Prediction of Dissolved Oxygen Content in Aquaculture Based on Clustering and Improved ELM

In the aquaculture industry, dissolved oxygen is an important water quality parameter index. The prediction of dissolved oxygen can reduce the operation cost of aquatic product management to a certain extent. In this paper, a hybrid method is proposed to predict the change of dissolved oxygen from the perspective of time series in aquaculture, which based on k-means clustering and improved Softplus extreme learning machine (SELM) with particle swarm optimization (PSO). We use k-means algorithm to divide the dataset into several clusters by calculating the similarity among variables, to find the periodic change rule and trend of variables. Softplus is employed as the activation function of ELM to make the model closer to the biological activation model. Meanwhile, partial least square (PLS) method is utilized to solve the strong collinearity among variables. In addition, we introduce PSO algorithm to optimize the model parameters. The experimental results show that our model can achieve better prediction performance and accuracy of prediction compared with other single models. Compared with the counterpart model, the improved model can tolerate some data loss and uncertain outliers of sensor time series. Our work provides an accurate predictive model framework for researchers to track dissolved oxygen.