Supply–Demand Prediction for Agile Manufacturing with Deep Neural Network

Abstract

Abstract Agile manufacturing represents modern production systems that rapidly respond to a fast-moving marketplace and connect customers to production. The success of an agile manufacturing system relies on multisource data analytics, which provide smart data-driven decision-making strategies to guide manufacturing and the supply chain system. In order to implement rapid manufacturing processes to respond to customer orders, supply–demand gap prediction plays a critical role. In this article, we study the problem of predicting supply–demand gap with respect to product types, categories, and spatiotemporal attributes. To this end, we propose and develop an end-to-end framework using an extendable deep neural network architecture for supply–demand gap reduction. The framework can process multiple customized input factors and automatically discover spatiotemporal supply–demand patterns from historical transaction data. A set of customized features are employed to build a general training model to predict future demand. Embedding layers are used to map high dimensional features into a smaller subspace for input data consolidation. Fully connected layers with activation functions are used to build the training architecture of the model. Customized data attributes can be concatenated from different layers in the deep learning neural network. In this way, multisource input data can be integrated with outputs of internal layers for a comprehensive demand prediction. Experiments were conducted to evaluate the network with real supply and demand data, which were acquired from warehouses of a manufacturing company. The experimental results demonstrated that the proposed network was able to reduce supply–demand gap and provide a practical solution to long-term customer demand prediction.