A Fuzzy Neural Network and Its Gradient-descent Algorithm for Prediction Intervals

Abstract

Recent years witness the tremendous growth of renewable energy from various sources such as solar, wind, tidal, wave energy, etc. As the data collected form the energy farms are critial for production and maintenance,  the prediction and forecasting task are consequently becoming essential in the field studies of solar energy, coastal and ocean engineering, etc. However, it is difficult to predict the energy parameters in long term and even in the short term due to their intermittent nature. This study aims to propose a solution to handle the issue using Interval type-2 fuzzy inference system, or IT2FIS. IT2FIS has been shown to be capable of handling uncertainty associated with the data. The proposed IT2FIS is a fuzzy neural network realizing Takagi-Sugeno-Kang inference mechanism. IT2FIS structure evolves automatically and the parameters are updated based on the meta-cognitive gradient descent algorithm. For performance evaluation studies, collected wind speed data were used. Using historical data and seasonal forecasting, the model provides short-term forecasting of wind energy parameter. The performance of IT2FIS is compared with existing state-of-the art fuzzy inference system approaches and results indicate the advantage of IT2FIS-based prediction.