Date Published: February 21, 2018
Publisher: Public Library of Science
Author(s): Yingyi Chen, Huihui Yu, Yanjun Cheng, Qianqian Cheng, Daoliang Li, Manabu Sakakibara.
A precise predictive model is important for obtaining a clear understanding of the changes in dissolved oxygen content in crab ponds. Highly accurate interval forecasting of dissolved oxygen content is fundamental to reduce risk, and three-dimensional prediction can provide more accurate results and overall guidance. In this study, a hybrid three-dimensional (3D) dissolved oxygen content prediction model based on a radial basis function (RBF) neural network, K-means and subtractive clustering was developed and named the subtractive clustering (SC)-K-means-RBF model. In this modeling process, K-means and subtractive clustering methods were employed to enhance the hyperparameters required in the RBF neural network model. The comparison of the predicted results of different traditional models validated the effectiveness and accuracy of the proposed hybrid SC-K-means-RBF model for three-dimensional prediction of dissolved oxygen content. Consequently, the proposed model can effectively display the three-dimensional distribution of dissolved oxygen content and serve as a guide for feeding and future studies.
Dissolved oxygen content plays a vital role in aquatic ecosystems because it has a substantial influence on water quality management, feed consumption, and energy expenditure . Proper control and management of dissolved oxygen content in crab pond aquaculture is crucial for the developing crabs and has a significant impact on the quality and quantity of the final product . An inappropriate dissolved oxygen content can cause crab hypoxia and even more severe disease . Therefore, establishing an efficient and accurate model for predicting dissolved oxygen content in crab aquaculture is important to provide a basis for water quality control and reduce aquaculture risk. Liu et al. and Yu et al. built dissolved oxygen content prediction models based on the machine learning method, which achieves time dimension forecasting without considering three-dimensional prediction [2, 3]. However, a three-dimensional prediction model for dissolved oxygen content can reveal changing trends and provide guidance for aquaculture. Therefore, in this study, we propose a hybrid three-dimensional (3D) dissolved oxygen content prediction model that can achieve more accurate results and overall guidance for dissolved oxygen content in crab ponds.
The SC-K-means-RBF model is a three-dimensional spatial interpolation method based on RBF incorporating subtractive clustering and K-means for dissolved oxygen content prediction and display. The method can identify changing trends and provide guidance for aquaculture. The SC-K-means-RBF interpolation method, which has a clear principle and simple structure, provides a simple method for 3D modeling the spatial distribution of dissolved oxygen content in an aquaculture pond. Combining the subtractive clustering method and K-means clustering method increases the accuracy of obtaining the number of hidden units for the RBF neural network. The results illustrate the validity of the proposed SC-K-means-RBF interpolation method by comparing the RMSE and MAE values of the proposed method with the standard RBF interpolation method. For example, the standard RBF interpolation achieves RMSE values from 0.5344 to 0.7022 with different hidden units, and the proposed SC-K-means-RBF interpolation method achieves lower RMSE values. The comparison of the prediction results of different traditional models validated the effectiveness and accuracy of the proposed hybrid SC-K-means-RBF model for the three-dimensional prediction of dissolved oxygen content.