Dynamic switched non-parametric identification of the human physiological response under virtual reality stimuli

In this work, it is proposed a Switched Differential Neural Networks structure (SDNN) to model the human physiological response in a virtual stimuli scenario. Two physiological variables are assessed: electrocardiography and electrodermal activity, which provide a reflex response after stimuli. The proposed approach is focused on the representation of two discrete primary states, relaxation and stress as the response of the virtual stimuli. A switched dynamic approach is set, in which the trigger of an stimuli generates a change in the heartbeat rate as well as in the skin conductivity, constructing the switch between the mentioned states. The SDNN allows to obtain a model structure whose dynamics corresponds to the rate of change of the physiological variables, given as result a particular class of uncertain switched systems. The proposed non-parametric identification in this switched structure is implemented and experimentally assessed showing appropriate convergence rates in, both, switching regions and the continuous states.