With the swift advancement of artificial intelligence technology, large-scale language models have demonstrated substantial potential in the teaching plan generation of education field. Nonetheless, current models still encounter challenges such as insufficient granularity of content, lack of expertise, and a low degree of structure in the task of generating lesson plans. This research introduces a fine-tuned EDU-GPT model based on logical chains and prompts, employing a dual-track parallel enhancement mechanism to amalgamate a high-quality knowledge base, thereby achieving high-quality and one-time generation of lesson plans. Experiment results reveal that the EDU-GPT significantly surpasses the GPT-4 baseline model in terms of granularity (8.90 points), accuracy (8.84 points), and structure (8.70 points) with a statistical significance (p < 0.001), offering a novel solution for the intelligent generation of lesson plans.