As high-performance plastics find expanding applications in automotive, electronics, and packaging applications, the efficiency and accuracy of formulation design have become critical bottlenecks limiting performance gains and the advancement of smart manufacturing. Conventional approaches rely heavily on expert intuition and trial-and-error experimentation, leading to prolonged development cycles, high costs, and difficulty in meeting multiobjective constraints. To address these challenges, we propose a big data-driven framework encompassing raw material compositions, processing parameters, and performance metrics, supported by a complete workflow of data preprocessing, feature engineering, multifeature constraint integration, formulation generation, and multiobjective optimization. Data preprocessing employs normalization, GAN-based imputation, and Isolation Forest algorithms to ensure consistency and integrity via outlier removal. Feature engineering constructs vectors from ternary diagram coordinates, process parameters, structural descriptors, and performance indicators, extracting high-dimensional representations via early concatenation, late fusion, and attention-based fusion. A domain-constrained reverse design framework based on a variational autoencoder enables the generation of controllable formulas. Multiobjective optimization is achieved using NSGA-II (nondominated classification genetic algorithm II) under performance - cost - sustainability constraints, with final selection guided by the Achievement Scalarization Function. Experimental validation shows that embedding compatibility rules and physicochemical constraints in the neural network improves the average hypervolume indicator by 0.06 and raises first-pass regulatory compliance by over 15%, confirming the framework’s industrial applicability.