<p>Copper smelting is challenged by the complex and variable composition of concentrates, along with dynamic interactions among key elements (Cu, Fe, S) and fluctuating process parameters. To address this gap, this study presents an integrated model merging metallurgical principles with machine learning, encompassing phase matching, equilibrium calculation, and product prediction. A dynamic phase matching based on elemental and phase molar masses enables analysis of the main Cu-Fe-S phase distribution in copper concentrates. A phase-based software for material and heat equilibrium calculations was developed, providing a reliable theoretical tool for process parameter optimization. By integrating the particle swarm optimization (PSO) algorithm to optimize the support vector machine (SVM), a prediction model with phase-based inputs for smelting products was constructed. The model achieved prediction errors within 1% (<i>R</i>² &gt; 0.98) for critical parameters of copper matte and copper slag in the smelting process, including their mass and component compositions. K-fold cross validation results indicated an average <i>R</i>² of 0.989 ± 0.002. This integrated phase-based framework bridges the gap between raw material characterization and dynamic process control, providing a theoretical and technical basis for intelligent copper smelting.</p>