Discriminating Between Individuals Based on the Parameters of the Enhanced Model of the Human Operator
Kirill Zaychik, Frank M. Cardullo · AIAA Modeling and Simulation Technologies Conference · 2009
Acceptable results have been obtained using conventional techniques to model the generic human operator’s control behavior. However, little resear ch has been done in an attempt to identify an indiv idual based on his/her control behavior. The main hypothesis in current study is that different operators ma y exhibit different control behavior when performing a given control task. The inter-person differences might be manifested in the amount and frequency content of the non-linear component of the control behavior. This paper describes two enhancements to the structural models of the human operator, which allow personalization of the modeled control behavior. One of the proposed enhancements accounts for the “testing” control signals, which are introduced by a n operator for more accurate control of the system an d/or to adjust his/ her control strategy. This enha ncement uses the Artificial Neural Network (ANN), which can be fine-tuned to model the “testing” control behav ior of a given individual. ANN is characterized by a set o f weighting coefficients, which will vary for diffe rent operators. In order to further improve the ability of the Hess structural model to simulate the contro l behavior of a given individual, the second enhancem ent is introduced. The design of the latter is insp ired by observation of actual control signals of different operators: there may be a significant variation in levels of power at any given frequency of the control signal depending on an individual. This enhancement took the form of an equiripple filter, which conditions the power spectrum of the control signal before it is p assed through the plant dynamics block. The filter design technique uses Parks-McClellan algorithm, which allows to parameterize the desired levels of power at cert ain frequencies. The hypothesis here is that these levels of power will vary from person to person. In the 2007 MSTC paper on operator modeling (Genetic Algorithm based approach for parameter estimation of the Hess Operator Model), the authors laid the theoretical foundation for the prospectiv e parameter identification technique, which is driven by genetic algorithm. This paper uses an improved version of the method, which makes it possible to i dentify subject specific parameters of the structur al models and their enhancements. Metric to evaluate effectiv eness of the proposed enhancements is based on comparison of power spectrum densities of the simulated control signals against that of the precision model, which stems off the McRuer’s crossover model. It is demonstrated that both ANN and filter significantl y improve the performance of the model when matching an individual control data. Paper also contains det ails of implementation of the proposed model enhancements as well as details of the improved genetic algori thm driven PID technique .