Unconstrained evolutionary and gradient descent-based tuning of fuzzy-partitions for UAV dynamic modeling

Abstract

In this paper, a novel fuzzy identification method for dynamic modelling of quadrotors UAV is presented. The method is based on a special parameterization of the antecedent part of fuzzy systems that results in fuzzy-partitions for antecedents. This antecedent parameter representation method of fuzzy rules ensures upholding of predefined linguistic value ordering and ensures that fuzzy-partitions remain intact throughout an unconstrained hybrid evolutionary and gradient descent based optimization process. In the equations of motion the first order derivative component is calculated based on Christoffel symbols, the derivatives of fuzzy systems are used for modelling the Coriolis effects, gyroscopic and centrifugal terms. The non-linear parameters are subjected to an initial global evolutionary optimization scheme and fine tuning with gradient descent based local search. Simulation results of the proposed new quadrotor dynamic model identification method are promising.

Keywords

• fuzzy identification method
• dynamic modelling
• quadrotor uav
• christoffel symbols
• genetic algorithms
• non-linear parameters
• global evolutionary optimization