A Deep-Learning Based Precipitation Forecasting Approach Using Multiple Environmental Factors

• Published in: 2017 IEEE International Congress on Big Data (BigData Congress) pp. 193 - 200
• Main Authors: Pengcheng Zhang, Lei Zhang, Leung, Hareton, Jimin Wang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.06.2017
• Subjects: Big Data; Biological neural networks; Data mining; Data models; Deep-learning; Forecasting; multiple time series; Precipitation forecasting; Predictive models; Time series analysis
• DOI: 10.1109/BigDataCongress.2017.34

Precise precipitation forecasting can better reflect the changing trend of climate and also provide timely and efficient environmental information for management decision, as well as prevent the occurrence of floods or droughts. In the era of big data, this paper proposes a novel approach for precipitation forecasting based on deep belief nets, called DBNPF (Deep Belief Network for Precipitation Forecast). Through simulating neural connecting structure of human brain, Gaussian kernel function for data conversion, and back-propagation network for fine-tuning the entire network, the features of the data in the original space are mapped into the new feature space with semantic feature through the dimensionality reduction. The proposed approach can not only learn the hierarchical representation of raw data using a highly generalized way, fully mining the information hidden in the original data, but also make a more accurate description of the rule underlying the different kind of time series of big data. Seven kinds of environmental factors, which are very closely related to precipitation, are used as input vector, and the next 24 hours precipitation is used as the output vector. A set of dedicated experiments with data from Zunyi area of Guizhou Province is conducted to validate the feasibility and useability of the model. We also compare the deep-learning based approach with other traditional machine learning approaches. The experimental results show that the proposed approach can improve the precision of precipitation forecasting.