Procedure2Command: an AI-based Nuclear Power Plant Control Command Code Generation Prototype System

Nuclear power plant control systems are complex and require extremely high safety. Safety operation procedure flowcharts drawn by experts with many pages are often used as the guidebooks to help human operators avoid improper operations. To help operators better use these flowcharts and reduce their mental burden, here we propose a control command code generation prototype system, which involves artificial intelligence technologies, especially natural language processing, to automatically translate the nuclear power plant safety operation procedure flowcharts into executable codes. The proposed system includes three components: a flowchart decomposition tool, a text-code translation dataset, and a deep-learning based code generation model. Experimental results show that our proposed method can generate command codes with comparable accuracy as manual translation in a much higher inference speed.