User:Bwawsc/sandbox

In the early 1970's, a mainline meter was installed downstream of the toll plaza on the westbound San Francisco-Oakland Bay Bridge. This meter had three purposes: 1) to reduce congestion and accidents by limiting flow onto the five lanes of the bridge to match the capacity of these lanes; 2) to encourage car pooling and bus ridership by giving these vehicles preferential access to the bridge; and 3) to give bridge operators a way to control traffic during incidents such as stalls or accidents, thus allowing them to get tow trucks to the scene of the incident more quickly.

The signals themselves were mounted on a bridge spanning the 15 traffic lanes at the metering point, a few hundred feet downstream (west) of the toll booths. Each signal head was programmed to be visible only to the first one or two vehicles in line at the stop bar in the lane being controlled by that signal head (a significant challenge). Vehicles were detected by magnetometer vehicle detectors buried in the pavement. Separate detectors were used to detect waiting vehicles (queue detectors) and released vehicles (passage detectors). The painted stop bar (a 12" wide painted white stripe crossing all lanes) was used, in conjunction with programming the red signal head to "drop out" if the vehicle drifted too far forward, to keep waiting vehicles positioned over the queue detectors. The green signal was not restricted - we wanted the driver who drifted too far to be able to see and react to the green signal.

The results of the metering were measured at five detectors (one in each lane) located at the foot of the bridge incline, about a mile downstream of the metering signals. These detectors allowed for counting vehicles, as well as roughly measuring their speed by using an approximation based on total "on" time for each vehicle (i.e. the longer the vehicle "occupies" the detector, the slower it is going). The target was to control such that vehicle speeds were over 35-40 mph. A rough computation of the current capacity of the bridge in passenger car equivalents per minute (more on this later) was based on vehicle counts and speeds at this location, and the number of vehicles released was adjusted accordingly.

The middle two lanes were reserved for bus and car pool traffic. These signals were generally kept green except when a detected (or manually indicated) downstream incident required short-term restricling of all traffic onto the bridge. The total flow rate of vehicles released onto the bridge included bus and car pool vehicles - so if more of them were counted during a particular minute, fewer metered vehicles were released.

The signal "cycle rate" was limited, both maximum and minimum. Experiment indicated that drives could not react to a fast cycle (shorter than 5 seconds green-to-green) without losing synchronization to the signal; and that many drivers were not willing to wait more than 15 seconds for a green signal. As a result, the total span of control in each lane ranged from 4 vehicles per minute to 12 vehicles per minute.

Trucks and buses were considered to have a greater "impact" on the traffic flow than smaller commuter vehicles, due to their inherent inability to accelerate quickly. An attempt was made to identify trucks and buses as they passed the "passage" detectors (by looking for particular detector actuation "signatures"), and to count them as 1.5 (buses) or 2.5 (trucks) passenger car equivalents. As a result, every truck or bus detected caused fewer green signal actuations during the next minute.

As a result of all of this real-time manipulation, and adjustments based on experience, the meter appeared to have a great deal of "intelligence" and made very effective traffic control decisions. The start and end of metering was automatic, within limits - and we found that on rainy or foggy days, metering would start earlier and end later, reflecting the reduced bridge capacity caused by these conditions.

All of the automation of the metering system was accomplished by means of software running on a Data General Nova 1210 minicomputer with 8K of core memory and no operating system. Code was written in assemply language using a teletype, and punched onto paper tape (both source and assembled machine code). Every vehicle detector was a bit on a parallel input device, sampled 60 times per second, and every signal light was represented as a "0" (off) or "1" (on) on a parallel output device.