In intensive sheep farming, the lack and backwardness of environmental management technologies are key factors contributing to the deterioration of sheep house environments. Accurately predicting changes in sheep house environmental parameters are crucial for ensuring the healthy growth of sheep and improving the economic benefits of the sheep farming industry. To accurately understand the PM?.? concentration patterns within sheep houses, the wavelet transform (WT) was used to decompose and reconstruct sheep house environmental parameter data to eliminate data noise. The sparrow search algorithm (SSA) was then used to optimize the number of hidden layer neurons, learning rate, and batch size of the LSTM model. This approach also adjusted the input model parameters to avoid randomness in parameter selection and further improve model performance. Experimental results showed that the WT-SSA-LSTM model outperformed other prediction models in all metrics, with MAE, RMSE, MSE, NRMSE, and R2 reaching 0.3497 μg/m3, 0.6004 μg/m3, 0.3605 μg2/m?, 0.0057, and 0.9981, respectively. This demonstrated the high accuracy and stability of the proposed WT-SSA-LSTM prediction model, effectively providing guidance for monitoring and regulating PM?.? levels in intensive sheep farming facilities.