Systems Identification by Quasilinearization and by Evolutionary Programming

Abstract Two new methods for obtaining the values of the coefficients in the differential equations describing the dynamics of a system are developed. The first method is based on a quasilinearization procedure and is applicable in parameter identification problems where the plant is modeled by a system of linear differential equations, and noisy measurements of state and control variables are available. Computationally, this method is equivalent to a modification of the classical Newton-Raphson method. The second, a “directed random search” method, is based on a concept called evolutionary programming, and is also applicable for nonlinear problems. Using recorded flight test data of an experimental aircraft, the two methods are compared for accuracy and computational efficiency.