A Comparative Study of LSTM, GRU, BiLSTM and BiGRU to Predict Dissolved Oxygen

Abstract

In aquaculture, dissolved oxygen (DO) levels affect fish growth and survival. Automated monitoring and prediction of DO is challenging and becomes expensive if unnecessary sensors are used. This study aims to identify the optimal water and environmental parameters for DO prediction. Data from the fishpond station of Rajabhat Rajanagarindra University were pre-processed and used for training using LSTM, GRU, BiLSTM, and BiGRU. The performance of the models was evaluated and contrasted using three error measures. The results showed that GRU gave the best performance compared to the other models. In conclusion, the best parameters for DO prediction are water pH and water temperature.