Abstract: With the deep integration and development of the information technology and system management, the Monte Carlo simulation technology and stochastic gradient estimation methods have attracted increasing attention in the system management. Gradient estimation is a critical tool for optimizing the structure of complex systems, inferencing complex systems parameters, and measuring and controlling risks. Since gradient estimation provides more useful information than system performance estimation, the accurate estimation of the stochastic gradient has been widely studied in the simulation literature, and it is a central issue in the gradient-based optimization methods. This paper provides an overview of the commonly used stochastic gradient estimation methods and introduces the main challenges, solution ideas, and insights into management for their applications in system management, to provide theoretical and methodological foundations for the system management.

Key words: stochastic gradient estimation, Monte Carlo simulation, simulation optimization, system management

摘要： 随着信息技术与系统管理的深度融合发展，蒙特卡洛仿真技术和随机梯度估计方法在系统管理中受到越来越多的重视。梯度估计是优化复杂系统结构、估计复杂系统参数、度量和控制风险的重要工具。由于梯度估计包含比系统表现估计更多有用的信息，如何对随机梯度进行准确估计在随机仿真领域得到了广泛的关注，并且它是基于梯度的优化方法中的核心问题。对常用的随机梯度估计方法进行了梳理，并举例介绍了其在系统管理中应用的主要挑战、求解思路及对管理学研究的启示，以期为系统管理提供理论和方法学基础。

关键词: 随机梯度估计, 蒙特卡洛仿真, 仿真优化, 系统管理