Annals of Nuclear Energy 68 (2014) 193–203

Contents lists available at ScienceDirect

Annals of Nuclear Energy

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

Modeling and LQG/LTR control for power and axial power difference
of load-follow PWR core
Gang Li ⇑
School of Nuclear Science & Technology, Xi’an Jiaotong University, Xi’an 710049, China

a r t i c l e i n f o a b s t r a c t

Article history: The task of this investigation is to design a nonlinear Pressurized Water Reactor (PWR) core load follow-
Received 10 September 2013 ing control system for regulating the core power level and axial power difference. A two-point based non-
Received in revised form 10 January 2014 linear PWR core model without boron and with the power regulating rod and Axial Offset (AO) rod is
Accepted 13 January 2014
built. The linearized single-input single-output (SISO) or multiple-input multiple-output (MIMO) core
Available online 6 February 2014
model under Case 1 or Case 2 classified by two movable regions of power regulating rod is constructed.
The linear SISO or MIMO model is further augmented with additional integrators to cater to the control
Keywords:
design. The Linear Quadratic Gaussian with Loop Transfer Recovery (LQG/LTR) robust optimal control is
Load-follow PWR core
LQG/LTR
used to contrive a controller of the linear core model of each case. For Case 1 (Case 2), the nonlinear core
Robustness model and the LQG/LTR SISO (MIMO) controller construct the nonlinear PWR core load following control
SISO/MIMO control system. Two stability theorems are deduced to define that the nonlinear core load following control sys-
Stability analysis tem of each case is asymptotically stable. Finally, the control system of each case is simulated and the
simulation results show that the control system is effective.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction predictive control for the power and axial power difference of
load-follow PWR cores. However, the controllers in references
Developing load-follow operation of nuclear power plants (Khajavi et al., 2002; Khorramabadi et al., 2002; Chi and Cho,
(NPPs) is an inevitable trend for electricity generation. Some 2002; Torabi et al., 2011) are used to control the reactor power
researchers (Meyer et al., 1978; Chari and Rohr, 1997) have worked and the axial power difference is ignored; in references (Busi,
on the load following operation of NPPs. Given that xenon-induced 1981; Lin and Shen, 2000; Na et al., 2005), controlling tasks are
power oscillations occur, controlling the reactor power according achieved by changing the boron concentration and control rod
to practical or predictable load requirements on an electric net- speeds, this introduction of boron must produce lots of liquid
work and regulating the axial power difference within a certain waste and increase the control difficulty; for the control rod model
boundary are two important criteria in load following operation in the reference (Eliasi et al., 2011), the power regulating rod is
of NPPs. Therefore, it is necessary to improve the control technique only considered, this is different from the popular rod model with
of reactor power and axial power difference in the load-follow the power regulating rod and the Axial Offset (AO) regulating one,
mode for the safety, operability and availability of NPPs. and increases design difficulty of controller. Based on the consider-
Continuous efforts have been made to improve the load follow- ations in the paper, new control strategies with the Linear
ing operation of NPPs. Khajavi et al. (2002) and Khorramabadi et al. Quadratic Gaussian with Loop Transfer Recovery (LQG/LTR) design
(2002) designed the intelligent controllers to provide the tight con- and stability analysis are utilized to devise a nonlinear PWR core
trol of nuclear reactors; Chi and Cho (2002) contrived a H1 control load following control system for the core power level and axial
system of a Pressurized Water Reactor (PWR) core; Torabi et al. power difference, which include the core modeling with single-
(2011) devised a PWR core power control system by using the input single-output (SISO) and multi-input multi-output (MIMO)
quantitative feedback theory. Busi (1981) designed a nuclear reac- plants of ignoring boron concentration changes and selecting the
tor control system for the reactor coolant temperature and axial power regulating rod and AO rod as the input mechanism.
shape index by using the multivariable frequency response tech- The two-point based model of nonlinear PWR core is built in the
nique; Lin and Shen (2000) presented the neurocontrol technique, paper. When the power regulating rod and AO rod are in a same
Na et al. (2005) and Eliasi et al. (2011) proposed the model point (Case 1), the linear SISO model is constructed; when the
power regulating rod and AO rod in the core are not in a same point
(Case 2), the linear MIMO model is built. For Case 1, the change of
⇑ Tel.: +86 18292573638; fax: +86 029 82667802. the power regulating rod position is an input and the change of
E-mail address: [email protected]

0306-4549/$ - see front matter Ó 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.anucene.2014.01.022
194 G. Li / Annals of Nuclear Energy 68 (2014) 193–203

8 dT P
core power level is an output; for Case 2, the change of the power >
>
f1
¼ lf 10 Pr1  lX1 ðT f 1  T C11 Þ
regulating rod position and one of the AO rod position are two in-
< dt f1 f1
dT C11
¼ lX1 ðT f 1  T C11 Þ  s21 ðT C11  T 1Cin Þ ð2Þ
puts, the change of core power level and one of axial power differ- >
>
dt c1
: dT C12 X1
ence are two outputs. In addition, the linear SISO or MIMO model is dt
¼ l ðT f 1  T C11 Þ  s21 ðT C12  T C11 Þ
c1
further augmented with additional integrators to cater to the LQG/
LTR design.
q1 ¼ qrod1 þ af 1 ðT f 1  T f 10 Þ þ a2C1 ðT C11  T C110 Þ þ a2C1 ðT C12  T C120 Þ
þaX1 X f 10 ðX r1  X r10 Þ
The LQG/LTR control is utilized to design controllers of linear
SISO and MIMO models of the core. Major advantages of LQG/LTR ð3Þ
controller are that it possesses the strong robustness and can be 8
> q2 b
Xg
designed for single-variable or multi-variable plants. The strategy >
>
dP r2
¼ P þ bi cr2i
 Pr21
> dt
> K2 r2 K2
has been developed from the Linear Quadratic Gaussian (LQG) con- >
> i¼1
< dc
trol (Athans, 1971), which is an integrated design way based on the dt
r2i
¼ ki Pr2  ki cr2i ; i ¼ 1; 2;    ; g ð4Þ
linear quadratic regulator (LQR) and a state observer. However, the >
> r2 ¼ cI Pf 20 Pr2  kI Ir2
dI
>
>
introduction of a state observer in designing a LQG controller >
>
dt Eeff V 2 If 20
 
> dXr2
: cX Pf 20 kI r P
weakens the robustness of stability and performances of LQR that dt
¼ E V2X
Pr2 þ XI f 20 Ir2  kX þ E XRf 20V 2 Pr2 X r2
eff f 20 f 20 eff f
usually has an infinite gain margin and a phase margin from [60°,
1) (Safonov and Athans, 1977). To improve the robustness of a
8 dT P
>
>
>
f2
¼ lf 20 Pr2  lX2 ðT f 2  T C21 Þ
LQG control system, Athans (1986), Stein and Athans (1987) and < dt f2 f2

dT C21
Doyle and Stein (1981) have proposed and developed the LTR tech- ¼ lX2 ðT f 2  T C21 Þ  s22 ðT C21  T 2Cin Þ ð5Þ
>
>
dt c2
nology based on LQG. The method can recover the robustness of a >
: dT C22 ¼ X2 ðT  T Þ  2 ðT  T Þ
lc2 f 2 C21 s2 C22 C21
LQG/LTR control system of a model in the light of a robustness of dt

Target Feedback Loop (TFL) of the model. Therefore, this robust

control is suitable to design control systems of nonlinear plants q2 ¼ qrod2 þ af 2 ðT f 2  T f 20 Þ þ a2C2 ðT C21  T C210 Þ þ a2C2 ðT C22  T C220 Þ
such as the core by using the controller based on the linearized þaX2 X f 20 ðX r2  X r20 Þ
core model to control the nonlinear core model. ð6Þ
For Case 1 (Case 2), the nonlinear core model and the LQG/LTR
where model parameters in Eqs. (1)-(6) are shown in Table 1.
SISO (MIMO) controller construct the nonlinear PWR core load fol-
 Coupling model
lowing control system. Based on four lemmas, one criterion and
Point 1 and point 2 are coupled together through neutron
one definition (Rosenbrock, 1974; Bai and Pang, 1988; Isidori,
diffusion.
1995; Liu and Tang, 2007; Dorf and Bishop, 2009), two stability  
theorems are deduced and adopted to define that the nonlinear J 12 Sa 2J 2 U1  U2 4dv s
n12 ¼ ¼ 12 ¼ d ¼ 2 ðn1  n2 Þ ð7Þ
core load following control system under each case is asymptoti- Sa ðh=2Þ h h h=2 h
cally stable. where n12 – the change of neutron density of point 1 caused by neu-
Finally, the control system of each case is simulated and conclu- tron diffusion from point 1 to point 2, m3; ni – average neutron
sions are drawn. density of point i (i = 1, 2), m3; J12 – the density of neutron current
from point 1 to point 2, (m2 s)1; Sa – the equivalent cross section
2. Model PWR core area of core, m2; h – the core height, m; d – the diffusion coefficient
from point 1 to point 2, m; Ui – neutron flux of point i (i = 1, 2),
2.1. Nonlinear model (m2 s)1; vs – neutron speed, m/s; 4dvs/h2 – coupling coefficient.
Let F = 4dvs/h2, then
The two-point based model of the nonlinear PWR core is devel-
P12 ¼ FðP1  P2 Þ ð8Þ
oped. In modeling this core, it is divided into two equal halves
(bottom and top). The bottom (top) half core is treated as point 1 where P12 – the change of core power of point 1 due to neutron dif-
(point 2). Point 1 and point 2 are coupled together through neutron fusion, W; Pi – the core power of point i (i = 1, 2), W.
diffusion. Eq. (8) is normalized by
Pf 20
 Models of two points Pr12 ¼ FPr1  F Pr2 ð9Þ
Pf 10

According to the lumped parameter method, point 1 and point 2 Similarly, Eq. (10) is available.
are respectively modeled. Each point core model is established via Pf 10
Pr21 ¼ FPr2  F Pr1 ð10Þ
using the point kinetics equations with six groups of delayed neu- Pf 20
trons and reactivity feedbacks due to changes in fuel temperature,
 Control rod model
coolant temperature and xenon concentration (Schultz, 1961;
Kerlin et al., 1978; Edwards et al., 1990; Eliasi et al., 2011). The For this core, the power regulating rod is called by Mrod and the
model expressions of point 1 are in Eqs. (1)-(3), and ones of point AO rod denoted by AOrod. The block diagram of core control rod is
2 are in Eqs. (4)-(6). shown in Fig. 1, where Case 1 is that Mrod moves in point 2 and
8 AOrod does in an AO band of point 2; Case 2 is that Mrod moves
>
> Xg
>
>
dP r1
¼ q1 b
P þ bi cr1i
 Pr12 in point 1 and AOrod does in an AO band of point 2.
>
> dt K1 r1 K1
For Case 1, the control rod model is expressed by
>
> i¼1
>
< dcr1i 
¼ ki Pr1  ki cr1i ; i ¼ 1; 2;    ; g dqrod1 ¼ 0
dt ð1Þ ð11Þ
>
> dqrod2 ¼ aMrod dMrod þ aAOrod dAOrod
> r1 ¼ cI Pf 10 Pr1  kI Ir1
>
dI
>
> dt Eeff V 1 If 10
>
> For Case 2, the control rod model is denoted by
>
: dXr1 ¼ cX Pf 10 P r1 þ kI If 10 Ir1  ðkX þ rX Pf 10 Pr1 ÞX r1
dt E V1X
eff f 10 f 10X E R V1
eff f 
dqrod1 ¼ aMrod dMrod
ð12Þ
dqrod2 ¼ aAOrod dAOrod
 G. Li / Annals of Nuclear Energy 68 (2014) 193–203 195

Table 1
Model parameters in Eqs. (1)-(6).

cr1i, cr2i ith group normalized precursor concentration

Eeff Useful energy liberated per fission of 235U, J
g Delayed neutron group number, g = 6
Iri Normalized iodine concentration of point i (i = 1, 2)
Ifi0 Iodine concentration of point i (i = 1, 2) at the full core power, m3
Pfi0 Full core power of point i (i = 1, 2), W
Pri Normalized core power of point i (i = 1, 2)
Pr12 The change of Pr1 caused by neutron diffusion from point 1 to point 2
Pr21 The change of Pr2 caused by neutron diffusion from point 2 to point 1
Tfi Fuel average temperature of point i (i = 1, 2), °C
TC11 Coolant average temperature of point 1, °C
T1Cin Coolant inlet temperature of point 1, °C
TC12 Coolant outlet temperature of point 1, °C
T2Cin Coolant inlet temperature of point 2, °C, T2Cin = TC12
TC21 Coolant average temperature of point 2, °C
TC22 Coolant outlet temperature of point 2, °C
Tfi0 Fuel average temperature of point i (i = 1, 2) at the initial moment, °C
TC110 Coolant average temperature of point 1 at the initial moment, °C
TC120 Coolant outlet temperature of point 1 at the initial moment, °C
TC210 Coolant average temperature of point 2 at the initial moment, °C
TC220 Coolant outlet temperature of point 2 at the initial moment, °C
Vi Core volume of point i (i = 1, 2), m3
Xri Normalized xenon concentration of point i (i = 1, 2)
Xfi0 Xenon concentration of point i (i = 1, 2) at the full core power, m3
Xri0 Normalized xenon concentration of point i (i = 1, 2) at the initial moment
Ki Neutron generation time of point i (i = 1, 2), s
Rf Macroscopic thermal neutron fission cross section, m1
Xi Coefficient of heat transfer between fuel and coolant of point i (i = 1, 2), W/°C
afi Fuel temperature coefficient of reactivity of point i (i = 1, 2), °C1
aci Coolant temperature coefficient of reactivity of point i (i = 1, 2), °C1
aX1 Xenon coefficient of reactivity of point i (i = 1, 2)
b Effective delayed neutron fraction
bi ith Group effective delayed neutron fraction
cI (cX) Iodine (xenon) yield per fission
ki ith Delayed neutron group decay constant, s1
kI (kX) Iodine (xenon) decay constant, s1
lfi Fuel total heat capacity of point i (i = 1, 2), J/°C
lci Total heat capacity of coolant of point i (i = 1, 2), J/°C
qi Total reactivity of point i (i = 1, 2)
qrodi reactivity due to control rod movement in point i (i = 1, 2)
rX Microscopic thermal neutron absorption cross section of xenon, m2
si Residence time between both coolant nodes in point i (i = 1, 2), s

where AO – Axial Offset; Pf0 – the full core power, W; Pr – normal-

ized core power (power level); DI – the axial power difference (the
normalized AO).

2.2. Linearized model

One-group delayed neutron model is adopted and the coolant

inlet temperature is treated as a constant. The small perturbation
linearization methodology is utilized to linearize the nonlinear
core model. A variable parameter in Section 2.1 is denoted by Vp,
Vp e {i = 1, 2| Pri, Pr12, Pr21, cri, Iri, Xri, Tfi, TC11, TC12, TC21, TC22, qi, qrodi,
Pr, DI}. According to the small perturbation linearization methodol-
ogy, Vp = Vp0 + dVp, where Vp0 – the initial steady-state value of a
parameter; d – the change of this parameter from Vp0.

Fig. 1. Block diagram of core control rod.  Case 1 corresponding to SISO model

where aMrod – the reactivity coefficient of Mrod; aAOrod – the reactiv- From the level of control technology, Case 1 is different from
ity coefficient of AOrod; dMrod (dAOrod) – the position change of Case 2. For Case 1, dqrod1 = 0 and dPr1 is very small, which is caused
Mrod (AOrod) in the core, in units of fraction of core length. by neutron diffusion and reactivity feedbacks due to changes in
 Output model fuel temperature, coolant temperature and xenon concentration
of point 1. According to Eq. (14), dPr and dDI are generated mainly
AO ¼ ðPf 20 Pr2  Pf 10 P r1 Þ=ðPf 20 Pr2 þ Pf 10 Pr1 Þ ð13Þ
by dPr2, and dDI is also close to dPr. So controlling dPr as a direct
8 controlled output is equal to control dDI as an indirect controlled
< Pr ¼ PPf 10 Pr1 þ PPf 20 Pr2 state.
f0 f0
ð14Þ Since dPr1 is very small, regulating dPr by a controller is virtually
: DI ¼ AO  P ¼ Pf 20 P  Pf 10 P
r P r2 P r1
f0 f0 to regulate dPr2. Besides, regulating dPr2 can be carried out by only
196 G. Li / Annals of Nuclear Energy 68 (2014) 193–203

Table 2 matrix; Bm – the R142 input matrix; Cm – the R214 output matrix;
Parameters of G. Dm – the R22 zero matrix.
[a13 a12 . . . a1 a0] There exists a coupled relationship between dMrod and dDI (or
[b14 b13 . . . b1 b0] dAOrod and dPr) in Eq. (17). The goal of devising a controller is to
G [2417 1.054e6 2.125e7 1.718e8 6.904e8 1.422e9 1.442e9 7.516e8 decouple the two relationships.
1.922e8 1.89e7 4310 0.3196 1.254e5 1.578e10] G and GG at the power level 100% are calculated by using
[1 836.3 1.833e5 3.601e6 2.916e7 1.201e8 2.645e8 3.098e8 1.976e8 parameters from (IAEA, 1971) in which the total primary heat out-
6.557e7 8.98e6 1413 0.1096 4.267e6 8.235e11]
put is 2200 MW; the primary coolant inlet temperature is 285 °C;
the primary coolant outlet temperature is 317 °C; the primary
coolant average pressure is 15.5 Mpa and the primary coolant mass
Table 3 flow is 12861.1 kg/s. Parameters of G and GG are given in Tables 2
Parameters of GG. and 3, where j is an integer and 1ej = 10j.
GG [c13 c12 . . . c1 c0]
[d14 d13 . . . d1 d0]
2.3. Augment models for LQG/LTR design
g11 [2671 1.165e6 2.344e7 1.891e8 7.56e8 1.507e9 1.285e9 5.354e8
1.092e8 8.643e6 1192 0.03924 4.249e6 8.05e11]
[1 836.3 1.833e5 3.601e6 2.916e7 1.201e8 2.645e8 3.098e8 1.976e8
A plant is augmented with additional integrators proposed by
6.557e7 8.98e6 1413 0.1096 4.267e6 8.235e11] Athans (1986) to design an improved LQG/LTR controller of the
g12 [9405 4.103e6 8.268e7 6.685e8 2.687e9 5.536e9 5.612e9 2.925e9 plant which can remove the steady state error for input commands
7.48e8 7.357e7 1.677e4 1.244 4.881e5 6.143e10] such as a step. The augmented part of a model is shown in Fig. 2,
[The same as denominator of g11]
where G is with an additional integrator (1/s) and GG with two
g21 [2671 1.124e6 2.257e7 1.821e8 7.342e8 1.564e9 1.845e9
1.101e9 3.095e8 3.242e7 8468 0.7268 2.426e5 2.604e10] additional integrators; daddi (i = 1, . . ., 3) is a new input of an aug-
[The same as denominator of g11] mented model.
g22 [9405 3.956e6 7.965e7 6.44e8 2.591e9 5.352e9 5.456e9 2.86e9 7.352e8 An augmented model of G (GG) is denoted by Ga (GGa). The state
7.259e7 1.647e4 1.215 4.798e5 6.068e10]
equations of Ga and GGa are in following equation:
[The same as denominator of g11]

x_ ¼ Ax þ Bu
ð20Þ
moving Mrod. To ease the maneuver burden of AOrod, let AOrod y ¼ Cx þ Du
motionless and Mrod moveable in modeling the core for Case 1
where for Case 1, A = [Am Bm;0114 011], B = [0141;1], C = [Cm 0],
(dMrod – 0, dAOrod = 0).
D = Dm, y = dPr; for Case 2, A = [Am Bm;0214 022], B = [0142;1 0;0
For case 1, the transfer function and the state equation of the
1], C = [Cm 022], D = Dm, 0ij – a i  j zero matrix, y = [dRPr dRDI]T.
core are respectively calculated and represented by
A LQG/LTR controller GOC will be designed in the light of Ga or
P13 i
dPr i¼0 ai s
GGa. Hence, GOC with 1/s is a SISO (MIMO) controller of the linear
G¼ ¼ P14 ð15Þ or nonlinear core model of Case 1 (Case 2).
dMrod i¼0 bi s
i


x_ m ¼ Am xm þ Bm um
ð16Þ 3. LQG/LTR robust control
y m ¼ C m xm þ D m u m
where um = dMrod – the input; ym = dPr – the output; ai(i = 0, 1, . . ., 3.1. Object model
13) – numerator coefficients; bi(i = 0, 1, . . ., 14)-denominator coeffi-
cients; xm = [dPr1, dcr1, dIr1, dXr1, dTf1, dTC11, dTC12, dPr2, dcr2, dIr2, dXr2, In the LQG/LTR control context, the state equation of Ga or GGa is
dTf2, dTC21, dTC22]T – the 14  1 state array; Am – the R1414 system denoted by
matrix; Bm – the R141 input matrix; Cm – the R114 output matrix; 
x_ ¼ Ax þ Bu þ n
Dm – the R11 zero matrix. ð21Þ
 Case 2 corresponding to MIMO model y ¼ Cx þ Du þ N

For case 2, the transfer function and the state equation of the where n = CH – the object noise, where C is a real matrix and H is
core are respectively calculated and represented by the random process noise; N – the random measurement noise; H
       and N as zero-mean white noises are stable and mutually
dPr dMrod g 11 g 12 dMrod
¼ GG ¼ ð17Þ independent.
dDI dAOrod g 21 g 22 dAOrod
P13 i
i¼0 c i s
g ij ¼ P14 ði; j ¼ 1; 2Þ ð18Þ
i
i¼0 di s


x_ m ¼ Am xm þ Bm um
ð19Þ
y m ¼ C m xm þ D m u m
where um = [dMrod dAOrod]T – the 2  1 input array, dMrod is the
first input and dAOrod is the second one; ym = [dPr dDI]T – the
2  1 output array, dPr is the first output and dDI is the second
one; GG – the 2  2 transfer function matrix; g11 = dPr/dMrod,
g12 = dPr/dAOrod, g21 = dDI/dMrod, g22 = dDI/dAOrod; ci(i = 0, 1, . . .,
13) – numerator coefficients; di(i = 0, 1, . . ., 14)-denominator coeffi-
cients; xm = [dPr1, dcr1, dIr1, dXr1, dTf1, dTC11, dTC12, dPr2, dcr2, dIr2, dXr2,
dTf2, dTC21, dTC22]T – the 14  1 state array; Am – the R1414 system Fig. 2. Augmented part of a model.
 G. Li / Annals of Nuclear Energy 68 (2014) 193–203 197

Fig. 3. LQG/LTR control simple schematic of Ga or GGa.

3.2. LQG/LTR controller After defining C, Q0 and R0 that are decent, contriving the KF in
Fig. 3 is to solve Kf for making Pf meet Eq. (22) and converge
There are two problems to be solved for the LQG/LTR control. min{Pf}. Kf solved is expressed by
They are the Kalman filter (KF) estimation problem with distur-
bances and the linear quadratic regulator (LQR) problem that is a K f ¼ Pf C T R1
0 ð23Þ
state feedback optimal control problem. The LQG/LTR control simple  Solve LQR
schematic of Ga or GGa is shown in Fig. 3, where for Ga, yr = dRPr – the
reference input, dRPr is a desired change of power level; for GGa, After choosing appropriate Q and R, the LQR problem can be
yr = [dRPr dRDI]T – the 2  1 reference input array, dRPr is a desired handled via minimizing the optimal performance index
change of power level and dRDI denotes one of axial power differ- Z tf
ence; GOC is a LQG/LTR controller; Kf is the gain matrix of KF and Kc J ¼ lim E ½xT Qx þ uT Rudt ð24Þ
t f !1 0
is the state feedback gain matrix of LQR. The process of solving Kf,
Kc is the following. where Q – a weighting matrix of state; R – a weighting matrix of
input.
 Solve KF The optimal control input u ¼ K c ^
x can be obtained by mini-
mizing the index. Kc can be solved by
The estimation of x is ^ x and the estimate error is e ¼ x  ^ x.
Covariances of H, N and e are denoted by Q0 = E(HHT), R0 = E(NNT) K c ¼ R1 BT Pc ð25Þ
and Pf = E(eeT). Q0 and R0 are the parameters chosen subjectively, Pf
where Pc – a symmetric positive semidefinite matrix, which is the
is a measurement of estimating effect. The smaller Pf, the better the
solution of the Riccati equation
estimating effect; Pf is the symmetric positive semidefinite matrix
that meets the Riccati equation represented by AT Pc þ P c A  Pc BR1 BT Pc þ Q ¼ 0 ð26Þ
T
Pf A þ AP f  Pf C T R1
0 CP f þ CQ 0 C ¼ 0 T
ð22Þ According to the separation principle about the problems of
LQG/LTR control, Kf and Kc can be designed respectively.
GOC is expressed by

x 10
5 Nyquist Diagram Bode Diagram
2 150

N1 B1
1.5 N2 100 B2
Magnitude (dB)

1 50

0.5 0
Imaginary Axis

0 -50
-60

-0.5
Phase (deg)

-1
-90

-1.5

-2 -120
0 5000 10000 -5 0
10 10
Real Axis Frequency (rad/sec)

Fig. 4. Nyquist and Bode plots of Hf and GaGOC.

198 G. Li / Annals of Nuclear Energy 68 (2014) 193–203

120 multi-variable plant. Q = q2C Tm Cm where q2 is the recovery gain. R

SV1 is a real or a real matrix such as I.
100 SV2 Increasing q1 (q2) makes GOCGa (GaGOC) sufficiently approach Hc
(Hf) after choosing other free parameters in Method 1 (or Method
80
2).
Singular value (dB)

60 For a single-variable plant such as Ga, the robustness of a LQG/

LTR control system of Ga can be recovered well by designing a KF (a
40 LQR) to make Nyquist and Bode curves of GOCGa (GaGOC) approxi-
mate the corresponding curves of Hc (Hf) in a defined frequency
20
range. For a multi-variable plant such as GGa, the verifiable curves
0 are singular value curves instead of Nyquist and Bode curves.

-20
4. Design control systems
-40
Method 2 is used to design GOC for Case 1 and Case 2. According
-60 to Section 3, GOC is calculated by programming after in turn defin-
-6 -4 -2 0 2
10 10 10 10 10 ing C, Q0 and R0 to solve Kf and Q, R to solve Kc.
Frequency (rad/sec) For Ga, C = [CH CL]T where CL = [CmA1 m Bm]
1
, CH = C Tm [Cm-
C Tm ]1, Q0 = 1, R0 = 8; q2 = 1e8, Q = q2C Tm Cm, R = 10. Kf and Kc solved
Fig. 5. Singular value curves of Hf and GGaGOC.
are expressed by
Kf = [0.0257 0.0067 5.3664e7 3.1737e6 1.9182 0.0341
0.0617 1.8166 0.4918 4.3194e5 2.3098e4 129.3269 2.2437
GOC ¼ K c ðsI  A þ BK c þ K f CÞ1 K f ð27Þ
4.0532 0.1845]T;
where I stands for an identity matrix with a dimension and is used Kc = [1.9771e3 370.7071 2.3307e5 691.3108 1.85
some times below. 14.1329 14.1374 1.7888e3 335.4018 1.9082e5 616.5159
Ga (GGa) is also represented by Ga = CUB (GGa = CUB), where 1.6962 17.6350 17.616 4.2663e3].
U = (sI  A)1. In Fig. 3, there are four additional loop-breaking TFL of Ga is Hf. Nyquist and Bode plots of Hf and GaGOC are pre-
points i, i0 , ii and ii0 . i and ii in turn locate at the plant output and sented in Fig. 4, where N1 and B1 are the Nyquist plot and Bode
input, i0 and ii0 are at the input of Kf and the output of Kc respec- plot of Hf, N2 and B2 are the Nyquist plot and Bode plot of GaGOC.
tively. The return ratio function at i0 is the KF loop transfer function As be shown in Fig. 4, N2 sufficiently approaches N1 and B2 ade-
denoted by Hf = CUKf, and one at i is denoted by GaGOC (GGaGOC). quately approximates B1.
The return ratio function at ii0 is the LQR loop transfer function de- For GGa, C = [CH CL]T where CL = [CmA1 m Bm]
1
, CH = C Tm [Cm-
noted by Hc = KcUB, and one at ii is denoted by GOCGa (GOCGGa). C m ] , Q0 = diag([1 1]), R0 = 8 ⁄ diag([1 1]); q2 = 1e8, Q = q2C Tm Cm,
T 1

R = 10 ⁄ diag([1 1]), where daig() denotes a diagonal matrix. Kf

 Method 1: LQG/LTR design at the plant input and Kc solved are as follows.
Kf = [1.0382 0.6076;0.2184 0.234;1.9808e5
Hc as TFL of Ga (GGa) should be designed at first by solving Kc to 1.9809e5;1.1713e4 1.1716e4;63.0591 68.8237;1.0733
make Hc possess a desirable robustness. Then defining the final 1.3286;1.9156 2.4934;1.277 0.8011;0.2771 0.2311;2.1894e5
controller GOC through devising Kf makes GOCGa (GOCGGa) suffi- 2.1894e5;1.1708e4 1.1708e4;71.7807 60.9676;2.6403
ciently approach Hc. 1.0916;3.2318 0.108;0.0981 0.0699;0.0338 0.0173];
Kc = [2.886e3 426.0275 2.1088e5 794.4586 2.1259
 Method 2: LQG/LTR design at the plant output 16.2417 16.2280 3.7588 0.1461 6.5685e13 0.2686
7.3863e4 0.0077 0.0077 5.4189e3 7.2335;8.0148 0.5687
Hf as TFL of Ga (GGa) should be designed at first by solving Kf to 3.3788e12 1.0605 0.0028 0.0217 0.0267 2.7878e3
make Hf possess a desirable robustness. Then defining the final 212.2527 4.9031e6 390.1284 1.0731 11.1596 11.1547
controller GOC through devising Kc makes GaGOC (GGaGOC) suffi- 7.2335 1.0507e4].
ciently approach Hf. TFL of GGa is Hf. Singular value curves of Hf and GGaGOC are pre-
sented in Fig. 5, where SV1 denotes the singular value curves of Hf;
SV2 ones of GGaGOC. As be shown in Fig. 5, SV2 sufficiently ap-
3.3. Free parameters for LTR
proaches SV1.
For Case 1 (Case 2), the nonlinear core model and the LQG/LTR
C, Q0, R0, Q and R are free parameters which can be regulated
SISO (MIMO) controller (with 1/s) construct the nonlinear PWR
subjectively according to a design effect. Generally speaking, these
core load following control system.
parameters are defined as follows.
For Method 1, regulating Q and R is separately to improve con-
trol performances of a system and to limit the system control en- 5. Stability analysis of control system
ergy. Usually, C = B, Q0 = Q00 + q1I or C = I, Q0 = Q00 + q1BBT, where
Q00 is a real or a real matrix and q1 is the recovery gain. R0 is a real It is important to analyze the stability of the nonlinear core load
or a real matrix such as I. following control system under Case 1 or Case 2 from the stability
For Method 2, C is the real matrix selected subjectively for a theory level. Theorems 1 and 2 are deduced to analyze the stability
single-variable plant, such as C = B or I. For a multi-variable plant, of the two control systems, Theorem 1 can be used for Case 1 or
C is usually defined according to the empirical formula denoted by Case 2 and Theorem 2 can be used for Case 2.
C = [CH CL]T, where CL = [C m A1m Bm]
1
, CH = C Tm [CmC Tm ]1 (Athans, For Case 1, the transfer function and the state equation of the
1986); CL = [Dm  CmAm Bm] , CH = Am Bm[CmA1
1 1 1
m Bm]
1
[I  DmCL] control system are expressed respectively by H1 and (A1 B1 C1
(Brain and Eduardo, 1995). Q0 = I. R0 is a real matrix, but it is con- D1), where H1 is denoted by Eq. (28) and A1 is the system matrix.
nected with crossover frequencies of singular value curves of a For Case 2, the transfer function and the state equation of the con-
 G. Li / Annals of Nuclear Energy 68 (2014) 193–203 199

trol system are expressed respectively by H2 and (A2 B2 C2 D2), Theorem 1. The nonlinear core load following control system Hn1
where H2 is denoted by Eq. (29) and A2 is the system matrix. (Hn2) is asymptotically stable on the sufficient condition that the pair
(A, B) in Ga (GGa) is controllable, and for the control system H1 (H2),
H1 ¼ ðGa GOC Þ=ð1 þ Ga GOC Þ ð28Þ
there exists a positive define real symmetric matrix N , such that
H2 ¼ ðGGa GOC Þ=ðI þ GGa GOC Þ ð29Þ
ATi N þ NAi < 0; i ¼ 1; 2 ð31Þ
For Case 1, let the combined model of the nonlinear core model
at the power level corresponding to G and one integrator be de-
noted by Gna; for Case 2, let the combined model of the nonlinear Proof. Know that for the control system z_ ¼ Ai z, there is a positive
core model at the power level corresponding to GG and two inte- define real symmetric matrix N, such that ATi N þ NAi < 0. Take
grators be represented by GGna; Gna or GGna is called by a nonlinear V(z) = zTNz with V(0) = 0 and V(z) ? 1 as kzk ? 1 as a Lyapunov
core model. function of H1 or H2, then for z – 0, V(z) is positive define and
If the linear plant Ga (GGa) in H1 (H2) is replaced by the nonlin- _
VðzÞ ¼ z_ T Nz þ zT Nz_ ¼ ðAi zÞT Nz þ zT NAi z
ear core model Gna (GGna) at the power level corresponding to Ga ð32Þ
(GGa), then the linear core control system H1 (H2) is converted into ¼ zT ðATi N þ NAi Þz < 0; i ¼ 1; 2
the nonlinear core control system denoted by Hn1 (Hn2). Namely, According to Lemmas 2 and 3, the control system H1 (H2) is asymp-
Hn1 (Hn2) consists of Gna (GGna) and GOC. Ga (GGa) is the linear totically stable in a large range. Under the condition that the pair (A,
approximation of Gna (GGna). B) in Ga (GGa) is controllable according to Criterion 1 and H1 (H2) is
asymptotically stable, Hn1 (Hn2) is also asymptotically stable in the
5.1. Stability theories light of Lemma 1. h

In accordance with the literature (Isidori, 1995), Lemma 1 is Besides, the following theories are used to deduce Theorem 2 to
shown to define the stability of the nonlinear closed control system analyze the stability of the nonlinear core control system Hn2 under
Hn1 or Hn2. Case 2.
The inverse Nyquist array (INA) proposed by Rosenbrock (1974)
Lemma 1. Suppose the linear approximation of Gna (GGna) is is a diagonal dominance multivariable frequency method. According
asymptotically stabilizable, or the pair (A, B) in Ga (GGa) is control- to INA, the pre-decoupling controller that is constant or dynamic is
lable, then the controller GOC which asymptotically stabilizes the linear firstly designed to make the resulting forward path-transfer func-
approximation Ga (GGa) is also able to asymptotically stabilize the tion matrix of an original MIMO plant be diagonally dominant,
original nonlinear system Gna (GGna); or if H1 (H2) is asymptotically namely reduce parameter interactions of this plant as much as pos-
stable, then Hn1 (Hn2) is also asymptotically stable. sible. Then for improving dynamic performances of the decoupled
plant, a dynamic controller matrix of the plant is contrived by the
single-loop control technique such as the PID control. In the paper,
Criterion 1 Dorf and Bishop, 2009. For the system Ga or GGa, the the dynamic controller GOC can be regarded as the pre-decoupling
controllability matrix Pc of Ga or GGa is expressed by Eq. (30). If one of GGa and can undertake the decoupling control and dynamic
Pc is nonsingular, then Ga or GGa is controllable. control simultaneously. Hence, the design of a dynamic controller
matrix for the decoupled plant GGa by the use of GOC is ignored.
Pc ¼ ½ B AB . . . Ai2 B Ai1 B ð30Þ
Suppose that Dny is a large contour in the complex plane con-
1515 1632 sisting of the imaginary axis from s = ir to s = ir, together with a
where Pc – a R (R ) matrix and i = 15 (i = 16) for Case 1
(Case 2). semicircle of radius r in the right half plane. Let Qo(s) = (GGaGOC)-
For Case 1, if Pc is nonsingular, then Ga is controllable; for Case
1
= [q11(s) q12(s); q21(s) q22(s)], Qo e C22 and s e Dny.
2, if there exists a R1616 nonsingular matrix in Pc, then GGa is If there exists Eq. (33) (or Eq. (34)), Qo is row (or column) diag-
controllable. onal dominance.
According to Lyapunov direct method (Liu and Tang, 2007), X
2

there are Lemma 2 and Lemma 3 shown as follows. The two lem- jqii ðsÞj > dr i ðsÞ ¼
qij ðsÞ
; i ¼ 1; 2 ð33Þ
mas can be used to analyze the stability of H1 or H2. In addition, j¼1;j–i
Lemma 3 is the basis of deducing Theorem 1 to seek a Lyapunov
function of H1 or H2. X
2

jqii ðsÞj > dci ðsÞ ¼

qji ðsÞ
; i ¼ 1; 2 ð34Þ
j¼1;j–i
Lemma 2. For i = 1 or 2 (namely Case 1 or Case 2), consider the closed
control system described by Hi or z_ ¼ wðzÞ ¼ Ai z with w(0) = 0. If one of the two diagonal dominances of Qo is constructed, Qo is
Suppose that there exists a scalar function V(z) such that V(0) = 0; a diagonal dominance system, namely a decoupling control sys-
_
V(z) > 0 and VðzÞ < 0 for z – 0; V(z) ? 1 as kzk ? 1, then V(z) is a tem. This indicates that dPr can only be regulated by dMrod and
Lyapunov function and the equilibrium state ze = 0 is asymptotically is not affected by dAOrod during dPr following dRPr; dDI can only
stable in a large range. be regulated by dAOrod and is not affected by dMrod during dDI fol-
lowing dRDI. Namely, decoupling tasks by the introduction of GOC
to GGa are implemented.
Lemma 3. For i = 1 or 2 (namely Case 1 or Case 2), consider the closed
control system described by z_ ¼ Ai z, then the equilibrium state ze = 0 is
 Graphical criteria of diagonal dominance
asymptotically stable in a large range on the sufficient and necessary
condition that for a positive define real symmetric matrix q given
The diagonal dominance of Qo can be checked by analyzing the
discretionarily, there must be a positive define real symmetric matrix
row or column Gershgorin band.
N such that ATi N þ NAi ¼ q, namely ATi N þ NAi < 0.
The first stability theorem of the nonlinear core load following Definition 1 Rosenbrock, 1974. (Gershgorin band) Select a point s
control system under Case 1 or Case 2 is proposed as Theorem 1 on Dny and the corresponding point in the complex plane of qii(s) is
based on Criterion 1 and Lemmas 1, 2, 3. seen as a center, draw a circle with this radius dri(s) (or dci(s)). As s
200 G. Li / Annals of Nuclear Energy 68 (2014) 193–203

goes round Dny, the corresponding circles sweep out a band, which Table 4
is represented by a finite set of circles. This band is called by the Eigenvalues of N for Case 1.

row (or column) Gershgorin band. 1.2592e4 0.0838 3.4779e3

5.2720e4 0.1268 3.5116e3
If the row (or column) Gershgorin band excludes the origin, 0.0011 0.1586 6.2421e3
then the distance from the origin to a center corresponding to qii(s) 0.0012 0.3217 4.4410e4
0.0028 0.4130 4.4428e4
is greater than the radius dri(s) (or dci(s)). So Eq. (33) (or Eq. (34))
0.0093 1.2453 2.4846e6
holds, Qo or Qo1 is a diagonal dominance system, and GOC can 0.0131 1.9111 1.1581e7
be a pre-decoupling controller of GGa. 0.0590 2.7093 1.1609e7
In accordance with the literature (Bai and Pang, 1988; 0.0596 442.0172 8.5442e7
Rosenbrock, 1974), Lemma 4 as the INA stability criterion is shown 0.0834 582.2467 8.5442e7

to define the stability of the control system H2.

Lemma 4. If Gershgorin band of qii(s) (i = 1, 2) does not contain the Table 5

origin and the point (1, 0), H2 is asymptotically stable if and only if Eigenvalues of N for Case 2.

8.6225e4 1.3108 24.8030 6.8687e4

X
2 X
2
0.0018 1.4106 30.6339 1.6169e5
enc1  enc2 ¼ encqii ðsÞ  encð1Þ qii ðsÞ ¼ n0 ð35Þ 0.0065 2.3160 32.9072 7.4734e5
i¼1 i¼1
0.0072 2.5820 55.3270 7.4805e5
0.0346 3.7735 5.8524e4 8.7053e5
where enc1-the sum of the number of turns of encircling the origin with 0.0368 6.3159 5.9520e4 8.7186e5
0.9982 21.0198 6.2687e4 9.5234e7
anticlockwise by Gershgorin band of qii(s) (i = 1, 2); enc2-the sum of the 0.9990 21.1411 6.3239e4 1.1353e9
number of turns of encircling (1, 0) with anticlockwise by Gershgorin
band of qii(s) (i = 1, 2); n0 – the pole number in the right half complex
plane of the open-loop system GGa  GOC of H2.
5.2. Stability analysis
The second stability theorem of the core load following control
system for Case 2 is proposed as Theorem 2 based on Criterion 1 According to Section 5.1, steps of stability analysis for Hn1 (Case
and Lemmas 1, 4. 1) or Hn2 (Case 2) are as follows.

Theorem 2. The nonlinear core load following control system for Case  Case 1
2 is globally asymptotically stable on the sufficient condition that the
pair (A, B) in GGa is controllable, and the control system H2 is (1) Determine controllability of Ga
asymptotically stable.
The determinant value of Pc of Ga is calculated as 4.1595e110,
namely Pc is nonsingular. Therefore, Ga is controllable in accor-
Proof. If the control system H2 is asymptotically stable according
dance with Criterion 1.
to Lemma 4, then under the condition that the pair (A, B) in GGa
is controllable according to Criterion 1, Hn2 is also asymptotically
(2) Solve a Lyapunov function
stable in the light of Lemma 1. h

q11(s) q12(s)
10 0

-0.2
Imag

5 -0.4

-0.6

0 -0.8
-0.5 0 0.5 1 1.5 -0.2 -0.15 -0.1 -0.05 0

q21(s) q22(s)
0 20

-0.2 15
Imag

-0.4 10

-0.6 5

-0.8 0
0 0.1 0.2 0.3 0.4 -1 -0.5 0 0.5 1
Real Real

Fig. 6. INA with row Gershgorin bands for GGaGOC.

 G. Li / Annals of Nuclear Energy 68 (2014) 193–203 201

Fig. 7. Calculation schematic of the nonlinear core model for Case 1 or Case 2.

Through utilizing LMI (Linear Matrix Inequalities) control tool- the zero point of y-axis and the height of core is 366 cm. The
box (Gahinet et al., 1995) and Eq. (32), the real symmetric matrix N boundary point between point 1 and point 2 is at 183 cm and
is solved. Every eigenvalue of N is a positive number and shown in the position range of AOrod is [236 316] that is the AO band. Sup-
Table 4. Therefore, there is the positive define real symmetric pose that the initial position of Mrod for Case 1 is at 366 cm, one of
matrix N meeting Theorem 1. Mrod for Case 2 is at 183 cm, one of AOrod is at 316 cm.
In terms of Theorem 1, the nonlinear core load following control |dDI| 6 0.05. The calculation schematic of the nonlinear core model
system Hn1 is asymptotically stable. for Case 1 or Case 2 is shown in Fig. 7.
A magnitude margin (mm) and a phase margin (pm) of GaGOC
with a strong robustness are calculated as mm = 5.3261e3,  Case 1
pm = 113.3392°.
GOC with 1/s is applied to the SISO nonlinear core model. When
 Case 2 dRPr is 0.1 step change, a step response of dPr from the nonlinear
core model with GOC and 1/s is shown in Fig. 8, where L1 is a re-
(a) Determine controllability of GGa sponse of dPr. It can be seen that the control performances of GOC
with 1/s are satisfactory for the nonlinear core model. So GOC with
There exists a R1616 nonsingular matrix in Pc of GGa and the 1/s is capable of controlling the nonlinear core.
determinant value of the nonsingular matrix is calculated as Under the control of GOC with 1/s, change trajectories of param-
2.4588e42. Therefore, GGa is controllable in accordance with eters during the nonlinear core following a ramp load
Criterion 1. (100% ? 90% ? 100%) are shown in Fig. 9, where C1R1 is the de-
sired load change trajectory for the start of reactor full-power,
(b) Analysis based on Theorem 1 the middle of reactor shut-down and low-power, and the end of
reactor start-up operation with a ±0.0556%/min ramp, C1R2 is a
Through utilizing LMI (Linear Matrix Inequalities) control tool- change trajectory of the core power level, C1R3 denotes target low-
box (Gahinet et al., 1995) and Eq. (32), the real symmetric matrix N er limit and upper one of DI, C1R4 is a change trajectory of the axial
is solved. Every eigenvalue of N is a positive number and shown in
Table 5. Therefore, there is the positive define real symmetric ma-
trix N meeting Theorem 1. 0.1
In terms of Theorem 1, the nonlinear core load following control L1
system Hn2 is asymptotically stable. 0.09

0.08
Power level change

(c) Analysis based on Theorem 2

0.07

INA with row Gershgorin bands for GGaGOC corresponding to H2 0.06

is drawn by programming, and poles of GGaGOC are solved, n0 = 0; 0.05
INA with row Gershgorin bands for GGaGOC is shown in Fig. 6, the
0.04
inverse Nyquist curve with a row Gershgorin band of qii(s) (i = 1,
2) does not contain the origin and (1, 0). As can be seen in 0.03
Fig. 6, enc1 = enc2 = 0. According to Lemma 4, H2 is asymptotically 0.02
stable. In terms of Theorem 2, the core load following control
system Hn2 is asymptotically stable. 0.01

0
0 20 40 60 80 100
6. Simulation
Time (sec)
The direction from the bottom to the top of core is treated as the
positive direction of y-axis for simulation work, the core bottom is Fig. 8. Step response of dPr from nonlinear core model with GOC and 1/s.
202 G. Li / Annals of Nuclear Energy 68 (2014) 193–203

(a) 1.05 As can be seen in Fig. 9, C1R2 can follow C1R1 in real time, C1R4
Power level

1 C1R1 is in the target band of DI; the dynamic processes of positions of

C1R2
0.95 Mrod and AOrod are decent and acceptable in practice.
0.9
 Case 2
0.85
0 12 15 21 24
GOC with 1/s is applied to the MIMO nonlinear core model.
When dRPr is a 0.1 step change and dRDI is zero, step responses
0
Axial power

C1R3 of dPr, dDI from the nonlinear core model with GOC and 1/s are
difference

-0.05 C1R4
-0.1
shown in Fig. 10, where K1 is the step response of dPr; K2 one of
-0.15 dDI. As can be seen from K1 and K2, the output dPr can follow dRPr
-0.2 and the step response of dDI approaches zero. This indicates that
0 12 15 21 24 under the control of GOC and 1/s, dMrod is used to regulate the out-
put dPr and scarcely works on another output dDI; dAOrod hardly
Time (h)
works on dPr. Meanwhile, the response K1 possesses good dynamic
performances including a zero steady-state error, a zero overshoot
(b) and a short adjustment time.
366
Mrod (cm)
Position of

360 C1R5 When dRPr is zero and dRDI is a 0.01 step change, step responses
350 of dPr, dDI are shown in Fig. 10, where M1 is the step response of
340
dPr; M2 one of dDI. As can be seen from M1 and M2, the output
dDI can follow dRDI and the step response of dPr approaches zero.
330
0 12 15 21 24 This indicates that in the presence of GOC and 1/s, dAOrod is used to
regulate the output dDI and scarcely works on another output dPr;
dMrod hardly works on dDI. Besides, M2 possesses good dynamic
316
AOrod (cm)
Position of

performances including a zero steady-state error, a zero overshoot

C1R6
C1R7 and a short adjustment time.
Consequently, GOC with 1/s is capable of implementing the
236 decoupling control with desirable dynamic performances for the
0 12 15 21 24 nonlinear core.
Time (h) Under the control of GOC with 1/s, change trajectories of param-
eters during the core following a ramp load (100% ? 90% ? 100%)
Fig. 9. Trajectories of parameters during the core following a ramp load with a desired trajectory of DI (5% ? 6% ? 5%) are shown in
(100% ? 90% ? 100%). Fig. 11, where C2R1 is the desired load change trajectory that is
the same as C1R1, C2R4 is the desired change trajectory of DI with
power difference, C1R5 is a change trajectory of position of Mrod, a ±0.0056%/min ramp; C2R2 is a change trajectory of the core
C1R6 denotes lower boundary and upper one of AO band, C1R7 is power level, C2R3 denotes target lower limit and upper one of
a change trajectory of position of AOrod that is motionless. DI, C2R5 is a change trajectory of the axial power difference,

-4
x 10
8
0.1
Power level change

K1 6 M1

4
0.05
2

0
0
0 50 100 0 50 100

-3
x 10
Change of axial power difference

5
0.01
K2
M2
0

0.005
-5

-10 0
0 50 100 0 50 100

Time (sec) Time (sec)

Fig. 10. Step responses of dPr and dDI from the nonlinear core model with GOC and 1/s.
 G. Li / Annals of Nuclear Energy 68 (2014) 193–203 203

(a) 1.05 within a target band simultaneously, and make the core have the
load following ability in response to different load changes such
Power level

1
as a ramp.
0.95
C2R1
Acknowledgements
0.9 C2R2

0.85 The author would like to thank anonymous reviewers for their
0 12 15 21 24
valuable comments. The previous work was funded by the special-
Axial power difference

0.05 ized research fund for the Doctoral Program of Higher Education
from the Ministry of Education of China (No. 20100201110009;
0
Name: The Load Following Control Characteristic Research of Pres-
-0.05 surized Water Reactor Core) and partly supported by the research
C2R3
-0.1 project from the Electric Power Research Institute of Guangdong
C2R4
C2R5 Power Grid Corporation in China (Name: The Adaptive Research
0 12 15 21 24
of Frequency Modulation and Peak Shaving of Nuclear Plant). Based
on the previous work, the work of this paper is accomplished.
Time (h)

(b) References
C2R6
Position of
Mrod (cm)

183 Athans, M., 1971. The role and use of the stochastic linear-quadratic-Gaussian
problem in control system design. J. IEEE Trans. Autom. Control AC16 (6), 529–
552.
170
Athans, M., 1986. A tutorial on the LQG/LTR method. In: Proceedings of the
American Control Conference, pp. 1289–1296.
160
Busi, T., 1981. Multivariable feedback control applications in combustion
0 12 15 21 24
engineering high power density Pressurized Water Reactors. In: International
Topical Meeting on Advances Mathematical Methods for the Solution of Nuclear
316 Engineering Problems, pp. 316–325.
AOrod (cm)
Position of

C2R7
Bai, F.Z., Pang, G.Z., 1988. Multivariable Frequency Domain Theory and Design
Techniques. National Defence Industry Press, Beijing.
C2R8
Brain, D.O., Eduardo, A.M., 1995. Obtaining uniform singular values of augmented
systems using LQG/LTR. In: Proceedings of the American Control Conference,
236 pp. 461–465.
Chari, D.R., Rohr, P.C., 1997. System 80+™ standard plant design meeting load
0 12 15 21 24
follow requirements for the future. In: Proceedings of 11th Bi-Annual Pacific
Time (h) Basin Nuclear Conference, pp. 1177–1193.
Chi, S.G., Cho, N.Z., 2002. H1 control theory applied to xenon control for load-
Fig. 11. Trajectories of parameters during the core following a ramp load following operation of a nuclear reactor. J. Nucl. Technol. 137 (2), 127–138.
(100% ? 90% ? 100%) with a desired trajectory of DI (5% ? 6% ? 5%). Doyle, J.C., Stein, G., 1981. Multivariable feedback design: concepts for a classical/
modern synthesis. J. IEEE Trans. Automat. Control AC26 (1), 4–16.
Dorf, R.C., Bishop, R.H., 2009. Modern Control Systems, 11th ed. Publishing House of
C2R6 is a change trajectory of position of Mrod, C2R7 denotes lower Electronics Industry, Beijing, pp. 757–761.
Edwards, R.M. et al., 1990. State feedback assisted classical control: an incremental
boundary and upper one of AO band, C2R8 is a change trajectory of approach to control modernization of existing and future nuclear reactors and
position of AOrod. power plants. J. Nucl. Technol. 92 (2), 167–185.
As can be seen in Fig. 11, C2R2 (C2R5) can follow C2R1 (C2R4) in Eliasi, H. et al., 2011. Robust nonlinear model predictive control for nuclear power
plants in load following operations with bounded xenon oscillations. J. Nucl.
real time, C2R5 is in the target band of DI; the dynamic processes
Eng. Des. 241 (2), 533–543.
of positions of Mrod and AOrod are decent and acceptable in Gahinet, P., Nemirovski, A., et al., 1995. LMI Control Toolbox User’s Guide. The
practice. MathWorks Inc., Natick.
IAEA, 1971. Directory of Nuclear Reactors Vol. IX. International Atomic Energy
Agency, Vienna, pp. 27–34.
7. Conclusions Isidori, A., 1995. Nonlinear Control Systems, third ed. Springer, New York, pp. 137–
178.
Kerlin, T.W. et al., 1978. Theoretical and experimental dynamic analysis of the H.B.
The core load following control problem with xenon oscillations Robinson nuclear plant. J Nucl. Technol. 30 (3), 299–316.
is dealt with. The described core without boron is modeled in Khajavi, M.N. et al., 2002. A neural network controller for load following operation
terms of the two-point mode and the lumped parameter method. of nuclear reactors. J. Ann. Nucl. Energy 29 (6), 751–760.
Khorramabadi, S.S. et al., 2002. Emotional learning based intelligent controller for a
According to the power regulating rod moving in point 1 or point
PWR nuclear reactor core during load following operation. J. Ann. Nucl. Energy
2, the linearized SISO or MIMO model of the core is analyzed and 35 (11), 2051–2058.
calculated via the small perturbation linearization method for Case Lin, C., Shen, C.M., 2000. Neurocontrol of pressurized water reactors in load-follow
operations. J. Nucl. Technol. 132 (3), 389–402.
1 or Case 2. This modeling technique with the lumped parameter
Liu, B., Tang, W.S., 2007. Modern Control Theory. China Machine Press, Beijing, pp.
method and the small perturbation linearization one can be a ref- 157–186.
erence method of modeling other nonlinear plants. Meyer, C.E. et al., 1978. Improved load follow strategy for return-to-power
For Case 1, a LQG/LTR SISO controller with robustness is de- capability. J. Nucl. Technol. 41 (1), 27–35.
Na, M.G. et al., 2005. A model predictive controller for load-following operation of
signed; for Case 2, a LQG/LTR MIMO controller with robustness is PWR reactors. J. IEEE Trans. Nucl. Sci. 52 (4), 1009–1020.
designed, which possesses the abilities of decoupling control and Rosenbrock, H.H., 1974. Computer-Aided Control System Design. Academic Press,
dynamic control. For each case, the nonlinear core load following New York.
Schultz, M.A., 1961. Control of Nuclear Reactors and Power Plants, second ed.
control system is built by using the controller based on the linear- McGraw-Hill, New York.
ized core model to control the nonlinear core model. Theorem 1 Safonov, M.G., Athans, M., 1977. Gain and phase margins of multiloop LQG
(Theorem 2) is a reference way of analyzing the stability of the regulators. J. IEEE Trans. Autom. Control AC22 (2), 173–179.
Stein, G., Athans, M., 1987. The LQG/LTR procedure for Multivariable feedback
SISO (MIMO) control system of a nonlinear plant. control design. J. IEEE Trans. Automat. Control AC32 (2), 105–114.
The nonlinear PWR core load following control system of each Torabi, K. et al., 2011. Robust control of the PWR core power using quantitative
case can control the core power level and axial power difference feedback theory. J. IEEE Trans. Nucl. Sci. 58 (1), 258–266.