针对动态规划存在的"维数灾"问题,提出了一种求解大规模电力系统机组组合(unit commitment,UC)问题的策略迭代近似动态规划(policy iteration-approximate dynamic programming,PI-ADP)方法。采用策略迭代对动态规划过程中的值函数进行近似,替代了从可行状态中精确计算值函数的过程,避免了"维数灾"的发生。在值函数的近似过程中,利用了实际系统的运行约束有效压缩状态空间,减少可选择的启停动作,进一步降低了计算量。10~1 000机96时段系统的计算结果表明,所提方法能在较少时间内获得高质量的解,从而为求解大规模电力系统UC问题提供了参考。

ABSTRACT:To solve the ‘curses of dimensionality’ problem of dynamic programming, this paper proposed a policy iteration-approximate dynamic programming (PI-ADP) method for large scale power system unit commitment (UC) problems. Policy iteration was introduced to approximate the value functions in the dynamic programming procedure, replacing theprocess of calculating the value functions accurately from feasible states, and the ‘curses of dimensionality’was avoided. As approximating the value functions, the operation constraints of practical systems were used for compressing state space effectively and the alternative on/off actions were cut down, so that the calculation amount was reduced further. The results of the systems ranging in size of 10 to 1000 units 96 times show that a high quality solution can be achieved in less time by applying the proposed method. It provides reference for solving the large scale power system UC problems.

为了解决内河海事无人艇路径规划问题,提出了一种基于电子江图的路径遍历算法。该算法以分层的电子江图为基础,运用全局路径规划和局部路径规划的方法寻找近似可航路径。运用栅格法在复杂多变的内河环境中选择可航区域,运用Voronoi图对动态物标或可视为质点的碍航物建立航行路径集;将可航区域(或轻微碍航区域)与航行路径集公共区域记为可航路径;并运用贝塞尔曲线和二次规划数学方法进行优化。Matlab仿真结果表明,当障碍物位置坐标不同或目的地位置坐标不同时均可以生成近似可航路径;生成的不同近似可航路径均能被优化为最优安全可航路径,所以建立的环境模型以及使用的路径规划算法是有效、可行的。

In order to solve the problem of path-planning for inland maritime unmanned surface vessel, a traversal algorithm based on inland electronic chart is presented. The grids method in the algorithm is selected to choose navigation area based on the layered inland electronic chart, using global path planning and local path planning method to find approximate navigable path. The Voronoi diagram is selected to establish the navigation path set for dynamic objects or obstacles which can be regarded as particles. Bezier curves and quadratic programming mathematical methods are used to optimize the path of the navigable which is the public area of navigation area(or minor hinders navigation area)and the navigation path set. Matlab based simula-tions demonstrate that when the obstacle position coordinates or destination location coordinates can generate approximate navi-gable path. Generating different approximate navigable path can be optimized for optimal safety navigable path, so the construc

针对实际工业常见的定点跟踪控制问题,通过数学变换,将原系统最优跟踪控制问题转化为新系统最优调节问题,以跟踪误差作为新系统的状态量,引入ε-自适应动态规划算法(ε-ADP)求解HJB方程,并以两个BP神经网络分别用于近似性能指标函数和最优控制,从而得到ε-最优跟踪控制。仿真实验表明,所设计的控制器可以在有限时间内将状态跟踪到目标值,并使得性能指标函数近似最优。

In order to deal with the common fix-point tracking control problems in actual industrial systems, a mathematical transformation is developed to change the original system optimal tracking control problem to an optimal regulator problem of a new system. The state variables of the new sys-tem are the tracking error. ε-adaptive dynamic programming (ε-ADP) is used to solve HJB equation while two BP neural networks are used to approximate the performance index function and optimal control. Thus ε-optimal tracking control is obtained. Simulation results show that the controller de-signed can track a state to the target and make the performance index function converges to optimal.

质子交换膜燃料电池（PEMFC）系统具有明显的非线性和时变的特质，因此质子交换膜燃料电池的建模和优化控制问题是研究的重点。通过建立单体质子交换膜燃料电池的近似线性动态模型，并在此模型基础上，设计了基于双启发式动态规划（DHP）的质子交换膜燃料电池神经网络优化控制器。仿真结果表明，此近似线性模型有效地简化了非线性和时变的特质，在此模型基础上所设计的神经网络控制器具有更好的控制效果和控制精度。

Proton exchange membrane fuel cell(PEMFC) system had obvious nonlinear and time-varying characteristics, so the study of proton exchange membrane fuel cellsystem neural network optimization control was necessary. The approximation linear dynamic model of a single proton exchange membrane fuel cellwas established. Then based on the approximation linear model, the optimization control er based on the dual heuristic dynamic programming (DHP) of was designed. Simulation results show that the nonlinear and time-varying characteristics are effectively simplified by this approximate linear model, and the proposed neural network control er has better control effect and control accuracy.

针对移动机器人在有大型障碍物和运动空间相对狭窄的复杂环境中,人工势场法(APF)容易出现反复震荡、路径规划时间较长以及大型障碍物附近避障困难的问题,提出了在结合边缘探测法的APF路径规划基础上,加入自适应动态步长调整算法来克服APF的上述缺陷,实现移动机器人在复杂环境下的平滑路径规划,在确保路径近似最优的同时提高APF算法的收敛速度和路经规划的避障性能。实验结果证明了上述方法的有效性。

When the obstacles are large, or the complex environment space is relatively narrow, Artificial Potential Field method (APF)is prone to appear repeated shocks, long time planning and obstacle avoidance of difficulty nearby the large obstacles. This paper presents an adaptive dynamic step length adjustment method based on the APF path planning which is combined with the edge detection method to overcome the proposed defects of APF, achieving mobile robot smooth path planning in the complex environment. Hence it can not only improve APF algorithm convergence speed and the safety of path planning, but at the same time ensure the approximate optimum path. Experiments are carried out by simulation to verify the effectiveness of the afore-mentioned methods.