10.4122/1.1000000715
Matli, Chandra Sekhar
Chandra Sekhar
Matli
mcs380@yahoo.co.uk
Matli, Chandra Sekhar
Chandra Sekhar
Matli
mcs380@yahoo.co.uk
APPLICATION OF FUZZY INFERENCE SYSTEMS FOR WATER QUALITY MODELLING
XVI International Conference on Computational Methods in Water Resources
2006
2006
The Krishna river is subjected to a varying degree of pollution, caused by numerous
outfalls - municipal and industrial effluents and by other human activities. The
main sources leading to pollution in the river include municipal wastewater from
urban areas of Hyderabad and Nalgonda, and wastewater from a variety of industries.
Non-point runoff is generated by precipitation that washes and cleanses the air and
land surface and then transports, a variety of materials, such as sediment, animal
wastes, fertilizers and leaves, to the nearest natural or man made collection
channel. Hence, it is becoming increasingly evident, that to establish the goals of
the water quality management programme, regulating and controlling only point
source pollution is not sufficient. In addition to the point sources, considerable
pollution reaches the river from various land use activities during the monsoon
period. Runoff from the agricultural lands, unsewered rural and urban areas, etc.,
is the source of non-point source pollution in this part of the basin.
Seasonal variations in quantity and quality of Krishna river water are significant
due to non-uniform distribution of rainfall and hence the discharge in the river.
During the dry season the river water is polluted due to discharge of
treated/partially treated/untreated domestic and industrial wastewaters. In wet
season, the river receives pollutants from non-point sources in addition to
pollutants from point sources. As the flow variations in the river are very large,
the flows are classified as wet flows (June - November) and dry flows (December -
May) for developing regression equations. A generalized regression equation which
can be used for water quality predictions did not yield in good predictions. To
incorporate the influence of previous flow on the present load of the pollutant,
ANFIS models are developed. Also, to develop generalized models for the river
stretch, Fuzzy Inference System is used for the first time and its applicability is
tested. Using concepts of fuzzy sets the flows are classified as low and high. The
influence of previous flow on the present pollutant load could be incorporated by
giving suitable weights by the ANFIS model.
Application of ANFIS for developing a generalized model for the river reach under
study yielded good results. The correlation coefficients in the range of 0.6 to 0.9
and low RMSE indicate suitability of the models to the study area. The models
successfully explained the variation in loads for different flow conditions in the
river. The peak flow and falling stage conditions indicated the influence of
previous flow on the load due to delayed effect. The models are verified using
linear regression. Agreement between model predictions and observed measurements is
within the uncertainty of data. However, for all the pollutants the observed values
fall within 95% confidence bands.