In the context of the rapid development of the current aquaculture, the precise regulation of dissolved oxygen levels is crucial to improve the efficiency of high-density farming. Therefore, this paper developed a Temporal Convolutional Network (TCN) -Gated Recurrent Unit (GRU) dissolved oxygen prediction model based on VMD decomposition. The model utilizes the advantages of convolutional neural network and recurrent neural network in processing multi-feature time series data to improve the efficiency of predicting the dynamic change of dissolved oxygen level. By comparing the current popular time series prediction models, such as convolutional neural network (CNN), long short-term memory network (LSTM), GRU, CNN-LSTM-attention, CNN-GRU-attention, etc., and it is concluded that VMD-TCN-GRU model has the best prediction effect, and MSE, RMSE and R2 are 0.006, 0.080 and 0.989, respectively. Therefore, it is concluded that VMD-TCN-GRU model is competent for the task of multi-feature prediction of dissolved oxygen in complex culture environment.