Abstract: fusion，multi-dimensional application performance indicators show non-stationary，multi-scale coupling and dynamic evolution characteristics. It is difficult to give-consideration to local transient fluctuations and long-term dependence of cross-indicators in the detection process，resulting in insufficient ability to identify hidden faults. Therefore，a multi-dimensional application performance index anomaly detection method based on CNN transformer is proposed. Using multidimensional temporal index data as input for the CNN transformer combination model，a one-dimensional CNN is employed to extract local abnormal features such as short- term single index mutations and local linkage of multiple indicators through the synergistic effect of convolutional layers， activation function layers，and pooling layers；and under the collaborative effect of position encoding，multi-head attention， residual connections，and layer standardization in the Transformer encoder，capture global long-term dependencies across time and metrics；after processing the outputs of CNN and Transformer through a fully connected layer, multi-dimensional application