An ensembled method for predicting dissolved oxygen level in aquaculture environment

Dissolved oxygen (DO) level is an important indicator aquaculture quality. This study proposes an ensembled method, WTD-GWO-SVR, combining wavelet threshold denoising (WTD), grey wolf optimization (GWO), and support vector regression (SVR) for accurately predicting DO levels. Addressing challenges such as high noise, poor data quality, and non-linearity and non-stationary properties of time series data, our method integrates SVR for regression-based estimation, WTD for data denoising, and GWO for optimizing the SVR parameters and the Gaussian kernel's radial basis function. We collected a dataset using a variety of low-cost sensors in a real aquaculture setting. Our comprehensive evaluation on the dataset demonstrates that WTD-GWO-SVR achieved mean squared error, mean absolute error, and R2 values of 0.38%, 3.81%, and 99.73%, respectively. It also consistently outperformed the back-propagation neural network and the long short-term memory model. It also achieved superior computational time performance compared to these methods. The high throughput and accuracy of WTD-GWO-SVR make it a potential choice for DO level prediction in water quality monitoring systems.