Riverflow prediction using neuro-fuzzy systems

Amithirigala Jayawardena, Ranjan Sarukkalige, Honey Badrzadeh

Thursday 2 july 2015

15:05 - 15:20h at Central America (level 0)

Themes: (T) Water resources and hydro informatics (WRHI), (ST) Catchment hydrology

Parallel session: 12H. Water resources - Catchment

In this paper, we propose a data driven approach of daily river flow prediction and apply it to predict the daily discharge data at St. George gauging station across Balonne River, which is part of the Murray-Darling Basin, located in the Queensland State of Australia. Neuro-fuzzy systems synergizes the human-like reasoning style of fuzzy systems with the connectionist structure of artificial neural networks and derives benefits of neural networks as well as of fuzzy logic systems and removes their individual disadvantages by combining them on their common features. The mean daily discharges for 27 years from Feb 1985 to Jul 2012 (time series with a length of 10,080 data points), are taken from the Queensland Department of Natural Resources. As usual, the first 70 percent of the data are used for training while the remaining data are used for testing the model. The average daily discharge at the St. George station is 38.11 m^3/s and the maximum during the study period is 3683.19 m^3/s. Predictions made for several lead-time steps are satisfactory. However, the reliability of the predictions decreases as the lead-time increases which is expected.