Combining Statistical Process Control and Optimization via Simulation for Robust Sensor Network Design in the Presence of Sensor Measurement Error

点击翻译按钮获取中文摘要

For sensor network design, an optimization method is often used to determine how many sensors to install and where to physically install them. One stream of research is to assume a known functional relationship (usually a regression line) among sensor measurements with normally distributed noise. This assumption does not hold for systems with complex dynamics such as traffic transportation or environmental monitoring. Even when a process simulation is used to capture complex dynamics, it is often assumed that estimated performance measures from stochastic simulation are accurate and there is no false alarm (i.e., no sensor measurement error). A sensor network found under these assumptions may produce unacceptably high false alarm rates, which eventually makes the decision maker abandon the sensor network. This project develops statistical monitoring and controlling methods for raising an alarm when a sensor network of a complicated system detects abnormal behaviors such as a contaminant spill in a water quality monitoring network. Then it develops efficient simulation-based optimization algorithms to determine the number of sensors and their locations when stochastic simulation is used to estimate multiple performance measures. Finally this project combines the statistical methods and the simulation-based optimization to design the optimal sensor network while controlling false alarm rates. This research is interdisciplinary across manufacturing, quality control, simulation and environmental engineering; and the composition of the research team broadens the participation of underrepresented groups in research and teaching. The results from this research are applicable to many application areas and thus will benefit the U.S. economy and society. The objective of this project is to develop methods that will be useful in identifying an optimal sensor network quickly and accurately in the presence of sensor measurement error for a complicated system whose in-control and out-of-control observations are obtained through stochastic process simulation. This project considers a potentially large-scale process with general marginal and general correlation structure, which can broaden application fields of statistical process control (SPC) methods; and develops SPC techniques whose control limits neither require modeling of an underlying process nor trial-and-error calibration. It also develops a combined framework of SPC and discrete optimization via simulation when multiple performance measures exist. Finally it facilitates knowledge transfer from IE/OR to non-traditional IE/OR fields by applying the resulting combined algorithms to the water quality monitoring problem.