Loading [MathJax]/extensions/TeX/boldsymbol.js
Research article

Global optimality analysis and solution of the 0 total variation signal denoising model

  • Received: 05 December 2022 Revised: 19 January 2023 Accepted: 03 February 2023 Published: 08 February 2023
  • The total variation regularizer is diffusely emerged in statistics, image and signal processing to obtain piecewise constant estimator. The 0 total variation (L0TV) regularized signal denoising model is a nonconvex and discontinuous optimization problem, and it is very difficult to find its global optimal solution. In this paper, we present the global optimality analysis of L0TV signal denoising model, and design an efficient algorithm to pursuit its solution. Firstly, we equivalently rewrite the L0TV denoising model as a partial regularized (PL0R) minimization problem by aid of the structured difference operator. Subsequently, we define a P-stationary point of PL0R, and show that it is a global optimal solution. These theoretical results allow us to find the global optimal solution of the L0TV model. Therefore, an efficient Newton-type algorithm is proposed for the PL0R problem. The algorithm has a considerably low computational complexity in each iteration. Finally, experimental results demonstrate the excellent performance of our approach in comparison with several state-of-the-art methods.

    Citation: Shanshan Pan, Qianqian Dai, Huangyue Chen. Global optimality analysis and solution of the 0 total variation signal denoising model[J]. Mathematical Biosciences and Engineering, 2023, 20(4): 6932-6946. doi: 10.3934/mbe.2023299

    Related Papers:

    [1] Jiazhe Lin, Rui Xu, Xiaohong Tian . Transmission dynamics of cholera with hyperinfectious and hypoinfectious vibrios: mathematical modelling and control strategies. Mathematical Biosciences and Engineering, 2019, 16(5): 4339-4358. doi: 10.3934/mbe.2019216
    [2] Lili Liu, Xi Wang, Yazhi Li . Mathematical analysis and optimal control of an epidemic model with vaccination and different infectivity. Mathematical Biosciences and Engineering, 2023, 20(12): 20914-20938. doi: 10.3934/mbe.2023925
    [3] Pannathon Kreabkhontho, Watchara Teparos, Thitiya Theparod . Potential for eliminating COVID-19 in Thailand through third-dose vaccination: A modeling approach. Mathematical Biosciences and Engineering, 2024, 21(8): 6807-6828. doi: 10.3934/mbe.2024298
    [4] Zhiping Liu, Zhen Jin, Junyuan Yang, Juan Zhang . The backward bifurcation of an age-structured cholera transmission model with saturation incidence. Mathematical Biosciences and Engineering, 2022, 19(12): 12427-12447. doi: 10.3934/mbe.2022580
    [5] Amira Bouhali, Walid Ben Aribi, Slimane Ben Miled, Amira Kebir . Impact of immunity loss on the optimal vaccination strategy for an age-structured epidemiological model. Mathematical Biosciences and Engineering, 2024, 21(6): 6372-6392. doi: 10.3934/mbe.2024278
    [6] Martin Luther Mann Manyombe, Joseph Mbang, Jean Lubuma, Berge Tsanou . Global dynamics of a vaccination model for infectious diseases with asymptomatic carriers. Mathematical Biosciences and Engineering, 2016, 13(4): 813-840. doi: 10.3934/mbe.2016019
    [7] Tingting Xue, Long Zhang, Xiaolin Fan . Dynamic modeling and analysis of Hepatitis B epidemic with general incidence. Mathematical Biosciences and Engineering, 2023, 20(6): 10883-10908. doi: 10.3934/mbe.2023483
    [8] Jinliang Wang, Ran Zhang, Toshikazu Kuniya . A note on dynamics of an age-of-infection cholera model. Mathematical Biosciences and Engineering, 2016, 13(1): 227-247. doi: 10.3934/mbe.2016.13.227
    [9] Majid Jaberi-Douraki, Seyed M. Moghadas . Optimal control of vaccination dynamics during an influenza epidemic. Mathematical Biosciences and Engineering, 2014, 11(5): 1045-1063. doi: 10.3934/mbe.2014.11.1045
    [10] Linhua Zhou, Meng Fan, Qiang Hou, Zhen Jin, Xiangdong Sun . Transmission dynamics and optimal control of brucellosis in Inner Mongolia of China. Mathematical Biosciences and Engineering, 2018, 15(2): 543-567. doi: 10.3934/mbe.2018025
  • The total variation regularizer is diffusely emerged in statistics, image and signal processing to obtain piecewise constant estimator. The 0 total variation (L0TV) regularized signal denoising model is a nonconvex and discontinuous optimization problem, and it is very difficult to find its global optimal solution. In this paper, we present the global optimality analysis of L0TV signal denoising model, and design an efficient algorithm to pursuit its solution. Firstly, we equivalently rewrite the L0TV denoising model as a partial regularized (PL0R) minimization problem by aid of the structured difference operator. Subsequently, we define a P-stationary point of PL0R, and show that it is a global optimal solution. These theoretical results allow us to find the global optimal solution of the L0TV model. Therefore, an efficient Newton-type algorithm is proposed for the PL0R problem. The algorithm has a considerably low computational complexity in each iteration. Finally, experimental results demonstrate the excellent performance of our approach in comparison with several state-of-the-art methods.



    Cholera is a waterborne disease caused by Vibrio cholera [1]. It is well known that it can spread rapidly in countries without clean drinking water and developed public health infrastructure [2,3,4]. In 2017, Somalia faced one of the largest outbreaks in history, with 78,784 suspected cases, including 1,159 related deaths. A year later, the number of suspected cases and related deaths reported in Somalia fell due to improved disease surveillance and case management. However, a new cholera outbreak in Somalia began in January 2019 [5].

    For cholera, the interaction between environment and human is the most common pathway of transmission, that is, human typically is infected by ingesting water or food contaminated by vibrios from the environment [6,7,8]. On the other hand, close contacts with infected individuals (such as shaking hands and hugging) can also cause human infection, which indicates that the transmission route of human-to-human also exists [9]. A number of studies have shown that human-to-human transmission has a great impact on human infection that can not be ignored [10,11,12].

    In 2010, Tien et al. formulated a model with multiple pathways [13]. It was assumed that the incidence rate is bilinear, indicating that the incidence rate increases with the number of infected individuals and the concentration of vibrio in the environment. However, for environment-to-human transmission, considering the fact that the inhibition effect from behavioral changes of susceptible individuals and the swarming effect of vibrios, the bilinear incidence rate is unreasonable. In 2011, based on the work in [13], by introducing a saturation incidence rate β1B/(K+B) to describe the inhibition effect, in [9], Mukandavire et al. analyzed the following model:

    ˙S=μNβSIβ1SBK+BμS,˙I=βSI+β1SBK+B(γ+μ)I,˙B=ξIδB,˙R=γIμR, (1.1)

    where S, I and R stand for the densities of the susceptible, infected and removed individuals, respectively, and B stands for the concentration of V. cholera in contaminated environment. The parameter μ denotes the natural birth and death rates of human, β and β1 are the transmission rates of human-to-human and environment-to-human, respectively, K is the pathogen concentration that yields 50% chance of catching cholera, γ denotes the recovery rate, ξ is the contribution rate of each infected individual to the concentration of V. cholera shedding from infected individuals and δ is the net death rate of V. cholerae.

    In 2010, WHO recommended the oral vaccines should be used in areas with endemic cholera [14]. Subsequently, a number of cholera models with vaccination strategy have been proposed and analysed [15,16,17]. In 2015, Posny et al. proposed a new cholera model consisting of vaccination [18]. Model analysis shows that the vaccine can effectively control the spread of cholera. However, vaccine protection is not permanent. The existing oral cholera vaccine (OCV) can provide > % continuous protection, lasting for 2 years in the epidemic population [14].

    Motivated by the works of the WHO report [14] and Posny et al. [18], in this paper, we focus on the influence of multiple pathways, imperfect vaccination on cholera infection, and analyze the following model:

    ˙S=AμSϕS(βI+β1BK+B)S+ηV,˙V=ϕS(βI+β1BK+B)σV(μ+η)V,˙I=(βI+β1BK+B)(S+σV)(μ+γ+d)I,˙B=ξIδB,˙R=γIμR, (1.2)

    where V stands for the density of vaccinated individuals. Vaccination rate of susceptible individuals is ϕ, vaccine efficiency is σ and 1/η is the duration of vaccine protection. And other parameters have the same biological meanings as in system (1.1).

    The initial condition of system (1.2) is

    S(0)0,V(0)0,I(0)0,B(0)0,R(0)0, (1.3)

    and we can obtain that all solutions of system (1.2) remain positive for all t0.

    The organization of this paper is as follows. In Section 2, we show the existence of feasible equilibria. In Section 3, we establish the global stability of each of feasible equilibria of system (1.2) by constructing Liapunov functions. In Section 4, we consider the optimal control problem of cholera model with vaccination, quarantine, treatment and sanitation control strategies. In order to determine the optimal control strategy, we use Pontryagin's minimum principle. In Section 5, the model is used to fit the real disease situation of cholera outbreak in Somalia. Besides, we analyze the sensitivity of the basic reproduction number and solve the resulting optimality problem numerically. Finally, a brief discussion is given in Section 6 to end this work.

    For system (1.2), it is easy to conclude that there is always a disease-free equilibrium E0(S0,V0,0,0,0), where

    S0=A(μ+η)μ(μ+η+ϕ),V0=Aϕμ(μ+η+ϕ).

    Below, we first calculate the basic reproduction number R0 by using the method of the next generation matrix [19]. Let

    F=((βI+β1BK+B)(S+σV)0),V=((μ+γ+d)IξI+δB).

    Computing the Jacobian Matrix at E0, we have

    F=(βS0+σβV0β1S0+σβ1V0K00),V=(μ+γ+d0ξδ).

    It follows that

    FV1=(βS0+σβV0μ+γ+d+(β1S0+σβ1V0)ξKδ(μ+γ+d)β1S0+σβ1V0Kδ00).

    We obtain:

    R0=Aβ(μ+η+σϕ)μ(μ+γ+d)(μ+η+ϕ)+Aβ1ξ(μ+η+σϕ)μKδ(μ+γ+d)(μ+η+ϕ).

    If R0>1, system (1.2) has a disease-free equilibrium E0(S0,V0,0,0,0) and an endemic equilibrium E(S,V,I,B,R), where

    S=a0(K+B)2+a1(K+B)a2(K+B)2+a3(K+B)+a4,V=Aϕξ2(K+B)2a2(K+B)2+a3(K+B)+a4,I=δξB,R=γδμξB,

    here

    a0=A(μ+η)ξ2,a1=AξσB(δβ+β1ξ),a2=βδB[σδβB+(μ+η)ξ]+(ϕ+μ+η)μ,a3=σβ1Bξ[βδB+(ϕ+μ)ξ]+β1Bξ[σδβB+(μ+η)ξ],a4=σβ21B2ξ2,

    and B is the positive real root of the equation h(B)=0, where

    h(B)=b4B4+b3B3+b2B2+b1B+b0, (2.1)

    in which

    b0=k2δμξ2(μ+γ+d)(μ+ϕ+η)(1R0),b1=(μ+γ+d)[K2βδ2(μ+η)ξ+K2βσδ2(ϕ+μ)ξ+2Kδμ(μ+ϕ+η)ξ2+Kδβ1ξ2(μ+η)+Kσδβ1ξ2(μ+ϕ)]K2Aξβ2σδ22KAξ2βδ(μ+η+ϕσ)2KAββ1ξ2σδAξ3β1(μ+η+ϕσ)Aξ3β21σ,b2=(μ+γ+d)[K2β2σδ3+2Kβδ2(μ+η)ξ+2Kβσδ2(ϕ+μ)ξ+δμ(μ+ϕ+η)ξ2+2Kξββ1σδ2+(ϕ+μ)ξ2β1δσ+β1δξ2(μ+η)+σβ21δξ2]2kAξβ2σδ22Aξ2ββ1δσAβδξ2(μ+η+ϕσ),b3=(μ+γ+d)[2Kβ2σδ3+βδ2(μ+η)ξ+βσδ2(ϕ+μ)ξ+2ββ1σδ2ξ]Aξσβ2δ2,b4=(μ+γ+d)β2σδ3.

    Note that limB+h(B)=+, h(0)=b0<0 if R0>1, in this case, system (1.2) has a positive equilibrium E.

    Let N=S+V+I+R. Then ˙N=Aμ(S+V+I+R)dIAμN. It follows that

    lim supt+N(t)A/μ.

    Furthermore, we derive from the fourth equation of system (1.2),

    lim supt+B(t)Aξ/μδ.

    We therefore conclude that the set

    Ω={(S,V,I,B,R)R5+:0S(t)+V(t)+I(t)+R(t)Aμ,0B(t)Aξμδ}

    is positively invariant.

    In this section, we study the global stability of each of the equilibria to system (1.2). The approach of proofs is to use suitable Lyapunov function.

    Theorem 3.1. If R0<1, the disease-free equilibrium E0(S0,V0,0,0,0) of system (1.2) is globally asymptotically stable.

    Proof. Define

    W(t)=S0(SS01lnSS0)+V0(VV01lnVV0)+I+β1S0+σβ1V0KδB.

    Calculating the derivative of W(t) along positive solutions of system (1.2), one has

    ˙W(t)=(1S0S)(AμSϕS(βI+β1BK+B)S+ηV)+(1V0V)(ϕS(βI+β1BK+B)σV(μ+η)V)+(βI+β1BK+B)(S+σV)(μ+γ+d)I+β1S0+σβ1V0Kδ(ξIδB). (3.1)

    On substituting A=μS0+μV0, ϕS0=(μ+η)V0 into (3.1), we obtain that

    ˙W(t)=μS0(2S0SSS0)+ηV0(2S0VV0SV0SS0V)+μV0(3SS0VV0V0SS0V)+(μ+γ+d)(R01)I(β1S0+σβ1V0)B2K(K+B)0,

    and ˙W(t)<0 for all (S,V,I,B,R)(S0,V0,0,0,0). Therefore, by Lyapunov's stability Theorem [20], the equilibrium E0 is globally asymptotically stable.

    Theorem 3.2. If R0>1, the endemic equilibrium E(S,V,I,B,R) of system (1.2) is globally asymptotically stable.

    Proof. Define

    W1(t)=S(SS1lnSS)+V(VV1lnVV)+I(II1lnII)+β1S+σβ1V(K+B)δB(BB1lnBB).

    Calculating the derivative of W1(t) along positive solutions of system (1.2), one has

    ˙W1(t)=(1SS)(AμSϕS(βI+β1BK+B)S+ηV)+(1VV)(ϕS(βI+β1BK+B)σV(μ+η)V)+(1II)((βI+β1BK+B)(S+σV)(μ+γ+d)I)+β1S+σβ1V(K+B)δ(1BB)(ξIδB). (3.2)

    On substituting

    A=(S+σV)(βI+β1BK+B)+μS+μV,ϕS=σV(βI+β1BK+B)+(μ+η)V,(μ+γ+d)I=(S+σV)(βI+β1BK+B),δB=ξI

    into (3.2), we have

    ˙W1(t)=(μ+βI)S(2SSSS)+ηV(2SVVSVSSV)+(μ+σβI)V(3SSVVVSSV)+β1SBK+B(4SSBIIBK+BK+B(K+B)ISBSB(K+B)I)+σβ1VBK+B(5SSVSSVBIIBK+BK+B(K+B)IVBVB(K+B)I)Kβ1(S+σV)(BB)2(K+B)2(K+B)0,

    and ˙W1(t)<0 for all (S,V,I,B,R)(S,V,I,B,R). Therefore, by Lyapunov's stability Theorem [20], the equilibrium E is globally asymptotically stable.

    We consider the optimal control problem of cholera model with vaccinate, quarantine, treatment and sanitation control strategies:

    ˙S=A(μ+u1(t))S(β(1u2(t))I+β1(1u3(t))BK+B)S+ηV,˙V=u1(t)S(β(1u2(t))I+β1(1u3(t))BK+B)σV(μ+η)V,˙I=(β(1u2(t))I+β1(1u3(t))BK+B)(S+σV)(μ+d+γ+u4(t))I,˙B=ξI(δ+u5(t))B,˙R=(γ+u4(t))IμR. (4.1)

    Where u1(t) is a vaccination strategy aimed to the susceptible individuals; u2(t) is a quarantine strategy that can reduce the transmission of human-to-human; u3(t) is another kind of quarantine strategy that can reduce the transmission of environment-to-human; u4(t) is therapeutic treatment aimed to the infected people, u5(t) is a sanitation strategy aimed at killing vibrios in contaminated water.

    Define a control function set as U={ui(t)i=1,,5}, and X=(S,V,I,B,R). The admissible trajectories of set X are given by

    X={X(.)W1,1([0,T);R5)(1.3)and(4.1)aresatisfied}.

    Define

    U={U()L([0,T];R5)|0ui(t)uimax1,i=1,,5,t[0,T]},

    where uimax(i=1,,5) denote the upper bounds for the efforts of vaccination, quarantine strategy, another kind of quarantine strategy, treatment and sanitation, respectively.

    The objective functional

    Q(X(),U())=T0g(X(t),U(t))dt.

    The function g is called the running payoff function [20]. The objective of the optimal control problem is to minimize the objective functional

    Q(X(),U())=minX(),U()X×UQ(X(),U()). (4.2)

    The first question that must be addressed is the existence of the optimal control pair. According to the Filippov-Cesari existence theorem [20], we obtain the following result.

    Theorem 4.1. There exists a \boldsymbol{U}^{*}(\cdot) such that the objective functional in (4.2) is minimized.

    To apply Pontryagin's minimum principle [21], we need to introduce the adjoint vector function \lambda(t) = (\lambda_{S}(t), \lambda_{I}(t), \lambda_{V}(t), \lambda_{B}(t), \lambda_{R}(t)), to define the Hamiltonian:

    \begin{aligned} H(\textbf{X},\textbf{U},\lambda)& = g(\textbf{X}(t),\textbf{U}(t))\\ &+\lambda_{S}\left(A-(\mu+u_{1}(t)) S-\left(\beta(1-u_{2}(t))I+\frac{\beta_{1}(1-u_{3}(t))B}{K+B}\right)S+\eta V\right)\\ &+\lambda_{V}\left(u_{1}(t)S- \left(\beta (1-u_{2}(t))I+\frac{ \beta_{1}(1-u_{3}(t))B}{K+B}\right)\sigma V-(\mu +\eta)V\right)\\ &+\lambda_{I}\left[\left(\beta(1-u_{2}(t))I+\frac{\beta_{1}(1-u_{3}(t))B}{K+B}\right)(S +\sigma V)-(\mu+d+\gamma+u_{4}(t))I\right]\\ &+\lambda_{B}\left[\xi I-(\delta+u_{5}(t))B\right]\\ &+\lambda_{R}\left[(\gamma+u_{4}(t))I-\mu R\right]. \end{aligned}

    The adjoint functions must satisfy

    \lambda_{S}^{'} = -\frac{\partial H}{\partial S},\quad \lambda_{V}^{'} = -\frac{\partial H}{\partial V},\quad \lambda_{I}^{'} = -\frac{\partial H}{\partial I}, \quad \lambda_{B}^{'} = -\frac{\partial H}{\partial B}, \quad \lambda_{R}^{'} = -\frac{\partial H}{\partial R}.

    That is,

    \begin{equation} \begin{aligned} \frac{d\lambda_{S}}{dt}& = (\mu+u_{1}(t))\lambda_{S} +\left(\beta(1-u_{2}(t))I+\frac{\beta_{1}(1-u_{3}(t))B}{K+B}\right)(\lambda_{S}-\lambda_{I}) -u_{1}(t)\lambda_{V}-\frac{\partial g}{\partial S}\\ \frac{d\lambda_{V}}{dt}& = (\mu+\eta)\lambda_{V}+\left(\sigma\beta(1-u_{2}(t))I +\frac{\sigma\beta_{1}(1-u_{3}(t))B}{K+B}\right)(\lambda_{V}-\lambda_{I})-\eta\lambda_{S}-\frac{\partial g}{\partial V},\\ \frac{d\lambda_{I}}{dt}& = \beta(1-u_{2}(t))S(\lambda_{S}-\lambda_{I})+ \sigma\beta(1-u_{2}(t))V(\lambda_{V}-\lambda_{I})+(\mu+\gamma+u_{4}(t)+d)\lambda_{I}\\ &-\xi\lambda_{B}-u_{4}(t)\lambda_{R}-\frac{\partial g}{\partial I},\\ \frac{d\lambda_{B}}{dt}& = \frac{\beta_{1}(1-u_{3}(t))S K}{(K+B)^{2}}(\lambda_{S}-\lambda_{I})+\frac{\sigma\beta_{1}(1-u_{3}(t))V K}{(K+B)^{2}}(\lambda_{V}-\lambda_{I})+(\delta+u_{5}(t))\lambda_{B}-\frac{\partial g}{\partial B},\\ \frac{d\lambda_{R}}{dt}& = \mu\lambda_{R}-\frac{\partial g}{\partial R}, \end{aligned} \end{equation} (4.3)

    with transversality condition

    \begin{equation} \begin{aligned} \lambda_{S}(T) = \lambda_{V}(T) = \lambda_{I}(T) = \lambda_{B}(T) = \lambda_{R}(T) = 0, \quad j = 1, 2. \end{aligned} \end{equation} (4.4)

    Moreover, the characterizations of the optimal controls are based on

    \begin{aligned} \frac{\partial H}{\partial u_{i}} = 0, \quad \quad\quad\;\; i = 1,\cdots,5. \end{aligned}

    In order to explore the sensitivity of the cost function to the optimal control solution, we consider two different cost functions for the running payoff function g(\textbf{X}(t), \textbf{U}(t)) . If we choose the running payoff function

    \begin{equation} \begin{aligned} g_{1}(\textbf{X}(t),\textbf{U}(t)) = I+\frac{C_{11}}{2}u_{1}(t)^{2}+\frac{C_{21}}{2}u_{2}(t)^{2} +\frac{C_{31}}{2}u_{3}(t)^{2}+\frac{C_{41}}{2}u_{4}(t)^{2}+\frac{C_{51}}{2}u_{5}(t)^{2}, \end{aligned} \end{equation} (4.5)

    where C_{i1}(i = 1, \cdots, 5) are the weight constants for the control strategies. The C_{11}u_{1}^{2}/2 , C_{21}u_{2}^{2}/2 , C_{31}u_{3}^{2}/2 , C_{41}u_{4}^{2}/2 , C_{51}u_{5}^{2}/2 define the appropriate costs function associated with these controls [20]. We can obtain that

    \begin{equation} \begin{aligned} \widetilde{u}_{11}& = \frac{(\lambda_{S1}-\lambda_{V1})S}{C_{11}}, \quad \widetilde{u}_{21} = \frac{\beta SI(\lambda_{I1}-\lambda_{S1})+\sigma\beta VI(\lambda_{I1}-\lambda_{V1})}{C_{21}},\\ \widetilde{u}_{31}& = \frac{\beta_{1} SB(\lambda_{I1}-\lambda_{S1})+\sigma\beta_{1} VB(\lambda_{I1}-\lambda_{V1})}{C_{31}(K+B)},\quad \widetilde{u}_{41} = \frac{(\lambda_{I1}-\lambda_{R1})I}{C_{41}},\\ \widetilde{u}_{51}& = \frac{\lambda_{B1}B}{C_{51}}. \end{aligned} \end{equation} (4.6)

    In addition, if we choose another running payoff function

    \begin{equation} \begin{aligned} g_{2}(\textbf{X}(t),\textbf{U}(t))& = I+C_{12}\left(u_{1}(t)+u_{1}(t)^{2}\right)+C_{22}\left(u_{2}(t)+u_{2}(t)^{2}\right)+C_{32}\left(u_{3}(t)+u_{3}(t)^{2}\right)\\ &+C_{42}\left(u_{4}(t)+u_{4}(t)^{2}\right)+C_{52}\left(u_{5}(t)+u_{5}(t)^{2}\right), \end{aligned} \end{equation} (4.7)

    where C_{i2}(i = 1, \cdots, 5) are the weight constants for the control strategies. C_{12}\left(u_{1}(t)+u_{1}(t)^{2}\right) , C_{22}\left(u_{2}(t)+u_{2}(t)^{2}\right) , C_{32}\left(u_{3}(t)+u_{3}(t)^{2}\right) , C_{42}\left(u_{4}(t)+u_{4}(t)^{2}\right) , C_{52}\left(u_{5}(t)+u_{5}(t)^{2}\right) define the appropriate costs function associated with these controls [23]. We can obtain that

    \begin{equation} \begin{aligned} \widetilde{u}_{12}& = \frac{(\lambda_{S2}-\lambda_{V2})S-C_{12}}{2C_{12}}, \quad \widetilde{u}_{22} = \frac{\beta SI(\lambda_{I2}-\lambda_{S2})+\sigma\beta VI(\lambda_{I2}-\lambda_{V2})-C_{22}}{2C_{22}},\\ \widetilde{u}_{32}& = \frac{\beta_{1} SB(\lambda_{I2}-\lambda_{S2})+\sigma\beta_{1} VB(\lambda_{I2}-\lambda_{V2})-C_{32}}{2C_{32}(K+B)},\quad \widetilde{u}_{42} = \frac{(\lambda_{I2}-\lambda_{R2})I-C_{42}}{C_{42}},\\ \widetilde{u}_{52}& = \frac{\lambda_{B2}B-C_{52}}{2C_{52}}. \end{aligned} \end{equation} (4.8)

    Where \lambda_{Sj}, \lambda_{Vj}, \lambda_{Ij}, \lambda_{Bj}, \lambda_{Rj}(j = 1, 2) satisfy the equations (4.3) and (4.4). Based on this fact, we obtain \widetilde{u}_{ij}(i = 1, \cdots, 5, j = 1, 2) . Further, we have

    u_{ij}^{*} = \max[0,\min(\widetilde{u}_{ij},u_{imax})].

    Next, in Section 5, we apply the forward-backward sweep method to solve it numerically [15,20].

    In this section, system (1.2) is used to fit the real disease situation of cholera outbreak in Somalia. Besides, we analyze the sensitivity of R_{0} . As mentioned in Section 4, the optimal control problem needs to be solved by numerical simulation, we will show the numerical result. In addition, we list the values of parameters in Table 1.

    Table 1.  Table of biologically relevant parameter values (week).
    Parameter Description Value Source
    A Constant birth rate 7342 [5]
    \phi Vaccinate rate of susceptible 4.2836\times10^{-3} fitting
    \beta Transmission rate of human-to-human 4.3771\times10^{-10} fitting
    \beta_1 Transmission rate of environment-to-human 0.5959\times10^{-4} fitting
    K Concentration of V. cholera in environment 10^{6} [8]
    \mu Natural death rate of human 0.00038 [5]
    \eta Waning rate of vaccinate 0.0104 [14]
    \sigma Reduction rate of vaccine efficacy 0.5 [14]
    \gamma Recovery rate of infected individuals 1.5 [8]
    d Cholera mortality 0.006 [5]
    \xi Rate of release of V. cholerae 70 [8]
    \delta Natural death rate of V. cholera 0.197 [8]
    Initial values Description Value Source
    S(0) Initial susceptible population 12316000 [22]
    V(0) Initial vaccinated population 0 [22]
    I(0) Initial infected population 192 [22]
    B(0) Initial concentration of vibrios 205740 fitting
    R(0) Initial recovered population 147 fitting

     | Show Table
    DownLoad: CSV

    In this subsection, we use system (1.2) to fit the real disease situation of cholera outbreak in Somalia. The new cholera outbreak in Somalia began in January 2019, and the first round of oral cholera vaccination (OCV) activity started on June 22, 2019, the 25th week of 2019, so our numerical simulation starts from the 25th week of 2019.

    In addition, the data we obtained is the cumulative number of cases since December 2017, so our initial case number is 7,994 [22]. By using the Markov Chain Monte Carlo method, we can get fitting results (see Figure 1). It is shown that the solutions of system (1.2) are in good agreement with the actual cholera cases, which verifies the rationality of the model established in this paper.

    Figure 1.  Cumulative cases in Somalia between weeks 25 and 39 of 2019, in which the red dots represent the reported data, the blue curve is the solution of model (1.2).

    In this subsection, we use the Latin hypercube sampling (LHS) method to analyze the sensitivity of R_{0} [24]. Through the analysis of LHS samples, we obtain the Partial Rank Correlation Coefficients (PRCC) with respect to R_{0} (see Figure 2). It is easy to see that \beta_{1}, \xi, \eta, \beta, \sigma are positive correlative variables with R_{0} ; and \phi, \gamma are negative correlative variables with R_{0} .

    Figure 2.  Tornado plot of partial rank correlation coefficients in respect to R_{0} .

    In this subsection, we will show the optimal control results. Based on empirical values, we set u_{1max} = 0.7, u_{2max} = 0.9, u_{3max} = 0.6, u_{4max} = 0.5, u_{5max} = 0.8 , respectively. Meanwhile, we assume that the costs for vaccination, treatment, quarantine and sanitation per unit of time is roughly the same.

    We first consider the running payoff function g_{1}(\textbf{X}(t), \textbf{U}(t)) . The following set of values for the cost parameters

    \begin{equation*} \label{51} \begin{split} C_{11} = 1,\quad C_{21} = 1,\quad C_{31} = 1,\quad C_{41} = 1, \quad C_{51} = 1. \end{split} \tag{5.1} \end{equation*} (5.1)

    The optimal control solution is shown in Figure 3(a). The vaccination strategy u_{1}(t) can be reduced 80 weeks later from the beginning of the cholera break, and the quarantine strategy u_{2}(t) , treatment strategy u_{4}(t) and sanitation strategy u_{5}(t) should be maintained in the whole process. Another quarantine strategy u_{3}(t) should be gradually increased over 1 weeks and maintained at a high level until 90 weeks.

    Figure 3.  The graph trajectories of five optimal control strategies based on different running payoff function, in which (a) g_{1}(\textbf{X}(t), \textbf{U}(t)) with the cost parameter (5.1), (b) g_{2}(\textbf{X}(t), \textbf{U}(t)) with the cost parameter (5.2).

    Next, we consider the running payoff function g_{2}(\textbf{X}(t), \textbf{U}(t)) . The following set of values for the cost parameters

    \begin{equation*} \label{52} \begin{split} C_{12} = 1,\quad C_{22} = 1,\quad C_{32} = 1,\quad C_{42} = 1, \quad C_{52} = 1. \end{split} \tag{5.2} \end{equation*} (5.2)

    Similarly, the optimal control solution is shown in Figure 3(b). The vaccination strategy u_{1}(t) can be reduced 40 weeks later from the beginning of the cholera break, and the quarantine strategy u_{2}(t) , treatment strategy u_{4}(t) should be maintained in the whole process. Another quarantine strategy u_{3}(t) should be gradually increased over 10 weeks and maintained at a high level until 90 weeks. In addition, the sanitation strategy u_{5}(t) is not recommended.

    As shown in Figure 3, the values obtained on the optimal control analysis is sensitive to the selected cost function. Therefore, there are different optimal control measures when considering different cost functions. Further, based on the optimal control measures, we can get the effects of the control strategies for the infected individuals. In order to make a comparison, we consider the effects of without any control measures (see Figure 4(a), Figure 5(a)). It is well known that vaccination is a effective measure for cholera prevention and control in a short term. In the following, we explore the influence of vaccine control alone for the infected individuals (see Figure 4(b), Figure 5(b)). By Figure 4 and Figure 5, we can conclude that combining multiple control strategies are most likely to yield the best results in fighting cholera, and the vaccine has a significant control effect on cholera.

    Figure 4.  The graph trajectories of I(t) based on the running payoff function g_{1}(\textbf{X}(t), \textbf{U}(t)) with the cost parameter (5.1).
    Figure 5.  The graph trajectories of I(t) based on the running payoff function g_{2}(\textbf{X}(t), \textbf{U}(t)) with the cost parameter (5.2).

    Therefore, with sufficient costs, vaccination should be combined with other prevention and control strategies to achieve better control in complex emergencies and endemic areas, as recommended by WHO in 2010. With limited costs, only vaccination strategy can control cholera to a great extent. Therefore, vaccination strategy is a feasible and effective method for countries such as Somalia and Yemen with high cholera prevalence and poor economy.

    In this paper, a cholera infection model with vaccination and transmission pathway has been discussed. Here, the total human population is divided into four subpopulation such as susceptible individuals, infected individuals, recovered individuals and vaccinated individuals. In addition, the vibrios in contaminated environment is introduced in the model. Furthermore, the global asymptomatic stability of the disease-free equilibrium and the endemic equilibrium have been completely established by using the Lyapunov's Stability Theorem. If R_{0} < 1 , the disease-free equilibrium is globally asymptomatically stable. If R_{0} > 1 , the endemic equilibrium is globally asymptomatically stable.

    In our paper, we consider the saturation incidence rate to describe the environment-to-human transmission way of cholera. However, if the incidence rate is considered as traditional bilinear, the system (1.2) becomes:

    \begin{equation} \begin{aligned} \dot S& = A-\mu S-\phi S-\left(\beta I+\beta_{1}B\right)S +\eta V,\\ \dot V& = \phi S-\left(\beta I+\beta_{1}B\right)\sigma V-(\mu+\eta)V,\\ \dot I& = \left(\beta I+\beta_{1}B\right)(S+\sigma V)-(\mu+\gamma+d)I,\\ \dot B& = \xi I-\delta B,\\ \dot R& = \gamma I-\mu R. \end{aligned} \end{equation} (6.1)

    In the following, we use the system (6.1) to fit the number of cases in Somalia from 25 to 39 weeks in 2019, and the parameter values obtained are shown in case 1 of Table 2. Further, the fitting results of systems (1.2) and (6.1) are shown in Figure 6(a). By calculating, we obtain the sum-of-squares error of system (1.2) is 6493.4, and that of system (6.1) is 16822. Therefore, it could be more practical to consider the saturation incidence rate to describe the environment-to-human transmission pathway to some extent.

    Table 2.  List of parameters.
    Parameter Case 1 Case 2 Source
    A 7342 - [5]
    \phi 0.5135\times10^{-3} 4.2836\times10^{-3} fitting
    \beta 2.4712\times10^{-10} 4.3771\times10^{-10} fitting
    \beta_1 0.0540\times10^{-9} 0.5959\times10^{-4} fitting
    \mu 0.00038 - [5]
    B(0) 189340 205740 fitting
    R(0) 155 146 fitting

     | Show Table
    DownLoad: CSV
    Figure 6.  Cumulative cases in Somalia between weeks 25 and 39 of 2019, in which the red dots represent the reported data, the blue curve is the solution of system.

    If we remove the growth and death rate, the system (1.2) becomes:

    \begin{equation} \begin{aligned} \dot S& = -\phi S-\left(\beta I+\frac{\beta_{1}B}{K+B}\right)S +\eta V,\\ \dot V& = \phi S-\left(\beta I+\frac{\beta_{1}B}{K+B}\right)\sigma V-\eta V,\\ \dot I& = \left(\beta I+\frac{\beta_{1}B}{K+B}\right)(S+\sigma V)-(\gamma+d)I,\\ \dot B& = \xi I-\delta B,\\ \dot R& = \gamma I. \end{aligned} \end{equation} (6.2)

    The parameter values are shown in case 2 of Table 2, and the results of system (1.2) and system (6.2) are shown in Figure 6(b). From Figure 6(b), we see that there is a small difference in our fitting results. Further, by calculating, we obtain that the sum-of-squares error of system (6.2) is 6481.3, which is close to system (1.2). The reason is that in the 14 weeks of fitting, the growth and death changes of individuals can be ignored. However, the threshold that determines whether a disease is prevalent changes with the growth and death rate. Compared with system (6.2), our model can more accurately describe the spread of cholera, and have more realistic representations of biological cholera infection.

    The authors wish to thank the reviewers and the Editor for their careful reading, valuable comments and suggestions that greatly improved the presentation of this work. This work was supported by the National Natural Science Foundation of China (Nos. 11871316, 11801340, 11371368), and the Natural Science Foundation of Shanxi Province (Nos. 201801D121006, 201801D221007).

    All authors declare no conflicts of interest in this paper.



    [1] R. Tibshirani, M. Saunders, S. Rosset, J. Zhu, K. Knight, Sparsity and smoothness via the fused lasso, J. R. Stat. Soc. Ser. B. Stat. Methodol., 67 (2005), 91–108. https://doi.org/10.1111/j.1467-9868.2005.00490.x doi: 10.1111/j.1467-9868.2005.00490.x
    [2] A. Guntuboyina, D. Lieu, S. Chatterjee, B. Sen, Adaptive risk bounds in univariate total variation denoising and trend filtering, Ann. Statist., 48 (2020), 205–229. https://doi.org/10.1214/18-AOS1799 doi: 10.1214/18-AOS1799
    [3] B. Fang, A. Guntuboyina, B. Sen, Multivariate extensions of isotonic regression and total variation denoising via entire monotonicity and hardy-krause variation, Ann. Statist., 49 (2021), 769–792. https://doi.org/10.1214/20-AOS1977 doi: 10.1214/20-AOS1977
    [4] L. Rudin, S. Osher, E. Fatemi, Nonlinear total variation based noise removal algorithms, Phys. D, 60 (1992), 259–268. https://doi.org/10.1016/0167-2789(92)90242-F doi: 10.1016/0167-2789(92)90242-F
    [5] I. W. Selesnick, A. Parekh, I. Bayram, Convex 1-d total variation denoising with non-convex regularization. IEEE Signal Process. Lett., 22 (2015), 141–144. https://doi.org/10.1109/LSP.2014.2349356
    [6] G. Yuan, B. Ghanem, \ell_0 tv: A sparse optimization method for impulse noise image restoration, IEEE Trans. Pattern Anal. Mach. Intell., 41 (2019), 352–364. https://doi.org/10.1109/TPAMI.2017.2783936 doi: 10.1109/TPAMI.2017.2783936
    [7] J. J. Liu, R. J. Ma, X. Y. Zeng, W. Q. Liu, An efficient non-convex total variation approach for image deblurring and denoising, Appl. Math. Comput., 397 (2021), 125977. https://doi.org/10.1016/j.amc.2021.125977 doi: 10.1016/j.amc.2021.125977
    [8] L. Condat, A direct algorithm for 1-d total variation denoising, IEEE Signal Process. Lett., 20 (2013), 1054–1057. https://doi.org/10.1109/LSP.2013.2278339 doi: 10.1109/LSP.2013.2278339
    [9] L. Dumbge, A. Kovac. Extensions of smoothing via taut strings, Electron. J. Stat., 3 (2009), 41–75. https://doi.org/10.1214/08-EJS216 doi: 10.1214/08-EJS216
    [10] J. Q. Fan, R. Z. Li, Variable selection via nonconcave penalized likelihood and its oracle properties, J. Amer. Statist. Assoc., 96 (2001), 1348–1360. https://doi.org/10.1198/016214501753382273 doi: 10.1198/016214501753382273
    [11] Z. B. Xu, H. Zhang, Y. Wang, X. Y. Chang, Y. Liang, \ell_{1/2} regularization, Sci. China Inf. Sci., 53 (2010), 1159–1169. https://doi.org/10.1007/s11432-010-0090-0 doi: 10.1007/s11432-010-0090-0
    [12] C. H. Zhang, Nearly unbiased variable selection under minimax concave penalty. Ann. Stat., 38 (2010), 894–945. https://doi.org/10.1214/09-AOS729
    [13] T. Zhang, Analysis of multi-stage convex relaxation for sparse regularization, J. Mach. Learn. Res., 11 (20104), 1081–1107.
    [14] H. Q. Du, Y. L. Liu, Minmax-concave total variation denoising. Signal, Image and Video Process., 12 (2018), 1027–1034. https://doi.org/10.1007/s11760-018-1248-2
    [15] I. W. Selesnick, Total variation denoising via the moreau envelope, IEEE Signal Process. Lett., 24 (2017), 216–220. https://doi.org/10.1109/LSP.2017.2647948 doi: 10.1109/LSP.2017.2647948
    [16] I. W. Selesnick, A. Lanza, S. Morigi, F. Sgallari, Non-convex total variation regularization for convex denoising of signals, J. Math. Imaging Vision, 62 (2020), 825–841. https://doi.org/10.1007/s10851-019-00937-5 doi: 10.1007/s10851-019-00937-5
    [17] M. Storath, A. Weinmann, L. Demaret, Jump-sparse and sparse recovery using potts functionals. IEEE Trans. Signal Process., 62 (2014), 3654–3666. https://doi.org/10.1109/TSP.2014.2329263
    [18] J. Frecon, N. Pustelnik, N. Dobigeon, H. Wendt, P. Abry, Bayesian selection for the l2-potts model regularization parameter: 1d piecewise constant signal denoising, IEEE Trans. Signal Process., 65 (2017), 5215–5224. https://doi.org/10.1109/TSP.2017.2715000 doi: 10.1109/TSP.2017.2715000
    [19] R. B. Potts, Some generalized order-disorder transformations, Math. Proc. Cambridge Philos. Soc., 48 (1952), 106–109. https://doi.org/10.1017/S0305004100027419 doi: 10.1017/S0305004100027419
    [20] T. Blumensath, M. E. Davies, Iterative thresholding for sparse approximations, J. Fourier Anal. Appl., 14 (2008), 629–654. https://doi.org/10.1007/s00041-008-9035-z doi: 10.1007/s00041-008-9035-z
    [21] Z. S. Lu, Iterative hard thresholding methods for \ell_0 regularized convex cone programming, Math. Program., 147 (2014), 125–154. https://doi.org/10.1007/s10107-013-0714-4 doi: 10.1007/s10107-013-0714-4
    [22] Z. S. Lu, Y. Zhang, Sparse approximation via penalty decomposition methods, SIAM J. Optim., 23 (2013), 2448–2478. https://doi.org/10.1137/100808071 doi: 10.1137/100808071
    [23] W. Y. Cheng, Z. Chen, Q. J. Hu, An active set barzilar-borwein algorithm for \ell_0 regularized optimization, J. Global Optim., 76 (2020), 769–791. https://doi.org/10.1007/s10898-019-00830-w doi: 10.1007/s10898-019-00830-w
    [24] S. L. Zhou, L. L. Pan, N. H. Xiu, Newton method for \ell_0 regularized optimization, Numer. Algor., 88 (2021), 1541–1570. https://doi.org/10.1007/s11075-021-01085-x doi: 10.1007/s11075-021-01085-x
    [25] A. Beck, N. Hallak, Proximal mapping for symmetric penalty and sparsity, SIAM J. Optim., 28 (2018), 496–527. https://doi.org/10.1137/17M1116544 doi: 10.1137/17M1116544
    [26] R. T. Rockafellar, R. J-B. Wets, Variational Analysis, Springer, Berlin, 1997. https://doi.org/10.1007/978-3-642-02431-3
    [27] M. Storath, A. Weinmann, Fast partitioning of vector-valued images, SIAM J. Imag. Sci., 7 (2014), 1826–1852. https://doi.org/10.1137/130950367 doi: 10.1137/130950367
  • This article has been cited by:

    1. Harendra Pal Singh, Sumit Kaur Bhatia, Riya Jain, Yashika Bahri, 2021, Chapter 1, 978-981-16-1695-2, 1, 10.1007/978-981-16-1696-9_1
    2. Hongyan Zhao, Shaofen Zou, Xia Wang, Yuming Chen, Dynamic behaviors of a cholera model with nonlinear incidences, multiple transmission pathways, and imperfect vaccine, 2024, 70, 1598-5865, 917, 10.1007/s12190-024-01994-9
    3. Jingyun Shen, Shengfu Wang, Linfei Nie, Dynamics of a reaction–diffusion epidemic model with general incidence and protection awareness for multi-transmission pathways, 2023, 74, 0044-2275, 10.1007/s00033-023-02144-0
  • Reader Comments
  • © 2023 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(1700) PDF downloads(80) Cited by(1)

Figures and Tables

Figures(3)  /  Tables(1)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog