In the domain of industrial control, supervisory control and data acquisition (SCADA) systems are essential for real-time monitoring and efficient data acquisition. However, as industrial systems grow in scale and complexity, conventional tag configuration methods face challenges in balancing precision and operational efficiency. Addressing these challenges requires innovative solutions. The rapid evolution of generative artificial intelligence, particularly large language models (LLMs), offers a transformative approach. This study introduces a structured prompt optimization strategy, termed structured tag engineering prompt (STEP), to increase the ability of LLMs to generate high-quality tag files. To validate the STEP method, we assessed five mainstream LLMs on basic tag generation tasks via the CodeBERTScore and pass@k metrics. The results revealed that performance of all models has been improved, thus validating the effectiveness of the proposed optimization method. On the basis of these findings, a tag generation framework grounded in the STEP method was developed and validated through case studies and practical industrial scenarios. These validations confirmed the STEP method's applicability, demonstrating its value and potential to advance prompt engineering for SCADA systems. In summary, this study contributes to the automation and intelligence of industrial control systems while providing unique insights through the application of LLMs combined with prompt engineering in addressing complex industrial tasks.