Research article Special Issues

Modified artificial rabbits optimization combined with bottlenose dolphin optimizer in feature selection of network intrusion detection

  • For the feature selection of network intrusion detection, the issue of numerous redundant features arises, posing challenges in enhancing detection accuracy and adversely affecting overall performance to some extent. Artificial rabbits optimization (ARO) is capable of reducing redundant features and can be applied for the feature selection of network intrusion detection. The ARO exhibits a slow iteration speed in the exploration phase of the population and is prone to an iterative stagnation condition in the exploitation phase, which hinders its ability to deliver outstanding performance in the aforementioned problems. First, to enhance the global exploration capabilities further, the thinking of ARO incorporates the mud ring feeding strategy from the bottlenose dolphin optimizer (BDO). Simultaneously, for adjusting the exploration and exploitation phases, the ARO employs an adaptive switching mechanism. Second, to avoid the original algorithm getting trapped in the local optimum during the local exploitation phase, the levy flight strategy is adopted. Lastly, the dynamic lens-imaging strategy is introduced to enhance population variety and facilitate escape from the local optimum. Then, this paper proposes a modified ARO, namely LBARO, a hybrid algorithm that combines BDO and ARO, for feature selection in the network intrusion detection model. The LBARO is first empirically evaluated to comprehensively demonstrate the superiority of the proposed algorithm, using 8 benchmark test functions and 4 UCI datasets. Subsequently, the LBARO is integrated into the feature selection process of the network intrusion detection model for classification experimental validation. This integration is validated utilizing the NSL-KDD, UNSW NB-15, and InSDN datasets, respectively. Experimental results indicate that the proposed model based on LBARO successfully reduces redundant characteristics while enhancing the classification capabilities of network intrusion detection.

    Citation: Fukui Li, Hui Xu, Feng Qiu. Modified artificial rabbits optimization combined with bottlenose dolphin optimizer in feature selection of network intrusion detection[J]. Electronic Research Archive, 2024, 32(3): 1770-1800. doi: 10.3934/era.2024081

    Related Papers:

    [1] Shihe Xu, Junde Wu . Qualitative analysis of a time-delayed free boundary problem for tumor growth with angiogenesis and Gibbs-Thomson relation. Mathematical Biosciences and Engineering, 2019, 16(6): 7433-7446. doi: 10.3934/mbe.2019372
    [2] H. J. Alsakaji, F. A. Rihan, K. Udhayakumar, F. El Ktaibi . Stochastic tumor-immune interaction model with external treatments and time delays: An optimal control problem. Mathematical Biosciences and Engineering, 2023, 20(11): 19270-19299. doi: 10.3934/mbe.2023852
    [3] Yuting Ding, Gaoyang Liu, Yong An . Stability and bifurcation analysis of a tumor-immune system with two delays and diffusion. Mathematical Biosciences and Engineering, 2022, 19(2): 1154-1173. doi: 10.3934/mbe.2022053
    [4] Alessandro Bertuzzi, Antonio Fasano, Alberto Gandolfi, Carmela Sinisgalli . Interstitial Pressure And Fluid Motion In Tumor Cords. Mathematical Biosciences and Engineering, 2005, 2(3): 445-460. doi: 10.3934/mbe.2005.2.445
    [5] Jiaxin Nan, Wanbiao Ma . Stability and persistence analysis of a microorganism flocculation model with infinite delay. Mathematical Biosciences and Engineering, 2023, 20(6): 10815-10827. doi: 10.3934/mbe.2023480
    [6] Maria Vittoria Barbarossa, Christina Kuttler, Jonathan Zinsl . Delay equations modeling the effects of phase-specific drugs and immunotherapy on proliferating tumor cells. Mathematical Biosciences and Engineering, 2012, 9(2): 241-257. doi: 10.3934/mbe.2012.9.241
    [7] Marek Bodnar, Monika Joanna Piotrowska, Urszula Foryś . Gompertz model with delays and treatment: Mathematical analysis. Mathematical Biosciences and Engineering, 2013, 10(3): 551-563. doi: 10.3934/mbe.2013.10.551
    [8] Qiaoling Chen, Fengquan Li, Sanyi Tang, Feng Wang . Free boundary problem for a nonlocal time-periodic diffusive competition model. Mathematical Biosciences and Engineering, 2023, 20(9): 16471-16505. doi: 10.3934/mbe.2023735
    [9] Avner Friedman, Harsh Vardhan Jain . A partial differential equation model of metastasized prostatic cancer. Mathematical Biosciences and Engineering, 2013, 10(3): 591-608. doi: 10.3934/mbe.2013.10.591
    [10] Hui Cao, Dongxue Yan, Ao Li . Dynamic analysis of the recurrent epidemic model. Mathematical Biosciences and Engineering, 2019, 16(5): 5972-5990. doi: 10.3934/mbe.2019299
  • For the feature selection of network intrusion detection, the issue of numerous redundant features arises, posing challenges in enhancing detection accuracy and adversely affecting overall performance to some extent. Artificial rabbits optimization (ARO) is capable of reducing redundant features and can be applied for the feature selection of network intrusion detection. The ARO exhibits a slow iteration speed in the exploration phase of the population and is prone to an iterative stagnation condition in the exploitation phase, which hinders its ability to deliver outstanding performance in the aforementioned problems. First, to enhance the global exploration capabilities further, the thinking of ARO incorporates the mud ring feeding strategy from the bottlenose dolphin optimizer (BDO). Simultaneously, for adjusting the exploration and exploitation phases, the ARO employs an adaptive switching mechanism. Second, to avoid the original algorithm getting trapped in the local optimum during the local exploitation phase, the levy flight strategy is adopted. Lastly, the dynamic lens-imaging strategy is introduced to enhance population variety and facilitate escape from the local optimum. Then, this paper proposes a modified ARO, namely LBARO, a hybrid algorithm that combines BDO and ARO, for feature selection in the network intrusion detection model. The LBARO is first empirically evaluated to comprehensively demonstrate the superiority of the proposed algorithm, using 8 benchmark test functions and 4 UCI datasets. Subsequently, the LBARO is integrated into the feature selection process of the network intrusion detection model for classification experimental validation. This integration is validated utilizing the NSL-KDD, UNSW NB-15, and InSDN datasets, respectively. Experimental results indicate that the proposed model based on LBARO successfully reduces redundant characteristics while enhancing the classification capabilities of network intrusion detection.



    Over the past few decades, considerable attention has been paid to the rigorous analysis of mathematical models describing tumor growth and great progress has been achieved. Most work in this direction focuses on the sphere-shaped or nearly sphere-shaped tumor models; see [1,2,3,4,5,6,7,8,9] and the references therein. Observing that the work concerning models for tumors having different geometric configurations from spheroids is less frequent, in this paper we are interested in the situation of tumor cord–a kind of tumor that grows cylindrically around the central blood vessel and receives nutrient materials (such as glucose and oxygen) from the blood vessel [10]. The model only describes the evolution of the tumor cord section perpendicular to the length direction of the blood vessel due to the cord's uniformity in that direction. Assume that the radius of the blood vessel is r0 and denote by J and Γ(t) the section of the blood vessel wall and the section of the exterior surface of the tumor cord, respectively, then J={xR2;|x|=r0}. We also denote by Ω(t) the region of the section of the tumor cord, so that Ω(t) is an annular-like bounded domain in R2 and Ω(t)=JΓ(t). The mathematical formulation of the tumor model under study is as follows:

    cσt(x,t)Δσ(x,t)+σ(x,t)=0,xΩ(t),t>0, (1.1)
    Δp(x,t)=μ[σ(ξ(tτ;x,t),tτ)˜σ],xΩ(t),t>0, (1.2)
    {dξds=p(ξ,s),tτst,ξ=x,s=t, (1.3)
    σ(x,t)=ˉσ,np(x,t)=0,xJ,t>0, (1.4)
    νσ(x,t)=0,p(x,t)=γκ(x,t),xΓ(t),t>0, (1.5)
    V(x,t)=νp(x,t),xΓ(t),t>0, (1.6)
    Γ(t)=Γ0,τt0, (1.7)
    σ(x,t)=σ0(x),xΩ0,τt0, (1.8)
    p(x,t)=p0(x),xΩ0,τt0. (1.9)

    Here, σ and p denote the nutrient concentration and pressure within the tumor, respectively, which are to be determined together with Ω(t), and c=Tdiffusion /Tgrowth is the ratio of the nutrient diffusion time scale to the tumor growth (e.g., tumor doubling) time scale; thus, it is very small and can sometimes be set to be 0 (quasi-steady state approximation). Assume that the time delay τ is reflected between the time at which a cell commences mitosis and the time at which the daughter cells are produced and ξ(s;x,t) represents the cell location at time s as cells are moving with the velocity field V, then the function ξ(s;x,t) satisfies

    {dξds=V(ξ,s),tτst,ξ|s=t=x. (1.10)

    In other words, ξ tracks the path of the cell currently located at x. (1.3) is further derived from (1.10) under the assumption of a porous medium structure for the tumor, where Darcy's law V=p holds true. Because of the presence of time delay, the tumor grows at a rate that is related to the nutrient concentration when it starts mitosis and a combination of the conservation of mass and Darcy's law yields (1.2), in which μ represents the growth intensity of the tumor and ˜σ is the nutrient concentration threshold required for tumor cell growth. Additionally, ˉσ is the nutrient concentration in the blood vessel, ˉσ>˜σ, V, κ and ν denote the normal velocity, the mean curvature and the unit outward normal field of the outer boundary Γ(t), respectively, n denotes the unit outward normal field of the fixed inner boundary J, and γ is the outer surface tension coefficient. Thus, the boundary condition σ=ˉσ on J indicates that the tumor receives constant nutrient supply from the blood vessel, νσ=0 on Γ(t) implies that the nutrient cannot pass through Γ(t), np=0 on J means that tumor cells cannot pass through the blood vessel wall, p=γκ on Γ(t) is due to the cell-to-cell adhesiveness, and V=νp on Γ(t) is the well-known Stefan condition representing that the normal velocity of the tumor cord outer boundary Γ(t) is the same with that of tumor cells adjacent to Γ(t). Finally, σ0(x), p0(x), Γ0 are given initial data and Ω(t)=Ω0 for τt0.

    Before going to our interest, we prefer to recall some relevant works. Models for the growth of the strictly cylindrical tumor cord were studied in [11,12,13]. For the model (1.1)–(1.9) without the time delay, if c=0, Zhou and Cui [14] showed that the unique radially symmetric stationary solution exists and is asymptotically stable for any sufficiently small perturbations. Meanwhile, if c>0, Wu et al. [15] proved that the stationary solution is locally asymptotically stable provided that c is small enough. On the other hand, Zhao and Hu [16] considered the multicell spheroids with time delays. For the case c=0, they analyzed the linear stability of the radially symmetric stationary solution as well as the impact of the time delay.

    Motivated by the works [14,15,16], here we aim to discuss the linear stability of stationary solutions to the problem (1.1)–(1.9) with the quasi-steady-state assumption, i.e., c=0, and investigate the effect of time delay on tumor growth. Our first main result is given below.

    Theorem 1.1. For small time delay τ, the problem (1.1)–(1.9) admits a unique radially symmetric stationary solution.

    Next, in order to deal with the linear stability of the radially symmetric stationary solution, denoted by (σ,p,Ω), where Ω={xR2: r0<r=|x|<R}, we assume that the initial conditions are perturbed as follows:

    Ω(t)={xR2:r0<r<R+ερ0(θ)},τt0,σ(r,θ,t)=σ(r)+εw0(r,θ),p(r,θ,t)=p(r)+εq0(r,θ),τt0.

    The linearized problem of (1.1)–(1.9) at (σ,p,Ω) is then obtained by substituting

    Ω(t):r0<r<R+ερ(θ,t)+O(ε2), (1.11)
    σ(r,θ,t)=σ(r)+εw(r,θ,t)+O(ε2), (1.12)
    p(r,θ,t)=p(r)+εq(r,θ,t)+O(ε2) (1.13)

    into (1.1)–(1.9) and collecting the ε-order terms. Now, we can state the second main result of this paper.

    Theorem 1.2. For small time delay τ, the radially symmetric stationary solution (σ,p,Ω) of (1.1)–(1.9) with c=0 is linearly stable, i.e.,

    max0θ2π|ρ(θ,t)|Ceδt,t>0 (1.14)

    for some positive constants C and δ.

    Remark 1.1. Compared with results of the problem modeling the growth of tumor cord without time delays in [14], the introduction of the time delay does not affect the stability of the radially symmetric stationary solution even under non-radial perturbations. However, as we shall see in Subsection 3.3, the numerical result shows that adding time delay would result in a larger stationary tumor. Moreover, the stronger the growth intensity of the tumor is, the greater the influence of time delay on the size of the stationary tumor is.

    Remark 1.2. Compared with results of the nearly sphere-shaped tumor model with time delays in [16], which state that the radially symmetric stationary solution is linearly stable for small μ in the sense that limtmax0θ2π|ρ(θ,t)(a1cosθ+b1sinθ)|=0 for some constants a1 and b1, the radially symmetric stationary solution of tumor cord with the time delay is linearly stable for any μ>0 in the normal sense.

    This paper is organized as follows. In Section 2, we give the proof of Theorem 1.1 by first transforming the free boundary problem into an equivalent problem with fixed boundary and then applying the contraction mapping principle combined with Lp estimates to this fixed boundary problem. In Section 3, we prove Theorem 1.2 and Remark 1.1 by first introducing the linearization of (1.1)–(1.9) at the radially symmetric stationary solution (σ,p,Ω), and then making a delicate analysis of the expansion in the time delay τ provided that τ is sufficiently small. A brief conclusion in Section 4 completes the paper.

    In this section, we study radially symmetric stationary solutions (σ,p,Ω) to the system (1.1)–(1.9), which satisfy

    Δrσ(r)+σ(r)=0,σ(r0)=ˉσ,σ(R)=0,r0<r<R, (2.1)
    Δrp(r)=μ[σ(ξ(τ;r,0))˜σ],p(r0)=0,p(R)=γR,r0<r<R, (2.2)
    {dξds(s;r,0)=pr(ξ(s;r,0)),τs0,ξ(s;r,0)=r,s=0, (2.3)
    Rr0[σ(ξ(τ;r,0))˜σ]rdr=0, (2.4)

    where Δr is the radial part of the Laplacian in R2.

    Before proceeding further, let us recall that the modified Bessel functions Kn(r) and In(r), standard solutions of the equation

    r2y+ry(r2+n2)y=0,r>0, (2.5)

    have the following properties:

    In+1(r)=In1(r)2nrIn(r),Kn+1(r)=Kn1(r)+2nrKn(r),n1, (2.6)
    In(r)=12[In1(r)+In+1(r)],Kn(r)=12[Kn1(r)+Kn+1(r)],n1, (2.7)
    In(r)=In1(r)nrIn(r), Kn(r)=Kn1(r)nrKn(r),n1, (2.8)
    In(r)=nrIn(r)+In+1(r), Kn(r)=nrKn(r)Kn+1(r),n0, (2.9)
    In(r)Kn+1(r)+In+1(r)Kn(r)=1r,n0 (2.10)

    and

    In(r)>0,Kn(r)<0.

    Proof of Theorem 1.1 In view of (2.5), the solution of (2.1) is clearly given by

    σ(r)=ˉσI0(r)K1(R)+I1(R)K0(r)I0(r0)K1(R)+I1(R)K0(r0). (2.11)

    Introducing the notations:

    ˆr=rr0Rr0,ˆσ(ˆr)=σ(r),ˆp(ˆr)=(Rr0)p(r),ˆξ(s,ˆr,0)=ξ(s,r,0)r0Rr0,

    (2.1)–(2.4) reduces to the following system after dropping the "^" in the above variables:

    2σr2+Rr0r(Rr0)+r0σr=(Rr0)2σ,σ(0)=ˉσ,σ(1)=0, (2.12)
    {2pr2+Rr0r(Rr0)+r0pr=μ(Rr0)3[σ(r+1(Rr0)30τpr((Rr0)ξ(s;r,0)+r0)ds)˜σ],p(0)=0,p(1)=γ(Rr0)R, (2.13)
    {dξds(s;r,0)=1(Rr0)3pr((Rr0)ξ(s;r,0)+r0),τs0,ξ(s;r,0)=r,s=0, (2.14)
    10[r(Rr0)+r0][σ(r+1(Rr0)30τpr((Rr0)ξ(s;r,0)+r0)ds)˜σ]dr=0. (2.15)

    It is clear that (2.12) can be solved explicitly. For convenience, we extend the solution of (2.12) outside [0, 1]:

    σ(r;R)={ˉσI0(r(Rr0)+r0)K1(R)+I1(R)K0(r(Rr0)+r0)I0(r0)K1(R)+I1(R)K0(r0),0r1,ˉσI0(R)K1(R)+I1(R)K0(R)I0(r0)K1(R)+I1(R)K0(r0),1<r2. (2.16)

    Assume that Rmin and Rmax are positive constants to be determined later and r0<Rmin<Rmax. For any R[Rmin,Rmax], we will prove that p is also uniquely determined by applying the contraction mapping principle.

    Noticing that 0 is a lower solution of (2.14), but there is no assurance that ξ(s;r,0)1 for τs0, we suppose ξ(s;r,0)[0,2] and take

    X={pW2,[0,2];pW2,[0,2]M},

    where M>0 is to be determined. For each pX, we first solve for ξ from (2.14) and substitute it into (2.13), then the following system

    {2ˉpr2+Rr0r(Rr0)+r0ˉpr=μ(Rr0)3[σ(r+1(Rr0)30τpr((Rr0)ξ(s;r,0)+r0)ds)˜σ],ˉp(1)=γ(Rr0)R,ˉpr(0)=0 (2.17)

    allows a unique solution ˉpW2,[0,1]. Applying the strong maximum principle combined with the Hopf lemma to (2.1) shows that σ(r;R)ˉσ. Thus, integrating (2.17), we obtain

    1r(Rr0)+r0ˉprL[0,1]μ2(Rmaxr0)2(ˉσ+˜σ), (2.18)
    ˉpL[0,1]γ(Rmaxr0)Rmin+μ4R2max(Rmaxr0)(ˉσ+˜σ), (2.19)
    2ˉpr2L[0,1]3μ2(Rmaxr0)3(ˉσ+˜σ). (2.20)

    Define the mapping

    Lp(r)={ˉp(r),0r1,ˉp(1)+ˉp(1)(r1),1<r2,pX,

    then LpW2,[0,2] and LpW2,[0,2]2ˉpW2,[0,1]. Combining (2.18)–(2.20), we find

    LpW2,[0,2]2{μ2(Rmaxr0)2(ˉσ+˜σ)+3μ2(Rmaxr0)3(ˉσ+˜σ)+γ(Rmaxr0)Rmin+μ4R2max(Rmaxr0)(ˉσ+˜σ)}M1. (2.21)

    If we choose MM1, then LpX by (2.21) and L maps X to itself.

    We now show that L is a contraction. Given p(1),p(2)X, one can first get ξ(1),ξ(2) from the following two systems:

    {dξ(1)ds(s;r,0)=1(Rr0)3p(1)r((Rr0)ξ(1)(s;r,0)+r0),τs0,ξ(1)(s;r,0)=r,s=0, (2.22)
    {dξ(2)ds(s;r,0)=1(Rr0)3p(2)r((Rr0)ξ(2)(s;r,0)+r0),τs0,ξ(2)(s;r,0)=r,s=0. (2.23)

    Integrating (2.22) and (2.23) with regard to s over the interval [τ,0] and making a subtraction yield

    |ξ(1)ξ(2)|τ(Rr0)3maxτs00r1[|p(1)r((Rr0)ξ(1)+r0)p(2)r((Rr0)ξ(1)+r0)|+|p(2)r((Rr0)ξ(1)+r0)p(2)r((Rr0)ξ(2)+r0)|]τ(Rr0)3p(1)p(2)W2,[0,2]+τM(Rr0)2maxτs00r1|ξ(1)ξ(2)|

    for all τs0 and 0r1. Consequently,

    maxτs00r1|ξ(1)ξ(2)|τ(Rr0)3τM(Rr0)p(1)p(2)W2,[0,2]. (2.24)

    Next, we substitute ξ(1),ξ(2) into (2.17) and solve for ˉp(1) and ˉp(2), respectively, then it follows from (2.17) that (ˉp(1)ˉp(2))(1)=0, r(ˉp(1)ˉp(2))(0)=0 and

    2r2(ˉp(1)ˉp(2))Rr0r(Rr0)+r0r(ˉp(1)ˉp(2))=μ(Rr0)3[σ(r+1(Rr0)30τp(1)r((Rr0)ξ(1)(s;r,0)+r0)ds)σ(r+1(Rr0)30τp(2)r((Rr0)ξ(2)(s;r,0)+r0)ds)].

    Using (2.24), we derive

    1r(Rr0)+r0r(ˉp(1)ˉp(2))L[0,1]μ2(Rr0)2σ(r+1(Rr0)30τp(1)r((Rr0)ξ(1)+r0)ds)σ(r+1(Rr0)30τp(2)r((Rr0)ξ(2)+r0)ds)L[0,1]μ2(Rr0)σrL[0,2]0τ(p(1)r((Rr0)ξ(1)+r0)p(2)r((Rr0)ξ(2)+r0))dsμτ2(Rr0)σrL[0,2](p(1)p(2)W2,[0,2]+(Rr0)p(2)W2,[0,2]maxτs00r1|ξ(1)ξ(2)|)M2τp(1)p(2)W2,[0,2]

    and similarly,

    ˉp(1)ˉp(2)L[0,1]M3τp(1)p(2)W2,[0,2],2r2(ˉp(1)ˉp(2))L[0,1]M4τp(1)p(2)W2,[0,2],

    where

    M2=μˉσRmax2r0(Rmaxr0)3(Rminr0)2Mτ,M3=μˉσR3max4r0(Rmaxr0)2(Rminr0)2Mτ,M4=3μˉσRmax2r0(Rmaxr0)4(Rminr0)2Mτ.

    Here, we employed the fact that

    σrL[0,2]=σrL[0,1]ˉσRmaxr0(Rmaxr0)2

    by (2.1) and σˉσ. Let M5=M2+M3+M4, then M5 is independent of τ and

    ˉp(1)ˉp(2)W2,[0,1]M5τp(1)p(2)W2,[0,2],

    which together with Lp(1)(1)=Lp(2)(1)=γ(Rr0)R and (Lp(1))(0)=(Lp(2))(0)=0 implies that

    Lp(1)Lp(2)W2,[0,2]2M5τp(1)p(2)W2,[0,2].

    Hence, if τ is sufficiently small such that 2M5τ<1, then we derive a contracting mapping L. The existence and uniqueness of p are therefore obtained.

    It suffices to prove that there exists a unique R[Rmin,Rmax] satisfying (2.15). Substituting (2.16) into (2.15), we find that it is equivalent to solving the following equation for R:

    G(R,τ)=10r(Rr0)+r0R+r0[σ(r+1(Rr0)30τpr((Rr0)ξ(s;r,0)+r0)ds)˜σ]dr=0.

    Clearly,

    G(R,0)=10(σ(r;R)˜σ)r(Rr0)+r0R+r0dr=10σ(r;R)r(Rr0)+r0R+r0dr˜σ2.

    Using Lemma 3.1 and Theorem 3.2 in [14] and the condition ˉσ>˜σ, we know that

    limRr0G(R,0)=ˉσ˜σ2>0,limRG(R,0)=˜σ2<0,G(R,0)R<0,

    which implies that the equation G(R,0)=0 has a unique solution, denoted by RS, and

    G(12(RS+r0),0)>0,G(32RS,0)<0.

    Since

    G(R,τ)R=G(R,0)R+2G(R,η)Rττ+O(τ2),0ητ

    when τ is sufficiently small, G(R,τ)R and G(R,0)R have the same sign. Thus, G(R,τ) is monotone decreasing in R. Using the fact that G(R,τ) is continuous in τ, we further have

    G(12(RS+r0),τ)>0,G(32RS,τ)<0.

    Hence, when τ is sufficiently small, the equation G(R,τ)=0 has a unique solution R. Taking Rmin=12(RS+r0) and Rmax=32RS, we complete the proof of the theorem.

    This section is devoted to the linear stability of the radially symmetric stationary solution (σ,p,Ω) of the problem (1.1)–(1.9) and the effect of time delay on the stability and the size of the stationary tumor. Let (σ,p,Ω(t)), given by (1.11)–(1.13), be solutions to (1.1)–(1.9), and denote by er, eθ the unit normal vectors in r, θ directions, respectively. Written in the rectangular coordinates in R2,

    er=(cosθ,sinθ)T,eθ=(sinθ,cosθ)T.

    Using the notation ξ1(s;r,θ,t), ξ2(s;r,θ,t) for the polar radius and angle of ξ(s;r,θ,t), respectively, we have

    ξ(s;r,θ,t)=ξ1(s;r,θ,t)er(ξ)=ξ1(s;r,θ,t)(cosξ2(s;r,θ,t),sinξ2(s;r,θ,t))T.

    Expand ξ1, ξ2 in ε as

    {ξ1=ξ10+εξ11+O(ε2),ξ2=ξ20+εξ21+O(ε2), (3.1)

    then we derive from (1.3) and (1.13) that

    {dξ10ds=pr(ξ10),tτst,ξ10|s=t=r; (3.2)
    {dξ11ds=2pr2(ξ10)ξ11qr(ξ10,ξ20,s),tτst,ξ11|s=t=0; (3.3)
    {dξ20ds=0,tτst,ξ20|s=t=θ; (3.4)
    {dξ21ds=1ξ210qθ(ξ10,ξ20,s),tτst,ξ21|s=t=0. (3.5)

    It is evident that ξ20θ. Noticing that the equation for ξ10 is the same as that for ξ in the radially symmetric case, ξ10 is independent of θ.

    Substituting (1.11)–(1.13) and (3.1)–(3.5) into (1.1), (1.2), (1.4)–(1.6), using the mean-curvature formula in the 2-dimensional case for the curve r=ρ(θ):

    κ=ρ2+2ρ2θρρθθ(ρ2+(ρθ)2)3/2

    and collecting the ε-order terms, we obtain the linearized system in BR×{t>0}:

    Δω(r,θ,t)=ω(r,θ,t),ω(r0,θ,t)=0,ωr(R,θ,t)+σ(R)ρ(θ,t)=0, (3.6)
    {Δq(r,θ,t)=μσr(ξ10(tτ;r,t))ξ11(tτ;r,θ,t)μw(ξ10(tτ;r,t),θ,tτ),qr(r0,θ,t)=0,q(R,θ,t)+γR2(ρ(θ,t)+2ρθ2(θ,t))=0, (3.7)
    ρ(θ,t)t=qr(R,θ,t)2pr2(R,θ,t)ρ(θ,t). (3.8)

    Due to the presence of the time delay, the linearization problem (3.6)–(3.8) cannot be solved explicitly. Assume that ω,q, ρ and ξ11 have the following Fourier expansions:

    {ω(r,θ,t)=A0(r,t)+n=1[An(r,t)cosnθ+Bn(r,t)sinnθ],q(r,θ,t)=E0(r,t)+n=1[En(r,t)cosnθ+Fn(r,t)sinnθ],ρ(θ,t)=a0(t)+n=1[an(t)cosnθ+bn(t)sinnθ],ξ11(s;r,θ,t)=e0(s;r,t)+n=1[en(s;r,t)cosnθ+fn(s;r,t)sinnθ]. (3.9)

    Substituting (3.9) into (3.6)–(3.8) yields the following system in BR×{t>0}:

    {2Anr2(r,t)+1rAnr(r,t)n2r2An(r,t)=An(r,t),An(r0,t)=0,Anr(R,t)+σ(R)an(t)=0, (3.10)
    {2Bnr2(r,t)+1rBnr(r,t)n2r2Bn(r,t)=Bn(r,t),Bn(r0,t)=0,Bnr(R,t)+σ(R)bn(t)=0, (3.11)
    {2Enr2(r,t)+1rEnr(r,t)n2r2En(r,t)=μσr(ξ10(tτ;r,t))en(tτ;r,t)μAn(ξ10(tτ;r,t),tτ),Enr(r0,t)=0,En(R,t)+γ(1n2)R2an(t)=0, (3.12)
    {2Fnr2(r,t)+1rFnr(r,t)n2r2Fn(r,t)=μσr(ξ10(tτ;r,t))fn(tτ;r,t)μBn(ξ10(tτ;r,t),tτ),Fnr(r0,t)=0,Fn(R,t)+γ(1n2)R2bn(t)=0, (3.13)
    {ens(s;r,t)=2pr2(ξ10)en(s;r,t)Enr(ξ10,s),tτst,ens=t=0, (3.14)
    {fns(s;r,t)=2pr2(ξ10)fn(s;r,t)Fnr(ξ10,s),tτst,fns=t=0, (3.15)
    dan(t)dt=2pr2(R)an(t)Enr(R,t), (3.16)
    dbn(t)dt=2pr2(R)bn(t)Fnr(R,t). (3.17)

    Since it is impossible to solve the systems (2.1)–(2.4) and (3.10)–(3.17) explicitly and the time delay τ is actually very small, in what follows, we analyze the expansion in τ for (2.1)–(2.4) and (3.10)–(3.17).

    Let

    R=R0+τR1+O(τ2),σ=σ0+τσ1+O(τ2),p=p0+τp1+O(τ2),An=A0n+τA1n+O(τ2),Bn=B0n+τB1n+O(τ2),En=E0n+τE1n+O(τ2),Fn=F0n+τF1n+O(τ2),an=a0n+τa1n+O(τ2),bn=b0n+τb1n+O(τ2).

    Substitute these expansions into (2.1)–(2.4) and (3.10)–(3.17). Since an(t) and bn(t) have the same asymptotic behavior at , we will only make an analysis of an(t). For this, we discuss the expansions of R, σ, p, An, En and an. Since the equations for the expansions of σ, p, An, En and an are the same as those in [16], here we only compute the expansions of the boundary conditions of σ, p, An and En.

    Expansions of the boundary conditions of σ:

    It follows from (2.10) and (2.11) that

    σ(r)=ˉσK1(R)I0(r)+I1(R)K0(r)I0(r0)K1(R)+I1(R)K0(r0)=ˉσK1(R0)I0(r)+I1(R0)K0(r)I0(r0)K1(R0)+I1(R0)K0(r0)+τˉσR1R0I0(r0)K0(r)K0(r0)I0(r)[I0(r0)K1(R0)+I1(R0)K0(r0)]2+O(τ2),

    which implies

    σ0(r)=ˉσK1(R0)I0(r)+I1(R0)K0(r)I0(r0)K1(R0)+I1(R0)K0(r0), (3.18)
    σ1(r)=ˉσR1R0I0(r0)K0(r)K0(r0)I0(r)[I0(r0)K1(R0)+I1(R0)K0(r0)]2. (3.19)

    By the boundary conditions in (2.1), we find

    σ0(r0)+τσ1(r0)+O(τ2)=ˉσ,σ0r(R0)+τ2σ0r2(R0)R1+τσ1r(R0)+O(τ2)=0.

    Expansions of the boundary conditions of p:

    One obtains from the boundary conditions in (2.2) that

    p0r(r0)+τp1r(r0)+O(τ2)=0,p0(R0)+τp0r(R0)R1+τp1(R0)+O(τ2)=γR0τγR1(R0)2+O(τ2).

    Expansion of (2.4):

    In view of (4.31) in [16], there holds

    0=Rr0[σ(ξ(τ;r,0))˜σ]rdr=Rr0[σ0(r)˜σ]rdr+τR0r0(σ0r(r)p0r(r)+σ1(r))rdr+O(τ2). (3.20)

    Using (3.18), we compute

    Rr0[σ0(r)˜σ]rdr=Rr0(ˉσK1(R0)I0(r)+I1(R0)K0(r)I0(r0)K1(R0)+I1(R0)K0(r0)˜σ)rdr=ˉσr0I1(R0)K1(r0)I1(r0)K1(R0)I0(r0)K1(R0)+I1(R0)K0(r0)+˜σ2[r20(R0)2]+τR1(ˉσI0(r0)K1(R0)+I1(R0)K0(r0)˜σR0)+O(τ2). (3.21)

    A combination of (3.20) and (3.21) gives

    τ[ˉσR1I0(r0)K1(R0)+I1(R0)K0(r0)˜σR0R1+R0r0(σ0r(r)p0r(r)+σ1(r))rdr]+ˉσr0I1(R0)K1(r0)I1(r0)K1(R0)I0(r0)K1(R0)+I1(R0)K0(r0)+˜σ2[r20(R0)2]+O(τ2)=0. (3.22)

    Expansions of the boundary conditions of An:

    We derive from the boundary conditions in (3.10) that

    A0n(r0,t)+τA1n(r0,t)+O(τ2)=0,0=A0nr(R0+τR1,t)+τA1nr(R0,t)+[σ0(R0+τR1)+τσ1(R0)][a0n(t)+τa1n(t)]+O(τ2)=A0nr(R0,t)+σ0(R0)a0n(t)+τ(2A0nr2(R0,t)R1+A1nr(R0,t)+σ0r(R0,t)R1a0n(t)+σ0(R0)a1n(t)+σ1(R0)a0n(t))+O(τ2).

    Expansions of the boundary conditions of En:

    Substituting the expansion of En into the boundary conditions in (3.12) yields

    E0nr(r0,t)+τE1nr(r0,t)+O(τ2)=0,0=E0n(R0+τR1,t)+τE1n(R0,t)+γ1n2(R0+τR1)2[a0n(t)+τa1n(t)]+O(τ2)=E0n(R0,t)+γ1n2(R0)2a0n(t)+τ(E0nr(R0,t)R1+E1n(R0,t)2γ1n2(R0)3R1a0n(t)+γ1n2(R0)2a1n(t))+O(τ2).

    Collecting all zeroth-order terms in τ leads to the following system for r0<r<R0:

    2σ0r21rσ0r=σ0,σ0(r0)=ˉσ,σ0r(R0)=0, (3.23)
    2p0r21rp0r=μ(σ0˜σ),p0r(r0)=0,p0(R0)=γR0, (3.24)
    ˉσr0I1(R0)K1(r0)I1(r0)K1(R0)I0(r0)K1(R0)+I1(R0)K0(r0)+˜σ2[r20(R0)2]=0, (3.25)
    {2A0nr21rA0nr+(n2r2+1)A0n=0,A0n(r0,t)=0,A0nr(R0,t)+σ0(R0)a0n(t)=0, (3.26)
    {2B0nr21rB0nr+(n2r2+1)B0n=0,B0n(r0,t)=0,B0nr(R0,t)+σ0(R0)b0n(t)=0, (3.27)
    {2E0nr21rE0nr+n2r2E0n=μA0n,E0nr(r0,t)=0,E0n(R0,t)=γn21(R0)2a0n(t), (3.28)
    {2F0nr21rF0nr+n2r2F0n=μB0n,F0nr(r0,t)=0,F0n(R0,t)=γn21(R0)2b0n(t), (3.29)
    da0n(t)dt=2p0r2(R0)a0n(t)E0nr(R0,t), (3.30)
    db0n(t)dt=2p0r2(R0)b0n(t)F0nr(R0,t). (3.31)

    A direct calculation gives

    p0r(r)=μˉσI0(r0)K1(R0)+I1(R0)K0(r0){K1(R0)(r0rI1(r0)I1(r))+I1(R0)(K1(r)r0rK1(r0))}+μ˜σr2μ˜σr202r, (3.32)
    2p0r2(R0)=μˉσR0[(R0)2r20]2r0R0[I1(R0)K1(r0)I1(r0)K1(R0)](R0)2+r20I0(r0)K1(R0)+I1(R0)K0(r0), (3.33)
    A0n(r,t)=ˉσa0n(t)hn(r0,R0)R0Kn(r0)In(r)In(r0)Kn(r)I0(r0)K1(R0)+I1(R0)K0(r0), (3.34)
    A0nr(r,t)=ˉσa0n(t)hn(r0,R0)R0[I0(r0)K1(R0)+I1(R0)K0(r0)]1hn(r0,r), (3.35)
    2A0nr2(R0,t)=ˉσa0n(t)hn(r0,R0)gn(r0,R0)R0[I0(r0)K1(R0)+I1(R0)K0(r0)], (3.36)

    where

    hn(r0,x)=1In(r0)(nxKn(x)Kn+1(x))Kn(r0)(nxIn(x)+In+1(x)),gn(r0,x)=In(r0)[(1+n(n1)x2)Kn(x)+1xKn+1(x)]Kn(r0)[(1+n(n1)x2)In(x)1xIn+1(x)]. (3.37)

    Let η0n=E0n+μA0n, then we find from (3.26) and (3.28) that η0n satisfies

    2η0nr21rη0nr+n2r2η0n=0,

    whose solution is

    η0n(r,t)=C1(t)rn+C2(t)rn

    and thus,

    E0n(r,t)=η0n(r,t)μA0n(r,t)=C1(t)rn+C2(t)rnμA0n(r,t), (3.38)

    where C1(t) and C2(t) are to be determined by the boundary conditions in (3.28). By (2.10), (3.34), (3.35) and (3.37), we get

    C1(t)=μˉσa0n(t)hn(r0,R0)nR0[(R0)2n+r2n0]n(R0)n[In(R0)Kn(r0)In(r0)Kn(R0)]+rn0I0(r0)K1(R0)+I1(R0)K0(r0)+(n21)γa0n(t)(R0)2n+r2n0(R0)n2, (3.39)
    C2(t)=μˉσa0n(t)hn(r0,R0)rn0(R0)nnR0[(R0)2n+r2n0]nrn0[In(R0)Kn(r0)In(r0)Kn(R0)](R0)nI0(r0)K1(R0)+I1(R0)K0(r0)+(n21)γa0n(t)(R0)2n+r2n0r2n0(R0)n2. (3.40)

    Using (3.35) and (3.38)–(3.40), we further derive

    E0nr(r,t)=μˉσa0n(t)hn(r0,R0)R0[I0(r0)K1(R0)+I1(R0)K0(r0)]{rn0rn1+(R0)2nrn1(R0)2n+r2n0+n(R0)nrn1r2n0rn1(R0)2n+r2n0[In(R0)Kn(r0)In(r0)Kn(R0)]+1hn(r0,r)}+n(n21)γa0n(t)(R0)n2rn1r2n0rn1(R0)2n+r2n0. (3.41)

    Substituting (3.33) and (3.41) into (3.30) yields

    da0n(t)dt=Un(r0,R0)a0n(t), (3.42)

    whose solution is explicitly given by

    a0n(t)=a0n(0)exp{Un(r0,R0)t}. (3.43)

    Here,

    Un(r0,R0)=μˉσhn(r0,R0)I0(r0)K1(R0)+I1(R0)K0(r0)[2r0(I1(r0)K1(R0)I1(R0)K1(r0))hn(r0,R0)[(R0)2r20]+n(R0)2(R0)2nr2n0(R0)2n+r2n0(In(r0)Kn(R0)In(R0)Kn(r0))2rn0(R0)n2(R0)2n+r2n0]γn(n21)(R0)3(R0)2nr2n0(R0)2n+r2n0. (3.44)

    It was proven in Lemma 4.4 of [14] that Un(r0,R0)<0 for any n0. Thus, we have the following:

    Lemma 3.1. For any n0, there exists δ>0 such that |an0(t)||an0(0)|eδt for all t>0.

    Lemma 3.1 shows that a0n(t) decays to 0 exponentially at +; hence, when τ=0, the radially symmetric stationary solution is asymptotically stable for all μ>0.

    Recalling that R=R0+τR1+O(τ2), in order to see the effect of the time delay τ on the size of the stationary tumor, in this subsection we discuss the sign of R1 by a theoretical analysis combined with numerical simulations.

    We obtain from (3.22) that

    ˉσR1I0(r0)K1(R0)+I1(R0)K0(r0)˜σR0R1+R0r0(σ0r(r)p0r(r)+σ1(r))rdr=0, (3.45)

    then by using (2.6)–(2.10), (3.18), (3.19), (3.25) and (3.32), one can solve (3.45) to obtain

    R1=μˉσ2R0T(r0,R0)S(r0,R0), (3.46)

    where

    T(r0,R0)=1r20[I0(r0)K1(R0)+I1(R0)K0(r0)]2+r20(R0)2r204(R0)2r20[I1(R0)K1(r0)I1(r0)K1(R0)]2,S(r0,R0)=2r0(R0)2r20[I1(r0)K1(R0)I1(R0)K1(r0)][I0(r0)K1(R0)+I1(R0)K0(r0)]+1(R0)2.

    Since

    U0(r0,R0)=μˉσ[I0(r0)K1(R0)+I1(R0)K0(r0)]2S(r0,R0)

    by (3.44), we know S(r0,R0)<0. Additionally, we numerically compute the function T(r0,R0) and find it is positive (see Figure 1). Hence, it follows from (3.46) that R1>0 and R1 is monotone increasing in μ.

    Figure 1.  The graph of min0.0001r0<R0T(r0,R0) with R0[0.0001,10].

    Remark 3.1. The discussion above indicates that the presence of the time delay leads to a larger stationary tumor. Furthermore, the bigger the tumor aggressive parameter μ is, the greater the effect of time delay on the size of the stationary tumor is.

    Now, we tackle the system consisting of all the first-order terms in τ for r0<r<R0:

    2σ1r21rσ1r=σ1,σ1(r0)=0,σ1r(R0)+2σ0r2(R0)R1=0, (3.47)
    {2p1r21rp1r=μσ0rp0r+μσ1,p1r(r0)=0,p1(R0)=γR1(R0)2p0r(R0)R1, (3.48)
    ˉσR1I0(r0)K1(R0)+I1(R0)K0(r0)˜σR0R1+R0r0(σ0r(r)p0r(r)+σ1(r))rdr=0, (3.49)
    {2A1nr21rA1nr+(n2r2+1)A1n=0,A1n(r0,t)=0,2A0nr2(R0,t)R1+A1nr(R0,t)+σ0(R0)a1n(t)+σ1(R0)a0n(t)=0, (3.50)
    {2B1nr21rB1nr+(n2r2+1)B1n=0,B1n(r0,t)=0,2B0nr2(R0,t)R1+B1nr(R0,t)+σ0(R0)b1n(t)+σ1(R0)b0n(t)=0, (3.51)
    {2E1nr21rE1nr+n2r2E1n=μσ0rE0nr+μA0nrp0rμA0nt+μA1n,E1nr(r0,t)=0,E1n(R0,t)=γn21(R0)2a1n(t)E0nr(R0,t)R12γn21(R0)3R1a0n(t), (3.52)
    {2F1nr21rF1nr+n2r2F1n=μσ0rF0nr+μB0nrp0rμB0nt+μB1n,F1nr(r0,t)=0,F1n(R0,t)=γn21(R0)2b1n(t)F0nr(R0,t)R12γn21(R0)3R1b0n(t), (3.53)
    da1n(t)dt=2p0r2(R0)a1n(t)3p0r3(R0)R1a0n(t)2p1r2(R0)a0n(t)2E0nr2(R0,t)R1E1nr(R0,t), (3.54)
    db1n(t)dt=2p0r2(R0)b1n(t)3p0r3(R0)R1b0n(t)2p1r2(R0)b0n(t)2F0nr2(R0,t)R1F1nr(R0,t). (3.55)

    To obtain the asymptotic behavior of a1n(t) as , by (3.54) and the boundedness of the modified Bessel functions In(r) and Kn(r) on [r0,R0], it suffices to analyze E1nr(R0,t). For this purpose, in view of (3.52), we first compute A1n(r,t). Solving (3.50) yields

    A1n(r,t)=ˉσhn(r0,R0)R0In(r0)Kn(r)Kn(r0)In(r)I0(r0)K1(R0)+I1(R0)K0(r0)Vn(r0,R0,R1,a0n(t),a1n(t)) (3.56)

    with

    Vn(r0,R0,R1,a0n(t),a1n(t))=a0n(t)R1hn(r0,R0)gn(r0,R0)a1n(t)+a0n(t)R1I0(R0)K0(r0)I0(r0)K0(R0)I0(r0)K1(R0)+I1(R0)K0(r0),

    where we have employed (2.8)–(2.10), (3.19), (3.36) and (3.37). Furthermore,

    A1nr(r,t)=ˉσhn(r0,R0)Vn(r0,R0,R1,a0n(t),a1n(t))R0[I0(r0)K1(R0)+I1(R0)K0(r0)]1hn(r0,r). (3.57)

    Next, being similar to the computation of E0n, we set η1n=E1n+μA1n, then we derive from (3.50), (3.52) and (3.56) that

    {2η1nr21rη1nr+n2r2η1n=μσ0rE0nr+μA0nrp0rμA0nt,η1nr(r0,t)=μA1nr(r0,t),η1n(R0,t)=E1n(R0,t)+μA1n(R0,t). (3.58)

    For brevity, we introduce the differential operator Ln=rr1rr+n2r2 and write η1n=u(1)n+u(2)n+u(3)n+u(4)n, where u(1)n, u(2)n, u(3)n and u(4)n solve the following problems, respectively:

    {Lnu(1)n=μσ0rE0nr,u(1)nr(r0,t)=0,u(1)n(R0,t)=0; (3.59)
    {Lnu(2)n=μA0nrp0r,u(2)nr(r0,t)=0,u(2)n(R0,t)=0; (3.60)
    {Lnu(3)n=μA0nt,u(3)nr(r0,t)=0,u(3)n(R0,t)=0; (3.61)
    {Lnu(4)n=0,u(4)nr(r0,t)=μA1nr(r0,t),u(4)n(R0,t)=E1n(R0,t)+μA1n(R0,t). (3.62)

    Let us first estimate u(1)n. By (3.18), (3.41) and (3.59), we have

    Lnu(1)n=μˉσa0n(t)K1(R0)I1(r)I1(R0)K1(r)I0(r0)K1(R0)+I1(R0)K0(r0){n(n21)γ(R0)n2rn1r2n0rn1(R0)2n+r2n0+μˉσhn(r0,R0)R0[I0(r0)K1(R0)+I1(R0)K0(r0)][1hn(r0,r)+rn0rn1+(R0)2nrn1(R0)2n+r2n0+n(R0)nrn1r2n0rn1(R0)2n+r2n0(In(R0)Kn(r0)In(r0)Kn(R0))]}. (3.63)

    Based on the properties of the modified Bessel functions In(r) and Kn(r), the righthand side of (3.63) is less than Q(n)a0n(t) when r0r<R0. Here, Q(n) denotes a polynomial function of n. Similar estimates can be established for u(2)n and u(3)n by (3.60) and (3.61).

    Lemma 3.2. Consider the elliptic problem

    Δω(x,t)+n2|x|2ω(x,t)=b(x,t),xΩR, (3.64)
    nω||x|=r0=0,ω||x|=R=0, (3.65)

    where ΩR={xR2:r0<|x|<R}. If b(x,t)=b(|x|,t) and b(,t)L2(ΩR), then the problem (3.64) and (3.65) admits a unique solution ω in H2(ΩR) with estimates

    ω(,t)H2(ΩR)C(Rr0|b(r,t)|2rdr)1/2; (3.66)
    νω(,t)L(BR)C(Rr0|b(r,t)|2rdr)1/2, (3.67)

    where the constant C in (3.66) and (3.67) is independent of n.

    The lemma can be proven by combining the proofs of [16, Lemma 4.6] and [17, Lemma 3.2]. The details are omitted here.

    Lemma 3.2 ensures the existence and uniqueness of u(k)n in H2(Ω) for k=1,2,3. Furthermore, there holds

    |u(1)nr(R0,t)|+|u(2)nr(R0,t)|+|u(3)nr(R0,t)|Ceδt. (3.68)

    Obviously, the solution u(4)n to the problem (3.62) has the form:

    u(4)n(r,t)=C5(t)rn+C6(t)rn, (3.69)

    where C5(t) and C6(t) are determined by the boundary conditions in (3.62). Using (2.10), (3.41), (3.56), (3.57) and the boundary conditions in (3.52), we get

    C5(t)=a1n(t)(R0)2n+r2n0{μˉσhn(r0,R0)R0[I0(r0)K1(R0)+I1(R0)K0(r0)][rn0n+(R0)n(Kn(r0)In(R0)In(r0)Kn(R0))]+γ(n21)(R0)n2}+H1(r0,R0,R1)a0n(t), (3.70)
    (3.71)

    where , are functions of , and .

    Now, since

    we derive

    (3.72)

    By (3.33), (3.57) and (3.69)–(3.72), we obtain from (3.54) that

    where is a known function of , , , and satisfies

    (3.73)

    Thus, using Lemma 3.1, (3.68) and (3.73) gives

    (3.74)

    In addition, for , Lemma 3.1 implies

    Therefore, applying [16, Lemma 4.7] to (3.74) yields

    (3.75)

    i.e., decays exponentially as . Noticing that and have the same asymptotic behavior, we also have

    (3.76)

    Proof of Theorem 1.2 The desired result (1.14) follows from Lemma 3.1, (3.75) and (3.76). The proof is complete.

    Remark 3.2. The results on tumor cord without time delays in [14] show that the radially symmetric stationary solution is asymptotically stable under nonradially symmetric perturbations. Here, our Theorem 1.2 says that such asymptotic stability does not be affected by small time delay.

    In this paper, we have investigated the effects of a time delay in cell proliferation on the growth of tumor cords, where the domain is a bounded subset in and its boundary consists of two disjoint closed curves, one fixed and the other moving and a priori unknown. The existence, uniqueness and linear stability of the radially symmetric stationary solution were studied.

    Here are some interesting findings. 1) Adding the time delay would not change the stability of the radially symmetric stationary solution when compared with the same system without delay [14], but adding the time delay would result in a larger stationary tumor. The bigger the tumor growth intensity is, the greater impact that time delay has on the size of the stationary tumor. 2) By the result of [16], we know that for tumor spheroids with the same time delay, there exists a threshold for the tumor aggressiveness constant such that only for , the radially symmetric stationary solution is linearly stable under non-radial perturbations. For tumor cords, however, from Theorem 1.2 we saw that the radially symmetric stationary solution is always linearly stable, regardless of the value of . It showed that there is an essential difference between tumor cords and tumor spheroids with the same time delay.

    We think that the linear stability analysis for the full system without quasi-steady state simplification, i.e., , may be very challenging, which we expect to solve in future work.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    The authors would like to thank the reviewers for their valuable suggestions. This work was partly supported by the National Natural Science Foundation of China (No. 12161045 and No. 12261047), Jiangxi Provincial Natural Science Foundation (No. 20224BCD41001 and No. 20232BAB201010), the Science and Technology Planning Project from Educational Commission of Jiangxi Province, China (No. GJJ2200319), the Scientific Research Project of Hunan Provincial Education Department, China (No. 22B0725, No. 23B0670 and No. 23C0234), and the Research Initiation Project of Hengyang Normal University, China (No. 2022QD01).

    The authors declare there is no conflict of interest.



    [1] M. H. Nasir, S. A. Khan, M. M. Khan, M. Fatima, Swarm intelligence inspired intrusion detection systems—a systematic literature review, Comput. Networks, 205 (2022), 1389–1286. https://doi.org/10.1016/j.comnet.2021.108708 doi: 10.1016/j.comnet.2021.108708
    [2] T. Dokeroglu, A. Deniz, H. E. Kiziloz, A comprehensive survey on recent metaheuristics for feature selection, Neurocomputing, 494 (2022), 269–296. https://doi.org/10.1016/j.neucom.2022.04.083 doi: 10.1016/j.neucom.2022.04.083
    [3] M. Rostami, K. Berahmand, E. Nasiri, S. Forouzandeh, Review of swarm intelligence-based feature selection methods, Eng. Appl. Artif. Intell., 100 (2021), 104210. https://doi.org/10.1016/j.engappai.2021.104210 doi: 10.1016/j.engappai.2021.104210
    [4] O. O. Akinola, A. E. Ezugwu, J. O. Agushaka, R. A. Zitar, L. Abualigah, Multiclass feature selection with metaheuristic optimization algorithms: a review, Neural Comput. Appl., 34 (2022), 19751–19790. https://doi.org/10.1007/s00521-022-07705-4 doi: 10.1007/s00521-022-07705-4
    [5] A. A. Heidari, S. Mirjalili, H. Faris, I. Aljarah, M. Mafarja, H. Chen, Harris hawks optimization: algorithm and applications, Future Gener. Comput. Syst., 97 (2019), 849–872. https://doi.org/10.1016/j.future.2019.02.028 doi: 10.1016/j.future.2019.02.028
    [6] J. Kennedy, R. Eberhart, Particle swarm optimization, in Proceedings of ICNN'95 - International Conference on Neural Networks, 4 (1995), 1942–1948. https://doi.org/10.1109/ICNN.1995.488968
    [7] S. Mirjalili, S. M. Mirjalili, A. Lewis, Grey wolf optimizer, Adv. Eng. Software, 69 (2014), 46–61. https://doi.org/10.1016/j.advengsoft.2013.12.007 doi: 10.1016/j.advengsoft.2013.12.007
    [8] J. Nasiri, F. M. Khiyabani, A whale optimization algorithm (WOA) approach for clustering, Cogent Math. Stat., 5 (2018), 1483565. https://doi.org/10.1080/25742558.2018.1483565 doi: 10.1080/25742558.2018.1483565
    [9] Z. Sadeghian, E. Akbari, H. Nematzadeh, H. Motameni, A review of feature selection methods based on meta-heuristic algorithms, J. Exp. Theor. Artif. Intell., 35 (2023), 1–51. https://doi.org/10.1080/0952813X.2023.2183267 doi: 10.1080/0952813X.2023.2183267
    [10] L. Wang, Q. Cao, Z. Zhang, S. Mirjalili, W. Zhao, Artificial rabbits optimization: a new bio-inspired meta-heuristic algorithm for solving engineering optimization problems, Eng. Appl. Artif. Intell., 114 (2022), 105082. https://doi.org/10.1016/j.engappai.2022.105082 doi: 10.1016/j.engappai.2022.105082
    [11] A. Srivastava, D. K. Das, A bottlenose dolphin optimizer: an application to solve dynamic emission economic dispatch problem in the microgrid, Knowledge-Based Syst., 243 (2022), 108455. https://doi.org/10.1016/j.knosys.2022.108455 doi: 10.1016/j.knosys.2022.108455
    [12] R. Ramalingam, B. Saleena, S. Basheer, P. Balasubramanian, M. Rashid, G. Jayaraman, EECHS-ARO: energy-efficient cluster head selection mechanism for livestock industry using artificial rabbits optimization and wireless sensor networks, Electron. Res. Arch., 31 (2023), 3123–3144. https://doi.org/10.3934/era.2023158 doi: 10.3934/era.2023158
    [13] Y. Wang, Y. Xiao, Y. Guo, J. Li, Dynamic chaotic opposition-based learning-driven hybrid Aquila Optimizer and artificial rabbits optimization algorithm: framework and applications, Processes, 10 (2022), 2703. https://doi.org/10.3390/pr10122703 doi: 10.3390/pr10122703
    [14] D. Dangi, S. T. Chandel, D. K. Dixit, S. Sharma, A. Bhagat, An efficient model for sentiment analysis using artificial rabbits optimized vector functional link network, Expert Syst. Appl., 225 (2023), 119849. https://doi.org/10.1016/j.eswa.2023.119849 doi: 10.1016/j.eswa.2023.119849
    [15] S. Kumar, S. Gupta, S. Arora, Research trends in network-based intrusion detection systems: a review, IEEE Access, 9 (2021), 157761–157779. https://doi.org/10.1109/ACCESS.2021.3129775 doi: 10.1109/ACCESS.2021.3129775
    [16] H. Alazzam, A. Sharieh, K. E. Sabri, A feature selection algorithm for intrusion detection system based on pigeon inspired optimizer, Expert Syst. Appl., 148 (2020), 113249. https://doi.org/10.1016/j.eswa.2020.113249 doi: 10.1016/j.eswa.2020.113249
    [17] Q. M. Alzubi, M. Anbar, Y. Sanjalawe, M. A. Al-Betar, R. Abdullah, Intrusion detection system based on hybridizing a modified binary grey wolf optimization and particle swarm optimization, Expert Syst. Appl., 204 (2022), 117597. https://doi.org/10.1016/j.eswa.2022.117597 doi: 10.1016/j.eswa.2022.117597
    [18] A. Alzaqebah, I. Aljarah, O. Al-Kadi, R. Damaševičius, A modified grey wolf optimization algorithm for an intrusion detection system, Mathematics, 10 (2022), 999. https://doi.org/10.3390/math10060999 doi: 10.3390/math10060999
    [19] M. Injadat, A. Moubayed, A. B. Nassif, A. Shami, Multi-stage optimized machine learning framework for network intrusion detection, IEEE Trans. Netw. Serv. Manage., 18 (2020), 1803–1816. https://doi.org/10.1109/TNSM.2020.3014929 doi: 10.1109/TNSM.2020.3014929
    [20] J. Lee, J. Pak, M. Lee, Network intrusion detection system using feature extraction based on deep sparse autoencoder, in 2020 International Conference on Information and Communication Technology Convergence (ICTC), (2020), 1282–1287. https://doi.org/10.1109/ICTC49870.2020.9289253
    [21] M. D. Mauro, G. Galatro, G. Fortino, A. Liotta, Supervised feature selection techniques in network intrusion detection: a critical review, Eng. Appl. Artif. Intell., 101 (2021), 104216. https://doi.org/10.1016/j.engappai.2021.104216 doi: 10.1016/j.engappai.2021.104216
    [22] Y. Li, W. Xu, W. Li, A. Li, Z. Liu, Research on hybrid intrusion detection method based on the ADASYN and ID3 algorithms, Math. Biosci. Eng., 19 (2021), 2030–2042. https://doi.org/10.3934/mbe.2022095 doi: 10.3934/mbe.2022095
    [23] T. Wang, H. Zhou, H. Liu, Multi-label feature selection based on HSIC and sparrow search algorithm, Math. Biosci. Eng, 20 (2023), 14201–14221. https://doi.org/10.3934/mbe.2023635 doi: 10.3934/mbe.2023635
    [24] A. Dahou, M. A. Elaziz, S. A. Chelloug, M. A. Awadallah, M. A. Al-Betar, M. A. Al-qaness, et al., Intrusion detection system for ioT based on deep learning and modified reptile search algorithm, Comput. Intell. Neurosci., 2022 (2022), 6473507. https://doi.org/10.1155/2022/6473507 doi: 10.1155/2022/6473507
    [25] M. Imran, S. Khan, H. Hlavacs, F. A. Khan, S. Anwar, Intrusion detection in networks using cuckoo search optimization, Soft Comput., 26 (2022), 10651–10663. https://doi.org/10.1007/s00500-022-06798-2 doi: 10.1007/s00500-022-06798-2
    [26] H. Xu, Y. Lu, Q. Guo, Application of improved butterfly optimization algorithm combined with black widow optimization in feature selection of network intrusion detection, Electronics, 11 (2022), 3531. https://doi.org/10.3390/electronics11213531 doi: 10.3390/electronics11213531
    [27] H. Xu, Y. Hu, W. Cao, L. Han, An improved jump spider optimization for network traffic identification feature selection, CMC-Comput. Mater. Continua, 76 (2023), 3239–3255. https://doi.org/10.32604/cmc.2023.039227 doi: 10.32604/cmc.2023.039227
    [28] H. Xu, K. Przystupa, C. Fang, A. Marciniak, O. Kochan, M. Beshley, A combination strategy of feature selection based on an integrated optimization algorithm and weighted k-nearest neighbor to improve the performance of network intrusion detection, Electronics, 9 (2020), 1206. https://doi.org/10.3390/electronics9081206 doi: 10.3390/electronics9081206
    [29] F. Qiu, H. Xu, F. Li, Applying modified golden jackal optimization to intrusion detection for Software-Defined Networking, Electron. Res. Arch., 32 (2024), 418–444. https://doi.org/10.3934/era.2024021 doi: 10.3934/era.2024021
    [30] A. Berta, Whales, Dolphins, and Porpoises: A Natural History and Species Guide, University of Chicago Press, 2020. https://doi.org/10.7208/9780226183220
    [31] L. Sun, M. M. Li, J. C. Xu, Binary harris hawk optimization and its feature selection algorithm, Comput. Sci., 50 (2023), 277–291. https://doi.org/10.11896/jsjkx.220300269 doi: 10.11896/jsjkx.220300269
    [32] M. Chawla, M. Duhan, Levy flights in metaheuristics optimization algorithms–a review, Appl. Artif. Intell., 32 (2018), 802–821. https://doi.org/10.1080/08839514.2018.1508807 doi: 10.1080/08839514.2018.1508807
    [33] J. Li, Q. An, H. Lei, Q. Deng, G. G. Wang, Survey of lévy flight-based metaheuristics for optimization, Mathematics, 10 (2022), 2785. https://doi.org/10.3390/math10152785 doi: 10.3390/math10152785
    [34] P. Yuan, T. Zhang, L. Yao, Y. Lu, W. Zhuang, A hybrid golden jackal optimization and golden sine algorithm with dynamic lens-imaging learning for global optimization problems, Appl. Sci., 12 (2022), 9709. https://doi.org/10.3390/app12199709 doi: 10.3390/app12199709
    [35] W. Long, J. Jiao, M. Xu, M. Tang, T. Wu, S. Cai, Lens-imaging learning Harris hawks optimizer for global optimization and its application to feature selection, Expert Syst. Appl., 202 (2022), 117255. https://doi.org/10.1016/j.eswa.2022.117255 doi: 10.1016/j.eswa.2022.117255
    [36] I. M. El-Hasnony, S. I. Barakat, M. Elhoseny, R. R. Mostafa, Improved feature selection model for big data analytics, IEEE Access, 8 (2020), 66989–67004. https://doi.org/10.1109/ACCESS.2020.2986232 doi: 10.1109/ACCESS.2020.2986232
    [37] B. Venkatesh, J. Anuradha, A review of feature selection and its methods, Cybern. Inf. Technol., 19 (2019), 3–26. https://doi.org/10.2478/cait-2019-0001 doi: 10.2478/cait-2019-0001
    [38] O. Almomani, A feature selection model for network intrusion detection system based on PSO, GWO, FFA and GA algorithms, Symmetry, 12 (2020), 1046. https://doi.org/10.3390/sym12061046 doi: 10.3390/sym12061046
    [39] T. Le, Y. Kim, H. Kim, Network intrusion detection based on novel feature selection model and various recurrent neural networks, Appl. Sci., 9 (2019), 1392. https://doi.org/10.3390/app9071392 doi: 10.3390/app9071392
    [40] K. Hussain, M. N. M. Salleh, S. Cheng, R. Naseem, Common benchmark functions for metaheuristic evaluation: a review, Int. J. Inf. Vis., 1 (2017), 218–223. http://dx.doi.org/10.30630/joiv.1.4-2.65 doi: 10.30630/joiv.1.4-2.65
    [41] N. M. Yusof, A. K. Muda, S. F. Pratama, A. Abraham, A novel nonlinear time-varying sigmoid transfer function in binary whale optimization algorithm for descriptors selection in drug classification, Mol. Diversity, 27 (2023), 71–80. https://doi.org/10.1007/s11030-022-10410-y doi: 10.1007/s11030-022-10410-y
    [42] K. Zhang, Y. Liu, F. Mei, G. Sun, J. Jin, IBGJO: improved binary golden jackal optimization with chaotic tent map and cosine similarity for feature selection, Entropy, 25 (2023), 1128. https://doi.org/10.3390/e25081128 doi: 10.3390/e25081128
    [43] R. D. Ravipati, M. Abualkibash, Intrusion detection system classification using different machine learning algorithms on KDD-99 and NSL-KDD datasets-a review paper, Int. J. Comput. Sci. Inf. Technol., 11 (2019), 65–80. https://doi.org/10.2139/ssrn.3428211 doi: 10.2139/ssrn.3428211
    [44] M. Tavallaee, E. Bagheri, W. Lu, A. A. Ghorbani, A detailed analysis of the KDD CUP 99 data set, in 2009 IEEE Symposium on Computational Intelligence for Security and Defense Applications, (2009), 1–6. https://doi.org/10.1109/CISDA.2009.5356528
    [45] T. Su, H. Sun, J. Zhu, S. Wang, Y. Li, BAT: deep learning methods on network intrusion detection using NSL-KDD dataset, IEEE Access, 8 (2020), 29575–29585. https://doi.org/10.1109/Access.6287639 doi: 10.1109/Access.6287639
    [46] M. K. Ngueajio, G. Washington, D. B. Rawat, Y. Ngueabou, Intrusion detection systems using support vector machines on the kddcup'99 and nsl-kdd datasets: a comprehensive survey, in Intelligent Systems and Applications, 543 (2022), 609–629. https://doi.org/10.1007/978-3-031-16078-3_42
    [47] N. Moustafa, J. Slay, UNSW-NB15: a comprehensive data set for network intrusion detection systems (UNSW-NB15 network data set), in 2015 Military Communications and Information Systems Conference (MilCIS), (2015), 1–6. https://doi.org/10.1109/MilCIS.2015.7348942
    [48] N. Moustafa, J. Slay, The evaluation of network anomaly detection systems: statistical analysis of the UNSW-NB15 data set and the comparison with the KDD99 data set, Inf. Secur. J.: Global Perspect., 25 (2016), 18–31. https://doi.org/10.1080/19393555.2015.1125974 doi: 10.1080/19393555.2015.1125974
    [49] M. S. Elsayed, N. A. Le-Khac, A. D. Jurcut, InSDN: a novel SDN intrusion dataset, IEEE Access, 8 (2020), 165263–165284. https://doi.org/10.1109/ACCESS.2020.3022633 doi: 10.1109/ACCESS.2020.3022633
    [50] M. Abdallah, N. A. L. Khac, H. Jahromi, A. D. Jurcut, A hybrid CNN-LSTM based approach for anomaly detection systems in SDNs, in ARES'21: Proceedings of the 16th International Conference on Availability, Reliability and Security, (2021), 1–7. https://doi.org/10.1145/3465481.3469190
  • Reader Comments
  • © 2024 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(1210) PDF downloads(57) Cited by(6)

Figures and Tables

Figures(13)  /  Tables(10)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog