Abstract: Abstract: Tunnel-ventilated poultry houses have been widely used for the breeding industry in recent years. However, it is difficult to accurately predict online the room temperature of tunnel-ventilated poultry houses, due to the variations in the natural environment and the age of the poultry. This study aims to accurately predict the room temperature of poultry houses for the high accuracy of temperature regulation. A PCA-LM-NARX method was also proposed to construct the online prediction model for the room temperature of poultry houses. The resulting NARX neural network model was used to accurately predict the short-term indoor temperature, according to the recent environmental data of poultry houses. The environmental variables were selected using the PCA-LM-NARX method, according to the energy balance equation of the temperature system in the poultry house. Principal component analysis (PCA) was also utilized to screen out the key influencing factors on the indoor temperature of the poultry house from multiple environmental variables. A prediction model of room temperature was constructed with a NARX neural network structure. The LM algorithm was also used to optimize the model parameters. The hysteresis characteristics of indoor temperature were considered in poultry houses, due mainly to the heat conduction delay, air circulation delay, control equipment adjustment delay, and evaporative cooling. The optimal delay order of the NARX neural network was designed using the Bayesian information criterion. The normalization and rolling statistical methods were used to preprocess the measured data. The PCA-LM-NARX method was modified suitable for poultry of different ages and environmental changes during modeling. Furthermore, the moving window was used to remove the previous data with a large time span from the current moment in the neural network training set, and then the new data was added with a short time span from the current moment. The real-time updates were achieved on the training set, thereby reducing the negative impact of previous data on the prediction performance of the model. The modeling program was divided into two parts: offline modeling and online prediction. The offline modeling was installed on the host computer, in order to update the parameters of the prediction model in real-time using the data in the moving window. Once the LM algorithm met the requirements of accuracy, the optimal parameters were transmitted to the field controller through the fieldbus network. The online prediction program and room temperature prediction model were stored in the on-site controller for the online prediction of poultry house temperature. The field controller was used to transmit the optimal training dataset to the host computer through the fieldbus network, in order to update the parameters of the prediction model. Experimental results show that the prediction model with the PCA-LM-NARX method can be expected to predict the room temperature of the poultry house in the next 5, 15, and 30 min online, with the mean squared error (MSE) of 0.0220, 0.0472 and 0.0779 ℃², respectively. The average modeling duration of the running program on the i5-12500H CPU was 0.3321s, indicating the rapid real-time update of model parameters. Therefore, the PCA-LM-NARX method can be used to construct a high-precision prediction model for the room temperature of a tunnel-ventilated poultry house, in order to achieve the rapid online update of model parameters.