A New Technology for Forecasting Intermittent Demand in Manufacturing

间歇性需求的问题干扰了美国制造企业的高效生产和库存控制活动。虽然准确和及时的需求预测在计划生产、库存和劳动力以及经济批量方面的关键作用受到了相当大的关注，但绝大多数预测研究以及几乎所有商业预测软件都假设需求是“平稳的”，而不是断断续续的。第一阶段小型企业创新研究(SBIR)的结果为一种新的软件工具奠定了技术基础，该工具可用于预测间歇性需求，并适合在台式计算机上使用。第二阶段研究的目标是：(1)收集和分析更多真实世界的间歇性时间序列数据；(2)实施Croston预测间歇性需求的单序列方法，包括将其扩展到趋势数据和季节性数据；(3)开发了一种新的基于随机过程模型的多序列预测方法，该模型跨产品线中的各个项目“借力”；(4)确定最有效地利用销售人员对间歇性需求的判断，包括将其与统计预测相结合；(5)将第二阶段制订的预测方法结合起来，最终在台式计算机上运行的软件原型中加以实施。该项目将为间歇性需求预测问题提供一种最先进的计算机化解决方案。很大一部分美国制造企业将受益于这种需求预测带来的更高效率的生产。

The problem of intermittent demand interferes with efficient production and inventory control activities in US manufacturing firms. While the key role of accurate and timely demand forecasting in planning production, inventories and work force, and economic lot sizing has received considerable attention, the vast majority of forecasting research, and nearly all commercial forecasting software, assume that demand is "smooth" rather than intermittent. Results of the Phase I Small Business Innovation Research (SBIR) study establish the technical basis for a new software tool that can be used to forecast intermittent demand and that is suitable for use on a desktop computer. The objectives of the Phase II research focus on (1) gathering and analysis of more real-world intermittent time-series data; (2) implementation of a single-series method of Croston's for forecasting intermittent demand, including its extension to trending and seasonal data; (3) development of a new multi- series forecasting approach based on stochastic process models that "borrow strength" across the items in a product line; (4) determination of the most effective use of sales-force judgments of intermittent demand, including their combination with statistical forecasts' and (5) integration of the forecasting methods developed in Phase II, culminating in their implementation in a software prototype running on desktop computers. The project will provide a computerized state-of-the-art solution to the intermittent demand forecasting problem. A large percentage of US manufacturing firms stand to benefit from more efficient production made possible by such demand forecasting.