With the explosive growth of big data, efficiently mining valuable patterns from dynamic historical data has become a core challenge. Temporal Knowledge Graphs (TKGs), due to their ability to integrate semantic information with temporal dynamics, have become a key representation for handling complex historical graph-structured data. However, current TKG prediction methods face three major bottlenecks: (1) Insufficient interpretability: Prediction mechanisms lack transparency, making it difficult to gain trust in real-world applications; (2) Separated modeling of patterns: Periodic and causal patterns are modeled independently, failing to fully capture dynamic regularities; (3) Low Computational Efficiency: TKGs typically contain millions of edges, and existing models are limited by the lengthy time required for path extraction and training, severely restricting their practical utility. To address these bottlenecks, this paper proposes the Periodic-Causal Relation Path (PCRP) framework. The framework uses paths as the basic modeling unit, explicitly unifying the representation of periodicity and causality to enhance reasoning interpretability. An entity frequency weighting strategy is introduced to deeply integrate paths, enabling accurate capture of evolutionary patterns. For efficiency optimization, the graph structure is divided into multiple subgraphs according to timestamps, with multiple nodes assigned to independently extract paths and perform training within subgraphs using DistributedDataParallel (DDP), and AllReduce is employed to ensure consistent model optimization across nodes. Experimental results show that PCRP significantly improves prediction accuracy and efficiency on multiple benchmark datasets, demonstrating its superior large-scale data processing capability and application potential.