Artificial intelligence (AI) models have been successfully applied in modeling engineering problems, including civil, water resources, electrical, and structure. The originality of the presented chapter is to investigate a non-tuned machine learning algorithm, called self-adaptive evolutionary extreme learning machine (SaE-ELM), to formulate an expert prediction model. The targeted application of the SaE-ELM is the prediction of river water level. Developing such water level prediction and monitoring models are crucial optimization tasks in water resources management and flood prediction. The aims of this chapter are (1) to conduct a comprehensive survey for AI models in water level modeling, (2) to apply a relatively new ML algorithm (i.e., SaE-ELM) for modeling water level, (3) to examine two different time scales (e.g., daily and monthly), and (4) to compare the inspected model with the extreme learning machine (ELM) model for validation. In conclusion, the contribution of the current chapter produced an expert and highly optimized predictive model that can yield a high-performance accuracy.
Top1. Introduction
In this chapter, we develop highly novel machine learning (ML) models as optimal tools for water resource forecasting. ML approaches have been broadly researched using soft computing engineering principles (Coppin, 2004; Ford, 1987; Gevarter, 1987; Noureldin, El-Shafie, & Bayoumi, 2011; Russell & Norvig, 2010). The most common ML models are designed by artificial intelligence (AI) techniques identify the nonlinear, dynamic, nonstationary relationships min predictor data for both regression and classification problems (Nourani, Hosseini Baghanam, Adamowski, & Kisi, 2014; Yaseen, El-shafie, Jaafar, Afan, & Sayl, 2015). Advantageous mechanism of soft computing-based AI techniques have led to several applications of science and engineering optimization problems (e.g., sediment transport, evaporation rate, stream-flow, river water quality, ground water and etc.) (Nourani, Hosseini Baghanam, Adamowski, & Kisi, 2014; Yaseen, El-shafie, Jaafar, Afan, & Sayl, 2015). Generally, machine learning utilizing data-driven methodologies are designed with artificial neural network, fuzzy set, evolutionary computing to solve complex computational problems in science and engineering. In 1994, Zadeh coined the term soft computing and defined it as:
collection of methodologies that aim to exploit the tolerance for imprecision and uncertainty to achieve tractability, robustness, and low solution cost.
Based on Zadeh’s opinion, that we live in a pervasively imprecise and uncertain world and that precision and certainty carry a cost. Therefore, soft computing should be seen as a partnership of distinct optimization methods, rather than as a homogeneous body of concepts and techniques.
The water level sector has been explored by several AI techniques (Alvisi & Franchini, 2011; F. J. Chang & Chang, 2006; Elhatip & Kömür, 2008; Fallah-Mehdipour, Bozorg Haddad, & Mariño, 2013; Wei, 2012). Its history for water level modeling back to 1998 with the earliest research accomplished in water level forecasting (Thirumalaiah & Deo, 1998), who conducted real-time streamflow stage using artificial neural network. The results showed that modelling low flow superior the high value of forecasting. Campolo et al. (1999) studied river flood forecasting based on rainfall and water level variables using ANN model. The main goal of their research was to build a predictive model by including climatological information. The modelling of hourly basis showed remarkable results of accuracies. Liong et al. (2000) demonstrated the application of ANN in forecasting water level stage at Dhaka, Bangladesh. The input variables were investigated up to seven days’ lead time to forecast on step ahead water level. The authors revealed that the results made a desirable advanced warning forecasting tool. In 2002, the application of Support Vector Machines have been implemented for forecasting flood stage by (S. Y. Liong & Sivapragasam, 2002). The results of SVM were compared with that of ANN based on the antecedent records of water level “one-lead day to seven-lead day forecasting”. The improvements in maximum predicted water level errors by SVM over ANN for four-lead day to seven-lead day are 9.6 cm, 22.6 cm, 4.9 cm and 15.7 cm, respectively. Clearly, these studies showed the relevance of AI techniques for optimization of forecasting methodologies in area of water resources.