Artificial Neural Network based Predictive Modeling of Operator Functional State

In safety-critical Human-Machine systems, objective and quantitative prediction of the operator functional state (OFS) is critical to prevent the possible operator performance breakdown and thus the construction of the corresponding OFS predictive models becomes a necessity. However, the suitable OFS predictive model structure is unknown. In this work, the artificial neural networks were employed to construct three sorts of predictive models: the nonlinear input-output model, the nonlinear autoregressive model and the nonlinear autoregressive model with exogenous inputs. The predictive model performances were compared when the model order was varied gradually from 1 to 10. The modeling results of 5 volunteering subjects showed that the 1st-order nonlinear input-output model was suitable for the OFS predictive modeling. The preliminary results of this paper would provide a basis for the future dynamical predictive modeling of the OFS.