“Event” refers to a specific incident or occurrence that has a significant impact on human society and the natural world. Predicting such events helps reduce the risk of potential losses. Event prediction technology plays a vital role in ensuring safety, reducing risks, and controlling infectious diseases, among other aspects. Existing approaches to event prediction using temporal knowledge graphs face several key limitations. First, most studies model historical events as a discrete temporal point process to learn entity evolution, thereby overlooking the interactions among concurrent events that occur simultaneously. Second, while some methods attempt to capture interactions among concurrent events, they typically rely on simple static linear transformations to update node representations, failing to account for the richness of semantic information and the need to dynamically differentiate the importance of individual edges—even under the same relation type. Third, current prediction frameworks struggle to effectively integrate both temporal and semantic information from historical events, lacking a mechanism to dynamically perceive and weigh their relative importance during the reasoning process, especially for complex events. To overcome these shortages, we propose a temporal event prediction model (named D-FRGAT) that contains a third-level control mechanism to decay the temporal information, and integrates multi-relational attention networks and node feature adjustment techniques to capture the correlations among events over time better. Experimental results demonstrate that our method achieves superior performance on the MRR and ICEWS18 scores on five public datasets.