In the era of big data, AI has revolutionized recommendation tasks, particularly in healthcare, where precise medication recommendations are crucial for patient outcomes. Traditional methods struggle to interpret large EHR datasets, resulting in limited interpretability. While large-scale knowledge graph approaches improve interpretability, they are resource-intensive and often less accurate due to their complexity. To address these issues, we propose Med-RAIK, an interpretable model for medication recommendation. Med-RAIK integrates personalized and drug-to-drug interaction (DDI) knowledge graphs into a time-series model, enhancing both accuracy and interpretability. The model has three key components: a time-series module to capture temporal relationships in EHRs, a personalized knowledge graph module to dynamically update patient-specific insights, and a drug interaction module to incorporate critical DDI information for optimized medication selection. Additionally, by analyzing gradient values during backpropagation, Med-RAIK identifies significant medical features, further enhancing interpretability. Extensive experiments on the MIMIC-III and MIMIC-IV datasets demonstrate that Med-RAIK outperforms existing methods, offering accurate, safe, and interpretable medication recommendations.