Neural Dynamic Programming for Automotive Engine Control

汽车发动机控制的神经动态规划

• 批准号: 0355364
• 负责人: Derong Liu
• 金额: $ 10万
• 依托单位: University of Illinois at Chicago
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2004
• 资助国家: 美国
• 起止时间: 2004-08-15 至 2007-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0355364&HistoricalAwards=false
• 关键词: Neural; Dynamic; Programming; Automotive; Engine

Neural dynamic programming (NDP) is a scheme that provides approximate solutions to dynamic programming in which neural networks are used as a tool for function approximation. Such a scheme is applicable to problems that minimize/maximize a cost but for which a traditional dynamic programming approach is not feasible since the cost function is usually not known. In the present project, applications of NDP to automotive engine control will be studied and implemented. A challenging problem facing the automotive industry is to design vehicles that generate emissions satisfying the federal government's future emission regulations. The challenge here is to minimize the emission and at the same time to achieve better fuel economy and vehicle driveability. In the past few years, the automotive industry has engaged in efforts to develop engine control algorithms that will generate emissions satisfying the government's emission standards. Emissions generated by automobiles are one of the major sources for air pollution in the United States. Theoretically, emissions can be controlled to a minimum possible level by controlling the engine combustion process so that the air and fuel are mixed at certain desired ratio. This control problem, as it is known, turns out to be very difficult to solve. This is partly due to the complexity of modern automotive engines and due to the complexity of the fuel combustion process. In addition to achieving lower emissions, the automotive industry has also engaged in efforts to design cars that have better driveability and consume less fuelNDP implementation can be accomplished by adding a $1-10 chip developed under a previous NSF SBIR grant that, with additional training, could also be to reduce the cost of fuel flexibility an urgent strategic need.

神经动态规划（NDP）是一种为动态规划提供近似解的方案，其中神经网络用作函数逼近的工具。这种方案适用于最小化/最大化成本的问题，但由于成本函数通常未知，因此传统的动态规划方法不可行。在本项目中，将研究和实施NDP在汽车发动机控制中的应用。汽车工业面临的一个挑战性问题是设计出产生满足联邦政府未来排放法规的排放的车辆。这里的挑战是最大限度地减少排放，同时实现更好的燃油经济性和车辆驾驶性能。在过去的几年中，汽车工业致力于开发将产生满足政府排放标准的排放的发动机控制算法。汽车排放的废气是美国空气污染的主要来源之一。从理论上讲，通过控制发动机燃烧过程，使空气和燃料以一定的理想比例混合，可以将排放控制在尽可能低的水平。众所周知，这个控制问题很难解决。这部分是由于现代汽车发动机的复杂性以及燃料燃烧过程的复杂性。除了实现更低的排放量，汽车行业还致力于设计具有更好的驾驶性能和更少的燃料消耗的汽车。NDP的实施可以通过添加根据先前NSF SBIR拨款开发的1 -10美元芯片来完成，通过额外的培训，也可以降低燃料灵活性的成本，这是一个紧迫的战略需求。