Genetic Algorithm Based Approach for Parameters Estimation of the Hess Operator Model

Abstract

In the 2006 MSTC paper on operator modeling ( A Con spectus on Operator Modeling: Past, Present and Future) 1 , the authors laid the theoretical foundation for the prospective methodology of the real -time estimation of operator model parameters. The proposed approach heavily relies on hybrid intelligent compu ting techniques, including neural net work s, fuzzy inference systems and genetic algorithms . This paper addresses the first part of the approach which is to identify the parameters of a human operator model, which is necessary to establish a data base to b e used in the real -time application . Such an approach is categorized as a data driven model , which implies that time histories of the input/ output data are available . In the case of the operator model, this relationship is given by the disturbance and the operator con trol as predicted by the Hess 2 structural model . The primary task of the current investigation was to devise a methodology, which would automatically identify parameters of the Hess model given the knowledge of the input disturbance and actual (real) operator behavior . The identification process is accomplished by searching for a combination of the Hess model parameters that would drive the model response to match the actual operator behavior. The search engine utilized by the proposed methodol ogy is based on the Genetic Algorithm (GA) concept, - a powerful soft computing technique, often used as an alternative to conventional optimization algorithms. Preliminary , non -real -time results presented in this paper demonstrate high efficiency of the p roposed methodology . The estimated parameters are in good correlation with the nominal val ues: the estimation error is 0% in its minimum, and 15% in its maximum . Time domain comparison shows an exact match of the estimated and actual operator control signa ls. These results are accompanied by the appropriate graphs and tables.