key: cord-0987265-wj8rxqo1
authors: Guirao, Juan L. G.; Sabir, Zulqurnain; Raja, Muhammad Asif Zahoor; Baleanu, Dumitru
title: Design of neuro-swarming computational solver for the fractional Bagley–Torvik mathematical model
date: 2022-02-18
journal: Eur Phys J Plus
DOI: 10.1140/epjp/s13360-022-02421-3
sha: 940d568a04cc6c6499ff12869036ccf810ebaa72
doc_id: 987265
cord_uid: wj8rxqo1

This study is to introduce a novel design and implementation of a neuro-swarming computational numerical procedure for numerical treatment of the fractional Bagley–Torvik mathematical model (FBTMM). The optimization procedures based on the global search with particle swarm optimization (PSO) and local search via active-set approach (ASA), while Mayer wavelet kernel-based activation function used in neural network (MWNNs) modeling, i.e., MWNN-PSOASA, to solve the FBTMM. The efficiency of the proposed stochastic solver MWNN-GAASA is utilized to solve three different variants based on the fractional order of the FBTMM. For the meticulousness of the stochastic solver MWNN-PSOASA, the obtained and exact solutions are compared for each variant of the FBTMM with reasonable accuracy. For the reliability of the stochastic solver MWNN-PSOASA, the statistical investigations are provided based on the stability, robustness, accuracy and convergence metrics.

The fractional Bagley-Torvik mathematical model (FBTMM) has achieved the huge attention of the research community in recent years. The fractional kinds of derivatives represent the physical network dynamics, a rigid plate based on the Newtonian fluid, and the frequency-dependent systems of the damping properties [1] [2] [3] [4] . The numerical, approximate and analytical form of the FBTMM has been performed by many scientists and reported in [6] [7] [8] [9] [10] . While few other utmost deterministic and stochastic numerical schemes [11] [12] [13] [14] [15] [16] [17] [18] [19] are listed in Table 1 in terms of novel methodology exploited for the solutions, publication year, and necessary remarks to highlight their significance in the reported literature for FBTMM.

The present study is to solve the FBTMM by using a competent soft computing approach based on a Mayer wavelet neural network (MWNN) using the optimization procedures of particle swarm optimization (PSO) along with active-set algorithm (ASA), i.e., MWNN-PSOASA. The general form of the FBTMM is provided as [20] [21] [22] [23] :

where a α indicates the initial conditions, λ represents the derivative based on the fractional-order with 1.25, 1.5, and 1.75, v(τ ) is the solution of above Eq. (1), while a 1 , a 2 , and a 3 are the constant values The FBTMM represented in Eq. (1) was the pioneering work of Bagley and Torvik introducing on the motion of an absorbed plate using the Newtonian fluid [3] . a emails: juan.garcia@upct.es; jlgarcia@kau.edu.sa (corresponding author) b e-mail: zulqurnain_maths@hu.edu.pk c e-mail: rajamaz@yuntech.edu.tw d e-mail: dumitru@cankaya.edu.tr [11] in 1998 Podlubny's consecutive approximation Novel numerical solution [12] in 2002 Deterministic numerical scheme Convergence established [13] in 2007 Differential transform method Novel numerical solver [14] in 2008 Adomian decomposition method Novel analytical solution [15] in 2008 He's variational iteration method Viable analytic method [16] in 2009 Matrix approach of discretization Novel discretization [17] in 2010 Shooting collocation approach Efficient scheme [18] in 2010 Taylor collocation method Power series approach [19] in 2011 Genetic algorithms and neural networks Novel stochastic solver [20] in 2011 Neural networks and Swarm intelligence Viable stochastic solver [20] in 2012 Haar wavelets operational matrix Novel wavelets approach [22] in 2017 Sequential quadratic programing Fractional neural network [23] in 2020 Interior-point method Fluid dynamics problem [24] in 2020 Galerkin approximations Numerical scheme [25] in 2020 Exponential spline approximation Novel spline method [26] in 2020 Jacobi collocation methods Power series approach [27] in 2021 Generalized Bessel polynomial Power series method [28] in 2021 Quadratic finite element mentod Numerical computing [29] in 2021 Lie symmetry analysis method Numerical analysis

The stochastic computing solvers have been generally applied to the singular, nonlinear and dynamical systems based on the platform of neural network together with the swarming/ evolutionary optimization schemes [30] [31] [32] . The stochastic solvers have been applied in diverse applications, few of them are coronavirus SITR model [33, 34] , singular doubly differential systems [35] , fluid dynamics problems [36] , HIV infections modeling systems [37, 38] , and electric circuits model [39, 40] . The authors are motivated by keeping these stochastic-based applications to design a computing solver for the FBTMM. Therefore, the objective of study is to introduce a novel design and implementation of a neuro-swarming computational numerical procedure MWNN-PSOASA for numerical treatment of the fractional Bagley-Torvik mathematical model (FBTMM) by exploiting global search optimization procedures via particle swarm optimization (PSO) and local search via active-set approach (ASA), while Mayer wavelet kernel-based activation function used in neural network (MWNNs) modeling.

The novelty and significance of the research investigations are briefly described in this section. The literature review presented for FBTMM, one can decipher evidently that a large variant of deterministic solvers have been introduced by research community for solving the FBTMM while few studies of stochastic solvers are available for finding the approximate solutions for FBTMM, Therefore a novel fractional neural network is presented for the solution of FBTMM by exploiting the strength of fractional Mayer wavelet neural networks (MWNN) based modeling of the fractional derivative terms in ODE (1) and training of these networks are performed by hybrid heuristics having global search with PSO and ASA based local refinements, i.e., MWNN-PSOASA. The designed solver MWNN-PSOASA is used efficiently and effectively to solve the FBTMM numerically. The achieved form of the numerical results is compared with the accessible true/exact solutions, which shows the precision, reliability, constancy, and convergence of the designed solver MWNN-PSOASA. The reliability of the results based on the designed solver MWNN-PSOASA is further presented using the statistical procedures of Mean, semi interquartile range (S.I.R), Minimum (Min), standard deviation (STD), Theil's inequality coefficient (TIC), mean square error (MSE) and Maximum (Max). Besides, the accurate and reasonably stable outcomes of the FBTMM through the designed solver MWNN-PSOASA, robustness, smooth processes, and exhaustive pertinence are other significant perks of the scheme.

The organization of this study is considered as: The methodology of MWNN-PSOASA is accessible in Sect. 2. The performance operators are shown in Sect. 3. The comprehensive results detail is given in Section 4. Final remarks and upcoming research directions are given in the last section.

This section represents the methodology of the designed solver by using the Mayer wavelet neural network along with the optimization of PSOASA to solve the FBTMM. The genetic flow diagram of proposed MWNN-PSOASA for solving FBTMM is provided in Fig. 1 , in which the process blocks in four steps are presented. The construction of the FBTMM, merit function based on the mean square error, and the PSOASA optimization is also presented in this section.

In this section, the FBTMM solutions are signified byv(τ ), whereas, D (n)v (τ ) and D αv (τ ) provides the integer derivatives of order n and fractional form of the derivative. The mathematical formulations of these systems by means of continuous mapping in neural networks models are given as:v An objective function-based Mayer wavelet is written as:

The updated Eq. (2) using the above values is become as:

The combination of the MWNN with the optimization of PSOASA is used to solve the FBTMM based on the availability of appropriate W values. For the ANN weights, an objective function e F is given as:

Here e F−1 and e F−2 are the objective functions based on the FBTMM and the ICs of Eq. (1), respectively written as:

The parameter optimization, i.e., weights, for the MWNN models are obtained using the hybridization of computing procedures of particle swarm intelligence PSO as an efficacious global search aided with active set algorithms (ASAs) for efficient local refinement mechanism to solve the variants of FBTMM in equation (1). Particle swarm optimization is a computational swarm intelligence approach, which is used to optimize a model through the process of iteration to improve the applicant outcomes, i.e., candidate solutions of a specific optimization tasks, with respect to assume quality measures and constraints. The PSO normally solves a model by using the population of applicant outcomes called swarm and each candidate solution is represented by the particles. The PSO algorithms operate with the adjustment of these particles during each flight in search-space based on the mathematical representations of the particles velocity and position in terms of previous velocity, inertia weight of velocity, cognitive learning block via local best particle, and social learning mechanism via global search particle. The movement of the particles is affected by its local prominent based positions; however, it is also directed to the best-recognized positions, which are efficient as improved positions of other particles. This is projected to transfer the swarm to the best results. Additional necessary elaborative details, underlying theory, mathematical representation, scope, and applications in diversified fields can be seen in [41] [42] [43] and references mentioned in them. In recent decades, PSO is implemented to plant diseases diagnosis and prediction [44] , nonlinear Bratu systems governing the fuel ignition model [45] , identification of control autoregressive moving average systems [46] , reactive power planning [47] , and thermal cloaking and shielding devices [48] .

In order to control and speed up the convergence performance of the global search PSO, the optimization through the hybridization with local search method is implemented for speedy adjustment of the parameters. The active-set algorithm is one of the quick, rapid, and efficient local search schemes, which is famous to find the optimal performances in different fields. ASA is an effectual convex optimization tool that is implemented for unconstrained and constrained systems. Few prominent applications of utmost performance of the ASA include embedded system of predictive control [49] , pressure-dependent network of water supply systems [50] , local decay of residuals in dual gradient method [51] , nonlinear singular heat conduction model [52] , and warehouse location problem [53] . Therefore, in the presented study, a memetic computing paradigm PSOASA based on global search efficacy of PSO aided with speedy tuning of parameter with ASA are exploited for finding the known adjustable of MWNN models for solving the FBTMM in Eq. (1).

In this section, the solutions of three different variants based on the FBTMM are provided by using the integrated design heuristics of MWNN-PSOASA. The precision and convergence on the basis of sixty number of autonomous trails MWNN-PSOASA are presented using sufficient large number of graphical and numerical illustrations with elaborative details for solving the variants of the FBTMM. While the information/outcomes of different performance indices for the analysis of proposed MWNN-PSOASA are given in this section.

The mathematical form of the performance gages of TIC and MSE are presented to solve the FBTMM as follows:

where n is total number of grid points, i.e., τ m , m 1, 2, …, n, while v m is the reference solution for m th grid point whilev m is proposed approximate solution for the m th grid point. 

An error function is derived as:

Example 2 Consider an FBTMM given in Eq. (1) using the values of h(τ ) τ 2 + 4 τ π + 2, α 1.5 and a 1 a 2 a 3 1 is given as:

An error function is derived as:

Example 3 Consider an FBTMM given in Eq. (1) using the values of h(τ ) τ 2 + (3) (1.25)v (τ ) 0.25 +2, α 1.75 and a 1 a 2 a 3 1 is written as:

An error function is derived as:

The 10 −04 to 10 −12 , 10 −06 , to 10 −10 and 10 −04 to 10 −10 for each example of the FBTMM, respectively. One can conclude from these outcomes that around 75% independent trials of MWNN-PSOASA achieved precise level of the accuracy.

In order to authenticate the accuracy, precision, convergence, and efficiency analysis, the statistical results in terms of Min, median (Med), Mean, STD, Max, and S.I.R operators are tabulated in Table 2 which are calculated for sixty independent runs using Example-1 Min 1 × 10 -07 3 × 10 -07 2 × 10 -07 5 × 10 -08 6 × 10 -08 2 × 10 -07 2 × 10 -07 2 × 10 -07 2 × 10 -07 1 × 10 -07 Mean 3 × 10 -02 2 × 10 -02 1 × 10 -02 6 × 10-03 1 × 10 -02 2 × 10 -02 3 × 10 -02 3 × 10 -02 4 × 10 -02 4 × 10 -02 Max 2 × 10 -01 1 × 10 -01 6 × 10 -01 2 × 10 -01 3 × 10 -01 8 × 10 -01 1 × 10 -01 1 × 10 -01 2 × 10 -01 2 × 10 -01 Med 1 × 10 -04 3 × 10 -04 4 × 10 -04 4 × 10 -04 4 × 10 -04 5 × 10 -04 6 × 10 -04 8 × 10 -04 8 × 10 -04 8 × 10 -04 SIR 3 × 10 -04 6 × 10 -04 8 × 10 -04 9 × 10 -04 1 × 10 -03 1 × 10 -03 1 × 10 -03 1 × 10 -03 1 × 10 -03 1 × 10 -03 STD 2 × 10 -01 1 × 10 -01 8 × 10 -02 3 × 10 -02 5 × 10 -02 1 × 10 -01 1 × 10 -01 2 × 10 -01 2 × 10 -01 2 × 10 -01

Example-2 Min 2 × 10 -08 3 × 10 -08 9 × 10 -08 6 × 10 -08 7 × 10 -08 7 × 10-09 4 × 10 -08 7 × 10 -08 8 × 10-09 6 × 10 -08 Table 3 for all three examples of FBTMM. The sufficient large number of independent execution of MWNN-PSOASA achieved the FIT, TIC and MSE less than 10 −04, that prove the worth of design scheme for solving FBTMM.

Comparison of the outcomes of fractional MWNNs optimized with PSOASA for solving FBTMM has been made with reported results of state of the art deterministic and stochastic solver in order to access the performance rigorously. The absolute error of the reported numerical solver based on matric approach introduced by Podlubny [11] , sigmoidal fractional neural networks optimized with IPA (FNN-IPA) [22, 23] , sigmoidal neural networks optimized with GAs aided with pattern search (PS), i.e., GA-PS [54] and sigmoidal neural networks trained with particle swarm optimization (PSO) supported with PS, i.e., (PSO-PS) [20] are presented in Table 4 along with the proposed results of FMWNN-PSOASA. One can easily decipher from results presented in Table 4 , the values of the AE for FMWNN-PSOASA are comparable to state-of-the-art deterministic and stochastic numerical procedures for solving FBTMM. 

The current work investigations are to design a neuro-swarming computational numerical procedure for the fractional Bagley-Torvik mathematical model. The optimization procedures based on the global search particle swarm optimization and local search active-set approach using the activation function Mayer wavelet neural network have been applied to solve the fractional model. The proposed stochastic solver MWNN-GAASA efficiency is performed to solve three different variants based on the fractional order of the FBTMM. For the exactness of the stochastic solver MWNN-PSOASA, the comparison of the attained and exact solutions will be provided for each variant of the FFBTMMM. The AE values have been obtained in good measures that are calculated around 10 −06 to 10 −07 for each example of the FBTMM. For the reliability of the proposed stochastic solver MWNN-PSOASA, the statistical soundings are provided based on the stability, robustness, accuracy, and convergence. One can conclude from these outcomes that around 75% independent trials achieved precise level of the accuracy. Beside the advantage of accurate and reliable outcomes of designed MWNN-PSOASA, the limitation of slowness of operation of global search with PSO and then local search with ASA.

The FMWNN-PSOASA can be implemented to solve the fluid nonlinear models, fraction order systems, and fluid models [55] [56] [57] [58] [59] [60] [61] [62] [63] . Moreover, the used of heuristic methodologies having inherent strength of global as well as local search like differential evolution, backtracking search optimization algorithm, weights differential evolution, and their recently introduced variants are good alternative of integrated PSOASA.

Funding Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature. This paper has been partially supported by Fundación Séneca de la Región de Murcia grant numbers 20783/PI/18, and Ministerio de Ciencia, Innovación y Universidades grant number PGC2018-0971-B-100.

There is no data associated with this manuscript.

Conflict of interest All the authors of the manuscript declared that the authors have no conflicts to disclose.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Fractional calculus-a different approach to the analysis of viscoelastically damped structures

Fractional calculus in the transient analysis of viscoelastically damped structures

On the appearance of the fractional derivative in the behavior of real materials

On the fractional calculus model of viscoelastic behavior

General solution of the Bagley-Torvik equation with fractional-order derivative

On the Bagley-Torvik equation

A new operational matrix of Caputo fractional derivatives of Fermat polynomials: an application for solving the Bagley-Torvik equation

An investigation of fractional Bagley-Torvik equation

Analytical approach to a class of Bagley-Torvik equations

On the explicit solutions of fractional Bagley-Torvik equation arises in engineering

Fractional differential equations: an introduction to fractional derivatives, fractional differential equations, to methods of their solution and some of their applications

Numerical solution of the Bagley-Torvik equation

Solution of fractional differential equations by using differential transform method

Analytical solution of the linear fractional differential equation by Adomian decomposition method

Application of He's Variational Iteration Method to Solve Semidifferential Equations of ð '› th Order

Matrix approach to discretization of fractional derivatives and to solution of fractional differential equations and their systems

A collocation-shooting method for solving fractional boundary value problems

The solution of the Bagley-Torvik equation with the generalized Taylor collocation method

Solving an infectious disease model considering its anatomical variables with stochastic numerical procedures

Swarm intelligence optimized neural networks in solving fractional system of Bagley-Torvik equation

On Haar wavelet operational matrix of general order and its application for the numerical solution of fractional Bagley Torvik equation

Design of unsupervised fractional neural network model optimized with interior point algorithm for solving Bagley-Torvik equation

Integrated intelligence of fractional neural networks and sequential quadratic programming for Bagley-Torvik systems arising in fluid mechanics

Local discontinuous Galerkin approximations to fractional Bagley-Torvik equation

An exponential spline approximation for fractional Bagley-Torvik equation. Boundary Value Problem

Jacobi collocation methods for solving the fractional Bagley-Torvik equation

Approximating solutions to fractional-order Bagley-Torvik equation via generalized Bessel polynomial on large domains

Numerical solution of a fractional-order Bagley-Torvik equation by quadratic finite element method

On Lie symmetry analysis of certain coupled fractional ordinary differential equations

Applications of Gudermannian neural network for solving the SITR fractal system

Integrated intelligent computing paradigm for the dynamics of micropolar fluid flow with heat transfer in a permeable walled channel

A novel design of morlet wavelet to solve the dynamics of nervous stomach nonlinear model

A stochastic intelligent computing with neuro-evolution heuristics for nonlinear SITR system of novel COVID-19 dynamics

Integrated neuro-swarm heuristic with interior-point for nonlinear SITR model for dynamics of novel COVID-19

Numerical solution of doubly singular nonlinear systems using neural networks-based integrated intelligent computing

The intelligent networks for double-diffusion and MHD analysis of thin film flow over a stretched surface

A novel study of Morlet neural networks to solve the nonlinear HIV infection system of latently infected cells

Solving a class of biological HIV infection model of latently infected cells using heuristic approach

Integrated computational intelligent paradigm for nonlinear electric circuit models using neural networks, genetic algorithms and sequential quadratic programming

Design of nature-inspired heuristic paradigm for systems in nonlinear electrical circuits

Different Applications of PSO

Novel application of FO-DPSO for 2-D parameter estimation of electromagnetic plane waves

Particle swarm optimisation in designing parameters of manufacturing processes: A review

An optimized model based on convolutional neural networks and orthogonal learning particle swarm optimization algorithm for plant diseases diagnosis

Solution of the one-dimensional Bratu equation arising in the fuel ignition model using ANN optimised with PSO and SQP

Nature-inspired heuristic paradigms for parameter estimation of control autoregressive moving average systems

A new fractional particle swarm optimization with entropy diversity based velocity for reactive power planning

Particle swarm optimization-based algorithms for solving inverse problems of designing thermal cloaking and shielding devices

Machine learning-based warm starting of active set methods in embedded model predictive control

Content-based active-set method for the pressure-dependent model of water distribution systems

Local decay of residuals in dual gradient method applied to MPC studied using active set approach

A new stochastic computing paradigm for the dynamics of nonlinear singular heat conduction model of the human head

An active-set trust-region algorithm for solving warehouse location problem

Solution of fractional order system of Bagley-Torvik equation using evolutionary computational intelligence

A generalization of truncated M-fractional derivative and applications to fractional differential equations

The Marichev-Saigo-Maeda fractional calculus operators pertaining to the generalized k-Struve function

Analytical and approximate solutions of fractional partial differential-algebraic equations

On solutions of fractional order telegraph partial differential equation by cranknicholson finite difference method

Fractional calculus of the extended hypergeometric function

A modified invariant subspace method for solving partial differential equations with non-singular kernel fractional derivatives

New analytical solutions of conformable time fractional bad and good modified Boussinesq equations

System analysis of HIV infection model with 4 under non-singular kernel derivative

New complex and hyperbolic forms for Ablowitz-Kaup-Newell-Segur wave equation with fourth order