The present reservoir system I am working on has only one major reservoir for which inflow forecasting has to be undertaken. The catchment area upto the reservoir site is 62,245 km^2. There are around 6 stream gauging stations on the main river upstream of the dam and 5 stream gauging stations on the principal tributaries for which daily stream flow discharge and water level data are available. Further, time series of daily reservoir storage, levels and inflow volumes are also available for a period of 40 years. Also, daily rainfall data for the concurrent time period at around 50 rain gauge stations are available.
Now, the doubt in my mind is that what combination and which technique shall I adopt using these data.
1. Should I adopt time series of reservoir inflows of previous time periods to develop model to forecast inflows into reservoir?
2. Should I adopt discharges of upstream stream gauges in the model to forecast inflows into reservoir?
3. Should I go for a cause-effect model, wherein the lumped rainfall across the basin can be linked to reservoir inflows and utilized to forecast inflows into reservoir?
4. Also which artificial intelligence/soft computing/data mining technique(s) should I adopt for the data sets I have and for the above stated system description.
Your kind and valuable inputs will be appreciated and help be to proceed in solving my problem.
Regards!
Priyank Sharma
Can try deep learning, should work for such data, I have applied it on high noise electric load forecasting.
Most of the climate change models involving forecasting use rainfall as a predictor, and from that draw connections to the past record. Also plotting the past record may help show trends, and perhaps comparing both annual and seasonal trends may help. Land use, conversion of forests, urbanization, imperveous surfaces can alter runoff and baseflow, as well as changes in water uses for irrigation, industry and consumption. I would probably plot flow duration curves for all stations and tributaries, perhaps also look to see if there have been shifts or trends by decade in the curves, as changes in baseflow and high flows might suggest, with consideration of rainfall, land and water use changes. If you have GIS coverages and aerial photos of land use changes over the decades, that may help with forecasting changes. It is possible to have the same annual rainfall, yet, marked differences in discharge, depending on storm size, intensity and other factors. Of course, you might try correlating the amount of rainfall with water yield, but a basin of this size has substantial time delays. Since you already have substantial data, you are not trying to use the model to estimate discharge, but forecast considering past trends, changes in land use, etc. Personally, I would get all this data into Excel and where appropriate GIS. I would look for trends, considering also rainfall differences, in river station data and tributaries, and plot time series, daily, monthly, annual water year, and perhaps seasonal and by decade, etc., looking for indicators of change that can be used in identifying trends that may be applied forward into a forecast.
Models may be great for speed, but it is easy to input data, receive output, and for you to loose almost all connectivity to the data. Models also need validation. Perhaps a water balance model run by water year on the various stations would be helpful, factoring in rainfall, discharge, ET estimate (model like Thornthewaite-Mather) and remainder deep seepage losses (important in some geology types and conditions of loosing streams). If considering comparing daily station data, you should probably factor in delay in time series based on watershed size, time of travel. These differences melt away as time increases to annual or decade levels, but you should always note dates of extreme events relative to rainfall and flood peaks. On occassion, rainfall from extreme event is one month, peak discharge in next month due to time delay.
I would say forecasting is not an easy job, and 40 years is excellent, but sometimes the review of more distant past records can be revealing (dendrochronology, glacial cores, paleochannels, old or ancient historical records, geological dating). If the 40 years was during a 20 year drought cycle, it would be good to know this.
As already have been said, I think you should pay attention to possibles trends in your data and look up for seasonal data also. I think R is a great packaged for statistics/data analysis (look for functions of auto-correlation and trends).
If you are interested in modelling a extreme event to verify hydraulic security of your reservoir, I think you should use a rainfall-runoff model to forecast the inflow(HEC-HMS), using you past series to calibrate your model, together with a hydraulic model (HEC-RAS)
If you are interested in the mean behavior of the watershed, you could use statistics analysis to forecast you future series and a water balance model to predict the evaporation, yield and release in your reservoir. Also you could think in use a Monte-Carlo simulation together with this analysis.
CAMPOS, J. N. B. Modeling the Yield-Evaporation-Spill in the Reservoir Storage Process: The Regulation Triangle Diagram. Water Resources Management, v. 24, n. 13, p. 3487–3511, 2010.7
ÜNAL, N. E.; AKSOY, H.; AKAR, T. Annual and monthly rainfall data generation schemes. Stochastic Environmental Research and Risk Assessment, v. 18, n. 4, p. 245–257, 2004.
ALVES, J. M. B.; CAMPOS, J. N. B.; SERVAIN, J. Reservoir Management Using Coupled Atmospheric and Hydrological Models: The Brazilian Semi-Arid Case. Water Resources Management, v. 26, n. 5, p. 1365–1385, 2012.
CAMPOS, N.; STUDART, T.; FRANCO, S.; LUNA, R. Hydrological Transformations in Jaguaribe River Basin during 20th Century. Proceedings of the 20th Annual American Geophysical Union, p. 221–227, 2000.
The best forecasting soft I now is GMDHshell.com. Look at the WIKIpedia.
Adaptive neuro fuzzy inference systems can be used for simulating this state, due to that the ANFIS model integrates both neural networks and fuzzy logic principles in a single framework.
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication