Volume 36, 2019, pp. 27-42 Ratio Mathematica 27 Solution of two-point fuzzy boundary value problems by fuzzy neural networks Mazin Hashim Suhhiem* Basim Nasih Abood+ Mohammed Hadi Lafta++ Abstract In this work, we have introduced a modified method for solving second-order fuzzy differential equations. This method based on the fully fuzzy neural network to find the numerical solution of the two-point fuzzy boundary value problems for the ordinary differential equations. The fuzzy trial solution of the two-point fuzzy boundary value problems is written based on the concepts of the fully fuzzy feed-forward neural networks which containing fuzzy adjustable parameters. In comparison with other numerical methods, the proposed method provides numerical solutions with high accuracy. Keywords: Two-point fuzzy boundary value problem; fully fuzzy neural network; fuzzy trial solution; minimized error function; hyperbolic tangent activation function. ـــ .ــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــ * Department of Statistics, University of Sumer, Alrifaee, Iraq; mazin.suhhiem@yahoo.com + Department of Mathematics, University of Wasit, Alkut, Iraq; basimabood@yahoo.com ++ Department of Statistics, University of Sumer, Alrifaee, Iraq; mohamnedhadi@yahoo.com Received on March 5th, 2019. Accepted on April 29rd, 2019. Published on June 30th, 2019. doi:10.23755/rm.v36i1.455. ISSN: 1592-7415. eISSN: 2282-8214. ©Mazin Suhhiem et al. This paper is published under the CC-BY licence agreement. Mazin H. Suhhiem, Basim N. Abood, Mohammed H. Lafta 28 1. Introduction Many methods have been developed so far for solving fuzzy differential equations (FDEs) since it is utilized widely for the purpose of modelling problems in science and engineering. Most of the practical problems require the solution of the FDE which satisfies fuzzy initial conditions or fuzzy boundary conditions, therefore, the FDE must be solved. Many FDE could not be solved exactly, thus considering their approximate solutions is becoming more important. The theory of FDE was "first formulated by Kaleva and Seikkala..Kaleva was formulated FDE in terms of the "Hukuhara derivative" (H-derivative). Buckley and Feuring have given a very general formulation of a first order" fuzzy "initial value problem. They first find the crisp solution, make it fuzzy and then check if it satisfies the FDE. In 1990 researchers began using the artificial neural network (ANN) for solving ordinary differential equation (ODE) and partial differential equation (PDE) such as: Lee and Kang in [1]; Meade and Fernandez in [2,3]; Lagaris and Likas in [4]; Liu and Jammes in [5]; Tawfiq in [6]; Malek and Shekari in [7]; Pattanaik and Mishra in [8]; Baymani and Kerayechian in [9]; and other researchers. In 2010 researchers began using ANN for solving a fuzzy differential equation such as: Effati and Pakdaman in [10]; Mosleh and Otadi in [11]; Ezadi and Parandin in [12]. In 2012 researchers began using partially (non-fully) fuzzy artificial neural network(FANN) for solving a fuzzy differential equation such as Mosleh and Otadi in [13,14,15]. In (2016) Suhhiem [16] developed and used partially FANN for solving fuzzy and non-fuzzy differential equations. In this work, we have used fully feed forward fuzzy neural network to find the numerical solution of the two-point fuzzy boundary value problems for the ordinary differential equations. The fuzzy trial solution of the fuzzy boundary value problem is written as a sum of two parts. The first part satisfies the fuzzy boundary condition, it contains no fuzzy adjustable parameters. The second part involves fully fuzzy feed-forward neural networks which containing fuzzy adjustable parameters. Solution of two-point fuzzy boundary value problems by fuzzy neural networks 29 2 Basic definitions In this section, the basic notations which are used in fuzzy calculus are introduced Definition(𝟏),[𝟏𝟔]: The r - level ( or r - cut ) set of a fuzzy set à labeled by Ar is the crisp set of all x in X (universal set) such that : µÃ(x) ≥ r ; i. e. Ar = {x ∈ X ∶ µÃ(x) ≥ r , r ∈ [0,1] } . (1) Definition(𝟐), 𝐅𝐮𝐳𝐳𝐲 𝐍𝐮𝐦𝐛𝐞𝐫[𝟏𝟔]: A fuzzy number ũ is completely determined by an ordered pair of functions (u (r) , u (r)), 0 ≤ r ≤ 1, which satisfy the following requirements: 𝟏) u (r) is a bounded left continuous and non-decreasing function on [0,1]. 𝟐) u (r) is a bounded left continuous and non-increasing function on [0,1]. 𝟑) u (r) ≤ u (r) , 0 ≤ r ≤ 1. (2) The crisp number (a) is simply represented by: u (r) = u (r) = a , 0 ≤ r ≤ 1 . The set of all the fuzzy numbers is denoted by E1. Remark(𝟏),[𝟏𝟎]: For arbitrary ũ = (u , u) , ṽ = (v , v) and K ∈ R, the addition and multiplication by K For all r ∈ [0,1] can be defined as: 𝟏) (u + v) (r) = u (r) + v (r). 𝟐) (u + v) (r) = u (r) + v (r). 𝟑) (Ku) (r) = K u (r), (Ku) (r) = K u (r) , if K ≥ 0. 𝟒) (Ku) (r) = K u (r), (Ku) (r) = K u (r), if K < 0. (3) Remark(𝟐),[𝟏𝟔]: The distance between two arbitrary fuzzy numbers ũ = (u , u) and ṽ = (v , v) is given as: D (ũ , ṽ) = [∫ ( u (r) - v (r) 1 0 ) 2 dr + ∫ ( u (r) - v (r) 1 0 ) 2 dr] 1 2 (4) Remark(𝟑),[𝟏𝟔]: (E1,D) is a complete metric space. Mazin H. Suhhiem, Basim N. Abood, Mohammed H. Lafta 30 Definition (𝟑) , 𝐅𝐮𝐳𝐳𝐲 𝐅𝐮𝐧𝐜𝐭𝐢𝐨𝐧 [𝟏𝟔] : The function F: R ⟶ E1 is called a fuzzy function. We call every function defined in set à ⊆ E1 to B̃ ⊆ E1 a fuzzy function. Definition(𝟒),[𝟏𝟎]: The fuzzy function F: R ⟶ E1 is said to be continuous if: For an arbitrary t1 ∈ R and ϵ > 0 there exists a δ > 0 such that: |t - t1| < δ ⇒ D (F (t), F(t1)) < ϵ, where D is the distance between two fuzzy numbers. Definition (5),[16]: Let I be a real interval. The r-level set of the fuzzy function y: I → E1 can be denoted by: [y(x)]r = [y1 r(x), y2 r (x)], x ∈ I , r ∈ [0,1] (5) The Seikkala derivative yˊ(x) of the fuzzy function y(x) is defined by: [yˊ(x)]r = [(y1 r)ˊ(x), (y2 r )ˊ(x)], x ∈ I, r ∈ [0,1] (6) Definition (6),[𝟏𝟎]: let u and v ∈ E1. If there exist w ∈ E1 such that: u = v+w then w is called the H-difference (Hukuhara-difference) of u and v and it is denoted by w = u ⊝ v. In this work, the ⊝ sign stands always for H-difference, and let us remark that u ⊝ v ≠ u + (-1) v . Definition (7), 𝐅𝐮𝐳𝐳𝐲 𝐃𝐞𝐫𝐢𝐯𝐚𝐭𝐢𝐯𝐞[𝟏𝟐]: Let F : (a,b) → E1 and t0 ∈ (a,b).We say that F is H-differential (Hukuhara-differential) at x0, if there exists an element Fˊ(x0) ∈ E 1 such that for all h> 0 (sufficiently small), ∃ F (x0 +h)⊝F(x0), F(x0) ⊝ F (x0 - h) and the limits (in the metric D) lim h→0 F(x0 + h) ⊝F(x0) h = lim h→0 F(x0) ⊝ F(x0 − h) h = Fˊ(x0) (7) Then Fˊ(x0) is called fuzzy derivative (H-derivative) of F at x0. where D is the distance between two fuzzy numbers. Solution of two-point fuzzy boundary value problems by fuzzy neural networks 31 3 Fully fuzzy neural network [6,16] Artificial neural networks are learning machines that can learn any arbitrary functional mapping between input and output. They are fast machines and can be implemented in parallel, either in software or in hardware. In fact, the computational complexity of ANN is polynomial in the number of neurons used in the network. Parallelism also brings with it the advantages of robustness and fault tolerance. (i.e.) ANN is a simplified mathematical model of the human brain. It can be implemented by both electric elements and computer software. It is a parallel distributed processor with large numbers of connections It is an information processing system that has certain performance characters in common with biological neural networks. A fuzzy neural network or neuro-fuzzy system is a learning machine that finds the parameters of a fuzzy system (i.e., fuzzy set, fuzzy rules) by exploiting approximation techniques from neural networks. Combining fuzzy systems with neural networks. Both neural networks and fuzzy systems have some things in common. They can be used for solving problems (e. g. fuzzy differential equations, fuzzy integral equations, etc ). If all the adjustable parameters (weights and biases) are fuzzy numbers, then the fuzzy neural network is called fully fuzzy neural network; otherwise it is called partially fuzzy neural network. 4 Solution of FDEs by fully fuzzy neural network To solve any fuzzy ordinary differential equation, we consider a three- layered fully fuzzy neural network with one unit entry x, one hidden layer consisting of m activation functions and one unit output N(x). The activation function for the hidden units of our fully fuzzy neural network is the hyperbolic tangent function (s(∝) = tanh (∝)). Here the dimension of a fully fuzzy neural network is (1 × m × 1) (figure1). Mazin H. Suhhiem, Basim N. Abood, Mohammed H. Lafta 32 Figure1: (1 × m × 1) Fully fuzzy feed-forward neural network. For every entry x (where x ≥ 0) the mathematical operations in the fully fuzzy neural network can be described as: Input unit: x = x, (8) Hidden units : [zj]r = [[zj]r L , [zj]r U ] = [s ([netj]r L ) , s ([netj]r U )] (9) where [netj]r L = x [wj]r L + [bj]r L (10) [netj]r U = x [wj]r U + [bj]r U (11) Solution of two-point fuzzy boundary value problems by fuzzy neural networks 33 Output unit: [N(x)]r = [[N(x)]r L , [N(x)]r U] (12) Where [N(x)]r L=∑ min {mj=1 [vj]r L [zj]r L , [vj]r L [zj]r U , [vj]r U [zj]r L , [vj]r U [zj]r U } (13) [N(x)]r U=∑ max {mj=1 [vj]r L [zj]r L , [vj]r L [zj]r U , [vj]r U [zj]r L , [vj]r U [zj]r U } (14) Where [zj]r L = s (x [wj]r L + [bj]r L ) (15) [zj]r U = s (x [wj]r U + [bj]r U ) (16) 5 Description of the proposed method For illustration the proposed method, we will consider the two points fuzzy boundary value problems: y´´(x) = f (x, y(x) , y´(x) ) , x ∈ [a , b] (17) with the fuzzy boundary conditions: y(a) = A and y(b) = B, where A and B are fuzzy numbers in E1 with r-level sets: [A]r = [A , A] and [B]r = [B , B] . The fuzzy trial solution for this problem is: [yt(x)]r = b − x b − a [A]r + x − a b − a [B]r +(x − a) (x − b) [N(x)]r (18) This fuzzy trial solution by intention satisfies the fuzzy boundary conditions in (17). The error function that must be minimized for problem (17) is in the form: E= ∑ (Eir L + Eir U) g i=1 (19) where Mazin H. Suhhiem, Basim N. Abood, Mohammed H. Lafta 34 Eir L = [ [ d2yt (xi) dx2 ] r L − [f (xi , yt (xi), d yt (xi) dx )] r L ] 2 (20) Eir U = [ [ d2yt (xi) dx2 ] r U − [f (xi , yt (xi), d yt (xi) dx )] r U ] 2 (21) where {xi}i=1 g are discrete points belonging to the interval [a , b] (training set) and in the cost function (19), Er L and Er U can be viewed as the squared errors for the lower limits and the upper limits of the r – level sets, respectively. Now, to drive the minimized error function for problem (17): From (18) we can find: [yt(x)]r L = b − x b − a [A]r L + x − a b − a [B]r L +( x2 − (a + b)x + ab)[N(x)]r L (22) [yt(x)]r U = b − x b − a [A]r U + x − a b − a [B]r U +( x2 − (a + b)x + ab)[N(x)]r U (23) Then we get: d[yt(x)]r L dx = −1 b − a [A]r L + 1 b − a [B]r L +( x2 − (a + b)x + ab) d[N(x)]r L dx +(2x−a − b)[N(x)]r L (24) d[yt(x)]r U dx = −1 b − a [A]r U + 1 b − a [B]r U +( x2 − (a + b)x + ab) d[N(x)]r U dx +(2x−a − b)[N(x)]r U (25) Therefore, we have: [ d2yt (x) dx2 ] r L = ( x2 − (a + b)x + ab) d2[N(x)]r L dx2 +2(2x − a − b ) d[N(x)]r L dx + 2[N(x)]r L (26) [ d2yt (x) dx2 ] r U = ( x2 − (a + b)x + ab) d2[N(x)]r U dx2 +2(2x − a − b ) d[N(x)]r U dx + 2[N(x)]r U (27) Then (20) and (21) can be rewritten as: Solution of two-point fuzzy boundary value problems by fuzzy neural networks 35 Eir L = [( xi 2 − (a + b)xi + ab) d2[N(xi)]r L dx2 + 2(2xi − a − b ) d[N(xi)]r L dx + 2[N(xi)]r L – f(xi , b − xi b − a [A]r L + xi − a b − a [B]r L + ( xi 2 − (a + b)xi + ab)[N(xi)]r L , −1 b − a [A]r L+ 1 b − a [B]r L+( xi 2 − (a + b)xi + ab) d[N(xi)]r L dx + (2xi − a − b)[N(xi)]r L) ]2 (28) Eir U = [( xi 2 − (a + b)xi + ab) d2[N(xi)]r U dx2 + 2(2xi − a − b ) d[N(xi)]r U dx + 2[N(xi)]r U – f(xi , b − xi b − a [A]r U + xi − a b − a [B]r U + ( xi 2 − (a + b)xi + ab)[N(xi)]r U , −1 b − a [A]r U+ 1 b − a [B]r U+( xi 2 − (a + b)xi + ab) d[N(xi)]r U dx + (2xi − a − b)[N(xi)]r U) ]2 (29) Where [N(xi)]r L = ∑ min {mj=1 [vj]r L s (xi [wj]r L + [bj]r L ) , [vj]r L s (xi [wj]r U + [bj]r U ) , [vj]r U s (xi [wj]r L + [bj]r L ) , [vj]r U s (xi [wj]r U + [bj]r U ) } (30) [N(xi)]r L = ∑ max {mj=1 [vj]r L s (xi [wj]r L + [bj]r L ) , [vj]r L s (xi [wj]r U + [bj]r U ) , [vj]r U s (xi [wj]r L + [bj]r L ) , [vj]r U s (xi [wj]r U + [bj]r U ) } (31) d[N(xi)]r L dx = ∑ min {mj=1 [vj]r L [wj]r L s´ (xi [wj]r L + [bj]r L ) , [vj]r L [wj]r U s´ (xi [wj]r U + [bj]r U ) , [vj]r U [wj]r L s´ (xi [wj]r L + [bj]r L ) , [vj]r U [wj]r U s´ (xi [wj]r U + [bj]r U ) } (32) d[N(xi)]r U dx = ∑ max {mj=1 [vj]r L [wj]r L s´ (xi [wj]r L + [bj]r L ) , [vj]r L [wj]r U s´ (xi [wj]r U + [bj]r U ) , [vj]r U [wj]r L s´ (xi [wj]r L + [bj]r L ) , [vj]r U [wj]r U s´ (xi [wj]r U + [bj]r U ) } (33) d2[N(xi)]r L dx2 = ∑ min{mj=1 [vj]r L ([wj]r L )2s ́ ́ (xi [wj]r L +[bj]r L ) , [vj]r L ([wj]r U )2 s ́ ́ (xi [wj]r U + [bj]r U ) , [vj]r U ([wj]r L )2s ́ ́ (xi [wj]r L +[bj]r L ) , [vj]r U ([wj]r U )2s ́ ́ (xi [wj]r U + [bj]r U ) } (34) Mazin H. Suhhiem, Basim N. Abood, Mohammed H. Lafta 36 d2[N(xi)]r U dx2 = ∑ max{mj=1 [vj]r L ([wj]r L )2s ́ ́ (xi [wj]r L +[bj]r L ) , [vj]r L ([wj]r U )2 s ́ ́ (xi [wj]r U + [bj]r U ) , [vj]r U ([wj]r L )2s ́ ́ (xi [wj]r L +[bj]r L ) , [vj]r U ([wj]r U )2s ́ ́ (xi [wj]r U + [bj]r U ) } (35) where s´ and s ́ ́ are the first and second derivative of the hyperbolic tangent function. Then we substitute (28) and (29) in (19) to find the error function that must be minimized for problem (17). 6. Numerical example In this section, we will solve two problems about two-point fuzzy boundary value problem. We have used (1 × 10 × 1) fully fuzzy feed-forward neural network. The activation function of each hidden unit is the hyperbolic tangent activation function. The analytical solutions [ya(x)]r L and [ya(x)]r U has been known in advance. Therefore, we test the accuracy of the obtained solutions by computing the deviation: e (x , r) = |[ya(x)]r U − [yt (x)]r U| , e (x , r)= |[ya(x)]r L − [yt(x)]r L| To minimize the error function, we have used BFGS quasi-Newton method (For more details, see [16]). The computer programs which we have used in this work are coded in MATLAB 2015. Example (1): Consider the linear fuzzy boundary value problem: y´´ (x) − y´(x) = 1 . with x ∈ [0, 0.5] y(0) = [ 2 + r , 4 − r ] , y(0.5)= [5 + r , 7 − r] . where r ∈ [0, 1]. The analytical solutions for this problem are: [ya(x)]r L = ( 2 + r − 3 e0.5−1 ) + ( 3 e0.5−1 )ex [ya(x)]r U= ( 4 − r − 3 e0.5−1 ) + ( 3 e0.5−1 )ex The trial solutions for this problem are: [yt(x)]r L= (1 − 2x) (2 + r) + 2x (4 − r)+(x2 − 0.5 x ) [N(x)]r L Solution of two-point fuzzy boundary value problems by fuzzy neural networks 37 [yt(x)]r U= (1 − 2x) (5 + r) + 2x (7 − r) + (x2 − 0.5 x ) [N(x)]r U The fully fuzzy feed forward neural network has been trained by using a grid of ten equidistant points in [0, 0.5]. The error function that must be minimized for this problem will be: E= ∑ (Eir L + Eir U)11i=1 (36) where Eir L =[( xi 2 − 0.5xi) d2[N(xi)]r L dx2 + (4xi − 1) d[N(xi)]r L dx + 2[N(xi)]r L − ( xi 2 − 0.5xi) d[N(xi)]r L dx − (2xi − 0.5)[N(xi)]r L + 4r − 5 ]2 (37) Eir U =[( xi 2 − 0.5xi) d2[N(xi)]r U dx2 + (4xi − 1) d[N(xi)]r U dx + 2[N(xi)]r U − ( xi 2 − 0.5xi) d[N(xi)]r U dx − (2xi − 0.5)[N(xi)]r U + 4r − 5 ]2 (38) numerical solutions for this problem can be found in table (1). Table (1): Numerical result for example (1), x=1. r [yt(x)]r L e (x , r) [yt(x)]r U e (x , r) 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 9.946164141 10.04616401 10.14616481 10.24616458 10.34616447 10.44616422 10.54616396 10.64616391 10.74616385 10.84616389 10.94616389 3.29137e-7 1.96846e-7 9.95565e-7 7.63284e-7 6.60993e-7 4.09513e-7 1.47232e-7 9.75941e-8 3.39072e-8 7.52383e-8 7.39070e-8 11.94616425 11.84616411 11.74616478 11.64616385 11.54616387 11.44616389 11.34616391 11.24616382 11.14616384 11.04616386 10.94616386 4.33916e-7 2.93475e-7 9.70548e-7 3.95104e-8 5.67802e-8 7.56011e-8 9.53493e-8 1.15291e-8 2.63433e-8 5.26859e-8 4.56782e-8 Mazin H. Suhhiem, Basim N. Abood, Mohammed H. Lafta 38 Example (2): Consider the non-linear fuzzy boundary value problem: y´´ (x) = - (y´(x)) 2 . with x ∈ [0 , 2] y(0) = [ r , 2 − r ] , y(2) = [1 + r , 3 − r] and r ∈ [0 , 1]. The analytical solutions for this problem are: [ya(x)]r L = ln (x + 2 e−1 ) + r − ln 2 e−1 [ya(x)]r U= ln (x + 2 e−1 ) + 2 − r − ln 2 e−1 The trial solutions for this problem are: [yt(x)]r L = r 2−x 2 + (1 + r) x 2 + x (x − 2 ) [N(x)]r L [yt(x)]r U= (2 − r) 2−x 2 + (3 − r ) x 2 + x (x − 2 ) [N(x)]r U The fully fuzzy feed forward neural network has been trained by using a grid of ten equidistant points in [0, 2]. The error function that must be minimized for this problem will be: E= ∑ (Eir L + Eir U)11i=1 (39) where Eir L =[( xi 2 − 2xi) d2[N(xi)]r L dx2 + (4xi − 4 ) d[N(xi)]r L dx + 2[N(xi)]r L + (( xi 2 − 2xi) d[N(xi)]r L dx + (2xi − 2 )[N(xi)]r L + 0.5)2 ]2 (40) Eir L =[( xi 2 − 2xi) d2[N(xi)]r U dx2 + (4xi − 4 ) d[N(xi)]r U dx + 2[N(xi)]r U + (( xi 2 − 2xi) d[N(xi)]r U dx + (2xi − 2 )[N(xi)]r U + 0.5)2 ]2 (41) Then we use (39) to update the weights and biases. Numerical solution for this problem can be found in table (2). Solution of two-point fuzzy boundary value problems by fuzzy neural networks 39 Table (2): Numerical result for example (2), x=1. r [yt(x)]r L e (x , r) [yt(x)]r U e (x , r) 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.620114507 0.720114507 0.820114507 0.920114507 1.020114507 1.120114507 1.220114507 1.320114516 1.420114512 1.520114507 1.620114514 3.24734e-10 4.66221-10 2.03208e-10 3.80684e-10 4.09557e-10 3.50405e-10 4.59008e-10 9.46681e-9 5.06564e-9 8.21899e-10 7.88763e-9 2.620114507 2.520114507 2.420114507 2.320114513 2.220114514 2.120114508 2.020114507 1.920114507 1.820114507 1.720114514 1.620114508 8.46634e-10 9.79602e-10 6.85555e-10 6.62032e-9 7.59010e-9 1.74006e-9 9.00817e-10 9.21604e-10 4.99811e-10 7.15955e-9 1.02988e-9 For the above two problems we have [N(xi)]r L = ∑ min {10j=1 [vj]r L s (xi [wj]r L + [bj]r L ) , [vj]r L s (xi [wj]r U + [bj]r U ) , [vj]r U s (xi [wj]r L + [bj]r L ) , [vj]r U s (xi [wj]r U + [bj]r U ) } [N(xi)]r L = ∑ max {10j=1 [vj]r L s (xi [wj]r L + [bj]r L ) , [vj]r L s (xi [wj]r U + [bj]r U ) , [vj]r U s (xi [wj]r L + [bj]r L ) , [vj]r U s (xi [wj]r U + [bj]r U ) } d[N(xi)]r L dx = ∑ min {10j=1 [vj]r L [wj]r L s´ (xi [wj]r L + [bj]r L ) , [vj]r L [wj]r U s´ (xi [wj]r U + [bj]r U ) , [vj]r U [wj]r L s´ (xi [wj]r L + [bj]r L ) , [vj]r U [wj]r U s´ (xi [wj]r U + [bj]r U ) } Mazin H. Suhhiem, Basim N. Abood, Mohammed H. Lafta 40 d[N(xi)]r U dx = ∑ max {10j=1 [vj]r L [wj]r L s´ (xi [wj]r L + [bj]r L ) , [vj]r L [wj]r U s´ (xi [wj]r U + [bj]r U ) , [vj]r U [wj]r L s´ (xi [wj]r L + [bj]r L ) , [vj]r U [wj]r U s´ (xi [wj]r U + [bj]r U ) } d2[N(xi)]r L dx2 = ∑ min{10j=1 [vj]r L ([wj]r L )2s ́ ́ (xi [wj]r L +[bj]r L ) , [vj]r L ([wj]r U )2 s ́ ́ (xi [wj]r U + [bj]r U ) , [vj]r U ([wj]r L )2s ́ ́ (xi [wj]r L +[bj]r L ) , [vj]r U ([wj]r U )2s ́ ́ (xi [wj]r U + [bj]r U ) } d2[N(xi)]r U dx2 = ∑ max{10j=1 [vj]r L ([wj]r L )2s ́ ́ (xi [wj]r L +[bj]r L ) , [vj]r L ([wj]r U )2 s ́ ́ (xi [wj]r U + [bj]r U ) , [vj]r U ([wj]r L )2s ́ ́ (xi [wj]r L +[bj]r L ) , [vj]r U ([wj]r U )2s ́ ́ (xi [wj]r U + [bj]r U ) } 7 Conclusion In this work, we have introduced a modified method to find the numerical solution of the two-point fuzzy boundary value problems for the ordinary differential equations. This method based on the fully fuzzy neural network to approximate the solution of the second-order fuzzy differential equations. For future studies, one can extend this method to find a numerical solution of the higher order fuzzy differential equations. Also, one may use this method for solving a fuzzy partial differential equation. Solution of two-point fuzzy boundary value problems by fuzzy neural networks 41 References [1] H. Lee, I.S. Kang. Neural Algorithms for Solving Differential Equations. Journal of Computational Physics, 91, 110-131. 1990. [2] A.J. Meade, A.A. Fernandes. The Numerical Solution of Linear Ordinary Differential Equations by Feed-Forward Neural Networks. Mathematical and Computer Modeling, 19(12), 1-25. 1994. [3] A.J. Meade, A.A. Fernandes. Solution of Nonlinear Ordinary Differential Equations by Feed-Forward Neural Networks. Mathematical and Computer Modeling, 20(9), 19-44. 1994. [4] I.E. Lagaris, A. Likas. Artificial Neural Networks for Solving Ordinary and Partial Differential Equations. Journal of Computational Physics, 104, 1- 26. 1997. [5] Liu, Jammes. Solving Ordinary Differential Equations by Neural Networks. Warsaw, Poland. 1999. [6] Tawfiq. On Design and Training of Artificial Neural Network for Solving Differential Equations. Ph.D. Thesis, College of Education, University of Baghdad, Iraq. 2004. [7] A. Malek, R. Shekari. Numerical Solution for High Order Differential Equations by Using a Hybrid Neural Network Optimization Method. Applied Mathematics and Computation, 183, 260-271. 2006. [8] S. Pattanaik, R.K. Mishra. Application of ANN for Solution of PDE in RF Engineering. International Journal on Information Sciences and Computing, 2(1), 74-79. 2008. [9] M. Baymani, A. Kerayechian. Artificial Neural Networks Approach for Solving Stokes Problem, Applied Mathematics, 1, 288-292. 2010. [10] S. Effati, M. Pakdaman. Artificial Neural Network Approach for Solving Fuzzy Differential Equations. Information Sciences, 180, 1434-1457. 2010. [11] M. Mosleh, M. Otadi. Fuzzy Fredholm Integro-Differential Equations with Artificial Neural Networks. Communications in Numerical Analysis, Article ID cna-00128, 1-13. 2012. [12] S Ezadi, N. Parandin. Numerical Solution of Fuzzy Differential Equations Based on Semi-Taylor by Using Neural Network. Journal of Basic and Applied Scientific Research, 3(1s), 477-482. 2013. Mazin H. Suhhiem, Basim N. Abood, Mohammed H. Lafta 42 [13] M. Mosleh, M. Otadi. Simulation and Evaluation of Fuzzy Differential Equations by Fuzzy Neural Network. Applied Soft Computing, 12, 2817-2827. 2012. [14] M. Mosleh. Fuzzy Neural Network For Solving a System of Fuzzy Differential Equations. Applied Soft Computing, 13, 3597-3607. 2013. [15] M. Mosleh, M. Otadi. Solving the Second Order Fuzzy Differential Equations by Fuzzy Neural Network. Journal of Mathematical Extension, 8(1), 11-27. 2014. [16] Suhhiem. Fuzzy Artificial Neural Network for Solving Fuzzy and Non- Fuzzy Differential Equations. Ph.D. Thesis, College of Sciences, AL- Mustansiriyah University, Iraq. 2016.