Hindawi Publishing Corporation

Mathematical Problems in Engineering

Volume 2013, Article ID 895640, 10 pages
http://dx.doi.org/10.1155/2013/895640

Research Article
Fractional-Order Generalized
Predictive Control: Application for Low-Speed Control of
Gasoline-Propelled Cars

M. Romero,1 A. P. de Madrid,1 C. Mañoso,1 V. Milanés,2 and B. M. Vinagre3

1
Escuela Técnica Superior de Ingenierı́a Informática, UNED, Juan del Rosal, 16, 28040 Madrid, Spain
2
California PATH, University of California at Berkeley, Richmond, CA 94804-4698, USA
3
Industrial Engineering School, University of Extremadura, Avenida de Elvas s/n, 06071 Badajoz, Spain

Correspondence should be addressed to M. Romero; [email protected]

Received 9 November 2012; Accepted 22 January 2013

Academic Editor: Clara Ionescu

Copyright © 2013 M. Romero et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

There is an increasing interest in using fractional calculus applied to control theory generalizing classical control strategies as the
PID controller and developing new ones with the intention of taking advantage of characteristics supplied by this mathematical
tool for the controller definition. In this work, the fractional generalization of the successful and spread control strategy known
as model predictive control is applied to drive autonomously a gasoline-propelled vehicle at low speeds. The vehicle is a Citroën
C3 Pluriel that was modified to act over the throttle and brake pedals. Its highly nonlinear dynamics are an excellent test bed for
applying beneficial characteristics of fractional predictive formulation to compensate unmodeled dynamics and external disturba-
nces.

1. Introduction following the GL definition (1) will be used to implement

fractional operators:
Fractional calculus can be defined as a generalization of
derivatives and integrals to noninteger orders, allowing cal- ∞
𝛼
culations such as deriving a function to real or complex 𝐷𝛼 𝑓(𝑡)𝑡=𝑘ℎ = lim ℎ−𝛼 ∑(−1)𝑗 ( ) 𝑓 (𝑘ℎ − 𝑗ℎ) , 𝛼 ∈ R,
order [1, 2]. Although this branch of mathematical analysis ℎ→0 𝑗
𝑗=0
began 300 years ago when Liebniz and L’Hôpital discussed (1)
the possibility that 𝑛 could be a fraction 1/2 for 𝑛th derivative
𝑑𝑛 𝑦/𝑑𝑥𝑛 , it was really developed at the beginning of the
19th century by Liouville, Riemann, Letnikov, and other where 𝛼 is the fractional order of the derivative or integral, h is
mathematicians [3]. the differential increment—close to zero—, and 𝑗 varies from
Fractional-order operators are commonly represented by 0 to ∞ due to the infinite memory of fractional operators.
𝐷𝛼 that stands for 𝛼-th-order derivative. Negative values In order to describe the dynamical behaviour of systems,
of 𝛼 correspond to fractional-order integrals: 𝐷−𝛼 ≡ 𝐼𝛼 . the Laplace transform is often used. Expression (2) gives the
These operators can be evaluated using two general frac- Laplace transform of the GL definition under zero initial con-
tional definitions, Riemann-Liouville (RL) and Grünwald- ditions. Nevertheless, the discretization of (2) does not lead to
Letnikov (GL). Both definitions, continuous and discrete, a transfer function with a limited number of coefficients in z
are equivalent for a wide class of functions which appear in [4]. Thus, the so-called short memory principle [1] is applied,
real physical and engineering applications [1]. In this work, which means taking into account the behaviour only in the
discrete domain will be exclusively considered. Hence, in the recent past that corresponds to a n-term truncated series,
2 Mathematical Problems in Engineering

On the other hand, driver-assistance systems have been

a topic of active research during the last decades. They are
intended to reduce traffic accidents and traffic congestions
[34–37]. Open-loop cruise control (CC) systems are a well-
𝑡−2𝑡−1 𝑡 𝑡 + 𝑁1 𝑡 + 𝑁2 known class of driver-assistance systems, based on control-
ling the throttle pedal, that reduces driver workload and
Figure 1: Model-based predictive control analogy. improve vehicle safety [38].
Nowadays, the tedious task of driving in traffic jams
represents an unresolved issue in the automotive sector
[39] because commercial vehicles exhibit highly nonlinear
paying a penalty in the form of some inaccuracy [5]:
dynamics due to the behaviour of the vehicle engine at very
low speed. Therefore, it constitutes one of the most important
𝐿 {𝐷±𝛼 𝑓 (𝑡)} = 𝑠±𝛼 𝐹 (𝑠) , ∀𝛼 ∈ R. (2) control challenges of the automotive sector [40]. Recently,
approaches to resolve this problem have been studied both
Nowadays, this mathematical tool is more and more used in using experimental scaled-down vehicles [41] and using
control theory to enhance the system performance. Typical commercial vehicles [42, 43].
fractional-order controllers include the CRONE control [6] In this paper, an application of FGPC to the velocity con-
and the PI𝜆 D𝜇 controller [7, 8]. Advanced control system trol of a mass-produced car at very low speeds is described.
strategies have also been generalized: fractional optimal con- The goal is to highlight the beneficial characteristics of
trol [9–11], fractional fuzzy adaptive control [12], fractional FGPC to compensate unmodeled dynamics and external
nonlinear control [13], fractional iterative learning control disturbances using the proposed tuning method. These char-
[14], and fractional predictive control, the latter known acteristics were shown up in [32], where the lateral control of
as fractional-order generalized predictive control (FGPC), an autonomous vehicle is carried out by FGPC in the presence
which was initially proposed in [15]. of sensor noise and the effect of the communication network.
Model predictive control (MPC) is an advanced process The remainder of this paper is organized as follows:
control methodology in which a dynamical model of the plant Section 2 summarizes the fundamentals of fractional predic-
is used to predict and optimize the future behaviour of the tive control methodology. Section 3 includes the description
process over a time interval [16–18]. At each present time t, of the experimental vehicle, presents the design and tuning
MPC generates a set of future control signals 𝑢(𝑡 + 𝑘 | 𝑡) of the fractional predictive control, and shows the results of
based on the prediction of future process outputs 𝑦(𝑡 + 𝑘 | 𝑡) the experimental trial, including a comparison with integer-
within the time window defined by 𝑁1 (minimum costing order GPC controllers. Finally, Section 4 draws the main
horizon), 𝑁2 (maximum costing horizon), and 𝑁𝑢 (control conclusions of this work.
horizon). (With this notation, 𝑥(𝑡 + 𝑘 | 𝑡) stands for the value
of 𝑥 at time 𝑡 + 𝑘 predicted at time t.) However, only the first 2. Controller Formulation
element of the control sequence 𝑢(𝑡 | 𝑡) is applied to the sys-
tem input. When the next measurement becomes available The GPC control law is obtained by minimizing, possibly
(present time equal to 𝑡 + 1), the previous procedure is subject to a set of constraints, the cost function:
repeated to find new predicted future process outputs 𝑦(𝑡 +
𝑁2 𝑁𝑢
1 + 𝑘 | 𝑡 + 1) and calculate the corresponding system input 2
𝑢(𝑡+1 | 𝑡+1) with prediction time windows moving forward; 𝐽GPC (Δ𝑢, 𝑡) = ∑ 𝛾𝑘 (𝑟 (𝑡 + 𝑘)−𝑦 (𝑡+ 𝑘)) +∑ 𝜆 𝑘 Δ𝑢(𝑡+𝑘 − 1)2 ,
𝑘=𝑁1 𝑘=1
for this reason this kind of control is also known as receding
horizon control (RHC). Figure 1 depicts the analogy between (3)
predictive control and a car driver who calculates the car where 𝑟 is the reference, y is the output, u is the control
manoeuvre following a receding horizon strategy [16]. signal, 𝛾𝑘 and 𝜆 𝑘 are nonnegative weighting elements, Δ is
MPC has become an industrial standard that has been the increment operator, and it is assumed that u(t) remains
widely adopted during the last 30 years. With over 2000 constant from time instant 𝑡 + 𝑁𝑢 (1 ≤ 𝑁𝑢 ≤ 𝑁2 ) [29, 30].
industrial installations, this control method is currently the For the sake of simplicity in the notation (⋅ | 𝑡) is omitted,
most implemented for process plants [19]. It was originally since all expressions are referred to the present time t.
developed to meet the specialized control needs of petroleum Outputs are predicted making use of a CARIMA model
refineries [20, 21]. MPC technology can now be found in to describe the system dynamics:
a wide variety of application areas such as chemicals [22,
23], solar power plants [24], agriculture [25], or clinical 𝑇𝑐 (𝑧−1 )
anaesthesia supply [26]. Recent developments related to MPC 𝐴 (𝑧−1 ) 𝑦 (𝑡) = 𝐵 (𝑧−1 ) 𝑢 (𝑡) + 𝜉 (𝑡) , (4)
Δ
can be found in [27, 28].
Generalized predictive control (GPC) [29, 30] is one of where 𝐵(𝑧−1 ) and 𝐴(𝑧−1 ) are the numerator and denom-
the most representative MPC formulations. Its fractional- inator of the model transfer function, respectively, 𝜉(t)
order counterpart, FGPC, uses a real-order fractional cost represents uncorrelated zero-mean white noise, and 𝑇𝑐 (𝑧−1 )
function to combine the characteristics of fractional calculus is a (pre)filter to improve the system robustness rejecting
and predictive control into a versatile control strategy [31–33]. disturbance and noise [44, 45].
Mathematical Problems in Engineering 3

𝑑(𝑡) the resolution of two Diophantine equations. See [16–18] for

+ 𝑒(𝑡) 𝐵 + + more details:
𝑟(𝑡) 𝑇𝑐 𝑆𝑐 𝑢(𝑡) 𝑦(𝑡)
𝑆𝑐 Δ𝑅𝑐 𝐴 𝑁
− 𝑇𝑐 (𝑧−1 ) + ∑𝑖=𝑁
2
𝑘𝑖 Φ𝑖
𝑅𝑐 (𝑧−1 ) = 𝑁
1
,
∑𝑖=𝑁
2
1
𝑘𝑖 𝑧−𝑁2 +𝑖
(10)
𝑁
Figure 2: Closed-loop equivalent control schema. ∑𝑖=𝑁
2
𝑘𝑖 𝐹𝑖
𝑆𝑐 (𝑧−1 ) = 𝑁
1
.
∑𝑖=𝑁
2
1
𝑘𝑖 𝑧−𝑁2 +𝑖
Using model (4), the future system outputs 𝑦(𝑡 + 𝑘) are In GPC the weighting sequences 𝛾𝑘 and 𝜆 𝑘 are controller
predicted as 𝑦 = 𝑦𝐶 + 𝑦𝐹 , where 𝑦𝐶—forced response—is the parameters defined by the user. However, in FGPC these
part of the future output that depends on the future control sequences are obtained from the optimization process itself
actions Δ𝑢 (with 𝑦𝐶 = 𝐺 ⋅ Δ𝑢, and 𝐺 the matrix of the step and depend on the fractional integration orders 𝛼 (7) and 𝛽
response coefficients of the model), and 𝑦𝐹 —free response— (8) as well as the controller horizons.
is the part of the future output that does not depend on Δu Tuning GPC and FGPC means setting the horizon
(i.e., the evolution of the process exclusively due to its present parameters (𝑁1 , 𝑁𝑢 , 𝑁2 ) together with the weighting
state) [29]. sequences 𝛾𝑘 and 𝜆 𝑘 for GPC, and 𝛼 and 𝛽 for FGPC,
When no constraints are defined, the minimization of (3) respectively. This task is critical because closed-loop stability
leads to a linear time invariant (LTI) control law that can be depends on this choice. In GPC some thumb rules are usually
precomputed in advance. accepted [29]. In FGPC, these thumb rules are also adequate
FGPC generalizes the GPC cost function (3) making use for choosing the horizons [15, 46].
of the so-called fractional-order definite integration operator A FGPC-tuning method was proposed in [49]. Based
𝛼 𝑏
𝐼𝑎 (⋅) [15, 46, 47] (see the appendix): on optimization, the objective is the system to fulfil phase
margin, sensitivity functions, and some other robustness
𝑁 𝑁
𝐽FGPC (Δ𝑢, 𝑡) = 𝛼 𝐼𝑁12 [𝑒 (𝑡)]2 + 𝛽 𝐼1 𝑢 [Δ𝑢 (𝑡−1)]2 , ∀𝛼, 𝛽 ∈ R, specifications. (This tuning method has already been used
(5) to tune fractional-order PI𝜆 D𝜇 controllers successfully [50–
52].) In order to keep the dimension of the optimization
where 𝑒 ≡ 𝑟 − 𝑦 is the error. This cost function has been problem low, it is assumed that the horizon parameters
discretized with sampling period Δ𝑡 and evaluated using (𝑁1 , 𝑁𝑢 , 𝑁2 ) are given (for instance, following the thumb-
(A.2). rules previously announced), and only the two unknown
The FGPC cost function has an equivalent matrix form: parameters, the fractional orders 𝛼 and 𝛽, are used in the
optimization process. Thus, the function FMINCON of the
𝐽FGPC (Δ𝑢, 𝑡) ≃ 𝑒󸀠 Γ (𝛼, Δ𝑡) 𝑒 + Δ𝑢󸀠 Λ (𝛽, Δ𝑡) Δ𝑢, (6)
MATLAB optimization toolbox [53] can be used to solve the
where Γ and Λ are infinite-dimensional square real weighting corresponding optimization problem.
matrices which depend, by construction, on 𝛼 and 𝛽, respec-
tively: 3. Experimental Application
Γ ≡ Δ𝑡𝛼 diag (⋅ ⋅ ⋅ 𝑤𝑛 𝑤𝑛−1 ⋅ ⋅ ⋅ 𝑤1 𝑤0 ) (7) In this section, we present a practical application of FGPC.
We describe its design, tuning, and practical performance on
with 𝑤𝑗 = 𝜔𝑗 −𝜔𝑗−𝑛 , 𝑛 = 𝑁2 −𝑁1 , 𝜔𝑙 = (−1)𝑙 ( −𝛼
𝑙 ), and 𝜔𝑙 = 0, the longitudinal speed control of a commercial vehicle.
for all 𝑙 < 0;
3.1. Experimental Vehicle. The vehicle used for the experi-
Λ ≡ Δ𝑡𝛽 diag (⋅ ⋅ ⋅ 𝑤𝑁𝑢 −1 𝑤𝑁𝑢 −2 ⋅ ⋅ ⋅ 𝑤1 𝑤0 ) (8)
mental phase is a convertible Citroën C3 Pluriel (Figure 3)
with 𝑤𝑗 = 𝜔𝑗 − 𝜔𝑗−𝑛 , 𝑛 = 𝑁𝑢 − 1, 𝜔𝑙 = (−1)𝑙 ( −𝛽 ), and 𝜔𝑙 = 0, which is equipped with automatic driving capabilities by
𝑙 means of hardware modifications to permit autonomous
for all 𝑙 < 0.
actions on the accelerator and brake pedals. These modifica-
In absence of constraints, the minimization of this cost
tions let the controller’s outputs steer the vehicle’s actuators.
function leads to a LTI control law similar to the one of GPC
The car’s throttle is handled by an analog signal that
whose equivalent closed-loop schema is shown in Figure 2.
represents the pressure on the pedal, generated by an analog
See [46, 48] and the references therein for details.
card. The action over the throttle pedal is transformed into
𝑅𝑐 and 𝑆𝑐 are the controller polynomials obtained from
two analogue values—one of them twice the other—between
the model polynomials 𝐴 and 𝐵, and the controller parame-
0 and 5 V. A switch has been installed on the dashboard to
ters 𝑁1 , 𝑁𝑢 , 𝑁2 , 𝛼 and 𝛽, and 𝑑 stand for disturbance. From
commute between automatic throttle control and original
schema, it is easy to obtain
throttle circuit.
𝑅𝑐 Δ𝑢 (𝑡) = 𝑇𝑐 𝑟 (𝑡) − 𝑆𝑐 𝑦 (𝑡) . (9) The brake’s automation has been done taking into account
that its action is critical. In case of a failure of any of the
The value of polynomials 𝑅𝑐 and 𝑆𝑐 is obtained using the autonomous systems, the vehicle can be stopped by human
expressions (10). Φ and 𝐹 are two polynomials obtained from driver intervention. So an electrohydraulic braking system
4 Mathematical Problems in Engineering

identification process; it has been used for the purpose of

returning to the initial speed, 0 km/h.)
The model of the vehicle is obtained by means of an iden-
tification process using the MATLAB Identification Toolbox
[56], considering a normalized input—in the interval (0, 1)—
for the throttle pedal and the sampling time of GPS fixed at
200 ms:
5.1850𝑧−4
𝐺 (𝑧−1 ) = . (11)
1 − 0.7344𝑧−1 − 0.2075𝑧−2
The time-domain model validation is depicted in Figure 4. It
is observable that model (11) captures the vehicle dynamics
reasonably good (dash line) in comparison with the exper-
imental data (solid line), despite environment and circuit
perturbations.
Figure 3: Commercial Citroën C3 prototype vehicle.
3.3. Controller Design. This section describes the controller
design for the longitudinal speed control of the vehicle
is mounted in parallel with the original one, permitting to described previously. Transfer function (11) constitutes the
coexist the two braking system independently. More details starting point in the controller tuning, where beneficial char-
about throttle and brake automation can be found in [54, 55]. acteristics of fractional predictive formulation will be used to
Concerning the on-board sensor systems, a real-time compensate unmodeled dynamics and external disturbances.
kinematic-differential global positioning system (RTK- Other practical requirements have to be taken into
DGPS) that gives vehicle position with a 1 centimeter account during the design process. (1) The car response has
precision and an inertial unit (IMU) to improve the to be smooth to guarantee that its acceleration is less than
positioning when GPS signal fails are used to obtain the ±2 m/s2 , the maximum acceptable acceleration for standing
vehicle’s true position. The car’s actual speed and acceleration passengers [57]. (2) Control action 𝑢 is normalized and has to
are obtained from a differential hall effect sensor and a be in the interval [0, 1], where negative values are not allowed
piezoelectric sensor, respectively. These values are acquired as they mean brake actions.
via controller area network bus (CAN) and provide the Firstly, the horizons are chosen to capture the loop
necessary information to the control algorithm, which is dominant dynamics. We have taken a time window of 2
running in real-time in the on-board control unit (OCU), seconds ahead defined by 𝑁1 = 1 and 𝑁2 = 10, which is
generating the control actions to govern the actuators. appropriated in a heavy traffic scene (low speed). A wider
For the purpose of this work, the gearbox is always in time window supposes an increment of 𝑁2 that would lead
first gear forcing the car to move at low speed. The sampling to a system with an excessively slow response. On the other
interval was fixed by the parameters of GPS at 200 ms. hand, we have also considered the control horizon 𝑁𝑢 =
Therefore, the frequency of actions on the pedals is set to 5 Hz. 2, which represents an agreement between system response
Using these settings, the OCU can approximately perform an speed and comfort of the vehicle’s occupants. It is well-known
action every metre at a maximum speed of 20 km/h. that larger values of 𝑁𝑢 produce tighter control actions [16]
that could even make the system unstable.
3.2. Identification of the Longitudinal Dynamics. Due to the Moreover, we have used a prefilter 𝑇𝑐 (z −1 ) to improve the
gasoline-propelled vehicle dynamics at very low speeds are system robustness against the model-process mismatch and
highly nonlinear, and finding an exact dynamical model for the disturbance rejection. In [44] a guideline is given:
the vehicle is not an easy task. Nevertheless, as we have seen 𝑁1
previously, fractional predictive controller needs a CARIMA 𝑇𝑐 (𝑧−1 ) = (1 − 𝜌𝑧−1 ) , (12)
model of the plant to make the predictions. Therefore, an
identification process has to be carried out despite inevitable where 𝜌 is recommended to be close to the dominant pole of
uncertainties and circuit perturbations. (11).
Since the vehicle always remains in first gear, restricting Thus, the chosen prefilter has the following expression:
its speed at less than 20 km/h and acting a high engine brake
𝑇𝑐 (𝑧−1 ) = 1 − 0.9𝑧−1 . (13)
force, the identification process is only fulfilled for the throttle
pedal. Taking the brake pedal effect into account leads us to a Once the controller horizons and the prefilter are chosen, the
hybrid control strategy that is not the purpose of this paper. objective of the optimization process is finding the pair (𝛼, 𝛽)
The experimental vehicle response is shown in Figure 4 that fulfils some specified robustness criteria. In our case, we
(solid line), where the vehicle has been subjected to sev- shall impose the following.
eral speed changes by means of successive throttle pedal
actuations. (In Figure 4, the action of the brake pedal is (i) Maximize the phase margin (no specification is set on
also depicted but is not taken into consideration in the the gain margin).
Mathematical Problems in Engineering 5

0.2 20

Speed (km/h)
Actuators (−1, 1)

0.1 15

0 10

−0.1 5

−0.2 0
0 50 100 150 200 250 0 50 100 150 200 250

Time (s) Time (s)

Throttle Real speed

Brake Simulated speed
(a) (b)
Figure 4: Experimental vehicle response and time-domain model validation.

Gain margin Phase margin

𝛽 : 0.3
𝛼: −2.1
20 Gain: 10.58 𝛽 : 0.3
70 𝛼: −2.1
10 Phase: 60.54
Gain margin (dB)

Phase margin (deg)

0 60
−10
−20 50
−30 40
−40 30
−50
−60 20
−70 3 10
−80 1 2 2
3
3 0
2 1 −1 0 3
1
0 −1
−2 −3 −2 2 −1 0
−4 −4 −3
1
𝛼 𝛽 0 −1 −2 −2
−3 −3 𝛽
𝛼 −4 −4
Figure 5: FGPC gain margin versus 𝛼 and 𝛽.
Figure 6: FGPC phase margin versus 𝛼 and 𝛽.
(ii) Sensitivity function |𝑆(𝑗𝜔)| ≤ −30 dB for 𝜔 ≤ 0.01 rad/
s.
(iii) Complementary sensitivity function |𝑇(𝑗𝜔)| ≤ 0 dB 3.4. Experimental Results. The experimental trial was accom-
for 𝜔 ≥ 0.1 rad/s. plished at the Centre for Automation and Robotics (CAR;
joint research centre by the Spanish Consejo Superior de
(Phase margin maximization guarantees smooth system
Investigaciones Cientı́ficas and the Universidad Politécnica
output and robustness; sensitivity functions constraints give
de Madrid) private driving circuit using the Citroën C3
good noise and disturbance rejection.)
Pluriel described previously. The circuit has been designed
In order to initialize the optimization algorithm an initial
with scientific purposes and represents an inner-city area
seed (𝛼0 , 𝛽0 ) is needed. Figures 5 and 6 depict the closed loop
with straight-road segments, bends, and so on. Figure 8
magnitude and phase margins, respectively, in the interval
shows an aerial sight.
𝛼, 𝛽 ∈ [−3, 3]. We select 𝛼0 = −2.1 and 𝛽0 = 0.3 for their
To validate the proposed controller, various target speed
corresponding good gain and phase margins.
changes were set each 25 seconds, trying to keep the speed
The optimization process has been carried out in an
error close to zero. Moreover, the automatic gearbox was
interval of 20−30 seconds using a PC computer with Intel
always in first gear, avoiding any effect of gear changes and
Core 2 Duo T9300 2.5 GHz running MATLAB 2007a. The
forcing the car to move at low speed. Figure 9 depicts the
solution to the optimization problem is 𝛼∗ = −2.2456 and
responses of the vehicle, both actual—real time—(dot line)
𝛽∗ = 2.9271, for which the weighting sequences Γ and Λ
and simulated (dash-dot line). The FGPC controller accom-
are given in (14), with a phase margin of 76.76∘ (and a gain
plished all practical requirements which were set previously.
margin of 15.51 dB). The controller sensitivity functions meet
The vehicle response is stable, smooth, and reasonably good
the design specifications, as it is depicted in Figure 7:
in comparison with its simulation. It is important to remark
Γ = diag (−36.9671 −0.0406 −0.0683 −0.1273 −0.1273 that the positive reference changes are faster than the negative
one. This is mainly due to the fact that the braking manoeuvre
−0.7881 −0.7881 51.4711 82.8442 36.9411) has to be achieved by the engine brake force, and it is affected
by the slope of the circuit.
Λ = diag (0.0173 0.0090) . With respect to the comfort of the vehicle’s occupants,
(14) it is observable that vehicle acceleration always remains (in
6 Mathematical Problems in Engineering

Speed (km/h)
Sensitivity analysis 15
20
10
0 5
𝑆 (dB)

−20 0
0 10 20 30 40 50 60 70 80
−40 Time (s)
−60
Reference
10−3 10−2 10−1 100 101 102
Simulation FGPC
Frequency (rad/s) Experimental FGPC

(a) (a)

Control action [0-1]

0
0.2
−10
𝑇 (dB)

0.1

−20 0
0 10 20 30 40 50 60 70 80
Time (s)
10−3 10−2 10−1 100 101 102
(b)
Frequency (rad/s)

Acceleration (m/s2 )
0.4
(b) 0.2
0
Figure 7: Sensitivity functions. −0.2
−0.4
0 10 20 30 40 50 60 70 80
Time (s)

Simulation FGPC
Experimental FGPC
(c)

Figure 9: FGPC controller performance.

For this reason, we have tuned several GPC controllers

Figure 8: Private driving circuit at CAR. with different constant weighting sequences 𝛾 and 𝜆. Specifi-
cally, 𝜆 ∈ {10−6 , 10−1 , 101 , 105 } and 𝛾 = 1 (as the variation of
𝛾 does not affect the system dynamics considerably).
absolute value) below the maximum acceptable acceleration Using these settings, we have obtained two GPC con-
requirement, 2 m/s2 . It is due to the soft action over the throt- trollers that in practice turned out to be unstable although
tle vehicle actuator, satisfying the comfort driving requisites. they were stable in simulation. These controllers correspond
For comparison purposes, we have also tested the per- to 𝜆 = 10−6 and 𝜆 = 10−1 (labelled Experimental GPC
formance of several GPCs which were tuned using the same 1 and Experimental GPC 2 in Figure 10, resp.). Thus, they
horizons (𝑁1 = 1, 𝑁𝑢 = 2, and 𝑁2 = 10) and prefilter 𝑇𝑐 (13) were not able to compensate unmodeled dynamics and circuit
as FGPC. perturbations.
In practice, in GPC it is commonly assumed that the On the other hand, GPC controllers for 𝜆 = 101 and 𝜆 =
5
weighting sequences are constant, that is, 𝛾𝑘 = 𝛾 and 𝜆 𝑘 = 10 (labelled Experimental GPC 3 and Experimental GPC 4
𝜆. Under this assumption, it has not been possible to find in Figure 11, resp.) were stable in practice. It is well-known
a GPC controller that fulfils the robustness criteria using that higher values of 𝜆 give rise to smooth control actions,
and equivalent optimization method. (The set of dynamics increasing the closed loop system robustness [16]. However,
that can be found with constant weights is much smaller an excessively high value of 𝜆 could make the system response
than in the case of FGPC. Furthermore, trying to optimize too slow. It would mean, in practice, that our car could not
a GPC controller in the general case (𝛾𝑘 , 𝜆 𝑘 ) would lead to stop in time, and it would probably crash into the front car.
an optimization problem with an extremely high dimension. To quantify these results, we shall compare the prin-
On the other hand, in the case of FGPC one has to optimize cipal control quality indicators for the stable realizations
only two parameters, 𝛼 and 𝛽, and this automatically leads to (GPC 3, GPC 4, and FGPC) speed error (reference speed—
nonconstant weighting sequences; recall that GPC and FGPC experimental speed), softness of the control action, and
controllers share a common LTI expression, as was pointed acceleration. The last ones require to calculate the fast fourier
out in Section 2 [49].) transform (FFT) to estimate them.
Mathematical Problems in Engineering 7

Speed (km/h)
Speed (km/h)

15
20
10
10 5
0 0
0 2 4 6 8 10 12 14 16 18 0 10 20 30 40 50 60 70 80
Time (s) Time (s)

Reference Reference
Experimental GPC 1 Experimental GPC 3
Experimental GPC 2 Experimental GPC 4

(a) (a)

Control action [0-1]

Control action [0-1]

0.5 0.5

0 0
0 2 4 6 8 10 12 14 16 18 0 10 20 30 40 50 60 70 80
Time (s)
Time (s)
(b)
(b)
Acceleration (m/s2 )

2 Acceleration (m/s2 ) 1.5

1 1
0.5
0 0
−0.5
0 2 4 6 8 10 12 14 16 18 0 10 20 30 40 50 60 70 80
Time (s) Time (s)

Experimental GPC 1 Experimental GPC 3

Experimental GPC 2 Experimental GPC 4
(c) (c)

Figure 10: Unstable GPC controllers. Action over the throttle has Figure 11: Stable GPC controllers.
been limited to [0−0.5] for passengers safety during the experimen-
tal trial.
(ii) standard deviation:
It is well known that FFT (15) is an efficient algorithm to
compute the discrete fourier transform (DFT), F, 1 𝑁−1 2
𝜎=√ ∑ (𝑒 − 𝑒) , (18)
𝑁 𝑖=0 𝑖
𝑁−1
𝑈𝑘 = F (𝑢𝑘 ) = ∑ 𝑢𝑘 𝑒(2𝜋𝑁/𝑘𝑖 ) , 𝑘 = 0, . . . , 𝑁 − 1, (15)
𝑖=0 (iii) root mean square error:

where 𝑢𝑘 is the control action or acceleration value at time

𝑡𝑘 and 𝑁 the length of these signals. FFT yields the signal 1 𝑁−1 2
RMSE = √ ∑𝑒 , (19)
sharpness by means of a frequency spectrum analysis of the 𝑁 𝑖=0 𝑖
sampled signal.
In order to get a good indicator of the overall control where 𝑒𝑘 is the speed error at time 𝑡𝑘 . Moreover, we have also
action and acceleration signals with robustness to outliers, we used the median ̃𝑒.
have used the median 𝑢 ̃ of sequence 𝑈𝑘 . All of these control quality indicators are reflected in
1 1 Table 1.
̃) ≥
𝑃 (𝑈𝑘 ≤ 𝑢 ̃) ≥ .
∧ 𝑃 (𝑈𝑘 ≥ 𝑢 (16) One observes (see Figures 9 and 11) that the speed changes
2 2
of GPC 4 and FGPC are slower than the response of GPC
The following widely used statistics parameters have been 3, so they need more time to reach the steady state after
used to evaluate the speed error: speed changes. This is reflected in Table 1 where, in terms
of speed error, all statistics parameters of GPC 3 are better
(i) mean: than the GPC 4 and FGPC ones. However, it presents very
poor values in the control action and acceleration indicators
1 𝑁−1 due to the very large fluctuations of these signals, as we
𝑒= ∑𝑒 , (17)
𝑁 𝑖=0 𝑖 can see graphically in Figure 11. This undesirable behaviour
8 Mathematical Problems in Engineering

Table 1: Comparison of stable controllers.

Speed error Control action Acceleration

Contr.
Mean St. dev. Median RMSE FFT median FFT median
GPC 3 0.0474 2.1340 2.5539 2.1317 0.4397 1.9320
GPC 4 0.3955 2.5186 6.3434 2.5461 0.0149 1.0562
FGPC 0.4604 2.4119 5.8174 2.4523 0.0044 0.1652

compromises seriously the comfort of standing passengers, Using the GL definition (1) assuming that 𝐷1−𝛼 [𝑓(𝑥)] ≠ 0, the
bordering on the maximum acceptable acceleration, 2 m/s2 . fractional-order definite integrator operator 𝛼 𝐼𝑎𝑏 (⋅) has the
Furthermore, it could injurey the throttle actuator due to its following discretized expression with a sampling period Δt:
continuous and aggressive fluctuations in the control action.
𝛼 𝑏 󸀠
The FGPC controller shows the best behaviour in the 𝐼𝑎 𝑓 (𝑥) = Δ𝑥𝛼 𝑊 𝑓, (A.2)
steady state without overshoot and presenting the best values
in terms of the softness of the control action and acceleration, where
due to the precise parameters tuning carried out by the
󸀠
optimization method. FGPC takes advantage of its diversity 𝑊 = (⋅ ⋅ ⋅ 𝑤𝑏 𝑤𝑏−1 ⋅ ⋅ ⋅ 𝑤𝑛+1 𝑤𝑛 ⋅ ⋅ ⋅ 𝑤1 𝑤0 )
of responses (varying the fractional orders 𝛼 and 𝛽) to
meet the design specifications and to improve the system 𝑓 = ( ⋅ ⋅ ⋅ 𝑓 (0) 𝑓 (Δ𝑥) ⋅ ⋅ ⋅ 𝑓 (𝑎 − Δ𝑥) 𝑓 (𝑎) (A.3)
robustness against the model-process mismatch. 󸀠
⋅ ⋅ ⋅ 𝑓 (𝑏 − Δ𝑥) 𝑓 (𝑏) )

4. Conclusions with 𝑤𝑗 = 𝜔𝑗 − 𝜔𝑗−𝑛 , 𝑛 = 𝑏 − 𝑎, 𝜔𝑙 = (−1)𝑙 ( −𝛼

𝑙 ), and 𝜔𝑙 = 0,
for all 𝑙 < 0.
The longitudinal control of a gasoline-propelled vehicle at low
speeds (common situation in traffic jams) constitutes one of
the most important topics in the automotive sector due to the Conflicts of Interest
highly nonlinear dynamics that the vehicle presents in this The authors report no actual or potential conflict of interests
situation. in relation to this manuscript.
In this paper, the fractional predictive control strategy,
FGPC, has been used to solve this problem. Taking advantage
of its beneficial characteristics and its tuning method to com- Acknowledgments
pensate un-modeled dynamics, a FGPC controller has been
designed which has achieved closed loop stability following The authors wish to acknowledge the economical support of
the changes in the velocity reference. Moreover, practical the Spanish Distance Education University (UNED), under
requirements to guarantee standing passengers comfort have Project reference PROY29 (Proyectos Investigación 2012),
been also achieved by means of the appropriate parameters and the AUTOPIA Program of the Center for Automation
choice carried out by the optimization-based tuning, in spite and Robotics UPM-CSIC.
of inevitable uncertainties and circuit perturbations.
Finally, the comparison between the fractional predictive References
control strategy, FGPC, and its integer-order counterpart,
GPC, has shown that the task of finding the correct setting [1] I. Podlubny, Fractional Differential Equations, vol. 198 of Math-
for the weighting sequences 𝛾𝑘 and 𝜆 𝑘 is crucial. In FGPC, ematics in Science and Engineering, Academic Press, San Diego,
Calif, USA, 1999.
the fractional orders 𝛼 and 𝛽 allow us to find them keeping
the dimension of the optimization problem low, since only [2] K. B. Oldham and J. Spanier, The Fractional Calculus, vol. 111 of
Mathematics in Science and Engineering, Academic Press, New
two parameters have been optimized.
York, NY, USA, 1974.
[3] K. S. Miller and B. Ross, An Introduction to the Fractional
Calculus and Fractional Differential Equations, John Wiley &
Appendix Sons, New York, NY, USA, 1993.
[4] B. M. Vinagre, C. A. Monje, and A. J. Calderón, “Fractional
Fractional-Order Definite Integral Operator order systems and fractional order control actions,” in Proceed-
ings of the 41st Conference on Decision and Control. Tutorial
The fractional-order definite integral of function f (x) within
Workshop 2: Fractional Calculus Applications in Automatic
interval [a, b] has the following expression [47]: Control and Robotics, Las Vegas, Nev, USA, 2002.
[5] I. Podlubny, “Numerical solution of ordinary fractional dif-
ferential equations by the fractional difference method,” in
𝑏
𝛼 𝑏 Advances in Difference Equations (Veszprém, 1995), pp. 507–515,
𝐼𝑎 𝑓 (𝑥) ≡ ∫ [𝐷1−𝛼 𝑓 (𝑥)] 𝑑𝑥, 𝛼, 𝑎, 𝑏 ∈ R. (A.1)
Gordon and Breach, Amsterdam, The Netherlands, 1997.
𝑎
Mathematical Problems in Engineering 9

[6] A. Oustaloup, B. Mathieu, and P. Lanusse, “The CRONE control the IEEE Conference on Control Applications, pp. 1657–1662,
of resonant plants: application to a flexible transmission,” Glasgow, UK, August 1994.
European Journal of Control, vol. 1, pp. 113–121, 1995. [25] F. Han, C. Zuo, W. Wu, J. Li, and Z. Liu, “Model predictive
[7] I. Podlubny, “Fractional-order systems and 𝑃𝐼𝜆 𝐷𝜇 -controllers,” control of the grain drying process,” Mathematical Problems in
IEEE Transactions on Automatic Control, vol. 44, no. 1, pp. 208– Engineering, vol. 2012, Article ID 584376, 12 pages, 2012.
214, 1999. [26] D. A. Linkens and M. Mahfouf, “Generalized predictive control
[8] I. Petráš, “The fractional-order controllers,” Journal of Electrical (GPC) in clinical anaesthesia,” in Advances in Model-Based
Engineering, vol. 50, pp. 284–288, 1999. Predictive Control, D. W. Clarke, Ed., pp. 429–445, Oxford
[9] O. P. Agrawal, “A general formulation and solution scheme for University Press, Oxford, UK, 1994.
fractional optimal control problems,” Nonlinear Dynamics, vol. [27] H. Yang and S. Li, “A data-driven bilinear predictive controller
38, no. 1–4, pp. 323–337, 2004. design based on subspace method,” Asian Journal of Control, vol.
[10] O. P. Agrawal and D. Baleanu, “A Hamiltonian formulation 13, no. 2, pp. 345–349, 2011.
and a direct numerical scheme for fractional optimal control [28] X.-H. Chang and G.-H. Yang, “Fuzzy robust constrained model
problems,” Journal of Vibration and Control, vol. 13, no. 9-10, pp. predictive control for nonlinear systems,” Asian Journal of
1269–1281, 2007. Control, vol. 13, no. 6, pp. 947–955, 2011.
[11] C. Tricaud and Y. Q. Chen, “Solving fractional order optimal [29] D. W. Clarke, C. Mohtadi, and P. S. Tuffs, “Generalized predic-
control problms in RIOTS 95—a general purpose optimal tive control. Part I. The basic algorithm,” Automatica, vol. 23,
control problems solver,” in Proceedings of the 3rd IFAC Work- no. 2, pp. 137–148, 1987.
shop on Fractional Differentiation and Its Applications, Ankara, [30] D. W. Clarke, C. Mohtadi, and P. S. Tuffs, “Generalized
Turkey, 2008. predictive control. Part II. Extensions and interpretations,”
[12] M. O. Efe, “Fractional fuzzy adaptive sliding-mode control of a Automatica, vol. 23, no. 2, pp. 149–160, 1987.
2-DOF direct-drive robot arm,” IEEE Transactions on Systems, [31] M. Romero, I. Tejado, B. M. Vinagre, and A. P. de Madrid,
Man, and Cybernetics, Part B, vol. 38, no. 6, pp. 1561–1570, 2008. “Position and velocity control of a servo by using GPC of
[13] I. S. Jesus, J. T. Machado, and R. S. Barbosa, “Fractional order arbitrary real order,” in New Trends in Nanotechnology and
nonlinear control of heat system,” in Proceedings of the 3rd IFAC Fractional Calculus Applications, D. Baleanu, Z. B. Guvenc, and
Workshop on Fractional Differentiation and Its Applications, J. A. T. Machado, Eds., pp. 369–376, Springer, Dordrecht, The
Ankara, Turkey, 2008. Netherlands, 2009.
[14] Y. Li, Y. Chen, and H.-S. Ahn, “Fractional-order iterative [32] M. Romero, I. Tejado, J. I. Suárez, B. M. Vinagre, and A. P. De
learning control for fractional-order linear systems,” Asian Madrid, “GPC strategies for the lateral control of a networked
Journal of Control, vol. 13, no. 1, pp. 54–63, 2011. AGV,” in Proceedings of the 5th International Conference on
Mechatronics (ICM ’09), Málaga, Spain, April 2009.
[15] M. Romero, A. P. de Madrid, C. Mañoso, and R. Hernandez,
“Generalized predictive control of arbitrary real order,” in New [33] I. Tejado, M. Romero, B. M. Vinagre, A. P. de Madrid, and Y.
Trends in Nanotechnology and Fractional Calculus Applications, Q. Chen, “Experiences on an internet link characterization and
D. Baleanu, Z. B. Guvenc, and J. A. T. Machado, Eds., pp. 411– networked control of a smart wheel,” International Journal of
418, Springer, Dordrecht, The Netherlands, 2009. Bifurcation and Chaos, vol. 22, no. 4, 2012.
[16] E. F. Camacho and C. Bordons, Model Predictive Control, [34] R. Bishop, “A survey of intelligent vehicle applications world-
Springer, New York, NY, USA, 2nd edition, 2004. wide,” in Proceedings of the IEEE Intelligent Vehicles Symposium,
pp. 25–30, Dearborn, Mich, USA, 2000.
[17] J. A. Rossiter, Model Based Predictive Control. A Practical
Approach, CRC Press, New York, NY, USA, 2003. [35] G. Marsden, M. McDonald, and M. Brackstone, “Towards
an understanding of adaptive cruise control,” Transportation
[18] J. M. Maciejowski, Predictive Control with Constraints, Prentice Research Part C, vol. 9, no. 1, pp. 33–51, 2001.
Hall, New York, NY, USA, 2002.
[36] A. Vahidi and A. Eskandarian, “Research advances in intelligent
[19] J. Qin and T. Badgwell, “An overview of industrial model pre- collision avoidance and adaptive cruise control,” IEEE Transac-
dictive control technology,” in Proceedings of the International tions on Intelligent Transportation Systems, vol. 4, no. 3, pp. 132–
Conference on Chemical Process, J. C. Kantor, C. E. Garcia, and 153, 2003.
B. Carnahan, Eds., vol. 93 of AIChE Symposium Series, pp. 232–
[37] B. Van Arem, C. J. G. Van Driel, and R. Visser, “The impact
256, 1997.
of cooperative adaptive cruise control on traffic-flow character-
[20] C. R. Cutler and B. L. Ramaker, “Dynamic matrix control—a istics,” IEEE Transactions on Intelligent Transportation Systems,
computer control algorithm,” in Proceedings of Joint Automatic vol. 7, no. 4, pp. 429–436, 2006.
Control Conference, San Francisco, Calif, USA, 1980.
[38] T. Aono and T. Kowatari, “Throttle-control algorithm for
[21] W. L. Luyben, Ed., Practical Distillation Control, Van Nostrand improving engine response based on air-intake model and
Reinhold, New York, NY, USA, 1992. throttle-response model,” IEEE Transactions on Industrial Elec-
[22] T. Alvarez, M. Sanzo, and C. de Prada, “Identification and con- tronics, vol. 53, no. 3, pp. 915–921, 2006.
strained multivariable predictive control of chemical reactors,” [39] N. B. Hounsell, B. P. Shrestha, J. Piao, and M. McDonald,
in Proceedings of the IEEE Conference on Control Applications, “Review of urban traffic management and the impacts of new
pp. 663–664, Albany, NY, USA, September 1995. vehicle technologies,” IET Intelligent Transport Systems, vol. 3,
[23] J. M. Martı́n Sánchez and J. Rodellar, Adaptive Predictive no. 4, pp. 419–428, 2009.
Control. From the Concepts to Plant Optimization, Prentice Hall, [40] S. Moon, I. Moon, and K. Yi, “Design, tuning, and evaluation
Upper Saddle River, NJ, USA, 1996. of a full-range adaptive cruise control system with collision
[24] E. F. Camacho and M. Berenguel, “Application of generalized avoidance,” Control Engineering Practice, vol. 17, no. 4, pp. 442–
predictive control to a solar power plant,” in Proceedings of 455, 2009.
10 Mathematical Problems in Engineering

[41] L. Cai, A. B. Rad, and W. L. Chan, “An intelligent longitudinal [57] BECHTEL, “Compendium of executive summaries from the
controller for application in semiautonomous vehicles,” IEEE maglev system concept definition final reports,” Tech. Rep.,
Transactions on Industrial Electronics, vol. 57, no. 4, pp. 1487– U.S. Department of Transportation, 1993, http://ntl.bts.gov/
1497, 2010. DOCS/CES.html.
[42] V. Milanés, J. Villagrá, J. Godoy, and C. González, “Comparing
fuzzy and intelligent PI controllers in stop-and-go manoeuvres,”
IEEE Transactions on Control Systems Technology, vol. 20, no. 3,
pp. 770–778, 2011.
[43] I. Tejado, V. Milanés, J. Villagrá, J. Godoy, H. HosseinNia, and
B. M. Vinagre, “Low speed control of an autonomous vehicle by
using a fractional PI,” in Proceedings of the 18th World Congress
International Federation Automatic Control, pp. 15025–15030,
Milano, Italy, 2011.
[44] T.-W. Yoon and D. W. Clarke, “Observer design in receding-
horizon predictive control,” International Journal of Control, vol.
61, no. 1, pp. 171–191, 1995.
[45] C. Mañoso, A. P. de Madrid, M. Romero, and R. Hernández,
“GPC with structured perturbations: the influence of prefilter-
ing and terminal equality constraints,” ISRN Applied Mathemat-
ics, vol. 2012, Article ID 623484, 12 pages, 2012.
[46] M. Romero, A. P. de Madrid, C. Mañoso, and B. M. Vinagre,
“Fractional-order generalized predictive control: formulation
and some properties,” in proceedings of the 11th Interna-
tional Conference on Control, Automation, Robotics and Vision
(ICARCV ’10), pp. 1495–1500, Singapore, December 2010.
[47] M. Romero, A. P. de Madrid, and B. M. Vinagre, “Arbitrary real-
order cost functions for signals and systems,” Signal Processing,
vol. 91, no. 3, pp. 372–378, 2011.
[48] M. Romero, A. P. de Madrid, C. Mañoso, and B. M. Vinagre,
“A survey of fractional-order generalized predictive control,” in
Proceedings of the 51st Conference on Decision and Control, pp.
6867–6872, Maui, Hawaii, USA, 2012.
[49] M. Romero, I. Tejado, A. P. de Madrid, and B. M. Vinagre,
“Tuning predictive controllers with optimization: application
to GPC and FGPC,” in Proceedings of the 18th World World
Congress International Federation Automatic Control, pp. 10824–
10829, Milano, Italy, 2011.
[50] C. A. Monje, A. J. Calderón, B. M. Vinagre, Y. Chen, and V. Feliu,
“On fractional PI 𝜆 controllers: some tuning rules for robustness
to plant uncertainties,” Nonlinear Dynamics, vol. 38, no. 1–4, pp.
369–381, 2004.
[51] Y. Q. Chen, H. Dou, B. M. Vinagre, and C. A. Monje, “A robust
tuning method for fractional order PI controllers,” in Proceed-
ings of the 2nd IFAC Workshop on Fractional Differentiation and
Its Applications, Porto, Portugal, 2006.
[52] C. A. Monje, B. M. Vinagre, V. Feliu, and Y. Chen, “Tuning
and auto-tuning of fractional order controllers for industry
applications,” Control Engineering Practice, vol. 16, no. 7, pp.
798–812, 2008.
[53] Mathworks Inc., “Matlab optimization toolbox user’s guide,”
2007.
[54] V. Milanés, C. González, J. E. Naranjo, E. Onieva, and T.
de Pedro, “Electro-hydraulic braking system for autonomous
vehicles,” International Journal of Automotive Technology, vol.
11, no. 1, pp. 89–95, 2010.
[55] V. Milanés, D. F. Llorca, B. M. Vinagre, C. González, and M.
A. Sotelo, “Clavileno: evolution of an autonomous car,” in Pro-
ceedings of the 13th IEEE International Intelligent Transportation
Systems, Madeira, Portugal, 2010.
[56] Mathworks Inc., “Matlab identification toolbox user’s guide,”
2007.