Research article Special Issues

Model-free based control of a HIV/AIDS prevention model


  • Controlling an epidemiological model is often performed using optimal control theory techniques for which the solution depends on the equations of the controlled system, objective functional and possible state and/or control constraints. In this paper, we propose a model-free control approach based on an algorithm that operates in 'real-time' and drives the state solution according to a direct feedback on the state solution that is aimed to be minimized, and without knowing explicitly the equations of the controlled system. We consider a concrete epidemic problem of minimizing the number of HIV infected individuals, through the preventive measure pre-exposure prophylaxis (PrEP) given to susceptible individuals. The solutions must satisfy control and mixed state-control constraints that represent the limitations on PrEP implementation. Our model-free based control algorithm allows to close the loop between the number of infected individuals with HIV and the supply of PrEP medication 'in real time', in such a manner that the number of infected individuals is asymptotically reduced and the number of individuals under PrEP medication remains below a fixed constant value. We prove the efficiency of our approach and compare the model-free control solutions with the ones obtained using a classical optimal control approach via Pontryagin maximum principle. The performed numerical simulations allow us to conclude that the model-free based control strategy highlights new and interesting performances compared with the classical optimal control approach.

    Citation: Loïc Michel, Cristiana J. Silva, Delfim F. M. Torres. Model-free based control of a HIV/AIDS prevention model[J]. Mathematical Biosciences and Engineering, 2022, 19(1): 759-774. doi: 10.3934/mbe.2022034

    Related Papers:

    [1] Wenshuang Li, Shaojian Cai, Xuanpei Zhai, Jianming Ou, Kuicheng Zheng, Fengying Wei, Xuerong Mao . Transmission dynamics of symptom-dependent HIV/AIDS models. Mathematical Biosciences and Engineering, 2024, 21(2): 1819-1843. doi: 10.3934/mbe.2024079
    [2] Churni Gupta, Necibe Tuncer, Maia Martcheva . Immuno-epidemiological co-affection model of HIV infection and opioid addiction. Mathematical Biosciences and Engineering, 2022, 19(4): 3636-3672. doi: 10.3934/mbe.2022168
    [3] Divine Wanduku . A nonlinear multi-population behavioral model to assess the roles of education campaigns, random supply of aids, and delayed ART treatment in HIV/AIDS epidemics. Mathematical Biosciences and Engineering, 2020, 17(6): 6791-6837. doi: 10.3934/mbe.2020354
    [4] Tefa Kaisara, Farai Nyabadza . Modelling Botswana's HIV/AIDS response and treatment policy changes: Insights from a cascade of mathematical models. Mathematical Biosciences and Engineering, 2023, 20(1): 1122-1147. doi: 10.3934/mbe.2023052
    [5] Arni S. R. Srinivasa Rao, Kurien Thomas, Kurapati Sudhakar, Philip K. Maini . HIV/AIDS epidemic in India and predicting the impact of the national response: Mathematical modeling and analysis. Mathematical Biosciences and Engineering, 2009, 6(4): 779-813. doi: 10.3934/mbe.2009.6.779
    [6] Kangsen Huang, Zimin Wang . Research on robust fuzzy logic sliding mode control of Two-DOF intelligent underwater manipulators. Mathematical Biosciences and Engineering, 2023, 20(9): 16279-16303. doi: 10.3934/mbe.2023727
    [7] Gordon Akudibillah, Abhishek Pandey, Jan Medlock . Optimal control for HIV treatment. Mathematical Biosciences and Engineering, 2019, 16(1): 373-396. doi: 10.3934/mbe.2019018
    [8] Nawei Chen, Shenglong Chen, Xiaoyu Li, Zhiming Li . Modelling and analysis of the HIV/AIDS epidemic with fast and slow asymptomatic infections in China from 2008 to 2021. Mathematical Biosciences and Engineering, 2023, 20(12): 20770-20794. doi: 10.3934/mbe.2023919
    [9] B. M. Adams, H. T. Banks, Hee-Dae Kwon, Hien T. Tran . Dynamic Multidrug Therapies for HIV: Optimal and STI Control Approaches. Mathematical Biosciences and Engineering, 2004, 1(2): 223-241. doi: 10.3934/mbe.2004.1.223
    [10] Jeff Musgrave, James Watmough . Examination of a simple model of condom usage and individual withdrawal for the HIV epidemic. Mathematical Biosciences and Engineering, 2009, 6(2): 363-376. doi: 10.3934/mbe.2009.6.363
  • Controlling an epidemiological model is often performed using optimal control theory techniques for which the solution depends on the equations of the controlled system, objective functional and possible state and/or control constraints. In this paper, we propose a model-free control approach based on an algorithm that operates in 'real-time' and drives the state solution according to a direct feedback on the state solution that is aimed to be minimized, and without knowing explicitly the equations of the controlled system. We consider a concrete epidemic problem of minimizing the number of HIV infected individuals, through the preventive measure pre-exposure prophylaxis (PrEP) given to susceptible individuals. The solutions must satisfy control and mixed state-control constraints that represent the limitations on PrEP implementation. Our model-free based control algorithm allows to close the loop between the number of infected individuals with HIV and the supply of PrEP medication 'in real time', in such a manner that the number of infected individuals is asymptotically reduced and the number of individuals under PrEP medication remains below a fixed constant value. We prove the efficiency of our approach and compare the model-free control solutions with the ones obtained using a classical optimal control approach via Pontryagin maximum principle. The performed numerical simulations allow us to conclude that the model-free based control strategy highlights new and interesting performances compared with the classical optimal control approach.



    The SICA (Susceptible-Infected-Chronic-AIDS) compartmental model for HIV/AIDS transmission dynamics was proposed by Silva and Torres in their seminal paper [1]. Since then, the model has been extended to stochastic systems of differential equations [2,3], to fractional-order [4,5] as well as discrete-time dynamics [6], and applied with success to describe very different HIV/AIDS epidemics, like the ones in Cape Verde [7,8] or Morocco [9]. For a survey see [10].

    In this work, we consider a SICA epidemic problem of controlling the transmission of the human immunodeficiency viruses (HIV), by considering not only the medical treatment with multiple antiretroviral (ART) drugs, but also the pre-exposure prophylaxis (PrEP), which are medicines taken to prevent getting HIV infection. According to the Centers for Disease Control and Prevention, PrEP is highly effective for preventing HIV, when taken as prescribed, and reduces the risk of getting HIV from sex and from injection drug use by about 99 and 74%, respectively [11].

    The main contribution of this work is to propose a model-free based control algorithm that closes the loop between the infected individuals with HIV and PreP medication, in such manner that the medication is driven in 'real-time', according to the number of infected individuals that has to be asymptotically reduced. We highlight that model-free control offers the advantages of a simple Proportional-Integral-Derivative (PID) controller in the framework of model-free design, that is, one whose parameters can be easily tuned without a precise knowledge of the controlled epidemiological model.

    The model-free control methodology, originally proposed by Fliess and Join in [12], has been designed to control a priori any "unknown" dynamical system in a "robust" manner, and is referred to as "a self-tuning regulator" in [13]. This control law can be considered as an alternative to PI and PID controllers [14] and the performances are really satisfactory taking into account that the control is calculated based only on the information provided by the controlled input and the measured output signal of the controlled systems. This control law has been extensively and successfully applied to control many nonlinear processes: see, e.g., [12,15,16] and the references therein. In particular, some applications have been dedicated to the control of chemistry and biological processes [15,17,18,19,20], including the development of an artificial pancreas [21]. A derivative-free-based version of this control algorithm has been proposed by the first author in [22], for which some interesting capabilities of online optimization have been highlighted. To the best of our knowledge, the application of model-free control to SICA modeling has never been discussed before.

    Here, we compare the solutions obtained by the model-free control method with the corresponding solutions of an optimal control problem for HIV/AIDS transmission from [8], which has a mixed state-control constraint. In [8], the control system is based on a SICAE (Susceptible, HIV-Infected, Chronic HIV-infected under ART, AIDS-symptomatic individuals, E-under PrEP medication) model for the transmission of HIV in a homogeneously mixing population. The control u represents the fraction of susceptible individuals under PrEP, with 0u(t)1, that is, when u(t)=0, no susceptible individual takes PrEP at time t, and when u(t)=1 all susceptible individuals are taking PrEP at time t. The mixed state-control constraint refers to the fact that only people who are HIV-negative and at a very high risk of HIV infection should take PrEP, and also to the high costs of PrEP medication. Therefore, the number of susceptible individuals that takes PrEP, at each day, must be bounded by a positive constant. Moreover, the cost functional, which is aimed to be minimized, represents a balance between the number of HIV infected individuals and the costs associated with PrEP implementation.

    The paper is structured as follows. In Section 2, we propose a model-free control method and the procedure to minimize the HIV infected cases is described. In Section 3, we present some numerical simulations and provide a comparison of the results obtained using the model-free based approach with the ones in [8] from the Pontryagin maximum principle. Section 4 discusses and compares the results. Finally, some concluding remarks and possible directions for future work are given in Section 5.

    In this section, we propose our model-free based control method and apply it to an epidemiological problem of minimizing HIV-infected individuals.

    Model-free control was introduced in 2008 and 2009 by Fliess and Join in [23,24]. It is an alternative technique to control complex systems based on elementary continuously updated local modeling via unique knowledge of the input-output behavior. The key feature of this approach lies in the fact that the control system, which might be highly nonlinear and/or time-varying, is taken into account without any modeling procedure [25]. The model-free based control approach has been successfully implemented in concrete applications to diverse fields, ranging from intelligent transportation systems to energy management, etc., see [12] and references cited therein. To the best of our knowledge, no one as yet used this approach in the context of epidemiology.

    Consider a nonlinear dynamical system f:uy to control:

    {˙x=f(x,u),y=g(x), (2.1)

    where f is the function describing the behavior of a nonlinear system and xRn is the state vector. The proposed control is an application Cπ:(y,y)u, whose purpose is to control the output y of system (2.1) following an output reference y. In simulation, the system (2.1) is controlled in its "original formulation", without any modification or linearization.

    For any discrete moment tk,kN, one defines the discrete controller Cπ:(y,y)u as an integrator associated to a numerical series (Ψk)kN, symbolically represented by

    uk=Cπ(yk,yk)=Ψkt0Ki(ykyk1)dτ (2.2)

    with the recursive term

    Ψk=Ψk1+Kp(kαekβkyk1),

    where y is the output (or tracking) reference trajectory; Kp and Ki are real positive tuning gains; εk1=ykyk1 is the tracking error; and kαekβk is an initialization function, where kα and kβ are real positive constants. In practice, the integral part is discretized using, for example, Riemann sums.

    The set of the Cπ-parameters of the controller, is defined as the set of the tuning coefficients {Kp,Ki,kα,kβ}.

    Consider the problem of minimizing the number of infected individuals with HIV, given by the state trajectory I, through the control measure u associated to PrEP medication, and satisfying a mixed state-control constraint Suγmax, where γmax is a positive constant.

    The control sequence is divided into two steps, in order to manage, separately, the increasing u transient and the associated decreasing u transient that must decrease afterwards both the infected cases and the control input u to lower values on u and I. Concerning the control input u, it is expected that u(t>tf)=0, where tf is the time from which the medication is stopped.

    ● A first sequence, associated to the setting up of the medication, aims to progressively increase the medication (and thus start decreasing the infected states) until a certain threshold of infected cases is reached, above which the number of infected cases could be considered as stabilizable around a low value. This control sequence can be managed by a simple L-linear or Q-quadratic slope, such as

    (slope)u(t)=u0+Lt(quadratic)u(t)=u0+Qt2 (2.3)

    driven in open-loop, i.e., independently from any feedback of the infected state, that increases gradually the medication while satisfying the constraint on Suη due to the remaining high level of the susceptible cases. It appears crucial to accelerate the medication at the beginning, in order to reach rapidly γmax and allow a strong decreasing of the infected cases. This point will be discussed later in Section 3.

    ● Denote umax=maxu(t), the maximum value of u. The second sequence is associated to the decrease of the medication until the infected state is stabilized around a low value. This sequence is managed by our proposed model-free based control that interacts, in real-time, with the number of infected cases and, consequently, calculates the "optimal" medication u in order to decrease and stabilize the infected state. According to Eq (2.2), the control reads:

    uk=Ψkt0Ki(IkIk1)dτwithΨk=Ψk1+Kp(kαekβkIk1),

    where I denotes the infected cases reference that practically can be chosen as I=minI(t), where the minimum value of I can be reached online, updating the tracking reference and, therefore, ensuring that the control law is "synchronized" on the lowest value that can be reachable. It is worth to note that, depending on the behavior of the closed-loop, a saturation is added to bound the controlled u: 0uumax1.

    The implementation of the control scheme is depicted in Figure 1, where the control sequence starts from the transient slope or the quadratic function and then, once umax is reached, switches to the proposed model-free based controller. The parameters of the control sequence to be adjusted, comprise the Cπ-parameters set of the model-free control algorithm and the (u0,L,Q) parameters of the first sequence, depending if a linear or a quadratic slope is involved.

    Figure 1.  Proposed scheme to control a nonlinear system.

    Additional constraints. The slope can be adjusted according to a state-control constraint that determines the maximum number of susceptible individuals that take PrEP medication, at each instant of time. This constraint reads as S(t)u(t)<γmax, for all t, where γmax is the corresponding upper bound.

    To properly tune each sequence, in order to satisfy both the state-control constraint as well as to minimize the cost criteria, a derivative-free based optimization procedure can be applied [26].

    Consider the mathematical model for HIV/AIDS transmission in a homogeneously mixing population proposed in [8] and based in [1,7].

    The model subdivides human population into five mutually-exclusive compartments: susceptible individuals (S); HIV-infected individuals with no clinical symptoms of AIDS (the virus is living or developing in the individuals but without producing symptoms or only mild ones) but able to transmit HIV to other individuals (I); HIV-infected individuals under ART treatment (the so called chronic stage) with a viral load remaining low (C); HIV-infected individuals with AIDS clinical symptoms (A); individuals that are under PrEP medication (E). The total population at time t, denoted by N(t), is given by N(t)=S(t)+I(t)+C(t)+A(t)+E(t). The model assumptions are the following [1,8]. Effective contact with people infected with HIV is at a rate λ, given by

    λ=βN(I+ηCC+ηAA),

    where β is the effective contact rate for HIV transmission. The modification parameter ηA1 accounts for the relative infectiousness of individuals with AIDS symptoms, in comparison to those infected with HIV with no AIDS symptoms. Individuals with AIDS symptoms are more infectious than HIV-infected individuals (pre-AIDS) because they have a higher viral load and there is a positive correlation between viral load and infectiousness [27]. On the other hand, ηC1 translates the partial restoration of immune function of individuals with HIV infection that use ART correctly [28]. All individuals suffer from natural death, at a constant rate μ. HIV-infected individuals, with and without AIDS symptoms, have access to ART treatment. HIV-infected individuals with no AIDS symptoms I progress to the class of individuals with HIV infection under ART treatment C at a rate ϕ, and HIV-infected individuals with AIDS symptoms are treated for HIV, at rate α. HIV-infected individuals with AIDS symptoms A, that start treatment, move to the class of HIV-infected individuals I, moving to the chronic class C only if the treatment is maintained. HIV-infected individuals with no AIDS symptoms I that do not take ART treatment progress to the AIDS class A, at rate ρ. Individuals in the class C that stop ART medication are transferred to the class I, at a rate ω. Only HIV-infected individuals with AIDS symptoms A suffer from an AIDS induced death, at a rate d. The proportion of susceptible individuals that takes PrEP is denoted by ψ. It is assume that PrEP is effective, so that all susceptible individuals under PrEP treatment are transferred to class E. The individuals that stop PrEP become susceptible individuals again, at a rate θ. Susceptible individuals are increased by the recruitment rate Λ.

    Such model is given by the following system of ordinary differential equations:

    {˙S(t)=Λβ(I(t)+ηCC(t)+ηAA(t))N(t)S(t)μS(t)ψS(t)+θE(t),˙I(t)=β(I(t)+ηCC(t)+ηAA(t))N(t)S(t)(ρ+ϕ+μ)I(t)+αA(t)+ωC(t),˙C(t)=ϕI(t)(ω+μ)C(t),˙A(t)=ρI(t)(α+μ+d)A(t),˙E(t)=ψS(t)(μ+θ)E(t). (2.4)

    Recall that PrEP medication should only be administrated to people who are HIV-negative and at very high risk for HIV infection. Moreover, PrEP is highly expensive and it is still not approved in many countries. Therefore, the number of individuals that should take PrEP should be limited at each instant of time for a fixed interval of time [0,tf] [8]. The optimal control problem proposed in [8], and considered in this paper for comparison of results, takes into account this health public problem.

    The main goal of the optimal control problem is to determine the PrEP strategy ψ that minimizes the number of individuals with pre-AIDS HIV-infection I as well as the costs associated with PrEP. Let the fraction of individuals that takes PrEP, at each instant of time, be a control function, that is, ψu(t) with t[0,tf], and assume that the total population N is constant: the recruitment rate is proportional to the natural death rate, Λ=μN, and there are no AIDS-induced deaths (d=0). The controlled model is given by

    {˙S(t)=μNβN(I(t)+ηCC(t)+ηAA(t))S(t)μS(t)S(t)u(t)+θE(t),˙I(t)=βN(I(t)+ηCC(t)+ηAA(t))S(t)(ρ+ϕ+μ)I(t)+αA(t)+ωC(t),˙C(t)=ϕI(t)(ω+μ)C(t),˙A(t)=ρI(t)(α+μ)A(t),˙E(t)=S(t)u(t)(μ+θ)E(t). (2.5)

    Remark 1. All the parameters of the SICAE model (2.5) are fixed with the exception of the control function u(t). This system is deterministic and there is no uncertainty. However, the model-free based control proposed in this paper does not use these equations. They are only needed in the classical optimal control approach that is used here for comparison. The sensitivity analysis of the parameters of the SICA model, which is in the basis of the SICAE model (2.4), was studied before in [7].

    The classical optimal control problem proposed in [8], and that is considered here in comparison with the model-free control method, considers the cost functional

    J(u)=tf0[w1I(t)+w2u2(t)]dt, (2.6)

    where the constants w1 and w2 represent the weights associated with the number of HIV infected individuals I and on the cost associated with the PrEP prevention treatment, respectively. It is assumed that the control function u takes values between 0 and 1. When u(t)=0, no susceptible individual takes PrEP at time t; if u(t)=1, then all susceptible individuals are taking PrEP at time t. Let γmax denote the total number of susceptible individuals under PrEP for a fixed time interval [0,tf]. This constraint is represented by

    S(t)u(t)γmax,γmax0,for almost allt[0,tf], (2.7)

    which should be satisfied at almost every instant of time during the whole PrEP program.

    Let

    x(t)=(x1(t),,x5(t))=(S(t),I(t),C(t),A(t),E(t))R5.

    The classical optimal control problem proposed in [8] consists to find the optimal trajectory ˜x, associated with the control ˜u, satisfying the control system (2.5), the initial conditions

    x(0)=(x10,x20,x30,x40,x50),withx100,x200,x300,x400,x500,

    the constraint inequality (Eq 2.7), and where the control ˜uΩ,

    Ω={u()L(0,tf)|0u(t)1}, (2.8)

    minimizes the objective functional Eq (2.6).

    We evaluate the accuracy of our model-free proposed approach compared to the classical optimal control one when applied to the SICAE model described in Section 2.3.

    Let Te denote the final time of the u treatment such as u(tTe)=0 and I(Te) denotes the final value of the state "Infected" at the time Te. Regarding the "energy" of the control input u with respect to the behavior of the infected cases and the period of time Te for which the medication u is in effect, i.e., u(tTe)=0, let us consider the following cost criteria:

    ● Cost criterion for performances over the final time Te:

    Ju+I=Te0u2+I2dτ, (2.9)
    JI=Te0I2dτ. (2.10)

    ● Time-pondered cost criterion:

    JTeu+I=Te0τ(u2+I2)dτ, (2.11)

    which takes into account the effective period needed to stabilize the infected case, i.e., the period for which u>0.

    To perform the numerical simulations, we consider the following parameter values, borrowed from [8]: N=10200, μ=1/69.54, β=0.582, ηC=0.04, ηA=1.35, θ=0.001, ω=0.09, ρ=0.1, ϕ=1 and α=0.33. The weight constants take the values w1=w2=1.

    The initial conditions are given by

    S(0)=10000,I(0)=200,C(0)=0,A(0)=0andE(0)=0,

    and the mixed state-control constraint is

    S(t)u(t)2000,for almost allt[0,tf]. (3.1)

    In Table 1, we evaluate several cases with the cost criteria Ju+I, JI and JTeu+I, according to the final time of medication Te. We compare the constrained and unconstrained classical optimal control problems and the unconstrained and constrained model-free problems with two types of configurations: slope and quadratic initial transient. The classical optimal control problem corresponds to the one performed in [8]. Figures 26 illustrate several scenarios: the constrained case is not satisfied under slope, see Figure 2; the constraint inequality (Eq 3.1) is satisfied under slope, see Figures 3 and 4; the constrained cases are satisfied under quadratic function, see Figures 5 and 6.

    Table 1.  Evaluation of the cost criteria.
    Case Te Ju+I JI JTeu+I I(Te) max[0,tfinal]Su umax
    Unconstrained model-free 11.3 5.40104 5.40104 1.15105 31.12 3129 0.70
    Constrained model-free – slope (Ⅰ) 19.0 6.45104 6.45104 2.39105 29.80 1990 0.80
    Constrained model-free – slope (Ⅱ) 22.9 6.45104 6.45104 2.86105 28.25 2000 0.62
    Constrained model-free – quad. (Ⅰ) 16.9 5.83104 5.83104 1.83105 29.12 1989 0.70
    Constrained model-free – quad. (Ⅱ) 17.2 6.66104 6.66104 2.39105 32.29 1604 0.62
    Unconstrained classical OC 25.0 4.17104 4.17104 1.69105 21.95 9750 1
    Constrained classical OC 25.0 6.14104 6.14104 2.72105 24.23 1989 1

     | Show Table
    DownLoad: CSV
    Figure 2.  Evaluation of the unconstrained model-free based control with slope as first sequence.
    Figure 3.  Evaluation of the constrained model-free based control: slope as first sequence including the constraint inequality (Eq 3.1)–case I.
    Figure 4.  Evaluation of the constrained model-free based control: slope as first sequence including the constraint inequality (Eq 3.1)–case II.
    Figure 5.  Evaluation of the constrained model-free based control: quadratic as first sequence including the constraint inequality (Eq 3.1)–case I.
    Figure 6.  Evaluation of the constrained model-free based control: quadratic as first sequence including the constraint inequality (Eq 3.1)–case II.

    From Figure 2(a), we see that, although the model-free based control and the classical optimal control approaches propose completely different control functions, the associated number of HIV infected individuals I(t), t[0,25], are very similar. Interestingly, the control solution of the model-free based control is active in a much smaller interval of time that the one obtained with the classical optimal control approach: Te=11.3 versus Te=25 (see first line of Table 1). Analogous conclusions are taken from Figures 3(a)6(a).

    It should be noted, however, that in Figure 2(b), the control obtained from the model-free based approach does not satisfy the mixed state-control constraint S(t)u(t)2000 for all t. In order to satisfy this constraint, one must increase the time interval where the model-free control is active, see Figures 3(b)6(b). This depends on the configuration of the initial transient of the control (slope or quadratic), see Figures 3(a)6(a) and Table 1.

    The proposed control sequence can be considered as "quasi-optimal" in the sense that it does not obey to the Pontryagin maximum principle, so it is not an optimal control by definition, but it offers similar properties in terms of cost criteria minimization and reduction of the duration of treatment that is assured by our procedure.

    The sequence is fully parametrized thanks to the initial transient coefficients (L or Q) associated to umax, including the Cπ-parameters for the decreasing transient, that must be adjusted according to the evolution of the infected state. In particular, the maximum value on the product Su depends on the "speed" of the increasing transient as well as the final value I(Te), which depends on the "speed" of the increasing transient, the initial value u0, and the feedback control that stabilizes I through the decreasing transient. The transient slope plays a key role in the "accuracy" of the initial decrease of the infected state since a sufficient dose of the medication u must be injected to the population in order to maintain the infected state to a lower level. Therefore, the maximum value of u is a trade-off between the constraint Suγmax to be satisfied and the duration of the treatment. Figure 2 illustrates a rather quick treatment, involving thus a fast initial transient but the constraint Su2000 is not satisfied; slower medical treatment for which the injection of the medication u takes more time due to the constraint Su2000, can reduce the final asymptotic value I(Te) despite not necessarily fully reducing the cost criteria. The model-free based control aims to relax the treatment until u=0 is reached. A first tuning has been made according to the gained experience and a more precise tuning can be performed using [26]. The numerical evaluation of the cost criteria shows that our approach is globally better in terms of energy minimization. Moreover, the time-pondered criteria shows that the proposed control procedure is favorable to our approach taking into account the reduction of the duration of the treatment. These results illustrate afterwards that tuning the parameters of the proposed control sequence is a trade-off between considering minimizing the cost criteria, or minimizing the final value I(Te), depending of additional constraints.

    We remark that the model-free based control could have been also used to drive the initial transient instead of the slope or the quadratic function, but the current algorithm offers slower performances at the very beginning to initiate the increase of the medication that prevents it to deal with, for example, the constraint Su2000.

    We have considered a SICAE epidemiological model for HIV/AIDS transmission, proposing, for the first time in the literature, a model-free based approach to minimize the number of infected individuals. This approach consists in initializing PrEP medication, using a basic linear or quadratic function, and, after that, creating a direct feedback to control the decrease of infected individuals with respect to the considered measure of infected cases. Globally, the advantages of the proposed approach, when compared with the classical optimal control based on the Pontryagin Maximum Principle, is that it does not need any a priori knowledge of the model and a simple tuning of the proposed control sequence values allows good performances in terms of "energy" minimization as well as minimization of the medical treatment duration. We concluded that our control strategy highlights interesting performances compared with the classical optimal control approach used in [8].

    From a biological point of view, our application of the model-free based control approach allows to propose new solutions for the implementation of PrEP in the prevention of HIV transmission, considering the constraints associated to the limitations on the availability of medicines for HIV and on number of individuals that the health systems have capacity to follow up during their treatment. To the best of our knowledge, we were the first to apply the model-free based control approach in the context of epidemiology.

    Future work may include replacing the slope or the quadratic initial transient by an optimal control; improvement of the proposed model-free based control; implementation and comparison with the original Fliess-Join version of the model-free control, as it has been done, for example, for the glycemia control [14]. Stability issues regarding the closed loop are very important and a promising LMI framework dedicated to study the stability of optimization algorithms is also of interest [29,30].

    This work was partially supported by Portuguese funds through CIDMA, The Center for Research and Development in Mathematics and Applications of University of Aveiro, and the Portuguese Foundation for Science and Technology (FCT–Fundação para a Ciência e a Tecnologia), within project UIDB/04106/2020. Silva is also supported by the FCT Researcher Program CEEC Individual 2018 with reference CEECIND/00564/2018.

    The authors are grateful to Reviewers for several constructive suggestions and remarks.

    The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.



    [1] C. J. Silva, D. F. M. Torres, A TB-HIV/AIDS coinfection model and optimal control treatment, Discrete Contin. Dyn. Syst., 35 (2015), 4639–4663. doi: 10.3934/dcds.2015.35.4639. doi: 10.3934/dcds.2015.35.4639
    [2] J. Djordjevic, C. J. Silva, D. F. M. Torres, A stochastic SICA epidemic model for HIV transmission, Appl. Math. Lett., 84 (2018), 168–175. doi: 10.1016/j.aml.2018.05.005. doi: 10.1016/j.aml.2018.05.005
    [3] X. Wang, C. Wang, K. Wang, Extinction and persistence of a stochastic SICA epidemic model with standard incidence rate for HIV transmission, Adv. Differ. Equations, 2021 (2021). doi: 10.1186/s13662-021-03392-y. doi: 10.1186/s13662-021-03392-y
    [4] C. J. Silva, D. F. M. Torres, Stability of a fractional HIV/AIDS model, Math. Comput. Simul., 164 (2019), 180–190. doi: 10.1016/j.matcom.2019.03.016. doi: 10.1016/j.matcom.2019.03.016
    [5] M. Z. Ullah, D. Baleanu, A new fractional SICA model and numerical method for the transmission of HIV/AIDS, Math. Methods Appl. Sci., 44 (2021), 8648–8659. doi: 10.1002/mma.7292. doi: 10.1002/mma.7292
    [6] S. Vaz, D. F. M. Torres, A dynamically-consistent nonstandard finite difference scheme for the SICA model, Math. Biosci. Eng., 18 (2021), 4552–4571. doi: 10.3934/mbe.2021231. doi: 10.3934/mbe.2021231
    [7] C. J. Silva, D. F. M. Torres, A SICA compartmental model in epidemiology with application to HIV/AIDS in Cape Verde, Ecol. Complexity, 30 (2017), 70–75. doi: 10.1016/j.ecocom.2016.12.001. doi: 10.1016/j.ecocom.2016.12.001
    [8] C. J. Silva, D. F. M. Torres, Modeling and optimal control of HIV/AIDS prevention through PrEP, Discrete Contin. Dyn. Syst. Ser. S, 11 (2018), 119–141. doi: 10.3934/dcdss.2018008. doi: 10.3934/dcdss.2018008
    [9] E. M. Lotfi, M. Mahrouf, M. Maziane, C. J. Silva, D. F. M. Torres, N. Yousfi, A minimal HIV-AIDS infection model with general incidence rate and application to Morocco data, Stat. Optim. Inf. Comput., 7 (2019), 588–603. doi: 10.19139/soic.v7i3.834. doi: 10.19139/soic.v7i3.834
    [10] C. J. Silva, D. F. M. Torres, On SICA models for HIV transmission, in Mathematical Modelling and Analysis of Infectious Diseases, Studies in Systems, Decision and Control, 302 (2020), 155–179. doi: 10.1007/978-3-030-49896-2_6.
    [11] Centers for Disease Control and Prevention, Pre-Exposure Prophylaxis (PrEP), 2021. Available from: https://www.cdc.gov/hiv/risk/prep/index.html.
    [12] M. Fliess, C. Join, Model-free control, Int. J. Control, 86 (2013), 2228–2252. doi: 10.1080/00207179.2013.810345. doi: 10.1080/00207179.2013.810345
    [13] K. J. Aström, P. R. Kumar, Control: a perspective, Automatica, 50 (2014), 3–43. doi: 10.1016/j.automatica.2013.10.012. doi: 10.1016/j.automatica.2013.10.012
    [14] M. Fliess, C. Join, An alternative to PIs and PIDs: intelligent proportional-derivative regulators, Int. J. Robust Nonlin., (2021). doi: 10.1002/rnc.5657. doi: 10.1002/rnc.5657
    [15] O. Bara, M. Fliess, C. Join, J. Day, S. M. Djouadi, Toward a model-free feedback control synthesis for treating acute inflammation, J. Theor. Biol., 448 (2018), 26–37. doi: 10.1016/j.jtbi.2018.04.003. doi: 10.1016/j.jtbi.2018.04.003
    [16] K. Hamiche, M. Fliess, C. Join, H. Abouaïssa, Bullwhip effect attenuation in supply chain management via control-theoretic tools and short-term forecasts: a preliminary study with an application to perishable inventories, in 6th International Conference on Control, Decision and Information Technologies (CoDIT), (2019), 1492–1497. doi: 10.1109/CoDIT.2019.8820297.
    [17] O. Bara, M. Fliess, C. Join, J. Day, S. Djouadi, Model-free immune therapy: a control approach to acute inflammation, in 15th European Control Conference (ECC), 2016. arXiv: 1607.07259.
    [18] C. Join, J. Bernier, S. Mottelet, M. Fliess, S. Rechdaoui-Guérin, S. Azimi, et al., A simple and efficient feedback control strategy for wastewater denitrification, IFAC-PapersOnLine, 50 (2017), 7657–7662. doi: 10.1016/j.ifacol.2017.08.1167. doi: 10.1016/j.ifacol.2017.08.1167
    [19] T. MohammadRidha, C. Moog, E. Delaleau, M. Fliess, C. Join, A variable reference trajectory for model-free glycemia regulation, in SIAM Conference on Control & its Applications (SIAM CT15), (2015). doi: 10.1137/1.9781611974072.9.
    [20] S. Tebbani, M. Titica, C. Join, M. Fliess, D. Dumur, Model-based versus model-free control designs for improving microalgae growth in a closed photobioreactor: some preliminary comparaisons, in 24th Mediterranean Conference on Control and Automation (MED), IEEE, (2016), 683–688. doi: 10.1109/MED.2016.7535870.
    [21] T. MohammadRidha, M. Aït-Ahmed, L. Chaillous, M. Krempf, I. Guilhem, J. Y. Poirier, et al., Model free iPID control for glycemia regulation of type-1 diabetes, IEEE Trans. Biomed. Eng., 65 (2018), 199–206. doi: 10.1109/TBME.2017.2698036. doi: 10.1109/TBME.2017.2698036
    [22] L. Michel, A para-model agent for dynamical systems, preprint, 2018. arXiv: 1202.4707.
    [23] M. Fliess, C. Join, Commande sans modèle et commande à modèle restreint, in e-STA Sciences et Technologies de l'Automatique, SEE-Société de l'Electricité, de l'Electronique et des Technologies de l'Information et de la Communication, 5 (2008), 1–23. Available from: https://hal.inria.fr/inria-00288107v3/document.
    [24] M. Fliess, C. Join, Model-free control and intelligent PID controllers: towards a possible trivialization of nonlinear control? IFAC Proc. Vol., 42 (2009), 1531–1550. doi: 10.3182/20090706-3-FR-2004.00256. doi: 10.3182/20090706-3-FR-2004.00256
    [25] T. P. Nascimento, M. Saska, Position and attitude control of multi-rotor aerial vehicles: a survey, Annu. Rev. Control, 48 (2019), 129–146. doi: 10.1016/j.arcontrol.2019.08.004. doi: 10.1016/j.arcontrol.2019.08.004
    [26] M. Porcelli, P. L. Toint, BFO, a trainable derivative-free brute force optimizer for nonlinear bound-constrained optimization and equilibrium computations with continuous and discrete variables, ACM Trans. Math. Software, 44 (2017), 1–25. doi: 10.1145/3085592. doi: 10.1145/3085592
    [27] D. P. Wilson, M. G. Law, A. E. Grulich, D. A. Cooper, J. M. Kaldor, Relation between HIV viral load and infectiousness: a model-based analysis, Lancet, 372 (2008), 314–320. doi: 10.1016/S0140-6736(08)61115-0. doi: 10.1016/S0140-6736(08)61115-0
    [28] S. G. Deeks, S. R. Lewin, D. V. Havlir, The end of AIDS: HIV infection as a chronic disease, Lancet, 382 (2013), 1525–1533. doi: 10.1016/S0140-6736(13)61809-7. doi: 10.1016/S0140-6736(13)61809-7
    [29] M. Fazlyab, A. Ribeiro, M. Morari, V. M. Preciado, Analysis of optimization algorithms via integral quadratic constraints: nonstrongly convex problems, SIAM J. Optim., 28 (2018), 2654–2689. doi: 10.1137/17M1136845. doi: 10.1137/17M1136845
    [30] J. M. Sanz-Serna, K. C. Zygalakis, The connections between Lyapunov functions for some optimization algorithms and differential equations, SIAM J. Numer. Anal., 59 (2021), 1542–1565. doi: 10.1137/20M1364138. doi: 10.1137/20M1364138
  • This article has been cited by:

    1. Miled El Hajji, Amer Hassan Albargi, A mathematical investigation of an "SVEIR" epidemic model for the measles transmission, 2022, 19, 1551-0018, 2853, 10.3934/mbe.2022131
    2. Michel Fliess, Cédric Join, Alberto d'Onofrio, Feedback control of social distancing for COVID-19 via elementary formulae, 2022, 55, 24058963, 439, 10.1016/j.ifacol.2022.09.134
  • Reader Comments
  • © 2022 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(2616) PDF downloads(98) Cited by(2)

Figures and Tables

Figures(6)  /  Tables(1)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog