A styling optimization design method for new energy SUV wheel hubs based on positive emotional granularity

This study develops a styling optimization method assisted by positive emotional granularity (PEG) for new energy SUV wheel hubs. Scenario-based expert screening, imagery-word refinement, principal component analysis, and user ratings establish three imagery dimensions; morphological analysis and the analytic hierarchy process identify eight inputs. A particle swarm optimization–backpropagation neural network (PSO-BP) is compared with BP, support vector regression (SVR), sparrow search algorithm–optimized SVR (SSA-SVR), and ordinary least squares regression (OLS). The third-generation nondominated sorting genetic algorithm (NSGA-III) and entropy-weighted TOPSIS support candidate generation and selection. Across 20 matched random splits, PSO-BP achieves the lowest equally weighted three-target test root mean square error (0.1896). After correlation adjustment and Holm correction, only its comparison with OLS for modern dynamic imagery is significant ( \( \:p\) = 0.0315). Ten searches yield a final hypervolume of 1.10825 ± 0.00360. Within a fixed 354-candidate set, equal weighting and six local weight perturbations preserve the baseline top-ranked candidate, whereas surrogate changes and prediction errors affect rankings. The implemented design exhibits a maximum radial static stress of 175 MPa, below the 252 MPa yield strength. The decoupling–mapping–optimization–verification method supports early styling decisions, subject to sample coverage, predictive uncertainty, and engineering conditions.