
This paper studies a two-layer control strategy for optimal operational control which is prevalent in industrial production. The upper layer determines and adjusts the target set values, while the lower layer makes the loop output track the target value. In the two-layer structure optimal setting control system, the widely used PID controller is used in the bottom layer. Firstly, the parameters of the PID controller are obtained by solving linear matrix inequalities (LMI). Secondly, for industrial processes with nonlinear harmonic disturbances, a disturbance observer is designed to estimate these disturbances. Thirdly, the effects of disturbances or noises are minimized by dynamically adjusting the setting points. This method does not change the structure or parameters of the bottom controller, and thus meets the actual industrial requirements to a certain extent. Finally, in the numerical simulation section, the value of the performance index before set-points adjustment is compared with that after set-points adjustment.
Citation: Liping Yin, Yangyu Zhu, Yangbo Xu, Tao Li. Dynamic optimal operational control for complex systems with nonlinear external loop disturbances[J]. AIMS Mathematics, 2022, 7(9): 16673-16691. doi: 10.3934/math.2022914
[1] | Saim Ahmed, Ahmad Taher Azar, Ibraheem Kasim Ibraheem . Nonlinear system controlled using novel adaptive fixed-time SMC. AIMS Mathematics, 2024, 9(4): 7895-7916. doi: 10.3934/math.2024384 |
[2] | Aziz Belmiloudi . Time-varying delays in electrophysiological wave propagation along cardiac tissue and minimax control problems associated with uncertain bidomain type models. AIMS Mathematics, 2019, 4(3): 928-983. doi: 10.3934/math.2019.3.928 |
[3] | Guijun Xing, Huatao Chen, Zahra S. Aghayan, Jingfei Jiang, Juan L. G. Guirao . Tracking control for a class of fractional order uncertain systems with time-delay based on composite nonlinear feedback control. AIMS Mathematics, 2024, 9(5): 13058-13076. doi: 10.3934/math.2024637 |
[4] | Kamran Sabahi, Chunwei Zhang, Nasreen Kausar, Ardashir Mohammadzadeh, Dragan Pamucar, Amir H. Mosavi . Input-output scaling factors tuning of type-2 fuzzy PID controller using multi-objective optimization technique. AIMS Mathematics, 2023, 8(4): 7917-7932. doi: 10.3934/math.2023399 |
[5] | Mohsen Bakouri, Abdullah Alqarni, Sultan Alanazi, Ahmad Alassaf, Ibrahim AlMohimeed, Mohamed Abdelkader Aboamer, Tareq Alqahtani . Robust dynamic control algorithm for uncertain powered wheelchairs based on sliding neural network approach. AIMS Mathematics, 2023, 8(11): 26821-26839. doi: 10.3934/math.20231373 |
[6] | Honglei Yin, Bo Meng, Zhen Wang . Disturbance observer-based adaptive sliding mode synchronization control for uncertain chaotic systems. AIMS Mathematics, 2023, 8(10): 23655-23673. doi: 10.3934/math.20231203 |
[7] | Shutong Liu, Renming Yang . Adaptive predefined-time robust control for nonlinear time-delay systems with different power Hamiltonian functions. AIMS Mathematics, 2023, 8(12): 28153-28175. doi: 10.3934/math.20231441 |
[8] | Siyue Yao, Jin-E Zhang . Exponential input-to-state stability of nonlinear systems under impulsive disturbance via aperiodic intermittent control. AIMS Mathematics, 2025, 10(5): 10787-10805. doi: 10.3934/math.2025490 |
[9] | Xiao Yu, Yan Hua, Yanrong Lu . Observer-based robust preview tracking control for a class of continuous-time Lipschitz nonlinear systems. AIMS Mathematics, 2024, 9(10): 26741-26764. doi: 10.3934/math.20241301 |
[10] | Hadil Alhazmi, Mohamed Kharrat . Echo state network-based adaptive control for nonstrict-feedback nonlinear systems with input dead-zone and external disturbance. AIMS Mathematics, 2024, 9(8): 20742-20762. doi: 10.3934/math.20241008 |
This paper studies a two-layer control strategy for optimal operational control which is prevalent in industrial production. The upper layer determines and adjusts the target set values, while the lower layer makes the loop output track the target value. In the two-layer structure optimal setting control system, the widely used PID controller is used in the bottom layer. Firstly, the parameters of the PID controller are obtained by solving linear matrix inequalities (LMI). Secondly, for industrial processes with nonlinear harmonic disturbances, a disturbance observer is designed to estimate these disturbances. Thirdly, the effects of disturbances or noises are minimized by dynamically adjusting the setting points. This method does not change the structure or parameters of the bottom controller, and thus meets the actual industrial requirements to a certain extent. Finally, in the numerical simulation section, the value of the performance index before set-points adjustment is compared with that after set-points adjustment.
With the development of modern industry and the increasingly fierce competition in the world market, new requirements are continually put forward to the production control process by industrial departments, which not only require the output of controlled equipment to track its set value to the greatest extent, but also require good control of the operation of the whole industrial equipment to improve the production quality, efficiency and energy consumption within some range. In fact, the optimal operational control of many complex industrial processes consists of two layers-loop control layer and operation control layer [1], which is shown in Figure 1. In Figure 1, the goal of the loop control layer is to control the output of each loop on the production line, and the operation control layer is to dynamically adjust the set value of each loop to optimize the performance index such that the value of the production performance index (such as economic profit or production cost) run with the target range [2,3,4,5,6,7,8].
As we know, in some industrial production processes, once the production line is completed, the underlying controller structure is difficult to change, so it is unrealistic to redesign the bottom layer controller [9,10,11,12,13,14]. Moreover, the range of controller parameters adjustment is usually limited, and thus the optimal operation of the whole production line is not easy to be realized. At present, many literatures have studied other ways to improve the system performance on the premise of using fixed controller, for example, the fixed multi-loop PID control is adopted for the grinding process to reduce the particle size of ore and screen out valuable ore in [5] which shows that selecting appropriate set points also works to improve the whole performance. In [6], the fixed controller operation optimization method is applied for another industrial process called high concentration refining system in which the pulp quality target, economic target and energy consumption target were tracked respectively to set points. This method can significantly reduce the energy consumption for refining system and ensure ideal pulp quality. In order to make the optimization algorithm easier for goethite process, a fixed PID controller is also used in [14] to control the addition amount of zinc oxide and oxygen during operation, and an optimization method is adopted based on set point tracking strategy to convert complex state constraints into additional objectives, which can not only suppress disturbances, but also reduce raw material consumption in the production process.
On the other hand, many practical systems are influenced by input (output) constraints, multiple objective functions or noises [15,16,17]. When the basic site conditions change, the set point of each loop can be hardly found in real time if only based on the experience of engineers and operators, so the actual production can not always meet the technical requirements [18,19,20]. To this purpose, some intelligent algorithms are adopted to optimize performance indices. For example, in [18], in order to improve the sewage purification efficiency, an adaptive multi-objective differential evolution algorithm is used to find the appropriate set point to balance the working performance and operating cost of the machine. In [19], a hybrid intelligent control optimization method is introduced based on multi-objective evaluation in the roasting process of the shaft furnace. The set value of each control loop is adjusted through real-time evaluation and on-line correction, which not only improves the combustion efficiency of the shaft furnace, but also reduces the equipment faults. In [20], a blade pitch parameter tuning optimization method is proposed based on intelligent genetic algorithm which rearranges the mutation rate and crossover points according to the algorithm progress, and more successfully adjusts the blade pitch of wind turbine under high wind speed.
Different from those intelligent optimization algorithms, this paper theoretically analyzes the relationship between the PID controller and upper layer set points, and studies an optimal operational control method with nonlinear harmonic disturbances. By designing a disturbance observer and dynamically resetting the set-points, the outputs of the nonlinear control loops can be tracked to the desired set points without changing the structure or parameters of the controllers.
The remainder of the paper is organized as follows: The optimal operational control process and the system model will be described in Section 2. The bottom layer controller will be designed in Section 3 and the disturbance observer will be designed in Section 4. In Section 5, the set points will be dynamically adjusted. The effectiveness of the method is verified by simulation in Section 6, and the conclusion is drawn in Section 7.
Figure 1 is an industrial operation control process in which the operation indicators ξ∗ can be obtained through the planning and scheduling of performance indicators J. During operation, engineers will determine the set point r∗j(j=1,2,⋯,n) for each control loop according to experience, and the controllers will produce required inputs uj(t)(j=1,2,⋯,n) to ensure that its outputs yj(t)(j=1,2,⋯,n) track the set point r∗j such that the overall performance can be optimized. In general industrial processes, the performance index J represents profit and energy cost, which can be usually written as:
J=∫TfT0[qm(τ)(pu(τ)−pl(τ))−C0(τ)]dτ, | (2.1) |
where qm(τ) is the product quantity in time interval [T0,Tf], pu(τ) is the unit price function of the product at time τ, pl(τ) is the unit price of raw materials, and C0(τ) is the energy consumption.
In Figure 1, di(t) is the disturbance in each loop. When there exist disturbances, the dynamic set point can be re selected and applied to the operation control layer as follows [2]:
˙r(t)=h(r(t),y(t),ˆd(t),r∗(t)), | (2.2) |
where h(⋅) is a unknown function with respect to the design of operation index r∗(t), output y(t) and disturbance estimation ˆd(t), and r(t) is the reselected set point.
In the two-layer system, it is assumed that the dynamic model of each bottom loop can be expressed as follows:
Loopi:{˙xi(t)=Aixi(t)+Biui(t)+fi(xi(t))+di(t)yi(t)=Cixi(t)i=1,2,⋯,n, | (2.3) |
where xi∈Rni×1 is the state vector of each subsystem, yi∈Rmi×1 is the system output, ui∈Rpi×1 is the control input, Ai∈Rni×ni, Bi∈Rni×pi and Ci∈Rmi×ni are the system matrices of the subsystem respectively, while di∈Rni×1 is the disturbance, and fi is the nonlinear function which meets the following assumptions:
Assumption 2.1: For any x1(t) and x2(t), the nonlinear function fi(x(t)) satisfies
‖fi(x1(t))−fi(x2(t))‖≤‖Ui(x1(t)−x2(t))‖, | (2.4) |
where Ui are given Lipschitz constants and ||⋅|| stands for Euclidean norm for vectors.
The composite system for (2.3) can be written as:
{˙x(t)=Ax(t)+Bu(t)+f(x(t))+d(t),y(t)=Cx(t), | (2.5) |
where the composite vectors x, y, u are written as x=[xT1,⋯xTn]T, y=[yT1,⋯yTn]T, u=[uT1,⋯uTn]T; A, B and C are system matrices which are denoted as A=diag{A1,⋯,An}, B=diag{B1,⋯,Bn} and C=diag{C1,⋯,Cn}, d(t)=diag{d1,⋯,dn}.
Denote the tracking error as ε(t), then it can be expressed as follows:
ε(t)=r∗(t)−y(t). |
The residual e(t) between the setpoint r(t) and the actual control output y(t) can be defined as
e(t)=r(t)−y(t). |
If the disturbances are not considered, the relationship between the tracking error ε(t) and residual e(t) is as follows:
e(t)=ε(t)=r∗(t)−y(t). |
Due to technical and economic constraints, PID controllers are still widely used in many industrial processes [21,22,23,24,25,26,27,28,29]. Because PID controllers are simple in structure and easy to operate, in order to make the algorithm in this paper more adaptive, this paper still uses PID control in the loop control layer shown in Figure 1, and tries to use LMI algorithms to determine the controller parameters. The PID controller of the loop control layer can be expressed as follows:
u(t)=KPe(t)+KI∫t0e(τ)dτ+KD˙e(t). | (3.1) |
Substitute the PID control law (3.1) into (2.5), we get
˙x(t)=Ax(t)+B(KPe(t)+KI∫t0e(τ)dτ+KD˙e(t))+f(x(t)). |
In the ideal case (without considering disturbance), e(t)=ε(t)=r∗(t)−y(t), in order to make the algorithm more explicit, we assume that the state is measurable and let C=I, then
˙ε(t)=˙r∗(t)−˙x(t)=˙r∗(t)−A(r∗−ε)−f(x(t))−BKPe(t)−BKI∫t0e(τ)dτ−BKD˙e(t)=(A−BKP)ε(t)−BKI∫t0e(τ)dτ−BKD˙e(t)−f(x(t))+˙r∗(t)−Ar∗(t). | (3.2) |
Let s(t)=∫t0ε(τ)dτ, then
(˙ε(t)˙s(t))=(Δ−1Π−Δ−1BKII0)(ε(t)s(t))+Δ−1(˙r∗(t)−Ar∗0)+(Δ−1f(r∗−ε)0), | (3.3) |
where Δ=I+BKD, Π=A−BKP, and KP, KI, KD represent the proportional coefficient, integral coefficient and differential coefficient of PID controller, respectively.
Theorem 3.1. If there exist P>0, KP∈Rm×m, KI∈Rm×m and KD∈Rm×m such that the following two matrix inequalities hold
P=(P1P2P2TP3)>0, | (3.4) |
(Φψ1ψ2ψ3ψT1−I00ψT20−I0ψT300−I)<0, | (3.5) |
where
Φ=(Φ11Φ12ΦT12Φ22), | (3.6) |
ψ1=λP(Δ−10),ψ2=√2λ(U1T0),ψ3=P(Δ−1A0), |
Φ11=sym(P1Δ−1Π+P2),Φ12=−P1Δ−1BKI+ΠTΔ−TP2+P3,Φ22=sym(−PT2Δ−1BKI), |
then the composite system (3.3) is asymptotically stable.
Proof. Construct the following Lyapunov function:
V1=[εT(t)sT(t)][P1P2PT2P3][ε(t)s(t)]+1λ2∫t0‖U1(r∗−ε(τ))‖2−‖f(r∗−ε(τ))‖2dτ. | (3.7) |
Taking the derivative of V1(t) with respect to t yields
˙V1=[εT(t)sT(t)]Φ[ε(t)s(t)]+2[εT(t)sT(t)]P[Δ−10](˙r∗(t)−Ar∗)+2[εT(t)sT(t)]P[−Δ−10]f(r∗−ε(t))+1λ2[‖U1(r∗−ε(t))‖2−‖f(r∗−ε(t))‖2]≤[εT(t)sT(t)][Φ+λP2[Δ−1Δ−T000]PT+[2λ2U1TU1000]+P[Δ−1AATΔ−T000]PT][ε(t)s(t)]+r∗Tr∗+1λ2˙r∗TUT1U1˙r∗≤−σ0‖(εs)‖2+λmax, |
where λmax is the maximum eigenvalue of matrix r∗Tr∗+1λ2˙r∗TUT1U1˙r∗. Therefore, when ‖(εs)‖≥σ−120λmax12, ˙V1<0, and the state vector (εTsT) satisfies
‖(ε(t)s(t))‖≤max{σ−120λmax12,‖(ε(0)s(0))‖}. |
In the ideal case (without considering the disturbances), let θ(t) be the trajectory of the composite loop system (3.3), then
limt→∞θ(t)=θ0, |
where θ0 is a constant vector, and
limt→∞˙θ(t)=0, |
because of ˙s(t)=ε(t), we have
limt→∞ε(t)=0, |
the proof is completed.
According to the Schur complement lemma [30], it can be easily proven that (3.5) in Theorem 3.1
[˜Φ11˜Φ12˜ΦT12˜Φ22]<0. | (3.8) |
where
˜Φ11=Φ11+λ2P1Δ−1Δ−TP1+2λ2U1TU1+P1Δ−1AATΔ−TP1,˜Φ12=Φ12+λ2P1Δ−1Δ−TP2+P1Δ−1AATΔ−TP2,˜Φ22=Φ22+λ2P2Δ−1Δ−TP2+P2Δ−1AATΔ−TP2, |
according to (3.8), ˜Φ22<0 is a necessary condition for (3.5).
Since (3.5) is not a LMI in a strict sense, it is impossible to directly get the required parameters KP, KI and KD, so it is necessary to convert (3.5) in Theorem 3.1 into a LMI, and we can further obtain the following theorem:
Theorem 3.2. If there exist two matrices PI and PD such that
sym(−BPI−BPDBT)+λ2I+AAT<0, | (3.9) |
then PI and PD can be obtained by the (3.9), and the PID controller parameter KI can be obtained by solving
KI=PIP2, | (3.10) |
while KD can be obtained by solving
PIKTD=PD, | (3.11) |
Proof. By pre-multiplying ΔP−T2 and post-multiplying P−12ΔT, ˜Φ22<0 in Eq (3.8) is equivalent to
sym(−BKIP−12ΔT)+λ2I+AAT<0, | (3.12) |
denote PI and PD as:
PI=KIP−12,PD=KIP−12KD. | (3.13) |
It can be verified that (3.12) holds only when (3.9) holds. Obviously, (3.9) is a LMI related to PI and PD, so a set of P2, KI matrices can be calculated and Δ can be guaranteed to be invertible. For any α>0, P2=12αI can be selected.
Next, we will continue to calculate KP.
Theorem 3.3. If there exist matrices Q1>0, QP and parameters δ>0, λ>0 such that
Ψ=(ψ11ψ12+δ2IQ1Δ−1Δ−1AQ1U1TψT12+δ2Iψ220000QT10−α−1I000Δ−T00−λ−1I00ATΔ−T000I0U1QT10000I)<0, | (3.14) |
where
ψ11=sym(Δ−1AQT1−Δ−1BQTP),ψ12=−Δ−1BKI+α(Q1AT−QPBT)Δ−T,ψ22=sym(−PT2Δ−1BKI)+λ2P2Δ−1Δ−TP2+P2Δ−1AATΔ−TP2, |
with the constraint
α2λmax(Q1)<δ, | (3.15) |
then the parameter KP of PID controller (3.1) can be calculated by KP=QTPQ−T1.
Proof. Let Q1=P−11, Ω1=diag{P−11,I}, substitute the obtained P2, KI and KD into (3.8), pre-multiply Ω1 and post-multiply Ω1T, we have
(˜ψ11˜ψ12˜ψT12˜ψ22)<0, | (3.16) |
where
˜ψ11=sym(Δ−1AQT1−Δ−1BQTP)+Q1(P2+PT2)QT1+λ2Δ−1Δ−T+Δ−1AATΔ−T+2λ2Q1U1TU1Q1T,˜ψ12=−Δ−1BKI+(Q1AT−QPBT)Δ−TP2+Q1P3+λ2Δ−1Δ−TP2+Δ−1AATΔ−TP2,˜ψ22=sym(−PT2Δ−1BKI)+λ2P2Δ−1Δ−TP2+P2Δ−1AATΔ−TP2. |
According to the Schur complement lemma [30], (3.16) is equivalent to
Ψ=(ψ11ψ12+Q1P3Q1Δ−1Δ−1AQ1U1TψT12+PT3QT1ψ220000QT10−α−1I000Δ−T00−λ−1I00ATΔ−T000I0U1QT10000I)<0. | (3.17) |
Let Q3=Q1P3=δ2I, then when P3=δ2Q−11, Inequality (3.17) is equivalent to Inequality (3.14). According to the Schur complement lemma [30], if inequality
P3−14α2P−11>0, |
or
(δ2Q−21−14α2)Q1>0, |
holds and satisfies (3.15), it can be guaranteed that (3.4) holds.
In practical engineering applications, there often exist various disturbances in the system. In this paper, it is assumed that the external interference d(t) is
{˙ξ(t)=Aωξ(t)+Fg(ξ(t)),d(t)=Cωξ(t), | (4.1) |
where ξ(t) is the state vector of the exogenous system, ξ(t)∈Rm×1; Aω∈Rm×m, Cω∈Rm×m, F∈Rm×m are known constant matrices; g(ξ(t)) is a known Lipschitz continuous nonlinear function, which satisfies the following assumption:
Assumption 4.1. For any ξ1(t) and ξ2(t), the nonlinear function g(ξ(t)) satisfies
‖g(ξ1(t))−g(ξ2(t))‖≤‖ˉU(ξ1(t)−ξ2(t))‖, |
where ˉU is given Lipschitz constant and ||⋅|| stands for Euclidean norm for vectors.
Compared with those linear external disturbances in [31,32], the external system in this paper can not only describe linear signals such as constant load, harmonic disturbances, but also nonlinear signals in [33,34].
For the external disturbance System (4.1), this paper constructs the following disturbance observer:
ˆd(t)=Cωˆξ(t),ˆξ(t)=v(t)+Lx(t),˙v(t)=(Aω−LCω)ˆξ(t)−L(Ax(t)+Bu(t)+f(x(t)))+Fg(ˆξ(t)), | (4.2) |
where ˆξ(t) is the estimated value of ξ(t), v(t) is the auxiliary vector, and L is the nonlinear gain function of the disturbance observer.
Define the estimation error as eξ(t)=ξ(t)−ˆξ(t), then the derivative of eξ(t) can be written as
˙eξ(t)=˙ξ(t)−˙ˆξ(t)=Aωξ(t)−(Aω−LCω)ˆξ(t)−LCω+F(g(ξ(t))−g(ˆξ(t)))=(Aω−LCω)(ξ(t)−ˆξ(t))+F(g(ξ(t))−g(ˆξ(t)))=(Aω−LCω)eξ(t)+F(g(ξ(t))−g(ˆξ(t))). | (4.3) |
In the case of nonlinear harmonic disturbances (4.1), if we do nothing, the tracking error ε(t) will increase sharply compared with the tracking error in (3.2). Next, we will analyze the tracking error ε(t) and try to minimize it in the upper layer. It is noted that neither the structure nor the parameters of the bottom layer controllers will be changed.
In fact, if we substitute the PID controller (3.1) into (2.5), we will obtain
˙ε(t)=˙r∗(t)−˙x(t)=Δ−1Πε(t)−Δ−1BKIs(t)−Δ−1(BKPR(t)+BKI∫t0R(t)dτ+BKD˙R(t))+Δ−1˙r∗(t)−Δ−1Cωξ(t)−Δ−1Ar∗(t)−Δ−1f(x(t)), | (4.4) |
where R(t)=r(t)−r∗(t). Combined with (4.3) and (4.4), the following composite system can be obtained:
{˙ε(t)=Δ−1Πε(t)−Δ−1BKIs(t)−Δ−1(BKPR(t)+BKI∫t0R(τ)dτ+BKD˙R(t))+Δ−1˙r∗(t)−Δ−1Cωξ(t)−Δ−1Ar∗(t)−Δ−1f(x(t)),˙s(t)=ε(t),˙eξ(t)=(Aω−LCω)eξ(t)+F(g(ξ(t))−g(ˆξ(t))). | (4.5) |
In Section 4, for those nonlinear harmonic disturbances, we have designed a disturbance observer to estimate them. Next, to ensure the loop outputs yi(t) can still track the original target r∗ under these disturbances, we will try to dynamically adjust the set-points in the upper layer based on the values of the estimated disturbances so that the loop output can still track the original target value under the action of the original controller.
If we adjust the set points as follows:
{E˙z=E1z+E2,r(t)=z2(t)+r∗, | (5.1) |
where
E=[I000I0000],E1=[0I000IΔ−1BKIΔ−1BKPΔ−1BKD],z(t)=[∫t0R(τ)dτR(t)˙R(t)], |
E2=[00Δ−1Ar∗−Δ−1˙r∗(t)+Δ−1f(x(t))+Δ−1Cωˆξ(t)+K1ε(t)+K2s(t)]. |
Then we can obtain
Δ−1BKPR(t)+Δ−1BKI∫t0R(τ)dτ+Δ−1BKD˙R(t)=−Δ−1Ar∗+Δ−1˙r∗(t)−Δ−1f(x(t))−Δ−1Cωˆξ(t)−K1ε(t)−K2s(t). | (5.2) |
Substituting (5.2) into (4.5), the composite system composing of ε(t), s(t) and eξ(t) can be expressed as follows:
[˙ε(t)˙s(t)˙eξ(t)]=[Δ−1Π+K1K2−Δ−1BKI−Δ−1CωI0000Aω−LCω][ε(t)s(t)eξ(t)]+[00F](g(ξ(t))−g(ˆξ(t))). | (5.3) |
Let G11=(Δ−1Π−Δ−1BKII0),G22=(−Δ−1Cω0),G12=(I0),ˉF=(00F), then the composite System (5.3) can be further written as
˙ˉx(t)=Gˉx(t)+ˉF(g(ξ(t))−g(ˆξ(t))), |
where ˉx(t)=(ε(t)Ts(t)Teξ(t)T)T, G=(G11+G12ˉKG220Aω−LCω), ˉK=[K1K2].
Theorem 5.1. For a given τ>0, if there exist Γ1∈Rm×m, X∈Rl×m and Y∈Rm×m such that the following LMI holds:
(Ξ1Ξ200∗Ξ3FˉUT∗∗−τI0∗∗∗−τ−1I)<0, | (5.4) |
where
Ξ1=sym(G11Γ1+G12X),Ξ2=G22,Ξ3=sym(Aω−YCω), |
then the gains K1, K2 in (5.1) and the disturbance observer gain L in (4.2) can be obtained by:
[K1K2]=XΓ−11L=Y, | (5.5) |
and the composite System (4.5) is exponentially stable.
Proof. In order to prove the stability of the error composite System (5.3), we construct the following Lyapunov function: V2=ˉxT(t)ˉPˉx(t) where ˉP=[ˉP100I]>0. Taking the derivative of V2(t) with respect to t yields
˙V2=2ˉxTˉP˙ˉx=2ˉxTˉP(Gˉx+ˉF(g(ξ(t))−g(ˆξ(t))))=2ˉxTˉPGˉx+2ˉxTˉPˉF(g(ξ(t))−g(ˆξ(t)))=W(t)TΛ1W(t)+τ‖g(ξ(t))−g(ˆξ(t))‖2, |
where W(t)=(ˉx(t)g(ξ(t))−g(ˆξ(t))), Λ1=(sym(ˉPG)ˉPTˉF∗−τI).
According to Assumption 4.1,
‖g(ξ(t))−g(ˆξ(t))‖≤‖ˉU(ξ(t)−ˆξ(t))‖=‖ˉUeξ(t)‖≤‖ˉUHˉx(t)‖, |
then ˙V2≤WT(t)Λ2W(t) where Λ2=(sym(ˉPG)+τHTˉUTˉUHˉPTˉF∗−τI), H=(00I).
Let ˉP1=Γ−11. By using (5.5) and then pre- and post-multiplying (5.4) with diag {ˉP1,I,I,I}, we have
(Ψ1Ψ200∗Ψ3FˉUT∗∗−τI0∗∗∗−τ−1I)<0, | (5.6) |
where
Ψ1=sym[ˉP1(G11+G12ˉK)],Ψ2=ˉP1G22,Ψ3=sym(Aω−LCω). |
Substitute ˙ˉx(t)=Gˉx(t)+ˉF(g(ξ(t))−g(ˆξ(t))) into (5.6), we get
(sym(ˉPG)ˉPTˉFHTˉUTˉFTˉP−τI0ˉUH0−τ−1I)<0. |
According to the Schur complement lemma [30], it is easy to obtain
Λ2<0, |
denote χ1Δ=|λmax(Λ2)|>0, χ2Δ=λmax(ˉP) and get Λ2≤−χ1I, then
˙V2≤W(t)TΛ2W(t)≤−χ1‖ˉx(t)‖2≤−χ1χ−12V, | (5.7) |
from (5.7), we get
V2(ˉx(t))≤V2(ˉx(0))e−χ1χ2−1t, |
which means that System (5.3) is exponentially stable, i.e., the composite System (4.5) is exponentially stable in the presence of disturbances (4.1).
In summary, the algorithm in this paper can be listed as follows:
Step 1. Solve the LMI in (3.9) to get PI and PD;
Step 2. Let P2=(1/2)αI, get KI by KI=PIP2, and get KD through PIKTD=PD;
Step 3. Solve LMIs (3.14) and (3.15) to obtain Q1 and QP;
Step 4. Obtain KP from KP=QTPQ−T1 and construct controller (3.1);
Step 5. Solve Eq (5.4) to obtain X, Γ1 and Y, and get L and ˉK from L=Y and ˉK=XΓ−11. Construct a disturbance observer (4.2) and adjust the set point value according to (5.1).
Most zinc smelting enterprises use goethite method to precipitate iron from zinc sulfate solution. After the introduced indium post deposition liquid is passed into the reactor for chemical reaction, the post deposition liquid with ferrous iron will be obtained. The goethite method for ion concentration control is shown in Figure 2. In [14], the process of iron deposition consists of four ferric precipitation reactors. Different concentrations of oxygen need to be fed to each branch tube because the low ferric ion concentration declines during the reaction. However, those small part of oxygen retention existing in the pipe reacting with heat might affect the actual oxygen concentration. In addition, the air pressure and temperature of the intake pipe generate periodic fluctuations, and mutual friction and thermal fluctuation are also generated between the components of the pipe. According to the three chemical reaction processes in the iron precipitation process of goethite method, the concentration of oxygen which is fed into each pipe can be regarded as the system state, the input current as the control input, the concentration of ferrous iron after the reaction as the system output, and the factors such as thermal fluctuation, mutual friction between pipeline components and periodic fluctuation of air pressure and temperature can be regarded as uncertain perturbation. Then the chemical reaction rate of ferrous iron can be obtained in combination with the chemical reaction kinetics, the corresponding mass balance equation can be established through the principle of material conservation, and the mechanism model of the reactor in the iron precipitation process can be finally determined as [35]:
{˙x(t)=Ax(t)+Bu(t)+f(x(t),t)+d(t),y(t)=Cx(t), |
where A=[−11001100001−100−2−1], B=[1000010000100001], f=[−0.3sin(x1)00.3cos(x3)0], d(t)=[d1d2d3d4]; x is the oxygen concentration of each branch pipe; u is the control input current under the action of external disturbance d(t); f is the concentration of residual oxygen in the pipeline after chemical reaction, system output y is the concentration of ferrous iron after the reaction.
The schematic diagram of ion concentration control for goethite process can be illustrated as follows:
The initial state of the state vector is set to be [x1(0)x2(0)x3(0)x4(0)]=[0000]. Also we set the target value as [r∗1(t)r∗2(t)r∗3(t)r∗4(t)]=[107.453.717.93.6] [14].
Define the system performance index as
J=(y1(t)−r∗1(t))2+(y2(t)−r∗2(t))2+(y3(t)−r∗3(t))2+(y4(t)−r∗4(t))2, |
where r∗(t) is the set target value, y(t) is the system output ion concentration, and J is the sum of squares of tracking error.
According to Theorems 3.2 and 3.3, the controller parameters can be obtained by KI=[1.010500001.010500001.010500001.0105], KD=[1000010000100001], KP=[7.79680.9893001.01067.972900008.8254−1.998400−3.563417.0733].
For the harmonic disturbances d(t) as in (4.1), the parameters are selected as Aω=(0600−6000000600−60), Cω=(0.100000.100000.100000.1), F=(1000010000100001), g=[0.5sin(ξ1)−0.5cos(ξ2)0.5sin(ξ3)−0.5cos(ξ4)].
In the presence of exogenous disturbances, the comparison results of the system output ion concentration, the error between the tracking target and the actual ion concentration and the system performance indices are shown in Figures 3–5, respectively.
In the ideal conditions, the tracking error ε finally tends to zero. However, when there exist exogenous disturbances (4.1) in the system, if the set value is not adjusted, the ion concentration y output from the four reactors will fluctuate near the ideal value. As shown by the blue line in Figure 3, the fluctuation ranges of the four reactors are as follows: −12.9∼12.9(mmol/L), −15.0∼15.0(mmol/L), −15.05∼15.05(mmol/L), −11.95∼11.95(mmol/L). At the same time, the system tracking error ε of the four reactors fluctuates near zero. As shown by the blue line in Figure 4, the fluctuation ranges of tracking error ε are as follows: −12.8∼12.8(mmol/L), −15.1∼15.1(mmol/L), −15.7∼15.7(mmol/L), −11.38∼11.38(mmol/L). The blue line in Figure 5 is the performance index J of the system in the presence of exogenous disturbances (4.1). When there exist exogenous disturbances and the set value is not adjusted, the curve fluctuates frequently up and down.
It can be seen that in the presence of exogenous disturbances (4.1), if the controller is not changed and no adjustment is made in the upper layer, the system performance will deteriorate seriously.
According to Theorem 5.1, we can get
L=[15.1182000015.1182000015.1182000015.1182], K1=[−1.238×1030.178×103000.178×103−1.238×1030000−3.182×1030.128×10300−0.405×103−7.316×103], K2=[−3.201×1051.077×103001.077×103−3.201×1050000−9.679×1050.679×103000.679×103−9.679×105]. |
Figure 6 is the set point value adjustment. After the set value is readjusted, the ion concentration output y of the system quickly reaches the set point. At the same time, the tracking error ε of the system can not only tend to zero, but also respond quickly. The performance index J of the system gradually improves and the system reaches a stable state, as shown by the red line in Figures 3–5.
This paper mainly studies a two-level optimal setting control for nonlinear systems with exogenous disturbances. In order to make the disturbance model more suitable for engineering practice, the additional disturbances caused by external system nonlinearity is considered. Firstly, the PID controller is used in the bottom loop, and the PID controller parameters are obtained according to the Lyapunov theory and the LMI method. Secondly, for nonlinear harmonic disturbances, a disturbance observer is designed in the upper layer to estimate these nonlinear disturbances. Finally, the set point value is dynamically adjusted to minimize the impact caused by disturbances or noises. This method can track the target value well without changing the structure or parameters of those bottom layer controllers of the whole system.
In this paper, we only consider nonlinear loop systems satisfying Lipschitz conditions, and the external disturbances are of harmonic type. In fact, the system might be complex, there probably exist composite disturbances, and the practical systems will generally have requirements on inputs and outputs. So, we believe the work includes data-based modeling, multi-objectives, composite disturbances or input/output constraints will be quite promising.
The work reported was jointly supported by the National Natural Science Foundation of China under grants 61573190, 61973168 and Project 333 of Jiangsu Province under Grant BRA2020067.
The authors declare that there is no conflict of interests.
[1] | A. Wang, P. Zhou, H. Wang, Performance analysis for operational optimal control for complex industrial processes under small loop control errors, In: Proceedings of the 2014 international conference on advanced mechatronic systems, 2014. http://doi.org/10.1109/ICAMechS.2014.6911643 |
[2] |
T. Chai, S. J. Qin, H. Wang, Optimal operational control for complex industrial processes, Annu. Rev. Control, 38 (2014), 81–92. http://doi.org/10.1016/j.arcontrol.2014.03.005 doi: 10.1016/j.arcontrol.2014.03.005
![]() |
[3] |
L. Yin, H. Wang, X. Yan, H. Zhang, Disturbance observer-based dynamic optimal setting control, IET Control Theory A., 12 (2018), 2423–1432. http://doi.org/10.1049/iet-cta.2018.5013 doi: 10.1049/iet-cta.2018.5013
![]() |
[4] |
L. Yin, H. Wang, L. Guo, H. Zhang, Data-driven pareto-de-based intelligent optimal operational control for stochastic processes, IEEE T. Syst. Man Cy. Syst., 51 (2021), 4443–4452. http://doi.org/10.1109/TSMC.2019.2936452 doi: 10.1109/TSMC.2019.2936452
![]() |
[5] |
W. Dai, G. Huang, F. Chu, T. Chai, Configurable platform for optimal-setting control of grinding processes, IEEE Access, 5 (2017), 26722–26733. http://doi.org/10.1109/ACCESS.2017.2774001 doi: 10.1109/ACCESS.2017.2774001
![]() |
[6] |
M. Li, P. Zhou, H. Wang, T. Chai, Nonlinear multiobjective mpc-based optimal operation of a high consistency refining system in papermaking, IEEE T. Syst. Man Cy. Syst., 50 (2017), 1208–1215. http://doi.org/10.1109/TSMC.2017.2748722 doi: 10.1109/TSMC.2017.2748722
![]() |
[7] |
P. Zhou, T. Chai, H. Wang, Intelligent optimal-setting control for grinding circuits of mineral processing process, IEEE T. Autom. Sci. Eng., 6 (2009), 730–743. http://doi.org/10.1109/TASE.2008.2011562 doi: 10.1109/TASE.2008.2011562
![]() |
[8] |
Y. Jiang, J. Fan, T. Chai, J. Li, F. L. Lewis, Data-driven flotation industrial process operational optimal control based on reinforcement learning, IEEE T. Ind. Inform., 14 (2018), 1974–1989. http://doi.org/10.1109/TII.2017.2761852 doi: 10.1109/TII.2017.2761852
![]() |
[9] |
Y. Zhou, Q. Zhang, H. Wang, P. Zhou, T. Chai, Ekf-based enhanced performance controller design for nonlinear stochastic systems, IEEE T. Automat. Contr., 63 (2018), 1155–1162. http://doi.org/10.1109/TAC.2017.2742661 doi: 10.1109/TAC.2017.2742661
![]() |
[10] |
L. Dong, X. Wei, H. Zhang, Anti-disturbance control based on nonlinear disturbance observer for a class of stochastic systems, T. I. Meas. Control, 41 (2019), 1665–1675. http://doi.org/10.1177/0142331218787608 doi: 10.1177/0142331218787608
![]() |
[11] |
S. Xie, Y. Xie, F. Li, C. Yang, W. Gui, Optimal setting and control for iron removal process based on adaptive neural network soft-sensor, IEEE T. Syst. Man Cy. Syst., 50 (2020), 2408–2420. http://doi.org/10.1109/TSMC.2018.2815580 doi: 10.1109/TSMC.2018.2815580
![]() |
[12] | L. Guo, H. Wang, Stochastic distribution control system design: A convex optimization approach, London: Springer, 2010. http://doi.org/10.1007/978-1-84996-030-4 |
[13] |
Y. Liu, K. Fan, Q. Ouyang, Intelligent traction control method based on model predictive fuzzy pid control and online optimization for permanent magnetic maglev trains, IEEE Access, 9 (2021), 29032–29046. http://doi.org/10.1109/ACCESS.2021.3059443 doi: 10.1109/ACCESS.2021.3059443
![]() |
[14] |
X. Zhou, J. Zhou, C. Yang, W. Gui, Set-point tracking and multi-objective optimization-based pid control for the goethite process, IEEE access, 6 (2018), 36683–36698. http://doi.org/10.1109/ACCESS.2018.2847641 doi: 10.1109/ACCESS.2018.2847641
![]() |
[15] |
C. Liu, Z. Gong, K. L. Teo, J. Sun, L. Caccetta, Robust multi-objective optimal switching control arising in 1, 3-propanediol microbial fed-batch process, Nonlinear Anal. Hybri., 25 (2017), 1–20. http://doi.org/10.1016/j.nahs.2017.01.006 doi: 10.1016/j.nahs.2017.01.006
![]() |
[16] |
C. Liu, Z. Gong, H. W. J. Lee, K. L. Teo, Robust bi-objective optimal control of 1, 3-propanediol microbial batch production process, J. Process Contr., 78 (2019), 170–182. http://doi.org/10.1016/j.jprocont.2018.10.001 doi: 10.1016/j.jprocont.2018.10.001
![]() |
[17] |
B. Li, Y. Wang, K. Zhang, G. R. Duan, Constrained feedback control for spacecraft reorientation with an optimal gain, IEEE T. Aero. Elec. Sys., 57 (2021), 3916–3926. http://doi.org/10.1109/TAES.2021.3082696 doi: 10.1109/TAES.2021.3082696
![]() |
[18] |
J. F. Qiao, Y. Hou, H. G. Han, Optimal control for wastewater treatment process based on an adaptive multi-objective differential evolution algorithm, Neural Comput. Applic., 31 (2019), 2537–2550. http://doi.org/10.1007/s00521-017-3212-4 doi: 10.1007/s00521-017-3212-4
![]() |
[19] |
A. Yan, T. Chai, W. Yu, Z. Xu, Multi-objective evaluation-based hybrid intelligent control optimization for shaft furnace roasting process, Control Eng. Pract., 20 (2012), 857–868. http://doi.org/10.1016/j.conengprac.2012.05.001 doi: 10.1016/j.conengprac.2012.05.001
![]() |
[20] |
Z. Civelek, E. Cam, M. Luy, H. Mamur, Proportional-integral-derivative parameter optimisation of blade pitch controller in wind turbines by a new intelligent genetic algorithm, IET Renew. Power Gen., 10 (2016), 1220–1228. http://doi.org/10.1049/iet-rpg.2016.0029 doi: 10.1049/iet-rpg.2016.0029
![]() |
[21] | X. Cong, L. Guo, PID control for a class of nonlinear uncertain stochastic systems, In: 2017 IEEE 56th annual conference on decision and control (CDC), 2017. http://doi.org/10.1109/CDC.2017.8263728 |
[22] | K. Guo, J. Jia, X. Yu, L. Guo, Dual-disturbance observers-based control of uav subject to internal and external disturbances, In: 2019 Chinese automation congress (CAC), 2019. http://doi.org/10.1109/CAC48633.2019.8997330 |
[23] |
C. Zhao, L. Guo, Control of nonlinear uncertain systems by extended pid, IEEE T. Automat. Contr., 66 (2021), 3840–3847. http://doi.org/10.1109/TAC.2020.3030876 doi: 10.1109/TAC.2020.3030876
![]() |
[24] | C. Zhao, L. Guo, PID control for a class of non-affine uncertain systems, In: 2018 37th Chinese control conference (CCC), 2018. http://doi.org/10.23919/ChiCC.2018.8483587 |
[25] | S. Yuan, C. Zhao, L. Guo, Decentralized PID control of multi-agent systems with nonlinear uncertain dynamics, In: 2017 36th Chinese control conference (CCC), 2017. http://doi.org/10.23919/ChiCC.2017.8028765 |
[26] | P. Thampi, G. Raghavendra, Intelligent model for automating PID controller tuning for industrial water level control system, In: 2021 International conference on design innovations for 3Cs compute communicate control (ICDI3C), 2021. http://doi.org/10.1109/ICDI3C53598.2021.00039 |
[27] |
H. Tsukamoto, S. J. Chung, Robust controller design for stochastic nonlinear systems via convex optimization, IEEE T. Automat. Contr., 66 (2021), 4731–4746. http://doi.org/10.1109/TAC.2020.3038402 doi: 10.1109/TAC.2020.3038402
![]() |
[28] |
L. Guo, H. Wang, PID controller design for output pdfs of stochastic systems using linear matrix inequalities, IEEE T. Syst. Man Cy. B, 35 (2005), 65–71. http://doi.org/10.1109/TSMCB.2004.839906 doi: 10.1109/TSMCB.2004.839906
![]() |
[29] |
C. Zhao, L. Guo, PID controller design for second order nonlinear uncertain systems, Sci. China Inf. Sci., 60 (2017), 022201. http://doi.org/10.1007/s11432-016-0879-3 doi: 10.1007/s11432-016-0879-3
![]() |
[30] | P. Gahinet, A. Nemirovskii, A. J. Laub, M. Chilali, The LMI control toolbox, In: Proceedings of 1994 33rd IEEE conference on decision and control, 1994. http://doi.org/10.1109/CDC.1994.411440 |
[31] |
X. Wei, L. Dong, H. Zhang, X. Hu, J. Han, Adaptive disturbance observer-based control for stochastic systems with multiple heterogeneous disturbances, Int. J. Robust Nonlin., 29 (2019), 5533–5549. http://doi.org/10.1002/rnc.4683 doi: 10.1002/rnc.4683
![]() |
[32] |
X. Wei, S. Sun, Elegant anti-disturbance control for discrete-time stochastic systems with nonlinearity and multiple disturbances, Int. J. Control, 91 (2018), 706–714. http://doi.org/10.1080/00207179.2017.1291996 doi: 10.1080/00207179.2017.1291996
![]() |
[33] |
Z. Ding, Output regulation of uncertain nonlinear systems with nonlinear exosystems, IEEE T. Automat. Contr., 51 (2006), 498–503. http://doi.org/10.1109/TAC.2005.864199 doi: 10.1109/TAC.2005.864199
![]() |
[34] |
M. Lu, J. Huang, A class of nonlinear internal models for global robust output regulation problem, Int. J. Robust Nonlin., 25 (2015), 1831–1843. http://doi.org/10.1002/rnc.3180 doi: 10.1002/rnc.3180
![]() |
[35] |
Y. Xie, S. Xie, Y. Li, C. Yang, W. Gui, Dynamic modeling and optimal control of goethite process based on the rate-controlling step, Control Eng. Pract., 58 (2017), 54–65. http://doi.org/10.1016/j.conengprac.2016.10.001 doi: 10.1016/j.conengprac.2016.10.001
![]() |
1. | Hae Yeon Park, Jung Hoon Kim, Model-free control approach to uncertain Euler-Lagrange equations with a Lyapunov-based $ L_\infty $-gain analysis, 2023, 8, 2473-6988, 17666, 10.3934/math.2023902 |