This paper proposes a Temporal Graph Contrastive Learning method for Sequential Recommendation (TGCL4SR) to better understand user interests and predict future behaviors. The method leverages both local interaction sequences and global temporal graphs to capture item correlations and analyze user behaviors over time. A Temporal Item Transition Graph (TITG) is constructed to represent interactions between items in a sequence, with timestamps and user information as edge attributes. The TITG is augmented using dual transformations based on neighbor sampling and time disturbance to handle noise and scale issues. A Temporal item Transition graph Convolutional network (TiTConv) is designed to capture temporal patterns in the TITG. A Temporal Graph Contrastive Learning (TGCL) mechanism is introduced to enhance representation uniformity between augmented graphs from identical sequences. For local sequences, a temporal sequence encoder is used to incorporate time interval embeddings into the Transformer architecture. The model is trained with cross-entropy and contrastive losses, along with Maximum Mean Discrepancy (MMD) loss to align item representations from global graphs and local sequences. Extensive experiments on four real-world datasets show that TGCL4SR outperforms state-of-the-art baselines in sequential recommendation tasks. The method effectively handles temporal information, reduces noise impact, and improves the accuracy of user interest prediction.This paper proposes a Temporal Graph Contrastive Learning method for Sequential Recommendation (TGCL4SR) to better understand user interests and predict future behaviors. The method leverages both local interaction sequences and global temporal graphs to capture item correlations and analyze user behaviors over time. A Temporal Item Transition Graph (TITG) is constructed to represent interactions between items in a sequence, with timestamps and user information as edge attributes. The TITG is augmented using dual transformations based on neighbor sampling and time disturbance to handle noise and scale issues. A Temporal item Transition graph Convolutional network (TiTConv) is designed to capture temporal patterns in the TITG. A Temporal Graph Contrastive Learning (TGCL) mechanism is introduced to enhance representation uniformity between augmented graphs from identical sequences. For local sequences, a temporal sequence encoder is used to incorporate time interval embeddings into the Transformer architecture. The model is trained with cross-entropy and contrastive losses, along with Maximum Mean Discrepancy (MMD) loss to align item representations from global graphs and local sequences. Extensive experiments on four real-world datasets show that TGCL4SR outperforms state-of-the-art baselines in sequential recommendation tasks. The method effectively handles temporal information, reduces noise impact, and improves the accuracy of user interest prediction.