Predictive analytics for energy systems
Date of Issue2019
School of Computer Science and Engineering
In the research world today, one of the hottest and most researched fields is the time series problem. One of the solutions available is by employing a fuzzy neural network. Two components of fuzzy neural networks are the neural network and fuzzy system. Both of these approaches have their own strengths and limitations. By combining both of these, the limitation of each of them can be reduced to minimal. In this project, research and experimentation have been done to design a new model which incorporates fuzzy neural network and other supporting approaches. RIT2FIS (Recurrent Interval Type-2 Fuzzy Inference System) model is developed as the result of the research. RIT2FIS extends the idea of the fuzzy neural network by using an interval type-2 fuzzy system, memory neurons, and K-means algorithm. RIT2FIS is developed to solve the time series problem. RIT2FIS is tested with some benchmark and real-world problems including non-linear system identification problem, Mackey-glass time-series problem, and wind speed prediction problem. Experiment with different designs, methods, and parameters has been done to get the best performance for RIT2FIS. RIT2FIS has demonstrated superior performance in terms of testing accuracy compared to the other state-of-the-art models. The superior performance can be attributed to the fact that it is robust against error, its rule initialization, gradient descent backpropagation approach, and the ability to capture the past instances. Further research should be done to explore the area of an evolving model in terms of the number of rules and number of layers, which approaches the neural fuzzy system in a completely different direction than the one proposed in this report.
DRNTU::Engineering::Computer science and engineering
Final Year Project (FYP)
Nanyang Technological University