Locally recurrent Functional Link Fuzzy neural network and unscented H-infinity filter for shortterm prediction of load time series in energy markets

Abstract

The paper presents a locally recurrent Fuzzy neural architecture to forecast electrical loads in an energy market on a short-term basis. In recent years combination of recurrent filter neurons with Fuzzy neural networks has gained significance to provide the identification of the temporal nature of the time series data. Further to increase the dimension of the input space the consequent part of the fuzzy rules are augmented with functional link networks and this provides a better approximation of the input-output mapping. Besides to provide faster learning in comparison to the gradient descent or evolutionary techniques a robust H-infinity unscented Kalman filter is used. Some of the energy market load time series data are used for numerical experimentation to highlight the significant improvement in the prediction performance of the hybrid Functional Link Fuzzy neural network (FLFNN).