One of the most often used data science techniques in business, finance, supply chain management, production, and inventory planning is time-series forecasting. Due to the dearth of studies in the literature that propose unique weights and structure (WASD) based models for regression issues, the goal of this research is to examine the creation of such a model for time-series forecasting. Given that WASD neural networks have been shown to overcome limitations of traditional back-propagation neural networks, including slow training speed and local minima, a multi-function activated WASD for time-series (MWASDT) model that uses numerous activation functions, a new auto cross-validation method and a new prediction mechanism are proposed. The MWASDT model was used in forecasting the gross domestic product (GDP) for numerous nations to show off its exceptional capacity for learning and predicting. Compared to previous WASD-based models for time-series forecasting and traditional machine learning models that MATLAB has to offer, the new model has produced noticeably better forecasting results, especially on unseen data.
Citation: Spyridon D. Mourtas, Emmanouil Drakonakis, Zacharias Bragoudakis. Forecasting the gross domestic product using a weight direct determination neural network[J]. AIMS Mathematics, 2023, 8(10): 24254-24273. doi: 10.3934/math.20231237
[1] | Aliaa Burqan, Mohammed Shqair, Ahmad El-Ajou, Sherif M. E. Ismaeel, Zeyad AlZhour . Analytical solutions to the coupled fractional neutron diffusion equations with delayed neutrons system using Laplace transform method. AIMS Mathematics, 2023, 8(8): 19297-19312. doi: 10.3934/math.2023984 |
[2] | M. Sivakumar, M. Mallikarjuna, R. Senthamarai . A kinetic non-steady state analysis of immobilized enzyme systems with external mass transfer resistance. AIMS Mathematics, 2024, 9(7): 18083-18102. doi: 10.3934/math.2024882 |
[3] | Yudhveer Singh, Devendra Kumar, Kanak Modi, Vinod Gill . A new approach to solve Cattaneo-Hristov diffusion model and fractional diffusion equations with Hilfer-Prabhakar derivative. AIMS Mathematics, 2020, 5(2): 843-855. doi: 10.3934/math.2020057 |
[4] | Zihan Yue, Wei Jiang, Boying Wu, Biao Zhang . A meshless method based on the Laplace transform for multi-term time-space fractional diffusion equation. AIMS Mathematics, 2024, 9(3): 7040-7062. doi: 10.3934/math.2024343 |
[5] | Zeliha Korpinar, Mustafa Inc, Dumitru Baleanu . On the fractional model of Fokker-Planck equations with two different operator. AIMS Mathematics, 2020, 5(1): 236-248. doi: 10.3934/math.2020015 |
[6] | Ayyaz Ali, Zafar Ullah, Irfan Waheed, Moin-ud-Din Junjua, Muhammad Mohsen Saleem, Gulnaz Atta, Maimoona Karim, Ather Qayyum . New exact solitary wave solutions for fractional model. AIMS Mathematics, 2022, 7(10): 18587-18602. doi: 10.3934/math.20221022 |
[7] | Nisar Gul, Saima Noor, Abdulkafi Mohammed Saeed, Musaad S. Aldhabani, Roman Ullah . Analytical solution of the systems of nonlinear fractional partial differential equations using conformable Laplace transform iterative method. AIMS Mathematics, 2025, 10(2): 1945-1966. doi: 10.3934/math.2025091 |
[8] | Shabir Ahmad, Aman Ullah, Ali Akgül, Fahd Jarad . A hybrid analytical technique for solving nonlinear fractional order PDEs of power law kernel: Application to KdV and Fornberg-Witham equations. AIMS Mathematics, 2022, 7(5): 9389-9404. doi: 10.3934/math.2022521 |
[9] | Gulalai, Shabir Ahmad, Fathalla Ali Rihan, Aman Ullah, Qasem M. Al-Mdallal, Ali Akgül . Nonlinear analysis of a nonlinear modified KdV equation under Atangana Baleanu Caputo derivative. AIMS Mathematics, 2022, 7(5): 7847-7865. doi: 10.3934/math.2022439 |
[10] | Lynda Taleb, Rabah Gherdaoui . Approximation by the heat kernel of the solution to the transport-diffusion equation with the time-dependent diffusion coefficient. AIMS Mathematics, 2025, 10(2): 2392-2412. doi: 10.3934/math.2025111 |
One of the most often used data science techniques in business, finance, supply chain management, production, and inventory planning is time-series forecasting. Due to the dearth of studies in the literature that propose unique weights and structure (WASD) based models for regression issues, the goal of this research is to examine the creation of such a model for time-series forecasting. Given that WASD neural networks have been shown to overcome limitations of traditional back-propagation neural networks, including slow training speed and local minima, a multi-function activated WASD for time-series (MWASDT) model that uses numerous activation functions, a new auto cross-validation method and a new prediction mechanism are proposed. The MWASDT model was used in forecasting the gross domestic product (GDP) for numerous nations to show off its exceptional capacity for learning and predicting. Compared to previous WASD-based models for time-series forecasting and traditional machine learning models that MATLAB has to offer, the new model has produced noticeably better forecasting results, especially on unseen data.
In the present research, we prove existence results for a fourth-order differential equation system that takes the form:
{ϖ(4)(t)=f(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t)),a.e. t∈J=[0,1],ϖ(0)=ϖ0,ϖ′(0)=ϖ1andϖ″∈(BC), | (1.1) |
where f:[0,1]×R4n→Rn represents an L1 -Carathéodory function, ϖ0,ϖ1∈Rn and (BC) can be the boundary conditions that are given by one of the following:
(SL) Strum-Liouville boundary conditions on J
A0ϖ(0)−β0ϖ′(0)=r0,A1ϖ(1)+β1ϖ′(1)=r1. | (1.2) |
(P) Periodic boundary conditions on J
ϖ(0)=ϖ(1),ϖ′(0)=ϖ′(1), | (1.3) |
where (Ai)i∈{0,1}∈Mn×n(R), such that
∀i∈{0,1},∃κi≥0:⟨ϖ,Aiϖ⟩≥κi‖ϖ‖2,∀ϖ∈Rn |
∀i∈{0,1},ri∈R:βi∈{0,1},κi+βi>0. |
We refer to [1,2,3] for further findings that were achieved in the specific instance of a boundary value issue for only one differential equation of the fourth-order (n=1), for more details, please see [4,5,6]. Existence results for higher-order differential equations can be found in [7,8], and the general case of Nth order systems is discussed in [9,10,11].
The concept of the solution-tube of problem (1.1) is presented in this work; see [12,13,14]. This idea is inspired by [15] and [16], where solution-tubes for second and third order differential equations systems are defined, respectively, as follows:
{ϖ″(t)=f(t,ϖ(t),ϖ′(t)),a.e. t∈J,ϖ∈(BC), | (1.4) |
and
{ϖ‴(t)=f(t,ϖ(t),ϖ′(t),ϖ″(t)),a.e. t∈J,ϖ(0)=ϖ0,ϖ′∈(BC). | (1.5) |
We prove that the system (1.1) has solutions. For this system, we employ the concept of a solution tube, which extends to systems the ideas of lower and upper solutions to the fourth-order differential equations presented in [17,18,19].
The structure of this paper is given as follows: This article will utilize the notations, definitions, and findings found in Section 2. In Section 3, we provide the idea of a solution-tube to get existence results for fourth-order differential equation systems. We then go on to demonstrate the practicality of our results through two examples.
In this section, we recall some notations, definitions, and results that we will use in this article. The scalar product and the Euclidian norm in Rn are denoted by ⟨,⟩ and ‖⋅‖, respectively. Also, let Ck(J,Rn) be the Banach space of the k-times continuously differentiable functions ϖ associated with the norm
‖ϖ‖k=max{‖ϖ‖0,‖ϖ′‖0,...,‖ϖ(k)‖0}, |
where
‖ϖ‖0=max{ϖ(t):t∈J}. |
The space of integral functions is denoted by L1(J,Rn), with the usual norm ‖⋅‖L1. The Sobolev space of functions in Ck−1(J,Rn), where k≥1 and the (k−1)th derivative is denoted by Wk,1(J,Rn).
For ϖ0,ϖ1∈Rn, we have the following:
Cϖ0(J,Rn):={ϖ∈C(J,Rn):ϖ(0)=ϖ0}, |
C1ϖ0,ϖ1(J,Rn):={ϖ∈C1(J,Rn):ϖ(0)=ϖ0,ϖ′(0)=ϖ1}, |
CkB(J,Rn)={ϖ∈Ck(J,Rn):ϖ∈(BC)}, |
Wk,1B(J,Rn))={ϖ∈Wk,1(J,Rn)):ϖ∈(BC)}, |
Ck+1ϖ0,B(J,Rn)={ϖ∈Ck+1(J,Rn):ϖ(0)=ϖ0,ϖ(k)∈(BC)}, |
Wk+1,1ϖ0,B(J,Rn))={ϖ∈Wk+1,1(J,Rn)):ϖ(0)=ϖ0,ϖ(k)∈(BC)}, |
Ck+2ϖ0,ϖ1,B(J,Rn)={ϖ∈Ck+2(J,Rn):ϖ(0)=ϖ0,ϖ′(0)=ϖ1,x(k)∈(BC)}, |
Wk+2,1ϖ0,ϖ1,B(J,Rn))={ϖ∈Wk+2,1(J,Rn)):ϖ(0)=ϖ0,ϖ′(0)=ϖ1,ϖ(k)∈(BC)}. |
Definition 2.1. A function f:J×R4n→Rn is called an L1-Carathéodory function if
(i) For every (ϖ,y,q,p)∈R4n, the function t↦f(t,ϖ,y,q,p) is measurable;
(ii) The function (ϖ,y,q,p)↦f(t,ϖ,y,q,p) is continuous for a.e. t∈J;
(iii) For every r>0, there exists a function hr∈L1(J,[0,∞)) such that ‖f(t,ϖ,y,q,p)‖≤hr(t) for a.e. t∈J and for all (ϖ,y,q,p)∈D, where
D={(ϖ,y,q,p)∈R4n:‖ϖ‖≤r, ‖y‖≤r, ‖q‖≤r, ‖p‖≤r}. |
Definition 2.2. A function F:C3(J,Rn)×J→L1(J,Rn) is integrally bounded, if for every bounded subset B⊂C3(J,Rn), there exists an integral function hB∈L1(J,[0,∞)) so that ‖F(ϖ,α)(t)‖≤hB(t), for ∀t∈J,(ϖ,α)∈B×J.
The operator NF:C3(J,Rn)×J→C0(J,Rn) will be associated with F and defined by
NF(ϖ)(t)=∫t0F(ϖ,α)(s)ds. |
We now state the following results:
Theorem 2.1. [20] Let F:C3(J,Rn)×J→L1(J,Rn) be continuous and integrally bounded, then NF is continuous and completely continuous.
Lemma 2.1. [21] Let E be a Banach space. Let v:J→E be an absolutely continuous function, then for
{t∈J:v(t)=0andv′(t)≠0}, |
the measure is zero.
Lemma 2.2. [22] For w∈W2,1(J;R) and ε≥0, assume that one of the next properties is satisfied:
(i) w″(t)−εw(t)≥0; for almost every t∈J,κ0w(0)−ν0w′(0)≤0,κ1w(1)+ν1w′(1)≤0; where κi,νi≥0,max{κi,νi}>0;i=0,1;andmax{κ0,κ1,ε}>0,
(ii) w″(t)−εw(t)≥0; for almost every t∈J,ε>0,w(0)=w(1), w′(1)−w′(0)≤0,
(iii) w″(t)−εw(t)≥0; for almost every t∈[0,t1]∪[t2,1],ε>0,w(0)=w(1), w′(1)−w′(0)≤0, w(t)≤0,t∈[t1,t2].
Then w(t)≤0, ∀t∈[0,1].
Lemma 2.3. [22] Let f∈C(J×R2n,Rn) be a L1-Carathéodory function (see definition in [22]). Consider the following problem:
{ϖ″(t)=f(t,ϖ(t),ϖ′(t)),a.e. t∈J,ϖ∈(BC). | (2.1) |
Let ε>0, and (z,N) a solution-tube of (2.1) given in Definition 2.3 of [22]. If ϖ∈W2,1B(J,Rn) satisfies
Π(t)=⟨ϖ(t)−z(t),ϖ″(t)−z″(t)⟩+‖ϖ′(t)−z′(t)‖2‖ϖ(t)−z(t)‖−⟨ϖ(t)−z(t),ϖ′(t)−z′(t)⟩2‖ϖ(t)−z(t)‖3−ε‖ϖ(t)−z(t)‖≥N″(t)−εN(t), |
a.e. on
{t∈J:‖ϖ(t)−z(t)‖>N(t)}. |
Then
‖ϖ(t)−z(t)‖≤N(t) for every t∈J. |
Now, we recall some properties of the Leray Schauder degree. The interested reader can see [23,24].
Theorem 2.2. Let E be a Banach space and U⊂E is an open bounded set. We define K∂U(¯U,E)={f:¯U→E, where f is compact and f(ϖ)≠ϖ, for every ϖ∈∂U}, the Leary-Schauder degree on U of (Id−f) is an integer deg(Id−f,U,0) satisfying the following properties:
(i) (Existence) If deg(Id−f,U,0)≠0, then ∃ϖ∈U, s.t.,
ϖ−f(ϖ)=0. |
(ii) (Normalization) If 0∈U, then deg(Id,U,0)=1.
(iii) (Homotopy invariance) If h:¯U×J→E is a compact such that ϖ−h(ϖ,α)≠0 for each (ϖ,α)∈∂U×J, then
deg(Id−h(.,α),U,0)=deg(Id−h(.,0),U,0), for every α∈J. |
(iv) (Excision) If V⊂U is open and ϖ−f(ϖ)≠0 for all ϖ∈¯U ∖V, then
deg(Id−f,U,0)=deg(Id−f,V,0). |
(v) (Additivity) If U1,U2⊂U are disjoint and open, such that ¯U=¯U1∪U2 and ϖ−f(ϖ)≠0 for all ϖ∈∂U1∪∂U2, then
deg(Id−f,U,0)=deg(Id−f,U1,0)+deg(Id−f,U2,0). |
In this section, we define the solution-tube to the problem (1.1). This definition is important for our discussion about the existence results. A solution to this problem is a function ϖ∈W4,1(J,Rn) satisfying (1.1). Now, we define the tube solution of problem (1.1), where the functions z∈W4,1(J,Rn) and N∈W4,1(J,[0,∞) are chosen before studying the existence of this problem.
Definition 3.1. Let (z,N)∈W4,1(J,Rn)×W4,1(J,[0,∞)). The couple (z,N) is solution-tube of (1.1), if
(i) N″(t)≥0,∀t∈J.
(ii) For almost every t∈J and for all (ϖ,y,q,p)∈F,
⟨q−z″(t),f(,t,ϖ,y,q)−z‴(t)⟩+‖p−z‴(t)‖2≥N″(t)N4(t)+(N‴(t))2, |
where
F={(ϖ,y,q,p)∈R4n:‖ϖ−z(t)‖≤N(t),‖y−z′(t)‖≤N′(t),‖q−z″(t)‖=N″(t),⟨q−z″(t),p−z‴(t)⟩=N″(t)N‴(t)}. |
(iii) z(4)(t)=f(t,ϖ,y,z″(t),z‴(t)),a.e.t∈[0,1] such that N″(t)=0 and (ϖ,y)∈R2n, such that ‖ϖ−z(t)‖≤N(t) and ‖y−z′(t)‖≤N′(t).
(iv) With (1.2), we have
‖r0−(A0z″(0)−β0z‴(0))‖≤κ0N″(0)−β0N‴(0), |
‖r1−(A1z″(1)+β1z‴(1))‖≤κ1N″(1)+β1N‴(1). |
If (BC) is given by (1.3), then
z″(0)=z″(1), N‴(0)=N″(1),‖z‴(1)−z‴(0)‖≤N‴(1)−N‴(0). |
(v) ‖ϖ0−z(0)‖≤N(0), ‖ϖ1−z′(0)‖≤N′(0).
The next notation will be used
T(z,N)={ϖ∈C2(J, Rn):‖ϖ″(t)−z″(t)‖≤N″(t),‖ϖ′(t)−z′(t)‖≤N′(t)and‖ϖ‴(t)−z‴(t)‖≤N″(t)forallt∈J}. |
The next hypotheses will be used:
(F1) f:J×R4n→Rn is a L1-Carathéodory function.
(H1) There exists (z,N)∈W4,1(J,Rn)×W4,1(J,[0,∞)) a solution-tube of the main system (1.1).
The next family of problems should be considered to prove the general existence theorem that will be presented:
{ϖ(4)(t)−εϖ″(t)=fεα(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t)),a.e. t∈J,ϖ(0)=ϖ0, ϖ′(0)=ϖ1andϖ″∈(BC), | (3.1) |
where ε,α∈J and fεα:J×R4n→Rn is defined by
fεα(t,ϖ,y,q,p)={α(N″(t)‖q−z″(t)‖f1(t,ϖ,y,˜q,˘p)−ε˜q)−ε(1−α)z″(t)+(1−αN″(t)‖q−z″(t)‖)(z(4)(t)+N(4)(t)‖q−z″(t)‖(q−z″(t))),if‖q−z″(t)‖>N″(t),α(f1(t,ϖ,y,q,p)−εq)−ε(1−α)z″(t)+(1−α)(z(4)(t)+N(4)(t)N″(t)(q−z″(t))),otherwise, |
where (z,N) is the solution-tube of (1.1),
f1(t,ϖ,y,q,p)={f(t,ˉϖ,ˆy,q,p), if ‖ϖ−z(t)‖>N(t) and ‖y−z′(t)‖>N′(t),f(t,ϖ,y,q,p), otherwise, |
ˉϖ(t)=N(t)‖ϖ−z(t)‖(ϖ−z(t))+z(t), | (3.2) |
ˆy(t)=N′(t)‖y−z′(t)‖(y−z′(t))+z′(t), | (3.3) |
˜q(t)=N″(t)‖q−z″(t)‖(q−z″(t))+z″(t), | (3.4) |
˘p(t)=p+(N‴(t)−⟨q−z″(t),p−z‴(t)⟩‖q−z″(t)‖)(q−z″(t)‖q−z″(t)‖), | (3.5) |
and where we mean
N(4)(t)N″(t)(q−z″(t))=0on{t∈J:‖q(t)−z″(t)‖=N″(t)=0}. |
We associate with fεα the operator Fε:C3(J,Rn)×J→L1(J,Rn) defined by
Fε(ϖ,α)(t)=fεα(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t)). |
Similarly to the Lemma 3.3 and Propositions 3.4 in [20] and results in [25], we need the following auxiliary results:
Lemma 3.1. Assume (H1). If a function ϖ∈W4,1ϖ0,ϖ1,B(J,Rn) satisfies
⟨ϖ″(t)−z″(4)(t)⟩+‖ϖ‴(t)−z‴(t)‖2‖ϖ″(t)−z′′(t)‖−⟨ϖ′′(t)−z′′(t),ϖ′′′(t)−z′′′(t)⟩2‖ϖ′′(t)−z′′(t)‖3−ε‖ϖ′′(t)−z′′(t)‖≥N(4)(t)−εN′′(t), |
for a.e. t∈{t∈J:‖ϖ″(t)−z″(t)‖>N″(t)}, then ϖ∈T(z,N).
Proof. By assumption
ϖ′∈W3,1ϖ1,B(J,Rn), ϖ″∈W2,1B(J,Rn), |
and thus, from applying Lemma 2.3 to ϖ″, we obtain
‖ϖ″(t)−z″(t)‖≤N″(t),∀t∈J. |
On
{t∈J:‖ϖ′(t)−z′(t)‖>N′(t), ‖ϖ′(t)−z′(t)‖′≤‖ϖ″(t)−z″(t)‖≤N″(t).} |
The function
t→‖ϖ′(t)−z′(t)‖−N′(t), |
is nonincreasing on J. Since
‖ϖ′0−z′(0)‖≤N′(0), |
we get
‖ϖ′(t)−z′(t)‖≤N′(t),∀t∈J, |
hence
‖ϖ(t)−z(t)‖′≤‖ϖ′(t)−z′(t)‖≤N′(t). |
The function
t→‖ϖ(t)−z(t)‖−N(t), |
is nonincreasing on J and since
‖ϖ(0)−z(0)‖≤N(0), |
we obtain
‖ϖ(t)−z(t)‖≤N(t),∀t∈J. |
Proposition 3.1. Assume (F1) and (H1) hold. Then the operator Fε that was defined earlier is continuous and integrally bounded.
Proof. First, we will prove that Fε is integrally bounded. If ϖ∈B, where B is a bounded set of C3(J,Rn), ∃K>0 that satisfies ‖ϖ(i)(t)‖≤K, ∀t∈J, where i=0,1,2,3. Then fεα(t,.,.,.,.) is bounded in E, it can be observed that
‖Fε(ϖ,α)(t)‖=‖fεα(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))‖≤max{‖f(t,ϖ,y,q,p)‖,(ϖ,y,q,p)∈E}+|N″(t)|+‖z″(t)‖+‖z(4)(t)‖+|N(4)(t)|, |
for all α∈J and almost every t∈J, where
E={(u,y,q,p)∈R4n:‖u‖≤‖z‖0+‖N‖0,‖y‖≤‖z′‖0+‖N′‖0,‖q‖≤‖z″‖0+‖N″‖0,‖p‖≤2‖ϖ‴‖0+‖z‴‖0+‖N‴‖0}. |
As f is L1-Carathéodory, z∈W4,1(J,Rn) and N∈W4,1(J,[0,∞)), it is easy to see that Fε is integrally bounded.
In order prove the continuity, we should firstly prove that if (ϖp,αp)→(ϖ,α) in C3(J,Rn)×J, then
fεαp(t,ϖp(t),ϖ′p(t),ϖ″p(t),ϖ‴p(t))→fεα(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))a.e. t∈J. | (3.6) |
Using the fact that f is L1-Carathéodory, and from the definition of fεα, it can be concluded that (3.6) is true a.e. on {t∈J:‖ϖ″(t)−z″(t)‖≠N″(t)}. Then, by Lemma 2.1 and Proposition 3.5 in [22], we easily show that ˘ϖ‴n(t)→ϖ‴(t) on
{t∈J:‖ϖ″(t)−z″(t)‖=N″(t)>0}, |
where ˘ϖ‴n(t), is defined as (3.5). Then, (3.6) is satisfied on
{t∈J:‖ϖ″(t)−z″(t)‖=N″(t)>0}. |
For
A={t∈J:‖ϖ″(t)−z″(t)‖=N″(t)=0}, |
where ϖ″(t)=z″(t), and by Lemma 2.1, it is not hard to see that ϖ‴(t)=z‴(t), N‴(t)=0 and N(4)(t)=0, ∀t∈A, which means,
fεα(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))=α(f1(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))−εϖ″(t))+(1−α)(z(4)(t)−εz″(t))=αf1(t,ϖ(t),ϖ′(t),z″(t),z‴(t))−εz″(t)+(1−α)z(4)(t), |
a.e. on A. By the solution tube hypothesis (Definition 3.1 condition (iii)), we have
fεα(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))=αz(4)(t)+(1−α)z(4)(t)−εz″(t)=z(4)(t)−εz″(t), |
a.e. on A. Consequently, (3.6) must be true a.e. on J. Using the Lebesgue-dominated convergence theorem, and since Fε is integrally bounded, the proof can be concluded.
Now, we can obtain our general existence result. We follow the method of proof given in [20].
Theorem 3.1. Assume (F1), (H1), and the following conditions are satisfied:
(Hk) For every solution ϖ of the related system (3.1), ∃K>0, so that
‖ϖ‴(t)‖<K,∀t∈J. |
Then, problem (1.1) has a solution ϖ∈W4,1(J,Rn)∩T(z,N).
Proof. We first show that if (ϖ,N)∈W4,1ϖ0,ϖ1,B(J,Rn)×W4,1(J,[0,∞)) is a solution of (3.1), then
‖ϖ″(t)−z″(t)‖≤N″(t),∀t∈J. |
For the set
{t∈J:‖ϖ″(t)−z″(t)‖>N″(t)}. |
By the definition of ~ϖ″ and ˘ϖ‴(t) (as (3.4) and (3.5)), we have
‖~ϖ″(t)−z″(t)‖=N″(t), | (3.7) |
<~ϖ″(t)−z″(t),˘ϖ‴(t)−z‴(t)>=N″(t)N‴(t). |
Also
‖˘ϖ‴(t)−z‴(t)‖2=‖ϖ‴(t)−z‴(t)‖2+(N‴(t))2−⟨ϖ″(t)−z″(t),ϖ‴(t)−z‴(t)⟩2‖ϖ″(t)−z″(t)‖2. |
Then, by (H1), we obtain
⟨ϖ″(t)−z″(t)−z(4)(t)⟩+‖ϖ‴(t)−z‴(t)‖2‖ϖ″(t)−z″(t)‖−⟨ϖ″(t)−z″(t),ϖ‴(t)−z‴(t)⟩2‖ϖ″(t)−z″(t)‖3−ε‖ϖ″(t)−z″(t)‖=⟨ϖ″(t)−z″(t),fεα(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))+εϖ″(t)⟩‖ϖ″(t)−z″(t)‖+1‖ϖ″(t)−z″(t)‖(‖ϖ‴(t)−z‴(t)‖2−⟨ϖ″(t)−z″(t),ϖ‴(t)−z‴(t)⟩2‖ϖ″(t)−z″(t)‖2)−ε‖ϖ″(t)−z″(t)‖=⟨ϖ″(t)−z″(t),αN″(t)‖ϖ″(t)−z″(t)‖(f1(t,ϖ(t),ϖ′(t),~ϖ″(t),˘ϖ‴(t))−z(4)(t))⟩‖ϖ″(t)−z″(t)‖+⟨ϖ″(t)−z″(t),(1−αN″(t)‖ϖ″(t)−z″(t)‖)N(4)(t)(ϖ″(t)−z″(t))‖ϖ″(t)−z″(t)‖⟩‖ϖ″(t)−z″(t)‖−ε⟨ϖ″(t)−z″(t),α(~ϖ″(t)−z″(t))−(ϖ″(t)−z″(t))⟩‖ϖ″(t)−z″(t)‖+‖˘ϖ‴(t)−z‴(t)‖2−(N‴(t))2‖ϖ″(t)−z″(t)‖−ε‖ϖ″(t)−z″(t)‖=α‖ϖ″(t)−z″(t)‖⟨~ϖ″(t)−z″(t),f1(t,ϖ(t),ϖ′(t),~ϖ″(t),˘ϖ‴(t))−z(4)(t)⟩+N(4)(t)(1−αN″(t)‖ϖ″(t)−z″(t)‖)−ε‖ϖ″(t)−z″(t)‖+ε‖ϖ″(t)−z″(t)‖−ε⟨ϖ″(t)−z″(t),α(~ϖ″(t)−z″(t))⟩‖ϖ″(t)−z″(t)‖+‖˘ϖ‴(t)−z‴(t)‖2−(N‴(t))2‖ϖ″(t)−z″(t)‖≥α‖ϖ″(t)−z″(t)‖(N″(t)+(N‴(t))2−‖˘ϖ‴(t)−z‴(t)‖2)+N(4)(t)−αN″(t)‖ϖ″(t)−z″(t)‖−αεN″(t)+‖˘ϖ‴(t)−z″(t)‖2−(N‴(t))2‖ϖ″(t)−z″(t)‖=N(4)(t)−αεN″(t)+(1−α)(‖˘ϖ‴(t)−z‴(t)‖2−(N‴(t))2)‖ϖ″(t)−z″(t)‖≥N(4)(t)−εN″(t), |
on
{t∈J:‖ϖ″(t)−z″(t)‖>N″(t)}. |
Using Lemma 3.1, it can be observed that any solutions to system (3.1) are in T(z,N) and then, in U, where
U={ϖ∈C3(J,Rn):‖u(i)‖0≤‖z(i)‖0+‖N(i)‖0+1,i=1,0,2;‖ϖ‴‖0≤K}. |
Fix ε∈J such that the operator Lε:C1B(J,Rn)→C0(J,Rn) given by
Lε(ϖ)(t)=ϖ′(t)−ϖ′(0)−ε∫t0ϖ(s)ds |
is invertible.
Consider the linear operator D:C3ϖ0,ϖ1,B(J,Rn)→C1B(J,Rn) defined by
D(ϖ)=ϖ″. |
It can be easily confirmed that D is invertible.
A solution to (1.1) is a fixed point of the operator
K=D−1oL−1εoNFε:C3(J,Rn)×J→C3ϖ0,ϖ1,B(J,Rn)⊂C3(J,Rn). |
Using Proposition 3.1 and Theorem 2.1, and since the operators D and Lε are continuous, it can be concluded that K is completely continuous and fixed point free on ∂U. Let
K0:C3(J,Rn)×J→C3(J, Rn) |
by K0(ϖ,α)=αK(ϖ,0). Because Fε(.,0) is integrally bounded, there exists an open bounded set K⊂C3(J,Rn), where
U⊂K and K0(C3(J,Rn)×J)⊂K, |
it can be implied from the homotopic and the excision properties of the Leray-Schauder theorem that
1=deg(Id,K,0)=°(Id−K0(.,1),K,0)=deg(Id−K(.,0),K,0)=°(Id−K(.,0),U,0)=deg(Id−K(.,1),U,0). |
As a result, there exists a solution ϖ∈T(z,N) for α=1 to (3.1), which also can solve (1.1) by definition of fε1. The proof is complete.
Now, following from our general existence theorem (Theorem 3.1), other existence results will be presented. We will consider the following assumptions:
(H2) There exist a function γ∈L1(J,[0,∞)) and a Borel measurable function Ψ∈C([0,∞),[1,∞)) s.t.
(ⅰ) ‖f(t,ϖ,y,q,p)‖≤γ(t)Ψ(‖p‖),∀t∈J and ∀(ϖ,y,q,p)∈R4n, where ‖ϖ−z(t)‖≤N(t), ‖y−z′(t)‖≤N′(t) and ‖q−z″(t)‖≤N″(t),
(ⅱ) ∀c≥0, we have
∫∞cdτΨ(τ)=∞. |
(H3) There exist, a function γ∈L1(J,[0,∞)) and a Borel measurable function Ψ∈C([0,∞],]0,∞)) s.t.
(ⅰ) ‖⟨p,f(t,ϖ,y,q,p)⟩‖≤Ψ(‖p‖)(γ(t)+‖p‖),∀t∈J and ∀(ϖ,y,q,p)∈R4n, where ‖ϖ−z(t)‖≤N(t), ‖y−z′(t)‖≤N′(t) and ‖q−z″(t)‖≤N″(t),
(ⅱ) ∀c≥0, we have
∫∞cτdτΨ(τ)+τ=∞. |
(H4) ∃r,b>0, c≥0 and a function h∈L1(J,R) s.t. ∀t∈J, ∀(ϖ,y,q,p)∈R4n, where
‖ϖ−z(t)‖≤N(t),‖y−z′(t)‖≤N′(t),‖q−z″(t)‖≤N″(t), |
and ‖p‖≥r, then
(b+c‖q‖)σ(t,ϖ,y,q,p)≥‖p‖−h(t), |
where
σ(t,ϖ,y,q,p)=⟨q,f(t,ϖ,y,q,p)⟩+‖p‖2‖p‖−⟨p,f(t,ϖ,y,q,p)⟩⟨q,p⟩‖p‖3. |
(H5) ∃a≥0 and l∈L1(J,R) s.t.
‖f(t,ϖ,y,q,p)‖≤a(⟨q,f(t,ϖ,y,q,p)⟩+‖p‖2)+l(t), |
∀t∈J and ∀(ϖ,y,q,p)∈R4n, where
‖ϖ−z(t)‖≤N(t),‖y−z′(t)‖≤N′(t), |
and
‖q−z″(t)‖≤N″(t). |
Theorem 3.2. Assume (F1), (H1), and (\mathcal{H}2) aresatisfied.If (BC) isgivenby (1.2) with \max \left\{ \beta _{0}, \beta _{1}\right\} > $$ 0 $, then system $ (1.1) $ has at least one solution $ \varpi \in T(z, N)\cap W^{4,1}( \mathcal{J}, \mathbb{\ R}^{n}) $.
Proof. Theorem 3.1 will guarantee the existence of a solution if we can obtain a priori bound on the third derivative of any solution ϖ to (3.1). It is known that ϖ∈T(z,N) from the Theorem 3.1 proof. Therefore, since (BC) is given by (1.2) with max{β0,β1}>0, ∃k>0, s.t.
min{‖ϖ‴(0)‖,‖ϖ‴(1)‖}≤k. |
Now, let R>k such that
∫RkdsΨ(s)>L=‖γ‖L1+ε‖N″‖0+‖z(4)‖L1+‖N(4)‖L1. |
Suppose there exists t1∈[0,1] s.t. ‖ϖ‴(t1)‖≥R. Then, there exists t0≠t1∈[0,1] such that ‖ϖ‴(t0)‖=k and ‖ϖ‴(t)‖≥k, ∀t∈[t0,t1]. Let us assume that t0<t1. Thus, by (H2), almost everywhere on [t0,t1], we have
‖ϖ‴(t)‖′=⟨ϖ‴(t)⟩‖ϖ‴(t)‖≤‖ϖ(4)(t)‖≤‖f(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))‖+ε‖ϖ″(t)−z″(t)‖+‖z(4)(t)‖+|N(4)(t)|≤‖γ(t)‖Ψ(‖ϖ‴(t)‖)+ε‖N″(t)‖0+‖z(4)(t)‖+‖N(4)‖L1. |
So,
∫t1t0‖ϖ‴(t)‖′tΨ(‖ϖ‴(t)‖)dt≤L. |
Then, we have
∫t1t0‖ϖ‴(t)‖′tΨ(‖ϖ‴(t)‖)dt=∫‖ϖ‴(t1)‖‖ϖ‴(t0)‖dsΨ(s)≥∫RkdsΨ(s)>L, |
which contradict the assumptions. So, for any solution ϖ of (3.1), ∃R>0 s.t. ‖ϖ‴(t)‖<R, ∀t∈J.
If (H2) is replaced by (H3), extra assumptions are needed.
Theorem 3.3. Assume (F1), (H1), (H3), and (\mathcal{H}4) or (\mathcal{H}5) aresatisfied.Then,thereexistsasolution \varpi \in T(z, N)\cap W^{4,1}(\mathcal{J}, \mathbb{\ R}^{n}) to (1.1) $.
For this end, we need the next three Lemmas.
Lemma 3.2. [20] Let r,k≥0, N∈L1([0,1],R) and Ψ∈C([0,∞[,]0,∞[) be a Borel measurable function s.t.
∫∞rτdτΨ(τ)>‖N‖L1([0,1],R)+k. |
Then ∃K>0, s.t. ‖ϖ′‖0<K, ∀ϖ∈W2,1([0,1],Rn) satisfy:
(i) mint∈[0,1]‖ϖ′(t)‖≤r;
(ii) ‖ϖ′‖L1([t0,t1],R)≤k for every interval [t0,t1]⊂{t∈[0,1]:‖ϖ′(t)‖≥r};
(iii) |⟨ϖ′(t),ϖ″(t)⟩|≤Ψ(‖ϖ′(t)‖)(N(t)+‖ϖ′(t)‖) a.e. on
{t∈[0,1]:‖ϖ′(t)‖≥r}. |
Lemma 3.3. [20] Let r,ν>0, γ≥0 and N∈L1([0,1],R). Then there exists a nondecreasing function ω∈C[0,∞[,[0,∞[) s.t.
‖ϖ′‖L1([t0,t1],R)≤ω(‖ϖ‖0), |
and
mint∈[0,1]‖ϖ′(t)‖≤max{r,;ω(‖ϖ‖0)}. |
∀u∈W2,1([0,1],Rn) and
{t∈[t0,t1]:‖ϖ′(t)‖≥r}, |
the following inequality
(ν+γ‖ϖ(t)‖)σ0(t,ϖ)+γ⟨ϖ(t),ϖ′(t)⟩2‖ϖ(t)‖ϖ′(t)‖≥‖ϖ′(t)‖−N(t) |
is satisfied, where
σ0(t,ϖ)=⟨ϖ(t),ϖ″(t)⟩+‖ϖ′2‖ϖ′(t)‖−⟨ϖ′(t),ϖ″(t)⟩⟨ϖ(t),ϖ′(t)⟩‖ϖ′(t)‖3. |
Lemma 3.4. [20] Let K>0, and N∈L1([0,1],R). Then there exists an increasing function ω∈C([0,∞[,]0,∞[) s.t. ‖ϖ′‖L1([0,1],R)≤ω(‖ϖ‖0) for all ϖ∈W2,1([0,1],Rn) that satisfies
‖ϖ″(t)‖≤k(⟨ϖ(t),ϖ″(t)⟩+‖ϖ′(t)‖2)+N(t), |
for almost every t∈[0,1].
Proof of Theorem 3.3. Similarly to the previous proof, we need Theorem 3.1 to prove that the third derivative of all solutions ϖ to (3.1) is bounded. Let ϖ be a solution to (3.1), where ϖ∈T(z,N) from Theorem 3.1 proof. We obtain from (H3),
|⟨ϖ‴(t),ϖ(4)(t)⟩|≤|⟨ϖ‴(t),f(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))⟩|+(ε‖ϖ″(t)−z″(t)‖+‖z(4)(t)‖+|N(4)(t)|)‖ϖ‴(t)‖≤(γ(t)+‖ϖ‴(t)‖)Ψ(‖ϖ‴(t)‖)+(ε|N″(t)|+‖z(4)(t)‖+|N(4)(t)|)‖ϖ‴(t)‖≤(Ψ(‖ϖ‴(t)‖)+‖ϖ‴(t)‖)+(γ(t)+‖ϖ‴(t)‖+ε|N″(t)|+‖z(4)(t)‖+|N(4)(t)|), |
for almost every t∈[0,1]. Thus, condition (iii) of Lemma 3.2 is verified, where
ψ(τ)=Ψ(τ)+τandN(τ)=γ(τ)+ε|N″(τ)|+‖z(4)(τ)‖+|N(4)(t)|. |
Therefore, it is enough to prove that conditions (i) and (ii) are verified. (H4) guarantees that a.e. on
{t∈[0,1]:‖ϖ‴(t)‖≥r}, |
we have
σ0(t,ϖ″)=⟨ϖ″(t)⟩+‖ϖ‴(t)‖2‖ϖ‴(t)‖−⟨ϖ‴(t)⟩⟨ϖ″(t),ϖ‴(t)⟩‖ϖ‴(t)‖3=ασ(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))+(1−α)‖ϖ‴(t)‖ +(1−α)⟨ϖ″(t)+(ε+N(4)(t)N″(t))(ϖ″(t)−z″(t))⟩‖ϖ‴(t)‖ −(1−α)⟨ϖ‴(t)+(ε+N(4)(t)N″(t))(ϖ″(t)−z″(t))⟩⟨ϖ″(t),ϖ‴(t)⟩‖ϖ‴(t)‖3≥ασ(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))+(1−α)‖ϖ‴(t)‖ −2(‖z″(t)‖+|N″(t)|)(‖z(4)(t)‖+ε|N″(t)|+|N(4)(t)|)r. |
Thus, we have
(b+c‖ϖ″(t)‖)σ0(t,ϖ″)+c⟨ϖ″(t),ϖ‴(t)⟩2‖ϖ″(t)‖‖ϖ‴(t)‖≥α‖ϖ‴(t)‖+b(1−α)‖ϖ‴(t)‖−h(t)−δ0,(t), |
where
δ0(t)=2r(b+c‖z″(t)‖+c|N″(t)|)(‖z″(t)‖+|N″(t)|)(‖z(4)(t)‖+ε|N″(t)|+|N(4)(t)|). |
If we take
z=minα∈[0,1]{α+b(1−α)},ν=bzandθ=cη, |
we can apply Lemma 3.3 to ϖ″([0,1],Rn). Thus, conditions of Lemma 3.2 are verified. Moreover, if (H5) holds, we have
‖ϖ(4)(t)‖≤α‖f(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))‖+ε‖ϖ″(t)−z″(t)‖+‖z(4)(t)‖+|N(4)(t)|≤αa(⟨ϖ″(t),f(t,ϖ(t),ϖ′(t),ϖ″(t),ϖ‴(t))⟩+‖ϖ‴(t)‖2)+l(t) +ε|N″(t)|+‖z(4)(t)‖+|N(4)(t)|≤a(⟨ϖ″(t),ϖ(4)(t)⟩+‖ϖ‴(t)‖2)+ε|N″(t)|+‖z(4)(t)‖+|N(4)(t)| −a(1−α)⟨ϖ″(t),z(4)(t)+(N(4)(t)N″(t)+ε)(ϖ″(t)−z″(t))⟩≤a(⟨ϖ″(t),ϖ(4)(t)⟩+‖ϖ‴(t)‖2)+ε|N″(t)|+‖z(4)(t)‖+|N(4)(t)| +a(‖z″(t)‖+|N″(t)|)(‖z(4)(t)‖+|N(4)(t)|+εN″(t)). |
Therefore, if Lemma 3.4 is applied to ϖ″([0,1],Rn), all conditions of Lemma 3.2 are satisfied. As a result, for all solutions ϖ of (3.1), ‖ϖ‴‖0<K for some constant K>0.
From the previous results, we obtain the following consequence:
Corollary 3.1. Assume (F1), (H1), (H2), and (H4) or (H5) are satisfied. Then, we have a solution ϖ∈T(z,N)∩W4,1([0,1],Rn) to the system (1.1).
Remark 3.1. Definition 3.1 is associated to the definitions of lower and upper solutions to the fourth-order differential equation. These definitions are used in [17], and introduce them for problems (1.1) and (1.2).
Definition 3.2. Let n=1 and ϖ0=ϖ1=0.
A function κ∈C4(]0,1[)∩C3(J) is called a lower solution to (1.1) and (1.2), if
(i) κ(4)(t)≥f(t,κ(t),κ′(t),κ″(t),κ‴(t)) for every t∈J;
(ii) κ(0)=κ′(0)=0;
(iii) A0κ″(0)−β0κ‴(0)≤r0 and A1κ″(1)+β1κ‴(1)≤r1.
On the other hand, an upper solution to (1.1) and (1.2) is a function ν∈C4(]0,1[)∩C3(J) that satisfies (i)–(iii) with reversed inequalities.
Similarly to Remark 3.2 in [20], we consider the following assumptions:
(A) There exist lower and upper solutions, κ and ν, respectively, to (1.1) and (1.2), where κ≤ν .
(B) There exists a solution-tube (z,N) to (1.1) and (1.2).
(C) There exist lower and upper solutions, κ≤ν, to (1.1) and (1.2) s.t.
(i) κ″(t)≤ν″(t)) for all t∈J;
(ii) f(t,ν(t),ν′(t),q,p)≤f(t,ϖ,y,q,p)≤f(t,κ(t),κ′(t),q,p); ∀t∈J and (ϖ,y,q,p)∈R4n such that κ(t)≤ϖ≤ν(t) and κ′(t)≤ϖ′(t)≤ν′(t).
It can be easily checked that
● If (B) holds with z and N of class C4, and z(0)=N(0)=0, then (A) holds.
Indeed, κ=z−N and ν=z+N are respectively lower and upper solutions of (1.1) and (1.2). However, (A) does not imply (B).
Noting that (B) is more general than (C), see [17]; i.e.,
● If (C) is verified, then (B) is verified.
Taking z=(κ+ν)/2 and N=(ν−κ)/2. But, (B) does not imply (C) (ii) and κ(0)=ν(0)=0.
Next, we present two examples to illustrate the applicability of Theorem 3.3.
Example 3.1. Consider the following system:
{ϖ(4)(t)=ϖ‴(t)+‖ϖ‴(t)‖(‖ϖ″(t)‖2ϖ′(t)−⟨ϖ′(t),ϖ″(t)⟩ϖ″(t))−ξ,a.e. t∈J,ϖ(0)=0,ϖ′(0)=0,A0ϖ″(0)=0,A1ϖ″(1)+βt=1ϖ‴(1)=0, | (3.8) |
here ξ∈Rn,‖ξ‖=1, and Aiandβi are given before for i=0,1. Show that when z≡0,N(t)=t36, (z,N) is a solution-tube of (3.8). We have (H3) and (H4) are verified for
Ψ(τ)=3τ+1,γ(t)=0,b=1,c=0,r>0,h(τ)=2τr+τ5. |
Owing to the Theorem 3.3, the problem (3.8) has at least one solution ϖ s.t.
‖ϖ(t)‖≤t36, ‖ϖ(t)′‖≤t22 and ‖ϖ(t)′′‖≤t for all t∈J. |
Example 3.2. Consider the following system:
{ϖ(4)(t)=ϖ″(t)(‖ϖ‴(t)‖2+1)+φ(t),a.e. t∈J,ϖ(0)=0,ϖ′(0)=0,ϖ″(0)=ϖ″(1),ϖ‴(0)=ϖ‴(1), | (3.9) |
where φ∈L∞(J,Rn) with ‖φ‖L∞≤1. Show that for z≡0,N(t)=t22, (z,N) is a solution-tube of (3.9). We have (H3) and (H5) are verified when
Ψ(τ)=τ2+2, γ(t)=0,a=1,l(t)=3. |
By Theorem 3.3, the problem (3.9) has at least one solution ϖ s.t.
‖ϖ(t)‖≤t22, ‖ϖ′(t)‖≤t, ‖ϖ′′(t)‖≤1,∀t∈J. |
Our paper discusses the existence of solutions for fourth-order differential equation systems, focusing particularly on cases involving L1-Carathéodory functions on the right-hand side of the equations. We first, introduced the concept of a solution-tube, which is an innovative approach that extends the concepts of upper and lower solutions applicable to fourth-order equations into the domain of systems. It outlines the mathematical framework necessary to demonstrate that solutions exist for these types of differential equation systems under specified boundary conditions (such as Sturm-Liouville and periodic conditions). The paper stands on prior results regarding higher-order differential equations, providing a fresh perspective and methodology that can be used to explore further developments in the field. In addition to presenting the theoretical underpinnings, we also illustrated the practicality of our results with examples, contributing to the mathematical discourse on differential equations and our solutions, which ultimately serves as a scholarly contribution to understanding the dynamics of fourth-order systems and the existence of their solutions; please see [26,27].
Bouharket Bendouma: Conceptualization, formal analysis, Writing-original draft preparation; Fatima Zohra Ladrani and Keltoum Bouhali: investigation, Methodology; Ahmed Hammoudi and Loay Alkhalifa: Writing-review and editing. All authors have read and approved the final version of the manuscript for publication.
The researchers would like to thank the Deanship of Graduate Studies and Scientific Research at Qassim University for financial support (QU-APC-2024-9/1).
The authors declare that there is no conflict of interest.
[1] |
G. N. Kouziokas, A new W-SVM kernel combining PSO-neural network transformed vector and Bayesian optimized SVM in GDP forecasting, Eng. Appl. Artif. Intel., 92 (2020), 103650. https://doi.org/10.1016/j.engappai.2020.103650 doi: 10.1016/j.engappai.2020.103650
![]() |
[2] |
O. Cepni, I. E. Güney, N. R. Swanson, Nowcasting and forecasting GDP in emerging markets using global financial and macroeconomic diffusion indexes, Int. J. Forecasting, 35 (2019), 555–572. https://doi.org/10.1016/j.ijforecast.2018.10.008 doi: 10.1016/j.ijforecast.2018.10.008
![]() |
[3] |
T. E. Simos, V. N. Katsikis, S. D. Mourtas, P. S. Stanimirović, Unique non-negative definite solution of the time-varying algebraic Riccati equations with applications to stabilization of LTV systems, Math. Comput. Simulat., 202 (2022), 164–180. https://doi.org/10.1016/j.matcom.2022.05.033 doi: 10.1016/j.matcom.2022.05.033
![]() |
[4] |
S. D. Mourtas, V. N. Katsikis, C. Kasimis, Feedback control systems stabilization using a bio-inspired neural network, EAI Endorsed Trans. AI Robotics, 1 (2022), 1–13. https://doi.org/10.4108/airo.v1i.17 doi: 10.4108/airo.v1i.17
![]() |
[5] |
N. Premalatha, A. V. Arasu, Prediction of solar radiation for solar systems by using ANN models with different back propagation algorithms, J. Appl. Res. Technol., 14 (2016), 206–214. https://doi.org/10.1016/j.jart.2016.05.001 doi: 10.1016/j.jart.2016.05.001
![]() |
[6] |
S. X. Lv, L. Wang, Multivariate wind speed forecasting based on multi-objective feature selection approach and hybrid deep learning model, Energy, 263 (2023), 126100. https://doi.org/10.1016/j.energy.2022.126100 doi: 10.1016/j.energy.2022.126100
![]() |
[7] |
C. Huang, X. Jia, Z. Zhang, A modified back propagation artificial neural network model based on genetic algorithm to predict the flow behavior of 5754 aluminum alloy, Materials, 11 (2018), 855. https://doi.org/10.3390/ma11050855 doi: 10.3390/ma11050855
![]() |
[8] | S. Gayathri, A. K. Krishna, V. P. Gopi, P. Palanisamy, Automated binary and multiclass classification of diabetic retinopathy using Haralick and multiresolution features, IEEE Access, 8 (2020), 57497–57504. https://doi.org/10.1109/ACCESS.2020.2979753 |
[9] |
L. Chen, Z. Huang, Y. Li, N. Zeng, M. Liu, A. Peng, et al., Weight and structure determination neural network aided with double pseudoinversion for diagnosis of flat foot, IEEE Access, 7 (2019), 33001–33008. https://doi.org/10.1109/ACCESS.2019.2903634 doi: 10.1109/ACCESS.2019.2903634
![]() |
[10] |
M. R. Daliri, A hybrid automatic system for the diagnosis of lung cancer based on genetic algorithm and fuzzy extreme learning machines, J. Medical Syst., 36 (2012), 1001–1005. https://doi.org/10.1007/s10916-011-9806-y doi: 10.1007/s10916-011-9806-y
![]() |
[11] |
R. J. S. Raj, S. J. Shobana, I. V. Pustokhina, D. A. Pustokhin, D. Gupta, K. Shankar, Optimal feature selection-based medical image classification using deep learning model in internet of medical things, IEEE Access, 8 (2020), 58006–58017. https://doi.org/10.1109/ACCESS.2020.2981337 doi: 10.1109/ACCESS.2020.2981337
![]() |
[12] |
S. D. Mourtas, V. N. Katsikis, Exploiting the Black-Litterman framework through error-correction neural networks, Neurocomputing, 498 (2022), 43–58. https://doi.org/10.1016/j.neucom.2022.05.036 doi: 10.1016/j.neucom.2022.05.036
![]() |
[13] |
S. D. Mourtas, A weights direct determination neuronet for time-series with applications in the industrial indices of the federal reserve bank of St. Louis, J. Forecasting, 14 (2022), 1512–1524. https://doi.org/10.1002/for.2874 doi: 10.1002/for.2874
![]() |
[14] |
S. X. Lv, L. Peng, H. Hu, L. Wang, Effective machine learning model combination based on selective ensemble strategy for time series forecasting, Inf. Sci., 612 (2022), 994–1023. https://doi.org/10.1016/j.ins.2022.09.002 doi: 10.1016/j.ins.2022.09.002
![]() |
[15] |
Y. Zhang, D. Guo, Z. Luo, K. Zhai, H. Tan, CP-activated WASD neuronet approach to Asian population prediction with abundant experimental verification, Neurocomputing, 198 (2016), 48–57. https://doi.org/10.1016/j.neucom.2015.12.111 doi: 10.1016/j.neucom.2015.12.111
![]() |
[16] | Y. Zhang, Z. Xue, M. Xiao, Y. Ling, C. Ye, Ten-Quarter Projection for Spanish Central Government Debt via WASD Neuronet, In: International Conference on Neural Information Processing, Springer, 2017. 893–902. |
[17] |
F. Groes, P. Kircher, I. Manovskii, The U-shapes of occupational mobility, Rev. Econ. Stud., 82 (2015), 659–692. https://doi.org/10.1093/restud/rdu037 doi: 10.1093/restud/rdu037
![]() |
[18] |
I. N. Generalao, Measuring the telework potential of jobs: Evidence from the international standard classification of occupations, Philippine Rev. Econ., 58 (2021), 92–127. https://doi.org/10.37907/5erp1202jd doi: 10.37907/5erp1202jd
![]() |
[19] |
D. Lagios, S. D. Mourtas, P. Zervas, G. Tzimas, A weights direct determination neural network for international standard classification of occupations, Mathematics, 11 (2023), 629. https://doi.org/10.3390/math11030629 doi: 10.3390/math11030629
![]() |
[20] |
J. Garnitz, R. Lehmann, K. Wohlrabe, Forecasting GDP all over the world using leading indicators based on comprehensive survey data, Appl. Econ., 51 (2019), 5802–5816. https://doi.org/10.1080/00036846.2019.1624915 doi: 10.1080/00036846.2019.1624915
![]() |
[21] |
M. Marcellino, M. Porqueddu, F. Venditti, Short-term GDP forecasting with a mixed-frequency dynamic factor model with stochastic volatility, J. Bus. Econ. Stat., 34 (2016), 118–127. https://doi.org/10.1080/07350015.2015.1006773 doi: 10.1080/07350015.2015.1006773
![]() |
[22] |
C. Liu, W. Xie, T. Lao, Y. Yao, J. Zhang, Application of a novel grey forecasting model with time power term to predict China's GDP, Grey Syst. Theory Appl., 11 (2021), 343–357. https://doi.org/10.1108/GS-05-2020-0065 doi: 10.1108/GS-05-2020-0065
![]() |
[23] |
J. Yoon, Forecasting of real GDP growth using machine learning models: Gradient boosting and random forest approach, Comput. Econ., 57 (2021), 247–265. https://doi.org/10.1007/s10614-020-10054-w doi: 10.1007/s10614-020-10054-w
![]() |
[24] |
H. H. Kim, N. R. Swanson, Methods for backcasting, nowcasting and forecasting using factor-MIDAS: With an application to Korean GDP, J. Forecasting, 37 (2018), 281–302. https://doi.org/10.1002/for.2499 doi: 10.1002/for.2499
![]() |
[25] |
A. Richardson, T. van Florenstein Mulder, T. Vehbi, Nowcasting GDP using machine-learning algorithms: A real-time assessment, Int. J. Forecasting, 37 (2021), 941–948. https://doi.org/10.1016/j.ijforecast.2020.10.005 doi: 10.1016/j.ijforecast.2020.10.005
![]() |
[26] | Y. Zhang, D. Chen, C. Ye, Deep Neural Networks: WASD Neuronet Models, Algorithms, and Applications, CRC Press: Boca Raton, FL, USA, 2019. |
[27] |
T. E. Simos, V. N. Katsikis, S. D. Mourtas, A multi-input with multi-function activated weights and structure determination neuronet for classification problems and applications in firm fraud and loan approval, Appl. Soft Comput., 127 (2022), 109351. https://doi.org/10.1016/j.asoc.2022.109351 doi: 10.1016/j.asoc.2022.109351
![]() |
[28] |
T. E. Simos, S. D. Mourtas, V. N. Katsikis, Time-varying Black-Litterman portfolio optimization using a bio-inspired approach and neuronets, Appl. Soft Comput., 112 (2021), 107767. https://doi.org/10.1016/j.asoc.2021.107767 doi: 10.1016/j.asoc.2021.107767
![]() |
[29] | Y. Zhang, X. Yu, L. Xiao, W. Li, Z. Fan, W. Zhang, Weights and structure determination of articial neuronets, In: Self-Organization: Theories and Methods, New York, NY, USA: Nova Science, 2013. |
[30] |
G. P. Zhang, D. M. Kline, Quarterly time-series forecasting with neural networks, IEEE T. Neur. Network., 18 (2007), 1800–1814. https://doi.org/10.1109/TNN.2007.896859 doi: 10.1109/TNN.2007.896859
![]() |
[31] |
A. Tharwat, Classification assessment methods, Appl. Comput. Inf., 17 (2020), 168–192. https://doi.org/10.1016/j.aci.2018.08.003 doi: 10.1016/j.aci.2018.08.003
![]() |
1. | Peng E, Tingting Xu, Linhua Deng, Yulin Shan, Miao Wan, Weihong Zhou, Solutions of a class of higher order variable coefficient homogeneous differential equations, 2025, 20, 1556-1801, 213, 10.3934/nhm.2025011 |