In recent years, petroleum exploration has increasingly shifted toward deepwater and ultra-deep formations, with the most critical technical challenges being the narrow safe drilling window. Accurate wellbore pressure prediction is essential to prevent drilling hazards. Current prediction methods face significant limitations: physical models rely on idealized assumptions and empirical formulas with limited applicability, while machine learning approaches predominantly focus on single field prediction, neglecting multi-field coupling effects. To address these limitations, this paper innovatively proposes a coupled prediction method for wellbore pressure and temperature. The main contribution lies in integrating a customized gate control (CGC) multi-task framework with an efficient channel attention (ECA) optimized TCN-BiLSTM architecture. This framework employs ECA-optimized temporal convolutional networks (TCN) as expert modules to dynamically extract static and dynamic features, intelligently emphasizing key high-dimensional time-series channels. Simultaneously, customized gating units achieve fine-grained feature sharing, preserving positive transfer between temperature and pressure tasks while isolating negative transfer. Task-specific bidirectional long short-term memory (BiLSTM) networks are then utilized for the final coupled prediction. Extensive experimental results quantitatively demonstrate the modelʼs superiority. Compared to state-of-the-art single-task and physical models, the proposed method reduces the mean absolute percentage error (MAPE) by up to 94.98% for pressure and 95.28% for temperature, while improving the R2 metric by 26.84% and 20.38%, respectively. Furthermore, compared to traditional multi-task learning models, it reduces pressure and temperature MAPE by 63.05% and 71.88%, respectively, maintaining exceptional accuracy even during abrupt changes and intense drilling fluctuations. Validated in field applications at an oilfield, this approach provides a novel technical solution for precise prediction and coordinated monitoring of wellbore pressure and temperature during drilling operations.