The on-line monitoring parameters of nuclear power system are disturbed by noise during the process of acquisition and transmission, which leads to the random deletion of the final monitoring signal. The great disturbance for the operator to judge the type of accident was caused. Therefore, a sliding time window fault diagnosis model based on dynamic time warping (DTW) was proposed. The multivariate time series were built from on-line real-time monitoring and the existing accident standard series. The sliding time window was used to dynamically search the minimum cumulative distance in the standard series. The absence of monitoring parameters in the test series or the standard series leaded to the unequal length sequence, and DTW was used to deal the phenomenon. The pattern category of the time series depended on the minimum cumulative distance. This method is based on the basic principle of the accident in nuclear power system, and has strong explanation and robustness. At the same time, the model can be extended by other standard accident sequences. © 2019, Editorial Board of Atomic Energy Science and Technology. All right reserved.
