Based on water quality data from Litopenaeus vannamei aquaculture ponds in the same aquaculture farm during 2014-2018 and 2021-2024， this study selected key water quality parameters including total nitrogen （TN）， total phosphorus （TP）， active phosphorus （AP）， nitrate nitrogen （NO^-₃-N）， nitrite nitrogen （NO^-₂-N）， total ammonia nitrogen （TAN）， chemical oxygen demand （COD）， temperature （T）， and pH values to develop water quality prediction models for TAN and COD using temporal convolutional network （TCN） and gated recurrent unit （GRU）. A hybrid FC-TCN-GRU model architecture was constructed， which employed TCN for feature extraction and dimensionality reduction of data features， fed the processed data into GRU， and finally maped the results through fully connected layers （FC） to generate predictions. Mean absolute error （MAE）， mean squared error （MSE）， and coefficient of determination （R²） values of the FC-TCN-GRU model for TAN prediction were 0.255， 0.089 and 0.861， respectively， while achieved 1.750， 4.840 and 0.332 for COD prediction. Compared with PCA-LSTM， basic LSTM and basic GRU models， the FC-TCN-GRU model showed better predictive accuracy for both TAN and COD prediction. The model performs superior in TAN prediction， but it still needs improvement in COD prediction.