Addressing student problem behaviors, which profoundly affect learning environments and students’ long-term development, remains a critical challenge in K-12 education, prompting educators to seek AI-powered educational counseling systems. However, the practical adoption of these systems is hindered by insufficient grounding in educational expertise and opaque decision-making processes. This paper presents an explainable educational counseling system based on a large language model that leverages retrieval-augmented generation to ground AI-generated responses in educational expertise while explaining whether strategies are case-supported or model-derived. Through technical evaluation, our system achieves 88.7% accuracy in distinguishing strategy origins, substantially outperforming baseline approaches. A preliminary controlled experiment with 20 pre-service teachers further demonstrates that these explanations significantly enhance their trust in the AI-generated responses. This research provides a practical solution to building explainable educational counseling systems that ultimately support teachers in addressing student problem behaviors.