The accuracy of vehicle location plays a pivotal role in several applications for bus network operations such as service control, warnings about low bridges, accurate prediction of bus arrival times, and traffic signal priority, as well as when historical operational data are used to measure network performance. In the context of iBus, an automatic vehicle location and control system deployed on all 8,400 buses operated by London Bus Services Limited (London Buses), tests demonstrated that the solution based on the Global Positioning System (GPS) provides location accuracy within 10 to 12 m 95% of the time; this timing is sufficient to support operational needs. To meet that level of location accuracy, the system must operate with a high degree of availability and accuracy on all vehicles. The challenge is to install, maintain, and repair vehicles so that they can operate at the required levels of performance under harsh operating conditions. Buses have many components—including the odometer, gyrocompass, aerials, WiFi, and power units—that can fail. This paper presents three approaches that London Buses uses to identify vehicles that have faulty hardware. One technique has also proved to be beneficial in testing new navigation software releases and in identifying design and parameterization problems that affect the quality of the navigation solution. The results will interest those involved with testing, maintenance, and repair of GPS-based vehicle fleets. These methods have helped ensure that London Buses can successfully observe 98% of all operated bus stop visits.

No abstract

Two methods for generating smoothing splines are compared and applied to data from a fed-batch fermentation process. One method chose both the degree of the spline and its parameters by minimizing the generalized cross validation (GCV) function using a genetic algorithm (GA). The other method adjusted the smoothing spline to a specified chi-square goodness-of-fit, requiring prior knowledge of the measurement variability. The GCV/GA method led to excellent results with all the fermentation data records. The goodness-of-fit method gave a family of spline fits; splines with a low percentage fit extracted trends from the data, while for general use a 50% fit appeared satisfactory. The goodness-of-fit method executed more quickly than the GCV/GA method, but the GCV/GA method was more generally applicable as it chose both the degree of the spline and the amount of smoothing automatically.

No abstract