AUTHOR=Li  Dashe , Zhang  Xuan 

TITLE=Utilizing a Two-Dimensional Data-Driven Convolutional Neural Network for Long-Term Prediction of Dissolved Oxygen Content

JOURNAL=Frontiers in Environmental Science

VOLUME=Volume 10 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/environmental-science/articles/10.3389/fenvs.2022.904939

DOI=10.3389/fenvs.2022.904939

ISSN=2296-665X

ABSTRACT=The prediction of water quality is a basic and preventive measure for marine pasture fisheries. Dissolved oxygen (DO) is an important water quality parameter that directly affects the survival and growth of fish. Because of the uncertainty of the breeding environment, DO trends have the characters of complex nonlinear. So studies to improve the prediction accuracy of DO have always been popular. Based on this, a two-dimensional data-driven convolutional neural network model (2DD-CNN) is proposed to provide reference for monitoring the marine pasture water environment and providing early warning alerts. To fill missing values, we first propose a novel 
sequence score matching-filling (SSMF) algorithm based on matching sequences with similar historical data features. For converting a DO sequence into two-dimensional images, a means is put forward by using RGB image transformation. Then, the self-attention mechanism is applied to construct convolutional neural network (CNN) for solving local perception problems.Finally, DO samples from multiple marine pastures are validated and compared with model predictions. The results show that 2DD-CNN can efficiently extract DO characteristics, thus effectively improving the prediction accuracy of DO and providing good generalization ability.