Multi-model of ammonia nitrogen in aquaculture water based on EM algorithm.

This article focuses on developing a multi-model prediction method for ammonia nitrogen concentration in the intensive marine aquaculture of Stichopus japonicus, a valuable seafood species. Using water temperature, salinity, pH, and nitrite concentration as auxiliary variables, the study applies the Fuzzy C-means (FCM) clustering algorithm and Takagi-Sugeno (TS) fuzzy models, with parameters estimated via the Expectation Maximization (EM) algorithm, to construct and fuse fuzzy sub-models. Experimental data from Xinyulong Marine Biological Seed Technology Co., Ltd. in Dalian, China, demonstrate that this multi-model approach outperforms single-model methods such as radial basis function (RBF) neural networks, support vector machines (SVM), and stochastic configuration networks (SCN) in prediction accuracy. The findings suggest that the proposed method enables effective real-time monitoring of ammonia nitrogen, which is critical for maintaining water quality and improving aquaculture yield and health in Stichopus japonicus farming.