Artificial neural networks training via bio-inspired optimisation algorithms: modelling industrial winding process, case
- PDF / 3,887,292 Bytes
- 25 Pages / 595.276 x 790.866 pts Page_size
- 94 Downloads / 181 Views
METHODOLOGIES AND APPLICATION
Artificial neural networks training via bio-inspired optimisation algorithms: modelling industrial winding process, case study Malik Braik1 · Hussein Al-Zoubi2
· Heba Al-Hiary3
© Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract This research provides a study on how the weights of artificial neural networks (ANNs) can be automatically updated by applying bio-inspired algorithms, particularly using the particle swarm optimisation (PSO) algorithm, grasshopper optimisation algorithm (GOA) and grey wolf optimisation (GWO). These evolutionary computation algorithms were used to evolve the synaptic weights of ANNs to find a particular architecture of ANNs. The developed nonlinear models were targeted to the identification of a particular nonlinear prediction system, an industrial winding process, as a case study. These new models were referred, respectively, to as ANN-PSO, ANN-GOA and ANN-GWO. The proposed models were compared with other linear and nonlinear conventional models including least square error and multiple nonlinear regression methods, respectively, as well as other state-of-the-art models including multilayer perceptron-type NNs, radial basis function and recurrent local linear neuro-fuzzy. The performance of the developed models was assessed using several metric criteria. Comparison of the proposed ANN-PSO, ANN-GOA and ANN-GWO models with other traditional and state-of-the-art models asserts the efficacy of the proposed modelling approaches. Keywords Artificial neural network · Particle swarm optimisation · Grasshopper optimisation algorithm · Grey wolf optimisation · Industrial winding process · Multiple nonlinear regression
1 Introduction Machine learning (ML) algorithms have progressed substantially through the past two decades, from laboratory Communicated by V. Loia. Electronic supplementary material The online version of this article (https://doi.org/10.1007/s00500-020-05464-9) contains supplementary material, which is available to authorized users.
B
Hussein Al-Zoubi [email protected] Malik Braik [email protected] Heba Al-Hiary [email protected]
1
Department of Computer Science, Al-Balqa Applied University, Salt 19117, Jordan
2
Computer Science Department, School of Electrical Engineering and Information Technology, German Jordanian University, Amman 11180, Jordan
3
Department of Computer Information System, Al-Balqa Applied University, Salt 19117, Jordan
curiousness to a workable technology in prolonged commercial use. Artificial neural networks (ANNs), support vector machines (SVMs), fuzzy logic (FL) and neuro-fuzzy are well-known ML techniques that have proliferated as preferred methods of solving a broad range of problems in many areas. The research interest in ANN is related to its attractive benefits it manifests, like adaptation ability, learning capacity and its aptitude to generalise. As a score of this interest, ANN has been extensively used to solve an extensive set of problems such as in colour recognition Al-Azzeh et
Data Loading...
