EP0771447B1

EP0771447B1 - Detection and prediction of traffic disturbances

Info

Publication number: EP0771447B1
Application number: EP96914510A
Authority: EP
Inventors: Kjell Olsson
Original assignee: Dinbis AB
Current assignee: Dinbis AB
Priority date: 1995-05-19
Filing date: 1996-05-13
Publication date: 2004-02-25
Anticipated expiration: 2016-05-13
Also published as: WO1996036929A1; DE69631629D1; EP0771447A1; SE9501919D0; SE503515C2; DE69631629T2; SE9501919L

Description

The present invention relates to a method for detection and prediction of disturbances in the road traffic, e g the forming of traffic queues depending on overloading the road-net or incidents. In traffic management systems an important task is to avoid overloading, where traffic breakdown is introducing queues with reduced passability, increased risk for accidents and increased environmental problems. Incidents should be detected early to be able to reduce the damages. The object is to get wounded people to hospitals, to reduce the secondary related accidents and to manage the traffic in such a way that no unnecessary blockings arise, but that the road-net will be efficiently utilized. A background for basic technologies is given in the published international patent application WO 94/11839. The present invention presupposes the existence of knowledge of that technology.

To-day there is technology for detection of queues and incidents. Traditionally one has detected incidents using a method, which measures "occupancy" and possibly traffic flow or velocity and possibly differencies of the parameter between the the two recent time periods. Those measures have been studied at one sensor, and if the values put together according to any special algorithm have exceeded a threshold value, one has detected an incident. There are also examples, where one has used the method at two succeeding sensors and then has calculated the differences between the sensor values, before the values have been put together in an algorithm. The algorithms have been formed by "trial and error", i e one has tested and changed until one has no longer got a lot of false alarms, at the same time as one has not missed detection of many real incidents.

The drawbacks with those traditional methods compared to the present invention are the following:

- 1A. Traditional methodology.

The free-flow traffic over a sensor contains very large statistical variations, why measured values in one measuring period naturally might differ very much from the next period, i e without being a sign of any incident. At the use of two sensors (A and B), which are very close, one can understand the process as follows; if one during the same time period measures traffic at A and B and it takes a short time for traffic to move from A to B, there would be almost the same traffic measured at A and B, i e the parameters cannot change very much. If traffic anyhow have changed very much, that would be a sign of something unnormal, that an incident has occurred between A and B. If it is a longer distance between A and B, however, then one is measuring different traffic and then there are natural large variations.
Since one wants to detect incidents as fast as possible, one wants to measure during a very short time period. At the same time one wants to keep a long distance between the sensors to keep a low number, saving costs. This is a dilemma for the function.

- 1B. The invention.

In the present invention, prediction of traffic is used. As said above, the traffic at sensor B might vary much. If e g during one period, there is not a single car passing, although there were many cars passing during the period before, that might indicate that an incident has occurred, which prevents traffic to pass. But it can also be a natural gap in the traffic. If one by measuring traffic upstream, finds that there is a gap in the traffic, which will be measured later on at B, that can be predicted for B, - and then the measurement of 0 cars passing at B will not be a sign of an incident between A and B, but a confirmation that the traffic is as expected.

This method increases the freedom to measure during short time periods, and since one predicts, one is not losing time. Directly after the measurement at B, differences between predicted and measured values are obtained and conclusive differences indicate an incident. The distances between the sensors A and B can be increased too, and the number of sensors reduced. The requirement here is instead that there are possibilities to do reasonable predictions. Roughly speaking however, even weaker predictions should mean improvements, e g also if one couldn't predict exact 0 cars in the example above, anyhow a prediction of a reduction of flow would mean a less deviation from the measured value, than if one knows nothing, and by that one can reduce the risk for false alarms.

- 2A. Traditional methods.

Another drawback with the traditional methods are that they build very much on experimenting to find suitable algorithms with balanced parameters. It is very often difficult to understand why one way would be better than anyone of the others, and it is also difficult to prove. Neither does one know if it is possible to improve the method, by adding more tests and trimming.

- 2B. The invention.

In the present invention, the understanding of the traffic processes are utilized, and that in a direct way. An important part is the understanding of large traffic variations. Those can be regarded as results of stochastic processes, and if one measures e g flow at a sensor, then one experiences those as noise. Using short measurement periods, one obtaines relatively large variations around a given average. By utilizing knowledge about "noise", one can understand how to make use of the information in those "noise-variations", and not only regard those as something that destroy the possibilities to perform simple detections of incidents. By measurements one can get means and standard deviations and from theory and measurements one can create approximative distributive functions, i e one knows statistically rather much about the traffic variations. For example, assuming a normal distribution, a measured standard deviation can give information about the probability for a variation being larger than a given value. One also obtains an understanding about what one does not know and a measure of the uncertainty.
( And that it is not of any use with ever so long and complex parameter filled algorithms, in accordance with the traditional methodology.)
In the present invention the knowledge is utilized about probability for deviations of a certain order to set thresholds, which by that give the desired false alarm rate. It might also be that a deviation that originates from an incident is not large enough to exceed the threshold. Then one can wait untill the next measured deviation is received and examine if those two values together are that large that the probability requirement now is fulfilled, ie that one is now exceeding the corresponding threshold. This can be repeated successively, and if the natural variations are very large, the threshold will be large, and there might be required more incident-caused deviations for them to exceed their threshold. However time is running, the more measurement periods there are needed, and the incident detection should be fast to prevent serious secondary effects. In the invention e g the threshold can be automatically set in that way that a minimum of extra measurement periods need to be used.
It is also by this reason important to keep the influence of the natural traffic variations at a low level, and that is done in the invention, as said above, by the process, where it is not the variations from an average or the former value, which is regarded, - but the much smaller deviation between the predicted and measured values, which is determining the threshold level. By that the threshold level can be reduced significantly without increased false alarm rate, and an incident-originated deviation is then more easily exceeding the threshold and the incident will be detected faster.

- 3A. Traditional methods.

A third drawback with the traditional methodology is, that it is difficult to transfer from one situation, where it finally with trial and error, has been adapted to operation, to another situation. It might mean geographically, positions, eg transfer to another road section, where access-roads, intersections or number of lanes offer other traffic situations. It might mean changes of measuring time periods or other parameters. This effort can be very time-wasting and resource-consuming. One needs to observe the traffic in parallel with other accurate measuring means to get a key answer to compare with, giving possibility to change parameters in the algorithm towards better agreement with the reality. Also changes in the traffic situation might result in working through the process again, to get new better adapted parameters to put into the algorithm.

- 3B. The invention.

In the present invention much of the adaption can be done automatically. Besides that, the starting values can be well chosen from the origin. Independent of implementation, the topical deviations are measured, and the corresponding statistical measures are obtained, e g the standard deviation of the traffic deviations. Based on those measures the respective threshold values can be set automatically, and the method starts to generate incident detections, which the operator can observe are true or false. Since the method continuously measures the deviations, the statistical parameters can be successively updated and adapted to changes in the traffic situations.

Traffic breakdown and queues.

Overloading of the road-net, also if only for a short term traffic peak, is enough for generating traffic breakdown and queue build up. Those queues might then be maintained by a somewhat lower traffic flow, as the road-net capacity usually decreases by the queue-forming. For example, if there is a traffic peak on the motorway at the same time as there arrives a traffic peak on the access-road, not offering space enough for all the cars, the cars have to break to increase their respective gaps during the trial to merge the two traffic flows. Then the velocity might be decreased to very low values with small gaps between the cars, resulting in a low traffic flow. In some cases it might be significantly lower than the maximum flow, obtainable at higher speeds, and which is regarded as the road capacity.

Thus the result of overloading is queue-forming, which in principle reduce the road-net capacity. In the large cities this is occurring daily at rush-hours at mornings and afternoons, which implicates that the road-net is forced to its lowest capacity, when it is needed to be largest.

So it is very valuable to be able to manage traffic in such a way that traffic breakdown is avoided. Especially important is preventing traffic breakdown at critical places, where queues give rise to serious disturbances on one or another of the large traffic arterials. Preventing that, might be one task for the invention.

A key-function is prediction of traffic breakdown and queue-forming. By prediction a time-margin is obtained before the predicted problem really is happening. That time-margin can be used to implement actions, which prevent that the problem arise in the real world. Also in the detection process of queue-forming, it is interesting to utilize prediction. For example, if free-flow is predicted and a queue anyhow forms, then the sensors offer values, showing the real traffic situation (queue). The deviations between the predicted free-flow values and the measured values can therefore be used as an indication on the forming of a queue.
In this desciption of the invention, sometimes other words are used than "prediction", e g the word "expected". In general it is meant, that if a method contains a process, where a measured value shall be compared with another, e g earlier known, "corresponding value", then the notation of "corresponding value" often implies an association of a time direction of changed knowledge of the parameter, also if the value just have been obtained from historical values. In this paper therefore the notation "predict" is used including also estimations, that is not direct predictions, but is fulfilling a corresponding object. For example, the comparison value might be a mean-value or a mean-value plus a value based on a standard deviation, historically estimated value etc. Independent of which way that has been used to get the comparison value, the object is anyhow that this value constitutes a type of expected comparison value, by which the measured value can reach criteria for detection of a queue. Hereby the expected value has got a forward-associated function towards the measured value, and might be estimated in an equivalent process of a prediction, also when the expected value is estimated afterwards, i e after that the the measured value has been obtained.

A queue-detection according to the invention, can also be performed when queues are formed on links between sensors. This is also valid for the use of video-sensors, IR-sensors , radar and similar sensors, which e g with an image can cover a longer road distance than those few meters that traditional loop-sensors cover. However, in practice the video-sensor range is much shorter than the distance one "can see". The limitations in height-positions of the cameras implies e g that a bus can hide a long row of cars. Video-sensors, positioned at 0,5 to 1 km interval, therefore might only have a guaranteed coverage of their respective close area, and the larger part of the distance in between, has to be treated in the corresponding way as with loop-sensors.
Detections can be performed at downstream as well as upstream sensor. At the upstream sensor the queue is detected by the fact that the queue is within the direct measuring area of the sensor. Characteristics of a queue is that traffic is dense and the speed is lower than at the free-flow mode. It is known, when the flow is approaching the capacity limit of the road, that the velocity is decreasing, e g at an access-road, where the speed limit at the motorway is 70 km/h, the motorway speed might drop to 55 km/h, because of the increased traffic density. At further increase of traffic density, the traffic breaks down to a queue, which might got still lower speeds. According to the invention, the later traffic state might be surveyed by measurements for at least two measuring periods. It appears, when it is queue caused by traffic collapse, that the velocity and the flow are in-phase, i e both flow and velocity are increasing respectively decreasing together. When it is dense traffic however, including queues with moving queue-fronts at high speeds, then the velocity and flow are changing in reverse phases.
Using this method, one can also define at which velocity dense traffic typically is transferred into traffic collapse. Measurements by the inventor in Gothenburg, indicated the velocity breakpoint at the level of 55 km/h. That was well reproducable at the typical roadnet velocity limitation to 70 km/h.
There are other definitions of a queue state, e g where already one car is considerred building a queue if the time gap to the car in front is less than 2 seconds and the following car has a higher velocity need. Queues and queue-forming also get different process courses on ordinary roads with one lane, compared to two lanes and compared to motorways. Those queues that are most interesting for this patent, are such that are appearing on motorways and similar arterial roads for larger cities.
From the view of traffic management, the essential queues are those creating large problems. Therefore small groups of cars driving close, are considered as dense traffic. Also longer packets of cars are here considerred as dense traffic, when driving in somewhat reduced velocities compared with the free-flow velocity ( often the given speed-limit on signs ). Usually those car-packets are characterized in that the front of the packet is moving forward along the road ( "moving queue"). At velocities above the break-point, the traffic in such a packet is characterized by high flow and reasonable high velocity, why a calm (homogeneous) driving in such a packet might not constitute a direct traffic problem.
In near ranges of cities there are however a high density of on- and off-flows of the motorways, why a calm dense traffic is seldom appearing. Instead the traffic is characterized by transfers of lanes, "weaving", which rather cause a dense traffic to collapse, and result in queues with low velocity in the unstable queue-forming state of traffic.

On motorways there have been used "stretch-systems". Those systems are traditionally based on variable speed-limit signs, which are installed in intervals of 500 m along a motorway link. Sensors are measuring the traffic situation, and when trade is considered inhomgeneous, varying and reduced velocities etc., then the system is putting on the signs showing decreased speed-limits for a number of intervals. The purpose is that the traffic should be "calmed" and the velocity homogeneous at the now present lower speed limit. There are reports saying that large improvements have been achieved, in respect of both reductions of accidents and increased capacities. However there is a problem in the traditional methods in that, when the inhomogeneity in traffic is occurring, then the traffic collapse is already on its way, and when the measurements are obtained and detection achieved, then the collapse is a fact. Different trials might be done to improve the situation, e g by using much shorter measuring periods to get information faster.

In the present invention the traffic is instead successively predicted, and when the probability of collapse is above a certain given value, then the corresponding speed-limits are reduced on the signs. By this there are created time-margins for avoiding the traffic collapse, and the action influence on the traffic might be kept at a lower level. The method is the same as that used for queue- and incident detection.

The present invention can also be used for control of on-flow traffic, e g for control of "ramp-metering". In city-areas on-flow and off-flow traffic is a serious source of disturbances on the motorways, as is described above. There are often occurring overloading and traffic breakdown, and troublesome queues are forming. Prediction of the risk for overloading can be used for reduction of on-flow traffic on the considerred ramp, and for reducing of traffic on the motorway, e g by reduction of of on-flow trade on ramps upstream of the original on-flow ramp.

The prediction of traffic collapse at an on-ramp can be based on measurements at upstream sensors e g a sensor at the main road and a sensor at the access-road. Measurements of traffic by respective sensor can be used to predict the traffic a certain time-interval later on, equal to the travel time to the weaving area at the connection. By matching or synchronizing of measurements can e g occasions be predicted, when coinciding traffic peaks reach the access connection. The predicted flows are compared with the threshold values to obtain the prediction of overloading. One way to estimate the threshold value for the main road is illustrated as follows. The weaving capacity C_v= C₀ - a * I_e, where C₀ is a constant and I_e is the flow on the access road. The factor a shows that the capacity on the main road is not determined by a simple sum of the two flows. Both C₀ and a should be calibrated for the present access road. In an example the value for a two-lane motorway is C₀ = 4200/h and a = 2,5, i e a differs significantly from the value= 1. That fact is also observed from the value of a for a turn-off road, where e g a = 1,5. In both cases it is the weaving that causes the smaller total capacity at off- respectively on-flow. Those present algorithms have been shown good agreement down to small on-flow values. When traffic has broken down, other conditions are valid. At established queue-situations e g, the weaving is by each second, which consequently gives a corresponding relation between the traffic on the motorway and the access road. When queues are growing, reaching the upstream sensor, the algorithms, as those given above, and C₀ respectively a can be successively updated.

The queue-growth is determined by the difference between the flows behind and in front of the queue. The flow in from of the queue, might be estimated when needed, from a model for queue off-flow at the front of the queue. An example of a simple algorithm is Iaq = b * (g)^-0,5, where g is the gap between the cars at the queue-front and b is a constant, which value can be approximately stated and updated by measurements. Iaq can also be obtained from Iaq = (Va-Vq) / (Da - Dq), where V is velocity and D is the periodic distance for the cars, and a indicates parameters in front of the queue and q in the queue. Together with the relation I = V / D, the off-flow at the queue-front and the flow downstream the queue can be determined, and with information on the flow and the related velocity downstream the queue, also the growth and decay of the queue can be determined. The queue off-flow algorithm is valid for many usual situations, and the gap g can be obtained typically from relations between gap, flow and velocity at queue-states.

Determination of probabilities for queues and incidents.

At prediction of an event, the most interesting is not always to judge, if it would be the most probable outcome that the event occurs, i e if that probability is above 50%. If the risk for queue-forming is 30 % or the risk for an accident is 10 %, then that might be enough for actions to be taken to prevent the event from occurring, i e in spite of the largest probability being neither a queue nor an accident. Below, examples are given for the way to work with the probability determination according to the invention.

A typical distribution function within statistics is the Normal or Gaussian distribution. Assuming that one as approximately valid for the traffic on a certain part of the road-net, then the function can be calibrated from measurements and estimations of the variance of traffic around the average value. The probability for obtaining a certain value can be calculated or usually fetched from tables. Depending on the detection process, there might be a need for modifications of the distributions, or adaptions with the use of other distribution functions. The Rayleigh-distribution e g is interesting at envelope detection and filtered noise deviations. For illustrating the invention we use an approximative distribution of the "noise"- deviations according to P{y(t)≥x} = exp(-x²/σ²), i e the probability that a deviation y(t) is is larger than a given value x is exp(-x²/σ²). The amplitude normation is neglected.
Then we find that the probability during one measurement period for a deviation larger than the standard deviation, is exp(-1) = 37%. Also we can ask, how large the deviation need to be for the probability to be as small as 10^-4. From 10^-4 = exp(-9,2) it is obtained x = 3σ. This also implicates, that for 10⁴ measurement periods, there is a probability of 37 % to find a deviation above the 3σ value. If a large deviation indicates the possibility for an incident, then a threshold setting on x = 3σ means a false alarm rate of 10^-4.
In the example above x and y can be flows. Then incidents are expected to give rise to lower flow values downstream the queue, and by that, it is only single sided deviations that are needed to be considered, whereby the probability values only are 50 % of those above.

At accumulation of several measurement periods, the noise deviations are added squared, i e after n periods σ(n) = (n)^0,5 * σ(1). The signal is added linearily, i e in comparison s(n) = n* s(1). Thereby an integration effect is obtained s(n)/σ(n)= =n^0,5* s(1)/σ(1).
At false alarm rate 10^-4 , it is now obtained that the threshold related to one period, can be decreased at accumulated measurement periods according to (9,2/n)^0,5, i e x = 3 * n^-0,5 * σ. Thus the accumulated mean value has got a lower threshold.

In an applicable example e g, the flow of cars during a 30 seconds period is in average 16 cars and the related σ-value is 4. If an incident is blocking one of two lanes and queue-forming is reducing the flow to 8 cars, then the deviation is 8 cars = 2σ, and the condition is, for exceeding the threshold at an false alarm rate about 10^-4, that 2σ is larger than (9,2/n)^0,5 * σ. The number of measurement periods thus needs to be above 9.2/4, i e larger than 3. If the distribution instead had been simply linear, i e exp(-x/σ), then there had been needed more than 20 periods.

According to the invention one can instead use the deviations between the predicted and measured value as the base for the distribution function. Thereby the needed deviations can be considerable reduced. E g assume that 70 % of the deviation size can be predicted, then σ = 30 % of 4 giving the new σ =1,2. Then it is required that the signal is larger than 9,2^0,5* 1,2 = 3,6 cars. With the incident-caused deviation of 8 cars, there is now obtained an acceptable incident detection already at the first measurement period.
Giving several sensors in a large road network, it is even more important to keep the false alarm rate low. If with 100 sensors, one wants to limit the number of false alarms to one per day, one get in total 100* 24 * 60 * 2 periods, and the corresponding requirement on false alarm is 3,5 * 10^-6.

Traditional methods, according to reports, have problems with long detection times, which also was illustrated above. The innovation offers as shown very large savings of time.

Some examples are given below, for the use of the invented method in the traffic management process. In comparison, traditional methods for control of traffic e g at an access road, would mean an isolated pointlike control of traffic at the ramp.

With the invention and its use of prediction, there is created time-margins for control of traffic as connected in a road network. This is a large step forward, as traffic inherently is a network phenomenon. Problems arise in narrow sections, bottle-necks, - but the solution of the problem should be network based, otherwise the problem often is only transfered to another point.

Closely related to network control is managing of traffic between different routes. If at prediction of traffic on a road network, traffic problems are predicted e g overloading, then this problem often might be avoided, by taking actions in time, e g traffic can be guided to another route, whereby the traffic volume on the original route is decreased to a level below overloading.
Route guidance might e g be performed by the use of "VMS", variable message signs. The message might e g contain information about different grades of problems on the given route.
The larger the problem the larger the number of drivers that will consider choosing an alternative route. By sensor measurements directly connected to the position for the choice of routes, a fast feedback is obtained of the share of drivers, who chosed the alternative road route. That measure is also used for updating the value of strongness ofthe presently shown message, whereby the system successively stores an updated measure of the strongness for the respective messages. Thus the system beforehand can choose a message matching that share of the drivers, which is desireable for choosing a new route.
It is an ingredient of the invention to predict the result of the actions. That is important as no action should be chosen giving rise to new problems. Calibration and updating is performed by successively measuring the consequences of the actions, and then matching the stored value of strongness for a message to the actions. In this process a slower rate of updating is preferrably chosen, in a way that deviations are only partially changing the former value.

The innovation is also suitable for management of "park and ride", e g parking the car and taking the train or bus, - where the control information partly is based on predicted problems at the road net-work. Another area of use is the control of departure, e g information about traffic problems might influence some drivers to choose another transportation means or to delay the travel.

Claims

A method of traffic management for use in electronic traffic systems, where the method determines queue-forming of vehicles in a road network using sensor information from a number of sensors, where queue-forming may occur within the sensing area of a first sensor on a route upstream or downstream of a second sensor, or queue-forming may occur on a route between two sensors, and the said queue-forming or risk for queue-forming being detected with the use of at least one of upstream or downstream sensor respectively, and measurements sensed by the sensors are processed by using algorithms having parameters characterised in:

a. using a downstream sensor according to at least one of b and c below,

b. detecting the arisen difference in traffic flow at queue-situations compared with no queue, the sensed traffic flow being indicative of said difference;

c. obtaining traffic relations from measurements by the upstream and downstream sensors and using the said relation together with updated or present traffic parameters to predict or estimate traffic parameter on said route and to predict or detect queue-forming or risk for queue-forming.

d. calibrating and updating at least one parameter of the algorithms by a process comparing predicted or estimated traffic parameter values with corresponding values obtained from measured values of the sensors, using a number of measurements related to a time span of a number of measurement time periods for the changes correcting the calibration and updating of the said algorithm parameters.
A method according to claim 1, indicating queue-forming, characterised in using at least one of a and b below,

a. comparing predicted or estimated traffic parameter values with traffic parameter values based on measurements for a corresponding time stamp and using the size of deviations for indication of a traffic breakdown starting queue-forming;

b. comparing present or predicted traffic flow values with certain corresponding threshold values and using the condition that if traffic still would have been free-flowing, the traffic flows would exceed said threshold values, and using that for indication of a traffic breakdown starting queue-forming;
A method according to claim 1 or 2, characterised in,

a. using an upstream sensor for detecting speed-decrease or density increase in relation to a free-flow situation using measurements of at least one traffic parameter of the parameter group consisting of traffic flow, traffic velocity and traffic density;

b. calibrating and updating at least one parameter of the algorithms in at least one of claim 1, 2 and 3 by using the said detection.
A method according to claim 1,2 or 3, characterised in that, the predicted or estimated flow valid for a given time stamp at the downstream sensor position, is compared with the corresponding measured flow for the same said time stamp or compared by using adapted to corresponding time periods, and wherein the difference between said predicted or estimated and said measured values, optionally accumulated over a number of measuring periods, is used as an indication on queue-forming; and wherein one or both of the traffic states, queue or not queue, are used in the said prediction or estimation of flows, which are used in the said comparison with the measured values.
A method according to one of the claims 1 up to 4, wherein the differences between the predicted or estimated and measured values are regarded as members of stochastic distributions, dependent on the traffic situation, which may be free-flow traffic or traffic containing a queue, and wherein said differences, when larger than a standard deviation, σ, become more improbable the larger they are, inasmuch they are not indications of corresponding changes in the traffic situation; the method being further characterised by using a single difference or accumulated differences as members of such said stochastic distributions, and judging differences which are larger than a predetermined threshold, as an indication or probability of the occurrence of a corresponding change in the traffic situation.
A method according to one of the claims 1 up to 5, characterised by

a. estimating the probability for queue-forming by;

b. estimating values on at least one traffic parameter of the group flow, density and velocity from measured values, and regarding as members of stochastic distributions at least one of (b1) to (b3);

b1. the said estimated values,

b2. differences between the said estimated values and predetermined values.

b3. differences between the said estimated values and values on the same traffic parameter estimated from another set of measured values,

c. using a measure on the deviations of a said stochastic distribution; the measure according to at least one of (c1) to (c3); the stochastic distribution according to at least one of (c4) to (c5);

c1. a standard deviation,

c2. an upper quartile or another level-measure on the distribution,

c3. a predetermined measure

c4. a predetermined stochastic distribution.

c5. a stochastic distribution obtained from or adapted to the said members in (b),

d. relating a factor q1 * (the measure in c) to the probability of obtaining a smaller or larger value for the member in (b),

e. estimating the probability for the occurrence of queue-forming related to the value of a member in (b) by use of the probability relations in (d).
A method according to one of the claims 1 up to 6, further characterised by the determination of the occurrence of incidents, wherein the detection of a queue in a situation where the probability is small that it is due to a cause other than an incident is interpreted as an indication that an incident may have occurred;
A method according to one of the claims 1 up to 7, further characterised by the determination of the occurrence of incidents according to at least one of a, b and c:

a. wherein an unexpected stationary queue-front in the case of a traffic situation with low traffic flows is interpreted as an indication that an incident may have occurred;

b. wherein the occurrence of a non-stationary queue-front is interpreted as an indication that an incident may not have occurred;

c. wherein the occurrence of a queue in a traffic situation with large traffic flows may be due either to an incident or an overloading of a part of the traffic network, requiring the use of additional information to determine the case, the detection of substantially blocked traffic for a predetermined time interval, being interpreted as an indication of the occurrence of an incident in circumstances where an overloading of the network would be expected to give rise to a queue with limited but nevertheless significantly larger traffic passability.
A method according to one of the claims 1 up to 8, further characterised by using the standard deviation or a similar measure on deviation between the predicted or estimated values and corresponding measured values multiplied by a factor q to obtain a q-related measure of the probability that a single or accumulated measured deviations with the value (q * the measure of deviation) represents the presupposed traffic situation; wherein the q-relation to the probability is governed by the assumed distribution; and may be defined by use of a probability function or be stored with related values in tables.
A method according to one of the claims 1 up to 9, used to reduce the number of false alarms of incidents below a predetermined false alarm rate, further characterised in that the corresponding probability of a false alarm is determined related to a first deviation threshold level, wherein optionally an associated first q-value is used to determine the threshold level, (q* the measure of deviation) ; and the deviations between the predicted and measured flow-values are compared with the threshold value for acceptance of a possible incident detection; and when accumulated deviations are used, those deviations are compared using a second threshold value or a second q-value, related to the number of accumulated values;
A method according to claim 10, further characterised by using an approximate method wherein the said accumulated value divided by the root-square of the number of accumulated values is compared using said first threshold or said first q-value which then can be regarded as substantially independent from variations of the number of accumulated values.
A method according to one of the claims 6 up to 11, characterised in that the traffic situation is further predicted on the assumption that an incident has occurred and that measured deviations from this predicted situation are used to judge if an incident has occurred.
A method according to claim 10, 11 or 12, further characterised in that queue detection is performed in a manner corresponding to the detection of incidents.
A method according to one of the claims 1 up to 13, further characterised in that determination of the traffic situation on a given road-link is performed using flow relations, derived from upstream sensors, preferably a sensor on the main road and a sensor on the access road; and wherein prediction of the flow at the connection is synchronised over the same time period, such that the prediction matches two traffic flows merging in the connection point, thereby obtaining dynamic traffic conditions and the larger risk for traffic breakdown and queue-forming when larger traffic flows including flow peaks are merging, and wherein the predicted flows are compared with the threshold values;
A method according to claim 14, further characterised by estimating the threshold values for the main road approximately as C_v = C₀ - a * I_e , where I_e is the flow on the access road or on the exit road and "a" is the related factor and C_o is a constant; and wherein C_o and "a" may be empirically obtained and may additionally be updated for respective road link, by comparing the predicted values with the real ones obtained from measurements at the sensors.
A method according to one of the claims 1 up to 15, further characterised in that the off-flow in the front of a queue is approximated by using substantially the factors in b * (g)^-0,5 , where g is the gap between cars at the queue-front and b is a factor, which may be given as a constant, and which may be updated by comparison with measured values; and wherein the off-flow may be determined from (Va - Vq)/(Da - Dq), where V is the velocity and D is the periodic distance for the cars and "a" indicates the situation downstream the queue and "q" in the queue; and wherein those relations are used together with the traffic flows, I = V/D, for determination of the off-flow at the queue-front; and wherein optionally growth and decay of the queue may be determined by the use of information about the flow also behind the queue.
A method according to one of the claims 1 up to 16 for use in motorway systems, where signs can be controlled to display warnings concerning traffic problems, optionally with recommended speed limits, further characterised in that when traffic disturbances are predicted, cars present in the vicinity of the disturbance are given a warning recommending speed decrease, and if considered appropriate upstream signs are changed to give warnings, optionally with recommended speed decrease, alerting upstream cars to the imminent problem situation; and when traffic disturbances are detected, upstream cars are alerted, optionally with recommended lower speed limits, and the sign information may optionally be directed to individual lanes.
A method according to one of the claims 1 up to 17 for use in the control of access traffic, preferable with the use of ramp-metering, further characterised in that light signs at access roads are controlled on the basis of prediction of traffic disturbances such that said light signs change their red-green time intervals to decrease the on-flow during the relevant traffic load time interval, whereby the risk that traffic disturbances occur in reality is reduced; and wherein said light signs are controlled based on the detection of traffic disturbances to change the on-flow and reduce the traffic load thereby reducing or dissolving the disturbances.
A method according to claim 18, wherein several access roads along the same road or chosen route can be controlled, said method further characterised in that prediction of traffic disturbances at a certain time, one or more of upstream accesses are controlled giving reduced on-flow during an early time period, corresponding to the downstream traffic load time period, and thereby reducing the risk that traffic disturbances arise in reality; and wherein the choice of the manner in which the on-flow is reduced is determined by the consequences for the traffic on the road network at the respective access road; and wherein the result of upstream on-flow control at one access road may be corrected at downstream accesses at a corresponding later time; and upon detection of traffic disturbances, a corresponding method can be used at upstream accesses to reduce the traffic load and thereby reduce or dissolve the disturbance.
A method according to one of the claims 1 up to 19 for use in route guidance, wherein a road user may choose a route based on message information about predicted or detected downstream traffic disturbances, further characterised in that the response of the road users to the message information is predicted with prediction factors determined by a feedback method according to which downstream sensors supply measurement values indicative of the road user response to the message information; and wherein the prediction factors are successively updated during operation so that message information providing changes in the choice of route can be preselected in order to reduce or prevent predicted traffic disturbances arising in reality and the sensors also give feed-back concerning the response to the message information so that corrections can be implemented by the choice of new message information.
A method according to one of the claims 1 up to 20 for use in traffic control in a road network, further characterised in that results of actions can be predicted by use of prediction factors, which relate the action to the consequence of the action, in which the succeeding measurements of results are used to update the prediction factors;
A method according to claim 21, further characterised in that the method is used in at least one of (a) and (b):

a. wherein actions work on several links in a road network;

b. wherein several actions can be connected and the combined result be predicted by use of relevant prediction factors.