An evolving interval type-2 fuzzy inference system for renewable energy prediction intervals
Trong Trung Anh Nguyen · 2018
Renewable energy is fast becoming a mainstay in today's energy scenario.Some of the main sources of renewable energy are wind, solar in addition to wave, tidal power, etc.The renewable energy sources introduce a significant amount of uncertainty in the energy grid.The main reason is being the inability to forecast the exact energy that could be generated.Due to various natural and artificial factors, it is difficult to predict the amount of the sources in long-term and even in short-term, i.e., there is a lot of uncertainty involved in using these data to build an artificial prediction system.In this thesis, a solution using Interval Type-2 Neuro-Fuzzy Inference System (IT2FIS) is proposed.In the literature, IT2FIS has been shown to be capable of handling uncertainty associated with the data.However, there is a challenge in developing an evolving IT2FIS which can learn and evolve the architecture automatically.The first contribution in this thesis is the development of an evolving interval type-2 fuzzy inference system (McIT2FIS-GD) to handle the uncertainty in data.The system models input features as uncertainty by employing interval type-2 sets in rule antecedent.The rules are of Takagi-Sugeno-Kang type and the learning algorithm of the system uses first-order gradient descent approach.The system employs a computationally fast interval type-reduction and is capable of evolving its architecture and parameters based on data.The performance of McIT2FIS-GD has been evaluated on a set of benchmark function approximation problems.The results show that the proposed system is able to generalize the underlying functional relationship between the input and output.In most of the practical problems, the data are being collected over a period of time in a streaming fashion and the data distribution may change with time.The i