• A novel machine learning strategy integrating GRU recurrent neural network and elastic network model was developed to design counter pressure casting gating systems, achieving high prediction accuracy.
• The method eliminates trial-and-error iterations, reducing casting defect volume from 11.23 cm³ to 2.23 cm³ and eliminating internal defects via an internally cooled iron.
• Spearman’s correlation analysis and orthogonal experiments ensured high-quality feature selection and dataset construction, enhancing model reliability.
• Comparative simulations using EasyCast and ProCAST confirmed steady filling and top-down sequential solidification, validating the intelligent gating system design.