Deep learning-based procedure compliance check system for nuclear power plant emergency operation

Abstract

Operating procedures are strictly followed in nuclear power plant operation. However, under a highly stressful condition such as emergency operation, human error probability can increase, with operators making mistakes in complying with the complex operating procedures. This paper proposes a procedure compliance check (PCC) system to monitor operator action and detect procedural deviation. If an operator action does not match the related procedural instruction, the PCC system notifies the operator in order to help them to recognize the mistake. A procedural logic process is constructed by referring to colored Petri nets. In situations requiring complex decisions, the PCC system employs a deep learning algorithm to predict operator judgement. The system was tested with data from a compact nuclear simulator, and demonstrated its potential to detect procedural noncompliance.