SBIR PHASE I: A Neuro-Dynamic Programming Approach to Stochastic Control

这个小企业创新研究（SBIR）项目开发了使用神经动态规划（NDP）解决复杂随机控制问题的技术。 虽然随机控制问题在许多应用中是至关重要的，但由于其巨大的复杂性，还没有找到满意的解决方案。 然而，最近的进展，NDP可能使解决方案的随机控制问题，曾经被视为棘手。 第一阶段将制定一个方法框架的应用NDP和应用该技术的一个复杂的现实世界的随机控制问题的商业利益，即供应链管理。 这种方法利用所有的信息，是关键的最优控制，并预计将导致一个现实的建模通用的供应链。 它有望超越现有的随机控制策略，并将进行严格的比较，以建立NDP作为一个重要的和实用的技术。 在第一阶段之后，可以开发用户友好的商业软件产品，以使神经动态编程的广泛应用成为可能。 此外，根据供应链管理案例研究制定的战略将纳入现有的商业供应链管理软件，并向制造公司销售。

This Small Business Innovation Research (SBIR) project develops techniques for solving complex stochastic control problems using neuro-dynamic programming (NDP). Although stochastic control problems are critical in many applications, satisfactory solutions have not been found because of their great complexity. However, recent advances in NDP may enable solutions to stochastic control problems that were once dismissed as intractable. Phase I will develop a methodological framework for the application of NDP and apply the technology to a complex real-world stochastic control problem of commercial interest, namely, supply chain management. This approach exploits all information that is critical for optimal control and is expected to lead to a realistic modeling of a generic supply-chain. It is expected to surpass existing strategies for stochastic control, and a rigorous comparison will be made to establish NDP as an important and practical technology. After Phase I a user-friendly commercial software product can be developed to enable the widespread application of neuro-dynamic programming. Furthermore, strategies developed from the case study in supply-chain management would be integrated into existing commercial supply-chain management software and marketed to manufacturing companies.