Making accurate demand forecasting of iron ore of China is essential for resource planning, strategic investment, and risk management. However, the complex and nonlinear nature of iron ore demand, driven by the interplay of domestic economic factors and international market forces, poses significant challenges to conventional forecasting methods. This study proposes a hybrid forecasting framework that integrates the Archimedes Optimization Algorithm (AOA) with the Least Squares Support Vector Machine (LSSVM) to predict China's annual iron ore demand. A comprehensive dataset spanning 2014 to 2025 is constructed, comprising ten explanatory variables that capture both macroeconomic conditions, including GDP, real estate investment, urbanization rate, industrial output, and steel apparent consumption, and international market factors, including steel net exports, iron ore prices, the Baltic Dry Index, and the USD exchange rate. The AOA is employed to automatically search for the optimal LSSVM hyperparameters, overcoming the limitations of manual grid-search approaches. Using the first eight years as the training set and the remaining four years for out-of-sample evaluation, the proposed model is benchmarked against the standard LSSVM without optimization. Empirical results demonstrate that the AOA-LSSVM model consistently outperforms the baseline across all evaluation metrics, reducing the mean absolute error by 7.05%, the root mean square error by 7.47%, and the mean absolute percentage error by 0.42 percentage points. These findings confirm the efficacy of AOA in enhancing LSSVM's generalization capability and underscore the value of integrating meta-heuristic optimization with machine learning for mineral resource demand forecasting. The proposed framework offers policymakers and industry practitioners a reliable tool for generating accurate demand projections to inform import strategies and resource security planning.