AI-Driven Warehouse Networks: Integrating Probabilistic Forecasting with Dynamic Allocation

The rapid expansion of e-commerce has created an urgent demand for intelligent and precise warehouse management solutions. This study proposes an innovative framework integrating machine learning with optimization algorithms to systematically address two critical challenges: shipment volume prediction and multi-warehouse SKU allocation. To handle the non-Gaussian distribution, high volatility, and complex constraints characteristic of e-commerce data, we first developed a multimodal data preprocessing pipeline incorporating 3D visualization and robust interpolation techniques. This approach significantly enhanced data quality (missing value imputation error <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$&lt;3.2 \%$</tex>, outlier processing efficiency improved by 37 %). For shipment forecasting, a comparative analysis between RF and GBDT revealed RF's superior performance in both prediction accuracy (MAE <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{= 2 5. 7 3}$</tex> vs. GBDT's 26.86) and stability (residual SD <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$=60$</tex> vs. 103), particularly during high-volatility scenarios like promotional events. In warehouse optimization, our multiobjective model demonstrated that GA achieved better category affinity (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$228,073.39$</tex> vs. PSO's <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$45,383.38$</tex>) and constraint satisfaction rates, while PSO yielded higher resource utilization (storage capacity utilization 0.63 % vs. GA's 0.60 %). The hybrid optimization strategy reduced warehousing costs while improving operational efficiency by 15 %, providing a scalable technical pathway for smart supply chains. Future research will focus on real-time dynamic optimization and heterogeneous data fusion to enhance industrial applicability.