Improved Soft Computing Techniques In Prediction Of Groundwater Level

Abstract

Due to problems like population and pollution in the developing countries like India, the groundwater is depleting too fast. In order to manage the groundwater resources, groundwater level must be tracked. Groundwater estimation is a crucial aspect to understand the mechanism of groundwater resources. Modeling of ground water is a complex process and possesses non-linear features which cannot be solved using classical methods. Soft computing techniques such as Fuzzy Logic (FL), Genetic Algorithm (GA), Artificial Neural Network (ANN) and Particle Swarm Optimization (PSO) are gaining importance in solving problems of hydrological domains due to their strengths to deal with complex problems. FL is a powerful tool to handle imprecision and vagueness present in the dataset. ANN inspired by human learning can learn through examples and adjust weights according to the problem. GA and PSO are potential techniques for optimizations are motivated by nature. The objective of this PhD work is to investigate innovations in FL, ANN, GA and PSO and there applicability in the prediction of groundwater level. newline newlineOn the basis of available literature five models are considered, where different combinations of groundwater recharge and groundwater discharge are considered as input and groundwater level as output. To investigate the best model for prediction of groundwater level, these developed models are trained, tested and validated on the available data of ground water using ANN. FL is also applied on the developed models. It has been observed that FL performs better for two to four inputs and for larger number of inputs ANN performs better. newlineIn application of FL, optimization of fuzzy intervals is a challenging task. Till now it has been done using hit and trial method which is time consuming. This is the motivation for searching a technique, which can optimize the length of fuzzy intervals in a mathematical way. GA is an unconventional method of optimization is devised to optimize the length of fuzzy intervals. In this process, binary GA is used to adjust the length of fuzzy intervals and rule base is constructed with the help of Wang and Mendel method. This method is implemented on three groundwater models for prediction of groundwater level. Obtained results using fuzzy GA method for the groundwater level prediction are compared with the FL technique and found the noteworthy progress in the performance indicators. In quest of more accuracy, GA is replaced with PSO to adjust length of fuzzy intervals. Results obtained with this combined fuzzy PSO method are significantly improved for the prediction of groundwater level in comparison with other existing methods. newline newlineTo balance exploration and exploitation, the inertia weight plays a crucial part in PSO methodology. Static inertia weights are considered in fuzzy PSO method. The drawback with static weights is that at times PSO converges at local optima. Therefore, adaptive inertia weights are considered for PSO to conquer the problem. The inertia weights are vigorously adjusted with the help of feedback received from particles best positions. This adaptive PSO (APSO) is used to adjust the intervals of fuzzy sets. The results indicate that the fuzzy APSO performs better than fuzzy PSO and fuzzy GA approaches for the groundwater level prediction. newline newlineConstruction of fuzzy sets is one of the tricky tasks in application of FL. A computational method is devised for constructing fuzzy sets in the absence of expert knowledge. This method uses the concepts of central tendency. This study gives a solution to a critical issue in designing of fuzzy systems that is finding number of fuzzy sets. Proposed computational method helps in finding intervals and thereby fuzzy sets for fuzzy time series forecasting. Central tendencies based fuzzy method is implemented on the authentic data for the enrollments of University of Alabama, which is considered as benchmark problem in the field of fuzzy time series. The forecasted values are compared with the results of other methods to state its supremacy. Projected computational method furnished promising results over other existing methods for the benchmark data. newline newline