Artificial neural networks model design of Lorenz chaotic system for EEG pattern recognition and prediction

This paper presents the preliminary work of a multidisciplinary brain research program. The goal of this research program is to generate accurate and effective signals for non-invasive brain stimulation, and deliver a hardware prototype to monitor and treat motion related mental disease such as Parkinson's and Epilepsy. It was shown in previous research that Electroencephalogram (EEG) signals captured from brain activities demonstrate chaotic features. Artificial neural network (ANN) resembles brain biological neural network and can be used to simulate chaotic system. The trained ANN model can in turn be used to analyze and control brain activities. In order to investigate the chaotic phenomenons of EEG signals and develop function for automatic pattern recognition, large amount of EEG signals are required. However, EEG signals are prone to noise and the available data is very limited. It is possible to control and predict the time series outputs of chaotic systems with known equations. Therefore, in order to study the dynamic control of the brain neural networks, an ANN architecture is designed and optimized for implementing Lorenz attractor to simulate the chaotic states of EEG signals. The research includes chaotic system, ANN design and the optimization of ANN architecture, which is based on the consideration of hardware implementation. The designed ANN model is trained with Lorenz attractor outputs with a fixed set of system parameters and the optimized architecture is selected based on the training results of three training algorithms and 16 ANN architectures with different number of hidden neurons.