Latent Variable Structured Bayesian Network for Cyanobacterial Risk Pre-control

Cyanobacterial blooms increasingly pose threats to ecosystems and human health. This paper is aimed to propose systematic risk pre-control schemes by understanding the complex causalities between cyanobacteria and multiple influencing variables. This research remains a challenge for three reasons. Firstly, the time-series evolution of cyanobacteria is characterized by deep uncertainties and nonlinear dynamics. Secondly, latent variables with hidden information usually exist in this kind of complex aquatic system. Thirdly, it is difficult to identify an efficient pre-control scheme that specifies variables for preferential regulation. To address these problems, we propose a latent variable structured Bayesian network model and a corresponding parameter learning algorithm. The model is tested by real-time spatio-temporal data. The computational results reveal that the proposed model demonstrates better performance in terms of inference accuracy and degree of system understanding. Based on sensitivity analysis and combination-effect analysis, a systematic risk pre-control scheme is proposed for decision-makers to prevent cyanobacterial blooms under the scenario of global warming.