CONTENTS

1. INTRODUCTION

To meet the requirements of today's public transportation, streetcars and buses shall have priority over individual traffic. Although using different lanes, at traffic junctions they interfere with each other. Priority treatment of public transportation should not decrease the capacity of individual traffic flow as it is today.

It has been shown (Riedel, 1994) that a fully dynamic traffic junction control enables a better performance than the usual way of sequential or dynamic sequential control, especially when co-ordinating cars and streetcars on the same crossing.

For a fully dynamic traffic control the deterministic control algorithm developed at ETH (Riedel, 1991; Riedel and Brunner, 1993) can only be used when the input information (as the detector signals) is highly reliable. But mostly they are in some way corrupted. For this reason incoming detector signals have to be pre-processed before being used for estimating the state of the crossing. The state then is used as input for the controller.

Processing detector information consists of the following tasks:

Detector evaluations are the base of the observer used for fully dynamic traffic control, developed and implemented for a double traffic crossing in Zurich (Heimplatz). Figure 1 shows the control structure used.

Fig. 1. Block-diagram of the controlled system, consisting of plant, observer, and controller

This paper treats the observer, whereas the control algorithm has already been demonstrated by Riedel and Brunner (1993).

2. THE INPUT DATA

Figure 2 shows a typical sequence of sensor readings by the observer. The three lower curves 508.5 to 508.7 represent streetcar detector signals, 507.8 indicates the position of a switch, 507.7 is a current loop used for save streetcar detection, and 507.6 and 507.6' are ready-signals of streetcars being ready to leave the stop.

Fig. 2. A sequence of detector signals

Figure 3 shows a part of the combination of two crossings in Zurich, that has been taken for implementing the controller, with all detectors. The two junctions have the numbers 507 and 508, as indicated. The example that was shown in figure 2 follows the streetcar access from "Bellevue" towards "ETH".

Fig. 3. Detectors and its numbers at Heimplatz in Zurich

One can easily see in figure 2 that some signals are repeating and others are missing. This has to be corrected by the observer in a three-step-computation described in the next chapter.

3. DATA PROCESSING

3.1 Filtering against repetitions

A software lowpass-filter suppresses high repetition frequencies so that only impulses of a minimal duration may pass the filter. The effect of such filtering is demonstrated in figure 4, showing the average car rate per 5 minutes during one day.

Fig. 4. Filtered and unfiltered car rate

The rate of unreliable signals can be decreased in this step from about 30% to 20%.

3.2 Co-ordination of detectors

By tracking the vehicles through the system – done by a simulator co-ordinated with real world data, see Riedel and Brunner (1993) – erroneous, or missing impulses can be corrected. Measured values of an average speed between two sensors have to be used for comparing the sequence of arriving impulses to the simulation. If too many errors are detected by the observer, a vehicle is no considered any more.

Typical measured speed and travelling time diagrams for streetcars are shown in figure 5.

Fig. 5. Measured speed (upper) and travelling time (lower) diagrams

By this step the information processed by the observer becomes about 99% reliable. The more detectors track the vehicles the more reliable the information can become.

3.3 Supervision against failure

It is difficult to find out rapidly if a detector fails too often than usual. On the other hand it is important for the controller to go to a save, cyclic control as soon as the input data is not correct any more. To obtain such a behaviour, the detectors must be supervised by a third, "neutral "element. This is done by evaluating probabilistic data of each detector, as probabilities and correlated probabilities.

Probability measures can be the time between rising edges of the signal or the "on"-duration. Figure 6 shows measured data from a car detector and figure 7 that form a streetcar detector.

Fig. 6. Probability measures of car detector 507.3

Signals of car detectors are mostly short in time, and they mostly have a high repetition frequency because cars follow each other with short intervals. In contrary streetcars operate approximately according to a given schedule and follow each other with almost deterministic distances.

Fig. 7. Probability measures of streetcar detector 508.5

For correlated probabilities always two following measurements are evaluated together and can be depicted in a three-dimensional diagram as shown in figure 8 for a car sensor.

Fig. 8. Correlated probability measure of time differences of rising slopes for car detector 507.3

Here it is also possible to show distances and lengths of car platoons. The streetcars behave differently and more in a deterministic way, as shown in figure 9.

Fig. 9. Correlated probability measure of time differences of rising slopes for streetcar detector 508.5

By probability and correlated probability diagrams unusual sensor behaviour can be detected very quickly. This enables a protection of the controller against erroneous, capacity-decreasing sensor signals.

4. CONCLUSIONS

It has been shown a way how to pre-process detector signals that are sometimes erroneous and thus not reliable, to obtain information as reliable as possible. Three steps have been identified: filtering, simulation, and supervision of detector signals.

Only in case the detector signals are enough reliable that the few wrong impulses still led to the controller are not of importance any more, it is allowed to use a fully dynamic junction control. This is to obtain much shorter waiting times of vehicles with priority over other vehicles, without a significant increase of waiting time of all other vehicles.

Further research will be done on the theory of processing detector signals by filtering, simulating and observing them. It can be shown (Riedel, 1994) that a good observer allows the use of a simple control algorithm and vice versa. Because the proposed control algorithm (Riedel, 1991), basing on Dynamic Programming, is quite time consuming, a good estimator, not basing on a recursive algorithm, can make the controller run on slower processors than it is possible with today's implementation.

5. REFERENCES

Riedel, Th. (1991). Reglerentwurf für Verkehrskreuzungen, theoretische Aspekte (Some Theoretical Aspects for Designing Traffic Light Controllers). Automatic Control Lab Report No. 91-19, ETH Zürich (also presented with title "A Generic DEDS Control" at: Workshop on Discrete Event Systems, Amherst, MA, June 1991).

Riedel, Th., and U. Brunner (1993). Traffic Control Using Graph Theory. 12th IFAC World Congress, Sydney, July 1993 (appeared in: IFAC Control Engineering Practice).