EP1174842B1 - Method to create forecasted traffic data for traffic information - Google Patents

Abstract

Traffic data from various sources is forecast as a forecast horizon in a forecast procedure for place and time. Forecast quality and errors currently attained are continuously determined by a set point/actual comparison of forecast traffic data with traffic data currently detected after the expiry of a forecasting time period.

Description

The invention relates to a method for generating predicted traffic data for traffic information.

In traffic, traffic conditions can change very quickly due to different influences. Resulting traffic disturbances, e.g. Traffic jams represent a danger for the following road users. Current route announcements or route recommendations based on distance-related travel time losses inform the road users of such disturbances, also for safety reasons to avoid accidents.

Now every system has an inherent reaction time. Thus, traffic information systems and methods responding to current traffic readings, e.g. Speeds and traffic flows at locations of the road network, traffic information such as e.g. Traffic messages and travel time information generate, with varying speed on changes in the state of traffic, which are reflected in the traffic readings. In addition there are the delay times of the measuring systems and the delays in the transmission, the measured values to the downstream systems to be processed. Also when passing the generated traffic information to the road users or service providers, runtimes and delays occur.

Consequently, the timeliness of traffic information is always limited, even if it has been calculated on the basis of completely up-to-date traffic readings. There is therefore an urgent need for more up-to-date traffic information than can be offered today.

The timeliness of the generated traffic information can be increased by they are forecasted to some extent. Thus, the inherent system run times and response times can be reduced or even completely compensated.

In realizing this objective, there are the following problems:

The traffic situation is determined from traffic measurements. Consequently, these measured values are to be forecast. The forecasting method must work adaptively and be appropriately influenced in order to be able to react accordingly to slowly or rapidly changing conditions. In particular, it must be possible
select, evaluate or even correct the respective traffic data source to be used depending on the situation
be able to determine the forecast horizon to be achieved, adaptively during the term of the forecasting system and
for a follow-up quality control to determine the actually achieved forecast quality continuously updated.

Furthermore
If the procedure is to work independently of the distance, ie not dependent on connection points, etc., these must be taken into account as appropriate.
Are several physical quantities to be included that are coupled together, eg speeds and flows,
are traffic measurements of different characteristics and sources, synchronously clocked and asynchronously or event-triggered, detected at fixed or variable locations, to include and consistently use (induction loops, Floatin, Car Data (FCD), stationary detection systems).

Hydrographs or historical load curves, e.g. are also used by energy suppliers, are averaged over comparable weekdays courses of Verkehrsmeßwerten (v = speed, d = traffic density (vehicles / distance), f = traffic flow (vehicles / time)., Such mean functions over time are common only at the same time repetitive traffic events to make predictions on the basis of history.

Traffic simulation models, eg flow simulation calculations of systems partially Differential equations [Drivers, Helbing, Numerical Simulation of Macroscopic Traffic Equations, CISE 5, 89 (1999)], are used to predict traffic readings regarding locations and times when there are no readings.

These known methods alone do not provide a solution to the problems presented.

The object of the invention is therefore to solve the above-mentioned problems and to provide a method that can be used on all types of roads, both inside and outside for forecasting current traffic data. It should also be suitable for the creation of travel time forecasts, traffic forecast forecasts, detour recommendations as well as for direct vehicle control and traffic control, as well as for numerous other services.

This object is achieved according to the invention with the features of claim 1. Further embodiments will become apparent from the dependent claims.

Fig. 1:

ein Schema eines Systems zur adaptiven Regelung einer Verkehrsprognose durch fortlaufende Bestimmung des Prognosefehlers über eine Rückkopplung mit Verzögerung und Soll-/Istvergleich. δ₀ steht dabei für die Impulsfunktion

Fig. 2

ein erweitertes System zur adaptiven Regelung einer Verkehrsprognose durch einen zweifach kaskadierten Regelkreis mit zusätzlichem Soll-/Istabgleich auf Produktebene. δ₀ steht dabei für die Impulsfunktion.

The invention will be explained in conjunction with the drawings. Showing:

Fig. 1:

a schematic of a system for adaptive control of a traffic forecast by continuously determining the forecast error on a feedback with delay and target / actual comparison. δ ₀ stands for the impulse function

Fig. 2

an extended system for the adaptive control of a traffic forecast by a double cascaded control loop with additional setpoint / actual adjustment on product level. δ ₀ stands for the impulse function.

Depending on the current traffic condition at a location of the transport network, a prognosis is more or less sustainable, since the traffic situation can take place in critical areas of the overall system in which the "system" can increasingly react chaotically. Suddenly occurring accidents are difficult to predict. The forecasting system will provide erroneous values in these cases.

The achievable forecasting quality depends on various factors. The forecasting method can therefore not work independently of the forecasting quality. It is therefore necessary to determine the quality of forecasting continuously and to make it available to the forecasting method used as a feedback control variable.

The ongoing measurement of the forecasting quality is also used for continuous quality control in order to determine situations in which the prognosis is not or only insufficiently possible. If the forecast quality decreases suddenly, this is in some cases not only a sign of a sudden change of the inherently chaotic system but of an unforeseeable disturbance on the traffic system, e.g. in the form of an accident.

It is therefore necessary
adapt the used forecasting methods adaptively to the respective system state,
determine the attainable forecast horizon suitably and adaptively
to continuously determine a forecasting quality control quality.

1. 1. Von verschiedenen Quellen eingehende Verkehrsdaten werden über ein Prognosesystem
   für einen bestimmten Zeitraum gerechnet vom aktuellen Zeitpunkt t der Berechnung der Prognose sowie einen gewissen Ortsbereich um Orte mit real vorliegenden Verkehrsdaten, den zeitlich-örtlichen Prognosehorizont t_p bzw. x_p, vorhergesagt. Dabei bestehen die Verkehrsdaten im einfachsten Fall aus Verkehrsmeßwerten aus dem Straßenverkehrsnetz, z.B. Verkehrsflüsse, -dichten oder-geschwindigkeiten. Sie können aber auch daraus abgeleitete Werte enthalten, z.B. gefilterte Verkehrsmeßwerte.
2. 2. Es wird die aktuell erreichte Prognosegüte bzw. der Prognosefehler durch Soll-/Ist Vergleich der prognostizierten Verkehrsdaten mit den nach Verstreichen des Prognosezeitraumes aktuell ermittelten Verkehrsdaten während des Prognose- und Produktionsbetriebs laufend aktuell bestimmt.
3. 3. Der gemessene Prognosefehler wird über eine Rückkopplung als Stellgröße zur adaptiven Anpassung des eingesetzten Prognoseverfahrens als auch des anvisierten Prognosehorizontes verwendet. Über eine Rückkopplung wird das Prognoseverfahren also laufend adaptiv gesteuert.
   Bei größer werdendem Prognosefehler wird dann der Prognosehorizont für das eingesetzte Simulationsverfahren entsprechend verringert und umgekehrt. Ebenso kann die Auswahl der für die Prognose zu verwendenden Verkehrsdatenquellen über den Prognosefehler gesteuert werden. Zusätzlich wird der Prognosefehler als Qualitätsmaß den prognostizierten Verkehrsdaten hinzugefügt, um nachfolgenden verarbeitenden Systemen eine Bewertung bzw. Auswahl zu ermöglichen.
4. 4. der Prognosefehler wird fortlaufend ausgegeben zur Qualitätskontrolle der Prognose. Er dient auch als Indikator für einen plötzlich aufgetretenen Störfall. Ebenso werden gemittelte Verläufe von historischen Verkehrsmeßwerten (Ganglinien) in der Nähe eines erkannten Störfalls korrigiert, z.B. die Flußwerte an stromabwärtigen Detektorstandorten verringert.
5. 5. die prognostizierten Verkehrsdaten werden sodann Verkehrsinformationssystemen zugeleitet, die dann die üblichen Verkehrsinformationen erzeugen, nur aktueller auf Basis der aktuelleren prognostizierten Datenbasis.

The method is divided into the following steps (FIG. 1):

1. 1. Traffic data received from various sources is passed through a forecasting system
   Calculated for a certain period calculated from the current time t of the calculation of the forecast and a certain area around places with real traffic data available, the temporal-local forecast horizon t _p or x _p , predicted. In the simplest case, the traffic data consists of traffic measurements from the road traffic network, eg traffic flows, densities or speeds. You can also use derived values, such as filtered traffic readings.
2. 2. The currently achieved forecast quality or the forecast error is determined by actual / actual comparison of the forecast traffic data with the traffic data currently determined after the expiry of the forecast period during the forecasting and production operation.
3. 3. The measured forecast error is used via a feedback as a manipulated variable for the adaptive adjustment of the used forecasting method as well as the envisaged forecast horizon. By means of feedback, the forecasting process is thus continuously adaptively controlled.
   As the forecast error increases, the forecast horizon for the simulation method used is then correspondingly reduced and vice versa. Likewise, the Selection of the traffic data sources to be used for the forecast are controlled by the forecast error. In addition, the forecast error is added as a measure of quality to the predicted traffic data to allow subsequent processing systems to score.
4. 4. The forecast error is continuously output for quality control of the forecast. It also serves as an indicator of a sudden accident. Similarly, averaged histories of historical traffic readings (hydrographs) in the vicinity of a detected accident are corrected, eg, the flux values at downstream detector sites are reduced.
5. 5. The predicted traffic data is then sent to traffic information systems, which then generate the usual traffic information, only more up-to-date on the basis of the more recent predicted database.

zwischen prognostizierten und aktuellen Verkehrsdaten in einer lokalen Umgebung, dr und ds von z.B.
x_p = 500m und t_l = -4 min respektive verwendet. Die lokale Glättung der prognostizierten Verkehrsdaten und der gegenwärtigen Verkehrsdaten wird über eine Ortsinterpolation vorgenommen. Die Gewichtsfunktion g(u,v) als Integrationskern wird für größere betragsmäßige Abweichungen x - r bzw. t - s kleiner und sorgt damit für die lokale Bewertung der Abweichungen bzgl. x und t. Sie kann z.B. eine Gaußfunktion sein oder auch einfach zu 1 gesetzt werden, was einem Rechteckfenster entspricht.The target / actual comparison of the forecast with the current traffic data is made via a similarity measure, which evaluates the current time with a local environment, the deviations of the data from each other. This can be done, for example, via a correlation or a distance metric. In the method, the distance metric is the mean square deviation $$e\left(x.t=0\right)={\displaystyle \underset{t_1}{\overset{t_2=0}{\int }}{\displaystyle \underset{-x_p}{\overset{x_p}{\int }}G\left(x-r.s-t\right){\left(w_{\mathrm{forecast}}\left(r.s\right)-w_{\mathrm{current}}\left(r.s\right)\right)}^{2}dr}ds}$$

between predicted and current traffic data in a local environment, dr and ds of eg
x _p = 500m and t _l = -4 min respectively used. The local smoothing of the predicted traffic data and the current traffic data is done via a local interpolation. The weight function g (u, v) as integration kernel becomes smaller for larger magnitude deviations x - r or t - s and thus ensures the local evaluation of the deviations with respect to x and t. It can be eg a Gaussian function or simply set to 1, which corresponds to a rectangular window.

In addition, the method is extended by a continuous generation of a measure of the prediction quality at the level of the products (traffic reports and travel times) (FIG. 2). These is used as an additional quality control and also as a manipulated variable. This results in a regulated variable forecasting horizon through a double-cascaded indirect or direct control loop. In principle, only the outer control loop can be operated for quality control or evaluation.

The measurement of the prognosis quality can also be done retrospectively, ie "offline", on the basis of archived measurement data or products (archived traffic reports and travel times) for a quality evaluation in order to determine situations in which the prognosis was not or only insufficiently possible.
To determine the forecast quality for the product data (external target / actual comparison in FIG. 2), similarity measures are used for traffic reports.

Claims (11)

1. A method of drawing up predicted traffic data for traffic information,
   characterised in that,
   on the basis of traffic data originating from various sources, traffic data are forecast in a prediction process for a location and a time as prediction horizon,

   - in that the currently achieved prediction quality or prediction error is determined on an ongoing basis during prediction and production operation by predicted/actual comparison of the predicted traffic data with the traffic data currently determined once the prediction period has passed,

   - in that the measured prediction error is used via feedback as a correcting variable for adaptive adaptation of the prediction process used and also of the target prediction horizon,

   - wherein, if the prediction error grows larger, the prediction horizon for the simulation process used is reduced accordingly and, if the prediction error diminishes, said horizon is increased,

   - and in that traffic information may then be compiled and output from the traffic data predicted in this way.
2. A method according to claim 1,
   characterised in that
   the traffic data consist of measured traffic values, such as traffic flow, density and speed.
3. A method according to any one of the preceding claims,
   characterised in that
   the traffic data contain variables derived or calculated from measured traffic values, such as filtered measured traffic values.
4. A method according to any one of the preceding claims,
   characterised in that
   the predicted/actual comparison of the traffic data is performed via a similarity measure, and in particular via a correlation or a distance metric between locally smoothed versions of the predicted and current traffic data.
5. A method according to any one of the preceding claims,
   characterised in that
   the standard deviation $$E\left(x,t=0\right)={\displaystyle \underset{t_1}{\overset{t_2=0}{\int }}{\displaystyle \underset{-z_p}{\overset{z_p}{\int }}g\left(x-r,s-t\right){\left(w_{\mathrm{prediction}}\left(r,s\right)-w_{\mathrm{current}}\left(r,s\right)\right)}^{2}dr}ds}$$

   

   
   between predicted and current traffic data in a local environment dx and dt is used as distance metric, with a monotonously falling weighting function g(u, v) in both arguments.
6. A method according to any one of the preceding claims,
   characterised in that
   local smoothing of the predicted traffic data is effected via interpolation.
7. A method according to any one of the preceding claims,
   characterised in that
   the prediction error is used as a fault indicator, to generate corresponding traffic reports or for taking corrected account of current or historic measured traffic data from detectors surrounding the particular location x under consideration.
8. A method according to any one of the preceding claims,
   characterised in that
   the traffic data sources to be used for the prediction are selected via the prediction error.
9. A method according to any one of the preceding claims,
   characterised in that
   the currently achieved prediction quality or prediction error is determined on an ongoing basis during prediction and production operation on the basis of traffic products, e.g. traffic reports, by predicted/actual comparison of the traffic products determined from predicted traffic data with the traffic products currently determined once the prediction period has passed (Fig. 2).
10. A method according to any one of the preceding claims,
    characterised in that
    prediction quality is determined as a function of location and time on the basis of archived traffic data or archived traffic products.
11. A method according to any one of the preceding claims,
    characterised in that
    the measured traffic values for the traffic conditions of the road traffic network are detected by stationary and/or mobile, synchronously and/or asynchronously transmitting detectors.

Images