Abstract: Transient statistics is a significant process in nuclear power plant, and is closely related to the operation monitoring and life extension of nuclear power plant. Transient classification is the key sub-process of transient statistic which can classify the operational transients into design transient. The existing automatic classification algorithms have two limitations: 1) the recognition accurate is low; 2) the algorithms training needs large operational data, which is lack for new nuclear power plant. Therefore, the manual classification method is still used in most nuclear power plants. In present work, a combined classification algorithm of rule-based and data-based is established for the nuclear power plant transient classification issue. In present algorithm, the nuclear power changes are adopted in the rule-based classification algorithm, and the Knearest neighbor (KNN) method is used to measure the distance between the operational transient and design transient. Although the power based rule-based algorithm has physical meaning, some transients with the same start power and end power cannot be classified in the rule-based algorithm. Therefore, a data-based method based on the dynamic time warping algorithm was proposed in present work. The distance between the key parameter time series data, such as the temperature, pressure and flow rate in the reactor coolant system, was calculated in the dynamic time warping algorithm, separately. Then an equivalent distance was employed to describe the difference between the operational transient and design transient. Moreover, a similarity factor was established by using the transformation function. The influences of transient parameter weights, signal-to-noise ratio on the algorithm were investigated, and the parallel efficiency of the algorithm was tested. The results indicate that present algorithm has high parallel computing efficiency, and has a high recognition accuracy rate when the signal-to-noise of the signal is higher than 30 dB. Moreover, the operational data of a nuclear power plants is used to verify the algorithm. The results show that present algorithm can not only identify conventional operational transients accurately, but also capture the transients with small fluctuations, which is hard to classify in the manual method. Meanwhile, present algorithm can works in the case that some data is missing. The classification suggestions with the similarity can be given. Therefore, present algorithm can effectively classify the operational data into design transients. Comparing with traditional method, present algorithm solved the problem that a large training data is required. Present algorithm can be widely used in the pressurized water reactor nuclear power plant.