Water quality prediction method based on preferred classification

Abstract

Water quality monitoring and prediction are important parts of Cyber Physical Systems. Considering the complexity, diversity, and strong non-linearity of water quality data, a single water quality prediction model is difficult to have a significant effect on different data. To solve this problem, a new water quality prediction method based on the preferred classification is proposed in this study. A preferred classifier is established to integrate back propagation neural network, support vector machines for regression and long short-term memory due to the fact that these three prediction models can take into account the different characteristics of water quality data. When new data input, the proposed method preferentially selects the prediction model that is most suitable for the data, and then uses the selected model for prediction. Finally, the proposed method is applied in two actual datasets: Songhua River and Victoria Bay. Experimental results demonstrate that the water quality prediction method based on preferred classification achieves better performance than any of the three single prediction models.