EP3813034A1 - Method for determining an adapted current cycle time value for a system - Google Patents

Abstract

The invention relates to a method for determining an adjusted current remaining time value for an installation, comprising the steps a. Providing a plurality of previous remaining time values each with at least one associated confidence in a specific time period (S1), comprising b.
at least one last previous remaining time value from the plurality of previous remaining time values, which immediately precedes the current remaining time value in time in the specific period of time; c. Receiving a current remaining time value with at least one current confidence (S2); and d. determining the adjusted current remaining time value on the basis of the plurality of previous remaining time values with the respective associated confidences as a function of the current confidence (S3). Furthermore, the invention is directed to a corresponding computer program and a determination unit.

Description

1. Technical field

The invention relates to a method for determining an adjusted current remaining time value for an installation. The invention also relates to a computer program product and a corresponding determination unit.

2. State of the art

Residual value predictions are known from the prior art. The residual time value in residual value predictions is the duration until the next switchover point in a system. The remaining time value can also be viewed as the remaining time and the duration as the remaining time. The system can be any technical system, such as a traffic light system.

If the residual value forecast is known and correct, the duration up to a switchover point in a technical system can be counted down linearly as the time progresses. The residual second value output by the forecast models can, however, be significantly influenced by various factors, for example fluctuating due to statistical uncertainties. Furthermore, unforeseen or unexpected events can occur and thus cause a new or changed residual value forecast that is not on the previous forecast line.

As an example, the residual value forecast of a traffic light system is mentioned below as a system in which the duration a changeover time of a signal group in the future is considered. With regard to the traffic light system, for example, a public transport registration or a detector triggered by the road user can significantly influence the residual value.

The user of the residual value forecast expects a linear countdown of the residual time value up to the point in time of the switchover. The user can, for example, be a driver who is on the direct route to the traffic light system. It is therefore desirable to avoid fluctuations in the displayed remaining time that are perceived as random and thereby improve the credibility of the behavior of the prognosis system.

Up to now, known approaches have usually provided the user with the output of a neural model, such as switching time and confidence. However, the forecast is simply hidden as soon as a given confidence value is not reached.

The disadvantage of the known approaches, however, is that they insufficiently take into account the above-mentioned factors, such as events. The remaining time predictions therefore have a low level of reliability.

The present invention therefore addresses the objective technical problem of providing a method for determining an adapted current remaining time value for a system, which method is more reliable and efficient.

3. Summary of the Invention

1. a. Bereitstellen einer Mehrzahl vorheriger Restzeitwerte mit jeweils mindestens einer zugehörigen Konfidenz in einem bestimmten Zeitraum, aufweisend
2. b. mindestens einen letzten vorherigen Restzeitwert aus der Mehrzahl der vorherigen Restzeitwerte, welcher dem aktuellen Restzeitwert zeitlich in dem bestimmten Zeitraum unmittelbar vorausgeht;
3. c. Empfangen des aktuellen Restzeitwerts mit mindestens einer aktuellen Konfidenz; und
4. d. Bestimmen des angepassten aktuellen Restzeitwerts auf Basis mindestens eines vorherigen Restzeitwerts der Mehrzahl der vorherigen Restzeitwerte mit den jeweiligen zugehörigen Konfidenzen in Abhängigkeit von der aktuellen Konfidenz.

The above-mentioned object is achieved according to the invention by a method for determining an adjusted current remaining time value for a system, comprising the steps:

1. a. Providing a plurality of previous remaining time values, each with at least one associated confidence in a specific period of time, having
2. b. at least one last previous remaining time value from the plurality of previous remaining time values, which immediately precedes the current remaining time value in time in the specific period of time;
3. c. Receiving the current remaining time value with at least one current confidence; and
4. d. Determining the adjusted current remaining time value on the basis of at least one previous remaining time value of the plurality of previous remaining time values with the respective associated confidences as a function of the current confidence.

Accordingly, the invention is directed to a method for determining an adjusted current remaining time value for a plant. The remaining time value is the duration until the next switchover point in the system, as explained above. The remaining time value can also be viewed as the remaining time and the duration as the remaining time. In the case of a traffic light system such as a traffic light as a system, the traffic light is switched from one operating state to another operating state at the time of switching, for example the traffic light changes from a green light to a red light. In other words, done the signal change in this example at the time of the switchover.

In a first step, the input data set is received. The input data set can be received via one or more interfaces by the determination unit below, such as a computer unit.

The input data set has a plurality of previous remaining time values with the respective confidences in a specific period of time. In other words, a time series with previous or previous known remaining time values is considered. The time series consequently also has a first remaining time value and a last remaining time value in the period.

Furthermore, the input data record also has a current remaining time value with its confidence. The current remaining time value is adapted by the method according to the invention and provided as an adapted current remaining time value.

The confidence is the certainty or probability that the deviation of the forecast remaining time from the actual or true remaining time does not exceed a specified value. 100% confidence means that the expected deviation of the remaining time forecast from the actual remaining time is always within the specified deviation interval limits. With a confidence of 0%, no conclusions can be drawn about the expected deviation from the actual switchover time.

The current remaining time value is adapted on the basis of at least one previous remaining time value from the plurality of remaining time values with the associated confidence as a function of the current confidence.

In this step, the last previous remaining time value can be used as the directly preceding remaining time value for adjustment. In addition, the further history can also be used implicitly via the internal state, for example if linear extrapolation is carried out within the dead band. In this case, the start value of this extrapolation can be further in the past, if necessary it can be the first value of the current cycle.

Accordingly, in other words, the history of the previous remaining time values and their confidences are evaluated as a function of the confidence of the current remaining time. As a result, all or part of the previous remaining time values can be used for adjustment.

The method according to the invention uses both the predicted remaining time and the confidence of the prediction from several time steps in order to achieve a smoothing of the current prediction. Consequently, a user advantageously perceives a steady countdown of the current remaining time value, which he can for example get displayed after the determination. The method can therefore also be referred to as a smoothing method.

If, for example, a discrete event occurs and the prognosis then leaves a predefined dead band, the adjusted current remaining time value is determined. With reference to the above exemplary light signal system, the discrete event can be a public transport telegram input or a detector pulse. The determination can be made by replacing the continuous forecast with a combination of linearly extrapolated previous forecast and the result of the updated forecast of a neural network. In one embodiment, the remaining time values and confidences are received by a computing unit or memory unit via at least one input interface. This refinement has proven to be particularly advantageous with regard to efficient and reliable access to the data. The storage unit can be designed as a volatile or non-volatile storage medium, such as a database or cloud. The computing unit can be designed as any computing unit.

In a further embodiment, the remaining time values and confidences are output values of a neural network. Accordingly, the predicted remaining time and the confidence of the remaining time prediction are output from a neural network, i.e. from machine learning.

• Bereitstellen des aktuellen Restzeitwerts als angepassten aktuellen Restzeitwert, falls die mindestens eine aktuelle Konfidenz des aktuellen Restzeitwerts einen festgelegten maximalen Grenzwert überschreitet;
• Bestimmen des angepassten aktuellen Restwerts durch Extrapolieren des letzten vorherigen Restzeitwerts, falls die mindestens eine aktuelle Konfidenz des aktuellen Restzeitwerts einen festgelegten minimalen Grenzwert unterschreitet,

und
Bereitstellen des extrapolierten Restzeitwerts als angepassten aktuellen Restzeitwert; oder

• Bestimmen des angepassten aktuellen Restzeitwerts durch Extrapolieren des letzten vorherigen Restzeitwerts, falls die mindestens eine aktuelle Konfidenz des aktuellen Restzeitwerts einen bestimmten Prozentsatz der vorherigen Konfidenz des letzten vorherigen Restzeitwerts unterschreitet und

Bereitstellen des extrapolierten Restzeitwerts als angepassten aktuellen Restzeitwert.In a further embodiment, the method also has the step

• Providing the current remaining time value as an adjusted current remaining time value if the at least one current confidence of the current remaining time value exceeds a specified maximum limit value;
• Determining the adjusted current remaining value by extrapolating the last previous remaining time value if the at least one current confidence of the current remaining time value falls below a specified minimum limit value,

and
Providing the extrapolated remaining time value as an adjusted current remaining time value; or

• Determining the adjusted current remaining time value by extrapolating the last previous remaining time value if the at least one current confidence of the current remaining time value falls below a certain percentage of the previous confidence of the last previous remaining time value and

Providing the extrapolated remaining time value as an adjusted current remaining time value.

Accordingly, the current remaining time value is provided directly with an extremely high level of confidence without adjustment.

The last remaining time value with an acceptable confidence level is extrapolated linearly if the confidence level is low.

• linear extrapoliert solange der aktuelle Restzeitwert innerhalb eines Totbandes um den extrapolierten Restzeitwert bleibt, wobei die Weite des Totbandes von der Konfidenz der aktuellen und bisherigen Prognosewerte abhängt
• die lineare Extrapolation wird anteilig um die Abweichung von der aktuellen Modellausgabe korrigiert, wobei der Anteil der Korrektur von den Konfidenzen der Prognosewerte abhängt.

The current remaining time value is given with an acceptable confidence level

• linearly extrapolated as long as the current remaining time value remains within a dead band around the extrapolated remaining time value, the width of the dead band depending on the confidence of the current and previous forecast values
• the linear extrapolation is proportionally corrected for the deviation from the current model output, the proportion of the correction depending on the confidence levels of the forecast values.

In a further embodiment, the extrapolation includes the linear counting down of a predetermined cycle time of the system by a counting unit. Accordingly, the current remaining time value can be counted down by seconds, minutes or other cycle times, for example of the traffic signal system. The counting unit enables reliable extrapolation.

In a further embodiment, the maximum limit value is a confidence greater than 90%, preferably 95%.

In a further embodiment, the minimum limit value is a confidence between 20% and 30%.

In a further embodiment, the system is a light signal system or other system in the area of traffic. These limit values have proven to be particularly advantageous with regard to the reliability of the adjusted current remaining time value as output value.

• Ausgeben des angepassten aktuellen Restzeitwerts und/oder zugehöriger Daten auf einer Anzeigeeinheit,
• Speichern des angepassten aktuellen Restzeitwerts und/oder zugehöriger Daten in einer Speichereinheit, und
• Übermitteln des angepassten aktuellen Restzeitwerts und/oder zugehöriger Daten an eine Recheneinheit.

In a further embodiment, the method also has the step
Carrying out a measure, the measure being selected from the group consisting of:

• Outputting the adjusted current remaining time value and / or associated data on a display unit,
• Storing the adjusted current remaining time value and / or associated data in a memory unit, and
• Transmission of the adjusted current remaining time value and / or associated data to a computing unit.

Accordingly, one or more measures can be initiated after the adapted current remaining time value has been determined as the output value of the method according to the invention. The measures can be carried out simultaneously, one after the other or also in stages.

First of all, the output value can be displayed to the user on a display unit of a computing unit. In contrast to the prior art, the user advantageously perceives the output of the prognosis as a steady countdown. This improves confidence in the underlying forecast system. Furthermore, the output value can be stored, the output value itself or can be transmitted in the form of a corresponding message or message to another unit, such as a terminal, a control unit or other computing unit. The receiving computing unit can also initiate further appropriate measures after receipt. Further measures are route planning, starting an engine of an autonomous vehicle or other control measures.

The invention also relates to a determination unit. The method according to the invention is accordingly carried out by a determination unit. The determination unit is any arithmetic unit. In addition to determining the adjusted current remaining time value, the determining unit can also initiate one or more of the above measures itself. As a result, the measures are advantageously taken promptly and efficiently.

The invention also relates to a computer program product with a computer program, the means for carrying out the method described above when the computer program is executed on a program-controlled device.

A computer program product, such as a computer program means, can for example be provided or delivered as a storage medium such as a memory card, USB stick, CD-ROM, DVD, or also in the form of a downloadable file from a server in a network. This can be for example take place in a wireless communication network by transmitting a corresponding file with the computer program product or the computer program means. In particular, a control device such as an industrial control PC or a programmable logic controller or a programmable logic controller, PLC for short, or a microprocessor for a smart card or the like can be considered as the program-controlled device.

4. Brief description of the drawings

FIG 1

zeigt ein erfindungsgemäßes Ablaufdiagramm des Verfahrens zum Bestimmen eines angepassten aktuellen Restzeitwerts für eine Anlage.

FIG 2

zeigt eine beispielhafte Restzeitprognose gemäß dem Stand der Technik ohne Glättung.

FIG 3

zeigt eine beispielhafte Restzeitprognose gemäß einer Ausführungsform der Erfindung mit Glättung.

In the following detailed description, presently preferred embodiments of the invention are further described with reference to the following figures.

FIG 1

shows a flowchart according to the invention of the method for determining an adjusted current remaining time value for an installation.

FIG 2

shows an exemplary remaining time prognosis according to the prior art without smoothing.

FIG 3

shows an exemplary remaining time prognosis according to an embodiment of the invention with smoothing.

5. Description of the Preferred Embodiments

Preferred embodiments of the present invention are described below with reference to the figures.

Figure 1 represents a flowchart of the method according to the invention with the method steps S1 to S3 schematically. The individual method steps can each be carried out by the determination unit or also its sub-units become. In the first two steps S1 and S2, the input data records are received, previous forecasts and the current forecast, as explained in detail above.

The forecasts can be output information from a neural network and include the forecast remaining time until the signal change and the confidence in the remaining time forecast.

1. 1) einen Einbruch der prognostizierten Restzeit bei geringer Konfidenz mit anschließender Erholung
2. 2) Jitter der Prognose mit abwechselnd steigenden und fallenden Restzeitprognosen
3. 3) Kurzfristige Spikes mit verringerter Konfidenz
4. 4) Stark schwankende Restzeitprognosen insbesondere am Beginn einer Phase

According to the prior art, only the current remaining time value of the remaining time forecast is output regardless of the associated current confidence. The adverse effects are in Figure 2 shown. These effects include, for example

1. 1) a collapse in the predicted remaining time with low confidence with subsequent recovery
2. 2) Jitter of the forecast with alternately rising and falling remaining time forecasts
3. 3) Short-term spikes with decreased confidence
4. 4) Strongly fluctuating remaining time forecasts, especially at the beginning of a phase

These effects are mitigated by the method according to the invention in step S3 as follows. In particular, the confidence of the current forecast is taken into account when determining the adjusted current remaining time value.

Smoothing out small fluctuations

The remaining time prognoses often show smaller fluctuations, which can lead to undesired sign changes in the gradient and to an uneven countdown. The According to one embodiment of the invention, these fluctuations can be smoothed by discarding the current value and accepting the previous value reduced by one for small relative deviations from the previous forecast minus one second, e.g. less than 20% of the last value reduced by one becomes.

The pseudocode can be represented as follows: $$\mathit{if}\left(\mathit{Absolute\; work}\left(\mathit{current\; forecast}-\left(\mathit{last\; forecast}-\mathit{1}\right)\right)<=\mathit{20th}\%\mathit{of\; the\; last\; forecast\; value}\right)\mathit{then\; current\; forecast}=\mathit{last\; forecast}-\mathit{1}$$

Smoothing out larger fluctuations

If the fluctuations in the remaining time prognoses are greater than the above small fluctuations, this prognosis can, according to one embodiment of the invention, be smoothed by a first-order IIR filter in which the filter coefficient depends on the confidence of the current prognosis. This advantageously leads to a quick adjustment of the prognosis with a high confidence, with only large steps being reduced, while with a lower confidence the adjustment takes place correspondingly more slowly.

The pseudocode can be represented as follows: $$\mathit{current\; forecast}=\mathit{last\; forecast}+\mathit{Rounded\; value}/\left(\left(\mathit{current\; forecast}-\mathit{last\; forecast}\right)*\mathit{Filter\; coefficient}*\mathit{Confidence\; of\; the\; current\; forecast}\right)$$

Integration of permanent small deviations from the smoothed forecast

The smoothing of small fluctuations in the remaining time forecast can result in a permanent constant deviation in the remaining time forecast. In order to reduce such constant deviations, according to one embodiment of the invention, the deviation between the original and the smoothed forecast can be integrated, this deviation being scaled using an integrator coefficient and the confidence of the current forecast value.

wenn integrierte Abweichung > 1 $$\mathrm{dann\; aktuelle\; Prognose}=\mathrm{aktuelle\; Prognose}+1\mathrm{und\; integrierte\; Abweichung}=\mathrm{integrierte\; Abweichung}-1$$

wenn integrierte Abweichung < -1 $$\mathrm{dann\; aktuelle\; Prognose}=\mathrm{aktuelle\; Prognose}-1\mathrm{und\; integrierte\; Abweichung}=\mathrm{integrierte\; Abweichung}+1$$

The pseudocode can be represented as follows: $$\mathrm{integrated\; deviation}=\mathrm{last\; integrated\; deviation}+\left(\mathrm{originally}\ddot{\mathrm{u}}\mathrm{related\; prognosis}-\mathrm{equal}\ddot{\mathrm{a}}\mathrm{killed\; forecast}\right)*\mathrm{Integrator\; Coefficiency}*\mathrm{Confidence\; of\; the\; current\; forecast}$$

if integrated deviation> 1 $$\mathrm{then\; current\; forecast}=\mathrm{current\; forecast}+1\mathrm{and\; integrated\; deviation}=\mathrm{integrated\; deviation}-1$$

if integrated deviation <-1 $$\mathrm{then\; current\; forecast}=\mathrm{current\; forecast}-1\mathrm{and\; integrated\; deviation}=\mathrm{integrated\; deviation}+1$$

Smoothing of spikes with reduced confidence

In the forecast remaining time, there may be brief jumps with a simultaneous drop in confidence up to spikes. In order to compensate for these short-term disturbances, the delivered forecast is discarded in the event of significant drops in confidence, e.g. to below 80% of the confidence of the last forecast according to one embodiment of the invention, and the last forecast is counted down further.

The pseudocode can be represented as follows:
if (confidence of the current forecast <80% of the confidence of the last forecast ) $$\mathit{then\; current\; forecast}=\mathit{last\; forecast}-\mathit{1}$$

Smoothing of forecasts with high or low absolute confidence

If a remaining time prognosis has a particularly high confidence, e.g. greater than 95% with which the prognosis can thus be assumed to be secure, then this prognosis is not further smoothed according to one embodiment of the invention, but can be taken over directly. Even without smoothing, this forecast is very close to the target value.

The pseudocode can be represented as follows:
if the confidence of the current prognosis> confidence of a prognosis assumed to be certain $$\mathit{then\; current\; forecast}=\mathit{originally}\ddot{\mathit{u}}\mathit{ual\; forecast}$$

If a forecast has a very low absolute confidence, for example <20%, and the last forecast has a higher confidence, then the last forecast remaining time is further counted down according to an embodiment of the invention.
if (confidence of the current prognosis <minimum necessary confidence) and (confidence of the last prognosis> minimum necessary confidence) $$\mathit{then\; current\; forecast}=\mathit{last\; forecast}-\mathit{1}$$

• Zu 1) Der Einbruch der prognostizierten Restzeit wird vermieden.
• Zu 2) Der Jitter der Restzeitprognose wird geglättet.
• Zu 3) Die kurzfristigen Spikes mit verringerter Konfidenz werden ignoriert.
• Zu 4) Die großen Schwankungen in der Restzeitprognose werden verringert.

Figure 3 shows the smoothed remaining time forecast according to an embodiment of the invention. The adverse effects can be reduced as follows:

• Regarding 1) The drop in the forecast remaining time is avoided.
• Regarding 2) The jitter of the remaining time forecast is smoothed.
• To 3) The short-term spikes with reduced confidence are ignored.
• To 4) The large fluctuations in the remaining time forecast are reduced.

The following pseudocode shows an exemplary implementation for the method according to an embodiment of the invention with smoothing in Python, with additional secondary conditions, for example no negative remaining time, being taken into account.

Claims (11)

Method for determining an adjusted current remaining time value for a plant, comprising the steps: a. Providing a plurality of previous remaining time values each with at least one associated confidence in a specific time period (S1), having b. at least one last previous remaining time value from the plurality of previous remaining time values, which immediately precedes the current remaining time value in time in the specific period of time; c. Receiving the current remaining time value with at least one current confidence (S2); and d. Determining the adjusted current remaining time value on the basis of at least one previous remaining time value of the plurality of previous remaining time values with the respective associated confidences as a function of the current confidence (S3). Method according to claim 1, wherein the remaining time values and confidences are received via at least one input interface from a computing unit or memory unit. Method according to Claim 1 or Claim 2, in which the remaining time values and confidences are output values of a neural network. Method according to one of the preceding claims, further comprising - Providing the current remaining time value as an adjusted current remaining time value if the at least one current confidence of the current remaining time value exceeds a specified maximum limit value; - Determination of the adjusted current remaining value by extrapolating the last previous remaining time value if the at least one current confidence of the current remaining time value falls below a specified minimum limit value,
and
Providing the extrapolated remaining time value as an adjusted current remaining time value; or - determining the adjusted current remaining time value by extrapolating the last previous remaining time value if the at least one current confidence of the current remaining time value falls below a certain percentage of the previous confidence of the last previous remaining time value and
Providing the extrapolated remaining time value as an adjusted current remaining time value. The method of claim 4, wherein the extrapolating comprises linearly counting down a predetermined cycle time of the plant by a counting unit. Method according to Claim 4 or Claim 5, the maximum limit value being a confidence greater than 90%, preferably 95%. A method according to claim 4 or claim 5, wherein the minimum limit value is a confidence between 20% and 30%. Method according to one of the preceding claims, wherein the system is a light signal system or other system in the area of traffic. Method according to one of the preceding claims, further comprising the step
Carrying out a measure, the measure being selected from the group consisting of: - Outputting the adjusted current remaining time value and / or associated data on a display unit, - Storing the adjusted current remaining time value and / or associated data in a storage unit, and - Transmission of the adjusted current remaining time value and / or associated data to an arithmetic unit. Determination unit for performing the method according to one of the preceding claims. Computer program product with a computer program which has means for carrying out the method according to one of Claims 1 to 9, when the computer program is executed on a program-controlled device.

Images