Improved Microservice Fault Prediction Model of Informer Network

It is of great significance to accurately predict the running state of the system. An improved Informer prediction model, PCI-Informer, is proposed to divide the sequence data into sub-sequence-level patches, and carry out feature extraction and integration to improve the processing ability and effect of the model on the sequence data. The multi-scale cause-and-effect convolution self-attention mechanism is used to realize multi-scale local feature fusion and improve the model's understanding of local information and modeling ability. The reverse residual feedforward network is introduced to enhance the ability of extracting and retaining local structure information. The experimental results of open source data set show that compared with the mainstream prediction model, PCI-Informer model achieves better prediction effect under different prediction steps.