Feature extraction and classification method for non-structural business data of power grid

The invention belongs to the technical field of power grid operation data processing and maintenance methods, and particularly relates to a feature extraction and classification method for non-structural business data of a power grid. Comprising the following steps of performing non-structural word vector feature fusion based on a double-layer Transform model; optimizing self-attention module input; obtaining a feature fusion vector; feature extraction and combination based on a convolutional neural network module; generating a combined feature map; and performing classification processing through an activation function. According to the method, through feature extraction, joint feature establishment and multi-head sub-neural network analysis, recognition and positioning of key position vectors and word meanings in the non-structural statement recognition process are improved, the complexity of non-structural western word vector expression is simplified, the data size in the analysis and recognition process o