Aiming at the problem of large amount of intelligent operation and maintenance KPI data and poor clustering effect, this paper proposes a fast KPI clustering method based on the IP-Kshape (IAE-PAA-Kshape) algorithm. First design an improved autoencoder algorithm (IAE), add a convolutional layer and a two-way LSTM network layer to the standard autoencoder, to achieve smooth denoising of KPI data and timing feature extraction; then the KPI data features are clustered based on the PAA-Kshape algorithm, and the PAA algorithm is used to perform dimension compression and Kshape algorithm to solve the drift problem of KPI sequences, which improves the clustering speed and accuracy of KPI data. Through experimental comparative analysis, it is proved that the method proposed in this paper can better realize the rapid clustering of KPI data, and the time efficiency and accuracy are better than traditional machine learning or deep learning methods.