CRML: A Convolution Regression Model With Machine Learning for Hydrology Forecasting

• Main Authors: Chen Chen, Qiang Hui, Qingqi Pei, Yang Zhou, Bin Wang, Ning Lv, Ji Li
• Format: Article
• Language: English
• Published: IEEE 2019-01-01
• Series: IEEE Access
• Subjects: Hydrology forecasting; convolution regression algorithm; flood process selection algorithm; gradient descent; machine learning
• Online Access: https://ieeexplore.ieee.org/document/8836461/

Hydrologic disasters often result in substantial property damage and casualties. Therefore, hydrology forecasting, especially the flooding, has become a hot research spot in all countries of the world. Based on the basic principle of flooding formation, this paper proposes a data-driven hydrology forecasting model, i.e., the CRML (Convolution Regression based on Machine Learning). This model could reflect the impact of hourly rainfall on the future flow changes and the flow changes are predicted by superimposing these impacts. First, our work is implemented on historical data onto the Xixian River Basin in Henan Province, China. Through the data filtering, the training set of our model is constructed by using the flood process selection algorithm proposed in this paper. Next, the gradient descent algorithm is used to update the weights of the model, and the optimal weights are verified by ten flooding events generated in the past ten years. Finally, the numerical results show that the qualified rates of our model in predicting flood peak flow and its arrival time is approximately 90% and 100%, respectively. Compared with the latest popular artificial intelligence schemes, our model structure is clear and concise. And combined with the physical meaning of the traditional model and machine learning technology, our model can accurately complete the task of long and short lead time hydrology forecasting.