Process Anomaly Diagnosis Using Hierarchical Color Maps

The patent segments the large number of statistical models into multiple groups based on their prediction accuracy characteristics. By dividing the models into groups (e.g., high-accuracy group, medium-accuracy group, low-accuracy group), the system can selectively apply different evaluation strategies to each group, thereby maintaining comprehensive diagnostic coverage while reducing the noise impact from low-accuracy models on the overall prediction result.

The patent applies local quality by evaluating and weighting different statistical models based on their local performance characteristics. Each model is assessed individually for its prediction accuracy and reliability, and weights are assigned accordingly. This allows high-accuracy models to have greater influence on the final diagnosis while low-accuracy models contribute less, thereby reducing noise without limiting the system's ability to handle complex processes with diverse characteristics.

The patent introduces a hierarchical dimension to the color mapping structure by organizing statistical models and their prediction results into multiple layers. The first layer provides an overview of all models, while subsequent layers provide progressively more detailed information. This dimensional organization allows operators to navigate from general to specific anomaly information systematically, improving visual recognition without requiring an overly complex flat structure.