DESIGN OF FUZZY LOGIC-BASED INTEGRATED ADAPTIVE DECAYED BRAIN EMOTIONAL LEARNING NETWORKS FOR ONLINE TIME SERIES PREDICTION

Abstract

The emotional neural network (ENN) is a new field in artificial intelligence systems (AISs). Although ENN was proven in control applications and successfully solved several control problems, it still suffers severe technical issues concerning prediction. This study aims to design a new model for an intelligent forecasting technique constructed by unified Adaptive Decayed Brain Emotional Learning (ADBEL) combined with a Neo-Fuzzy Neuron (NFN) network. In the literature, the ADBEL network is used to predict time series in online mode. Unlike other popular learning networks such as ar- artificial neural networks (ANNs), the ADBEL network offers lower computational time, less complexity, and fast learning, making it ideal for time series prediction in online applications. This thesis aims to further enhance the ADBEL network’s forecasting accuracy through three significant modifications in design. The first modification is its integration with a neo-fuzzy network in the orbitofrontal cortex section. The result is the Neo-Fuzzy integrated Adaptive Decayed Brain Emotional Learning (NF-ADBEL) network. The second modification is the integration with a neo-fuzzy network in two sections: the orbitofrontal cortex section and partially in the amygdala section. This modification leads to a new design: the Expanded Neo-Fuzzy integrated Adaptive Decayed Brain Emotional Learning (ENF-ADBEL) network. The third modification is to design a fuzzy logic-based parameter adjustment model for the ADBEL network, resulting in the F-ADBEL model. The F-ADBEL model can set the learning parameters (namely, alpha, beta, and gamma) of the online mode’s ADBEL network. Root mean squared error and correlation coefficient criteria are used to evaluate the models. The chaotic time series, namely the Mackey-Glass, Lorenz, Rossler, Disturbance Storm Time Index, and the Narendra dynamic plant identification problem, were used for applications and validation. Stochastic series such as wind speed and wind power series are applied to validate the designed models mentioned above. Furthermore, we conducted a comparison between the developed models and the ADBEL and other models. The created models were tested in a MATLAB programming environment and showed superior performance compared to other state-of-the-art predictors.