1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2:
3: \documentclass{iopart}
4: \usepackage{epsfig,subfigure}
5: \usepackage{citesort}
6: \newcommand{\DIR}{figures}
7:
8: \begin{document}
9: \jl{1}
10:
11:
12: \def\NEW#1{{\bf{#1}}\marginpar{$\longleftarrow$ {\bf NEW!}}}
13: \def\ADD#1{{\bf{#1}}\marginpar{$\longleftarrow$ {\bf ADD!}}}
14: \def\WOLFGANG#1{{\bf{#1}}\marginpar{$\longleftarrow$ {\bf WOLFGANG!}}}
15:
16: \title{Optimization potential of a real highway network: an empirical study}
17:
18:
19: \author{Wolfgang Knospe$^1$, Ludger Santen$^2$, Andreas Schadschneider$^3$
20: and Michael Schreckenberg$^1$}
21:
22: \address{$^1$ Theoretische Physik, Fakult\"at 4,
23: Gerhard-Mercator-Universit\"at Duisburg,
24: Lotharstr. 1, 47048 Duisburg, Germany}
25:
26: \address{$^2$ Fachrichtung Theoretische Physik, Universit\"at des
27: Saarlandes, Postfach 151150, 66041 Saarbr\"ucken, Germany}
28:
29: \address{$^3$ Institut f\"ur Theoretische Physik,
30: Universit\"at zu K\"oln,
31: Z\"ulpicher Str. 77, 50937 K\"oln, Germany }
32:
33: \date{\today}
34:
35: \pacs{45.70, 05.60, 02.50}
36:
37:
38:
39: \begin{abstract}
40: Empirical observations and theoretical studies indicate that the
41: overall travel-time of vehicles in a traffic network can be optimized
42: by means of ramp metering control systems. Here, we present an
43: analysis of traffic data of the highway network of
44: North-Rhine-Westfalia in order to identify and
45: characterize the sections of the network which limit the performance,
46: i.e., the bottlenecks. It is
47: clarified whether the bottlenecks are of topological nature or if they are
48: constituted by on-ramps. This allows to judge possible optimization
49: mechanisms and reveals in which areas of the network they have to be
50: applied.
51: \end{abstract}
52:
53:
54:
55: \section{Introduction}
56:
57:
58: In the last few years, the increasing amount of vehicular traffic has led to a
59: more and more oversaturated freeway network. In contrast to the past,
60: it is no longer possible to react on this additional amount of traffic
61: by construction of new highways.
62: Therefore, in order to use the existing network more efficiently,
63: several control
64: strategies like signal control, variable message signs, route
65: guidance, motorway-to-motorway control and ramp
66: metering (Papageorgiou, 1995) have been proposed for the optimization
67: of the highway's performance and for the amelioration of the system's
68: traffic state.
69:
70: Empirical studies of parts of a highway network (Salem, 1995) have
71: revealed the benefits of ramp metering strategies for the total mean speed
72: and the total travel-time for both, the vehicles on the highway {\it and}
73: the queuing vehicles on the on-ramp.
74: These empirical findings are underscored by several theoretical
75: investigations of simplified (Kolomeisky, 1998; Appert, 2001) as
76: well as well as of more realistic (Popkov, 2001; Huberman, 1999;
77: Barlovic, submitted) traffic models. In
78: particular, studies of the asymmetric simple exclusion process
79: reveal the existence of a maximal current phase which obviously is the most
80: desirable state. It can be obtained for large input and
81: small output rates and allows to operate a highway in the
82: optimal regime. The possible capacity gain is even
83: higher in more realistic models which reproduce the empirically
84: observed meta-stable high-flow states. In finite systems these
85: states are temporally quite stable and, therefore, it seems
86: to be possible to optimize at least short highway sections in
87: this manner.
88:
89:
90: Moreover, in (Huberman, 1999), a simple simulation setup consisting
91: of a two-lane highway segment
92: with an on-ramp is presented. It has been shown that the
93: travel-time of all vehicles can be optimized at a finite injection
94: rate of the on-ramp in the sense that its variance is minimized.
95:
96: This improvement of the system's throughput can be explained by
97: considering fluctuations of the traffic flow. Obviously, free flow is
98: the most desirable traffic state since large flows can be
99: obtained. However, at increasing traffic demand, synchronized traffic
100: forms in the vicinity of on-ramps, which enables a large amount of
101: vehicles to drive with a velocity considerably larger than in a
102: jam but smaller than in free flow, and leads to a flow
103: comparable to free flow (Kerner, 2001).
104: Synchronized states are not very stable, i.e., already
105: small perturbations can cause wide jams. The importance of
106: synchronized traffic for an optimized usage of the network
107: capacity is due to its role as a precursor of traffic jams:
108: If synchronized traffic is observed at a bottleneck
109: control mechanisms will help to stabilize
110: this phase and, thus, ensure large throughputs of the
111: highway at large densities.
112: In particular, by means of ramp metering, it is possible to tune the
113: injection rate
114: of the on-ramp in order to affect the vehicle-vehicle interactions.
115: Increasing the interactions will improve the vehicle coupling and help to
116: synchronize them (Ariaratnam, 2001), however, strong interactions
117: lead to perturbations
118: of the flow and, thus, to wide jams. The variation of the on-ramp flow
119: therefore helps to adjust the interactions and to stabilize the
120: synchronized state. As a result, the perturbations are minimized, and
121: the performance of the highway is optimized.
122:
123: In this study we present an extensive analysis of a highway network
124: which gives insight about the spreading of jams in the
125: network. In contrast to
126: former analyses which are restricted to only small parts of a highway
127: system, here, the coverage of the network with inductive loops allows
128: to analyse the overall traffic state on a global scale. This enables
129: us to identify and characterize its bottlenecks. In particular, the
130: question is raised whether bottlenecks are of topological nature or if
131: they are
132: constituted by on-ramps with a large inflow.
133: Topological bottlenecks are static whereas dynamical bottlenecks (like
134: ramps) in principle can be influenced externally, e.g., by controlling
135: their strength.
136: Moreover, the properties
137: of the bottlenecks may give evidence if
138: the network's state can be optimized by local control devices only
139: or whether global traffic management strategies are
140: necessary (Brockfeld, 2001).
141:
142:
143: \begin{figure}
144: \begin{center}
145: \includegraphics[width=0.63\linewidth]{\DIR/nrw_bw.eps}
146: \caption{Sketch of the highway network of North
147: Rhine-Westfalia and position of the main highways, the A1, the A3, the A57 and
148: the A40.}
149: \label{nrw}
150: \end{center}
151: \end{figure}
152:
153:
154:
155: \begin{figure}
156: \begin{center}
157: \includegraphics[width=0.63\linewidth]{\DIR/nrw_loops.eps}
158: \caption{Position of the inductive loops (black dots). As one can
159: see, the inductive loops concentrate on the inner region of the
160: network where most of the traffic volume can be found.}
161: \label{nrw_loops}
162: \end{center}
163: \end{figure}
164:
165:
166: \section{Network characteristics}
167:
168: We examine the highway network of North Rhine-Westfalia (see
169: Fig.~\ref{nrw} for a sketch of the network) which has
170: a total length of
171: about $2~500~km$ and includes $67$ highway intersections and $830$ on- and
172: off-ramps. The data set is provided by about $3~500$ inductive loops
173: which send minute aggregated data of the flow, the occupancy and the
174: velocity online (that is every minute) via permanent lines from the
175: traffic control centers in Recklinghausen and in Leverkusen. For the
176: sake of simple data handling, the driving directions are classified
177: into south/west and north/east.
178:
179:
180: As one can see in Fig.~\ref{nrw_loops}, most of the detectors are
181: concentrated on the inner network. This is justified by the
182: distribution of the traffic volume, as we will show later.
183: The
184: recording of the data started on 10-10-2000. For the analysis of the
185: network, a period of $265$ days from 10-10-2000 to 07-01-2001 was considered.
186:
187:
188: The highway network of North Rhine-Westfalia has an average traffic
189: load of about $30~000$ veh/24h per measurement section for one
190: driving direction. Note that the traffic volume is calculated
191: per measurement section and thus sums the flow of two or
192: three lanes. However, there
193: are large differences between the highways (Fig.~\ref{kfz}).
194: Obviously, only a few sections with very large traffic
195: volumes exist which concentrate on the main urban areas. In detail,
196: the K\"olner Ring, especially the section between the highway
197: intersections Kreuz Leverkusen and Kreuz K\"oln Ost, large parts of
198: the A3 and the
199: A40, the A57 near Neuss and the A2 near the highway intersection Kreuz
200: Duisburg have an average load of $40~000-80~000$ veh/24h.
201: The more the highways leave these conurbations, the less is the traffic
202: volume which can be measured.
203: In addition, nearly the same values of the traffic volumes can be
204: found for both, the driving directions south/west and north/east.
205:
206:
207:
208: \begin{figure}[h]
209: \begin{center}
210: \includegraphics[width=0.63\linewidth]{\DIR/kfz_bw.eps}
211: \caption{Number of vehicles per measurement location per $24$
212: hours and driving direction.}
213: \label{kfz}
214: \end{center}
215: \end{figure}
216:
217:
218:
219:
220: \begin{figure}[h]
221: \begin{center}
222: \includegraphics[width=0.63\linewidth]{\DIR/lkwshare_bw.eps}
223: \caption{Relative number of trucks per measurement section.}
224: \label{lkwshare}
225: \end{center}
226: \end{figure}
227:
228:
229:
230:
231: About $15\%$ of the vehicles are trucks.
232: Fig.~\ref{lkwshare} shows the distribution
233: of the relative number of trucks per measurement section.
234: Obviously, the share of trucks on the traffic volume
235: increases with the distance to the conurbations, while
236: the absolute number of trucks does not change significantly on the main
237: highways. Thus, the long-distance traffic is mainly dominated by
238: trucks, while at short distances commuter traffic leads to large
239: traffic loads of the network.
240:
241:
242:
243:
244:
245:
246:
247: \section{Bottleneck localization}
248:
249: Since the traffic volumes are simple integrated quantities, it is possible
250: that a large traffic volume can indicate that a two-lane road is often
251: jammed, while the capacity of a three-lane road is still not reached.
252: Therefore, in order to allow an evaluation of the traffic load, the
253: probability to find a jam is calculated. This has the advantage that
254: traffic volumes which are larger than the highway capacity can
255: be related directly to a large jam probability, and thus bottlenecks of
256: the highway network can clearly be identified.
257: Due to the size of the analysed data set it is necessary
258: to introduce a simple criterion to identify jams. Our criterion
259: is the following: If the density
260: of at least one minute is larger than $50\%$ (the density is
261: given as occupancy which is the percentage of time a detector
262: is covered by a vehicle), a jam is supposed. This criterion
263: was motivated by former empirical studies (Knospe, 2002)
264: that indicated that
265: densities higher than $50\%$ are typically neither observed
266: in free-flow nor in synchronized traffic. The threshold value as
267: well as the chosen time interval are, of course, somehow arbitrary.
268: Our results are, however, rather insensitive upon an increase of
269: the density threshold. By contrast, longer time intervals may
270: lead to problems, because jams are often not compact but
271: include regions of free flow traffic. This implies that the
272: density threshold has to be lowered if longer time intervals
273: are used, but lower density threshold lead to problems in
274: discriminating between jams and synchronized traffic.
275:
276:
277: The usage of the density as an indicator for a jam has some advantages
278: compared to the mean velocity and the flow. First, the flow as well as
279: the velocity depend on speed limits which vary in the
280: network. Therefore, it is not
281: possible to distinguish whether a small velocity is due to a large
282: density or a speed limit.
283: Second, a small flow can refer to both a jam or just free flow
284: with only a few cars. The jam probability is therefore calculated as
285: the relation of the number of days a jam was found to the total number of
286: days of the observation period ($265$ days).
287:
288: Fig.~\ref{pjam034} represents the jam probability of the
289: network for the driving direction south/west. Obviously, large jam
290: probabilities can be found in a region of the
291: inner network, where the conurbations of the Ruhrgebiet (region 1), the areas of
292: Dortmund (region 2), D\"usseldorf and Krefeld (region 3) and K\"oln
293: (region 4) are located.
294: As one can see in Fig.~\ref{nrw_loops}, most of the inductive loops are
295: concentrated in this regions, which are indeed the most crowded
296: sectors of the network.
297: In contrast, in regions which are less equipped with
298: counting devices, only few jams can be found.
299:
300:
301: Large differences of the jam probability
302: between the areas can be observed.
303: Only $10\%-30\%$ of the observation days were jammed in the area of
304: the Ruhrgebiet (region 1), that is every third day a jam occurred. In
305: contrast, in the areas of Dortmund (region 2), Krefeld and
306: D\"usseldorf (region 3) and K\"oln (region 4), more than $50\%$ of the
307: days showed
308: jams. Especially in
309: the area of Dortmund, between the Westhofener Kreuz and the Kreuz
310: Dortmund/Unna, nearly $5$ days a weak jams can be measured.
311:
312: Next, we analyse {\it when} the jams typically occur. First, we
313: expect significant differences between the weekend and
314: working days.
315:
316: \begin{figure}[h]
317: \begin{center}
318: \includegraphics[width=0.63\linewidth]{\DIR/pjam034_bw.eps}
319: \caption{Jam probability for the driving direction south and
320: west. The encircled regions are the Ruhrgebiet (region 1), the area of
321: Dortmund (region 2), the area of Krefeld and D\"usseldorf (region 3)
322: and the area of K\"oln (region 4).}
323: \label{pjam034}
324: \end{center}
325: \end{figure}
326:
327:
328:
329: \begin{figure}[h]
330: \begin{center}
331: \includegraphics[width=0.63\linewidth]{\DIR/pjam034_wednesday_bw.eps}
332: \caption{Jam probability on wednesday for the driving
333: directions south and west.}
334: \label{pjam034_wednesday}
335: \end{center}
336: \end{figure}
337:
338:
339: Indeed, the exclusion of weekends leads to an increased jam
340: probability on weekdays compared to the
341: probability averaged over the whole observation period.
342: Second, we regarded the daily variations of the jam
343: probabilities. During the week from monday to friday
344: only minor differences in the jam probability exist, but
345: one has to distinguish the period from monday to
346: thursday and the friday as we will show below.
347: The jam probabilities we obtained are qualitatively
348: the same as in case of an average over the whole
349: data set (see Fig.~\ref{pjam034_wednesday}),
350: the same four regions show a large jam probability
351: that is the Ruhrgebiet and the areas
352: of Dortmund, Krefeld and D\"usseldorf and K\"oln.
353: The actual value of the jam probability, however, is increased and the
354: difference to the remainder of the network is now more pronounced.
355: During the weekend the jam probability is considerably
356: reduced. Nevertheless, there are a few locations in the network where high jam
357: probabilities can be observed on sundays, i.e., between the Westhofener
358: Kreuz and the Kreuz Dortmund/Unna.
359: Remarkably the same classification scheme was found to be
360: appropriate for city data as discussed
361: in (Chrobok, 2000; Chrobok, 2001).
362:
363:
364: \begin{figure}[p]
365: \begin{center}
366: \includegraphics[width=0.63\linewidth]{\DIR/pjam034_6_11_bw.eps}
367: \caption{Jam probability for the driving direction south and west
368: averaged over the whole observation period for the time interval
369: $6$-$11$ a.m.}
370: \label{pjam034_611}
371: \end{center}
372: \end{figure}
373:
374: Next, we analyse the intraday variations of the data set.
375: We further divided the time-series into five time intervals, in order to
376: capture the main rush-hours. The first interval
377: covers the time from $0$ to $6$ a.m., the second interval from $6$ to
378: $11$ a.m., the third interval from $11$ a.m. to $1$ p.m., the fourth
379: interval from $1$ to $8$ p.m. and the fifth interval from $8$ to $12$
380: p.m. Indeed, the division of a day is somehow arbitrary, but
381: a finer discretization of a day will not change the results
382: significantly. Moreover, analyses of city data revealed the existence
383: of basically two rush-hour peaks in the time-series of the traffic
384: volume at about $8$ a.m.~and $4$ p.m.~(Chrobok, 2000; Chrobok, 2001).
385:
386:
387: \begin{figure}[p]
388: \begin{center}
389: \includegraphics[width=0.63\linewidth]{\DIR/pjam034_13_20_bw.eps}
390: \caption{Jam probability for the driving direction south and west
391: averaged over the whole observation period for the time interval $1$-$8$ p.m.}
392: \label{pjam034_1320}
393: \end{center}
394: \end{figure}
395:
396:
397:
398: Figures~\ref{pjam034_611} and~\ref{pjam034_1320}
399: show the jam probability of the whole data set in
400: the driving direction south/west for
401: two time intervals. During the first time interval, that is $0$ to
402: $6$ a.m., no jams could be measured. The main traffic volume occurs in
403: the second and in the fourth interval, which include the rush-hours.
404: %7In the time
405: From $11$ a.m.~to $1$ p.m.~only a few jams are visible, while
406: from $8$ p.m.~to $12$ p.m.~just one region shows jams, namely the
407: region between the Westhofener Kreuz and the Kreuz Dortmund/Unna.
408: The jams therefore occur mainly in the morning and afternoon
409: rush-hours. There are, however, parts of the network where
410: the load is so large, that the probability to find a jam is high
411: throughout the whole day.
412:
413: The same picture can be drawn from the analysis of the data set
414: classified in single days
415: which are divided into the five time intervals.
416: Due to the exclusion of the weekends, the values of the jam
417: probability increase significantly.
418: The main disturbances of the network can be measured in the
419: rush-hours, while the regions $2$ and $4$ show continuously
420: a large traffic load.
421: Again, there are only small differences between the weekdays.
422: On fridays on the one hand, the load of the network in
423: the time between $6$ and $11$ a.m.~is smaller, but, on the other hand,
424: is
425: increased significantly between $1$ and $8$ p.m.~compared to monday to
426: thursday.
427: Moreover, on sundays, the large jam probability in the area of Dortmund
428: can be traced back to congestions between $1$ to $8$ p.m.
429:
430:
431: \begin{figure}[p]
432: \begin{center}
433: \includegraphics[width=0.63\linewidth]{\DIR/pjam002_bw.eps}
434: \caption{Jam probability for the driving directions north and east
435: averaged over the whole data set. The encircled regions are the
436: Ruhrgebiet (region 1), the area of
437: Dortmund (region 2), the area of Krefeld and D\"usseldorf (region 3)
438: and the area of K\"oln (region 4).}
439: \label{pjam002}
440: \end{center}
441: \end{figure}
442:
443:
444:
445: Analogously to the analysis of the data of the driving direction
446: south/west, the jam probability is calculated for the
447: direction north/east (Fig.~\ref{pjam002}).
448: Like for the driving direction south/west, the
449: same regions with large jam probabilities occur which can be identified as
450: the conurbations of the Ruhrgebiet (region 1), the areas of Dortmund
451: (region 2), Krefeld and
452: D\"usseldorf (region 3) and K\"oln (region 4).
453: In contrast, the region of the Ruhrgebiet (region 1) and the area Krefeld and
454: D\"usseldorf (region 3) have a larger jam probability in the direction
455: north/east, while in
456: the direction south/west the area of K\"oln (region 4) shows a
457: larger traffic
458: volume. In the area of Dortmund (region 2), a large jam probability
459: can be found in the
460: south of the Westhofener Kreuz.
461: Because north of the Westhofener Kreuz in the driving
462: direction south a large jam probability was measured, too, one
463: can clearly identify the Westhofener Kreuz as a dominant bottleneck.
464:
465: Again, the distinction between the days shows no difference to the
466: analysis of the whole observation period with the exception of a
467: larger jam probability in the single regions.
468: In addition, the main traffic volume can be measured in the morning
469: and afternoon rush-hours.
470:
471:
472: \section{Spatial extension of jams}
473:
474: Calculating the jam probability in the four regions, one can observe
475: that consecutive measurement sections are often jammed even at the
476: same time, which may be a consequence of jams that have a large spatial
477: extension. However, as one can see further, a sequence of jammed
478: detectors is always restricted by highway intersections, not by on-
479: and offramps. But more importantly a jam which branches
480: from one highway to
481: another highway via an intersection has been observed only in one case
482: (see below).
483:
484:
485:
486: In order to determine the spatial extension of a jam,
487: it is necessary to consider its temporal occurrence:
488: If for a given day a jam was detected at a detector, the
489: detector is considered to be active (1) if not as being passive (0).
490: In this way we generated %This results in
491: a binary time-series
492: which allows the calculation of the spatial correlation between
493: different measurement locations. The explicit calculation
494: of the
495: correlation of the density time-series, in contrast, has the
496: disadvantage that the
497: duration of jams decreases in upstream direction,
498: which leads to vanishing correlations.
499:
500: \begin{figure}[p]
501: \begin{center}
502: \includegraphics[width=0.63\linewidth]{\DIR/cc_034_4114074_bw.eps}
503: \caption{Spatial correlation of the activity in the driving directions
504: south and west near the highway intersection Kreuz Dortmund/Unna.
505: Correlation values larger than $0.3$ are shown.}
506: \label{cc034_4114074}
507: \end{center}
508: \end{figure}
509:
510: \begin{figure}[p]
511: \begin{center}
512: \includegraphics[width=0.63\linewidth]{\DIR/cc_002_6257028_bw.eps}
513: \caption{Spatial correlation of the activity in the driving directions
514: north and east near the highway intersection Kreuz Kaarst.
515: Correlation values larger than $0.3$ are shown.}
516: \label{cc002_6257028}
517: \end{center}
518: \end{figure}
519:
520:
521:
522: Although the activity gives only a rough estimate of the jamming
523: state at an inductive loop, the typical spatial extension of jams
524: can clearly be seen.
525: The large jam probability in region 2 in the driving direction
526: south/west can be related to a
527: jamming of large parts of the A1 between the Westhofener Kreuz and the
528: Kreuz Dortmund/Unna (Fig.~\ref{cc034_4114074}) since the correlation
529: shows large values.
530: The same picture can be drawn in the area of K\"oln (region 4) where
531: a jam may
532: cover a distance from the Kreuz K\"oln West
533: and from the Kreuz K\"oln Ost to the Kreuz Leverkusen.
534: The largest
535: extension of a jam can be found on the A40 (region 1) where strong correlations
536: between the Kreuz Duisburg and the Kreuz Bochum
537: are measured.
538: Obviously, there are also correlations with highways that are very
539: far away from the reference detector, but which are not connected by a
540: sequence of succeeding inductive loops with a large jam probability.
541: The correlations are simply a consequence of a large
542: traffic load of the network which leads to many jams on various
543: highway sections.
544: Since the number of jams
545: counted at the
546: reference detector is large, correlations of a free flow signal can be
547: excluded.
548:
549: The correlation analysis of the driving direction north/east reveals
550: strong values on the A57 near the Kreuz Kaarst
551: (Fig.~\ref{cc002_6257028}) and again on the A40 between the Kreuz Bochum
552: and the Kreuz Duisburg (Fig.\ref{cc002_5902002}).
553: A jam measured at the Westhofener Kreuz (Fig.\ref{cc002_4115017}) can
554: branch due to the highways intersection on both, the highway A45 and
555: the highway A1.
556:
557:
558:
559: The analysis of the activity allows for the daily occurrence of jams,
560: but temporal differences due to the morning or afternoon rush-hours
561: are omitted. Therefore, the activity is classified into the five
562: periods used for the jam probability calculation. This additional
563: consideration of the temporal occurrence of the activity allows a
564: better determination of the spatial correlation of the measurement
565: locations.
566:
567:
568:
569:
570:
571: The picture of large
572: congested areas on the highways can be confirmed by the classified
573: activity correlation analysis. Especially the A40 between the Kreuz
574: Bochum and the Kreuz Duisburg in the
575: directions east and west, the A57
576: in the vicinity of the Kreuz
577: Kaarst in the direction north and the area
578: of K\"oln in the direction south/west
579: (Fig.~\ref{cc034p_4133047}) show strong
580: correlations of
581: consecutive measurement locations.
582: Since the finer discretization of the activity time-series does
583: not change the results significantly, the classification of
584: the activity into five intervals is sufficient
585: for the proper recording of the temporal occurrence of jams.
586: Thus, although the classified activity is only a rough estimate for
587: the density time-series of a detector, the correlation analysis
588: nevertheless allows the determination of the spatial extension of jams.
589:
590: The correlation analysis supports the results drawn from the jam
591: probability calculation. Although jams can have a large spatial
592: extension, branching of a jam via a highway intersection is rarely
593: observed. As a consequence one can conclude that the bottlenecks are,
594: apart from the Westhofener Kreuz, a result of perturbations due to
595: on- and offramps rather then local capacity restrictions.
596: As one can see in Fig.~\ref{sources} and Fig.~\ref{sinks}, most of the
597: sources of the network with a large
598: injection rate are concentrated in the four regions. In addition, most
599: of the sinks can be found there~\footnote{The source and sink rate is
600: calculated by summing up the flow of all lanes at a measurement
601: section. The difference of the total flow between two succeeding
602: sections gives the loss or gain of the number of
603: vehicles.}. Therefore, traffic in these four
604: regions is determined by strong fluctuations of the traffic demand
605: and, thus, the flow, resulting in jams with a large spatial
606: extension. However, these jams are restricted to single highways
607: and do not affect traffic in other parts of the network.
608:
609: Nevertheless, an example of a topological bottleneck is given in
610: Fig.~\ref{lirich_before}. The bottleneck is located north of the junction
611: Oberhausen-Lirich on the A3 and is generated by a reduction from three
612: to two lanes in each driving direction. In September 2001 this
613: bottleneck was eliminated by the extension to three lanes.
614: Due to the road works, jams regularly emerged in the south of the
615: bottleneck. These jams could have an extension up to the Kreuz
616: Breitscheid, passing the Kreuz Kaiserberg undisturbed.
617: For comparison, Fig.~\ref{lirich_after} shows the same part of the
618: network after the road constructions have been finished. Obviously, the
619: probability to find a
620: jam on the A3 south of the junction Oberhausen-Lirich is reduced
621: drastically. A systematical impact on other parts of the highway
622: network, however, cannot be observed.
623:
624: \begin{figure}[p]
625: \begin{center}
626: \includegraphics[width=0.63\linewidth]{\DIR/cc_002_5902002_bw.eps}
627: \caption{Spatial correlation of the activity in the driving directions
628: north and east near AS Bochum-Zentrum.
629: Correlation values larger than $0.3$ are shown.}
630: \label{cc002_5902002}
631: \end{center}
632: \end{figure}
633:
634:
635:
636: \begin{figure}[p]
637: \begin{center}
638: \includegraphics[width=0.63\linewidth]{\DIR/cc_002_4115017_bw.eps}
639: \caption{Spatial correlation of the activity in the driving directions
640: north and east near the highway intersection Westhofener Kreuz.
641: Correlation values larger than $0.3$ are shown.}
642: \label{cc002_4115017}
643: \end{center}
644: \end{figure}
645:
646:
647:
648: \begin{figure}
649: \begin{center}
650: \includegraphics[width=0.63\linewidth]{\DIR/cc_034_period_4133047_bw.eps}
651: \caption{Spatial correlation of the classified activity in the driving
652: directions
653: south and west near the highway intersection Kreuz K\"oln-Nord.
654: Correlation values larger than $0.3$ are shown.}
655: \label{cc034p_4133047}
656: \end{center}
657: \end{figure}
658:
659: \begin{figure}
660: \begin{center}
661: \includegraphics[width=0.63\linewidth]{\DIR/sources_34.eps}
662: \caption{Source rate in the driving directions south and west.}
663: \label{sources}
664: \end{center}
665: \end{figure}
666:
667: \begin{figure}
668: \begin{center}
669: \includegraphics[width=0.63\linewidth]{\DIR/sinks_34.eps}
670: \caption{Sink rate in the driving directions south and west.}
671: \label{sinks}
672: \end{center}
673: \end{figure}
674:
675: \begin{figure}
676: \begin{center}
677: \includegraphics[width=0.63\linewidth]{\DIR/pjam002_october2000_zoom_bw.eps}
678: \caption{Jam probability in the driving direction north and east in
679: October 2000 during the road constructions north of the junction
680: Oberhausen-Lirich.}
681: \label{lirich_before}
682: \end{center}
683: \end{figure}
684:
685: \begin{figure}
686: \begin{center}
687: \includegraphics[width=0.63\linewidth]{\DIR/pjam002_october2001_zoom_bw.eps}
688: \caption{Jam probability in the driving direction north and east in
689: October 2001 after finishing the road construction at
690: Oberhausen-Lirich.}
691: \label{lirich_after}
692: \end{center}
693: \end{figure}
694:
695:
696:
697: \section{Conclusion}
698:
699: The analysis of the highway network of North Rhine-Westfalia leads to
700: the following results: The main traffic load concentrates on the inner
701: network where the conurbations of the Ruhrgebiet (region 1), the areas of
702: Dortmund (region 2), D\"usseldorf and Krefeld (region 3) and K\"oln
703: (region 4) are located. In these regions, traffic is predominantly determined
704: by commuter or local traffic. In contrast, the outer regions of the
705: network are dominated by long-distance traffic, indicated by a
706: large truck share on the total traffic volume.
707: Jams can be measured very often in the central part of the network
708: i.e., jam probabilities of more than $50\%$ have been measured.
709: This picture can be confirmed by a
710: more detailed analysis of the traffic data which takes the daily
711: occurrence of jams into account. Like in~(Chrobok, 2000; Chrobok, 2000),
712: the days can be classified into monday to thursday, friday, saturday
713: and sunday. However, the time a jam emerges is considered by the
714: subdivision of one day into five intervals. As a result, most of the
715: jams can be traced back on the morning and afternoon rush-hours, but
716: for the areas of K\"oln (region 4) and Dortmund (region 2), the
717: probability to find a jam is
718: large during the whole day. In addition, on fridays, the main traffic
719: load is shifted from the morning to the afternoon rush-hour.
720:
721: The correlation analysis of the activity takes the temporal occurrence
722: of a jam at consecutive measurement locations into account. Large
723: correlations between succeeding detectors in the four regions can be
724: measured. Nevertheless,
725: the jams are mainly restricted to single highways and do not branch on
726: other highways via intersections. The only bottleneck which can be
727: identified is the highway intersection Westhofener Kreuz.
728: A more detailed calculation of the
729: correlation by means of a classified activity confirms these results.
730:
731: The bottlenecks of the network are therefore predominantly on- and
732: offramps rather than topological peculiarities of the highway system. It is
733: the large in- or outflow at ramps that perturbes the stream of vehicles
734: on the main highway. As proposed in (Salem, 1995; Huberman, 1999), it
735: has to be expected that ramp metering systems are able to counteract
736: these destabilization of the flow and reduce the formation of jams.
737: In this way a restricted flow on the ramps may lead to a
738: significant increase of the capacity of the main highways.
739: Of special interest for a capacity optimization is the
740: existence of some ``hot spots'' in the network: The highest
741: jam probabilities are spatially and temporally well localized.
742: These are the parts of the network, where the flow
743: has to be optimized. Presently these sections self-organize in
744: a congested state, leading fluxes that are far below their capacity.
745: The situation can be improved a lot by controling the number
746: of entering cars and by optimizing the traffic stream at the
747: on- and off-ramps at this section.
748: The small number of bottlenecks that are present in the
749: network shows that it is possible
750: to improve the capacity of the network with a reasonable
751: technical effort. Of particular interest are sections where the
752: main input is from other, less crowded, highways. In these
753: cases a restricted input does not lead to a collapse of the
754: urban traffic.
755:
756: There is of course another way to avoid congestion in conurbations.
757: Our analysis shows that the jams are caused mainly by commuters.
758: This means that a better usage of public transport in the urban
759: areas would indeed reduce significantly the congestion at the
760: conurbations.
761:
762: Summarizing our analysis has shown that the bottlenecks of a highway
763: network can be found by means of a simple statistical analysis.
764: The distribution of the bottlenecks in the network indicates that
765: optimization strategies can be successfully applied in order to increase
766: the capacity of the network.
767:
768:
769:
770:
771: \vspace{0.2cm}
772:
773: {\bf Acknowledgments}:
774: The authors are grateful to the Landesbetrieb Stra\ss enbau NRW
775: for the data support and to the
776: Ministry of Economics and Midsize Businesses, Technology and Transport
777: as well as to the Federal Ministry of Education and Research of
778: Germany for the financial support (the latter within the BMBF project
779: ``DAISY'').
780:
781:
782:
783: \section*{References}
784: \begin{description}
785:
786: \item[Appert, C.,] and L.~Santen. (2001).
787: "Boundary Induced Phase Transitions in Driven Lattice Gases with
788: Metastable States."
789: {\em Physical~Review~Letters} 86, 2498--2501.
790:
791: \item[Ariaratnam J.~T.,] and S.~H.~Strogatz (2001).
792: "Phase Diagram for the Winfree Model of Coupled Nonlinear Oscillators."
793: {\em Physical~Review~Letters} 86, 4278--4281.
794:
795: \item[Barlovic, R.,] T.~Huisinga, A.~Schadschneider,
796: and M.~Schreckenberg. (submitted).
797: "Open boundaries in a cellular automaton model for traffic
798: flow with metastable states."
799:
800:
801: \item[Brockfeld E.,] R.~Barlovic, A.~Schadschneider,
802: and M.~Schreckenberg. (2001).
803: "Optimizing Traffic Lights in a Cellular Automaton Model for City Traffic."
804: {\em Physical Review E} 64, article number: 056132.
805:
806: \item[Chrobok~R.,] O.~Kaumann, J.~Wahle,
807: and M.~Schreckenberg. (2000).
808: "Three Categories of Traffic Data: Historical, Current, and Predictive."
809: In E. Schnieder, and U.~Becker (eds.),
810: {\em 9th IFAC Symposium Control in Transportation Systems.}
811: Braunschweig: IFAC, pp.~250--255.
812:
813:
814: \item[Chrobok~R.,] J.~Wahle, and
815: M.~Schreckenberg. (2001).
816: "Traffic Forecast Using Simulations of Large Scale Networks."
817: In B.~Stone, P.~Conroy, and A.~Broggi (eds.),
818: {\em 4th International IEEE Conference on Intelligent
819: Transportation Systems.},
820: Oakland: IEEE, pp.~434--439.
821:
822: \item[Huberman B.A.,] and D.~Helbing. (1999).
823: "Economics-based optimization of unstable flows."
824: {\em Europhysics Letters } 47, 196--202.
825:
826: \item[Kerner B.S.] (2001).
827: "Complexity of Synchronized Flow and Related Problems for Basic
828: Assumptions of Traffic Flow Theories."
829: {\em Networks and Spatial Economics} 1, 35--76.
830:
831:
832: \item[Knospe~W.,] L.~Santen, A.~Schadschneider,
833: and M.~Schreckenberg. (2002).
834: "Single-vehicle data of highway traffic: microscopic description
835: of traffic phases."
836: {\em Phys.~Rev.~E}, in press.
837:
838: \item[Kolomeisky A.~B.,] G.~M.~Sch\"{u}tz,
839: E.~B.~Kolomeisky, and J.~P.~Straley. (1998).
840: "Phase diagram of one-dimensional driven lattice gases with open boundaries."
841: {\em Journal of Physics ~A} 31, 6911--6919.
842:
843: \item[Papageorgiou M.] (1995).
844: "An Integrated Control Approach for Traffic Corridors."
845: {\em Transportation Research C} 3, 19--30.
846:
847:
848: \item[Popkov V.,] L.~Santen, A.~Schadschneider,
849: G.~M.~Sch{\"u}tz. (2001).
850: "Empirical evidence for a boundary-induced nonequilibrium phase
851: transition."
852: {\em Journal of Physics A} 34, L45--L52.
853:
854:
855:
856: \item[Salem H.~H.,] and M.~Papageorgiou. (1995).
857: "Ramp Metring Impact on Urban Corridor Traffic: Field Results."
858: {\em Transportation Research A} 29, 303--319.
859:
860: \end{description}
861:
862:
863:
864:
865: \end{document}
866:
867:
868:
869:
870:
871:
872:
873: