Abstract: China had the largest meat duck industry, accounting for over 82% of the global total slaughter volume of meat ducks. The air temperature in the meat duck breeding environment was crucial to the comfort and survival of the ducks. The air temperature was susceptible to multiple factors such as relative humidity and illumination intensity. Therefore, it was significant to grasp timely and accurately the change of the air temperature which could improve the quality and economic returns, and optimize breeding management. Grasping the trend of the air temperature value timely and accurately was the key to the high-density healthy breeding in duck houses. However, traditional temperature prediction methods had problems such as low prediction accuracy, poor robustness, and poor generalization ability, etc.To achieve accurate prediction of duck-house temperature, this study proposed a hybrid 1DCNN-BiLSTM-DQN model integrating a one-dimensional convolutional neural network (1DCNN), a bidirectional long short-term memory network (BiLSTM), and a deep Q-network (DQN).During the model construction, considering that the temperature time series signal contained multi-scale variation characteristics, the signal was decomposed into high-frequency and low-frequency components via the discrete Fourier transform (DFT).Given that the high-frequency component mainly reflected short-term fluctuations, the 1DCNN was used to extract local features from the high-frequency component; whereas the low-frequency component mainly reflected long-term fluctuations, the BiLSTM was used to extract long-sequence dependency features from the low-frequency component.Subsequently, the two sets of features were fused using a concatenation model, and finally the temperature prediction value was obtained through the mapping of a fully connected layer.Based on this, the DQN algorithm was introduced to construct an agent for iterative optimization of the model hyperparameters. The agent established an adaptive optimization mechanism with an 8-dimensional state space and a 12-dimensional action space, enabling dynamic optimization of key hyperparameters such as the learning rate, number of hidden layers, dropout rate, and network architecture, thereby improving the prediction robustness of the model under scenarios of seasonal transitions and extreme weather.The indoor and outdoor temperature data of net-raised duck houses were collected, and the model was used to predict the indoor and outdoor temperatures of the net-raised duck houses.The proposed model was used to predict the indoor and outdoor temperature variations of net-raised duck houses in Gaoyou City, Yangzhou City, Jiangsu Province, from March 22, 2025, to March 22, 2026. The results demonstrate that the 1DCNN-BiLSTM-DQN model achieves a coefficient of determination (R²) of 0.993 with an optimal input step size of 48.Its metrics, including MAE, and RMSE, are superior to those of traditional models such as the Temporal Convolutional Network and Transformer.Specifically, Under different weather conditions, compared to the fixed-hyperparameter 1DCNN-BiLSTM model, the 1DCNN-BiLSTM-DQN model exhibits the following improvements: on cloudy days, the MAE decreases from 0.29℃ to 0.18℃, the RMSE decreases from 0.36℃ to 0.23℃, and the R² increases from 0.77 to 0.91; on sunny days, the MAE decreases from 0.48℃ to 0.23℃, the RMSE decreases from 0.59℃ to 0.31℃, and the R² increases from 0.79 to 0.94; on rainy days, the MAE decreases from 0.54℃ to 0.33℃, the RMSE decreases from 0.70℃ to 0.46℃, and the R² increases from 0.72 to 0.89.The experimental results show that the proposed 1DCNN-BiLSTM combined prediction model achieves better prediction performance compared to traditional models such as BiLSTM, 1DCNN, TCN, and Transformer.In summary, the 1DCNN-BiLSTM-DQN model proposed in this study can predict the temperature in duck houses, providing data support for early environmental regulation and reducing the risk of environmental stress in meat ducks.