Multisensor monitoring and water quality prediction for live ornamental fish transportation based on artificial neural network

Abstract

The microenvironment of live ornamental fish transportation is a significant source of fish mortality. The transportation time and density of fish have a significant impact on water quality. The previous studies measured water quality parameters with traditional and non-real-time methods, which cannot give the abrupt changing patterns during live transportation of ornamental fish. In this study, water quality key parameters in the microenvironment of goldfish were monitored using a multisensor box during the simulation transportation experiment. According to the findings, nitrite was not identified, and pH remained within the acceptable goldfish range even after 48 h of monitoring, indicating that these factors did not affect the goldfish physiology during commercial transport. The collected data were correlated with physiological health and fish behaviour to identify the most impactful parameters. Data sets went through a data screening process (data correction, filtering); ammonia nitrogen and dissolved oxygen fed into the multilayer neural network and regression analysis, with time and density as input variables. The neural network successfully predicts the dissolved oxygen and ammonia nitrogen with mean absolute error (MAE) of 0.2306 and 0.00775. The regression model achieved good prediction accuracy for only ammonia nitrogen, with a mean absolute error (MAE) of 0.00484 and a relative average error of 4.58%. However, for future studies, the prediction model must account that the distinct physiology of the transported fish will result in distinct changes in water quality.