Dissolved Oxygen Prediction Based on PCA-LSTM

Abstract Dissolved oxygen in surface water is an essential assessment of water quality. Predicting the concentration of dissolved oxygen in a basin is essential for the integrated use of water resources and the prevention and control of water pollution, as it enables the prediction of water quality trends in advance. In accordance with the cyclical and non-linear character of the variation of dissolved oxygen, we present a PCA-LSTM combined with a principal component analysis (PCA) method and a long and short-term memory network (LSTM) to estimate the dissolved oxygen concentration in the short period of time. Firstly, the dissolved oxygen data extracted from the water quality monitoring platform were preprocessed, and then 8 external variables, which retained most of the information, were converted into 5 new variables and put into the LSTM network for training. Finally, the predictions of the pre-processed training set data were compared using both the LSTM and PCA-LSTM models respectively. Experiments demonstrate that the PCA-LSTM model not only simplifies the structure of the proposed network, but also has more accurate prediction results than the conventional LSTM. Its mean absolute errors as well as the mean squared errors are improved by 2.71% and 9.03% respectively compared with the traditional LSTM model.