Supply chain multidimensional demand forecasting and inventory optimization model combining LSTM and convolutional neural network

Abstract

To solve the problem that traditional forecasting models are difficult to effectively capture multi-dimensional spatiotemporal features and long-term dependencies, which leads to high inventory costs, this paper aims to develop an intelligent hybrid model (BO-CNN-LSTM) that fuses Bayesian optimization (BO), convolutional neural network (CNN) and long short-term memory network (LSTM) to realize high-precision demand forecasting and collaborative optimization of inventory strategy. The model automatically adjusts parameters through Bayesian optimization, extracts local spatiotemporal features by CNN, and models long-term dependencies of sequences with the help of LSTM, and finally generates the optimal decision by combining inventory cost function. Experiments on multi-source public datasets show that the proposed model is significantly superior to the traditional method in prediction accuracy, inventory efficiency and anti-noise interference ability, and effectively reduces the total supply chain cost to 624,000 yuan. In summary, this study provides a decision support framework with high accuracy, strong interpretability and easy deployment for supply chain management. Its modular integration idea and empirical effect have important reference value for promoting the intelligent development of supply chain. Although the model still has room for improvement in computational complexity and dynamic adaptability, it will be further optimized by introducing attention mechanism and reinforcement learning in the future.