An industrial product quality prediction method based on geometric preservation cross-scale difference

Abstract

The application provides an industrial product quality prediction method based on geometric preservation of cross-scale differences, and relates to the technical field of industrial product quality prediction, comprising: preprocessing the collected time series data of industrial process variables, constructing a sample sequence using a sliding window, and dividing into a training set, a validation set and a test set in chronological order; by constructing a double-branch coding architecture, trend characteristics and difference characteristics are independently processed; a geometric perception attention mechanism is introduced in each branch encoder to ensure that the hidden layer representation and the output target space are geometrically consistent, thereby enhancing the stability and interpretability of the model; further, cross-scale cross-attention is used to realize deep interaction and adaptive fusion of double-branch information, thereby significantly improving the comprehensive modeling capability for long-term trends and short-term dynamics in the industrial process.

Images