Artificial intelligence operation and maintenance knowledge base construction method and system based on multi-modal data fusion

Abstract

This invention relates to the fields of AI operation and maintenance and machine learning technologies, and discloses a method and system for constructing an AI operation and maintenance knowledge base based on multimodal data fusion. The method includes acquiring parallel data streams from new and old monitoring systems, performing time-series data alignment and normalization, semantically encoding fault cases to generate a set of semantically similar sample pairs, constructing three types of sample pairs, inputting the multimodal data into a joint compression encoder, constructing a joint contrastive loss function including type discrimination loss, semantic preservation loss, and cross-system alignment loss for encoder training, and finally jointly compressing the real-time data to generate a unified compressed knowledge base. The system includes eight modules: data acquisition, data alignment, semantic encoding, sample pair construction, joint compression encoding, loss function construction, encoder training, and knowledge base generation.

Images