Dissolved oxygen prediction using regularized extreme learning machine with clustering mechanism in a black bass aquaculture pond

Abstract

Dissolved oxygen (DO) is an important indicator of aquaculture water quality. The prediction accuracy of DO content is the key role in managing and controlling aquaculture water quality. However, potential trends of DO under various conditions (such as weather) are always overlooked. This study aims to develop a novel DO forecasting model using the optimized regularized extreme learning machine (RELM) with factor extraction operation and K-medoids clustering strategy in a black bass aquaculture pond. We adopt the leave-one-out cross (LOO) error validation to obtain the optimal regularization parameter of RELM and enhance the forecasting accuracy. We further adjust the activation function to accelerate the RELM. Next, we divide the time series into day and night segments, and construct the clustering mechanism with the K-medoids method to extract the different patterns of data streams under various weather conditions. The experiments on 14 days’ data from a real-world aquaculture pond demonstrate the efficiency and accuracy of our proposed DO prediction model. We believe that our research will facilitate the development of a forecasting tool for warning hypoxia in the near future, which combines intelligent prediction models and real-time data.