A Computational Intelligence Framework for Data-Driven Shape Optimization Using GRA and GA-BPNN

Abstract Traditional product shape optimization suffers from single-objective tuning, low prediction precision, and inefficient nonlinear mapping between design variables and performance. To address the limitations of conventional Backpropagation Neural Network (BPNN), such as weak optimization ability and low prediction accuracy in shape design, this paper proposes an intelligent optimization framework integrating Grey Relational Analysis (GRA) and Genetic Algorithm-optimized BPNN (GA-BPNN) for purple clay teapot shape design. First, a standardized teapot shape database is established, and morphological decomposition and factor analysis are performed to extract and screen key shape parameters. GRA is then applied to quantify correlations between parameters and objective dimensions, determining four core inputs for the neural network. On this basis, GA is used to optimize the initial weights and thresholds of BPNN, overcoming slow convergence and local optima issues. The GA-BPNN model achieves accurate nonlinear mapping from subjective design requirements to three-dimensional shape parameters and outputs the globally optimal shape combination. Experimental results and user tests show that the optimized scheme exhibits superior objective matching and user satisfaction, with significantly higher prediction accuracy and convergence efficiency than the standard BPNN. This work establishes a complete intelligent optimization workflow for product shape design, which supports high-precision multi-parameter optimization and nonlinear function mapping. The proposed hybrid method has strong scalability and can be extended to general shape optimization and multi-parameter decision-making tasks, providing a data-driven approach for intelligent design applications.