A Method of Rainfall Runoff Forecasting Based on Deep Convolution Neural Networks

• Published in: 2018 Sixth International Conference on Advanced Cloud and Big Data (CBD) pp. 304 - 310
• Main Authors: Li, Xiaoli, Du, Zhenlong, Song, Guomei
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.08.2018
• Subjects: Data models; deep convolutional belief network; Forecasting; Mathematical model; Neurons; prediction; Predictive models; rainfall runoff; Rivers
• DOI: 10.1109/CBD.2018.00061

The prediction of rainfall runoff is an ordinary function in hy-drological information process. As it bears the strong locality and nonlinearity, accurate prediction is challenging. In the pa-per a novel approach of rainfall runoff prediction based on con-volutional deep belief networks is proposed. The constructed deep learning machine better simulates the complex nonlinearity within data. Even if observation values are limited, it still main-tains very good prediction capability. The proposed model is tes-tified and validated in the Luo River Basin (Guangdong Prov-ince, China) for training and testing the prediction performance. At the same time, the traditional Xinanjiang rainfall runoff model was introduced to evaluate and compare results with the ones, made by the new model. Moreover, multiple forecasts (e.g. 1-day, 3-day or 5-day) achieved to demonstrate better model performance. The results prove that the currently proposed model could predict the runoff more accurately than the Xinan-jiang model.