An exact identification of a nonlinear complex structure is very important for its stability and control point of view. However since many accurate systems such as robotics and autonomous systems carrying dynamic and nonlinear behavior, it is a largely challenging task to obtain an accurate model with a less priority knowledge. Fuzzy neural network are adjusted in the mean while. So FNNs are extensively used to resolve the problems related to the classification as well as regression. FNNs use TSK type of fuzzy rules where the consequent parts of the rule are generally based on the linear terms. Therefore the FNNs could not be able to handle the chaotic time series like the nonlinear plant identification, stock market price prediction by providing an accurate mapping. So the consequent part of chaotic time series incorporate the output from FLANNs that yields an enlarged input dimension to handle the indecisive and chaotic variations of the time series database. Further to improve the training speed for the weights and to have an effectively high convergence rate as well as to hold a less computational load in network a second order Levenberg-Marquardt Algorithm is used. The performance of Trigonometric FLANNs is evaluated for computational efficient purpose providing excellent prediction accuracy.