A Contrastive Learning Based CNN-GRU Model for Time Series and Its Application for Water Quality Prediction
Yahong Lian, Jing Yun, Yun-Cheng Wang, Zhiwei Xu · 2022
Water quality monitoring is of great importance in sustainable development of intelligence city. One of the main challenge of this data is that the lack of labeled data. Therefore, the unsupervised representative framework which can produce reliable results could be of great value. It is key to applying contrastive learning, which can be used in environmental data, where there exists lots of unlabeled sequential data. This work proposes a contrastive learning based CNN -GRU model which exploits a data augmentation scheme in which new instances are generated by mixing two data instances with a mixing component and utilize CNN and GRU model to extract more useful representation from time series data. To show the effectiveness, we conduct comparative experiments on groups of time series dataset, and apply our model to a real-world water quality prediction task where the data comes from Erdaohezi, Tongliao, China.