Abstract: The monitoring parameters of the nuclear power system are randomly lost due to noise interference,which affects the judgement of the operators onthe severity of the accident. A diagnosis model of fracture size with tolerance parameter loss is proposed. The multiple time series which fracture size is known is selected as the standard series, and several sampling sites are built on the standard series based on the accident mechanism. The sliding dynamic time warping algorithm is adopted to find the minimum cumulative distance between the diagnosed multivariate time series and the standard sampling site, and all the minimum cumulative distances obtained are taken as the characteristic values of the fracture diagnosis model. The support vector machine is used as the prediction model to predict the size of the fracture, and the ensemble learning strategy is used to optimize the diagnosis results. Taking the right main steam pipeline as an example for verification, the results show that this method does not have high requirements for the integrity of sequencing sequences, and the evaluation error of the fracture with random loss of parameters is within 10%, which makes it better for the auxiliary operator to conduct the evaluation of the fracture.