Significant progress has been achieved in temporal knowledge graph (TKG) forecasting in recent years, particularly in modeling historical information to forecast repetitive or cyclic facts. Nevertheless, existing models struggle to forecast facts involving non-historical information that have not interacted with the query entity. To address this challenge, a TKG forecasting architecture based on Historical and Non-Historical Enhanced ContrastiveLearning (HNECL) is proposed. HNECL constructs a non-historical information encoder and integrates non-historical information with global and local historical information for multi-contrastive learning. It is entities semantically relevant with structurally unconnected that are forecast by the architecture. Tests on four public datasets show that HNECL works better than the best baseline models.