Advanced Control in Epidemics: a Practical

Identification Algorithm for Extended SVIR

Models with Vaccination and Social Mobility
Control ⋆
Marcus V. Americano da Costa ∗ Igor M. L. Pataro ∗∗
∗
National Institute of Science and Technology in Control and
Automation of Energy Processes, Department of Automation and
Systems, Federal University of Santa Catarina, Florianópolis-SC,
Brazil (e-mail: [email protected]).
∗∗
University of Almerı́a, CIESOL, ceiA3, Department of Informatics,
04120 Almerı́a, Spain (e-mail: [email protected]).

Abstract: Humanity has historical events of epidemics and pandemics that have caused millions
of deaths and have had a devastating impact on the economy, public health, education, and social
life. In this context, researchers from different areas have investigated solutions that mitigate the
effects and spread of the virus in epidemics/pandemics. In this paper, an identification algorithm
is proposed for an epidemiological model with two doses of vaccination that considers social
mobility as a control variable (stringency index). In addition, a practical model is developed
in order to ensure the applicability of the system since much of the data is not reported in
epidemics. Therefore, the identification algorithm is composed mainly of the practical equivalent
model, discrete analytical solution for the observable compartments, adaptive variable, and
optimization. As a result, the system was validated by simulations and was incorporated into
an optimal control strategy, improving its robustness in a hypothetical scenario of an epidemic.

Keywords: Epidemiological model; vaccine; social distancing; stringency index; identification

algorithm; advanced control.

1. INTRODUCTION It is noteworthy that professionals from the healthcare

sector, as well as researchers from various scientific disci-
Epidemics refer to a significant number of cases of a spe- plines such as mathematics, physics, and engineering, have
cific disease in a determined population caused seasonally proposed strategies encompassing forecasting, decision-
by the spread of viruses, bacteria, or other infectious making, vaccination, quarantine, social distancing, and
agents. When epidemics occur in a large geographic region, lockdown measures. These efforts are aimed at mitigating
we have a pandemic. Humanity has historical events of the velocity of disease transmission (Pandey and Medlock,
epidemics and pandemics that have caused millions of 2015; Morato et al., 2020b; Pataro et al., 2022). In this
deaths and had a devastating impact on the economy, context, the art of mathematical modeling arises as an ef-
public health, education, and social life. ficient technique to analyze and develop practical solutions
for dynamic systems like an epidemic.
Measures like social distancing, quarantine, the use of
Personal Protective Equipment (PPE) - masks, gloves, The first epidemiological model, created by Kermack and
clothes, etc. - and specific chemical or medical products McKendrick (1927), is a type of compartmental model
are employed to combat epidemics. The dynamic behavior used to study the spread of infectious diseases in a pop-
of the disease spread presents different waves of cases, ulation. The acronym SIR represents the three main cat-
with variations in severity depending on factors such as egories of individuals in the population: Susceptible (S),
government decrees, availability of medical resources, and active Infected (I), and Recovered or Removed (R). This
population adherence to these prevention measures. In ex- model is described by differential equations and assumes
treme situations, the World Health Organization (WHO) that once an individual recovers from the disease, they
declares an international emergency to obtain global and acquire immunity and cannot be reinfected.
public cooperation, mainly in scientific research, which Nonetheless, extended models have been developed with
is essential to finding answers to mitigate the effects of new compartments and control measures incorporated into
epidemics/pandemics. their equations. In this way, vaccination has been included
⋆ The authors would like to thank the National Council for Scientific in the SIR structure to produce its effects on dengue trans-
and Technological Development - CNPq - Brazil (PhD Program mission (Pandey and Medlock, 2015), measles (Roberts
Abroad/Process no 201143/2019-4 and no 200291/2023-8) for finan- and Tobias, 2000) and influenza A (H1N1) (Laetitiau
cial support. and Turinici, 2000) epidemics. In general, SIRD, SEIR,
and SEIRD models are used to predict the spread of RV , recovered RI , and deceased (D). The model scheme
the virus. However, epidemiological models become more is sketched in Figure 1.
complex according to the details desired to represent the
epidemic/pandemic dynamics, and the complexity is also
variable depending on the particular disease. For instance, S IS
it is possible to find models with a great number of classes
of individuals and/or other mathematical methods to sim- RI
ulate different diseases like Ebola (Asher, 2018), diarrhea
(Heaney et al., 2020), and COVID-19 (Pataro et al., 2021). RV V1 I V1
Based on the foregoing, this work proposes an epidemio-
logical model and a practical identification algorithm for D
applications of advanced control in epidemics. The proce-
dure is composed of two layers: the first one uses a discrete V2 I V2
analytical equation to calculate the “exact” parameters
that characterize the main dynamics of the disease, such as
transmissivity of infection, recovery, and fatality rates. The Figure 1. SVVIIIRRD model in a block diagram.
second one, employing adaptive variables and an optimiza-
tion problem to estimate other parameters, approximates A dynamic stringency index that comprises the social
the epidemic curves as closely as possible to the actual mobility responses due to isolation policies is incorporated
data, aiming to fit the dynamic model, maintaining the into the model, denoted as SVVIIIRRD+ψ, which is
model’s biological sense. In this proposed approach, com- expressed by the nonlinear differential equations 1 :
partments of two doses of vaccine and a stringency index dS βS S(IS + δV1 IV1 + δV2 IV2 )
= −(1 − ψ) − τ1 S
(associated with social mobility variables) are incorporated dt N
into the model and included in the identification algorithm dV1 βV V1 (IS + δV1 IV1 + δV2 IV2 )
to relate its real effects on the epidemic, allowing proper = τ1 S − (1 − ψ) 1 − ϕ1 ϵ1 V1 − τ2 V1
dt N
predictions and investigation for control strategies.
dV2 βV2 V2 (IS + δV1 IV1 + δV2 IV2 )
The paper is organized as follows: based on the SVIR+ψ = τ2 V1 − (1 − ψ) − ϕ2 ϵ2 V2
dt N
structure, Section 2 explains the proposed nominal model dIS βS S(IS + δV1 IV1 + δV2 IV2 )
and the parametric identification algorithm in detail. In = (1 − ψ) − (γS + αS )IS
dt N
Section 3, using the nominal model as a reference, the
dIV1 βV V1 (IS + δV1 IV1 + δV2 IV2 )
identification algorithm is performed and the practical = (1 − ψ) 1 − (γV1 + αV1 )IV1
dt N
model is validated. In Section 4, a hypothetical epidemic
dIV2 βV V2 (IS + δV1 IV1 + δV2 IV2 )
scenario is simulated with an adaptive optimal control = (1 − ψ) 2 − (γV2 + αV2 )IV2
system to demonstrate this type of technique as a solution dt N
to mitigate epidemics considering economic and social dRv
= ϕ1 ϵ1 V1 + ϕ2 ϵ2 V2
aspects. Finally, Section 5 states the final considerations dt
and conclusions. dRI
= γS IS + γV1 IV1 + γV2 IV2
dt
dD
= αS IS + αV1 IV1 + αV2 IV 2
2. EPIDEMIOLOGICAL MODELING dt
(1)
The SVIR structure is used in this work as the basis of wherein the parameter βj is the transmission rate; γj is the
mathematical modeling since its equations can adequately recovery rate, and the parameter αj denotes the observed
describe the dynamic behavior of a viral epidemic consid- mortality rate of the virus, removing some individuals from
ering the influence of the process of vaccination. The pro- the infected class Ij (with j ∈ {S, V1 , V2 }). Without com-
posed SVIR model describes the dynamics of a population promising the methodology of this work, we assume here
divided into compartments of susceptible (S), vaccinated, that the population has full and immediate adherence to
which may get infected (or susceptible vaccinated) (V ), the stringency measures. In other words, social distancing
and symptomatic (reported) infections (I). The removed responses coincide with stringency index ψ, which denotes
compartment (R) is composed of the individuals in whom restrictions on the people’s mobility, i.e., ψ = 1 stands for
the vaccine dosage presents total effectiveness, the recov- a complete lockdown, whereas ψ = 0 means no social dis-
ered people from the infections after a certain period and tancing. Note that these parameters may not be constant
death cases due to severe symptoms. and may vary according to the stage of the pandemic. It is
reasonable to assume that, in the case of the virus does not
The model proposed here was inspired by the COVID-19 mutate, the identified disease parameters tend to stabilize
pandemic. However, we can generalize it to other types of at a natural point. However, new events, such as changes
diseases. Thus, we subdivided the compartment V into the in medical treatments or new government protocols, can
susceptible individuals with one dose of vaccine (V1 ) and affect the disease dynamic and, consequently, modify the
two doses of vaccine (V2 ). Consequently, I is composed biological parameters. The parameter τj is the gain rate of
of infected individuals without vaccine IS , with one dose vaccination for the first (j = 1) and second (j = 2) doses
IV1 and with two doses IV2 from the compartments S, V1 with their period to obtain immunity ϕj and effectiveness
and V2 , respectively. Finally, we made explicit the removed
people in the total immune individuals due to vaccines 1 The argument t was neglected to simplify the writing.
ϵj . Lastly, δVj (j ∈ {1, 2}) is the reduction factor of infec- real data, different values for the same parameters affect
tivity by the vaccine. We consider ϵ1 = ηϵ2 with 0 < η < 1, the epidemic characterization.
and δV2 = κδV1 with κ ∈ [0, 1[.  dS β1 SI
 = −(1 − ψ) − Θ1
dt N

The initial population size (before the contagion) is de- 


noted by N0 , and, in this work, it is assumed that natural  dV1 = Θ − (1 − ψ) β2 V1 I − ϕ ϵ V − Θ



deaths balance the newborns, which holds that:  1 1 1 1 2
 dt N



N0 = S + V1 + V2 + IS + IV1 + IV 2 + RV + RI + D . (2)  dV2

 β3 V 2 I
| {z } | {z } | {z }  = Θ2 − (1 − ψ) − ϕ 2 ϵ2 V 2
 dt N

V R
I


Thus, the size of the total population exposed is calculated  dI

βeq I
= (1 − ψ) (S + V1 + V2 ) − (γeq + αeq )I
by N = N0 − D = S + Va + IS + RI , and Va = V1 + V2 +  dt N
IV1 + IV2 + Vϵ1 + Vϵ2 , in which Va is the number of all alive


 dRV
= ϕ1 ϵ1 V1 + ϕ2 ϵ2 V2

vaccinated people, and Vϵ1 and Vϵ2 are the totally immune 


 dt
individuals with one and two doses of vaccine, respectively. 
dRI


Note that RV = Vϵ1 + Vϵ2 .


 = γeq I


 dt
The theoretical SVVIIIRRD+ψ is the nominal model that 

 dD
= αeq I

will be used as a reference to validate the proposed iden- 
dt

tification algorithm. A practical model will be developed,
considering the data normally available during an epidemic (7)
(or pandemic). Therefore, in order to be able to describe the epidemic
behavior, especially for simulations, it is proposed a two-
IDENTIFICATION ALGORITHM step identification method for the parameters βeq (t), γeq (t)
and αeq (t). As the first step, this procedure includes a
The richness of data is essential to the quality and accuracy discrete analytical formulation derived from Equation (7)
of identified system models. However, in an epidemic, and, subsequently, an optimization algorithm based on the
much data and information are unavailable or not reg- Ordinary Least Square (OLS) minimization method to find
istered, such as what portion of the infected individuals the parameters βj (j ∈ {1, 2, 3}). In order to strictly adjust
is vaccinated and the number of their doses. Therefore, the equivalent parameters given in the previous layer,
the algorithm is developed from a practical point of view, caused by the numerical approximation of the original
considering the data that the public agencies currently analytical equations and the supposed data corrupted by
register. Thus, it is proposed to integrate the group of in- minor issues, a small bound can be defined for them.
fectious and reduce the structure model using the following Thus, the algorithm developed here ensures appropriate
equivalence, in which exist the time-varying parameters β parameters that characterize the epidemic’s reality since
and β0 , such that: analytical calculations are used from the SVVIRRD+ψ
IS + δV1 IV1 + δV2 IV2 = βI, (3a) model concepts, and it can adjust the model responses
βS S + βV1 V1 + βV2 V2 = β0 (S + V1 + V2 ). (3b) according to the official data.
As I = IS + IV1 + IV2 and defining βeq = β0 β, from As the published data are available with a sampling time of
Equation (1), it has: T1 and the proposed algorithm organizes time series packs
dI βeq I of T2 units of time used as calculus basis, the first layer
= (1 − ψ) (S + V1 + V2 ) − (γS + αS )IS (4) calculates the parameters as mean values from a discrete
dt N
−(γV1 + αV1 )IV1 − (γV2 + αV2 )IV2 . analytical mode as follows:
Similarly, we can make:
RI (kf ) − RI (ki )
γS IS + γV1 IV1 + γV2 IV2 = γeq (IS + IV1 + IV2 ) (5) γ̄eq (k) = Pi=kf (8)
and i=ki I(i)
αS IS + αV1 IV1 + αV2 IV2 = αeq (IS + IV1 + IV2 ), (6)
obtaining the practical equivalent model in relation to D(kf ) − D(ki )
Equation (1), in which β1 = βS β, β2 = βV1 β, β3 = βV2 β, ᾱeq (k) = Pi=kf (9)
and Θ1 is the vaccination rate applied on the susceptible i=ki I(i)
individuals, and Θ2 is the vaccination rate applied on the Pkf
susceptible vaccinated people with one dose. I(kf ) − I(ki ) + (γ̄eq + ᾱeq ) ki I(i)
β̄eq (k) = Pi=kf
Recent numerical algorithms have been applied to calcu- i=ki(1 − ψ(i))I(i)[S(i) + V1 (i) + V2 (i)]/N (i)
late the parameters of the COVID-19 model (Morato et al., (10)
2020a). Nonetheless, due to the degree of freedom for the wherein k = 1, 2, 3...j with j = Td /T2 being the number
parameters given by Equation (7), different parametric val- of time series packs delimited by the points ki = (k −
ues can produce equivalent results concerning the number 1)T2 /T1 + 1 and kf = kT2 /T1 + 1, and Td the epidemic
of susceptible, vaccinated, infected, immune, recovered, duration. Note that, for any generic parameter, χ(t) =
and deceased individuals. In other words, no unique solu- χ̄(k), in which t ∈ [(k − 1)T2 , kT2 ]. Therefore, the time
tion for the parameters solved by the numerical estimators window T2 as a tuning parameter directly influences the
exists. Although mathematical and graphical criteria have calculations. Moreover, ψ(i) is the known social mobility
been used to validate these dynamic models compared to index at time i.
Note that Equation (10) contains compartments that are in which Vτj is the (first (j = 1), or second (j = 2))
not reported in an epidemic (in general, cumulative and reported dose applied on the population. ϕj and ϵj are
active infected, recovered, and deceased individuals are information available in the scientific reports of the vac-
reported in practice) 2 . However, we can infer the sum S + cines. Therefore, the identification algorithm is composed
V1 +V2 = N0 −I −RV −RI −D from Equation (2), since the mainly of the practical equivalent model, discrete analyt-
only unknown data is related to RV . Thus, we propose the ical solution for the observable compartments, adaptive
estimation R˜V = ξVτ , wherein Vτ is the total vaccinated variable, and optimization.
people, with the adaptive variable 0 ≤ ξ ≤ 1 calculated
from R˜V , which is given by the practical equivalent model 3. VALIDATION
in the previous loop whose period is T2 (in each loop we
used the average of the ξ values calculated in the previous In this section, we use the Nominal Model as a reference
loop). to validate the identified Practical Model. The idea is to
In the second layer, the available real data from the same demonstrate the equivalence between the proposed models
interval i ∈ [ki , kf ] are used once again to minimize the and the applicability of the identification algorithm in
OLS problem. Providing the data (observed) vector I, RI real epidemic situations, as the unknown information is
and D in the optimization loop, a dynamic SVIRRD+ψ considered in the simulations.
model takes into account the decision variables βeq , γeq , Since government data sources disclose daily samples of
αeq , βj , and the initial values I(ki ), RI (ki ) and D(ki ) the pandemic dynamics, counting the infections cases,
to estimate the values of I, RI and D from ki to kf . deaths, and vaccinated people, the unit of time as the
At each interaction, the optimization layer minimizes the calculus basis used in the algorithm is day. Therefore,
error between the real data and its estimated values from it is intuitive to choose a sampling time of T1 = 1 day
the SVIRRD+ψ model choosing the decision variables for the discrete-time dynamics samples in the interval i.
accordingly. The error is then calculated as: Still, as the algorithm organizes the data in time series
packs composed of each T2 days, T2 is tuned as the
ErY (i) = (Y (i) − Ŷ (i, β̂eq , γ̂eq , α̂eq , βˆj ))2 , disease’s average incubation period. Thus, each estimated
epidemiological parameter χ̄(k) refers to the pandemic’s
ˆ RˆI , D̂} are estimated according to the
for which Ŷ = {I, dynamics behavior for each T2 days along the time.
SVIRRD+ψ model equations.
The procedure starts collecting the available real data from
Thus, the complete optimization problem is formulated as the first (k = 1) time series pack of T2 days split into
follows:
T1 sample periods within the interval points i ∈ [ki , kf ].
i=kf Using the number of active infected I(i), recovered RI (i),
min J=
X

w1 ErI (i) + w2 ErRI (i) + w3 ErD (i) ,(11)
and fatal D(i) cases, the epidemiological parameters are
βj ,βeq ,γeq ,αeq calculated in the first layer of the system by means of the
i=ki
Equations (8), (9) and (10). The obtained parameters de-
s.t.: δ β̄eq ≤ βeq ≤ δ β̄eq , fine the epidemiological characteristics properly since the
δγ̄eq ≤ γeq ≤ δγ̄eq , set of equations provides a unique solution. Nonetheless,
δ ᾱeq ≤ αeq ≤ δ ᾱeq , numerical approximation and the discretization procedure
β
j
≤ βj ≤ β j , with time series packs mislead the measurements. Thus,
these parameters are refined in the second layer, where
wherein δ and δ are the strict interval for the fine ad- the optimization algorithm also calculates the parameters
justment of equivalent epidemiological parameters on the βj and ensures the best values to improve the dynamics
optimization problem (we can define different intervals curve fitness. These stages run continuously in the loop
for each parameter according to the uncertainties). w1 , until the last collected data.
w2 and w3 are taken as positive weighting values tuning
parameters, used to normalize the magnitude order of We chose T1 = 1 and T2 = 14 days, which means that
the total cost concerning variables ErI (active infection the practical model would be simulated every day, and the
errors), ErRI (recovered case errors), and ErD (death case parameters would identified every two weeks. According
errors). Ensuring good predictions of these variables is an to the literature, we use mean values for the period to
essential strategy in terms of epidemic control design since obtain immunity (one and two doses) and effectiveness,
it must decrease the number of infected people to avoid the respectively, as ϕ1 = ϕ2 = 1/14 days−1 , ϵ1 = 80%
collapse of health systems and, naturally, preserve human and ϵ2 = 88%. Finally, we adjust the weighting gains
life. of the cost function in Equation (11) as w1 = 0.1,
w2 = 2 · 10−3 and w3 = 90, and define the uncertainty
Finally, the vaccination rates Θ1 and Θ2 are calculated by: intervals as 2.5% for the equivalent parameters (βeq , γeq ,
Vτ (i) − Vτj (i − 1) αeq ). This ensures real epidemiological characterization.
Θj (i − 1) = j , The simulations correspond to 490 days of an epidemic
T1
location of 14 million people, whose stringency index and
2 Generally, the published data are the accumulated infected, death, characteristics are designed as shown in Tables 1, 2 and 3.
recovered, and vaccinated individuals, being possible to calculate
new cases for these variables and the active infected people (I)
(assuming that the newborns balance the natural deaths). Moreover, In Figure 2, the active infected people dynamics are plot-
the stringency index is a composite measure based on social response ted from the Nominal Model and the identified Practical
indicators, which can be accessed on Our World in Data (2021). Model, where it can be noticed the precise fit between
 Table 1. Stringency Index. 4. CONTROL APPLICATION
ψ Time [day] ψ Time [day]
In this section, the model identification algorithm is as-
0.65 t < 84 0.40 126 ≤ t < 226
sociated with an optimal controller in order to update
0.60 84 ≤ t < 126 0.35 t ≥ 226
its predictor model in real-time using data from the last
Table 2. Vaccination rates for the Nominal T2 = 14 days, making the strategy robust and adaptive.
Model. The control scheme is illustrated in Figure 3, whose sam-
pling time is Ts = 7 days. Thus, the online identification
Parameter Time [day] works on a backward sliding mode since T2 > Ts .
τ1 = 0, τ2 = 0 t < 84
τ1 = 0.005, τ2 = 0 84 ≤ t < 126
τ1 = 0.0035, τ2 = 0.2 t ≥ 126

Table 3. Nominal Model parameters.

δV1 = 0.8; βV2 = 0.2; αS = 6.5 · 10−4 ;
δV2 = 0.7; γS = 0.1; αV1 = 1.5 · 10−4
βS = 0.4; γV1 = 0.2; αV2 = 6.5 · 10−4
βV1 = 0.3; γV2 = 0.25;
Figure 3. Optimal Control Scheme.
the curves. Similarly, the other observed compartments
(recovered and fatal cases) presented great fits.
The controller is designed to minimize the stringency
105
3.5 indexes and infections as:
Nominal Model
3 Identified Practical Model
[I] Total Active Infected

i=N
XH
2.5
min J = (λψ(i) + ωI(i)) , (13)
2
ψ
i=0
1.5 s.t.: 0.30 ≤ ψ ≤ 0.75,
1 I ≤ 120 · 103 ,
0.5 wherein NH = 14 days is the prediction horizon; λ = 4
0 and ω = 1/70000 weigh the stringency measures and
0 50 100 150 200 250 300 350 400 450
Day infections, respectively, balancing economic and health
aspects. The constraints related to ψ are due to natural
Figure 2. Active Infected individuals from the simulated
mobility restrictions (minimum) and essential services
Nominal Model and identified Practical Model.
(maximum) and to avoid the collapse of the health system
in the location. For the predictions, the system holds the
Although the compartments S, V1 , V2 , and RV are consid- last identified parameters as constant.
ered unknown for the purposes of a realistic simulation, the
proposed algorithm also fits their dynamics well. Despite The control strategy directly produces effects on the sus-
being small, the major difference between the models is ceptible, vaccinated susceptible, and active infected peo-
presented by the susceptible people. ple, being this last one the main compartment which has
been designed to mitigate the impact of the epidemic as
The results are numerically represented by the relative shown in Figure 4. The number of active infected indi-
standard deviation (RSD) shown in Table 4, defined as viduals slightly exceeds the limit at the 276o day because
s
PNn of the prediction errors. Using hard constraints and slack
2 100
k=1 (yn (k) − yp (k)) variables, one can ensure this control criterion.
RSD(%) = · , (12)
Nn − 1 ȳ 14
104
Epidemic Curve
wherein yn is the value given by the Nominal Model, yp is 12 Predictor Model
[I] Total Active Infected

Constraint
given by the identified Practical Model, ȳ is the nominal 10
mean value, and Nn is the number of the simulation points.
8

Table 4. Relative Standard Deviation between 4

the models. 2

Compartments 0
0 50 100 150 200 250 300 350 400 450
S V1 V2 I RV RI D Day
RSD [%] 4.43 2.16 2.41 0.42 1.20 0.39 0.30
Figure 4. Controlled Total Infected Active individuals.

The presented results lead to the conclusion that the Accordingly, the controller also impacts the dynamics
proposed algorithm can capture epidemic dynamics con- of Removed, Recovered, and Dead people (indirectly)
sidering vaccination with two doses and social mobility (see Figures 5 and 6), contributing to prevent deaths by
responses and can thus be useful in terms of robustness reducing the amount of infected people. Although the
and control strategy applications. system reads only RI , I, and D, the identified internal
model properly reproduces all compartment dynamics, as The system was validated by simulating a nominal model,
depicted in the figures. The sliding mode improves the considering realistic scenarios when some data are not
identified models. available, and using the adaptive formulation. Also, it was
demonstrated that control engineering is an alternative
Finally, the optimal stringency indexes calculated by the to dealing with epidemics/pandemics, offering promising
controller are shown in Figure 7. In this simulation, the solutions. Hard restrictions of isolation do not mean the
epidemic resulted in an accumulated infection and fatal best outcome in terms of fewer infections and deaths.
= 2.56 · 106 and AD = 1.60 · 104 , respectively,
cases of AI P Thus, optimal strategies indicate when and how intense
490
with SU = 1 ψ = 212.55. For a simple comparison, the control measures will be, balancing the economic and
Nominal Model simulated without control in the previous health aspects of society.
Section resulted in AI = 2.92 · 106 , AD = 1.83 · 104 and
SU = 212.55. This shows that stricter stringency indices REFERENCES
do not always result in a better solution at the end of an
epidemic. Asher, J. (2018). Forecasting Ebola with a regression
10
6 transmission model. Epidemics, 22, 50–55. doi:10.1016/
[RV],[RI] Removed and Recovered

10
Epidemic Curve - RV j.epidem.2017.02.009. The RAPIDD Ebola Forecasting
8
Predictor Model - RV Challenge.
Epidemic Curve - R
Predictor Model - RI
I Heaney, A.K., Alexander, K.A., and Shaman, J. (2020).
6 Ensemble forecast and parameter inference of childhood
4
diarrhea in chobe district, botswana. Epidemics, 30,
100372. doi:10.1016/j.epidem.2019.100372.
2 Kermack, W. and McKendrick, A. (1927). Contributions
to the mathematical theory of epidemics-i. Proceedings
0
0 50 100 150 200 250 300 350 400 450 of the Royal Society, 115A, 700–721.
Day Laetitiau, L. and Turinici, G. (2000). Individual vaccina-
Figure 5. Removed by Vaccine and Recovered individuals tion as nash equilibrium in a sir model with application
resulted by the controller. to the 2009–2010 influenza A (H1N1) epidemic in france.
Bulletin of Mathematical Biology, 77, 1955–1984. doi:
2
104 10.1007/s11538-015-0111-7.
Epidemic Curve Morato, M.M., Bastos, S.B., Cajueiro, D.O., and Normey-
Predictor Model
Rico, J.E. (2020a). An optimal predictive control strat-
1.5
[D] Fatal Cases

egy for covid-19 (SARS-CoV-2) social distancing policies

in brazil. Annual Reviews in Control, 50, 417–431. doi:
1
10.1016/j.arcontrol.2020.07.001.
Morato, M.M., Pataro, I.M.L., Americano da Costa, M.V.,
0.5
and Normey-Rico, J.E. (2020b). Optimal control con-
cerns regarding the COVID-19 (SARS-CoV-2) pan-
0
0 50 100 150 200 250 300 350 400 450 demic in Bahia and Santa Catarina, Brazil. Brazilian
Day Automation Conference, 2(1). doi:10.48011/asba.v2i1.
Figure 6. Deceased individuals resulted by the controller. 1673.
Our World in Data (2021). Stringency index.
0.8
https://ourworldindata.org/covid-stringency-index.
Optimal Control Pandey, A. and Medlock, J. (2015). The introduction
of dengue vaccine may temporarily cause large spikes
[ ] Stringency Index

0.7
in prevalence. Epidemiology and Infection, 143(6),
0.6
1276–1286. doi:10.1017/S0950268814001939.
0.5
Pataro, I.M.L., Oliveira, J.F., Morato, M.M., Amad,
A.A.S., Ramos, P.I.P., Pereira, F.A.C., Silva, M.S.,
0.4 Jorge, D.C.P., Andrade, R.F.S., Barreto, M.L., and
Americano da Costa, M. (2021). A control framework
0.3
0 50 100 150 200 250 300 350 400 450 to optimize public health policies in the course of the
Day covid-19 pandemic. Scientific Reports. doi:10.1038/
Figure 7. Optimal Stringency Index calculated by the s41598-021-92636-8.
controller. Pataro, I.M.L., Morato, M.M., Americano da Costa,
M.V., and Normey-Rico, J.E. (2022). Optimal con-
trol approach for the covid-19 pandemic in bahia and
5. CONCLUSIONS santa catarina, brazil. Journal of Control, Automa-
tion and Electrical Systems, 33(1), 49–62. doi:10.1007/
This work proposed an epidemiological model and a practi- s40313-021-00760-7.
cal identification algorithm for calculating epidemiological Roberts, M.G. and Tobias, M.I. (2000). Predicting and
parameters regarding restriction measures and vaccination preventing measles epidemics in new zealand: applica-
effects. The procedure uses an SVVIRRD+ψ structure and tion of a mathematical model. Epidemiology and Infec-
discrete analytical solutions to ensure correct parameters tion, 124(2), 279–287. doi:10.1017/S0950268899003556.
according to measured data and epidemiological concepts.