(A) retrieval methodology for similar NPP limiting conditions for operation cases based on domain specific natural language processing

Abstract

Nuclear power plants (NPPs) have technical specifications (Tech Specs) to ensure that the equipment and key operating parameters necessary for the safe operation of the power plant are maintained within limiting conditions for operation (LCO) determined by a safety analysis. The LCO of Tech Specs that identify the lowest functional capability of equipment required for safe operation for a facility must be complied for the safe operation of NPP. In nuclear power plants, compliance with technical specifications is used in the same sense as compliance with LCO. If LCO is not met, the required actions shall be taken. If required actions are not properly performed, NPP will be in violation of the technical specifications.In order to comply with the technical specifications, first, the related procedures are performed to prevent unsatisfactory conditions for LCO in advance, second, abnormalities in the LCO are immediately detected, and third, determination of whether LCO is met or not is achieved by considering the comprehensive situation of the power plant after detection, fourth, if it is determined to be dissatisfied, declare entry into required actions related to the LCO, fifth, take required actions within the completion time, and finally, document a series of processes. However, determination of whether LCO is met or not met is a high burden to main control room operators. The main reason is that the determination of whether LOC is met or not is an inherent determination part of the senior reactor operator considering the overall situation of the plant, and the determination may differ depending on each person's point of view. Therefore, in order to solve these problems, there have been many previous studies related to technical specifications. These studies can be broadly classified into those focusing on determination of whether LCO is met and those related to the detection of LCO abnormality.However, in the site investigation related to the technical specifications, it was confirmed that opertors have difficulty not in detecting the abnormality of LCO but in determining whether LCO is met or not met. In addition, since nuclear power plants have very complex and various systems, there are obvious limits to represent all possible cases in a power plant with expert knowledge, and the maintenance cost for continuously accumulating expert knowledge is also high.Therefore, in this paper, we propose a retrieval methodology for similar LCO cases to support determination of whether LCO is met or not. Through the retrieval methodology, it is possible to support rapid decision-making by providing retrieved results similar to user queries. In order to obtain the similarity between user queries and similar cases, natural language processing techniques were used to tokenize user queries and cases, and a nuclear domain dictionary was constructed to improve tokenization performance. Each tokenized word was converted into a number according to the weight using the TF-IDF technique, and the retrieval performance was significantly improved by adding a boolean retrieval model based on dictionary related to LCO to the vector space model. The suggested retrieval methodology is expected to be very helpful in determining of whether LCO is met or not met.