There are five operating modes of Nuclear Power Plants (NPPs): refueling, startup, low power, normal power, and shutdown. In these operating modes, the startup and the shutdown operating modes of NPPs are completely manually operated. From the Operational Performance Information System (OPIS) for NPPs, which is the overall database for controlling safety performance, human error under startup and shutdown was found to be 9% during last 20 years in South Korea. For reducing the operator’s load from the startup and shutdown operations of existing NPPs, it is necessary to develop an operator support system based on Artificial Intelligence (AI). Recently, AI technology has facilitated a breakthrough by accumulating data, advanced algorithms, and growing computing power. Among these factors, the key technology of the breakthrough is deep learning that leads current AI technology. In many technical fields, the development of automation and autonomous systems has been studied by using deep learning. Therefore, in this study, an automation system for the startup and shutdown in NPPs develop using deep learning. The automation system is based on an expert system due to characteristics of the startup and shutdown operating modes, and a variety of operating controls depending on each operator are simulated by deep learning. A feasibility study is conducted by using the Compact Nuclear Simulator (CNS) that is a simulator based on Westinghouse 3-loop NPPs. The target scenario for the feasibility study is bubble creation in a pressurizer under startup. In addition, a selected deep-learning algorithm is a Recurrent Neural Network (RNN), which is a robust method for time series analysis.