A Modified Brain Emotional Learning Model Inspired By Online Recurrent Memory Sequential Extreme Learning Machine Based On Neural Networks

• Published in: مجله مدل سازی در مهندسی Vol. 20; no. 70; pp. 1 - 21
• Main Authors: Mehdi Golshan, Mohammad Teshnehlab, Arash Sharifi
• Format: Journal Article
• Language: Persian
• Published: Semnan University 01.09.2022
• Subjects: brain emotional learning; neural networks; online sequential extreme learning machine; recurrent memory networks
• ISSN: 2008-4854; 2783-2538
• DOI: 10.22075/jme.2022.25125.2184

Predicting data, in the form of complex and chaotic time series, is one of the fundamental issues in various scientific and industrial fields. Data-driven models such as artificial neural networks and fuzzy neural networks compared to other models have been received more attention due to their special features. To develop and improve these models, the concepts of the mammalian brain limbic system are used. Therefore, the brain emotional learning machine is introduced. In this paper, the online sequential extreme learning machine is used as the main component in the processing centers of the brain emotional learning machine. To interact between processing centers, the online sequential extreme learning machine is designed in the form of a recurrent memory network with transfer learning ability. The proposed model is named the brain emotional learning based on recurrent memory online extreme learning machine (BEL-ORMS-ELM). To evaluate and compare the efficiency of the proposed model, the initial parameters of the models are adjusted according to the Mackey-Glass and Lorenz time series data under the same conditions. Different models are evaluated and compared based on the valid measurable criteria in regression problems prediction. The simulation results show that the proposed model with sigmoid and hyperbolic tangent activation function for Mackey-Glass and Lorenz time series test data has the highest performance criteria compared to similar online models. It also has acceptable performance for training data compared to similar models.