An Oxygen Forecasting Strategy for Waterless Live Fish Transportation Based on IPSO-GRU Method

Abstract

To improve the prediction accuracy and efficiency for short-term oxygen concentration trends during the process of waterless live fish transportation, a kind of short-term oxygen consumption prediction model is proposed by using the gated recurrent unit (GRU) neural network, and its parameters are optimized by improved particle swarm optimization technology (IPSO). By comparing the prediction accuracy and efficiency of prediction fusion algorithms IPSO-GRU, IPSO-LSTM, GRU, and LSTM (Long Short-Term Memory), it is concluded that the time-series prediction model IPSO-GRU has higher predicting accuracy in short-term oxygen forecasting, and its efficiency has been significantly improved. Through experimental comparison, the accuracy of IPSO-GRU is improved about 45.9%, 9.16% when it compares with GRU by error criteria MAPE and RMSE respectively, and also improved about 54.1%,15.3% than LSTM. In addition, the time cost of IPSO-GRU is greatly reduced in a predicting operation when it compares with LSTM and GRU methods. Therefore, the method IPSO-GRU provides an effective prediction and early-warning functions for oxygen consumption prediction management during fish waterless keep-alive transportation.