Knowledge-Driven Hallucination in Large Language Models: An Empirical Study on Process Modeling