Adaptive supervisory control in target tracking

Consider a field where sensors are randomly distributed in order to detect and track various targets that enter the field. It is assumed that the perimeter of the field is fully covered by sensors so that any target entering the field is detected. There is assumed to be a controller to enable or disable the sensors as the targets move within the field or leave the field. The controller works on a discrete time, so that at each pre-defined time epoch, all the sensors are scanned and tracking errors are measured. Should a sensor fail to observe any target or should the tracking error for a target exceed a threshold, the controller then alters the combination of the tracking sensors. A combination of two or three sensors defines a tracker, and each target is associated with a tracker. The tracking error basically defines how good a target is tracked. The error can be calculated in different ways, such as GDOP Error, Covariance Error, etc. We call this system adaptive supervisory control system.; Depending on the application, different types of optimal control problems can be formulated. We are interested in the applications where the command and control operate remotely and that the sensors are limited in their power source. Thus, the optimal control problem here is to optimize the sensory network power utilization (minimize the power consumption) and to minimize the cumulative tracking error for all the trackers (maximize the accuracy of target location). The adaptive supervisory control system is formulated in this thesis. The primary goal of the controller is to automatically generate the optimization problem based on the mission surveillance requirements and any unfolding events that occur in the surveillance area.; The adaptive controller has to choose, at each moment in time, the minimal set of sensors to activate in order to achieve the surveillance mission requirements. It performs two main functions: translation of the broad mission requirements into a precise system objective and optimization of the sets of sensors to achieve the system objective.; Sensor selection for perimeter detection and target tracking are two critical issues in order to obtain the optimal solution. Genetic algorithm may conduct these two important tasks, which bring us satisfactory result. Other algorithms, such as network flow methodology, integer-programming method, are explored to achieve a better solution.; A critical issue for the controller is the estimation of' target trajectory so that the controller can pro-actively activate proper sensors and establish the proper tracker. Target trajectory estimation can impact the control strategy especially in some special tracking problems such as “holes”. In this thesis, target path model is established by autoregressive-moving average (ARMA) model. By this model, target path estimation and forecasting are carried out, and thus the controller strategy and algorithms can be improved.