DE19907041A1 - Alternative route control for influencing road network involves determining non-static demand values for observable traffic parameters, deriving control intervention in successive steps - Google Patents

Abstract

The method involves forming local detection cross-sections using fixed detectors, detecting traffic-related measurement values and processing them to obtain traffic information from which the control intervention is derived in at least a complete process. Non-static demand values for observable traffic parameters, especially measurable parameters, are determined from the data in a processing step, and the control intervention is derived in a further step using the observed parameters.

Description

State of the art

Simple control algorithms for alternate routing, for example for Federal highways are based on "if then" rule bases with threshold value queries. The rules relate directly to the respective road network and are therefore not transferable.

By dividing the control task into a traffic model and Control model becomes largely independent of the concrete Network topology reached. The Dernbach-Koblenz method (FGSV, 1992) uses one simple traffic model for forecasting travel times; the turn rates are as regardless of the current traffic situation. The control model determines the circuit diagram by minimizing travel times.

CREMER (1993) replaces the simple traffic model with a dynamic one macroscopic model. MESSMER (1994) extends the control model so that instead of a circuit diagram, an optimal sequence of circuit diagrams is calculated. SACHSE (1998) defines one based on the Dernbach-Koblenz method Framework plan for control procedures and extends this by an estimate of the Kordon matrices (PLOSS, 1993) of the current traffic relations.

Theoretical considerations regarding the interaction of the real data, the matrix estimation and the prognosis showed that the usual procedures and models in their In addition to simple estimation errors, they also work together to form systematic errors can tend. These considerations were made during extensive data analyzes in the Project TABASCO (KELLER, 1998) and confirmed in SACHSE (1998). There is therefore the Need to extend the classic forecasting and estimation methods or to make it dispensable.

PAPAGEORGIOU (1990) introduces a controller instead of a traffic forecast, which controls the Difference of travel time on an alternative route pair regulates to zero. He differentiates thereby a "predictive" (prediction of travel time) and a "reactive" user optimum, at to which the travel time from the current edge travel times is added. In both Static setpoints are used; however, these do not allow Consideration of dynamic variables such as the compliance rate.  

Based on this prior art, the object of the present invention to replace the faulty models for determining the effects. this happens by using measured effects in a new controller module of the Control procedure. In this paper, therefore, the logical system architecture is out FGSV (1992) expanded using a controller model using dynamic setpoints. This provides an architecture for methods that are "adaptive", both with respect to one another changing system state, as well as previously modeled as time invariant Parameters such as the acceptance rates in network controls. In addition, from the measured effects using methods of artificial Intelligence derives an impact model that is based on the measured values at runtime Effects calibrated. The combined use of the measured effect and with the Model estimated from historical measured values allows an increase in the technical reliability of the process through fallback levels but also a Increased system quality through the use of model-based forecasts.

method Use of the measured effect

The method is based on measuring the effect at the decision point, the. Starting point for alternative routes. The intervention can consist of an individual information system as well as a collective information system, such as a sign for collective route recommendations (see (a) in Fig. 1). The detection is chosen so that the distribution of traffic flows on the left and right edges can be measured. This can be done, for example, with the measuring cross sections b, c and d shown in Fig. 1. The measurement cross sections can be designed, for example, as radar detectors. It is proposed to incorporate the measured values into the control process using automation technology methods. To this end, the previously known basic architecture for control processes is being expanded.

Extension of the basic architecture

This is the case in the previous basic architecture for procedures for influencing traffic Control procedure from a traffic model and a control model. The Traffic model is used for the local-time recording and forecast of traffic. The The control model implements and uses the traffic requirements in a target function the traffic model to determine the intervention.

An expansion of the basic architecture (see Fig. 2) makes it possible that the control model can no longer only rely on classic model-based methods without taking model uncertainties into account. The possibility is created to take into account the estimation errors of the traffic model and to use the actually measured effect instead of the error-prone model.

The observer corresponds to the traffic model. It becomes in the observer Traffic events at least at the current time, or also in the future estimated. The observer provides an estimated system state (t) based on that the measured data y (t) is determined. The algorithm corresponds to the control model. However, the algorithm does not immediately provide a circuit diagram or the Control intervention, but first dynamic setpoints s (t) for quantities that are as good as possible are measurable or observable. In addition to the classical optimization also other approaches from control engineering and / or Heuristics can be used.

The controller calculates the circuit diagram using the measured quantities or the control intervention u (t) so that the control error remains as small as possible. In this way, the traffic model and reality (system) are adapted to each other. Disturbances from z (t) that cannot be modeled explicitly can thus be taken into account implicitly become. This additional feedback of measurement information already enables simple approaches in the submodels observer, algorithm and controller a very good system behavior. This is done through a simulative validation of the below described network control shown.

The dynamic dynamic approach used to decouple the algorithm from the controller Setpoints can not only be used as an interface between the algorithm and the controller, but can also be used within a sub-model; this is illustrated by a heuristic algorithm for network control demonstrated. In addition, the Approach of the dynamic setpoints the potential in a cascade structure Coupling different control methods can be used (POSCHINGER ET AL, 1997).

Structure of the network control

According to the basic architecture, the network control consists of a traffic mo dell (observer), a control model (algorithm) and a controller. The Observer provides the current macroscopic traffic status. The algorithm provides a dynamic setpoint for the (measurable) vehicle current on the Main route. By varying the intervention, the controller ensures that the setpoint is adhered to as much as possible.

The traffic model

The traffic model (the observer) must be aware of the possibilities of detection to adjust. The current state of the art is the cross-sectional measurement of the Traffic flow and speed. In addition, the Lane and vehicle types are taken into account.

Section or network-related measurement methods, for example based on a License plate detection or individual vehicle-related data cannot yet be assumed.  

The observer becomes a real data preprocessing and a Kalman Estimation method based on CREMER (1979). The Kalman filter provides one closed (sub-) optimal approach for estimating model states.

Real data preprocessing is the interface to the control system in online operation or for simulation in offline validation. In line with TLS, the real data preprocessing delivers the traffic flow q _{m of} the past minute extrapolated to vehicles / h and the average speed v _{m of} the past minute in km / h at each measurement cross-section.

The estimation method is based on a location discretization through the segmentation of the road network. The minimum segment length is 200 m. The road network is also divided into subnetworks that are limited by measuring cross sections. There are no measuring cross sections within the subnetworks. The segments are dimensioned so that the minimum segment length is maintained and the number of segments per measurement cross section does not exceed three. This ensures observability (CREMER, 1979). The traffic variables speed v _i and density k _i are regarded as locally constant mean variables of segment i. Turning rates tr _mn are also introduced at nodes; these turn rates describe the proportion of the density k _{m of} the segment in front of the node whose target is segment n after the node. The formulas described in CREMER (1979) and patented in GR 97 P 8623 as a process are formally expanded to include the turn rates, which are understood as further system states. The sizes v _i , k _i , and tr _mn are estimated and are available to the control model. The estimation method described can be replaced by other estimation methods; Route-related measurement data available in the future can largely replace the estimation method.

The control model (algorithm) Traffic objective

The traffic objectives can be differentiated from the point of view of the Road users as unhindered as possible in a considered Network coverage (user optimum) and from the point of view of the operator of the process into one Optimization of the resulting global costs (system optimum).

The formulation of the road user-related objective is metrological and thus also not mathematically clearly ascertainable, since this depends on the character of the depends on individual road users. As an internal measure of quality of the process for example, the average individual travel time can be used. Depending on selected traffic model, other quality measures can also be used. The The operator's objectives can be formulated, for example, in a cost function; this usually consists of the weighted sum of quantities like the summed Travel time, the total pollutant emissions, possibly also the wear of the Road network etc.  The traffic objective is to be made available by utilizing existing otherwise unused, capacities can be achieved by traffic flows on Alternative routes will be relocated. This is done by sharing Traffic status information or by directly specifying detour recommendations en. The information can either be customized (using on-board devices) or transmitted collectively (using variable message signs). In both cases the information has only a recommendative character; the system operator is on the instructed voluntary compliance.

When using the system optimum as a switching criterion, the user optimum usually injured (WARDROP, 1952); it is therefore low if not disappearing acceptance rates. The system-optimal target function leads thus become ad absurdum, because due to the vanishing effect the user optimum also delivers better results than the "System-optimal" approach. Therefore, the user optimum is used as an example below considered.

An individual network control can consider the individual user optimum; it therefore has a greater potential for acceptance than the collective Network control. In principle, an individual network control allows finely metered Effect by transmitting the information only to selected vehicles.

In contrast, a collective network control can only be a very rough one step-like relationship between the control intervention and the effect achieved become. In this work, the special difficulties that arise as a result. the collective network control is taken into account by the control method without Limitation of transferability using the example of collective network control is developed.

approach

The algorithm is based on simple heuristic assumptions and usage classic control technology. The road network to be regulated is in alternative routes pairs divided. The average individual travel time is on each pair of routes or considered a variable proportional to it. To avoid Congestion will also increase traffic demand and the capacity of the routes considered. Based on an examination of the possible operating conditions formulas for calculating the dynamic setpoint for the vehicle current worked out on the main route.

Possible operating states

The traffic on the route pairs can depend on the load and the length difference of the routes can be divided into two or four areas, see Error! Reference source not found.

• - No diversion is required in area 1.  
• - In areas 2 and 3, overloading by diversion can be avoided become. If the user optimum is observed, areas 2 and 3 can only then take effect when the alternative and main route are roughly the same Have travel time in free traffic. With a longer alternative route it will Optimum user only met if traffic is already bound on the main route occurs; in this case only areas 1 and 4 are used.
• - Area 4 applies not only when there is a complete overload, but also when on one Route jam occurs. In area 4, the travel times of one Congestion compensation algorithm adapted to each other.

The following applies in areas 1-3:

q _d1 + q _d2 <c ₁ + c ₂

q _di route i request
c _i capacity route i

The total demand must be less than the total capacity of both routes; in addition, bound traffic (traffic jams) must not occur on any of the routes. For areas 1-3, the target value for the vehicles to be diverted is calculated as follows:

q _iS setpoint traffic flow route i

The congestion compensation algorithm

In area 4 of the operating states, an attempt is made according to the user optimum to adjust the travel time on the main and alternative route. After PAPAGEORGIOU (1990) is the "reactive" travel time calculated from the current edge travel times in approximately proportional to the "predictive" estimated actual travel time. The edge journey times can be calculated directly from the average speeds of the edges become. In bound traffic, the edge travel time is from the average density in Traffic jam, the length of which and depending on the outflow traffic volume (CREMER ET AL, 1993). Assuming that the travel time is one of the route length dependent share and the travel time in bound traffic, the Travel time therefore by an observable traffic jam length plus one of the route length dependent offset to be replaced.  

Assuming that the density in a traffic jam is constant, the quality measure can be calculated as:

i is _actual traffic _jam length on route
ls _i length of segment i

The formula

k _i density of segment i
k _max Maximum density (approx. 100 vehicles / (km * track))
also takes into account different densities, which are each related to the maximum density. The measure of quality

v _i speed of segment i
uses a sum of the segment-based travel time. In addition, depending on the condition monitoring used, other approaches for determining a quality measure can be used.

The dynamic setpoint for the vehicle current on the main route is calculated from a controller for the quality measure, for example the length of the traffic jam on the main route. The target dimension for the traffic jam on the main route is calculated from the sum of the traffic jam on the alternative route and an offset traffic jam length to take into account different route lengths and other parameterization:

l _1S = L ₀ + l ₂

l _1S setpoint for traffic jam length on Route 1,
l _{2Is the actual} traffic _jam length on Route 2,
L ₀ Adaptation for different route lengths

The setpoint for the quality measure is therefore also a dynamic setpoint.

The setpoint for the vehicle current on the main route is supplied by a proportional controller using the control difference and a feedforward control:

q _1S (t) = K (t). (l _1S (t) -l _1Is (t)) + S (t)

q _1S setpoint for the vehicle current on the main route
K controller gain
l _{1Is the} actual _traffic jam length on route 1
S feedforward control

Under the simplified assumption that there is only one traffic jam per route, there is Disturbance signaling from the vehicle current leaving this traffic jam; without additional disturbances, this is exactly the travel time current to be introduced into the main route, at which the accumulation length is constant. In the event that several separate Areas with limited traffic occur, calculated Feedforward control from a balance of the subareas.

The regulator approach

Any control technology approach can be used in the controller model. Three possible approaches are outlined below. These can be classified as follows:

• - A two-point controller uses the control error, whereby only two different discrete ones Interventions (circuit diagrams, target plans) can be used.
• - A multi-point control can use all possible manipulations but does not work to the control error, but only the setpoint.
• - A multi-point controller uses all possible control interventions as well as the Control error; this controller uses the added value of the extended architecture.

Two-point controller

If the difference between the target value for the vehicle current and the actual vehicle current on the main route exceeds a threshold value, the maximum possible vehicle current is diverted. This controller is based on the assumption contained in many publications that there are only two different circuit diagrams (not switched and redirection recommended) for collective network controls for an alternative route pair (e.g. PAPAGEORGIOU, 1990; MESSMER, 1994). The two-point controller follows the equation:

D Selected target plan
d _i target plan i

The two-point controller can be used as a two-point control without using the control error also in the previous basic architecture (FGSV, 1992) directly in a traffic jam compensation algorithm can be used (PAPAGEORGIOU, 1990).

Due to the system dead times, the two-point controller leads to large (from the dead times dependent) oscillating work movements. It is therefore an approach using a multi-point controller or a multi-point control makes sense.

Multi-point control

The use of a multi-point control is possible because there are two alternative routes There are two decision points at an interchange and a signpost equipped with several different target streams (e.g. three using prism technology) can be. So there are up to nine different goal plans.

Using the cordon matrix of traffic relations (VAN ZUYLEN ET AL, 1980; PLOSS, 1993) and assuming a compliance rate, the target plan can be determined in which the calculated current on the main route deviates as little as possible from the target current.

D (n + 1) = d _i with i, | Effect (d _i ) -q _1S | = min

Effect (d _i ) Effect of the destination plan i as a flow on the main route 1

It is assumed that the currents without switching are always the shortest route choose and the Kordonmatrix does not change when switching. During the first Point in a network with large differences in length, such as in Motorway subnetwork Munich North is realistic, the second point in reality is often not adhered to because local drivers use the secondary network (see SAXONY, 1998). The classic estimation methods must therefore be expanded; also must the acceptance rate with sufficient quality is available for this multi-point control to be able to use.

The multi-point control requires the control error or the real one Traffic flow is not. It can therefore also be assigned to the algorithm. This can be found in the context of a forecast-based optimization, for example at SACHSE  (1998). This example demonstrates that in the extended control process too previous control models and methods can be used.

Multipoint controller

This controller uses the added value of the process. For this approach, a sorting of the target plans with regard to their effect on redirected vehicles is required, starting with the circuit diagram without a redirected vehicle current up to the circuit diagram with the maximum redirectable vehicle current:

Effect (d _i ) <effect (d _j ) if i <j

In individual control processes, the circuit diagrams can be replaced by a continuous, but whole number of vehicles, to which the information is communicated. The individual vehicles could be selected based on the quality of the fulfillment of the individual user optimum. The switching procedure is as follows:

• - The control error between setpoint and actual value is calculated
• - If the error exceeds a certain amount, the next circuit diagram taken with greater effect in the event of a negative error of the next circuit diagram with less effect
• - If a connection has been made, the calculation is carried out for a certain period of time (e.g. 1 minute) blocked, so that the effect occurs on the measuring cross-sections can.

If the current circuit diagram d _{i is} switched, the circuit diagram is determined in the next step:

If the circuit diagram calculated in this way does not exist (d _-1 , d _{N + 1} ), the previous circuit diagram is used (d ₀ , d _N ). The bypass controller can be interpreted as an integral controller with delay element and limitation.

Multi-point controller with model knowledge

The multi-point controller can be combined with a multi-point control. The Added value results from a sudden change in the dynamic setpoint for the traffic volume, it is irrelevant whether this jump by a sudden disturbance or the change in the operating range is caused.  

The multi-point controller based on measured values can only iteratively respond to such a fault react while a control as feedforward control immediately, if not optimally, can react. The remaining difference between the target and actual value of the Traffic level is iterated by the multipoint controller using the real data balanced. The timing of this form of the control process is in . . . shown.

Validation methodology

The validation is based on a macroscopic traffic flow simulation. The Inputs to the simulation consist of the cord streams, which are from an estimate the Kordonmatrix and the Realdatenganglinien be won, as well as that of Control procedure determined control intervention. The effect of the goal plans is based on a separate simulation of traffic flows and assumed acceptance rates considered. The simulation in turn provides virtual measurement data that the Control procedures are made available. So the closed one Control loop are simulated.

The model does not have to be realistic in terms of quantity, but only in terms of quality replicate in order to be able to assess the behavior of the regulated system. Perhaps existing forecast inaccuracies of the traffic flow model are therefore not annoying.

By using different macroscopic equations in the internal and external evaluation ensures that the results are not based on Model similarities are falsified.

Test scenario

The control procedure is being tested on the Munich North Autobahn sub-network, consisting of the A9, A92 and A99 (see Fig. 5). The network can be divided into two alternative route pairs. In the following, the alternative route pair with the three-lane main route A9 from the AK Neufahrn to the AK München Nord and the two-lane alternative route A92, A99 will be considered. Six different target plans can be switched with different effects (SACHSE, 1998):

The larger the number of the circuit diagram, the greater the effect in the form of Number of vehicles diverted. The exact number of vehicles does not have to be known his; this is measured at the time of the procedure.

The traffic data are from 04/29/97 from 7:00 a.m. The model becomes so para metrized that an actually existing congestion at the AK Munich North is reproduced. Without alternative route control, this would go up to the AK Extend Neufahrn.

Results

Fig. 6 shows the travel times of the main and alternative routes when using the multi-point control. The resulting deviation between the target and actual value of the traffic volume causes a large deviation between the travel times of both routes via the quality measure controller.

Fig. 7 shows the results of the multi-variable control, which was introduced with the method described here. Only at 7:20 am there is a greater deviation in the travel time as a result of a 5 minute deviation in the target and actual traffic from 7:16 a.m. This results from a change in the operating range at 7:16 a.m. and the setpoint value that changes suddenly.

Fig. 8 shows the results of the multivariable control with model knowledge. The unfavorable behavior when changing the operating area can be avoided. This coupling of the use of measured real data and model knowledge is also a particular advantage of the method described here.

Error! Reference source not found. 9 shows the resulting spatial spreading of traffic jams. Segment 0 is located shortly after the AK Neufahrn, segment 30 just before the AK München Nord. In the contour lines are the switching operations are clearly recognizable. As a rule, there is no circuit diagram whose Effect corresponds exactly to the current setpoint. For this reason, the Forwarding regulator automatically switches back and forth between two circuit diagrams to average set the setpoint.

Summary and Outlook

Using network control as an example, it was shown that the extension of the Basic architecture the implementation of simple but effective control ver driving allowed.

The described network control method is based on the control model on simple heuristic approaches with which the user-related objective function of the shortest possible individual travel time is optimized. The control process requires neither an estimate of the cordon matrix nor an assumption of Compliance rates, still cross-sectional forecasts of traffic flow, still  a spatial-temporal forecast of the traffic flow. Anyway, or because of that, because error sources are often excluded, good results are achieved.

The Kalman filter used as an observer of the current traffic condition can if necessary, be replaced by other traffic models. The control model can be supported by model knowledge. Available cordon matrices in the future can be taken into account, for example. There were already several in the controller Approaches demonstrated; Of course, other approaches can also be used here, for example for a individual control method can be used. Because of the underlying The basic architecture is a modular control process, whose modules are largely independent of each other.  

literature

CREMER, M. (1979): The flow of traffic on expressways - models, monitoring, regulation. Springer, Berlin Heidelberg New York
CREMER, M .; MEISSNER, F .; SCHRIEBER, S. (1993): On Predictive Control Schemes in Dynamic Rerouting Strategies. Transportation and Traffic Theory, pp 407-426
FGSV (1992): Notes for control models of variable message sign systems in out-of-town areas. Research Society for Roads and Traffic, Cologne
JANKO, J. (1993): Problems of Travel Time Prediction in a Road Traffic Management System, Dissertation, TU Berlin
KELLER, H. (1998): TABASco validation of telematics applications in Bavaria, Scotland and other countries; Project and results in Bavaria. TASASCO consortium
KRUX, W .; OVERKAMP, K .; KATHMANN, TH .; OFFERMANN, F .; DETERMANN, D. (1997):
Alternation signage A9 / A92 / A99 of the traffic control system Munich North - adjustment and optimization of the parameters for the automatic operation of the signposting. Institute for Road Engineering, RWTH Aachen
MESSMER, A. (1994): Application of control engineering methods for dynamic route guidance in expressway networks. Dissertation, TU Munich
PAPAGEORGIOU, M. (1990): Dynamic modeling, assignment and route guidance in traffic networks. Transportation Research 24B, 471-495
PLOSS, G. (1993): A dynamic method for estimating traffic relationships from cross-sectional counts. Publications in the field of traffic engineering and traffic planning, Technical University of Munich, Munich
POSCHINGER, A .; CREMER, M .; KELLER, H. (1997): A Control Scheme for Variable Direction Signs Using Dynamic Reference Values. Proceedings of the IEEE Conference on Intelligent Transportation Systems, Boston, USA
SACHSE, T. (1998): Alternative route control in highway networks based on an extended analysis of the traffic flow. Dissertation, TU Munich
WARDROP, JG (1952): Some theoretical aspects of road traffic research. Proceedings of the Institute of Civil Engineers, part II / 1, pp. 325-378
vAN ZUYLEN, HJ; WILLUMSEN (1980): The Most Likely Trip Matrix Estimation from Traffic Counts. Transportation Research, Vol. 14B, pp. 281-293

Claims (13)

1. Method for determining control interventions in alternative route controls on road networks, local detection cross-sections being formed by means of stationary detectors, traffic-related measured values being recorded and processed for determining traffic information, the control intervention being derived from this in at least one complex method, characterized in that from the data In one processing step, non-static target values for observable traffic variables, in particular measurable traffic variables, are determined, from which the intervention is determined in at least one further processing step using the observed traffic variables. 2. The method according to claim 1, characterized in that the road network in Alternative route pairs is divided. 3. The method according to claim 2, characterized in that for the main route one Route tuples in a machining process a target value for the primary effect, in particular the traffic flow of the alternative route control is determined. 4. The method according to claim 3. characterized in that for determining the Setpoint evaluation criteria such as travel times and / or traffic jam lengths and / or emissions on the alternative routes are related. 5. The method according to claim 4, characterized in that the evaluation criteria are estimated using a traffic flow model, the state variables of which are min the speed, density, traffic flows and turn rates Decision points are. 6. The method according to claim 5, characterized in that the traffic flow model using a correction procedure expanded to include the turn rates, especially according to Wiener and Kalman, can be corrected on the basis of measured values. 7. The method according to claim 6, characterized in that the road network in length segments and virtual segments are divided at nodes. 8. The method according to claim 7, characterized in that the comparison of the Evaluation criteria on the route pairs is carried out by a controller, whereby the impact criterion of the main route and the alternative routes as setpoint, or actual value can be used. 9. The method according to claim 8, characterized in that the evaluation criteria in In order to achieve a high level of acceptance, the user needs to be used in the best possible way. 10. The method according to claim 9, characterized in that from the target values for observed traffic quantities and the actually observed traffic quantities by means of of a controller, especially using linear and non-linear approaches Control technology the circuit diagram is determined. 11. The method according to claim 10, characterized in that in claim 10 complex method of real data use described by at least one other Processes using model knowledge for improvement is supported.   12. The method according to claim 11, characterized in that for the Alternative route tuple circuit diagrams are formed, which are characterized by this to have different effects in the form of diverted vehicles. 13. The method according to claim 12, characterized in that the target plans accordingly their effects are ordered.