Short-Term Water Level Prediction Base on Meta-Cognitive Interval Type-2 Fuzzy Neural Networks

Many research has been conducted over the years on the fuzzy neural networks for water level prediction, but they have been largely unsuitable due to the high fluctuation and nonlinearity in reservoir water level. To enhance the predction accuracy and reduce the computation complecxity, this paper proposes a meta-cognitive interval type-2 fuzzy neural network (McIT2FNN) for short-term water level forecasting. The meta-cognitive learning mechanism is utilized to monitor and control the learning process, constructing an optimal set of rules based on knowledge novelty. Then, the Levenberg-Marquardt (LM) algorithm is utilized to adjust the parameters of the interval type-2 fuzzy neural network. The efficacy of the McIT2FNN is evaulated with acutual wave level data collected from a hydropower facility in Sichuan, China, the results indicate that McIT2FNN can obtian better prediction accuracy. Also the performance of McIT2FNN is compared with other models, demonstrating superior accuracy and faster convergence speed.