Neural Process Lett DOI 10.1007/s11063-017-9711-6
Determining Approximate Solutions of Nonlinear Ordinary Differential Equations Using Orthogonal Colliding Bodies Optimization Arnapurna Panda1 · Sabyasachi Pani1
© Springer Science+Business Media, LLC 2017
Abstract The solution of nonlinear Ordinary Differential Equations (ODE) finds potential applications in physics, economics, computing and engineering. Conventional approaches used for solving ODE are effective in case of 1st order or 2nd order problems. With increase in order the complexity associated with the problem increases. Thus instead of going for an exact solution, determining an approximate solution is also helpful. In this paper, solving ODE is handled as an optimization problem. Popular Fourier Series expansion is used as an approximation function. A hybrid algorithm Orthogonal Colliding Bodies Optimization (OCBO) is formulated by assembling good features of orthogonal array (exploration of solution in search space) and Colliding Bodies Optimization (bodies after collision quickly move to a position with minimal energy level on the search space). The coefficients of the Fourier series are computed with OCBO. Simulation studies are carried out to determine solution of popular practically used ODEs: Bernoulli Equation for flowing fluids, IntegroDifferential equation, Brachistochrone equation for gravity, current response of an oscillatory Tank circuit, voltage and current decay with time in an electrical circuit. Simulations are also carried out on three benchmark ODEs used for modelling the biological processes. Comparative analysis demonstrated the superior approximation of the proposed approach over Orthogonal PSO, water cycle algorithm and Harmonic search. Keywords Ordinary differential equations · Orthogonal array · Colliding bodies optimization · Orthogonal PSO · Water cycle algorithm
B
Arnapurna Panda
[email protected] Sabyasachi Pani
[email protected]
1
School of Basic Sciences, Indian Institute of Technology Bhubaneswar, Bhubaneswar, Odisha 751007, India
123
A. Panda, S. Pani
1 Introduction Solving ordinary differential equations(ODE) finds extensive applications in physics (formulation of Bernoulli equation), electrical engineering (characteristics of voltage/current variation with time in an circuit), mechanical engineering (vibrations in mass-spring systems) and many more. The ODE differs from partial differential equations (PDE) in terms of considering the derivative i.e. in ODE derivative is taken with a single independent variable where as in PDE it takes into account of multiple variables [1]. In this manuscript solving ODE is formulated as an optimization problem. The aim is to determine approximate solution for any order ODE. Earlier approaches for determining the approximate solution of differential equation(DE) involve numerical approximation methods: Taylor polynomials [2], Taylor expansion approach [3,4], Homotopy-perturbation method [5]. These methods have the limitation for convergence to local optimum while approximating multi-modal non-linearities. Therefore, nature inspired meta-heuristic optimization algorithms are used to determine the near global optimal solutions. In these algorithms a group of agent interact among themselves and work in an collaborative manner to find out the solution of the problem. Literature survey reveals some of these popular algorithms have been used for solving the DE includes: genetic programming [6], genetic algorithm [7], particle swarm optimization [8,9], cuckoo search and water cycle algorithm [10], Harmonic Search [11]. The Fourier Series is wide popular for approximating any linear/nonlinear functions. Babaei [9] used Fourier Series to determine of approximate solution of ODE. The optimal value of coefficients of Fourier series were computed with the particle swarm optimization. Recently Sadollah et al. [10] proposed the use of cuckoo search and water cycle algorithm to obtain the coefficients more accurately. The use of orthogonal array to generate offsprings in nature inspired algorithms have been effective for several real life applications. The algorithms employing this concept include Orthogonal Genetic Algorithm (OGA) [12], Orthogonal differential evolution (ODE) [13,14], Orthogonal particle swarm optimization (OPSO) [15,16]. In most of the cases it is observed that the accuracy of optimization is more for the solutions obtained with orthogonal array. Attempts have also recently been made to reduce the computational complexity of OPSO [17]. Kaveh and Mahdavi [18] in 2014, proposed a new nature inspired meta-heuristic algorithm known as Colliding Bodies Optimization (CBO). It is influenced by collision between bodies in a surface which leads to their moment to achieve a minimum energy position on the surface. The advantage of CBO over other meta-heuristics is: it’s parameter free and it does not use memory to save good solutions. It has been applied to solve structural design problems [19], resource allocation problem [19], identification of Hammerstein plant [20]. The enhanced version of the CBO (ECBO) [19] is reported for optimal design of Reinforced Concrete Structures. In this manuscript Fourier Series is used as an approximator to determine solution of ODE. An hybrid algorithm orthogonal colliding bodies optimization (OCBO) is formulated to accurately compute the coefficient of Fourier Series. Simulations are carried out for determining approximate solution of five real life nonlinear ODE problems: three belongs to science (Bernoulli Equation for flowing fluids, Integro-Differential equation, Brachistochrone equation for gravity) and two belongs to electrical engineering (current response of an oscillatory Tank Circuit, Voltage and current decay in an electrical circuit). The results obtained with the proposed method is compared with those obtained by OPSO, water cycle algorithm (WCA) and Harmonic search algorithm (HS).
123
Determining Approximate Solutions of Nonlinear Ordinary…
The ODEs plays a significant role in modelling the biological processes. Some real applications are malaria epidemics [26], disease of respiratory disorders [27], disease of hematopoiesis [27], disease of HIV [31]. Initially after use of stochastic gradient based methods, Mendes first applied simulated annealing (SA) [31] and evolutionary programming (EP) [32] for parameter estimation (to determine several rate constants of the mechanism) in irreversible inhibition of HIV proteinase. Then Moles et al. [33] used evolution strategies (ES) to estimate 36 parameters of a nonlinear biochemical dynamic model. All these methods though ensure solution they are their computational involvement is extensive. In 2007, Peifer and Timmer [34] proposed a multiple shooting method to estimate the parameters of ODEs for biochemical processes which has lower computational burden. Qian et al. [35] developed a nonlinear ODE model for yeast protein synthesis. In this model they opted Genetic programming (GP) to determine the structure of the model and employ Kalman filtering to determine the parameters in every iteration. Zhan et al. [36] proposed two models for determining parameters of known structure dynamical biological systems. One is using spline theory and Nonlinear Programming (NLP) with which the parameter estimation is treated as an optimization problem. Other one is use of differential evolution (DE) algorithm to estimate the parameter of ODE. With further analysis Zhan et al. [37] reported to improve the identification quality by using S-system representation. With certain assumptions they have applied it to yeast fermentation analysis. Inspired by the recent trends of research of ODEs role in the biological processes, in this manuscript we have used the OCBO algorithm to estimate the parameters of three benchmark ODEs used in biological modelling given by Shakeri and Dehghan in [29]. The performance of the proposed parameter estimation are compared with that obtained by OPSO, WCA and HS based models. The paper is organized as follows. Section 2 describes the problem formulation of determining the approximate solution of ODE with Fourier series. The creation of orthogonal array and corresponding pseudo code for implementation is given in Sect. 3. Formulation of hybrid algorithm OCBO and its step wise implementation is described in Sect. 4. For obtaining approximate solution of ODE, the steps involve in computing Fourier series coefficients with proposed OCBO are narrated in Sect. 5. The real life ODE problems for analysis, comparative algorithms, simulation environment and performance evaluation criteria are presented in Sect. 6. Discussions on the obtained results are highlighted in Sect. 7. Concluding remarks on the proposed methodology are given in Sect. 8.
2 Approximate Solution of Ordinary Differential Equations with Fourier Series Let z be a function of x. Then an ODE of order n is expressed by dz d 2 z dn z F x, z, , 2,..., n = 0 dx dx dx or F x, z, z , z , . . . , z n = 0
(1)
The Eq. (1) is subject to the boundary conditions: z(x0 ) = z 0 , .... .. z(xn ) = z n ,
z (x0 ) = z 0 , .... .. z (xn ) = z n
(2)
123
A. Panda, S. Pani
or the initial conditions
(n−1)
z(x0 ) = z 0 , z (x0 ) = z 0 , . . . z (n−1) (x0 ) = z 0
(3)
Following [23] the actual solution z(x) can be approximated using Fourier series as below: T 0) 0) (4) z(x) ≈ Z apx (x) = a0 + am cos mπ(x−x + bm sin mπ(x−x N N m=1
where T is the number of terms associate with sin and cos functions. The N is the length of the interval solution where [x0 , xn ] is the bound of the interval. The derivatives of Eq. (4) are given by T mπ mπ(x − x0 ) z (x) ≈ Z apx (x) = am sin − m=1 N N mπ mπ(x − x0 ) bm cos + N N
2 T mπ mπ(x − x0 ) z (x) ≈ Z apx (x) = − am cos m=1 N N mπ 2 mπ(x − x0 ) bm sin − N N ::
mπ n T mπ(x − x0 ) nπ n n + am cos z (x) ≈ Z apx (x) = m=1 N N 2 (5) mπ n mπ(x − x0 ) nπ bm sin + + N N 2 In order to generate approximate solution of (1), the Eqs. (17)–(5) are substituted in Eq. (1), which creates an error function E(x) given by n E(x) = Fˆ x, Z apx , Z apx , Z apx , . . . , Z apx (6) The aim is to determine the coefficients of Fourier Series i.e. a0 , am and bm such that the value of error function E(x) approaches zero value. The meta-heuristic algorithm are used to optimize the coefficient value such that the error becomes minimal.
3 Orthogonal Design Combination In meta-heuristic based approach to solve optimization problems, the ‘Orthogonal Design Combination’ play a vital role as it scatters the solutions uniformly in the search space which enhance the exploration capabilities of the algorithm. Suppose there are N factors and Q levels, that means N number of vectors are there and each vector has Q components. The general combination of N factors and Q levels is Q N which is very vast and difficult for calculation of fitness value. With Orthogonal design combination the resulting orthogonal array of N factors and Q levels is L M (Q N ) where M is the number of combinations of levels and L represents the Latin square. The L M (Q N ) is a matrix of M rows and N columns which is a representative set of all combinations Q N . It can be represented as L M (Q N ) = [xi, j ] M X N where the jth factor in ith combination is represented by xi, j . The xi, j takes value in 1, 2, . . . , Q. The combination is orthogonal due to some specific properties. If any two columns of L M (Q N ) are swapped then the resulting L M (Q N ) is also an orthogonal array. Secondly if
123
Determining Approximate Solutions of Nonlinear Ordinary…
we delete some columns of L M (Q N ) then the resulting matrix is also a orthogonal array with less number of columns. To understand the ‘Orthogonal Design Combination’ consider an example: Lets consider four factors and three levels, that means four variables having three components each. In general total number of possible combination is 34 = 81. With Orthogonal Design Combination results in an array L 16 (34 ) which have 16 combinations. Thus here instead of 81 only 16 combinations are to be tested. The representative 16 combinations are shown in Fig. 1.
3.1 Construction of Orthogonal Array Construction of orthogonal array has been well described in the literature [15,25]. A simple procedure is outlined in Algorithm 1 for creating the array. The columns of orthogonal array xi, j denoted by x j is categorized into two parts: Primary columns and Secondary columns. The primary columns are j = 1, 2, (Q 2 − 1)/(Q − 1) + 1, . . . , (Q P −1)/(Q −1)+1. Here Q is odd. The remaining columns are termed as secondary columns. The procedure to create the primary columns and secondary columns are highlighted in Algorithm 1. Algorithm 1 Creating Orthogonal Array L M (Q N ) 1: M ← Q P , P is a positive integer; 2: 3: 4: 5: 6: 7: 8:
P
−1 N ← QQ−1 ; loop: (1 ≤ i ≤ M) and (1 ≤ j ≤ N ) Step 1: Construct the Primary Columns. for p = 1 to P do p−1 j = Q Q−1−1 + 1; for i = 1 to Q P do xi, j = i−1 mod Q; P− p
Q
9: end for 10: end for 11: Step 2: Construct the Secondary Columns. 12: for p = 2 to P do 13: 14: 15: 16: 17: 18: 19: 20: 21: 22: 23:
p−1
j = Q Q−1−1 + 1; for u = 1 to j − 1 do for v = 1 to Q − 1 do x j+(u−1)(Q−1)+v = xu × v + x j mod Q; end for end for end for Step 3: Increment xi, j j ← j + 1; i ← i + 1; goto loop.
4 Orthogonal Colliding Bodies Optimization (OCBO) OCBO is a hybrid algorithm which contains good feature of Orthogonal Array and characteristics of Colliding Bodies Optimization (CBO). The Orthogonal Array explained in Sect. 3 elaborated the creation of new solutions by making orthogonal combination of the existing
123
A. Panda, S. Pani
Fig. 1 Example of orthogonal design combination
solutions. Thus it is helpful for exploring and determining the optimal solution by making suitable combination. The CBO [19] is based on the natural phenomenon of collision between bodies with which they try to achieve a position with minimal energy level on the search space. As shown in Fig. 2 a body with mass m 1 moving with velocity u 1 collide with another
123
Determining Approximate Solutions of Nonlinear Ordinary…
Fig. 2 Collision between two bodies of mass m 1 and m 2 moving with velocities u 1 and u 2 . With progressive collision bodies attend positions corresponding to minimal energy level
body m 2 with velocity u 2 . During collision conservation of momentum and kinetic energy takes place. After collision mass m 1 achieve a velocity u 1 and m 2 move with velocity u 2 . The CBO does not use memory for saving good solution. Another advantage is it does not posses any internal parameter tuning.
4.1 Implementation Steps of OCBO The motivation here is to design orthogonal array between bodies positions attended before collision and after collision. It is beneficial to obtain new positions which may explore the possibilities of attending minimum energy level. The detailed procedure is outlined as follows: Objective Minimize a function Q = f (w1 , w2 , . . . , w D ) in the range [wmin , wmax ]. Where D is the dimensionality which also reflect the number of variable to minimize simultaneously. Step wise procedure for determining optimal solution: 1. Initialize population Initialize B number of Colliding Bodies (CB) which represent the population. Each CB is of dimension D. The bodies position are given by ⎡ → ⎢ ⎢ W= ⎢ ⎣
⎤ ⎡ W1 w1,1 w1,2 ⎢ w2,1 w2,2 W2 ⎥ ⎥ ⎢ .. ⎥ = ⎢ .. .. . ⎦ ⎣ . . WB w B,1 w B,2
⎤ . . . w1,D . . . w2,D ⎥ ⎥ ⎥ .... .. : ⎦ . . . w B,D
(7)
where Wi = Wmin + r nd(Wmax − Wmin ) i = 1, 2, . . . , B
(8)
Here Wmin = [min(w1 ), min(w2 ), . . . min(w D )] and Wmax = [max(w1 ), max(w2 ), . . . max(w D )]. The r nd is a random number in range [0, 1]. 2. Compute the fitness of each body The fitness of ith body is calculated by substituting it’s position value in the fitness function given by Q i = f (Wi ) ∀i = 1, 2, . . . , B
(9)
123
A. Panda, S. Pani
3. Compute the mass of each body The mass of each body is calculated as 1 Qi
mi = B
1 j=1 Q j
∀i = 1, 2, . . . , B
(10)
4. Initial velocities of bodies The bodies are sorted in according to their mass in a decreasing order. They are classified into two groups: – Stationary bodies: The upper half of bodies having higher mass are termed as stationary bodies and their velocities before collision are considered as as zero. u i = 0; ∀i = 1, 2, . . . ,
B 2
(11)
– Moving bodies: The lower half of bodies are movable and their velocities before collision are B + 1, . . . , B (12) u i = Wi − Wi− B ; ∀i = 2 2 5. Bodies velocity after collision The velocity of each stationary body after collision is : m i+ B + m i+ B u i+ B B 2 2 2 ui = ; ∀i = 1, . . . (13) m i + m i+ B 2 2
Velocity of every moving body after collision is: m − m B i i− 2 u i B ; ∀i = + 1, . . . , B ui = m i + m i− B 2
(14)
2
The is known as coefficient of restitution. 6. Bodies position after collision The positions of stationary bodies after collision are
Winew = Wi + r nd ⊗ u i ∀i = 1, . . . ,
B 2
(15)
The moving bodies positions after collision are
Winew = Wi− B + r nd ⊗ u i ∀i = 2
B + 1, . . . , B 2
(16)
The r nd represent random numbers in range [-1, 1]. The ‘⊗’ denotes term by term multiplication. 7. Design Orthogonal Positions The updated position Winew and old position Wi are used to create orthogonal vectors. Creating the vectors are described in algorithm 1 of earlier Sect. 3.1. Total number of bodies for orthogonal combination is Q = 2B and each with dimension N = D. The orthogonal combination produces a Latin Square of M × N given by L M Q N = αi, j M×N where M = Q log N (N +1)
(17) The details to obtain L M Q N is described in Sect. 3. 8. Upgrade each body position with it’s best Orthogonal Position – Compute the fitness of all orthogonal combinations L M Q N .
123
Determining Approximate Solutions of Nonlinear Ordinary…
– Find out the Winew obtained from orthogonal combinations which is superior than Winew . Replace xinew by the orthogonal position xinew . 9. Calculate coefficient of restitution (COR) The COR is computed as =1−
gen gen max
(18)
where gen is the present generation and gen max is the maximum number of generation. 10. Termination Criteria The steps from 2 to 9 are repeated till a satisfactory minimal function value is achieved or the algorithm is set to run for a maximum number of generation.
5 Computing the Fourier Series Coefficients with Proposed OCBO The steps to compute the Fourier Series coefficients with proposed OCBO algorithm are narrated below: 1. Determine the number of terms for approximation in Fourier Series It is dependent on the non-linearity associated with the ODE. For complex non-linearity more number of terms are required for approximation. Based on the application the optimal value of number of sin-cos expansion required may be set by hit and trail method by the programmer. ex- If the number terms is 3 then total number of variable to optimize is 7 (a0 along with 3 sine and 3 cosine component). Therefore the dimension at the initialization D = 7. → 2. Initialization of bodies Initialize a W = B × D dimensional matrix that represent ‘B’ bodies each with ‘D’ dimension as described in Eqs. (7)–(8). 3. Bodies fitness evaluation For each of the body the error function E(x) in (6) represent the fitness. Suppose E(x) = a0 + a1 cos(π x) + b1 sin(π x) (19) →
Then each row of W representing a possible solution (i.e. a body) [a0 , a1 , b1 ]. With the →
progress in OCBO the positions of bodies in W are updated such that E(x) becomes minimum. Then the implementation steps 3–10 described in Sect. 4 are continued. Finally the obtained solution i.e. minimal value of E(x) and corresponding coefficient values are reported. Once the training is complete for twenty independent runs the performance evaluation of the obtained approximate solution is carried out using the criteria defined in Sect. 6.4 and obtained results are reported.
6 Simulation Analysis 6.1 Test Problems: Differential Equations Arising in Science and Engineering Problems – Example 1: Solving Bernoulli Equation
123
A. Panda, S. Pani
The Bernoulli Equation is famous and deals with the conservation of energy in the flowing fluids. It is represented by second order ODE [24] given by
z + (z )2 − 2e(−z) = 0 z(0) = 0, z(1) = 0
(20)
The true solution for the above ODE obtained with analytical method is z(x) = ln((x − 0.5)2 + 0.75)
(21)
For approximate solution using Fourier Series the number of expansion using sin-cos is taken as 3. The solution space is assumed to be in the range [0, 1] as earlier taken in the literature [9,10]. The approximate solution takes the form Z apx (x) = a0 + a1 cos(π x) + b1 sin(π x) + a2 cos(2π x) + b2 sin(2π x) + a3 cos(3π x) + b3 sin(3π x)
(22)
– Example 2: Solving Integro-Differential Equation The Integro-differential equation is mathematically given by x 1x ≥0 z + 2z + 5 , z(0) = 0 z(t)dt = 0 x <0 0
(23)
The solution interval is in range 0 to π. The exact solution of this integral value problem is 1 −x e sin(2x) (24) 2 For effective approximate solution using Fourier Series the number of terms in the sin-cos expansion is set as six. The approximate solution obtained is in the form z(x) =
Z apx (x) = a0 + a1 cos (x) + b1 sin (x) + a2 cos (2x) + b2 sin (2x) + a3 cos (3x) + b3 sin (3x) + a4 cos (4x) + b4 sin (4x) + a5 cos (5x) + b5 sin (5x) + a6 cos (6x) + b6 sin (6x)
(25)
– Example 3: Solving Brachistochrone Equation This nonlinear ODE is mathematically represented by
[1 + (z )2 ]z = k 2 z(0) = 0, z(1) = 2
(26)
where k 2 is unknown constant. The true solution mathematically is a parametric equation of cycloid 2 x(θ ) = k2 (θ − sin(θ )) (27) 2 z(θ ) = k2 (1 − cos(θ )) The analytical value of constant k 2 is 4.81. This constant value is also taken as an additional parameter and approximated by the optimization algorithm. In the sin-cos expansion number of terms taken is 5. The approximate expansion consist of 11 variable. Thus total number of parameter to optimize including k 2 is 12. The approximate solution is Z apx (x) = a0 + a1 cos (π x) + b1 sin (π x) + a2 cos (2π x) + b2 sin (2π x) + a3 cos (3π x) + b3 sin (3π x) + a4 cos (4π x) + b4 sin (4π x) + a5 cos (5π x) + b5 sin (5π x)
123
(28)
Determining Approximate Solutions of Nonlinear Ordinary… Fig. 3 A tank circuit in Ex. 4 which produces damped oscillations
– Example 4: Oscillatory circuit or Tank circuit In electrical engineering the circuit that produces voltage/current oscillations of any desired frequency is popularly known as Oscillatory circuit or Tank circuit. A simple tank circuit comprise of a switch (S), capacitor (C)and an inductor (L) is shown in Fig. 3. The capacitor is initially charged from a d.c. source. Then the switch is closed. The circuit behave as an oscillator and generate damped oscillations till the current decreases gradually and becomes zero. The frequency of oscillation is f =
2π
1 √
LC
(29)
The damped oscillation of current in the circuit is represented by second order differential equation given by z + 0.3z + z − 1 = 0 (30) z(0) = 0, z (0) = 0 The exact solution for this ODE using analytical calculation is given by ⎞ ⎛ ⎞ √ √ ⎛ √ −3 391 391x 391x cos sin 391 20 20 ⎠−⎝ ⎠ z(x) = 1 + ⎝ 3x 3x 20 20 e e
(31)
The number of expansions in Fourier series is set to six. The time interval is taken from 0 to 15 to observe the damping oscillations. The approximate solution is obtained in the form πx πx x Z apx (x) = a0 + a1 cos + a2 cos 2π 15 + b1 sin 2π x 15 3π 15 x + b2 sin 15 + a3 cos 15x + b3 sin 3π 15 4π x 4π x 5π x (32) + a4 cos 15 + b4 sin 15 + a5 cos 15 x x 6π x + a6 cos 6π + b5 sin 5π 15 15 + b6 sin 15 – Example 5: Voltage and Current Decay in a circuit In electrical engineering circuits the voltage (V) and current (I) decay with time (t) and their behavior is expressed in terms of ODE. One example of such ODE is given by dV = 2I − V dt dI = −I − V dt
(33) (34)
123
A. Panda, S. Pani
with I(0)=2, V(0)=2 are the initial conditions. The exact solutions are given by √ √ √ −t cos( 2t) V (t) = 2 2e−t sin( 2t) + 2e √ √ √ I (t) = 2e−t cos( 2t) − 2e−t sin( 2t)
(35)
The time interval for solution is considered between 0 to 1.5 seconds. Te approximate solution for the voltage and current are approximated with three sin-cos expansions given by πt πt 2πt Vapp (t) = a0 + a1 cos + b1 sin + a2 cos 1.5 1.5 1.5 2πt 3πt 3πt + a3 cos + b3 sin (36) + b2 sin 1.5 1.5 1.5 πt πt 2πt + b1 sin + a2 cos Iapp (x) = a0 + a1 cos 1.5 1.5 1.5 2πt 3πt 3πt + b2 sin + a3 cos + b3 sin (37) 1.5 1.5 1.5
6.2 Test Problems: Differential Equations Arising in Mathematical Biology The differential equations are extensively used in modeling several biological phenomenons. Some important applications are: effect of incubation delays on malaria epidemics [26], effect of CO2 concentration in a mammal causing a dynamic disease of respiratory disorders [27], density of mature cells in blood circulation leading to disease of hematopoiesis [27], the effect of concentration of a chemical component and temperature resulting an exothermic/irreversible reaction [28]. Shakeri and Dehghan in [29] proposed a generalized form of differential equation expressing the characteristics of all the above model given by u (m) (t) =
J m−1
μ jk (t)u (k) (α jk t + β jk ) + g(u) + f (t)
(38)
j=0 k=0
where g(u) = bu +
l
di u γi
(39)
i=1
The di ∈ R, γi > 1 and γi ∈ N , l ∈ N . The α jk ∈ [0, 1] and β jk ∈ R. The u (m) is the mth derivative of function u. The Eq. (38) is associated with initial conditions m−1
cik u (k) (0) = λi , i = 0, 1, . . . , m − 1
(40)
k=0
In [30] Yildirim et al. applied variational iteration method for a reliable analysis of this generalized differential equation for mathematical biology. In this paper the proposed Fourier series-OCBO based approximation model is used to determine the effective solutions of this generalized differential equation. – Example 6: First Order Differential Equation The first order differential equation used in [29] for modelling the biological processes is given by 1 t u (t) = −2tu(t − ) − t 2 u(t) + 2 sin(t)u + u 3 (t) + f (t) (41) 2 2
123
Determining Approximate Solutions of Nonlinear Ordinary… Table 1 Comparative results of Ex. 1, Ex. 2 and Ex. 3: obtained coefficients of Fourier Series after optimized with OCBO, OPSO, WCA and HS (Reported best values after 20 independent runs) Ex. Ex. 1
Ex. 2
Ex. 3
where
Coefficients
OCBO
OPSO
WCA
HS
a0
2.2512E-02
2.2436E-02
2.2508E-02
2.2528E-02
a1
8.3125E-05
8.3150E-05
8.3129E-05
8.3142E-05
b1
−0.3290
−0.3328
−0.3250
−0.3188
a2
−2.2490E-02
−2.2468E-02
−2.2550E-02
−2.2608E-02
b2
−1.6735E-04
−1.6750E-04
−1.6727E-04
−1.6742E-04
a3
−8.3096E-05
−8.3202E-05
−8.3114E-05
−8.3150E-05
b3
3.6882E-03
3.6850E-03
3.6871E-03
3.6864E-03
a0
1.9998E-02
2.0004E-02
1.9992E-02
1.9989E-02
a1
1.1000E-02
1.1022E-02
1.0097E-02
1.0092E-02
b1
3.8360E-02
3.8367E-02
3.8354E-02
3.8351E-02
a2
−7.9945E-04
−7.9951E-04
−7.9942E-04
−7.9938E-04
b2
1.1888E-01
1.1886E-01
1.1876E-01
1.1879E-01
a3
−3.6250E-02
−3.6254E-02
−3.6245E-02
−3.6239E-02
b3
4.7975E-02
4.7987E-02
4.7968E-02
4.7970E-02
a4
−4.2281E-02
−4.2285E-02
−4.2277E-02
−4.2268E-02
b4
6.5365E-02
6.5371E-02
6.5358E-02
6.5365E-02
a5
2.7610E-02
2.7616E-02
2.7608E-02
2.7598E-02
b5
4.9890E-02
4.9902E-02
4.9892E-02
4.9896E-02
a6
2.0755E-02
2.0761E-02
2.0748E-02
2.0742E-02
b6
−4.1470E-03
−4.1478E-03
−4.1467E-03
−4.1456E-03
a0
0.8683
0.8690
0.8680
0.8677
a1
−1.0000
−1.0020
−0.9996
−0.9985
b1
0.5370
0.5374
0.5366
0.5362
a2
0.0630
0.0636
0.0627
0.0622
b2
0.2754
0.2758
0.2749
0.2750
a3
−0.0370
−0.0375
−0.0369
−0.0366
b3
0.0884
0.0890
0.0882
0.0878
a4
0.0690
0.0702
0.0682
−0.0678
b4
0.0895
0.0912
0.0884
0.0881
a5
0.0370
0.0376
0.0368
0.0363
b5
−0.0184
−0.0188
−0.0182
−0.0175
k2
4.8224
4.8418
4.8556
4.8810
t t 7 f (t) = − e− 2 −2 −t 4 − t 3 + 23 t 2 − 23 t − 23 − 2te− 2 − 4 −t 2 + 45 2 2 3 3t t t + e− 8 −6 − 16 − 4t + 1 + 2 sin(t)e− 4 −2 − t4 − 2t + 1
(42)
with initial condition u(0) = e−2 . The exact solution is given by t u(t) = −t 2 + 1 − t e− 2 −2
(43)
123
A. Panda, S. Pani Table 2 Comparative results for Ex. 4 and Ex. 5: obtained coefficients of Fourier Series after optimized with OCBO, OPSO, WCA and HS (reported best values after 20 independent runs ) Ex. Ex. 4
Ex. 5
Coefficients
OCBO
OPSO
WCA
HS 0.8852
a0
0.8865
0.8880
0.8861
a1
−0.1648
−0.1664
−0.1652
−0.1645
b1
0.2529
0.2538
0.2533
0.2522
a2
0.3277
0.3280
0.3278
0.3272
b2
0.2070
0.2055
0.2068
0.2080
a3
0.1610
0.1598
0.1606
0.1622
b3
−0.4635
−0.4642
−0.4637
−0.4640
a4
−0.6636
−0.6639
−0.6637
−0.6634
b4
−0.1966
−0.1963
−0.1965
−0.1960
a5
−0.5220
−0.5223
−0.5218
−0.5216 0.1912
b5
0.1910
0.1910
0.1909
a6
−0.0235
−0.0240
−0.0238
−0.0234
b6
0.0926
0.0930
0.0926
0.0928
a0
0.6620
0.6623
0.6618
0.6615
a1
0.8185
0.8187
0.8179
0.8177
b1
1.3270
1.3277
1.3265
1.3268
a2
0.4820
0.4822
0.4821
0.4815
b2
8.5915E-02
8.5921E-02
8.5909E-02
8.5911E-02
a3
3.7430E-02
3.7433E-02
3.7436E-02
3.7428E-02
b3
−0.1135
−0.1133
−0.1137
−0.1139
a0
0.8385
0.8392
0.8381
0.8378
a1
1.3625
1.3633
1.3621
1.3616
b1
−1.0562
−1.0567
−1.0557
−1.0553
a2
−9.8535E-02
−9.8542E-02
−9.8533E-02
−9.8528E-02
b2
−0.4600
−0.4611
−0.4597
−0.4591
a3
−0.1025
−0.1027
−0.1021
−0.1019
b3
1.1920E-02
1.1931E-02
1.1915E-02
1.1912E-02
Voltage
Current
The approximate solution is represented in the form u apx (t) = a0 + a1 cos (πt) + b1 sin (πt) + a2 cos (2πt) + b2 sin (2πt) + a3 cos (3πt) + b3 sin (3πt) + a4 cos (4πt) + b4 sin (4πt) + a5 cos (5πt) + b5 sin (5πt)
(44)
– Example 7: Second Order Differential Equation The second order differential equation from [29] for analyzing the biological processes is represented by t 1 + (t − 1)u(t) − 2u + u 2 (t) + f (t) (45) u (t) = e−t u t − 5 3
123
Determining Approximate Solutions of Nonlinear Ordinary… Table 3 Comparative results of biological systems Ex. 6, Ex. 7 and Ex. 8: obtained coefficients of Fourier Series after optimized with OCBO, OPSO, WCA and HS (Reported best values after 20 independent runs) Ex. Ex. 6
Ex. 7
Ex. 8
Coefficients
OCBO
OPSO
WCA
HS
a0
0.02360
0.02364
0.02361
0.02357
a1
0.08133
0.08128
0.08131
0.08126
b1
0.00602
0.00604
0.00603
0.00602
a2
0.00680
0.00682
0.00676
0.00675
b2
0.01821
0.01821
0.01819
0.01818
a3
0.03461
0.03463
0.03460
0.03461
b3
−0.00570
−0.00568
−0.00572
−0.00573
a4
−0.00383
−0.00380
−0.00384
−0.00387
b4
−0.02195
−0.02192
−0.02196
−0.02195
a5
−0.00694
−0.00694
−0.00696
−0.00699 0.00151
b5
0.00152
0.00154
0.00152
a0
0.33341
0.33342
0.33340
0.33339
a1
−0.04320
−0.04318
−0.04321
−0.04324 0.12042
b1
0.12046
0.12050
0.12043
a2
0.07530
0.07534
0.07529
0.07528
b2
−0.01342
−0.01340
−0.01344
−0.01347
a3
−0.02242
−0.02238
−0.01345
−0.01346
b3
−0.03873
−0.03870
−0.03875
−0.03876
a4
−0.01390
−0.01388
−0.01392
−0.01393
b4
0.01371
0.01375
0.01368
0.01367
a5
0.00410
0.00414
0.00405
0.00406
b5
0.00282
0.00284
0.00277
0.00277
a0
0.55030
0.55028
0.55032
0.55035
a1
−0.35231
−0.35228
−0.35233
−0.35235
b1
0.34333
0.34328
0.34335
0.34335
a2
−0.05042
−0.05036
−0.05045
−0.05048
b2
−0.02280
−0.02280
−0.02282
−0.02286
a3
−0.19565
−0.19562
−0.19568
−0.19570
b3
−0.00610
−0.00606
−0.00613
−0.00612
a4
−0.00033
−0.00030
−0.00034
−0.00034 0.14310
b4
0.14310
0.14305
0.14311
a5
0.04635
0.04634
0.04635
0.04638
b5
−0.00060
−0.00060
−0.00061
−0.00061
where 1 t 1 t 2 t 2 t f (t) = − 41 sin t 3 sin1 3 cos t 2 − 49 cos t 2 1 t 13 − + − 2 sin 3 − 3 cos 2 t + 9 sin 3 + 4 cos 2 1 1 − 16 sin 2t − 10 + sin 9t + 23 cos 6t − e−t 16 cos 3t − 15 with initial conditions
(46)
3u(0) + 6u (0) = 2 −2u(0) + u (0) = − 21
(47)
123
A. Panda, S. Pani
(a) 0.05 Exact OCBO OPSO WCA HS
0 −0.05
Z(x)
−0.1 −0.15 −0.2 −0.25 −0.3
(b)
0
0.2
0.4
x
0.6
0.8
1
0.3
Exact OCBO OPSO WCA HS
0.25 0.2
Z(x)
0.15 0.1 0.05 0 −0.05 −0.1
0
0.5
1
1.5
2
2.5
3
3.5
x
Fig. 4 Response matching of the true solution and approximate solutions obtained by OCBO and comparative algorithms: a Example 1 b Example 2
The exact solution is given by u(t) =
t 1 t 1 sin + cos 2 3 3 2
(48)
The approximate solution using proposed approach here also takes the form defined in Eq. (44). – Example 8: Third Order Differential Equation The third order exothermic reaction from [29] used for modelling the biological processes is represented by t t t u (t) = u (t) + et−1 u −u + f (t) (49) 3 3 2
123
Determining Approximate Solutions of Nonlinear Ordinary…
(a) 2.5 2
Z(x)
1.5 1 0.5
Exact OCBO OPSO WCA HS
0 −0.5
0
0.2
0.4
x
0.6
0.8
(b) 1.8
1
Exact OCBO OPSO WCA HS
1.6 1.4 1.2
Z(x)
1 0.8 0.6 0.4 0.2 0 −0.2
0
5
10
x
15
Fig. 5 Response matching of the true solution and approximate solutions obtained by OCBO and comparative algorithms: a Example 3 b Example 4
where
8 7 f (t) = − et−1 − 2187 t +
2 5 27 t
+
5 4 81 t
−
16 3 27 t
+ 13 t 2 − 43 t + 3
(50)
+
893 7 48 t
−
with initial conditions
5847 5 16 t
−
305 4 48 t
+ 374t 3
+
235 2 4 t
−
290 3 t
+9
⎧ ⎨ u(0) − u (0) − 2u (0) = 5 u (0) − u (0) = 7 ⎩ 2u (0) + 3u (0) = −6
(51)
u(x, t) = −t 8 + 3t 6 + t 5 − 4t 4 + t 3 − 2t 2 + 3t
(52)
The exact solution is given by
123
A. Panda, S. Pani
(a) 2.5
Exact OCBO OPSO WCA HS
Vapp(t)
2
1.5
1
0.5
0
0
0.5
1
1.5
time (t)
(b) 2.5
Exact OCBO OPSO WCA HS
2
Iapp(t)
1.5 1 0.5 0 −0.5 −1
0
0.5
time (t)
1
1.5
Fig. 6 Response matching of the true solution and approximate solutions obtained by OCBO and comparative algorithms for Example 5: a Approximation Voltage b Approximation Current
Here also the approximate solution by the proposed method is represented in the form defined in Eq. (44).
6.3 Comparative Algorithms The performance of the proposed OCBO is compared with Orthogonal PSO [15], Water Cycle Algorithm (WCA) [10] and Harmonic Search [11]. Sadollah et al. in 2015 [10], have reported the superior performance of WCA for solving ODE over cuckoo search and particle swarm optimization(PSO). In literature [11], the Harmonic Search has reported better performance than genetic algorithm and PSO for determining approximate solutions of ODE of longitudinal fins (having rectangular, trapezoidal and concave parabolic profiles). In all four algorithms the initial population size is taken for Ex. 1–Ex. 2 is 50, Ex. 3–Ex. 5 is 200. The maximum number of iteration allowed to run for Ex. 1–Ex. 3 is 500, Ex. 4 is
123
Determining Approximate Solutions of Nonlinear Ordinary…
(a) 0.15
Exact OCBO OPSO WCA HS
0.1
U(t)
0.05
0
−0.05
−0.1
0
0.2
0.4
t
0.6
0.8
1
(b) 0.46 0.44
U(t)
0.42 0.4 0.38 0.36
Exact OCBO OPSO WCA HS
0.34 0.32
0
0.2
0.4
t
0.6
0.8
1
Fig. 7 Response matching of the true solution and approximate solutions obtained by OCBO and comparative algorithms for biological systems: a Example 6 b Example 7
100, Ex. 5 is 200, Ex. 6 is 100, Ex. 7 is 500 and Ex. 8 is 200. The total number of fitness evaluation from Ex. 1–Ex. 3 are 1,00,000, for Ex. 4 is 20,000, for Ex. 5 is 40,000, for Ex. 6 is 20,000, for Ex. 7 is 1,00,000 and Ex. 8 is 40,000. The rest parameter settings for WCA and HS are kept same as given in [10] and [11] respectively. The settings for orthogonal array using proposed OCBO are elaborated in Sect. 3. The parameter settings for OPSO is kept same as given in [15]. As all the algorithms used in this context are meta-heuristic in nature therefore twenty independent runs of each algorithm is considered. The best, average, worst and standard deviation values are considered during performance evaluation.
6.4 Simulation Environment The simulation studies are carried out in MATLAB R2011 platform. All the four algorithms are set to run in a Dell Latitude E5430 laptop with Intel i7-3540M 3GHz CPU, 8GB RAM in Windows 8.1 (64-bit) environment.
123
A. Panda, S. Pani 1.2 1
U(t)
0.8 0.6 0.4 0.2
Exact OCBO OPSO WCA HS
0 −0.2
0
0.2
0.4
0.6
0.8
1
t
Fig. 8 Response matching of the true solution and approximate solutions obtained by OCBO and comparative algorithms for biological system Example 8
6.5 Performance Evaluation In order to evaluate the performance of the algorithms the following criteria are considered: – Response matching The response matching between true analytical solution and obtained approximate solution reveals the closeness of the solutions. – Generalized distance It is the Euclidean distance between points lying on the exact and approximate solution [10]: " # S # GD = $ (53) (z − Z )2 s
s
s=1
where z s represent points true analytical solution, Z s are points approximate solution and S is the total number of points in both the curves. The minimum value of G D represents the approximate solution is nearest to the true solution. – Estimation of the Coefficients The best values of Fourier Coefficients obtained by the comparative algorithms reveals an near effective value of the approximation to the ODE. – Run time The average run time for fixed number of fitness evaluation is a measure of computational complexity of the algorithm.
7 Results and Discussion The results obtained over 20 independent runs by proposed OCBO, OPSO, WCA and HS are reported here for five examples of science-engineering problems described in Sect. 6.1 and three examples of biological modelling problems given in Sect. 6.2. The approximated value of Fourier coefficients obtained for science ODE problems (Ex. 1–Ex. 3) are reported in Table 1 and ODEs for electrical engineering problems are reported in Table 2. Similarly the approximated value of coefficients obtained for biological modelling problems are presented
123
Determining Approximate Solutions of Nonlinear Ordinary… Table 4 Comparative results of Generalized Distance obtained for 20 independent runs with OCBO, OPSO and WCA and HS Ex.
GD value
Ex. 1
Best
2.86E-06
3.26E-06
3.55E-06
4.26E-06
Average
5.42E-06
6.10E-06
6.49E-06
8.62E-06
Ex. 2
Ex. 3
Ex. 4
Ex. 5
OCBO
OPSO
WCA
HS
Worst
8.22E-06
8.86E-06
9.71E-06
12.51E-06
SD
2.21E-06
2.43E-06
2.45E-06
3.20E-06
Best
9.18E-05
1.02E-04
2.35E-04
4.12E-04
Average
3.51E-04
5.57E-04
7.83E-04
8.83E-04
Worst
5.99E-04
7.83E-04
1.21E-03
1.35E-03
SD
1.67E-04
1.96E-04
2.12E-04
2.54E-04
Best
6.62E-04
8.76E-04
1.14E-03
1.74E-03
Average
1.21E-03
1.85E-03
2.56E-03
3.12E-03
Worst
3.65E-03
4.98E-03
6.22E-03
7.31E-03
SD
6.11E-04
8.25E-04
1.03E-03
1.62E-03
Best
3.66E-03
3.80E-03
3.80E-03
4.12E-03
Average
4.54E-03
4.62E-03
4.64E-03
7.56E-03
Worst
7.66E-03
8.02E-03
8.10E-03
10.11E-03
SD
9.08E-04
1.04E-03
1.08E-03
2.21E-03
Voltage Best
4.08E-04
4.18E-04
4.25E-04
5.74E-04
Average
7.62E-03
8.12E-03
9.52E-03
1.24E-02
Worst
1.59E-02
1.98E-02
2.17E-02
2.63E-02
SD
2.38E-03
3.12E-03
5.34E-03
6.88E-03
Current
Ex. 6
Ex. 7
Ex. 8
Best
2.61E-03
3.24E-03
3.75E-03
4.32E-03
Average
9.25E-03
1.14E-02
1.23E-02
1.45E-02
Worst
1.42E-02
1.94E-02
2.73E-02
2.94E-02
SD
3.45E-03
4.12E-03
6.48E-03
6.75E-03
Best
6.22E-04
7.35E-04
8.14E-04
8.94E-04
Average
7.42E-04
8.54E-04
9.45E-04
9.88E-04
Worst
2.68E-03
3.12E-03
3.57E-03
3.99E-03
SD
1.18E-04
1.75E-04
2.16E-04
2.46E-04
Best
7.68E-04
8.07E-04
9.13E-04
9.84E-04
Average
9.32E-04
9.86E-04
1.23E-03
1.72E-03
Worst
2.43E-03
2.40E-03
2.68E-03
2.85E-03
SD
1.57E-04
1.74E-04
1.94E-04
2.10E-04
Best
1.25E-03
1.71E-03
2.53E-03
2.91E-03
Average
2.20E-03
2.80E-03
3.64E-03
4.05E-03
Worst
3.12E-03
3.91E-03
4.81E-03
5.42E-03
SD
1.02E-03
1.11E-03
1.23E-03
1.18E-03
123
A. Panda, S. Pani Table 5 Comparative results of run time obtained for 20 independent runs with OCBO, OPSO, WCA and HS
Ex.
Comp. time
OCBO
Ex. 1
Best
12.4510
15.3222
9.0202
10.1240
Average
12.6870
15.5260
9.2042
10.3250
Worst
10.4246
Ex. 2
Ex. 3
Ex. 4
Ex. 5
Ex. 6
Ex. 7
Ex. 8
OPSO
WCA
HS
12.8010
15.7500
9.3721
SD
0.1778
0.2140
0.1760
0.1642
Best
15.9212
18.6234
11.4116
12.8073
Average
15.9844
18.7015
11.4730
12.9124
Worst
16.1006
18.8425
11.5488
13.0112
SD
0.0816
0.1226
0.0798
0.1304
Best
14.7314
17.4014
10.6225
11.9700
Average
14.7862
17.4621
10.6310
11.9762
Worst
14.8405
17.5123
10.6375
11.9805
SD
0.0658
0.0712
0.0850
0.0755
Best
16.6324
19.0210
12.1022
13.2120
Average
16.7514
19.2631
12.2310
13.5024
Worst
16.8620
19.4100
12.3532
13.8218
SD
0.1148
0.1964
0.1255
0.2506
Best
15.7344
18.5017
11.3008
12.5842
Average
15.7912
18.5869
11.3120
12.5961
Worst
15.8413
18.6210
11.3224
12.6210
SD
0.0714
0.0902
0.0821
0.0935
Best
14.9115
17.7526
10.9025
12.2133
Average
14.9743
17.7920
10.9504
12.3025
Worst
15.0520
17.8621
10.9805
12.3843
SD
0.0591
0.0635
0.0412
0.0762
Best
15.2325
17.9624
11.4110
12.5312
Average
15.2947
18.0328
11.4562
12.6024
Worst
12.6840
15.3504
18.1004
11.4987
SD
0.0621
0.0658
0.0425
0.0780
Best
17.5328
20.4020
14.0020
15.7125
Average
17.5836
20.4621
14.0436
15.7720
Worst
17.6440
20.5203
14.0866
15.8262
0.0524
0.0584
0.0412
0.0645
SD
in Table 3. In all the eight examples it is observed that the obtained coefficients by the meta-heuristic algorithms are nearby close to each other. The plot of true analytical response and the proposed approximate response obtained with Fourier Series coefficient (values obtained by the four meta-heuristic algorithms reported in Tables 1 and 2) for all the five examples of science-engineering problems are shown in Figs. 4, 5 and 6 (Fig. 4 for Ex. 1–2, Fig. 5 for Ex. 3–4 and Fig. 6 representing approximate voltage and current value in Ex. 5). Similarly the response matching for the three biological modelling problems are presented in Figs. 7 and 8 (Fig. 7 for Ex. 6–7 and Fig. 8 corresponds to Ex. 8). Close response matching between the obtained curves with respect to the exact solution justifies the significance of the proposed method. In most of the cases on a close look
123
Determining Approximate Solutions of Nonlinear Ordinary…
at the curves it is observed that the red line (OCBO response) is close to the exact analytical curve, the blue line (OPSO response) marginally overshoots it. The green (WCA response) and dotted line (HS response) are slightly lower than the true response (under-fitting). The similar trend is also observed in Tables 1, 2 and 3. In these problems true parameters are unknown. But comparison between most parameters (in Tables 1, 2 and 3) reveals higher values for OPSO and lower values for WCA and HS compared to the OCBO values. The best, average, worst and standard deviation values of G D achieved over 20 independent runs are presented in Table 4. In all the eight examples, among the four algorithms the OCBO based learning results in lower G D values which are highlighted in bold letters. Except Ex. 4 in all the rest cases mostly the superiority in terms of G Dare OCBO followed by OPSO, WCA and HS. In case of Ex. 4 the GD values of OPSO and WCA are approximately equal, though from Fig. 5b the blue line overshoots where as green line under-fits. Therefore in this case the approximate average distance between elements of green and blue curve are same from the black solid curve (exact) though they lie on both sides of the exact solution. The computational complexity is not a matter in this case as the applications taken here are offline in nature. However to know the comparative value of complexity run time of the four algorithms obtained over the 20 independent runs are presented in Table 5. In all the eight examples it is observed that by combining the orthogonal array the overall computational time of the algorithm increases. It is due to the time required to generate orthogonal combination. Thus the run time of WCA is lower followed by HS, than that of the orthogonal algorithms (OCBO and OPSO). Earlier it has been reported that the CBO has lower computational complexity than the PSO for solving multi-objective [21] and constrained optimization problems [22]. In this case also after combining with the orthogonal array the complexity of OCBO in all the five examples reported lower than OPSO.
8 Conclusion In this paper an attempt is made to determine approximate solution of nonlinar ordinary differential equation (ODE). Fourier series expansion is used as the approximator and it’s coefficients are optimized by meta-heuristic algorithms. A hybrid algorithm OCBO is derived by combining orthogonal array with colliding bodies optimization. Simulation studies were carried out on three popular applications of ODE in science, two applications in electrical engineering and three applications on biological process modelling. Comparative analysis is carried out with results obtained by Orthogonal PSO, water cycle algorithm and Harmonic Search over twenty independent runs. Superior performance of OCBO is reported over counter part algorithms in terms of close response matching of the approximated solution with the exact analytical solution, lower generalized distance value. It is also observe that adding orthogonal array enhance the accuracy but simultaneously computational complexity of the process also increase slightly due to incorporation of additional steps of creating the array. Thus complexity of WCA and HS are lower. However in the eight examples the proposed OCBO has lower complexity than the OPSO algorithm. Compliance with Ethical Standards Conflict of interest The authors declare that they have no conflict of interest. Human Participants or Animals Further, this manuscript does not contain any studies with human participants or animals performed by any of the authors.
123
A. Panda, S. Pani
References 1. Trench WF (2001) Elementary differential equations with boundary value problems. Brooks/Cole Thomson Learning, Pacific Grove, ISBN: 978-0534360900 2. Yalcinbas S, Sezer M (2000) The approximate solution of high-order linear Volterra–Fredholm integrodifferential equations in terms of Taylor polynomials. Appl Math Comput 112:291–308 3. Darania P, Ebadian A (2007) A method for the numerical solution of the integro differential equations. Appl Math Comput 188:657–668 4. Darania P, Ivaz K (2008) Numerical solution of nonlinear VolterraFredholm integro-differential equations. Comput Math Appl 56:2197–2209 5. Roul P, Meyer P (2011) Numerical solutions of systems of nonlinear integro differential equations by Homotopy-perturbation method. Appl Math Model 35:4234–4242 6. Cao H, Kang L, Chen Y (2000) Evolutionary modeling of systems of ordinary differential equations with genetic programming. Genet Program Evolvable Mach 1:309–337 7. Mateescu GD (2006) On the application of genetic algorithms to differential equations. Romanian J Econ Forecast 2:5–9 8. Lee ZY (2006) Method of bilaterally bounded to solution Blasius equation using particle swarm optimization. Appl Math Comput 179:779–786 9. Babaei M (2013) A general approach to approximate solutions of nonlinear differential equations using particle swarm optimization. Appl Soft Comput 13:3354–3365 10. Sadollah A, Eskandar H, Yoo DG, Kim JH (2015) Approximate solving of nonlinear ordinary differential equations using least square weight function and metaheuristic algorithms. Eng Appl Artif Intell 40:117– 132 11. Sadollah A, Choi Y, Yoo DG, Kim JH (2015) Metaheuristic algorithms for approximate solution to ordinary differential equations of longitudinal fins having various profiles. Appl Soft Comput 33:360– 379 12. Zhang Q, Leung YW (1999) An orthogonal genetic algorithm for multimedia multicast routing. IEEE Trans Evolut Comput 3:53–62 13. Gong W, Cai Z, Ling CX (2006) ODE: a fast and robust differential evolution based on orthogonal design. In: Proceedings of 19th international australian joint conference on artificial intelligence. In Advances in Artificial Intelligence, Springer, Berlin Heidelberg, pp 709–718 14. Gong W, Cai Z, Jiang L (2008) Enhancing the performance of differential evolution using orthogonal design method. Appl Math Comput 206(1):56–69 15. Ho SY, Lin HS, Liauh WH, Ho SJ (2008) OPSO: orthogonal particle swarm optimization and its application to task assignment problems. IEEE Trans Syst Man Cybern Part A Syst Hum 38(2):288–298 16. Zhan ZH, Zhang J, Li Y, Shi YH (2011) Orthogonal learning particle swarm optimization. IEEE Trans Evolut Comput 15(6):832–847 17. Baghel V, Nanda SJ, Panda G (2011) New GOPSO and its application to robust identification. In: Proceedings of international conference on energy, automation and signal (ICEAS-2011), IEEE, pp 1–6 18. Kaveh A, Mahdavi VR (2014) Colliding bodies optimization: a novel meta-heuristic method. Comput Struct 139:18–27 19. Kaveh A, Mahdavi VR (2015) Colliding Bodies optimization. Springer International Publishing, Berlin, ISBN: 978-3319196589 20. Panda A, Pani S (2014) A new model based on colliding bodies optimization for identification of hammerstein plant. In: Proceedings of 11th Indian conference (INDICON-2014), IEEE Indian Council, pp 1–5 21. Panda A, Pani S (2016) Multi-objective colliding bodies optimization. In: Proceedings of fifth international conference on soft computing for problem solving, Springer Singapore, pp 651–664 22. Panda A, Pani S (2016) A symbiotic organisms search algorithm with adaptive penalty function to solve multi-objective constrained optimization problems. Appl Soft Comput 46:344–360 23. Kreyszig E (2015) Advanced engineering mathematics, Tenth edn. Wiley, Hoboken, ISBN: 9788126554232 24. Boyce WE, DiPrima RC (2005) Elementary differential equations and boundary value problems, Eighth edn. Wiley, Hoboken, ISBN: 978-0471433385 25. Leung YW, Wang Y (2001) An orthogonal genetic algorithm with quantization for global numerical Optimization. IEEE Trans Evolut Comput 5(1):41–53 26. Ross R (1911) The prevention of malaria. Murray, London 27. Glass L, Mackey C (1979) A simple model for phase locking of biological oscillators. J Math Biol 7:339352
123
Determining Approximate Solutions of Nonlinear Ordinary… 28. Dugard L, Verriest EI (1997) Stability and control of time-delay systems. In: Lecture notes in control and information sciences, vol. 228, Springer 29. Shakeri F, Dehghan M (2008) Solution of delay differential equations via a homotopy perturbation method. Math Comput Model 48(3):486–498 30. Yildirim A, Kocak H, Tutkun S (2012) Reliable analysis for delay differential equations arising in mathematical biology. J King Saud Univ Sci 24(4):359–365 31. Mendes P, Kell DB (1998) Non-linear optimization of biochemical pathways: applications to metabolic engineering and parameter estimation. Bioinformatics 14:869–883 32. Mendes P (2001) Modeling large biological systems from functional genomic data: parameter estimation. In: Kitano H (ed) Foundations of systems biology. MIT Press, Cambridge, pp 163–186 33. Moles C, Mendes P, Bangle JR (2003) Parameter estimation in biochemical pathway: a comparison of global optimization methods. Genome Res 13(11):2467–2474 34. Peifer M, Timmer J (2007) Parameter estimation in ordinary differential equations for biochemical processes using the method of multiple shooting. IET Syst Biol 1(2):78–88 35. Qian L, Wang H, Dougherty ER (2008) Inference of noisy nonlinear differential equation models for gene regulatory networks using genetic programming and Kalman filtering. IEEE Trans Signal Process 56(7):3327–3339 36. Zhan C, Situ W, Yeung LF, Tsang PW, Yang G (2014) A parameter estimation method for biological systems modelled by ode/dde models using spline approximation and differential evolution algorithm. IEEE/ACM Trans Comput Biol Bioinform (TCBB) 11(6):1066–1076 37. Zhan C, Li BY, Yeung LF (2015) Structural and practical identifiability analysis of S-system. IET Syst Biol 9(6):285–93
123