Performance analysis of fuzzy rule based classification system for transient identification in nuclear power plant

•An interpretable fuzzy system with acceptable accuracy can be used in nuclear power plant.•This system is worthy of being used as a redundant system for transient identification.•Deaerator level gives quicker response to fuzzy system, to classify transients in steam water system.•Increase in the number of input variables does not necessarily increase the efficiency of a fuzzy system.•Helps in operator guidance by reducing information overloading. Even though fuzzy rule based classification system (FRBCS) has been useful in event identification, it has led to strong clash in terms of better interpretability along with adequate percentage of accuracy. Basically for classification in nuclear power plant (NPP) which receives data within a cycle time of few milliseconds, either the accuracy or the interpretability of the FRBCS would get jeopardized. Online event identification of any abnormality or transient using FRBCS becomes really critical for the plant which has such a short cycle time. For such cases, the output from a FRBCS may not be conducive to classify the event every cycle. Thus, it is necessary to monitor the output of a classification system for certain amount of cycles till the static nature is attained. This gives a high level of confidence on the classifier output to be accurate. A FRBCS can produce this level of confidence by choosing the best input features with high interpretability and acceptable accuracy. The best feature selection out of a lot of input variables and preparing the rule base is again a very critical and challenging task in FRBCS. There is always a dilemma on judiciously choosing the number of input features for the FRBCS to achieve an optimized interpretable and accurate fuzzy system. It is always advisable to select least number of features with proper output error margin for a FRBCS. On adding extra features along with some rules as input to the system, certainly increases the accuracy of the system but degrades its interpretability. The other thing which is of equal importance is to keep an agile notice on the output which has to be within the output error margin. The output error margin denotes the response time and static nature of the classifier output. Here, the classification could be confirmed only after observing the output for a subsequent number of cycles. Both the number of inputs and the output error margin needs to be properly balanced in order to get an efficient online FRBCS for such kind of low cycle time