Chaotic Time Series Forecasting using Locally Quadratic Fuzzy Neural Models

Time series forecasting in highly nonlinear and chaotic systems is a challenging research area with a variety of applications in economics, environmental sciences and various fields of engineering. This paper presents a novel Locally Quadratic Fuzzy Neural Model (LQFNM) to forecast the behavior of highly nonlinear and chaotic time series. It is based on the idea of approximating a nonlinear function with interpolated local quadratic models using a tree construction algorithm. A fast heuristic learning algorithm is integrated in the model to derive the structure as well as the parameters of the Locally Quadratic Models. Four different case studies are conducted in which the performance of the method is evaluated through comparisons with other techniques available in literature. The results confirm the accuracy and reliability of the presented method. The proposed LQFNM can be applied to time series forecasting in a wide range of real world applications. Key-Words: Locally Quadratic Neural Fuzzy Model, Forecasting, Chaotic Time Series