In this paper, we study the necessary conditions as well as sufficient conditions for optimality of stochastic SEIR model. The most distinguishing feature, compared with the well-studied SEIR model, is that the model system follows stochastic differential equations (SDEs) driven by Brownian motions. Hamiltonian function is introduced to derive the necessary conditions. Using the explicit formulation of adjoint variables, desired necessary conditions for optimal control results are obtained. We also establish a sufficient condition which is called verification theorem for the stochastic SEIR model.
Project of Shandong Province Higher Educational Science and Technology ProgramJ18KA221National Natural Science Foundation of China61973185Natural Science Foundation of Shandong ProvinceZR2018MF016Development Plan of Young Innovation Team in Colleges and Universities of Shandong Province2019KJN011Shandong Province Key Research and Development Program2018GGX103054)Shandong Academy of Sciences2018BSHZ20081. Introduction
New corona viruses are very harmful to people. Especially, COVID-19 is currently being spread around the world. It has seriously affected people’s lives. Many countries have made good efforts to deal with it and prevent it. At present, the COVID-19 epidemic in China has been basically controlled. It is well known that SEIR models are widely used to model the spreading of infectious diseases in a population. Up to now, many researchers from the world have achieved a variety of both theoretical results and applications, see [1, 2, 7, 9, 15, 16, 18, 20, 21] and the references therein.
However, it is worth pointing that out that SEIR models in the existing literatures are deterministic models. As we know the stochastic events are inevitable in practice, and the stochastic effects that may lead to significant changes, thus, the stochastic SEIR models maybe better to be applied to describe the COVID-19 epidemic. Motivated by the actual situation in reality and the lack of theory, this paper studies the optimal control of stochastic SEIR model. To the best of our knowledge, there were few literatures about the optimal control of epidemic model in the stochastic case. Our main objective is to derive necessary conditions for optimality of the stochastic SEIR model by using the stochastic maximum principle (SMP).
Stochastic optimal control problems have received considerable research attention in recent years due to wide applicability in a number of different fields such as physics, biology, economics, and management science. As it is well known, dynamic programming principle (DPP) and SMP are two main tools to study stochastic control problems. SMP, which provides a necessary condition of an optimal control in stochastic optimal control problems known as the stochastic version of Pontryagin’s type [3–6, 8, 11–14, 19], has been the tool predominantly used to study the stochastic optimal control problems and some stochastic differential game problems.
For example, by using SMP, Xu and Shi [17] obtain the feedback form of optimal control for linear-quadratic-Gaussian (LQG) problems to study stochastic large population system with jump diffusion processes. The standard SMP involves solving the adjoint equation and minimizing the Hamiltonian function. We also followed this in our paper. We should point out that the SEIR model studied in [10] is deterministic case, while our SEIR model is stochastic case, that is, the main difference between our model and the model studied in [10].
The organization of this paper is as follows. Section 2 is devoted to the problem formulation and assumptions. Necessary conditions for optimality are introduced in Section 3. Section 4 aims to prove that the necessary conditions presented in Section 3 are also the sufficient conditions for optimality. Finally, we end our work with some concluding remarks in Section 5.
2. Problem Statement and Assumptions
Throughout this paper, let T>0 be a fixed time horizon and Ω,ℱ,F;ℙ be a given complete filtered probability space, on which independent standard one-dimensional Brownian motions Wit,1≤i≤40≤t≤T are defined. The superscript τ denotes the transpose of vectors or matrices. We suppose that the filtration ℱt≥0 is generated by the independent standard one-dimensional standard Brownian motions Wit,1≤i≤40≤t≤T.
Let L20,T;ℝ denote the set of Lebesgue measurable functions ψ·:0,T⟶ℝ such that ∫0Tψt2dt<+∞. We write φ.∈Lℱ20,T;ℝ if φ.:0,T×Ω⟶ℝ is an ℱt-adapted square-integrable process (i.e., B∫0Tφt2dt<+∞). Let U be nonempty subsets of ℝ. We introduce the admissible control set as(1)U≔v·vt∈U,0≤t≤T;v·∈Lℱ20,T;ℝ.
Now, we introduce our stochastic SEIR model. Let St,Et,It, and Rt represent the number of individuals in the susceptible, exposed, infectious, and recovered compartments at time t, respectively. The total population is denoted by Nt=St+Et+It+Rt. Let ut denote the fraction of susceptible individuals being vaccinated per unit of time.
We input the random disturbance proportionally to each variable value in the model and get the following dynamical system:(2)dSt=bNt−μSt−βStIt−utStdt+σ1StdW1t,dEt=βStIt−μ+ϵEtdt+σ2EtdW2t,dIt=ϵEt−μ+γ+aItdt+σ3ItdW3t,dRt=γIt−μRt+utStdt+σ4RtdW4t,S0&=S_0,E0=E_0,I0=I_0,R0=R_0.
The parameters in the disease transmission model is described as follows: b is the natural birth rate, μ represents the natural death rate, a denotes the death rate due to the disease, and β represents the incidence coefficient of horizontal transmission, and let ϵ be the rate at which the exposed individuals become infectious, and γ is removal rate. Note that the rate of transmission of the disease is βStIt. In the above model, σ=σ1,σ2,σ3,σ4 denotes the random disturbance proportionally to each variable value in the model. The parameters in the model are supposed to be constants for simplicity. For more information about the disease transmission model, we refer the reader to [2, 9, 13] and references within.
The expected cost functional is given by(3)Ju=E∫0TQIt+Gu2tdt,where Q and G are given constants.
The optimal control problem under consideration is as follows.
Problem (P): the objective of the control problem is to find admissible control u∗∈U such that
(4)Ju∗=infu∈UJu.
A control that solves this problem is called optimal.
3. Necessary Conditions for Optimality
This section focuses on the necessary optimality conditions of Problem (P).
In order to apply the necessary conditions for optimal control in the form of maximum principle, we first introduce some notations. Assume that(5)x≔SEIR,A≔b−μbbb0−μ+ϵ000ϵ−μ+γ+a000γ−μ,f1x≔−βSIβSI00,f2x≔−S00S,W≔W1W2W3W4,σx≔diagσ1S,σ2E,σ3I,σ4R,L=0,0,Q,0.
Therefore, stochastic SEIR model (2) can be written as(6)dxt=Axt+f1xt+f2xtutdt+σxtdWt,x0=S0,E0,I0,R0τ.
The corresponding cost functional is (7)Ju=E∫0TLxt+Gu2tdt.
Let x∗,u∗ be the optimal pair of Problem P. The standard Hamiltonian function is given by(8)Hx,p,q,u=p,Ax+f1x+f2xu+σxq+Lx+Gu2,where the adjoint variable p.,q. satisfies (9)dpt=−Hxtdt+qtdWt,pT=0,0,0,0,where Ht:=Hx∗t,pt,qt,u∗t.
Next, we want to obtain the adjoint variable p.,q. explicitly. Let p=pS,pE,pI,pRτandq=qS,qE,qI,qRτ. According to (9), p,q are explicitly given by(10)dpSt=−b−μ−βI∗t−u∗tpStdt−βI∗tpEtdt+σ1qStdW1t,dpEt=−bpSt−μ+ϵpEt+ϵpItdt+σ2qEtdW2t,dpIt=−bpSt−βS∗tpSt+βS∗tpEt−μ+γ+apIt+γpRt+Qdt+σ3qItdW3t,dpRt=−bpSt−μpRt+u∗tpRtdt+σ4qRtdW4t.
Next, we evaluate the necessary condition for the optimal control. By (8), we have(11)H_ux,p,q,u=−Sp_S+Sp_R+2Gu=0,which means(12)u∗t=12GpSt−pRtS∗t.
Now, we summarize the above discussion with the main result of this article.
Theorem 1.
Let x∗,u∗ be the optimal pair of Problem P with x∗:=S∗,E∗,I∗,R∗τ. Then, u∗. fulfills (12), where pS.,pR. admits (10).
4. Sufficient Conditions for Optimality
In this section, we will establish the sufficient maximum principle (also called verification theorem) of Problem P. That is to say, u∗. given in (12) is also the sufficient condition of Problem P.
Theorem 2.
Assume that u∗. fulfills (12) with state trajectory x∗:=S∗,E∗,I∗,R∗τ which id given such that there exist solutions pS.,pR. to the adjoint equation (10). Then, x∗,u∗ is the optimal pair of Problem P.
Proof.
For any u∈U, we consider(13)Ju−Ju∗=E∫0TQIt−I∗t+Gu2t−u∗t2dt.
Applying IÔ’s formula top.,x.−x∗., we obtain(14)EpT,xT−x∗T−p0,x0−x∗0=E∫0Tpt,dxt−x∗t+∫0Txt−x∗t,dpt+∫0Tqtσxt−σx∗tdt=E∫0Tpt,Axt−x∗t+f1xt−f1x∗t+f2xtut−f2x∗tu∗tdt+E∫0Tqtσxt−σx∗tdt−∫0Txt−x∗t,Hxtdt=E∫0THxt,pt,qt,ut−Lxt−Gu2t−Ht−Lx∗t−Gu∗t2dt−E∫0THxt,pt,qt,ut−Ht−xt−x∗t,Hxtdt=E∫0THxt,pt,qt,ut−Ht−xt−x∗t,Hxtdt−E∫0TLxt−x∗t+Gu2t−u∗t2dt.
Combining (6), (9), (13), with (14), one has(15)0=E∫0THxt,pt,qt,ut−Ht−xt−x∗t,Hxtdt−Ju+Ju∗≥−Ju+Ju∗,where, in the last step, we have used the condition of u∗· which fulfills (12).
Therefore, Ju∗≤Ju. Hence, we draw the desired conclusion.
5. Conclusion
Maria do Rosário de et al. [10] considered an optimal control problem with mixed control-state constraint for a SEIR epidemic model of human infectious diseases. Motivated by their pioneering work and the lack of theory, this paper is concerned with the necessary conditions (also, the sufficient conditions) for optimality of the stochastic SEIR model. The model system follows SDEs driven by Brownian motions and with the corresponding cost. It is the first attempt to study this kind of control problem in our technical note, to the authors’ knowledge. Using the explicit formulation of adjoint variables, we obtain the desired necessary conditions for optimal control results.
Some interesting topics deserve further investigations. On the one hand, one may determine the optimal control strategies for the stochastic delayed SIR model and compare it with that presented in this work. On the other hand, we shall investigate some more realistic but complex models, such as considering the effects of impulsive perturbations on the system. We leave these investigations in our future work.
Data Availability
The data used to support the findings of the study are available from the corresponding author upon request.
Conflicts of Interest
The authors declare that there are no conflicts of interest regarding the publication of this paper.
Acknowledgments
This research was partially supported by the Project of Shandong Province Higher Educational Science and Technology Program (Grant no. J18KA221), National Natural Science Foundation of China (Grant no. 61973185), Natural Science Foundation of Shandong Province (Grant no. ZR2018MF016), Development Plan of Young Innovation Team in Colleges and Universities of Shandong Province (Grant no.2019KJN011), Shandong Province Key Research and Development Program (Grant no. 2018GGX103054), and Young Doctor Cooperation Foundation of Qilu University of Technology (Shandong Academy of Sciences) (Grant no. 2018BSHZ2008).
AbdelhadiA.HassanL.Optimal control strategy for SEIR with latent period and a saturated incidence rate201320131410.1155/2013/706848ArmbrusterB.BeckE.Elementary proof of convergence to the mean-field model for the SIR process201675232733910.1007/s00285-016-1086-12-s2.0-85006985108BensoussanA.Lectures on stochastic control1982Berlin, New YorkSpringer162BismutJ.-M.Conjugate convex functions in optimal stochastic control197344238440410.1016/0022-247x(73)90066-82-s2.0-0000929070CadenillasA.A stochastic maximum principle for systems with jumps, with applications to finance200247543344410.1016/s0167-6911(02)00231-12-s2.0-0037121523HaussmannU. G.General necessary conditions for optimal control of stochastic systems19766304810.1007/bfb0120743HanX.LiF.MengX.Dynamics analysis of a nonlinear stochastic SEIR epidemic system with varying population size20182018120KushnerH. J.Necessary conditions for continuous parameter stochastic optimization problems197210355056510.1137/0310041LiuQ.JiangD.ShiN.HayatT.AhmadB.Stationary distribution and extinction of a stochastic SEIR epidemic model with standard incidence2017476586910.1016/j.physa.2017.02.0282-s2.0-85014481371Maria do Rosário deP.KornienkoI.MaurerH.Optimal control of a SEIR model with mixed constraints and L1 costCONTROLO’2014–Proceedings of the 11th Portuguese Conference on Automatic ControlJanuary 2015Guimaraes, Portugal13514510.1007/978-3-319-10380-8_142-s2.0-84907302225PengS.A general stochastic maximum principle for optimal control problems199028496697910.1137/03280542-s2.0-0025462369PengS.Backward stochastic differential equations and applications to optimal control199327212514410.1007/bf011959782-s2.0-0003318214ShiJ.WuZ.Maximum principle for forward-backward stochastic control system with random jumps and applications to finance201023221923110.1007/s11424-010-7224-82-s2.0-77952417631ShiJ.WuZ.A risk-sensitive stochastic maximum principle for optimal control of jump diffusions and its applications201131B2419433SherborneN.MillerJ. C.BlyussK. B.KissI. Z.KissI. Z.Mean-field models for non-Markovian epidemics on networks201876375577810.1007/s00285-017-1155-02-s2.0-85022098767WangX.PengH.ShiB.Optimal vaccination strategy of a constrained time-varying SEIR epidemic model2019673748XuR.ShiJ.-Nash mean-field games for linear-quadratic systems with random jumps and applications201911411110883510.1080/00207179.2019.16519402-s2.0-85071173206YiX.GuoR.QiY.Stabilization of chaotic systems with both uncertainty and disturbance by the UDE-based control method202081624716247710.1109/access.2020.2983674YongJ.ZhouX.19993New York, NY, USASpringer-VerlagZhaoD.ZhangT.YuanS.The threshold of a stochastic SIVS epidemic model with nonlinear saturated incidence201644337237910.1016/j.physa.2015.09.0922-s2.0-84945183164ZhangX.WangK.Stochastic SEIR model with jumps201423913314310.1016/j.amc.2014.04.0612-s2.0-84900834468