Soft Computing Methods for Groundwater Level Prediction: A Survey

Abstract

Population growth and pollution are causing
groundwater depletion in developing countries like
India. Monitoring groundwater levels is crucial for
effective water resource management. Accurate
estimation of groundwater resources is essential, but
groundwater modeling is inherently non-linear and
cannot be solved using traditional methods alone.
Consequently, soft computing technologies such as
Fuzzy Logic, Genetic Algorithms, and Artificial Neural
Networks (ANN) are gaining importance in hydrological
studies.
Fuzzy Logic is adept at handling imprecise and
ambiguous datasets, while ANN, inspired by human
learning, can learn from examples and adjust weights
accordingly. Genetic Algorithms, mimicking natural
evolutionary processes, offer innovative solutions. This
thesis investigates the development of Fuzzy Logic (FL),
ANN, and other methodologies like FPSO in predicting
groundwater levels. Four models are evaluated,
incorporating different combinations of groundwater
recharge and discharge as inputs, with groundwater level
as the output.
ANN is trained, tested, and validated using groundwater
datasets to identify the most effective model for
groundwater level prediction. FL works optimally with
two inputs, while ANN performs better with more
inputs. Fuzzy interval optimization within FL has
traditionally been challenging, often relying on hit-ormiss
approaches. However, Genetic Algorithms offer a
solution by adjusting fuzzy interval length, with
methodologies like Wang and Mendel's rule basis
creation enhancing performance.
The fuzzy Genetic Algorithm method for estimating
groundwater levels outperforms traditional FL methods,
providing more accurate predictions. Creating fuzzy sets
in FL without specialized knowledge poses a challenge,
prompting the development of computational methods.
This thesis proposes a methodology based on central
tendency concepts to determine the appropriate number
of fuzzy sets.
This approach effectively identifies intervals and fuzzy
sets for fuzzy time series forecasting. Chennai's
reservoir rainfall is modeled using a central tendenciesbased
fuzzy approach, serving as a benchmark challenge
for fuzzy time series analysis. Comparative evaluations
against other methods demonstrate the superiority of the
proposed computational technique, yielding promising
results for benchmark datasets.