Stream flow analysis and modelling using artificial intelligence techniques / Mohammed A. B. Seyam

Mohammed , A. B. Seyam (2016) Stream flow analysis and modelling using artificial intelligence techniques / Mohammed A. B. Seyam. PhD thesis, University of Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (1892Kb)
Preview	PDF (Thesis PhD) Download (11Mb) | Preview

Abstract

The reliable prediction of stream flow (SF) is an important aspect in the planning, design and management of surface water and rivers systems. This prediction can be performed using either process-based or data driven-based models (DDMs). Several modelling approaches fall under DDMs, such as statistical and artificial intelligence (AI) techniques. AI includes artificial neural networks (ANNs), support vector machines (SVM) and other techniques. The main goal of this research is to develop and employ a group of efficient AI-based models for predicting the real-time hourly stream flow (Q) in the downstream area of the Selangor River basin, taken here as the paradigm of humid tropical rivers in Southeast Asia. The Q of this river is yet to be subjected to prediction using AI. Despite intensive applications of monthly and daily SF prediction using AI over the last two decades, the prediction of Q is rare, particularly in small rivers in humid tropical regions, such as the Selangor River. The significance of this research lies in the uniqueness of the considered process and the novelty of the applied methodology in the modelling process. The performance of AI-based models can be improved through the integration of the hydrological description of SF in the modelling process through estimation of lag time (Lt) and analysis of the long-term changes of SF regimes which verified considerable changes may potentially result in increasing the probability of floods occurring in future. The integration process is essential to the selection of input and output variables of AIbased models and the lag intervals between them. The modelling process are performed in two phases to explore the possibility of improving the performance of AI-based models through the accurate timing of the model variables based on Lt estimation by two approaches, namely, the correlation coefficient and hydrological graphical approaches. Through the two modelling phases, four AI techniques, which include three types of ANNs, namely, the multi-layer perceptron network, radial basis function network, and generalized regression neural networks, along with SVM, are employed to develop six AI-based models to predict the Q. Three scenarios were employed to achieve six combinations of input variables, the first adopts RF and the second adopts WL while the third adopts both WL and RF as input variables. A total of 8753 patterns of Q, water level, and rainfall hourly records representing a one-year period (2011) were utilized in the modelling process. The performance evaluation of the developed AI-based models shows that high correlation coefficient (R) between the observed and predicted Q is achieved by most of the developed models. For example, R in SVM-M6 model is 0.992 and 0.953 for the training and testing data sets, respectively. The developed AI-based models were efficiently employed in some hydrological applications, such as Q prediction, analysis of the influence of both water level and rainfall on Q and estimation of the missing records of Q. They also were employed in flood early warning throughout the advanced detection of hydrological conditions that could lead to formations of floods.

Actions (For repository staff only : Login required)