured liquid level is calculated according to step 5

[0161]

[0162] 6) According to the time delay calculation expression obtained in step 6, the current time delay is calculated:

[0163]

[0164] 7) According to the hysteresis time L' obtained in the above step 6, update the Smith predictor time delay parameter.

[0165] Simulation experiment

[0166] In the simulation time of 1000s, the real hysteresis time L dynamically changes up and down based on 0.5s. According to the real-time data of the actual opening degree and the measured liquid level at the corresponding time, the hysteresis time L' is repeatedly calculated according to the above i) process. The comparison between the identified hysteresis time result and the actual real hysteresis time is shown in Figure 9 .

[0167] There are some errors between the identified hysteresis time result and the real hysteresis time result, but they are basically within the acceptable range. According to the identified hysteresis time L' result, the model time delay parameter of the Smith predictor is updated in real time, and the crystallizer liquid level control based on the hysteresis time identification of the Smith prediction compensation is realized.

[0168] When the set liquid level is changed from 64mm to 84mm at 250s, it can be known from Figure 7 that the real pure hysteresis time of the system at this time is 1s. The PI control, the Smith prediction compensation control (the time delay parameter is 0.5s), and the improved Smith prediction compensation control based on the hysteresis time identification are compared respectively. The simulation results of the measured liquid level following the set liquid level under different control algorithms are shown in Figure 10 .

[0169] Compared with the traditional Smith prediction control, the improved crystallizer liquid level control system based on time identification in the technical solution can change more quickly and in real time following the set liquid level, has a small overshoot, and quickly reaches a stable state.

[0170] At 500s, a step disturbance is added to the process disturbance. The PI control, the Smith prediction compensation control (the time delay parameter is 0.5s), and the improved Smith prediction compensation control based on the hysteresis time identification are compared respectively. The simulation results of the measured liquid level following the set liquid level under different control algorithms are shown in Figure 11 .

[0171] According to the simulation results of Figure 11 , the conventional Smith prediction control cannot eliminate the steady-state error of the disturbance response, resulting in that the measured liquid level cannot well follow the set liquid level. The improved Smith prediction compensation crystallizer liquid level control system based on the hysteresis time identification can not only eliminate the steady-state error problem of the disturbance response, but also achieve good control accuracy.

[0172] The present technology realizes complete decoupling of setting response and disturbance response by improving the structure of Smith predictor, and solves the problem that Smith prediction control cannot eliminate the stable error of disturbance for crystallizer objects containing integral elements by designing a specific structure of H(s). Moreover, a new lag time identification method is designed and implemented, which can adjust the time lag parameter of the Smith predictor online. The above three simulation results show that the crystallizer liquid level control system based on the improved Smith prediction compensation of lag time identification of the present technical solution can effectively identify the change of lag time online; when the set value changes, compared with PI control and Smith prediction control, the present technical solution can follow the set liquid level with smaller overshoot and faster speed; when sudden disturbances such as crystallization and shedding occur, the present technical solution does not have problems such as steady-state error.

Claims

1. A crystallizer level control method based on improved Smith prediction compensation, which establishes a pre-estimation and compensation for the dynamic response of the control process by setting a Smith prediction compensation link in the level control system, thereby reflecting the controlled variable with time delay in advance to the level regulator, realizing the advance action of the level regulator, characterized in that, Includes the following steps: S1: Determine the time delay of the liquid level control system based on the punch signal; S2: The Smith predictor structure is set up by decoupling the set response and the disturbance response, forming a Smith predictor structure with dual controllers: a tracking controller and a disturbance controller. S3: Determine the parameters of each component in the tracking controller and disturbance controller respectively, and perform Smith-based prediction compensation based on the determined parameters. Step S1 is as follows: S11: According to the set sampling duration, periodically acquire the actual stopper opening signal and the measured liquid level signal to form their respective time series signals; S12: Perform anomaly processing on the time series signal of the actual stopper opening and the time series signal of the measured liquid level to obtain the stopper opening fluctuation signal and the liquid level fluctuation signal, respectively. S13: Perform time-frequency conversion on the stopper opening fluctuation signal and the liquid level fluctuation signal, and determine the stopper opening phase and liquid level phase at the set frequency of the punch signal; and determine the actual phase difference of the punch signal accordingly. S14: Determine the theoretical lag phase of the punch signal based on the set frequency of the punch signal and the structural characteristics of the crystallizer object. S15: Based on the actual phase difference of the punch signal determined in step S13 and the theoretical lag phase of the punch signal determined in step S14, the time delay time is determined.

2. The crystallizer level control method based on improved Smith prediction compensation according to claim 1, characterized in that: Step S14 specifically involves determining the theoretical lag phase of the punch signal based on the proportional differential link structure of the crystallizer object and the set frequency of the punch signal, as follows: In the formula, The theoretical lag phase of the punch signal, in degrees; f0: The set frequency of the punch signal, in Hz; L: Time delay, unit: seconds (S).

3. The crystallizer level control method based on improved Smith prediction compensation according to claim 1, characterized in that: The disturbance controller in step S2 completes the structural setup based on proportional control to characterize steady-state error and integral control to reduce steady-state error.

4. The crystallizer level control method based on improved Smith prediction compensation according to claim 3, characterized in that: The disturbance controller described herein has the following specific structure: In the formula, k1, k2, k3: Optimizable parameters; S: Laplace differential operator; k4 = k2 + k3 * L′; in, L′: Identification lag time, unit: s; Δph: The actual phase difference; f o : Frequency of the punch signal, unit: Hz; L min Minimum time delay, unit: seconds (S); L max Maximum time delay, unit: seconds (S).

5. The crystallizer level control method based on improved Smith prediction compensation according to claim 1, characterized in that: The determination of each parameter in the disturbance controller in step S3 is completed under the constraint that the steady-state error is zero.

Images