Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42

Contents lists available at ScienceDirect

Journal of the Taiwan Institute of Chemical Engineers

journal homepage: www.elsevier.com/locate/jtice

Full cycle dynamic optimisation maintaining the operation margin of

acetylene hydrogenation fixed-bed reactor
Fu-Ming Xie, Feng Xu*, Zhi-Shan Liang, Xiong-Lin Luo
Department of Automation, China University of Petroleum Beijing, 102249, China

A R T I C L E I N F O A B S T R A C T

Article History: In an acetylene hydrogenation reactor, the slowly falling catalyst activity affects ethylene plant performance
Received 19 September 2019 in an operating cycle. To maximize the integral economic benefit in a regeneration cycle or achieve the lon-
Revised 17 December 2019 gest regeneration cycle, the full cycle operation optimisation must be a long time-scale dynamic optimisation
Accepted 18 January 2020
for which the method of control vector parameterization is adopted. Because of various uncertain disturban-
Available online 5 February 2020
ces, some distances must be reserved for process constraints and the desired effect of full cycle optimisation
is hardly fully achieved. Thus, based on the strictly slowly-time-varying catalyst deactivation model, the
Keywords:
operation margin consumption is estimated. Furthermore, a modified full cycle operation optimisation strat-
Process engineering systems
Dynamic optimisation
egy is presented which takes the operation margin consumption as additional constraints. To obtain the max-
Operation margin imum integral economic benefit or the longest regeneration cycle, we present a realizable full cycle
Control vector parameterisation optimized operation plan that considers the sufficient operation space reserved for uncertain disturbances.
Acetylene hydrogenation reactor © 2020 Taiwan Institute of Chemical Engineers. Published by Elsevier B.V. All rights reserved.

1. Introduction To improve industrial competitiveness, operation optimisation pro-

vides a unified framework for reducing production costs, meeting safety
1.1. Literature review requirements and environmental regulations, improving product quality,
and reducing product variability. In most chemical processes, operation
Acetylene hydrogenation reactors are used to convert acetylene in optimisation is always treated as steady-state optimisation. But for a
high concentration ethylene stream. Acetylene hydrogenation reactor is slowly-time-varying process such as acetylene hydrogenation reactor, it
a complex nonlinear system, which can be described by two modelling should be treated as a long time-scale dynamic process. Thus, the full
approaches the pseudo-homogeneous and heterogeneous model. For cycle operation optimisation must be a dynamic optimisation. Dynamic
simplified modelling, one-dimensional pseudo-homogeneous model has optimisation is commonly used in aerospace system [29], electric [22]
been widely accepted [9,40,41]. Acetylene hydrogenation reactors exten- and petrochemical process systems [10,35]. Considering the complexity
sively utilise improved bimetallic catalyst and the pseudo-homogeneous and uncertainty, the dynamic optimisation of chemical process cannot be
model is also built [12,26]. The accuracy of heterogeneous model is solved easily. As an application of dynamic optimisation, acetylene hydro-
higher than that of pseudo-homogeneous model, but the heterogeneous genation process is discussed from the optimisation model to the solving
model suffers from high computation complexity [18,36]. methods [38,39]. In petrochemical process, only the numerical method
Since the catalyst activity is usually decreasing slowly with time, a can be used to solve the dynamic optimisation problem. Considering the
catalyst deactivation model should be built [1,17,20,30,31]. Several model complexity, the method of control vector parameterisation (CVP)
catalyst deactivation models of acetylene hydrogenation reactors have is widely used for solving the dynamic optimisation problem, where the
been proposed, which mainly considered that the decreasing catalyst control vector is parameterised by finite parameters in each time interval
activity is related to the temperature and poisoning effect [4]. In practi- [23,27]. CVP has been discussed in detail by many scholars [3,5,6,11,37].
cal processes, the fitted curves from industrial data are often adopted However, in the actual process, because of various uncertain disturbances,
for catalyst deactivation modelling, which is of no universality. Hence, some distances must be reserved for process constraints and the desired
the deactivation kinetics model based on two-dimensional heteroge- effect of full cycle optimisation is hardly fully achieved [19,43].
neous model has been proposed, and the optimal economic benefits Because of the slow-time-varying characteristic, with the changes of
could be obtained by operation optimisation, which only considers the slowly-time-varying parameters, the operating point slowly deviates
steady-state operation (the objective function is diurnal economic ben- from the steady-state optimum design point. The performance of chem-
efit) and did not consider margin consumption [42]. ical plant will become worse, even some process constraints may be vio-
lated. Moreover, various uncertain disturbances may also perturb the
* Corresponding author. operating point from the desired steady-state optimum design point.
E-mail addresses: xufeng@cup.edu.cn, xufengfxzt@sohu.com (F. Xu).

https://doi.org/10.1016/j.jtice.2020.01.008
1876-1070/© 2020 Taiwan Institute of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
30 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42

Nomenclature hs steady state equations of the fast-time-varying

system
Parameters N time interval of CVP
a specific surface area of catalyst(m
1) pa, pb, pc acetylene, ethylene and hydrogen pressure in
t number of axial discretization in the reactor the reactor (kPa)
v number of radial discretization in the reactor ps, pg the catalyst and the gas phase average gas pres-
v1 frequency of disturbance sure in the reactor (kPa)
s catalyst porosity pi1, pi2, pi3 gas pressure in the 1
3 stage reactor(kPa)
h1, h2 efficient factor of the ethylene and ethane R dimensionless radius of the reactor
generated reaction ra, rb, rc reaction rate of acetylene, ethylene and hydro-
λsR , λgR the catalyst and the gas phase heat conductiv- gen in the reactor (mol m
3 s
1)
ity coefficient of reactor radial direction rg reaction rate of green oil in the reactor
(W m
1 K
1) (mol m
3 s
1)
cp heat capacity (J kg
1 K
1) r1, r2, r3 reaction rate in the 1
3 stage reactor
DRa, DRb, DRc acetylene, ethylene and hydrogen diffusion (mol m
3 s
1)
coefficient of reactor radial (m2 s
1) Ts, Tg the catalyst phase and the gas phase tempera-
E1, E2, E3 activation energy of reaction coefficients of ture in the reactor (K)
the ethylene, ethane and green oil generated T1, T2, T3 temperature in the 1
3 stage reactor(K)
reaction (J¢mol
1) t time of fast-time-varying system(minute)
Ed activation energy of catalyst deactivation U control variable
(kJ kmol
1) u optimisation variable
DH1, DH2 heat of the ethylene and ethane generated x dynamic state variables of the fast-time-varying
reaction (J mol
1) system
hw interphase heat transfer coefficient x steady state variables of the fast-time-varying
(W m
1 K
1) system
k1, k2, k3 reaction coefficients reaction coefficients of Z dimensionless axial of the reactor
the ethylene, ethane and green oil generated
reaction
kd, Kd catalyst deactivation coefficients
Kp proportional gain of PID controller
KI integration time of PID controller For this problem, sufficient margins must be added for design variables
KD differentiation time of PID controller to allow sufficient distances between the operating point and process
Ma, Mb, Mc acetylene, ethylene and hydrogen price coeffi- constraints and prevent process variables from breaking process con-
cient (RMB kPa
1 day
1) straints after slowly-time-varying parameters become “worse”. The cur-
Rg gas constant (J mol
1 K
1) rent research on margin mainly focused on enhancing the operation
t2, t3 step change and one-order inertia link time flexibility of chemical plants involving controllability evaluation [16],
constant of disturbance dynamic economic analysis [34], resiliency index [13] and design mar-
v1,v2,v3 sine wave, step change, and one-order inertia gin [32]. The chemical process is designed to ensure safe operation and
link amplitude of disturbance product quality. However, the design margins are usually determined
according to the worst-case influence of various uncertainties, while the
Variables changes of slowly-time-varying parameters from the initial state to the
u catalyst activity “worst” state is a slowly dynamic process. Many researchers have pro-
e broadened constraint posed simultaneous process and control integration design methods for
r average density (kg m
3) heat exchanger network and fluid catalytic cracking unit, which can
dEj ,dRj The conversion factors of bounds in model 7 and 8 ensure the satisfaction for process performance under the various
v1 time of slowly-time-varying system(day) uncertainties [19,43].
Gf maximum operating period(day) To coordinate the control and optimisation problems under the
Cg Green oil concentration (mole fraction) uncertain disturbances, a “back-off” method for estimating the influence
d the optimal design variables of uncertain disturbances is proposed, and the distance between operat-
Dd process margin ing points and constrain boundaries is defined [24,25]. The “back-off”
Ddc control margin method is early applied to improve the steady state optimisation effect
Ea,Eb,Ec the function of acetylene, ethylene and hydro- for dynamic processes [2,7]. Furthermore, the optimisation of complex
gen economic benefits nonlinear models is enabled using “back-off” method [15,21,28]. The
e steady state error stochastic simulation using a given probability distribution is used by
DF The upper limits of economic benefits “back-off” method, improving the computational efficiency [8,33]. In
DF1 The upper limits of diurnal catalyst activity recent years, the “back-off” method has been applied to the optimisa-
decreased tion method of integrated design, optimisation and control [14]. For a
f equality constraint of the slowly-time-varying slowly-time-varying system, the “back-off” method can offset the effect
system of uncertainties. However, it still needs to combine certain effect of the
g inequality constraints of the slowly-time-vary- slowly-time-varying parameters, when coordinating the control and
ing system optimisation problems.
DgjE ,DgjR sum of surplus and control margins solved using
model 3 and 6 1.2. Our contributions
h dynamic state equations of the fast-time-varying
system The systematic calculation of operation optimisation relies on the
precise mathematical models. Particularly during the long period of
 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42 31

the full cycle operation optimisation using complex nonlinear mod- product quality requirements. Compared with other uncertainties in
els, a small deviation caused by the initial modelling error can greatly process, the decreasing of catalyst activity is inevitable but predictable,
affect the entire operation during operating cycle. In this paper, a and it has a significant effect on product [18]. Thus, the modelling of
mechanistic modelling approach is used based on acetylene hydro- acetylene hydrogenation reactor focuses on catalyst deactivation
genation intrinsic kinetics and catalyst deactivation kinetics. During modelling based on the intrinsic kinetics. Catalyst activity decreasing
the operation optimisation of an acetylene hydrogenation reactor, is a complex process involving the surface temperature variation of
the equivalent catalyst activity is generally assumed in all internal catalyst, the accumulated oligomer of secondary reaction and the
sites of the reactor for most of research [9,12,18,26,36,40,41]. How- change in molecular structure [38]. For catalyst activity, the greatest
ever, in this paper, considering the integrity of the acetylene hydro- short-run influence is from the temperature variation of the catalyst
genation reactor and the complexity of dynamic running, the internal surface. Nonetheless, the long-run influence is from the clogging of the
catalyst activity is handled as distributed parameter. The acetylene surface micro-pores by the accumulated oligomer of the secondary
hydrogenation reactor is expected to be operated for the maximum reaction product (green oil) leading to slowly catalyst deactivation.
integral economic benefit in certain regeneration cycle or the longest The catalyst activity u is defined as a distributed parameter related
regeneration cycle, whose outlet acetylene concentration is the main to the radial and axial direction distance, that is u (Z, R). Correspond-
constraint of operation optimisation. Sometimes, for maintaining the ing to the catalyst activity of an arbitrary point in the reactor, the
long-time steady operation of the ethylene plant, the longer regener- oligomer concentration is also defined as a distributed parameter
ation cycle is more preferred than economic benefit. The full cycle related to the radial and axial direction distance, that is Cg(Z,R). On
operation optimisation is solved via the CVP method on a long time- the basis of the oligomer concentration of arbitrary point in the reac-
scale. Considering that the time scale of full cycle operation dynamic tor, the kinetic equation of slowly catalyst deactivation is derived.
optimisation is in days and the time scale of process controllers is in Primarily, based on energy conservation and intrinsic kinetics, the
minutes or seconds, the model of acetylene hydrogenation reactor following reaction rate equation of oligomer (green oil) producing is
and its control system are running on short time-scale whereas the obtained [42]:
full cycle operation optimisation is solved on a long time-scale. How- 2
@Cg ðZ; R; t Þ kd psak ðZ; R; t Þpsck ðZ; R; t Þ
ever, if the full cycle operation optimisation and the control system ¼
3 ð1Þ
are both carried out on a unified platform, it can cause computational @t 1 þ Kd ps ðZ; R; t Þ
ak
complexity. Thus, the two systems should run on different platforms
p is the gas partial pressure in the reactor. Subscripts a, b, c represent
and some variables of them are shared.
acetylene, ethylene and hydrogen respectively. Subscript k denotes
The design margins are determined by the worst-case influence of
the sequence number of reactor bed with k = 1, 2, 3. Superscripts s
various uncertainties, and must make up the influence of slowly-time-
and g represent the variable of gaseous and solid phase equation for
varying parameters and uncertain disturbances. Thus, one part of mar-
the heterogeneous model.
gin consumption is a long time-scale dynamic process affected by
For the equation of green oil concentration, two sides of Eq. (1) are
slowly-time-varying parameters, at the same time, another part of
integrated over t. Furthermore, the temperature is considered as the
margin consumption is serving for control system to resist the uncer-
key factor for increasing adjustability of catalyst activity. Based on
tain disturbances under short time-scale. In our study, considering the
thermal deactivation kinetics, the reaction rate equation of catalyst
dynamic processes of the slowly-time-varying parameters and the
deactivation is represented as Eq. (2), where T is temperature in the
uncertain disturbances under different time scales, the consumed mar-
reactor [42].
gins involving process margins (for slowly-time-varying parameters)  
and control margins (for operation and control) in an operation cycle @uk ðZ; R; tÞ
sd
E

are estimated through on-line optimisation. ¼
kd e RgTk ðZ;R;tÞ Cg ðZ; R; t Þ ð2Þ
@t
The maximum potential benefit from operation optimisation is
determined by the distances between the current operating point Then, by substituting the equation of Cg (green oil concentration)
and process constraints, which is related to the operation margins of into Eq. (2), the following reaction rate equation of catalyst deactiva-
design variables. Considering the various disturbances in chemical tion is obtained:
 
processes, some distances must be reserved for fluctuations of pro- Z t s 2
pak ðZ; R; zÞpsck ðZ; R; zÞ
Ed
@uk ðZ; R; tÞ
R T s ðZ;R;tÞ
cess variables. Thus, the operation optimisation must not fully ¼
kd e g
k dz ð3Þ
@t 0 ð1 þ Kd ps ðZ; R; zÞÞ3
explore the operation space from design margins. The maximum ak

potential benefit is related to the design margins minus the con-

sumed process margins, while the achievable benefit is related to the
design margins minus the consumed process and control margins. In 2.2. Structure of full cycle operation optimisation
our study, the economic benefits of operation optimisation (including
the maximum potential benefit and achievable benefit) are estimated The process-based model for the three stages of reactor bed is the
on-line according to the distances between the operating point and foundation for full cycle operation optimisation. We substitute the
active constraints. Furthermore, a modified full cycle operation opti- Eq. (3) into the heterogeneous model of the acetylene hydrogenation
misation strategy is presented which takes the operation margin con- reactor to obtain a stricter and more suitable model, whose detailed
sumption as additional constraints. A full cycle optimized operation description is shown in the Appendix A.
plan maintaining sufficient operation space can be obtained. The simulation of fast-time-varying system and the optimisation of
slowly-time-varying system based on mechanism models are imple-
2. Modelling and structure of full cycle operation dynamic mented simultaneously. In a fast-time-varying system, PID controllers
optimisation should be used to compensate for the bad affection of uncertain distur-
bances. We design the control systems in which the temperature and
2.1. Preliminary work hydrogen inlets of the three stage reactors are the manipulated variables
of six PID controllers. The PID controller parameters are shown in Table 1.
As a representative slowly-time-varying process, with the effect of
slowly-time-varying parameter (catalyst activity), the acetylene To ensure the simultaneous operation of slowly-time-varying sys-
hydrogenation reactor is expected to be operated for longer regenera- tem and the fast-time-varying system, we propose the following
tion cycle or more economic benefit at the same time satisfying the structure of simulation implementation on double platforms at the
32 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42

Table 1 3. Full cycle operation optimisation of acetylene hydrogen

Parameters of the controllers. reactor
Operation variable Proportional Integral Derivative
gain time time For acetylene hydrogenation reactors, we want more economical
benefit and longer regeneration cycle while ensuring the quality of
Temperature First stage reactor 0.6 6
inlets Second stage reactor 0.6 7 outlet ethylene product. With longer regeneration cycle of catalyst,
Third stage reactor 0.5 7 the inlet hydrogen can be increased and the inlet temperature for
Hydrogen First stage reactor 30 8 0.1 activity compensation can be reduced but leading to increased eco-
inlets Second stage reactor 30 7 0.1
nomic consumption. Our operation optimisation will coordinate the
Third stage reactor 30 5 0.1
incompatibility between the two problems. The full cycle operation
optimisation is a long time-scale dynamic optimisation whose time
same time. The long time-scale unit of slowly-time-varying system in unit is day.
days is represented by G. To obtain the catalyst activity of the G day, Generally speaking, there are two main types of method for solv-
the reaction rate of catalyst deactivation is integrated, as shown in ing dynamic optimisation, namely numerical method and analytical
Eq. (4). method. However, efficiently obtaining a high-quality solution for
such dynamic optimisation is challenging and difficult. Especially
uk ðZ; R; GÞ ¼ uk ðZ; R; 0Þ when the problem contains large-scale models, solving the Hamilto-
  nian for complex processes is almost impossible through analytical
Z G Z
psak 2 ðZ; R; zÞpsck ðZ; R; zÞ
Ed t

R

kd e
s
g T ðZ;R;tÞ
k
 3 dzdc methods when the calculus of variations is used. By time discretiza-
0 0
1 þ psak ðZ; R; zÞ tion, the numerical method transforms the infinite dimensional
dynamic optimisation problem to finite dimensional nonlinear pro-
ð4Þ gramming problem [3,5,6,11,23,27,37]. Although an optimal solution
However, considering the computation complexity, we ignore the for the Pontryagin’s maximum principle cannot be obtained, the
short time-scale dynamic fluctuation of process variables in one day. numerical method is simple and easy be solved, which is widely used
Thus, we take the mean values of process variables in one day into in the chemical processes. The model of acetylene hydrogenation
Eq. (4), and Eq. (4) is transformed into a discrete form as Eq. (5). reactor is considered as a high non-linear complex model. Hence, the
dynamic optimisation problem is only solved using numerical meth-
uk ðZ; R; GÞ ¼ uk ðZ; R; 0Þ ods. In addition, to simplify the calculation of dynamic optimisation,
  the distributed parameter system should transform the original dis-
G
X
R
Ed
X i
psak 2 ðZ; R; jÞpsck ðZ; R; jÞ
s
g T ðZ;R;iÞ

kd e k
 3 ð5Þ tributed parameter system into an equivalent large-scale lumped sys-
i¼1 j¼1 1 þ psak ðZ; R; jÞ tem. Through radial and axial discretisation, the numbers of the
radial and axial interval are represented by v and t , respectively, as
The structure of simulation implementation is shown in Fig. 1. The shown in Eq. (6). Hence, the dynamic optimisation model easier to be
optimised operation variable u is sent to the fast-time-varying system, solved is obtained.
which maintains the normal running of the reactor through the PID 
½0; Z  ! ½0; Z1 Þ; ½Z1 ; Z2 Þ; ⋯; ½Zλ
1 ; Zλ Þ; ⋯; ½Zv
1 ; Zv Þ
controllers. Correspondingly, when the fast-time-varying system is ð6Þ
½0; R ! ½0; R1 Þ; ½R1 ; R2 Þ; ⋯; ½Rλ
1 ; Rλ Þ; ⋯; ½Rv
1 ; Rv Þ
running on steady state through the PID controller, the steady state
variables represented as x are transferred to the slowly-time-varying In addition, considering both the simulation implementations of
system. In addition, during the on-line margin estimation, the state fast-time-varying and slowly-time-varying systems, Gf is defined as
variable of the fast-time-varying system are transferred to the on-line the cycle of catalyst regeneration, and short time-scale state variables
margin estimator to calculate the surplus margins. Then, the optimised and long-time-scale state variables x; x are represented as follows:
variables that ensure a certain operation margin are obtained using the  
x ¼ x1 ; x2 ; ⋯; xG ; ⋯; xGf
optimiser. Simulation software gPROMS is used to build the model and   ð7Þ
control system, and the optimizer is built on MATLAB software. x ¼ x1 ; x2 ; ⋯; xG ; ⋯; xGf

Fig. 1. Structure of slowly-time-varying system.

 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42 33

The CVP method divides the time duration of the whole dynamic Table 2
process into several time intervals. The optimisation horizon is Constraints of the full cycle operation optimisation.

divided into several time intervals, which are shown in Eq. (8). Operation variable Initial Trajectory constraint
          constraint
0; Gf ! 0; G1 ; G1 ; G2 ; ⋯; Gλ
1 ; Gλ ⋯; GN
1 ; GN ð8Þ
Temperature T1s ð0; R; Gλ Þ [320,325] ½T1s ð0; R; Gλ
1 Þ
e1 ; T1s ð0; R; Gλ
1 Þ þ e1 
N is the number of time intervals which determines discrete preci- Inlets (K) T2s ð0; R; Gλ Þ [320,325] ½T2s ð0; R; Gλ
1 Þ
e2 ; T2s ð0; R; Gλ
1 Þ þ e2 
sion. The discrete optimisation variable is shown in Eq. (9). T3s ð0; R; Gλ Þ [321.5,322.5] ½T3s ð0; R; Gλ
1 Þ
e3 ; T3s ð0; R; Gλ
1 Þ þ e3 
  Hydrogen psc1 ð0; R; Gλ Þ [28.5,29.2] ½psc1 ð0; R; Gλ
1 Þ
e4 ; psc1 ð0; R; Gλ
1 Þ þ e4 
uðGÞ ¼ u1 ðG1 Þ; u2 ðG2 Þ; ⋯; uλ ðGλ Þ; ⋯; uN ðGN Þ ð9Þ Inlets psc2 ð0; R; Gλ Þ [13.8,14.2] ½psc2 ð0; R; Gλ
1 Þ
e5 ; psc2 ð0; R; Gλ
1 Þ þ e5 
(kPa) psc3 ð0; R; Gλ Þ [0.77,0.81] ½psc3 ð0; R; Gλ
1 Þ
e6 ; psc2 ð0; R; Gλ
1 Þ þ e6 
uλ(Gλ) represents the trajectory of u(G) in [Gλ
1, Gλ), whose value
is invariable for piecewise constant strategy. For the fast-time-vary-
ing system in a full cycle, the equations of process-based model and
control system model for the acetylene hydrogen reactor in Table 1 stages of the acetylene hydrogen reactor and outlet acetylene concentra-
are represented as h(&), as shown in Eq. (10). tions of the end stage. The constraints of the full cycle operation optimi-

T sation include the initial and trajectory constraints as shown in Table 2.
hðx; u; tÞ ¼ h1 ðx1 ; u1 ; t Þ; h2 ðx2 ; u1 ; t Þ; ⋯; hG ðxG ; uλ ; tÞ; ⋯; hGf xGf ; uN ; t Relative to initial constraints, trajectory constraints are continually broad-
ð10Þ ened for long-run decreasing catalyst activity effect. Broadened constraint
e depends on the reaction rate of decreasing catalyst activity and the con-
By CVP, the optimisation variables are converted into a set of param- trol process of the fast-time-varying system. In addition, there is a neces-
eters, that is, the original dynamic optimisation problem is transformed sary constraint for process requirements that the acetylene fraction of
into the finite dimensional nonlinear programming problem. The non- total gases of reactor outlet is lower than 5 £ 10
6.
linear programming problem an approximation of the original dynamic The optimiser uses CVP to transform the original full cycle opera-
optimisation problem, and their difference is eliminated while N tends tion optimisation problem into a nonlinear programming problem,
to be infinite, but increasing N increases the computation. which can be solved using the sequential quadratic programming
(SQP) method. However, considering the variations in the operating
3.1. Operation optimisation for economic benefit
conditions due to the full cycle trajectories determined by the optimi-
sation framework, the tuning PID parameters under specific conditions
In this section, we implement the full cycle operation optimisation to
have to perform well under the full cycle trajectory. Therefore, it is
maximise the integral economic benefit in certain regeneration cycle. For
necessary to optimize the PID parameters of the controller. Admit-
constructing the function of ethylene economic benefits in the G day, the
tedly, the PID parameters of 6 controllers can be substituted into the
radial average ethylene partial pressure of the end stage reactor outlet is
optimisation framework as optimisation variables, but this method
multiplied by the ethylene price coefficient Mb, as shown in Eq. (11).
greatly increases the computational burden of the full cycle operation
Mb X
Rt optimisation. Therefore, we propose a method to obtain the subopti-
Eb ðGÞ ¼ psb3 ð1; R; GÞ ð11Þ mal solution using an iterative method. The steps are as follows:
t R¼0

Similarly, by calculating the summation of the hydrogen and acety-

(1) Solve model 1 under PID parameters are shown in Table 1;
lene pressure difference between the inlet and outlet of the reactor, the
(2) Fix the obtained optimisation variables and optimize the PID
economic benefit consumption in G day is obtained, as shown in Eq. (12).
parameters in model 1;
Ma X
Rt X
3
 (3) Fix the PID parameters obtained from optimisation of 2 and then
Ea ðGÞ ¼ psak ð1; R; GÞ
psak ð0; R; GÞ solve model 1;
t R¼0 k¼1
ð12Þ (4) Repeat steps 2 and 3 until the obtained PID parameters and opti-
Mc X
Rt X
3

Ec ðGÞ ¼ psck ð1; R; G
Þ
psck ð0; R; GÞ misation variables are no longer changed.
t R¼0 k¼1

The economic benefit of the full-cycle is obtained by integrating The optimized PID parameter trajectories are shown in Fig. 2.
the diurnal economic benefits, which are the economic benefit differ- The results of the full cycle operation optimisation are shown in
entials of reactant gases, as shown in Eq. (13). Table 3, and the optimal operation trajectory is shown in Fig. 3(a). The
Z Gf economic benefits of one year are obtained by solving the optimisation


J1 ¼
Eb ðGÞ
Ea ðGÞ
Ec ðGÞ dG ð13Þ model. In addition, considering that the acetylene fraction in the total
0 outlet gases is a hard constraint which cannot be violated, the regenera-
The inlet hydrogen partial pressures and inlet temperatures of tion cycle without optimisation is only 76 days, and economic benefit of
three reactor stages are chosen as independent variables of dynamic no optimisation is only 2.78 £ 105<. No optimisation refers to the nor-
optimisation, and the dynamic optimisation model is represented as mal operation results of acetylene hydrogenation reactor under normal
follows: conditions. Reactor operation data under normal conditions are shown
Model 1 in Appendix C. For economic benefits and regeneration cycles, the
results of dynamic optimisation whose objective is the maximum inte-
min J1
gral economic benefit in certain regeneration cycle are obviously supe-
rior to the results without optimisation.
s:t: f ðx; u; GÞ ¼ 0
The number of time intervals is a critical parameter in full cycle
operation optimisation based on the CVP method. The obtained eco-
hðx; u; tÞ ¼ 0
nomic benefits with a different number of time intervals are arranged
gðx; u; GÞ0 based on size order as N = 10 < 18 < 30, as shown in Table 3. Notably,
the deviation between the obtained economic benefits of N = 18 and
In the above model, f(&) represents the catalyst deactivation model in N = 30 is less than approximately 1.5%. Considering the computation
the slowly-time-varying system and g(&) represents the constraints of complexity and optimisation accuracy, the suitable number of time
dynamic optimisation, including the operation horizons of the three intervals is approximately N = 20
30.
34 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42

Fig. 2. Optimal curve of PID parameters.

Table 3 3.2. Operation optimisation for regeneration cycle

Parameters and results of the full cycle operation optimisation.

Optimisation parameters and results Value In coordination with the other operation requirement in the eth-
ylene industry, the economic benefit is sometimes considered the
The number of time interval (N) 10 18 30
secondary optimisation objective compared with regeneration cycle.
Economic benefit of full cycle 1.43 £ 106 1.55 £ 106 1.57 £ 106
optimisation (</a) The system influenced by slowly-time-varying parameters is stopped
Economic benefit of no 2.78 £ 105 2.78 £ 105 2.78 £ 105 for catalyst regeneration when it reaches the regeneration cycle Gf.
optimisation (</a) Therefore, the full cycle operation optimisation model whose objec-
tive is the longest regeneration cycle is represented as follows:
 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42 35

Fig. 3. Operation trajectory of full cycle optimisation: a) Operation trajectory of solved model 1; b) Operation trajectory of solved model 2.
36 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42

Model 2 the third bed increased. The simulation on the optimised strategies of
model 1 can obtain more benefits but shorter regeneration cycle,
min J2 ¼
G f
while the other simulation on the optimised strategies of model 2
can maintain a longer regeneration cycle but less benefits. Exact opti-
s:t: f ðx; u; GÞ ¼ 0
misation plan should depend on practical production requirement.
hðx; u; tÞ ¼ 0
4. Full cycle operation optimisation maintaining operation
gðx; u; GÞ0 margin

In the above model, Gf is defined as an integer variable, we should The off-line dynamic optimisation strategy whose objective is the
solve a mixed-integer dynamic optimisation problem. However, con- maximum integral economic benefit in certain regeneration cycle or
sidering that there is one integer variable Gf, this mixed-integer the longest regeneration cycle can be used to determine an optimal
dynamic optimisation problem can be simplified. To solve the mixed- solution for the ideal experiment. However, considering the various
integer dynamic optimisation problem, we can treat the integer vari- disturbances in chemical processes, some distances must be reserved
able Gf as a continuous variable to transform the problem into a con- for the fluctuations of process variables and that the optimum operat-
tinuous dynamic optimisation problem. ing point is not on some process constraints. Accordingly, the design
The optimal operation trajectory is shown in Fig. 3(b), and the margin for reactor length must be added for sustained system opera-
curve of economic benefit is shown in Fig. 7. For the obtaining opti- tion throughout the regeneration cycle. Accordingly, a full cycle oper-
mal regeneration cycle, we must suffer from a certain loss of eco- ation optimisation strategy that keeps operation margin and fully
nomic benefits in the late operation phase. Hence, the optimal uses surplus margin is proposed.
solution of the model 2 is 218.74 days, although the economic bene-
fits obtained are less than those obtained with model 1. 4.1. Estimation of process and control margin
In comparison with model 1, the operation trajectory of model 2 is
smoother for maintaining catalyst activity. The temperature opera- The margin is defined as the value added on the nominal value of the
tion strategies of the two optimisation models are generally design variable because of process uncertainties. The total design margin
decreased for reducing the reaction rate of intrinsic activity deactiva- consists of two parts: process margin and control margin. Process margin
tion. Correspondingly, the hydrogen inlet operation strategies of the is designed for compensating the effect of the change in slowly-time-
two optimisation models generally increase for compensating eco- varying parameters, whereas control margin is designed for operation
nomic loss with decreasing catalyst activity. In Fig. 3(a), considering and control. During the design phase of chemical processes, sufficient
that the catalyst activity in the first bed is close to the constraint margins must be added for design variables to ensure sufficient distances
boundary in the later period of regeneration cycle, the hydrogenation between the operating point and process constraints and prevent the
of the first bed decreases slightly to reduce the oligomer accumula- process variables from breaking process constraints after slowly-time-
tion and maintain certain activity of the catalyst. It increases the acet- varying parameters become “worse”. However, the design margins are
ylene at the inlet of the second bed, thus increasing the inlet usually determined according to the worst-case influence of various
temperature of the second bed of the reactor. In Fig. 3(b), the full uncertainties, while the changes of slowly-time-varying parameters
cycle operation optimisation for the longest regeneration cycle choo- from the initial state to the “worst” state is a slowly dynamic process.
ses the strategy with slowly operating inlet temperature and inlet Hence, the margin consumption is a long dynamic process affected by
hydrogen. In comparison with the initial operating point, the maxi- slowly-time-varying parameters and unmeasured parameters. As shown
mum operation deviation of inlet temperature is less than §1 K and in Fig. 4, the process margin consumption is increasing with time, but
the maximum operation deviation of the inlet hydrogen is less than control margin consumption fluctuates within a certain range. After the
20 kmol h
1. The catalyst activity of the first and second beds would cycle, the surplus margin can be reduced to 0, and the sum of the process
become very low and the hydrogenation reaction is hardly running margin and the control margin is equal to the design margin.
in the later period of the regeneration cycle. For extending the regen- To meet the constraints and for more economic benefit optimisa-
eration period, the third bed of the reactor should be used as far as tion, the estimation of process margin and control margin is signifi-
possible to ensure the percentage of acetylene at the outlet. There- cant for slowly-time-varying chemical process. Process margin in G
fore, the inlet temperature of the first and second beds of the reactor day can be obtained by solving the steady state optimisation model
decreased obviously, and the inlet temperature and hydrogenation in as shown in Model 3.

Fig. 4. Margin consumption.

 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42 37

reduction, only the reactor length is treated as design variable in the

acetylene hydrogenation reactor system. Then, to obtain the optimal
design variables d, we solved the steady state optimisation via model
4 for the economic benefits.
Model 4
min J1
u;d

s:t: f ðx; u; GÞ ¼ 0

hs ðx; u; t; dÞ ¼ 0

gðx; u; G; dÞ0
In addition, to satisfy operation and control the needs of long-
term running, the control margin must be designed by solving the
short time-scale dynamic optimisation model. The operation of
control system must be considered to design the control margin.
Hence, the dynamic equation of the PID control system is
accounted in the short time-scale dynamic optimisation model, as
Fig. 5. Margin consumption at 180-day operation.
shown in model 5.
Model 5
min J3 ¼
Eb ðGÞ þ Ea ðGÞ þ Ec ðGÞ
u;dþDdþDdc
Model 3
min J3 ¼
Eb ðGÞ þ Ea ðGÞ þ Ec ðGÞ s:t: fðx; u; GÞ ¼ 0
u;dþDd

hs ðx; U; t; d þ Dd þ Ddc Þ ¼ 0
s:t: f ðx; u; GÞ ¼ 0
gðx; u; G; d þ Dd þ Ddc Þ0
hs ðx; u; t; d þ DdÞ ¼ 0
Z t

deðtÞ
gðx; u; G; d þ DdÞ0 U ðtÞ ¼ KP eðtÞ þ K
1
I eðtÞdt þ KD
0 dt

DdðGÞ0 eðtÞ ¼ u
xðtÞ

In model 3, d represents the optimal design variables and Dd rep-
resents the process margin of optimal design variables. To facilitate Dd c ðGÞ; DdðGÞ0
the representation of control margin estimator, we transform the In model 5, Ddc(G) represents the control margin of the optimal
fast-time-varying dynamic system into quasi-steady-state equations design variables in the G day. The solved result of the model 5 at the
as hs, which contain no control systems. During the actual process 180-day operation and the percentage of consumed process margin,
operation, with the variation of slowly-time-varying parameters, the consumed control margin and surplus margin in the total margin
process margin of the G day is represented as Dd(G). For complexity design are shown in Fig. 5.
The total design margin is defined as the summation of the process
and control margin in the Gf day, represented as d(Gf) + Dd(Gf) + Ddc(Gf).
The percentage of process margin continuously increases until operation
termination due to catalyst reactivation, and finally obtains approxi-
mately 80% of the total margin design. By contrast, the percentage of con-
trol margin fluctuates within approximately 20
30% of the total margin
design. Exactly 80% of the total margin is remained at the reactor begin-
ning and decreasing to 0 in the 180 day.
The benefit curve of the operation optimisation based on the
above design margin is shown in Fig. 6. Although the total eco-
nomic benefit is 4.75 £ 105 < and less than the results of model 1,
the operation optimisation based on the margin saved can ensure
that the system steadily runs even when faced with uncertain dis-
turbances.
For estimating the process and control margin of the model whose
object is the longer regeneration cycle, considering that the regenera-
tion cycle is a full period variable, the minimum diurnal decreased
activity is considered as object function to replace the regeneration
cycle. In addition, to obtain diurnal decreased activity, the average of
diurnal catalyst activity is extracted from the distributed parameter
Fig. 6. Economic benefit curve of full cycle operation optimisation.
variable. The optimisation model is shown in model 6.
38 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42

constraint and objective function are obtained. The Lagrangian func-

tion is shown in Eq. (14).
P
DF ¼
mj DgjE
P ð14Þ
DF1 ¼
mj DgjR
The steady state operating points of models 3 and 6 are on the
boundary of active constraint; hence, gj ð ¢ Þ ¼ 0, while process margin
consumption exactly compensate for the influence of slowly-time-
varying parameters. The sum of surplus and control margins solved
using model 3 and 6 can move operating points inside the constraint,
represented as DgEj ; DgRj < 0. The upper limits of economic benefits
and diurnal decreased activity which are obtained by operation opti-
misation are equivalent to their declining quantity affected by sur-
plus margin, represented as DF and DF1.
Then, we can construct a function that describes the distance
between operating points and constrained boundary, which is
defined as the margin consumption function. The object function of
model 1 is added by the margin consumption function for structuring
Fig. 7. Economic benefit curve of the full cycle operation optimisation for regeneration the full cycle operation optimisation model which considers margin
cycle.
consumption, as shown in model 7.
Model 7
Model 6 X E
min J5 ¼ J1 þ dj gj2 ð ¢ Þ
1 X
Rt X
Zv X
Rt X
Zv 
min J4 ¼ uðG
1; Z; RÞ
uðG; Z; RÞ j
u;dþDdþDdc vt R¼R1 Z¼Z1 R¼R1 Z¼Z1
s:t: f ðx; u; GÞ ¼ 0
s:t: f ðx; u; GÞ ¼ 0
hðx; u; tÞ ¼ 0
hs ðx; U; t; d þ Dd þ Dd c Þ ¼ 0
gðx; u; GÞ0
gðx; u; G; d þ Dd þ Dd c Þ0
0 < dj gj2 ð ¢ Þ < DF
E
Z t

deðtÞ E
UðtÞ ¼ KP eðtÞ þ K
1 eðtÞdt þ KD The conversion factors of constraints are represented as dj ; and
I
dt P E
0
the margin consumption function is represented as j dj gj2 ð ¢ Þ. Con-
sidering the optimisation results of model 5, we can estimate the con-
eðtÞ ¼ u
xðtÞ
trol and process margin in the G day online. Then, we can calculate
the economic benefit reduction caused by the influence of control
Dd c ðGÞ; DdðGÞ0
and process margin. Their sum is equivalent to the minimum eco-
By solving model 6, the operation optimisation that ensures suffi- nomic benefit loss in the G day, which is then substituted into model
cient control margin is shown in Fig. 7, the economic benefit is less 7 to be solved for the full cycle operation optimisation problem. The
than that in model 2 and regeneration cycle is only 203 days. solved results are shown in Table 4, and the economic benefits of
1.49 £ 106 < are obtained by full cycle operation optimisation main-
4.2. Full cycle operation optimisation maintaining operation margin taining operation margin. In comparison with the results of the full
cycle operation optimisation without margin maintaining, consider-
The operation optimisation in Section 4.1 is based on the estima- ing the saved necessary control margin, the economic benefits of full
tion of the design margin. However, the part of total design margin is cycle operation optimisation maintaining operation margin are less
not consumed until the maximum regeneration cycle. If the part of than that of the full cycle operation optimisation without margin
the design margin can be fully exploited, the full cycle operation opti- maintaining. The economic benefit curve of reactor with respect to
misation can bring more economic benefits under the condition of time is shown in Fig. 6.
ensuring sufficient margin. However, the part of the design margin For ensuring sufficient control margin for operation and control
cannot be immediately but instead slowly released with decreasing and simultaneously taking full advantage of surplus margin, the opti-
catalyst activity. mized benefit curve of the model 7 is slightly less than that of model
The Lagrange multiplier vector of inequality constraints g(¢) in 1 and far more than that of model 5, thus achieving mutual balance
models 3 and 6 of steady state optimisation is defined as m, which between economic benefit and control system requirements.
contains active constraint mj  0 and non-active constraint mi = 0. For possible process uncertainty, the three forms of common dis-
Base on the Kuhn
Tucker condition, the equations between active turbances are considered, including sine wave, step change, and the

Table 4
Comparison with anti-interference performances of models 1 and 7.

Optimisation parameters and results Interference function

Process disturbance model (v)
v1sin v1t v2(t
t2) v3 ð1
e
t=t3 Þ

Economic benefit of the full cycle operation optimisation without margin maintaining (</a) 1.55 £ 106 1.39 £ 106 1.45 £ 106 1.43 £ 106
Economic benefit of the full cycle operation optimisation maintaining operation margin (</a) 1.49 £ 106 1.42 £ 106 1.46 £ 106 1.45 £ 106
 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42 39

Table 5
Comparing with anti-interference performances of model 2 and model 8.

Optimisation parameters and results Interference function

Process disturbance model(v)
v1sin v1t v2(t
t2) v3 ð1
e
t=t3 Þ

Economic benefit of the full cycle operation optimisation without margin maintaining (</a) 6.44 £ 105 3.68 £ 105 4.68 £ 105 4.36 £ 105
Economic benefit of the full cycle operation optimisation maintaining operation margin (</a) 5.56 £ 105 4.81 £ 105 5.45 £ 105 5.18 £ 105
Regeneration cycle of the full cycle operation optimisation without margin maintaining (day) 218.74 208.43 211.02 209.83
Regeneration cycle of the full cycle operation optimisation maintaining operation margin (day) 218.36 212.84 215.41 214.19

output of one-order inertia object, and they are added in the short is longer than that without margin maintaining under disturbances.
time-scale dynamic optimisation model. In Table 4, v1 and v1 repre- Comprehensively, the full cycle operation optimisation maintaining
sent the amplitude and frequency of the sine wave form disturbance; operation margin can produce optimal economic benefits and long
v2 and t2 are the amplitude and occurrence time of the step distur- regeneration cycle with disturbances.
bance; v3 and t3 are the amplitude and time constant when the dis-
turbance is in the form of a one-order inertia object. 5. Conclusions
Faced with the same disturbances, the economic benefit of the full
cycle operation optimisation maintaining operation margin is more Based on the established process-based model of acetylene hydro-
than that of the full cycle operation optimisation without margin genation reactors, this work discusses the feasibility of the full cycle
maintaining. The full cycle operation optimisation maintaining oper- operation optimisation and proposes a long time-scale dynamic opti-
ation margin is of superior performance for anti-disturbance. misation model based on the estimation of operation margin. For
Similar to Section 3.2, the full cycle operation optimisation model coordinating the long time-scale dynamic optimisation and short
maintaining operation margin for a longer regeneration cycle solves time-scale dynamic control, the structure of double platform is used
a mixed-integer dynamic optimisation problem. Hence, based on the for simulation implementation with a slowly-time-varying system.
mixed-integer dynamic optimisation problem which can be simpli- The full cycle operation optimisation of the acetylene hydrogena-
fied, the optimisation method in Section 3.2 is utilised for the full tion reactors is solved based on CVP, and we discussed the optimal
cycle operation optimisation maintaining operation margin. The range for the number of time interval. In addition, for the full cycle
object function of model 2 is added by the margin consumption func- operation optimisation problem, whose object is the regeneration
tion for realizing the full cycle operation optimisation which consid- cycle, a method is proposed to solve the continuous problem and
ers margin consumption, as shown in model 8. determine the optimal solution in the feasible domain.
Model 8 Based on the method of margin estimation, a full cycle operation
X R
min J6 ¼ J2 þ dj gj2 ð ¢ Þ optimisation maintaining operation margin is proposed ensuring
j preferable optimized results with uncertain disturbances. Although
the full cycle operation optimisation without margin maintaining
s:t: f ðx; u; GÞ ¼ 0 obtains the best effect under the ideal condition, the full cycle opera-
tion optimisation maintaining operation margin obtains the achiev-
hðx; u; tÞ ¼ 0 able better effect under disturbance condition. In comparison with
the steady-state operation optimisation that ensures the design mar-
gðx; u; GÞ0 gin, the full cycle operation optimisation maintaining operation mar-
gin can obtain more economic benefits or longer regeneration cycle.
X
0< dRj gj2 ð ¢ Þ < DF1
j Declaration of Competing Interest
R
The conversion factors of constraints are represented as dj , while
P R There are no conflicts of interest.
the margin consumption function is represented as j dj gj2 ð ¢ Þ. Simi-
lar to Section 3.2, the integer variable Gf is treated as continuous vari-
able to solve for the continuous dynamic optimisation problem as the Acknowledgements
form of model 8.
The optimal regeneration cycle is approximate to the model 2, and This work is supported by the National Natural Science Founda-
the obtained benefits are slightly less than that of model 2. In com- tion of China (21676295).
parison with model 6, model 8 has obvious advantages in regenera-
tion cycle and economic benefits. Appendix A. Process-based model of acetylene hydrogenation
Similar to Table 4, the three forms of disturbances are respectively based on deactivation kinetics
added in the reactor model. The optimised solution of economic ben-
efits and regeneration cycle is shown in Table 5. The kinetics equation of acetylene hydrogenation, ethylene
Although the benefit of the full cycle operation optimisation hydrogenation and oligomer of secondary reaction product are
maintaining operation margin is less than that of the full cycle opera- respectively shown in Eqs. (15)
(17), respectively.
tion optimisation without margin maintaining when there are no dis-

E1
turbances, but under the disturbance of sine wave, step change, and s

r1 ðZ; R; t Þ ¼ k1 eRgTk psck ðZ; R; t Þuk ðZ; R; t Þ ð15Þ
one-order inertia link the optimized benefit of the full cycle operation
optimisation maintaining operation margin is more than that of the
E2
s psbk ðZ; R; tÞpsck ðZ; R; tÞ
full cycle operation optimisation without margin maintaining. More-
r2 ðZ; R; tÞ ¼ k2 eRgTk ð16Þ
psak ðZ; R; tÞ
over, the optimised regeneration cycle maintaining operation margin
40 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42

Fig. 8. Distribution of catalyst activity in the reactor.

90
0.05 1.0

0.04
0.5

50day operation
0.03 0.006
Center of reactor Acetylene Hydrogenation
R
Cg

0.0 180 0
0.005 Reactor
0.02
0.004 Wall of reactor Center of reactor
0.60 0.65 0.70 0.5
0.01

10day operation Wall of reactor

0.00 1.0
0.2 0.4 0.6 0.8 1.0 270
Z

Fig. 9. Distribution of green oil concentration in the reactor.

E3
s psak 2 ðZ; R; tÞpsck ðZ; R; tÞ The thermal balance equation for fluid phase is shown as follows:

r3 ðZ; R; tÞ ¼ k3 eRgTk  3 ð17Þ
1 þ psak ðZ; R; tÞ @Tkg ðZ; R; tÞ
srg cpg
@t " #
@T g ðZ; R; tÞ 2 g
g @ Tk ðZ; R; tÞ
g
1 @Tk ðZ; R; tÞ ð23Þ
¼
rg ucpg k þ λR þ
Generating rate of acetylene, ethylene, hydrogen and oligomer @Z @R2 R @R
(green oil) are shown in Eqs. (18)
(21), respectively.  s g 
þ hw a Tk ðZ; R; tÞ
Tk ðZ; R; tÞ
r a ¼ h1 r 1 ð18Þ
The mass balance equation for catalyst phase is shown as follows:
rb ¼
h1 r1 þ h2 r2 ð19Þ @ps ðZ; R; tÞ
ð1
s Þ ik
 @t  ð24Þ
rc ¼ h1 r1 þ h2 r2 ð20Þ ¼ kgi a pgik ðZ; R; tÞ
psik ðZ; R; tÞ þ ð1
s Þu ri

rg ¼
r3 ð21Þ The thermal balance equation for catalyst phase is shown as fol-
lows:
The mass balance equation for fluid phase is shown as follows:
@T s ðZ; R; tÞ  
@pg kðZ; R; tÞ ð1
s Þrs cps k ¼ hw a Tkg ðZ; R; tÞ
Tks ðZ; R; tÞ
s i " @t #
@t X2 ð25Þ
s @ Tk ðZ; R; tÞ 1 @Tks ðZ; R; tÞ
2 s
" g #
@pgi kðZ; R; tÞ @2 pi kðZ; R; tÞ 1 @pgi kðZ; R; tÞ þλR þ þ hj uð
rj Þð
DHj Þ
ð22Þ @R 2 R @R
¼
u þ dDRi þ j¼1
@Z @R2 R @R
g  g  As shown in Fig. 8, based on the heterogeneous model, the radial
þ ki a pi kðZ; R; tÞ
pi kðZ; R; tÞ
s
and axial catalyst activity in the reactor is presented on the 10th and
 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42 41

Rg gas constant 8.314

50th day of operation. Considering the accumulated green oil and
(J mol
1 K
1)
temperature variation, the catalyst activity on the 50th day of opera- t2, t3 step change and one- 200, 1.25
tion is lower than that on the 10th day of operation. Considering the order inertia link time
acetylene hydrogenation reaction depth, the catalyst activity con- constant of
sumption at the front of the reactor is more than the catalyst activity disturbance
v1, v2, v3 sine wave, step change, 0.1, 0.05, 0.2
consumption at the end of reactor. As temperature differs between and one-order inertia
the reactor wall (radius R = 1) and the centre (radius R = 0), the radial link amplitude of
catalyst activities have slight deviation. disturbance
Relative to activity, the radial and axial concentration of green oil
is the dominant factor of long-run falling catalyst activity, as shown
in Fig. 9. Oppositely, the concentration of accumulated green oil on Appendix C. Operation data of the acetylene hydrogenation
the 50th day of operation is greater than the concentration on the reactor under normal conditions
10th day of operation. Similar to the activity, as temperature differs
between the reactor wall and centre, the radial rate of green oil pro-
Operating variable Value
ducing slight deviates. Therefore, the radial and axial values of activ-
ity and green oil in the reactor can be accurately calculated by Hydrogen partial pressure of the first bed inlet / kPa 29.55
Temperature of the first bed inlet / K 320.8
establishing a process-based model of catalyst activity.
Acetylene partial pressure of the first bed inlet / kPa 29.55
Ethylene partial pressure of the first bed inlet / kPa 1684.35
Appendix B. Parameter values used in the process model Hydrogen partial pressure of the second bed inlet / kPa 14
Temperature of the second bed inlet / K 321.8
Hydrogen partial pressure of the third bed inlet / kPa 0.79
a specific surface area of 29.556 Temperature of the third bed inlet / K 322.8
catalyst(m
1)
t number of axial discreti- 26
zation in the reactor
References
v number of radial discre- 13
tization in the reactor
v1 frequency of 3 [1] Albers P, PietschI J, Parker SF. Poisoning and deactivation of palladium catalysts. J
Mol Catal A Chem 2001;173(1):275–86.
disturbance
[2] Bahri PA, Romagnoli JA, Bandoni JA, Barton GW. Back-off calculations in optimis-
s catalyst porosity 0.35
ing control: a dynamic approach. Comput Chem Eng 1995;19:699–708.
h1,h2 efficient factor of the 3.461, 4.26 £ 10
3 [3] Binder T, Cruse A, Villar CAC, Marquardt W. Dynamic optimisation using a wave-
ethylene and ethane let based adaptive control vector parameterization strategy. Comput Chem Eng
generated reaction 2000;24(2):1201–7.
λsR ,λgR the catalyst and the gas 0.578, 0.374 [4] Borodzinski A, Cybulski A. The kinetic model of hydrogenation of acetylene
ethy-
phase heat conductiv- lene mixtures over palladium surface covered by carbonaceous deposits. Appl
ity coefficient of reac- Catal A 2000;198(1):51–66.
tor radial direction [5] Chen TWC, Vassiliadis VS. Inequality path constraints in optimal control: a finite
(W m
1 K
1) iteration e-convergent scheme based on pointwise discretization. J Process Contr
cp heat capacity 1.724 2005;15(3):353–62.
(J kg
1 K
1) [6] Chen X, Du W, Tianfield H, Qian F. Dynamic optimisation of industrial processes
DRa, DRb, DRc acetylene, ethylene and 2.75 £ 10
5, 3.42 £ 10
5, with nonuniform discretization-based control vector parameterization. IEEE T
Autom Sci Eng 2014;11(4):1289–99.
hydrogen diffusion 7.75 £ 10
6
[7] Figueroa JL, Bahri PA, Bandoni JA, Romagnoli JA. Economic impact of disturbances
coefficient of reactor
and uncertain parameters in chemical processes - a dynamic back-off analysis.
radial (m2 s
1)
Comput Chem Eng 1996;20(4):453–61.
E1, E2, E3 activation energy of 66.89, 52.8, 2100 [8] Galvanin F, Barolo M, Bezzo F, Macchietto S. A backoff strategy for model-based
reaction ethylene, experiment design under parametric uncertainty. AIChE J 2009;56(8):2088–102.
ethane and green oil [9] Gobbo R, Soares RP, Lansarin MA, Ferreira MP. Modeling, simulation, and optimi-
generated reaction sation of a front-end system for acetylene hydrogenation reactors. Braz J Chem
(J mol
1) Eng 2004;21(4):545–56.
Ed activation energy of cat- 9.9 [10] Hartwich A, Marquardt W. Dynamic optimization of the load change of a large-
alyst deactivation scale chemical plant by adaptive single shooting. Comput Chem Eng 2010;34
(kJ kmol
1) (11):1873–89.
DH1, DH2 heat of the ethylene and 174.3,136.7 [11] Hartwich A, Schlegel M, Wu € rth L, Marquardt W. Adaptive control vector parame-
ethane generated terization for nonlinear model-predictive control. Int J Robust nonlin 2008;18
(8):845–61.
reaction (J mol
1)
[12] Huang W, Mccormick JR, Lobo RF, Chen JG. Selective hydrogenation of acetylene
hw interphase heat transfer 0.127
in the presence of ethylene on zeolite-supported bimetallic catalysts. J Catal
coefficient
2007;246(1):40–51.
(W m
1 K
1) [13] Karafyllis I, Kokossis A. On a new measure for the integration of process design
k1, k2, k3 reaction coefficients of 2.64 £ 103, 1.58 £ 102, and control: the disturbance resiliency index. Chem Eng Sci 2002;57(5):873–86.
the ethylene, ethane 2.29 £ 10
2 [14] Koller RW, Ricardez-Sandoval LA, Biegler LT. Stochastic back-off algorithm for
and green oil gener- simultaneous design, control and scheduling of multi-product systems under
ated reaction uncertainty. AIChE J 2018;64(13):2379–89.
kd, Kd catalyst deactivation 6.73 £ 10
3, 3.09 £ 10
3 [15] Kookos IK, Perkins JD. Control structure selection based on economics: generali-
coefficients zation of the back-off methodology. AIChE J 2016;62(9):3056–64.
Kp proportional gain of PID In Fig. 2 [16] Kuhlmann A, Bogle I. Controllability evaluation for nonminimum phase-processes
controller with multiplicity. AIChE J 2001;47(11):2627–31.
KI integration time of PID In Fig. 2 [17] Kuhn M, Lucas M, Claus P. Precise recognition of catalyst deactivation during acet-
controller ylene hydrogenation studied with the advanced Temkin reactor catalyst. Catal
Commun 2015;72(2):170–3.
KD differentiation time of In Fig. 2
[18] Luo XL, Liu JX, Xu F, Zuo X. Hydrogenation, two-dimensional dynamic modeling
PID controller
and analysis of acetylene hydrogenation reactor. CIESC J 2008;59(6):1454–61.
Ma, Mb, Mc acetylene, ethylene and 2.597 £ 103, 1.35 £ 103, [19] Luo XL, Xia CK, Sun L. Margin design, online optimisation, and control approach of a
hydrogen price coeffi- 1.685 £ 102 heat exchanger network with bypasses. Comput Chem Eng 2013;53(11):102–21.
cient [20] Mccue AJ, Mcritchie CJ, Shepherd AM, Anderson JA. Cu/Al2O3 catalysts modified
(RMB kPa
1 day
1) with Pd for selective acetylene hydrogenation. J Catal 2014;319:127–35.
[21] Mehta S, Ricardez-Sandoval LA. Integration of design and control of dynamic systems
(continued) under uncertainty: a new back-off approach. Ind Eng Chem Res 2016;55(2):485–98.
42 F.-M. Xie et al. / Journal of the Taiwan Institute of Chemical Engineers 108 (2020) 29
42

[22] Moayedi S, Davoudi A. Unifying distributed dynamic optimisation and control of [33] Shi J, Biegler LT, Hamdan I, Wassick JM. Optimisation of grade transitions in poly-
islanded dc microgrids. IEEE T Power Electr 2017;32(3):2329–46. ethylene solution polymerization process under uncertainty. Comput Chem Eng
[23] Morison KR, Sargent RWH. Optimisation of multistage processes described by dif- 2016;95:260–79.
ferential-algebraic equations. Num Anal 1986:86–102. [34] Soliman M, Swartz C, Baker R. A mixed-integer formulation for back-off under
[24] Narraway LT, Perkins JD. Selection of process control structure based on linear constrained predictive control. Comput Chem Eng 2008;32(10):2409–19.
dynamic economics. Ind Eng Chem Res 1993;32(11):2681–92. [35] Srinivasan B, Palanki S, Bonvin D. Dynamic optimisation of batch processes: I.
[25] Narraway LT, Perkins JD, Barton GW. Interaction between process design and process Characterization of the nominal solution. Comput Chem Eng 2003;27(1):1–26.
control: economic analysis of process dynamics. J Process Contr 1991;1:243–50. [36] Szukiewicz M, Kaczmarski K, Petrus R. Modeling of fixed-bed reactor: two models
[26] Pachulski A, Schodel R, Claus P. Kinetics and reactor modeling of a Pd-Ag/Al2O3 cata- of industrial reactor for selective hydrogenation of acetylene. Chem Eng Sci
lyst during selective hydrogenation of ethyne. Appl Catal 2012;445-446(6):107–20. 1998;53(1):149–55.
[27] Pollard GP, Sargent RWH. Off line computation of optimum controls for a plate [37] Teo KL, Jennings LS, Lee HWJ, Renboch V. The control parameterization enhancing
distillation column. Automatica 1970;6(1):59–76. transform for constrained optimal control problems. J Aust Math Soc Ser B Appl
[28] Rafiei-Shishavan M, Mehta S, Ricardez-Sandoval LA. Simultaneous design and Math 1999;40(3):314–35.
control under uncertainty: a back-off approach using power series expansions. [38] Tian L, Jiang D, Qian F. Simulation and optimisation of acetylene converter with
Comput Chem Eng 2017;99:66–81. decreasing catalyst activity. CIESC J 2012;63(1):185–92.
[29] Raghunathan AU, Gopal V, Subramanian D, Biegler LT, Samad T. Dynamic optimi- [39] Tian L, Jiang D, Qian F. Reactor system switch strategy for acetylene hydrogena-
sation strategies for three-dimensional conflict resolution of multiple aircraft. J tion process. CIESC J 2015;66(1):373–7.
Guid Contr Dynam 2004;27(4):586–94. [40] Tu F, Qing HY, Luo XL, Zuo X. Advanced process control of acetylene hydrogenation
[30] Rahimpour MR, Dehghani O, Gholipour MR, Yancheshmeh MS, Haghighi SS, Shariati reactor (I): construct dynamic model. Control Instrum Chem Ind 2003;30(1):20–4.
A. A novel configuration for Pd/Ag/a-Al2O3 catalyst regeneration in the acetylene [41] Weiss G. Modeling and control of the acetylene converter. J Process Contr 1996;6
hydrogenation reactor of a multi feed cracker. Chem Eng J 2012;198-199:491–502. (1):7–15.
[31] Ravanchi MT, Sahebdelfar S. Pd-Ag/Al2O3 catalyst: stages of deactivation in tail- [42] Xie FM, Xu F, Liang ZS, Luo XL, Shi FY. Full-cycle operation optimisation of acety-
end acetylene selective hydrogenation. Appl Catal, A 2016;525:197–203. lene hydrogenation reactor. CIESC J 2018;69(3):1081–91.
[32] Seider WD, Brengel DD, Provost AM, Widagdo S. Nonlinear analysis in process [43] Xu F, Jiang HR, Wang R, Luo XL. Influence of design margin on operation optimisation
design. Why overdesign to avoid complex nonlinearities? Ind Eng Chem Res and control performance of chemical processes. Chin J Chem Eng 2014;22(1):51–8.
1990;29(5):805–18.