SBIR PHASE I: Advanced Adaptive Critic Designs for Optimization and Control of Complex Nonlinear Dynamic Systems

这个小企业创新研究（SBIR）第一阶段项目将设计一个基于自适应批评的控制器，用于优化控制复杂的大规模、多目标系统，这些系统表现出非线性动力学。目前实现的自适应批评控制器通常在小规模问题上工作得很好，但在复杂的大规模问题上收敛速度很慢。此外，需要优化性能与竞争目标使得自适应批评控制器的学习和收敛困难。第一阶段将在一个典型但具有挑战性的问题上演示这种自适应批评控制器：在标准驾驶周期内优化自动驾驶特性。控制器将优化油门控制、加速、减速和制动，以最小化整个过程中的能量消耗，同时以最小的时间跨度穿越周期。这类似于长途卡车司机寻找最佳的油门踏板和刹车控制顺序，以便在最短的时间内从纽约到洛杉矶，消耗最少的燃料（并在两者之间的丘陵和山脉中行驶）。这种自适应临界控制器的商业应用有望在半导体制造和飞行控制以及汽车控制问题中得到应用。

This Small Business Innovation Research (SBIR) Phase I project will design an adaptive critic based controller for use in optimizing control of complex large-scale, multi-goal systems which exhibit nonlinear dynamics. Currently implemented Adaptive Critic controllers often work well on small-scale problems, but convergence is slow on complex large-scale problems. In addition, the need to optimize performance with competing goals makes learning and convergence of adaptive critic controllers difficult. Phase I will demonstrate this adaptive critic controller on a typical but challenging problem: optimization of auto driving characteristics over a standard driving cycle. The controller will optimize throttle control, acceleration, deceleration, and braking so as to minimize energy expenditure over the course while also traversing the cycle in a minimum time span. This is similar to a long-haul truck driver searching for an optimal sequence of gas pedal and brake controls to go from New York to Los Angeles in minimum time and burning minimum fuel (and traveling through hills and mountains in between). Commercial applications of adaptive critic controllers of this kind are expected semiconductor manufacturing and flight control as well as in the automotive control problem.