A closed-loop control method for automatic switching of high-precision preparation of rare gases
By using an automatic switching closed-loop control method and a fuzzy neural network tuning PID controller during the refining of rare gases, the control inaccurate problem caused by sensor data packet loss is solved, and the purity of krypton xenon and the stability of the preparation process are improved.
Patent Information
- Application Number
- CN202410355722.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-27
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-03-27
AI Technical Summary
During the existing rare gas refining process, packet loss in sensor data transmission leads to inaccurate temperature and pressure control, affecting the purity of krypton xenon.
Using the closed-loop control method of automatic switching, data is collected through pressure sensors and temperature sensors, packet loss detection is performed, appropriate PID algorithm (prediction function PID algorithm or generalized open-loop response PID algorithm) is selected for control, and PID is used using fuzzy neural networks. The controller parameters are tuned to improve the reliability of the control signal.
It improves the accuracy of temperature and pressure control and the stability of the system, enhances the reliability of the output signal, and ensures the high purity preparation of krypton xenon.
Smart Images

Figure CN118034035B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of rare gas refining, and in particular to a closed-loop control method for automatic switching of high-precision preparation of rare gases. Background Art
[0002] Krypton-xenon is a rare gas with very stable chemical properties and is not easy to react with other substances. Its content in nature is extremely rare, with a content of only about 0.000087% in the air. As an inert gas, it plays a role in protecting objects and preventing oxidation in the fields of semiconductors, automobile industry, and medical treatment. After crude extraction, krypton-xenon needs to enter a refining tower for refining to obtain high-purity krypton-xenon. The existing refining process generally includes sending the krypton-xenon concentrate into a crude krypton-xenon tower. A crude krypton-xenon tower condenser evaporator is provided at the top of the crude krypton-xenon tower. The reflux liquid required for distillation is generated by cooling with liquid nitrogen. A crude krypton-xenon tower reboiler controlled by a power control cabinet is provided at the bottom of the tower to generate rising steam at the bottom of the crude krypton-xenon tower. The top of the crude krypton-xenon tower obtains an oxygen delivery system, and the bottom obtains a krypton-xenon concentrate that is sent to a post-krypton-xenon separation tower C2902 for distillation. The pressure and temperature in the process directly affect the purity of krypton-xenon. The sensor may transmit inaccurate data or lose data during data transmission, that is, the system may experience packet loss. At this time, temperature and pressure control may result in untimely and inaccurate regulation. Summary of the invention
[0003] The purpose of the present invention is to provide a closed-loop control method for automatic switching of high-precision preparation of rare gases. The present invention performs packet loss detection and selects different algorithms for control according to the amount of packet loss to improve the accuracy of temperature and pressure control, and uses fuzzy neural networks to perform parameter tuning on PID controllers, thereby further improving the reliability of output signals.
[0004] The technical solution of the present invention is a closed-loop control method for automatic switching of high-precision preparation of rare gas, which is carried out according to the following steps:
[0005] Step S1: The pressure sensor and the temperature sensor collect the pressure and temperature data in the refining tower, and input the pressure and temperature data into the PID controller. The PID controller adjusts the adjustment parameters and sets the first packet loss threshold δ P1 and the second packet loss threshold δ P2 ;
[0006] Step S2: PID controller performs packet loss detection to obtain packet loss amount δ LT , if 0<δ LT <δ P1 , the PID controller uses the prediction function PID algorithm for control; if δ P1 <δ LT <δ P2PID controller uses generalized open-loop response PID algorithm for control; if δ LT >δ P2 , then reduce the flow of pressure and temperature data, keep the system running at a low speed, and issue an alarm;
[0007] Step S3: inputting the output of the prediction function PID algorithm or the generalized open-loop response PID algorithm into the fuzzy neural network to adjust three adjustment parameters of the PID controller;
[0008] Step S4: The adjusted PID controller outputs temperature and pressure control signals. The control signals are used for krypton-xenon refining after system compensation and are fed back to the PID controller at the same time.
[0009] In the above-mentioned closed-loop control method for automatic switching of high-precision rare gas preparation, the calculation process of the generalized open-loop response PID algorithm is as follows:
[0010] Step A1: Set the process model of the controlled object as:
[0011]
[0012] Among them, y(z), u(z), d i (z) are the output variables, input variables and z-transform of the ith disturbance in the process model; G R (z) is the impulse transfer function of the process model; G i (z) is the impulse transfer function of the ith disturbance;
[0013] Step A2: If G i The unstable pole of (z) is also G R (z) is the extreme point, then the process model of step A1 is written as:
[0014]
[0015] in, a 0 =1; β 0 ≠0; b 0 ≠0;
[0016] Step A3: The process model of step A2 is further simplified as follows:
[0017]
[0018] in, β 0 ≠0; b 0 ≠0;
[0019] Step A4: When packet loss occurs, the output of the PID controller is expressed as:
[0020]
[0021] Among them, y f (z) is the output of the system in the controller; δ fn (z) is the number of consecutive packet losses; 0<δ fn (z)<δ fmax ; δ fmax is the maximum number of consecutive packet losses;
[0022] Substituting into the simplified process model of step A3, we get:
[0023]
[0024] When δ fn When (z) = 0, When δ fn When (z) = 0, When δ fn (z) = δ fmax hour,
[0025] Step A5: Build the Smith estimator with packet loss compensation:
[0026]
[0027] in:
[0028]
[0029] Define generalized input as:
[0030]
[0031] in,
[0032] The transfer function of the generalized open-loop system is:
[0033]
[0034] Among them, H(z -1 )=M(z -1 )D(z -1 )+N + (z -1 )T(z -1 );
[0035] In the generalized open-loop response PID algorithm, the expression of the feedback controller is as follows:
[0036]
[0037] Where K is the steady-state gain of the generalized open-loop system, expressed as N + (1) / H(1), and So the generalized input is:
[0038]
[0039] The control quantity calculation formula of the generalized open-loop response PID algorithm is derived as follows:
[0040]
[0041] Step A6: Substitute the parameter F(z) fed back to the PID controller into the generalized open-loop response PID algorithm, expressed as:
[0042]
[0043] In the aforementioned closed-loop control method for automatic switching of high-precision preparation of rare gas, the calculation process of the prediction function PID algorithm is carried out according to the following steps:
[0044] Step B1: Set the prediction model at time k+t as:
[0045] k+ty m (k) = y u (k)+y h (k);
[0046] Where: y m (k) is the predicted output of the model at that moment; y u (k) is the output of the model under the control quantity; y h (k) is the control action output;
[0047] Step B2: Select the first-order exponential as the reference trajectory and introduce it in the form of feedforward:
[0048]
[0049] Among them, y f (k+t) is the reference trajectory value corresponding to time k+t; c(k) is the set value at time k; N c is the number of polynomial expansions; b j (k) is the coefficient of the polynomial; t is time; β t is the attenuation coefficient corresponding to the system approaching the set value; y p (k) is the actual output value of the process corresponding to time k;
[0050] Step B3: Estimate the error and calibrate the model input. The specific formula is as follows:
[0051]
[0052] Among them, e(k+1) is the output error of the system at time k+1; β j (k) is the input spectrum of the polynomial fitting coefficients. The finite-time control input is a specific family of functions related to the system object and the set value trajectory;
[0053] Structuring and normalizing the control input, using the relevant basis function f j The linear combination of describes the control input at any time:
[0054]
[0055] Among them, u(k+t) is the control quantity corresponding to the time k+t; f j (t) is the value of the jth basis function at time t; μ j is the linear combination coefficient of the basis function;
[0056] Step B4: Select the first-order inertia delay link as the model of the refining tower temperature and pressure control object to describe the dynamic characteristic function between the heat flow rate and the controlled temperature and pressure:
[0057]
[0058] Among them, K m is the prediction gain; T m is the time constant; T d is the lag time;
[0059] The discrete difference equation corresponding to the zero-order holder model is as follows:
[0060]
[0061] Among them, x m (k+h p ) represents the k+h in the model p , the output value prediction function within the time is optimized by the algorithm and the prediction output is described by the following formula:
[0062] y m (k+h p )=y m (k+h p )+e(k+h p );
[0063] Step B5: Get the output of the prediction function PID algorithm under the refining tower temperature and pressure at time k:
[0064]
[0065] In the aforementioned closed-loop control method for automatic switching of high-precision preparation of rare gas, the PID calculation part in the prediction function PID algorithm and the generalized open-loop response PID algorithm adopts a fuzzy PID algorithm, and the specific steps are as follows:
[0066] Step C1: Get the input value r(t) and the system output value y(t), obtain the error e(t) by the difference between the two, and calculate the error change rate ec(t). The calculation formula is as follows:
[0067] ec(t)=e(t)-e(t-1);
[0068] Step C2: Fuzzify e(t) and ec(t) respectively and correspond them to the corresponding domains;
[0069] Step C3: Select the membership functions of e(t) and ec(t), and calculate the membership of e(t) and ec(t) in their corresponding fuzzy subsets respectively;
[0070] Step C4: construct a fuzzy rule base, and perform fuzzy reasoning on the adjustment parameters Kp, Ki, and Kd of the PID control according to the fuzzy rules to obtain the fuzzy control table of the proportional Kp, integral Ki, and differential Kd of the response;
[0071] Step C5: After quantizing the fuzzy control table, the control output signals of proportion Kp, integral Ki and differential Kd are obtained.
[0072] In the closed-loop control method for automatic switching of high-precision rare gas preparation described above, in step S3, the calculation process of the fuzzy neural network is performed according to the following steps:
[0073] Step D1: Set the system y(k) and performance index J:
[0074] y(k)=f[Y(k-1),u(k-1-d),U(k-2-d)];
[0075] J = 0.5 × [r(k) - y(k)] 2 ;
[0076] Where d is the hysteresis, Y(k-1) = [y(k-1), y(k-2), …, y(kn)], y(k) is the system output at time k, U(k-2-d) = [u(k-2-d), …, u(k-2-dm)], u(k) is the system input at time k, and r(k) is the system expected value at time k;
[0077] Step D2: Use the gradient descent method to adjust the three parameters of the fuzzy neural network. The formula is as follows:
[0078]
[0079]
[0080]
[0081] Among them, α is the learning rate, J represents the performance index of the system, q and q+1 represent the qth and q+1th moments respectively. represents the center of the output fuzzy set in the lth control rule at the qth time, Represents the i-th input language variable x in the l-th rule at the q-th time i (k) is the center of the membership function, Represents the i-th input language variable x in the l-th rule at the q-th time i (k) The width of the membership function;
[0082] Step D3: Let Then the system input u(k) at time k is written as:
[0083]
[0084] The derivative is:
[0085]
[0086]
[0087]
[0088] Three setting parameters are obtained through calculation.
[0089] In the closed-loop control method for automatic switching of high-precision rare gas preparation, the fuzzy neural network calculation process also includes compensation amount determination, which is performed according to the following steps:
[0090] Set up a simplified linear function of the mathematical model of the controlled object:
[0091]
[0092] The parameters are obtained by online recursive least squares method. The recursive formula is as follows:
[0093]
[0094] Among them, h(k)=[-y(k-1)...-y(kn)u(k-1)...u(km)] T , P(k) and K(k) are all intermediate variables. are the parameters in the mathematical model of the system;
[0095] According to the mathematical model of the system, the output of the system in the future d+1 time can be obtained, and the output prediction formula is as follows:
[0096]
[0097] ...
[0099]
[0100]
[0101] Among them, a 1 ,a 2 ,…,a n ; b 1 ,b 2 ,…,b m is the parameter of the mathematical model of the system, and d is the lag time. The trust coefficient e is used here to construct the compensation value Δμ, where 0≤ε≤1. The constructed compensation value Δμ is as follows:
[0102]
[0103] in, Successive recursion to obtain K and ε are the proportionality coefficient and the trust coefficient respectively.
[0104] In the closed-loop control method for automatic switching of high-precision rare gas preparation described above, the construction process of the fuzzy neural network in step S3 is carried out according to the following steps:
[0105] Step E1: Introduce BP neural network into the topology structure;
[0106] Step E2: Initialize the activation function center value, width and output layer weight, and select the learning efficiency and inertia factor;
[0107] Step E3: Perform fuzzy reasoning on the PID controller and calculate the error, and adjust the excitation function parameters and weights according to the error;
[0108] Step E4: Repeat step E3 until the learning is completed.
[0109] Compared with the prior art, the present invention collects the pressure and temperature data in the refining tower through the pressure sensor and the temperature sensor, and the PID controller performs packet loss detection according to the data and the control signal to obtain the specific packet loss amount. LT <δ P1 , the PID controller uses the prediction function PID algorithm with higher computing speed and less calculation to control; if δP1 <δ LT <δ P2 PID controller adopts the generalized open-loop response PID algorithm with stronger compensation ability, better stability and adjustable robustness for control; if δ LT >δ P2 , the flow of pressure and temperature data is reduced, the system is kept running at a low speed, and an alarm is issued at the same time; the output of the prediction function PID algorithm or the generalized open-loop response PID algorithm is input into the fuzzy neural network to adjust the three adjustment parameters of the PID controller, and the adjusted PID controller outputs temperature and pressure control signals, which are used for krypton-xenon refining after system compensation and are fed back to the PID controller at the same time; the present invention selects different algorithms for control according to the packet loss amount, improves the accuracy of temperature and pressure control, and uses the fuzzy neural network to adjust the parameters of the PID controller, thereby further improving the reliability of the output signal. BRIEF DESCRIPTION OF THE DRAWINGS
[0110] Figure 1 It is a schematic diagram of the process of the present invention;
[0111] Figure 2 It is the system structure diagram of PFC-PID algorithm and GORC-PID algorithm;
[0112] Figure 3 It is a flow chart of the fuzzy PID algorithm;
[0113] Figure 4 It is the control schematic diagram of fuzzy PID algorithm;
[0114] Figure 5 It is the control schematic diagram of fuzzy neural network;
[0115] Figure 6 It is a schematic diagram of the algorithm of fuzzy neural network;
[0116] Figure 7 It is a schematic diagram of the topological structure of the BP neural network;
[0117] Figure 8 It is a schematic diagram of the learning process of the BP neural network. DETAILED DESCRIPTION
[0118] The present invention is further described below in conjunction with the accompanying drawings and embodiments, but they are not intended to limit the present invention.
[0119] Embodiment: A closed-loop control method for automatic switching of high-precision rare gas preparation, as shown in the attached Figure 1 As shown in the attached Figure 1 As shown, follow the steps below:
[0120] Step S1: The pressure sensor and the temperature sensor collect the pressure and temperature data in the refining tower, and input the pressure and temperature data into the PID controller. The PID controller sets the adjustment parameters and sets the first packet loss threshold δ P1 and the second packet loss threshold δ P2 ;
[0121] Step S2: PID controller performs packet loss detection, definition:
[0122]
[0123]
[0124] where ε k ,φ k represents a random variable, when ε k ,φ k When at least one of them is not 0, record the packet loss at that time δ LT ; as attached Figure 2 As shown, if 0<δ LT <δ P1 , the PID controller uses the prediction function PID algorithm (PFC-PID algorithm) for control; if δ P1 <δ LT <δ P2 ; The PID controller uses the generalized open-loop response PID algorithm (GORC-PID algorithm) for control; if δ LT >δ P2 , the flow of pressure and temperature data is reduced to keep the system running at a low speed and an alarm is issued, where δ P1 =0.01%,δ P2 =0.05%;
[0125] The PFC-PID algorithm can predict the trend of future system changes based on the model, and can make corresponding changes in time according to the future trend of changes. The calculation amount of this control algorithm is greatly reduced, which makes the system response speed faster, and it has strong robustness to overcome external interference. Then, by using the characteristics of PID control, the advantages of PFC control and PID control are integrated to further improve the control effect and control quality of the system; therefore, at this time, there are two controllers in the temperature, pressure, and pressure control system: PFC controller and PID controller. The control loop composed of PID controller and controlled object is the generalized controlled object of PFC controller. The control quantity calculated by PFC controller is the set value of PID controller, and PFC controller performs the set value optimization work; due to the adoption of transparent control strategy, when the system is in advanced control state, PID controller and PFC controller run simultaneously. The calculation process of PFC-PID algorithm is carried out according to the following steps:
[0126] Step B1: The model function output and the model free output constitute the refining tower temperature and pressure control prediction model. The control basis and composition basis of the model prediction algorithm is the prediction model, which mainly predicts the process output value of the system at the future moment through the past state of the object and the input state of the system. The expression of the prediction model at the k+t moment is:
[0127] y m (k) = y u (k)+y h (k);
[0128] Where: y m (k) is the predicted output of the model at that moment; y u (k) is the output of the model under the control quantity; y h (k) is the control action output;
[0129] Step B2: In order to smoothly control the temperature and pressure of the refining tower, it is necessary to design a reference trajectory according to the expected output value and the actual output value of the system during the control process, so as to shorten the error between the actual value and the set value as much as possible, and reduce the difference between the system output value and the tracking set value. The reference trajectory of the gradually stable refining tower temperature and pressure control system usually selects the first-order exponent:
[0130]
[0131] Among them, y f (k+t) is the reference trajectory value corresponding to time k+t; c(k) is the set value at time k; N c is the number of polynomial expansions; b j (k) is the coefficient of the polynomial; t is time; β t is the attenuation coefficient corresponding to the system approaching the set value; y p (k) is the actual output value of the process corresponding to time k;
[0132] Step B3: There will be a deviation between the system output value and the set value during the prediction process. The reference trajectory is introduced in the form of feedforward, the error is estimated by the estimator, the model input is corrected, and the accuracy of the input in the prediction process is improved. The specific formula is as follows:
[0133]
[0134] Among them, e(k+1) is the output error of the system at time k+1; β j (k) is the input spectrum of the polynomial fitting coefficients. The finite-time control input is a specific family of functions related to the system object and the set value trajectory;
[0135] Structuring and normalizing the control input, using the relevant basis function fj The linear combination of describes the control input at any time:
[0136]
[0137] Among them, u(k+t) is the control quantity corresponding to the time k+t; f j (t) is the value of the jth basis function at time t; μ j is the linear combination coefficient of the basis function;
[0138] Step B4: Select the first-order inertia delay link as the model of the refining tower temperature and pressure control object to describe the dynamic characteristic function between the heat flow rate and the controlled temperature and pressure:
[0139]
[0140] Among them, K m is the prediction gain; T m is the time constant; T d is the lag time;
[0141] The discrete difference equation corresponding to the zero-order holder model is as follows:
[0142]
[0143] Among them, x m (k+h p ) represents the k+h in the model p , the output value prediction function within the time is optimized by the algorithm and the prediction output is described by the following formula:
[0144] y m (k+h p )=y m (k+h p )+e(k+h p );
[0145] Step B5: Get the output of the PFC-PID algorithm under the refining tower temperature and pressure at time k:
[0146]
[0147] GORC is an advanced control algorithm, which is regarded as a simplified model predictive control. At the same time, the GORC control algorithm has low requirements on the model, simple parameter setting, good tracking set value capability and anti-interference ability, and the robustness is adjustable online. GORC control can effectively deal with problems such as pure lag and large inertia of the process. This control scheme does not require high accuracy of the model, and GORC control has good tracking set value capability and anti-interference ability, which can improve the tracking speed of the system and further improve the system performance; the advantages of GORC control and PID control are integrated to further improve the control effect and control quality of the system; therefore, the GORC controller and PID controller run simultaneously in the system, and the calculation process of the GORC-PID algorithm is carried out according to the following steps:
[0148] Step A1: Set the process model of the controlled object as:
[0149]
[0150] Among them, y(z), u(z), d i (z) are the output variables, input variables and z-transform of the ith disturbance in the process model; G R (z) is the impulse transfer function of the process model; G i (z) is the impulse transfer function of the ith disturbance;
[0151] Step A2: If G i The unstable pole of (z) is also G R (z) is the extreme point, then the process model of step A1 is written as:
[0152]
[0153] in, a 0 =1; β 0 ≠0;
[0154] b 0 ≠0;
[0155] Step A3: The process model of step A2 is further simplified as follows:
[0156]
[0157] in, β 0 ≠0; b 0 ≠0;
[0158] Step A4: When packet loss occurs, the output of the PID controller is expressed as:
[0159]
[0160] Among them, y f (z) is the output of the system in the controller; δ fn (z) is the number of consecutive packet losses; 0<δ fn (z)<δ fmax ; δ fmax is the maximum number of consecutive packet losses;
[0161] Substituting into the simplified process model of step A3, we get:
[0162]
[0163] When δ fn When (z) = 0, When δ fn When (z) = 0, When δ fn (z) = δ fmax hour,
[0164] Step A5: Build a Smith estimator for packet loss compensation, where
[0165]
[0166] Define generalized input as:
[0167]
[0168] in,
[0169] The transfer function of the generalized open-loop system is:
[0170]
[0171] Among them, H(z -1 )=M(z -1 )D(z -1 )+N + (z -1 )T(z -1 );
[0172] In the GORC-PID algorithm, the expression of the feedback controller is as follows:
[0173]
[0174] Where K is the steady-state gain of the generalized open-loop system, expressed as N +(1) / H(1), and So the generalized input is:
[0175]
[0176] The control quantity calculation formula of the GORC-PID algorithm is derived as follows:
[0177]
[0178] Step A6: Substitute the parameter F(z) fed back to the PID controller into the GORC-PID algorithm, expressed as:
[0179]
[0180] Since complex systems have many problems such as nonlinearity, variable influencing factors, and response delay, it is difficult for conventional PID controllers to effectively control them. Therefore, the PID calculation part of the PFC-PID algorithm and the GORC-PID algorithm uses a fuzzy PID algorithm, as shown in the attached Figure 3 and attached Figure 4 As shown, the specific steps are as follows:
[0181] Step C1: Get the input value r(t) and the system output value y(t), obtain the error e(t) by the difference between the two, and calculate the error change rate ec(t). The calculation formula is as follows:
[0182] ec(t)=e(t)-e(t-1);
[0183] Step C2: Fuzzify e(t) and ec(t) respectively and correspond them to the corresponding domains;
[0184] Step C3: Select the membership functions of e(t) and ec(t), and calculate the membership of e(t) and ec(t) in their corresponding fuzzy subsets respectively;
[0185] Step C4: A fuzzy control rule base and a knowledge base are established based on the operator's experience and knowledge of thermal technology experts, and fuzzy reasoning is performed on the adjustment parameters Kp, Ki, and Kd of the PID control based on fuzzy rules and fuzzy mathematics to obtain the fuzzy control tables of the proportional Kp, integral Ki, and differential Kd of the response. For example, the reasoning process for the proportional Kp is as follows:
[0186] (1)Synthetic reasoning
[0187] The input of the inference system is the error E and error change rate EC between the measured value PV of the refining tower temperature and pressure and the expected value SP of the refining tower temperature and pressure, and the output is the proportional coefficient Kp of the PID controller. The mathematical description of its synthetic inference process is:
[0188]
[0189]
[0190] (2) De-gelatinization of PID controller output parameters Kp, Ki, and Kd
[0191] In fuzzy control, there are many defuzzification algorithms, mainly including: centroid method, maximum membership method, mean method and other algorithms, among which the centroid method and maximum membership method are the most widely used. According to the needs of the temperature and pressure system production process, this embodiment uses the centroid method defuzzification algorithm to obtain the fuzzy control table of PID adjustment parameters Kp, Ki, and Kd: from the error E, the error change rate EC and the membership value of the proportional factor Kp, integral factor Ki and differential factor Kd of the PID controller, and then according to the control rules of Kp, Ki and Kd of PID adjustment, the control table of the proportional Kp, integral Ki and differential Kd of the PID controller is obtained, as shown in the following Tables 1-3 respectively.
[0192]
[0193] Table 1
[0194]
[0195] Table 2
[0196]
[0197] Table 3
[0198] Step C5: After quantizing the fuzzy control table, the control output signals of proportion Kp, integral Ki and differential Kd are obtained. According to the control table of proportional factor Kp, it can be seen that the horizontal axis is the domain range [-1212] corresponding to the error change rate, which is the temperature and pressure signals collected by the on-site refining tower temperature and pressure transmitter, which are transmitted to the computer and then quantized by the computer and mapped to the interval [-12 12]; here, only the quantization factor needs to be determined, and the size of the quantization factor is determined by the temperature and pressure variation range of the refining tower collected by the transmitter and the fuzzy action domain; the ordinate is the domain range corresponding to the error [-1212], and the variation range of the error signal only needs to be quantified to obtain the corresponding fuzzy domain range; after the quantization factor is determined, the actual values of the error E and the error change rate EC are mapped to the corresponding domain interval for approximate processing, and the error and the error change rate are mapped to the value within the fuzzy domain range through the nearest rule. This operation process is called the quantization processing of the fuzzy domain; the remaining two output integral coefficients Ki and differential coefficients Kd of the PID controller are processed by the same processing process; the output of the fuzzy logic reasoning system is a fuzzy quantity, and an accurate physical quantity is obtained after proportional quantization operation. Referring to the control table of the proportional coefficient Kp, the integral coefficient Ki and the differential coefficient Kd, the control output signals of the three output parameters of the fuzzy logic reasoning system, namely, proportional, integral and differential, are obtained. After quantization operation, it represents an exact physical signal.
[0199] Step S3: inputting the output of the prediction function PID algorithm or the generalized open-loop response PID algorithm into the fuzzy neural network to adjust three adjustment parameters of the PID controller;
[0200] As attached Figure 5 and attached Figure 6 As shown, the calculation process of the fuzzy neural network to obtain the setting parameters in step S3 is carried out according to the following steps:
[0201] Step D1: Set the system y(k) and performance index J:
[0202] y(k)=f[Y(k-1),u(k-1-d),U(k-2-d)];
[0203] J = 0.5 × [r(k) - y(k)] 2 ;
[0204] Where d is the hysteresis, Y(k-1) = [y(k-1), y(k-2), …, y(kn)], y(k) is the system output at time k, U(k-2-d) = [u(k-2-d), …, u(k-2-dm)], u(k) is the system input at time k, and r(k) is the system expected value at time k;
[0205] Step D2: Use the gradient descent method to adjust the parameters of the fuzzy neural network. The formula is as follows:
[0206]
[0207] Among them, α is the learning rate, J represents the performance index of the system, q and q+1 represent the qth and q+1th moments respectively. represents the center of the output fuzzy set in the lth control rule at the qth time, Represents the i-th input language variable x in the l-th rule at the q-th time i (k) is the center of the membership function, Represents the i-th input language variable x in the l-th rule at the q-th time i (k) The width of the membership function;
[0208] Step D3: Let
[0209]
[0210] Then the system input u(k) at time k is written as:
[0211]
[0212] According to the derivation rule, we have
[0213]
[0214]
[0215]
[0216] Substituting the above three equations into step D2, we get:
[0217]
[0218]
[0219]
[0220] The compensation setting parameters are obtained through calculation.
[0221] Determine the compensation amount: Assume that the mathematical model of the controlled object can be simplified into a linear function:
[0222] The parameters of the simplified model can be obtained by online recursive least squares method. The recursive formula is as follows:
[0223]
[0224] Among them, h(k)=[-y(k-1)...-y(kn)u(k-1)...u(km)] T , P(k) and K(k) are both intermediate variables, are the parameters in the mathematical model of the system;
[0225] According to the mathematical model of the system, the output of the system in the future d+1 time can be obtained, and the output prediction formula is as follows:
[0226]
[0227] ...
[0229]
[0230]
[0231] The above formula is the established prediction mathematical model, where a 1 ,a 2 ,…,a n ; b 1 ,b 2 ,…,b m is the parameter of the mathematical model of the system, d is the lag time, and the trust coefficient e is used to construct the compensation amount Δμ, where 0≤ε≤1; the constructed compensation amount Δμ is as follows:
[0232]
[0233] in, By recursion, we can get K and ε are the proportionality coefficient and the trust coefficient respectively.
[0234] The calculation process of the fuzzy neural network in step S3 is performed according to the following steps:
[0235] Step E1: Introduce BP neural network into the topology structure; Figure 7 As shown, the topological structure of the BP neural network includes:
[0236] (1) Input layer
[0237] This layer contains two nodes, namely displacement error and error rate, which play the role of transferring input variables to the lower layer. The input and output expressions are:
[0238]
[0239]
[0240] Among them, x i is the i-th input variable
[0241] (2) Fuzzy layer
[0242] This layer contains 2 inputs, each of which corresponds to 7 linguistic variable values, and has 14 nodes. The input is converted into a fuzzy vector after fuzzification processing in this layer, and then the membership function of each linguistic variable value is generated. The input and output expressions are:
[0243]
[0244]
[0245] In the formula, c ij is the membership function center, σ ij is the membership function basis width, is the jth linguistic variable value of the ith input variable.
[0246] (3) Fuzzy reasoning layer
[0247] This layer contains 49 nodes, each of which corresponds to a fuzzy rule. The nodes in this layer use the fuzziness of the error and error change rate of the fuzzification layer and use this as the premise criterion for fuzzy reasoning output. The input and output expressions are:
[0248]
[0249]
[0250] In the formula, i 1 ={1,2,…,7},i 2 ={1,2,…,7}, l is a regular point, l = 1,2,…,49
[0251] (4) Normalization layer
[0252] This layer contains 49 nodes like the third layer. Its function is to normalize the output of the fuzzy inference layer. The input and output expressions are:
[0253]
[0254]
[0255] Where l = 1, 2, ..., 49
[0256] (5) Output layer
[0257] This layer contains 3 nodes, whose function is to perform defuzzification operations and generate output control quantities. The input and output expressions are:
[0258]
[0259]
[0260] In the formula, k = 1, 2, 3, the output result is
[0261] Optimize some parameters in the fuzzy neural network PID control algorithm. Given the system output value is r(k) and the actual system output value is y(k), the correction error function is defined as
[0262]
[0263] The learning algorithm for the output layer network weights is:
[0264]
[0265] In the formula, α is the inertia factor
[0266] The gradient optimization method is used to adjust the center and base width of the membership function of the fuzzy layer:
[0267]
[0268]
[0269] Where η is the learning rate;
[0270] As attached Figure 8 As shown, the learning steps of BP neural network are:
[0271] (1) Initialization: assign values to all weights and thresholds and set them to the smallest arbitrary number;
[0272] (2) Provide input sample training set x = {x 1 ,x 2 ,…,x n} and the expected output response d = {d 1 ,d 2 ,...,d n};
[0273] (3) Calculate the outputs of neurons in layers other than the input layer;
[0274] (4) Calculate error function and cost function;
[0275] (5) Adjust the weights of the hidden layer and output layer:
[0276] (6) Return to step (3) until the error control requirements are met.
[0277] Step E2: Initialize the activation function center value, width and output layer weight, and select the learning efficiency and inertia factor;
[0278] Step E3: Perform fuzzy reasoning on the PID controller and calculate the error, and adjust the excitation function parameters and weights according to the error;
[0279] Step E4: Repeat step E3 until the learning is completed.
[0280] Step S4: The adjusted PID controller outputs temperature and pressure control signals. The control signals are used for krypton-xenon refining after system compensation and are fed back to the PID controller at the same time.
Claims
1. A closed-loop control method for automatic switching of high-precision preparation of rare gases, Features: Follow these steps: Step S1: The pressure sensor and the temperature sensor collect the pressure and temperature data in the refining tower, and input the pressure and temperature data into the PID controller. The PID controller adjusts the adjustment parameters and sets the first packet loss threshold δ P1 and the second packet loss threshold δ P2 ; Step S2: PID controller performs packet loss detection to obtain packet loss amount δ LT , if 0<δ LT <δ P1 , the PID controller uses the prediction function PID algorithm for control; if δ P1 <δ LT <δ P2 PID controller uses generalized open-loop response PID algorithm for control; if δ LT >δ P2 , then reduce the flow of pressure and temperature data, keep the system running at a low speed, and issue an alarm; Step S3: inputting the output of the prediction function PID algorithm or the generalized open-loop response PID algorithm into the fuzzy neural network to adjust three adjustment parameters of the PID controller; Step S4: The adjusted PID controller outputs temperature and pressure control signals. The control signals are used for krypton-xenon refining after system compensation and are fed back to the PID controller at the same time.
2. The closed-loop control method for automatic switching of high-precision rare gas preparation according to claim 1, Features: The calculation process of the generalized open-loop response PID algorithm is as follows: Step A1: Set the process model of the controlled object as: Among them, y(z), u(z), d i (z) are the output variables, input variables and z-transform of the ith disturbance in the process model; G R (z) is the impulse transfer function of the process model; G i (z) is the impulse transfer function of the ith disturbance; Step A2: If G i The unstable pole of (z) is also G R (z) is the extreme point, then the process model of step A1 is written as: Among them, a 0 =1; b 0 ≠0; b 0 ≠0; Step A3: The process model of step A2 is further simplified as follows: in, β 0 ≠0; b 0 ≠0; Step A4: When packet loss occurs, the output of the PID controller is expressed as: Among them, y f (z) is the output of the system in the controller; δ fn (z) is the number of consecutive packet losses; 0<δ fn (z)<δ fmax ; δ fmax is the maximum number of consecutive packet losses; Substituting into the simplified process model of step A3, we get: When δ fn When (z) = 0, When δ fn When (z) = 0, When δ fn (z) = δ fmax hour, Step A5: Build the Smith estimator with packet loss compensation: in: Define generalized input as: in, The transfer function of the generalized open-loop system is: where, H(z -1 ) = M(z -1 )D(z -1 ) + N + (z -1 )T(z -1 ); In the generalized open-loop response PID algorithm, the expression of the feedback controller is as follows: Where K is the steady-state gain of the generalized open-loop system, expressed as N + (1) / H(1), and So the generalized input is: The control quantity calculation formula of the generalized open-loop response PID algorithm is derived as follows: Step A6: Substitute the parameter F(z) fed back to the PID controller into the generalized open-loop response PID algorithm, expressed as:
3. The closed-loop control method for automatic switching of high-precision rare gas preparation according to claim 1, Features: The calculation process of the prediction function PID algorithm is carried out in the following steps: Step B1: Set the prediction model at time k+t as: and m (k)=y u (k)+y h (k); Where: y m (k) is the predicted output of the model at that moment; y u (k) is the output of the model under the control quantity; y h (k) is the control action output; Step B2: Select the first-order exponential as the reference trajectory and introduce it in the form of feedforward: Among them, y f (k+t) is the reference trajectory value corresponding to time k+t; c(k) is the set value at time k; N c is the number of polynomial expansions; b j (k) is the coefficient of the polynomial; t is time; β t is the attenuation coefficient corresponding to the system approaching the set value; y p (k) is the actual output value of the process corresponding to time k; Step B3: Estimate the error and calibrate the model input. The specific formula is as follows: Among them, e(k+1) is the output error of the system at time k+1; β j (k) is the input spectrum of the polynomial fitting coefficients. The finite-time control input is a specific family of functions related to the system object and the set value trajectory; Structuring and normalizing the control input, using the relevant basis function f j The linear combination of describes the control input at any time: Among them, u(k+t) is the control quantity corresponding to the time k+t; f j (t) is the value of the jth basis function at time t; μ j is the linear combination coefficient of the basis function; Step B4: Select the first-order inertia delay link as the model of the refining tower temperature and pressure control object to describe the dynamic characteristic function between the heat flow rate and the controlled temperature and pressure: Among them, K m is the prediction gain; T m is the time constant; T d is the lag time; The discrete difference equation corresponding to the zero-order holder model is as follows: Among them, x m (k+h p ) represents the k+h in the model p , the output value prediction function within the time is optimized by the algorithm and the prediction output is described by the following formula: y m (k+h p )=y m (k+h p )+e(k+h p ); Step B5: Get the output of the prediction function PID algorithm under the refining tower temperature and pressure at time k:
4. The closed-loop control method for automatic switching of high-precision rare gas preparation according to claim 1, Features: The PID calculation part in the prediction function PID algorithm and the generalized open-loop response PID algorithm adopts the fuzzy PID algorithm, and the specific steps are as follows: Step C1: Get the input value r(t) and the system output value y(t), obtain the error e(t) by the difference between the two, and calculate the error change rate ec(t). The calculation formula is as follows: ec(t)=e(t)-e(t-1); Step C2: Fuzzify e(t) and ec(t) respectively and correspond them to the corresponding domains; Step C3: Select the membership functions of e(t) and ec(t), and calculate the membership of e(t) and ec(t) in their corresponding fuzzy subsets respectively; Step C4: construct a fuzzy rule base, and perform fuzzy reasoning on the adjustment parameters Kp, Ki, and Kd of the PID control according to the fuzzy rules to obtain the fuzzy control table of the proportional Kp, integral Ki, and differential Kd of the response; Step C5: After quantizing the fuzzy control table, the control output signals of proportion Kp, integral Ki and differential Kd are obtained.
5. The closed-loop control method for automatic switching of high-precision rare gas preparation according to claim 1, Features: In step S3, the calculation process of the fuzzy neural network is carried out according to the following steps: Step D1: Set the system y(k) and performance index J: y(k)=f[Y(k-1),u(k-1-d),U(k-2-d)]; J=0.5×[r(k)-y(k)] 2 ; Where d is the hysteresis, Y(k-1) = [y(k-1), y(k-2), …, y(kn)], y(k) is the system output at time k, U(k-2-d) = [u(k-2-d), …, u(k-2-dm)], u(k) is the system input at time k, and r(k) is the system expected value at time k; Step D2: Use the gradient descent method to adjust the three parameters of the fuzzy neural network. The formula is as follows: Among them, α is the learning rate, J represents the performance index of the system, q and q+1 represent the qth and q+1th moments respectively. represents the center of the output fuzzy set in the lth control rule at the qth time, Represents the i-th input language variable x in the l-th rule at the q-th time i (k) is the center of the membership function, Represents the i-th input language variable x in the l-th rule at the q-th time i (k) The width of the membership function; Step D3: Let Then the system input u(k) at time k is written as: The derivative is: Three setting parameters are obtained through calculation.
6. The closed-loop control method for automatic switching of high-precision rare gas preparation according to claim 5, Features: The fuzzy neural network calculation process also includes compensation amount determination, which is performed in the following steps: Set up a simplified linear function of the mathematical model of the controlled object: The parameters are obtained by online recursive least squares method. The recursive formula is as follows: Among them, h(k)=[-y(k-1)…-y(kn) u(k-1)…u(km)] T , P(k) and K(k) are all intermediate variables. are the parameters in the mathematical model of the system; According to the mathematical model of the system, the output of the system in the future d+1 time can be obtained, and the output prediction formula is as follows: Among them, a 1 ,a 2 ,…,a n ; b 1 ,b 2 ,…,b m is the parameter of the mathematical model of the system, and d is the lag time; the trust coefficient e is used here to construct the compensation Δμ, where 0≤ε≤1, and the compensation Δμ is as follows: in, Successive recursion to obtain K and ε are the proportionality coefficient and the trust coefficient respectively.
7. The closed-loop control method for automatic switching of high-precision rare gas preparation according to claim 1, Features: The construction process of the fuzzy neural network in step S3 is carried out according to the following steps: Step E1: Introduce BP neural network into the topology structure; Step E2: Initialize the activation function center value, width and output layer weight, and select the learning efficiency and inertia factor; Step E3: Perform fuzzy reasoning on the PID controller and calculate the error, and adjust the excitation function parameters and weights according to the error; Step E4: Repeat step E3 until the learning is completed.
Citation Information
Patent Citations
Networked control method for nonlinear two-time-scale system (NTTSS) with random packet loss
CN104570727A
Boiler combustion system control method based on multivariable PID-PFC
CN105700357A
PID-PFC control method optimized by employing fuzzy network in coking furnace pressure system
CN107065541A
Ship course controller based on fuzzy neural network PID algorithm and control method thereof
CN108536005A
Optimized network control device and method based on PLC remote monitoring terminal
CN108628270A
