A method for predicting predictable road surface temperatures

Abstract

The present invention relates to a method for predicting the expected road surface temperature for at least one road section, in which the surface temperature of the road surface in the road section is measured via at least one sensor, ambient environmental information is associated with the measurement and transmitted to the backend along with the surface temperature, and the relationship between the ambient environmental information and the surface temperature is learned by machine learning, thereby enabling the estimation of the expected road surface temperature. According to the present invention, a range including the maximum and minimum possible road surface temperatures that can occur depending on the location is generated as an output value.

Description

Technical Field

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

Background Art

[0002] From Patent Document 1, a method for generating a digital map is known, in which an estimated value depending on the position of a physical measurement quantity is recorded in the digital map. Measurement data related to road conditions can be collected via a vehicle fleet, and data related to, for example, temperature can also be collected. These data can be transmitted to a server using mobile radio and can be processed at the server. Since a digital map including the collected and associated data can be transmitted to a vehicle, a road condition depending on a location on the map can be provided to a driving support system. Depending on this, the driving support system can turn on or off functions and thus can react to ambient environmental conditions.

[0003] Patent Document 2 discloses a method for predicting an expected road surface temperature for at least one road section as described at the beginning, in which the surface temperature of the road surface in the road section is measured via at least one sensor. This measurement is associated with ambient environmental information and transmitted to the backend together with the surface temperature. Since the relationship between the ambient environmental information and the surface temperature is learned by machine learning, the expected road surface temperature can be estimated as a fixed value respectively.

[0004] Patent Document 3 discloses enabling an autonomous driving function depending on temperature conditions.

[0005] A drawback is that the system is only supplied with data that has already been measured. Therefore, it cannot predict future conditions. However, future conditions are advantageous for economic, sustainable, and forward-thinking processing, particularly regarding the adaptation of functions provided by driver assistance systems. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] German Patent Application Publication No. 102015000394 Specification [Patent Document 2] German Patent Application Publication No. 102019135751 Specification [Patent Document 3] German Patent Application Publication No. 102020004018 [Overview of the project] [Problems that the invention aims to solve]

[0007] The objective of this invention is to provide a method for predicting road surface temperature that can also detect future data that cannot be directly measured on-site. [Means for solving the problem]

[0008] According to the present invention, this problem is solved by a method for predicting an expected road surface temperature, having the features described in claim 1, in particular the feature portion of claim 1. Advantageous configurations and variations will become apparent from the dependent claims of claim 1.

[0009] At the core of the method according to the present invention, the surface temperature of the road surface in a road section is measured via at least one sensor, ambient environmental information is associated with the measurement and transmitted to the backend along with the surface temperature, and the relationship between the ambient environmental information and the surface temperature is learned by machine learning, thereby enabling the estimation of an expected road surface temperature. The output value provides a range including the maximum and minimum possible values ​​of the road surface temperature that can occur depending on the location. Thus, the uncertainty in predicting the value can be quantified and estimation can be performed more effectively. In contrast, in known systems, only the surface temperature value is used as a so-called point estimate. The method provided by the present invention allows for advantageous verification of how well the estimated value approximates the actual value.

[0010] This method for predicting foreseeable road surface temperatures for at least one road section makes it possible, for example, to create a current road surface temperature map and to predict how the data in the road surface temperature map will behave over a certain period of time. This type of map can be advantageous for road management, for example, for winter management by road managers to ensure the safety of infrastructure. In particular, it improves the functionality of vehicle assistance systems. For example, having accurate information on surface conditions and future values ​​can avoid premature shutdown of the system, thereby significantly improving the availability of Level 3 to 5 driving functions. Therefore, this method can advantageously alleviate the current limitations on the use of autonomous driving, thereby allowing autonomous driving functions to be used in a much wider range of situations and enabling the early identification of important safety information. An advantage is that by predicting road surface temperatures over time, road conditions can be inferred, allowing infrastructure managers to maintain roads better, more quickly, and more cheaply. Therefore, the high data density can fill gaps in existing knowledge. This offers a particular advantage over conventionally known fixed-point measurements.

[0011] By using machine learning, it becomes possible to identify different states for specific local conditions, such as moisture or freezing, their thresholds, and critical road surface temperatures. This allows for more accurate predictions about road conditions. Consequently, both external and internal vehicle applications can be adapted to match driving behavior to road conditions.

[0012] The actual identification of road surface temperature based on available data can first be performed in the backend. Therefore, this identification is expected to depend on various requirements such as transmission capacity and online computing power. Based on the collected data, a machine learning model can be trained, which can output a prediction range, for example, by using conformal prediction, thereby predicting the uncertainty of the machine learning model. Therefore, the data is preferably output as range data.

[0013] Initial training of a machine learning model can be performed in a development vehicle equipped with appropriate sensors capable of measuring road surface temperature. The backend can, in this case specifically, create an ambient environment map by making predictions using information or data from multiple different vehicles.

[0014] Preferably, a conformal prediction algorithm can be used to detect predictable road surface temperatures for future points in time, thereby detecting uncertainties in the machine learning data. Therefore, current road surface temperatures and the temporal progression of road surface temperatures at specific geographical locations can be experimentally estimated, and with high data density, these estimates can be made very accurately. In particular, the output of the prediction range is utilized by imposing limits on the machine learning. In this case, for example, a split conformal prediction, a full conformal prediction, or an intermediate method between these two approaches, such as cross-conformal prediction or jackknife, can be used as the conformal prediction approach.

[0015] According to a highly advantageous development of this idea, ambient information may include geographical location and time. This allows data to be associated with a specific location at a particular time. Ambient information may originate, for example, from the measurement sensor system on the vehicle itself. Furthermore, ambient information from meteorological observatories or measurement data from road administrators can be used. Ambient information may include conditions such as solar radiation intensity, precipitation, snow depth, and especially temperature. To achieve this, data can be transmitted to a backend server using sensors and / or cameras, as well as wireless connections, such as mobile radio. All of this data, including data from the German Meteorological Service, can be collected anonymously in the backend.

[0016] In this advantageous configuration, additional data can be measured using the vehicle's onboard sensor system, and this additional data is transmitted to the backend along with ambient environmental information and road surface temperature. This provides improved geographical coverage of meteorological data. In particular, because the sensor system is placed only at a very short distance to the road surface, road surface data can be identified with great accuracy.

[0017] According to a highly advantageous development of this idea, road administrator data can also be used and transmitted to the backend. Similarly, publicly available weather data from fixed-point observations can also be used. All of this data improves data density, allowing the system to operate more accurately.

[0018] In another advantageous configuration, conformal prediction algorithms can aggregate a large amount of data and divide it into training, test, and calibration sets with the same distribution.

[0019] A highly advantageous development of this idea would lead to machine learning being performed based on a training set.

[0020] In a favorable configuration, a score function can be defined that defines how anomalous a given dataset is compared to previous datasets. A calibration score can be calculated using a predefined set of calibrations. In this case, a quantile function defined for the calibration score can be calculated. This quantile function can be used to calculate the prediction range, particularly for new examples, thus satisfying marginal coverage. In other words, this means calculating a probability, which can be expressed as 1-α (where α is in the range [0,1]) that the prediction range contains the exact labels.

[0021] This type of approach offers the advantage of only requiring the splitting of the dataset into training and test sets. For example, the AI ​​model is trained on the training dataset. Calibration datasets are optional, and selecting these datasets is difficult. Furthermore, learning the scoring function can be omitted because this scoring function can be trained or selected.

[0022] In another advantageous configuration, a map including information on road surface temperature is created using data from a group of vehicles. Further, the created map can be provided to each vehicle within the group of vehicles, and these vehicles need to include a subsystem, especially for transmitting or receiving data that contributes to the determination of the map.

[0023] According to a very advantageous development form of the idea, each time the current road surface temperature and the future road surface temperature are detected, it is possible to enable or disable the autonomous driving function. That is, the autonomous driving function can be utilized over a wider range, and the frequency of stopping the autonomous driving function can be reduced. This especially includes the availability of driving functions from level 3 to 5, and this availability is significantly improved. Therefore, in order to expand autonomous driving, that information can be utilized. Furthermore, infrastructure managers can utilize that information to better maintain the road network they manage.

[0024] A further advantageous configuration of the method according to the invention for predicting the expected road surface temperature will also become apparent from the embodiments described in more detail below with reference to the drawings.

Brief Description of the Drawings

[0025] [Figure 1] Shows a schematic diagram of the sequence of the method.

Mode for Carrying Out the Invention

[0026] Figure 1 schematically shows one possible sequence of this method. In the step denoted by reference numeral 1, onboard data from multiple vehicles can be measured and stored, and this onboard data is measured, for example, through the respective sensor systems on the vehicles. This data can be combined with offboard data detected outside the vehicle in the step denoted by reference numeral 2 via a communication module. This so-called offboard data is thought to include ambient environmental conditions and geographical location data that are not measured by the vehicle, as represented in step 3. Here, weather service data can be supplied to the aforementioned data by the step denoted by reference numeral 4. Returning to step 2, the road surface temperature can be determined, and thus the data can be determined via artificial intelligence.

[0027] For example, by aggregating ambient environment variables at a given time t in the backend, an AI model can be used to predict the characteristics of the road surface at a future time t+n (where n is a natural number). Here, for example, the following predictions can be made: When the range is wide, the AI ​​model becomes uncertain, and in this case, the model's decision becomes inappropriate. To improve this, it is desirable to collect more data or homogenize the data. When the range is narrow, the reliability of the AI ​​model becomes very high, and in this case, it can make a very good approximation to reality.

[0028] In other words, in the AI ​​model, the relationship between ambient environment variables and road surface temperature is established, and the uncertainty of the model can be quantified using conformal prediction. Therefore, in summary, a method can be provided to find, provide, and predict geographically localized road surface temperatures.

Claims

1. A method for predicting the road surface temperature for at least one road section, In a method in which the surface temperature of the road surface in the road section is measured via at least one sensor, ambient environmental information is associated with the measurement and transmitted to the backend along with the surface temperature, and the relationship between the ambient environmental information and the surface temperature is learned by machine learning, thereby enabling the estimation of a predictable road surface temperature, A method characterized in that an output value is generated which includes the maximum and minimum possible values ​​of the road surface temperature depending on the location.

2. The method according to claim 1, characterized in that a conformal prediction algorithm is used to detect the predictable road surface temperature for a future point in time, and the uncertainty of the machine learning data is detected.

3. The method according to claim 1 or 2, characterized in that the surrounding environment information is geographical location and time.

4. The method according to claim 1, 2, or 3, characterized in that additional data is measured using the vehicle's onboard sensor system, and the additional data is transmitted to the backend together with the ambient environment information and the road surface temperature.

5. The method according to any one of claims 1 to 4, characterized in that road administrator data is used and similarly transmitted to the backend.

6. The conformal prediction algorithm is characterized by aggregating a large amount of data and dividing it into a training set, a test set and a calibration set according to the same distribution, as described in any one of claims 2 to 5.

7. The method according to claim 6, characterized in that the machine learning is performed based on the training set.

8. The method according to claim 6 or 7, characterized in that a score function is defined, the score function defines how anomalous a given dataset is compared to a preceding dataset.

9. The method according to any one of claims 1 to 8, characterized in that a map including information on the road surface temperature is created using data from a group of vehicles.

10. The method according to any one of claims 1 to 9, characterized in that the autonomous driving function is enabled or disabled each time the current road surface temperature and the future road surface temperature are determined.

Images