<p>Industrial twin screw refrigeration compressor (TSRC) oil temperature is directly related to the compressor’s efficient, stable operation and fault prevention. In a refrigeration system, the lubricating oil plays a pivotal role by reducing rotor friction and sealing the compression chamber. Abnormal oil temperature can trigger a chain of failures such as excessive wear, refrigerant leakage, and blockages in the lubrication system. TSRC systems are complex, high-dimensional, and dynamically changing, with strongly nonlinear behavior. To address these challenges, this study proposes a novel temporal convolutional network (TCN) architecture with dual-attention mechanisms to improve forecasting performance in this domain. The model integrates a self-attention mechanism with a Gaussian decay function to prioritize temporally relevant observations while gradually diminishing the influence of distant past data and incorporates a squeeze-and-excitation (SE) module to dynamically recalibrate the importance of each feature channel. These enhancements enable the TCN to capture long-range temporal dependencies and key feature interactions more effectively. The results indicate that R² improved by 0.8%. The MSE, RMSE, and MAE are reduced by 13.3%, 6.9%, and 7.6%, respectively. This offers a novel and adaptable solution for maintenance in industrial refrigeration systems.</p>