SBIR Phase I: Accurate Demand Forecasts of Spare Parts Inventories

SBIR第一阶段：准确的备件库存需求预测

• 批准号: 1246888
• 负责人: Nelson Hartunian
• 金额: $ 15万
• 依托单位: Smart Software Inc
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2013
• 资助国家: 美国
• 起止时间: 2013-01-01 至 2013-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1246888&HistoricalAwards=false
• 关键词: SBIR; Phase; Accurate; Demand; Forecasts

The innovation is a new method for forecasting the demand for spare and service parts whose demand is intermittent or slow moving. Intermittent demand consists mostly of zero values, with nonzero demands of random sizes mixed in at random times, making it extremely difficult to forecast. The innovation will allow parts providers to operate in the "sweet spot" that balances the costs of keeping unused parts on the shelf against the costs created by not having parts available when needed. The research will develop the new forecasting algorithm, embed it in a prototype software product, and document its greater accuracy compared to conventional methods. The evaluation of accuracy will be based on extensive computational experiments using both synthetic data (to discover which data features are critical) and a library of over 100,000 real-world demand histories provided by existing customers.The broader/commercial impact will be reduction of a multi-billion dollar drag on the US economy: mismanagement of parts inventories. Parts inventories are the second largest item on the balance sheets of many companies, and spending on them amounts to roughly 8% of the US Gross Domestic Product (GDP). Improved parts forecasting can lead to increases in parts availability by over 10%, and simultaneous reductions of over 15% in inventory costs. These improvements will benefit not only the vendors of parts but their customers, whose supply chains will become more reliable and whose operations will have reduced down time.

该方法是对间歇性或缓慢变化的备件需求进行预测的一种新方法。间歇性需求主要由零值组成，随机大小的非零需求在随机时间混合，使得预测极其困难。这一创新将允许零部件供应商在“最佳点”运营，平衡将未使用零部件放在货架上的成本与在需要时没有零部件可用所产生的成本。该研究将开发新的预测算法，将其嵌入到原型软件产品中，并证明其与传统方法相比具有更高的准确性。准确性评估将基于广泛的计算实验，使用合成数据（以发现哪些数据特征是关键的）和现有客户提供的超过100,000个真实需求历史的库。更广泛的商业影响将是减少对美国经济的数十亿美元拖累：零部件库存管理不善。零部件库存是许多公司资产负债表上的第二大项目，其支出约占美国国内生产总值（GDP）的8%。改进的零件预测可以使零件的可用性增加10%以上，同时减少15%以上的库存成本。这些改进不仅有利于零件供应商，也有利于他们的客户，他们的供应链将变得更加可靠，他们的操作将减少停机时间。