DAILY STREAMFLOW FORECASTING USING SIMPLIFIED RULE-BASED FUZZY LOGIC SYSTEM

In this study, a simplified fuzzy logic system with uniform partitions in the input space is proposed for forecasting the daily streamflow of four river systems in Malaysia. The proposed simplified fuzzy logic system was calibrated (trained) using back-propagation (BP) and recursive prediction error (RPE) algorithms. For each catchment, the calibration data set consisted of three consecutive years of daily rainfall and streamflow records. Verifications of the calibrated models were done using the data set of the following year. The performances of the simplified fuzzy logic system and the normal fuzzy logic system are compared, with each model having the same number of adjustable parameters. The results are also compared with the auto-regressive with exogenous input model. This study has shown that the proposed RPE algorithm performed better than the more popular BP algorithm. The results show that all the simplified fuzzy logic system models registered better performance measures for the calibration data sets. However, variable results were obtained for the predictions of the verification data sets