River Water Level Forecasting with Adaptive ARIMA and Extreme Learning Machine Models

Abstract

This dissertation delves into the predictive capabilities of two prominent modelling techniques, Extreme Learning Machine window (ELM) and the Adaptive Autoregressive Integrated Moving Average (ARIMA), for short-term forecasting of river water levels. With increasing environmental uncertainties, accurate predictions of water levels are crucial for effective water management and flood prevention. Through rigorous data processing and model training, this research employs recent river data to evaluate the performance of both models over four forecasting horizons: 1-day, 3-day, 5-day, and 7-day. The evaluation metrics, including Root Mean Squared Errors (RMSE), Mean Absolute Deviation (MAD), and Mean Squared Errors (MSE), revealed insightful patterns about the accuracy and reliability of each model. Further, the distribution of forecast errors was analysed to understand the consistency and potential biases in predictions. The thesis findings indicate nuanced differences in the performance of Adaptive ELM and Adaptive ARIMA, shedding light on the specific conditions and scenarios where one model may outperform the other. This comparative analysis serves as a comprehensive guide for researchers and practitioners in selecting the most suitable model for river water level forecasting under varying circumstances. The insights from this study also pave the way for future research opportunities, exploring the integration of both models or the incorporation of additional data sources to enhance forecasting accuracy.