Description of Business Problem
Run-of-river (ROR) hydro projects harness the energy in flowing water to generate
electricity. There is total dependency on the flow rate measured in cusecs. As there is
an absence of a dam and reservoir, the power generation is susceptible to the effects of
daily and seasonal flow infrequencies. SinoHydro is a global Hydro Project construction
company, looking for construction projects in the next quarter. It is looking to bid for six
ROR hydro projects and wants us to forecast the river flows for next three years for the
- Madison River, West Yellowstone
- Chang Jiang, Han Kou
- Pigeon River, Ontario
- Funder River, Logager
- Piper’s Hole, Newfoundland
- Mitta Mitta River, Tallandoon
The data is available for the years 1981 – 1990 and consists of the mean monthly flow rate.
Flow rate predictions for three years will also be used to estimate expected power
generation and revenue. Over prediction in such a scenario is expensive and shall be
penalised as we do not wish the company to take up projects that are unprofitable.
We plotted the monthly flows of each river against time to understand the trend and the
seasonality (shown for two rivers below) which would help us in choosing an appropriate
model for forecast. These would also give an insight into the external factors if any at play.
Chang Jiang: We observe no trend in the average monthly flows and a seasonality of 12 as
shown in the plots below. The peaks are observed from the months of June to august which
coincide with the rainy season.
Funder: We observe no trend in the average monthly flows and a seasonality of 12 can be
observed even though the change is slight since the values are small as shown in plots below
The viability of methods was evaluated based on the time series components available in
the series. Following our initial analysis, we concluded that the data has no trend and has
only level, noise and seasonality. For a metric like river flow this conclusion is very intuitive
because barring external factors the flow rate is not expected to grow year on year. Hence
the models highlighted in green have been finalised to try on the river flow data.