Adaptive Neuro-fuzzy Inference System on Downstream Water Level Forecasting

To optimize water resource management, a better reservoir operation system is required. However, all flood control decisions depend on many variables; it is never an easy task. For example, water level prediction under tidal effects is one of the essential judgments in flood control problems, and it determines that reservoir release water or not. Therefore, in order to predict the optimal reservoir drainage, avoidable of inundating downstream area, reliable water level prediction system is necessary. A five-layer ANFIS model with three input and one output variables is built in this paper. Differ from other researches in the past, the estuary tide is considered as an input variable in this ANFIS model. Since the downstream area is located in tideland, the tide impacts water level as well as the other two inputs, rainfall and drainage of reservoir. Predicting downstream water level accurately is very valuable for reservoir to manipulate drainage in flood season, and reservoir could control flood efficiently thus. The ten past years of 16 typhoon events with over four thousand hourly data are collected and used to train ANFIS model. Some successful results are displayed in this paper, and it demonstrates that ANFIS is appropriate for forecasting water level.