AEWS-Net: An Auditor Early-Warning Indicator Set Construction Network Based on Contrastive Learning

Through the construction of early warning index set of electric power management environment audit, it can effectively support electric power enterprises to optimize audit process and improve audit efficiency. However, the traditional way of building audit early warning index set is mainly formulated by experts, which cannot be flexibly changed according to the complex and changeable business environment. In addition, audit data is often derived from different areas or sectors, which presents challenges for the effective processing of multiple sources of data. Therefore, we use deep learning technology to propose an end-to-end multi-source data processing network AEWS-Net (Auditor Early-Warning indicator Set construction Network) based on contrast learning and self-attention mechanism to efficiently screen and construct the set of audit early warning indicators. The model uses convolution view and self-attention view contrast learning methods to learn sequence information of multi-source audit data to obtain richer feature representations, and uses multi-source feature attention mechanism to learn the importance of different audit data sources to the early warning indicator set. Through the performance and ablation experiments on the power audit data, the superiority of the proposed method in the task of index set screening and construction and the rationality of the model architecture are proved.