Method for predicting expected road surface temperature

Abstract

The invention relates to a method for predicting an expected road surface temperature of at least one road section, in which a surface temperature of a road surface in the road section is measured by means of at least one sensor, in which environmental information is associated to the measurement and sent together with the surface temperature to a rear end, machine learning is used to learn a relationship between environmental information and surface temperature such that an expected road surface temperature can be estimated. According to the invention, an interval having a maximum possible value and a minimum possible value of the road surface temperature is generated as an output value depending on the position.

Description

TECHNICAL FIELD

[0001] The invention relates to a method for predicting an expected road surface temperature of at least one road section according to the preamble of claim 1. BACKGROUND

[0002] DE 10 2015 000 394 A1 discloses a method for generating a digital map, wherein location-dependent estimates of physical measured variables are recorded in the digital map. Measurement data about road conditions and, for example, data about temperatures can be collected by a fleet of vehicles. These data can be transmitted by mobile radio to a server and processed therefrom. The digital map with associated collected data can be transmitted to a vehicle, so that road conditions can be provided to a driver assistance system depending on the location on the map. The driver assistance system can accordingly switch on or off functions, so that it responds to environmental conditions.

[0003] DE 10 2019 135 751 A1 describes a general method for predicting an expected road surface temperature of at least one road section, wherein a surface temperature of a road surface in the road section is measured by at least one sensor. Environmental information is associated to the measurement and sent to a backend together with the surface temperature. Machine learning is used to learn a connection between the environmental information and the surface temperature, so that an expected road surface temperature can be estimated as a fixed value.

[0004] DE 10 2020 004 018 A1 describes the activation of an autonomous driving function depending on temperature conditions.

[0005] The disadvantage is that only data that has already been measured is fed into the system. Thus, future situations cannot be predicted. However, especially for the adaptation of functions provided by driver assistance systems, future situations facilitate taking action in an economic, sustainable and forward-looking manner. SUMMARY

[0006] It is an object of the present invention to provide a method for predicting an expected road surface temperature, wherein future data that cannot be measured directly on site can also be determined.

[0007] According to the invention, this object is achieved by a method for predicting an expected road surface temperature having the features of claim 1, in particular in the characterizing part of claim 1. Advantageous embodiments and improvements are given in the dependent claims.

[0008] In the core of the method according to the application, the surface temperature of the road in the road section is measured by at least one sensor, wherein environmental information is associated to the measurement and sent to the backend together with the surface temperature, wherein a machine learning is used to learn the connection between the environmental information and the surface temperature, so that an expected road surface temperature can be estimated. An interval is provided as an output value, which provides the maximum and minimum possible values of the road surface temperature depending on the location. Thus, the uncertainty in the value prediction can be quantified and better estimated. On the other hand, in known systems, only the value of the surface temperature is used as a so-called point estimator. With the method now provided, it can advantageously be ensured that the estimate is close to the actual value.

[0009] The method for predicting an expected road surface temperature of at least one road section makes it possible, for example, to generate a current road surface temperature map and to predict how the data of the road surface temperature map will behave over a certain period of time. For example, this type of map can be advantageous for road management, in particular for winter management of road operators, to ensure the safety of the infrastructure. In particular, however, the operability of vehicle assistance systems is enhanced. For example, this can significantly increase the availability of level 3 to 5 driving functions, since accurate information about the surface state and future values is available, so that the system can be prevented from switching off too early. Thus, the method can advantageously extend the current use restrictions of automated driving, the result of which is that automated functions are used in significantly more cases and safety-critical infrastructure can be displayed at an early stage. Advantageously, infrastructure operators can also maintain roads better, faster and more economically, since conclusions about the condition of the road can be drawn by predicting the road surface temperature over a period of time. Thus, the high data density can make up for previous knowledge gaps. This in particular offers advantages over previously known fixed measurements.

[0010] By using machine learning, it is possible to determine a surface temperature that represents a limit value of different aggregation states, for example wet or icy, or is essential for the conditions in a specific location. Thus, a more accurate prediction of the condition of the road can be made. Both the outside-in and the inside-out application can therefore react in an adaptive manner to adapt the driving behavior to the condition of the road.

[0011] For example, the actual detection of the road surface temperature based on the available data can only be carried out in the backend. Thus, the detection can depend on different requirements, for example transmission capacity and online calculation capacity. Based on the collected data, a machine learning model can be trained, which outputs a prediction interval, for example by using conformal prediction, so that the uncertainty of the machine learning model is predicted. Thus, the data is preferably output as data of an interval.

[0012] The initial training of the model for machine learning can take place on a development vehicle with a suitable sensor system that can measure the road surface temperature. In this case, the backend is in particular provided with this information or data by a plurality of different vehicles, so that an environmental map can be created.

[0013] Preferably, the expected road surface temperature at a future point in time can be determined by a conformal prediction algorithm, in which the uncertainty of the data of the machine learning is determined. Thus, the current road surface temperature at a specific geographic location and the temporal change of the road surface temperature can be estimated very precisely by experiments and by a large data density. In particular, it uses the output of a prediction interval that defines the boundaries of the machine learning. For example, a split conformal prediction, a full conformal prediction or a method between these two methods, such as a cross-conformal prediction or a Jacknife, can be used as a conformal prediction.

[0014] According to a very advantageous refinement of the idea, it can be provided that the environmental information is a geographic location and a time. Thus, the data can be assigned to a specific location at a specific time. For example, the environmental information can come from a measurement sensor system on the vehicle itself. In addition, environmental information from a weather station or measurement data from a road operator can be used. The environmental information can include states such as solar intensity, precipitation, snowfall, in particular air temperature. For this purpose, sensors and / or cameras and wireless connections, for example by mobile radio, can be used to transfer the data to a backend server. The sum of all these data, including data from the German Weather Service, can be collected anonymously at the backend.

[0015] In this case, according to an advantageous embodiment, it can be provided that further data are measured by a vehicle's on-board sensor system, which are sent to the backend together with the environmental information and the surface temperature. Thus, an improved geographical coverage of the weather data can be provided. In particular, the sensor system can be located at a very small distance from the road surface, so that in particular the data of the road surface can be specified very precisely.

[0016] According to a very advantageous refinement of the idea, it can be provided that data from a road operator are used and sent to the backend. In this case, publicly available weather data from stationary measurements can also be used. All these data increase the data density, so that the system can work more precisely.

[0017] A further advantageous embodiment can provide that the conformal prediction algorithm aggregates a set of data and divides it into a training set, a test set and a calibration set in an equal distribution.

[0018] In this case, according to a very advantageous refinement of the idea, it can be provided that the machine learning is based on the training set.

[0019] In this case, according to an advantageous embodiment, a definition quantile function can be provided which defines the anomaly of a data set with respect to a previous data set. Using a previously defined calibration set, a calibration score can be calculated. In this case, a definition quantile function of the calibration score can be calculated. The quantile can be used in particular to calculate a prediction interval for a new example such that the marginal coverage is satisfied in particular. In other words, this means a calculation of probabilities, wherein the probability that the prediction interval includes the correct label can be expressed with 1 - a, wherein a is in [0, 1].

[0020] An advantage of this approach is that it only needs to be split into a training set and a test set. For example, an AI model is trained on the training data set. The calibration data set can be omitted, although it is difficult to select this data set. Furthermore, the learning of the score function can be omitted, since this is trained or can be selected.

[0021] A further advantageous embodiment can provide for creating a map with road surface temperature information, wherein data from a fleet of vehicles is used. Furthermore, the created map can be provided to any vehicle in the fleet, wherein these vehicles must include a subsystem in particular in order to transmit or receive data which contributes to the determination of the map.

[0022] According to a very advantageous refinement of the idea, it can be provided that, depending on the determined current and future road surface temperatures, an autonomous driving function is enabled or not. Thus, autonomous driving functions can be used more extensively and must be deactivated less often. This includes in particular the availability of driving functions of level 3 to 5, which will increase significantly. Thus, these information can be used to expand autonomous driving. Furthermore, infrastructure operators can use these information to better maintain their road network. BRIEF DESCRIPTION OF DRAWINGS

[0023] Further advantageous embodiments of the method for predicting an expected road surface temperature according to the invention can also be derived from the embodiments described in more detail below with reference to the drawings.

[0024] In the drawings:

[0025] Figure 1 A schematic diagram showing the process of the method. DETAILED DESCRIPTION

[0026] Figure 1A possible procedure of the method is schematically shown. In a step marked 1, on-board data from a plurality of vehicles can be measured and stored, which are measured for example by a corresponding sensor system on the vehicle. Through a communication module, the data can be combined with offline data determined in a step marked 2. These so-called offline data can include data from environmental conditions and geographical locations, which are not measured by the vehicle, as in step marked 3. In this case, data from a weather provider can be transmitted to the data through a step marked 4. Back to step 2, from this it is possible to determine the road surface temperature, wherein the data can be determined by artificial intelligence.

[0027] For example, the environmental variables at a point in time t can be aggregated in the backend and an AI model can be used to predict the road surface temperature behavior at a future point in time t+n, where n is a natural number. For example, the following predictions can be made. In a very large interval, the AI model is not safe; here the model has difficulty making a decision. To improve, more data should be collected or the data should be homogenized. In a very small interval, the AI model is very safe and can well approximate reality.

[0028] In other words, the AI model links environmental variables to road surface temperature, wherein the uncertainty of the model can be quantified by conformal prediction. In summary, it is thus possible to provide a method for determining, providing and predicting road surface temperature for a geographical location.

Claims

1. A method for predicting the expected pavement temperature of at least one road segment, wherein, The surface temperature of the road surface in the road segment is measured using at least one sensor, wherein environmental information is correlated with the measurement and the environmental information is sent together with the surface temperature to a backend, wherein machine learning is used to learn the relationship between the environmental information and the surface temperature, enabling the estimation of the expected road surface temperature. Its features are, The output value is a range containing the maximum and minimum possible values ​​of the road surface temperature, based on the location.

2. The method according to claim 1, Its features are, The expected road surface temperature at future points in time is determined using a conformal prediction algorithm, where the uncertainty of the machine learning data is determined.

3. The method according to claim 1 or 2, Its features are, The environmental information refers to geographical location and time.

4. The method according to claim 1, 2 or 3, Its features are, Other data is measured by the vehicle's onboard sensor system, and this data, along with the environmental information and the surface temperature, is sent to the backend.

5. The method according to any one of claims 1 to 4, Its features are, Use data from road operators and send it to the backend.

6. The method according to any one of claims 2 to 5, Its features are, The conformal prediction algorithm aggregates a set of data and divides the set of data into a training set, a test set, and a calibration set with an equal distribution.

7. The method according to claim 6, Its features are, Machine learning is performed based on the training set.

8. The method according to claim 6 or 7, Its features are, Define a fractional function that defines the anomaly of the dataset relative to the previous dataset.

9. The method according to any one of claims 1 to 8, Its features are, Create a map with road surface temperature information, using data from the vehicle fleet.

10. The method according to any one of claims 1 to 9, Its features are, The autonomous driving function is enabled or disabled based on the determined current and future road surface temperatures.

Images