METHOD, COMPUTER PROGRAM AND DEVICE FOR PROCESSING SIGNALS

Description

[0001] The present invention relates to a method, a computer program with instructions, and a device for processing signals in a continuous data provision process. The invention further relates to a means of transportation in which a method or device according to the invention is used.

[0002] Modern vehicles, such as automobiles, typically contain a multitude of sensors that provide signals relating to various vehicle components. In addition to sensor signals, modeled variables are also exchanged within the vehicle; these are not measured but calculated using an internal model. Other signals include control inputs that specify the direction of movement for actuators installed in the vehicle. These signals can be used, among other things, to make data-driven aging predictions. Such signals can also be transmitted as telematics data to an external server for further processing. This transmission generally occurs via mobile networks.

[0003] When transmitting data, it's important to consider that vehicles travel through cities and towns with varying levels of available mobile network coverage, such as Wi-Fi, 2G, 3G, 4G, or, in the future, 5G. Different levels of utilization across individual cell towers also lead to bandwidth variations both temporally and spatially. High bandwidth is required for transmitting vehicle signals, primarily CAN message data, as CAN signals typically have a temporal resolution of 10 ms. To ensure data transmission even in areas with poor mobile network coverage, it may be necessary to transmit only selected or reduced amounts of data. Therefore, lossless or lossy data compression algorithms are regularly employed.

[0004] Against this background, DE 10 2016 100 302 A1 describes a method for providing telematics data from vehicles. In this method, a parameter definition of a processed parameter to be calculated by an electronic control unit is received from a remote server. According to the parameter definition, the processed parameter is generated based on an unprocessed parameter produced by the electronic control unit. The processed parameter is then sent to a vehicle data buffer for upload to the remote server. Before the upload, the data is processed by an algorithm and undergoes lossy data compression.

[0005] EP 2 573 727 A1 describes a telematics on-board unit for a vehicle. The telematics on-board unit includes means for collecting vehicle usage data, means for transmitting collected vehicle usage data or derived, analyzed vehicle usage data to a telematics service platform, and means for identifying a driver using the vehicle and providing driver identification. Data compression takes place before the data is transmitted to the telematics service platform.

[0006] However, simply compressing the data often cannot achieve the necessary reduction in data volume. Furthermore, it would be desirable to be able to adjust the compression level to the available bandwidth.

[0007] Against this background, DE 10 2019 219 922 A1 describes a method for transmitting a plurality of signals. In this method, the plurality of signals are recorded within a time window. Signals exhibiting a similar waveform within that time window are grouped together. One signal from each group is selected as its representative. Subsequently, transmission data is sent, containing the representative for each group as well as transformation information for each signal within that group.

[0008] The article by C. Guyeux et al.: "Introducing and Comparing Recent Clustering Methods for Massive Data Management in the Internet of Things", Journal of Sensor and Actuator Networks, Vol. 8 (2019), provides an overview of common and advanced clustering methods, compares them, and performs a detailed analysis of their performance depending on the amount of data, the type of data collected or its heterogeneity, and the noise level.

[0009] It is an object of the invention to provide solutions for processing signals in a process of continuous data provision that allow for easy adjustment of a compression level of a data compression.

[0010] This problem is solved by a method having the features of claim 1, by a computer program with instructions according to claim 7, and by a device having the features of claim 8. Preferred embodiments of the invention are the subject of the dependent claims.

[0011] According to a first aspect of the invention, a method for processing signals in a continuous data provision process comprises the following steps: Sequencing the signals into segments; determining at least one statistical characteristic for each segment; clustering the signals based on the determined statistical characteristics using a clustering algorithm; determining representatives for the clusters; providing the representatives for transmission; and transmitting the provided representatives. where the number of clusters is automatically adjusted to a changing available bandwidth by forming many clusters and thus transmitting many representatives in the case of high bandwidth, forming a medium number of clusters and thus transmitting a smaller number of representatives in the case of medium bandwidth, and forming few clusters and thus transmitting few representatives in the case of low bandwidth.

[0012] According to another aspect of the invention, a computer program contains instructions which, when executed by a computer, cause the computer to perform the following steps for processing signals in a process of continuous data provision: Sequencing the signals into segments; determining at least one statistical characteristic for each segment; clustering the signals based on the determined statistical characteristics using a clustering algorithm; determining representatives for the clusters; providing the representatives for transmission; and transmitting the provided representatives. where the number of clusters is automatically adjusted to a changing available bandwidth by forming many clusters and thus transmitting many representatives in the case of high bandwidth, forming a medium number of clusters and thus transmitting a smaller number of representatives in the case of medium bandwidth, and in In the case of low bandwidth, few clusters are formed and therefore few representatives are transmitted.

[0013] The term "computer" is to be understood broadly. In particular, it also includes microcontrollers, embedded systems, and other processor-based data processing devices.

[0014] The computer program can, for example, be made available for electronic retrieval or be stored on a computer-readable storage medium.

[0015] According to a further aspect of the invention, a device for processing signals in a continuous data provision process comprises a memory containing instructions and a processor, wherein the processor is configured to perform the following steps for processing signals in a continuous data provision process when the instructions are executed: Sequencing the signals into segments; determining at least one statistical characteristic for each segment; clustering the signals based on the determined statistical characteristics using a clustering algorithm; determining representatives for the clusters; providing the representatives for transmission; and transmitting the provided representatives. where the number of clusters is automatically adjusted to a changing available bandwidth by forming many clusters and thus transmitting many representatives in the case of high bandwidth, forming a medium number of clusters and thus transmitting a smaller number of representatives in the case of medium bandwidth, and forming few clusters and thus transmitting few representatives in the case of low bandwidth.

[0016] In the solution according to the invention, the vehicle data to be transmitted is reduced by adaptively adjusting the quality of an evolutionary signal clustering method to the bandwidth, such that lossy data compression occurs depending on the available bandwidth. Using signal clustering, vehicle signals are grouped into clusters. Only one representative of each cluster is used for data transmission, thereby achieving significant data reduction in a highly correlated signal space. The clustering algorithm is configured so that the number of resulting clusters is automatically adjusted to the available bandwidth. With a high bandwidth, many clusters can be generated, resulting in a correspondingly large number of representatives. In this case, a large amount of data is transmitted.When bandwidth is limited, the algorithm is configured to create fewer clusters. This results in fewer representatives and less data being transmitted.

[0017] If no network is available for data transmission, the data can be stored in an available data buffer. This data buffer is designed to bridge at least short periods of time spent in an area without network coverage. Ideally, the data buffer is sized so that no data is lost during a period of, for example, two hours without a connection to the wireless network.

[0018] If the data buffer is full and transmission still cannot occur, the signals can be clustered again to create fewer clusters and therefore less data in the buffer. In this case, a greater loss of information is accepted.

[0019] According to the invention, the signals are sequenced into sections, and at least one statistical feature is determined for each section. The signals are then clustered based on these determined statistical features.

[0020] In many cases, the data consists of very high-resolution measurements, for example, in the automotive sector with CAN bus data. Simply clustering the time series of these signals does not yield usable results. There are several reasons for this. Firstly, the signals have different resolutions, which means that a direct comparison is only possible with a very high time and computational effort, even if the signals are very similar, such as the front right and front left wheel speeds. Furthermore, the signals are so dynamic that the algorithm does not assign them to a common cluster in the high-resolution representation, even though they clearly correspond to the same clusters to a human eye. Finally, clustering the original time series is so memory-intensive that it is only possible in sequences, for example, in segments with a duration of ten minutes each.However, attempts with such sections yielded poor results. According to the invention, the database can also be divided into small sequences. These sequences can, for example, have a duration of ten minutes or even hours. Statistical characteristics are then calculated for these sequences; that is, statistical, artificial parameters are aggregated from the time intervals. These characteristics serve as input data for a clustering algorithm. The result is clustered signals. These clusters can be used as the basis for further processing steps. Preferably, a refined database is used in which the input data are equidistant and of the same length. Since only simple mathematical operations are required, the clustering algorithm can be implemented on the signal acquisition side of the vehicle. This enables data-efficient storage.

[0021] According to one aspect of the invention, hyperparameters of the clustering algorithm are set to adapt the number of clusters to the available bandwidth. Hyperparameters influence the clustering result; that is, different hyperparameter settings result in different cluster strengths and cluster counts. Which hyperparameters are available and what effect they have on the number of clusters depends on the chosen clustering algorithm and the specific signals to be clustered. The hyperparameter settings for different available bandwidths can, for example, be determined experimentally beforehand.

[0022] According to one aspect of the invention, a feature space of specific statistical features is transformed into a lower-dimensional space before clustering. Preferably, this transformation is into a one-dimensional representation. The transformation into a lower-dimensional space results in high-quality data compression for signal description. The resulting reduced data set is particularly advantageous for the correct identification of identical signals within the existing signal space, as it facilitates machine processing of the data and supports error-free signal assignment.

[0023] According to one aspect of the invention, principal component analysis is applied to specific statistical features, or at least one specific statistical feature is selected, to transform the feature space. Principal component analysis, also known as principal axis transformation, is ideally suited for structuring large datasets by approximating the existing statistical variables with a smaller number of the most informative principal components possible. Alternatively, it is possible to use only one specific statistical feature or a reduced selection of statistical features, e.g., the mean of certain time periods. Suitable results can also be achieved with this approach. Which statistical features are best suited for a specific application can be determined empirically. Preferably, the selection of statistical features is adaptable during operation.

[0024] According to one aspect of the invention, at least one statistical feature is a mean, a maximum, a minimum, or a quantile. The quantile can, in particular, be a quartile, i.e., the quantiles Q0.25, Q0.5, and Q0.75, also referred to as the lower quartile, middle quartile, and upper quartile, respectively. All these statistical features are well suited for subsequent clustering. Naturally, a selection or subset of statistical features can also be determined.

[0025] According to one aspect of the invention, a density-based clustering method, a partitioning clustering method, or a hierarchical clustering method is used for clustering the signals. For example, a DBSCAN algorithm can be used as a density-based clustering method. A K-means algorithm is a suitable partitioning clustering method. Examples of suitable hierarchical clustering methods are agglomerative clustering or a mean-shift algorithm. The use of hierarchical clustering methods has the advantage that no prior knowledge of the number of clusters is required. Furthermore, the shape of the clusters is not restricted. Preferably, silhouette coefficients are used to assess the quality of the clustering.

[0026] A method or device according to the invention is particularly advantageous when used in a (semi-)autonomous or manually controlled means of transport. The means of transport can be, in particular, a motor vehicle, but also a ship, an aircraft (e.g., a Volocopter), a construction machine, etc. Application in mobile production machines is also possible. The data to be transmitted can be used, for example, for telematics services. In the future, this data can also be used for predictive services, such as predictive maintenance. For this purpose, it is advantageous to have data from the entire vehicle lifespan available. It is more important to have data covering the entire vehicle lifespan for evaluation than to have data that is exceptionally high-resolution in terms of both time and discretization but incomplete.

[0027] Further features of the present invention will become apparent from the following description and the attached claims in conjunction with the figures. Fig. 1 schematically shows a method for processing signals in a continuous data provision process; Fig. 2 shows a first embodiment of a device for processing signals in a continuous data provision process; Fig. 3 shows a second embodiment of a device for processing signals in a continuous data provision process; Fig. 4 schematically represents a means of transport in which a solution according to the invention is implemented; Fig. 5 schematically shows a series of signals to be preprocessed; Fig. 6 schematically shows the signals from Fig. 5 After completion of preprocessing; Fig. 7 schematically shows a subdivision of the preprocessed signals into sections; Fig. 8 illustrates an extraction of feature vectors from the sections; Fig. 9 illustrates a transformation of the feature vectors into a statistical feature space; Fig. 10 illustrates a transformation of the feature space of the statistical features into a one-dimensional representation; Fig. 11 illustrates clusters generated on the basis of the one-dimensional representation of the statistical features; Fig. 12 illustrates an example of clustering with many clusters, a medium number of clusters, and few clusters; and Fig. 13 shows an associated silhouette index for different numbers of clusters resulting from the clustering.

[0028] To better understand the principles of the present invention, embodiments of the invention are explained in more detail below with reference to the figures. It is understood that the invention is not limited to these embodiments and that the described features can also be combined or modified without leaving the scope of protection of the invention as defined in the appended claims.

[0029] Fig. 1 Figure 1 schematically shows a method for processing signals in a continuous data provision process, e.g., sensor signals, modeled quantities, or manipulated variables. In this example, the signals are first preprocessed. During this preprocessing, the signals are sequenced into sections.10 For each section, at least one statistical feature is then determined, e.g., a mean, a maximum value, a minimum value, or a quantile. A feature space of the determined statistical features can then optionally be transformed into a lower-dimensional space.12 For this purpose, for example, principal component analysis can be applied to the statistical features, or at least a specific statistical feature can be selected. The signals are then clustered using a clustering algorithm based on the determined statistical features.13To adapt the number of clusters to the available bandwidth, hyperparameters of the clustering algorithm can be set. The clustering algorithm can, for example, implement a density-based clustering method, a partitioning clustering method, or a hierarchical clustering method. Representatives are then determined for the clusters resulting from clustering 13. At least the signals determined as representatives are finally made available for transmission 15.When clustering 13 of the signals, the number of clusters is automatically adjusted to a changing available bandwidth by forming many clusters and thus transmitting many representatives in the case of high bandwidth, forming a medium number of clusters and thus transmitting a smaller number of representatives in the case of medium bandwidth, and forming few clusters and thus transmitting few representatives in the case of low bandwidth.

[0030] Fig. 2 Figure 1 shows a simplified schematic representation of a first embodiment of a device 20 for processing signals in a continuous data provision process, e.g., of sensor signals, modeled quantities, or manipulated variables. The device 20 has an input 21 through which signals Si from various sensors 41i can be received, two of which are shown as examples. A sequencing module 22 and an analysis module 23 are provided for preprocessing the signals Si. The sequencing module 22 is configured to sequence the signals Si into segments. The analysis module 23 determines at least one statistical feature for each segment, e.g., a mean, a maximum value, a minimum value, or a quantile. The analysis module 23 can optionally be configured to transform a feature space of the statistical features into a lower-dimensional space after the features have been determined, e.g.,by applying principal component analysis to the statistical features or by selecting at least one specific statistical feature. A cluster module 24 then clusters the signals S i using a clustering algorithm based on the selected statistical features. Hyperparameters of the clustering algorithm can be set, for example, to adapt the number of clusters to the available bandwidth. The clustering algorithm can implement, for example, a density-based clustering method, a partitioning clustering method, or a hierarchical clustering method. The cluster module 24 is also configured to determine representatives R i for the clusters resulting from the clustering process. At least the signals S i determined as representatives R i are finally made available for transmission via an output 27 of the device 20.The cluster module 24 is configured to automatically adjust the number of clusters to a changing available bandwidth when clustering the signals S i, by forming many clusters and thus transmitting many representatives R i in the case of high bandwidth, forming a medium number of clusters and thus transmitting a smaller number of representatives R i in the case of medium bandwidth, and forming few clusters and thus transmitting few representatives R i in the case of low bandwidth.

[0031] The sequencing module 22, the analysis module 23, and the cluster module 24 can be controlled by a control module 25. Settings of the sequencing module 22, the analysis module 23, the cluster module 24, or the control module 25 can be changed via a user interface 27. The data generated in the device 20 can be stored in a memory 26 of the device 20 as needed, for example, for later evaluation or for use by the components of the device 20. The sequencing module 22, the analysis module 23, the cluster module 24, and the control module 25 can be implemented as dedicated hardware, for example, as integrated circuits. Of course, they can also be partially or completely combined or implemented as software running on a suitable processor, such as a GPU or a CPU.Input 21 and output 27 can be implemented as separate interfaces or as a combined bidirectional interface.

[0032] Fig. 3 Figure 3 shows a simplified schematic representation of a second embodiment of a device 30 for processing signals in a continuous data provision process. The device 30 comprises a processor 32 and a memory 31. For example, the device 30 is a microcontroller or an embedded system. Instructions are stored in the memory 31 which, when executed by the processor 32, cause the device 30 to perform the steps according to one of the described methods. The instructions stored in the memory 31 thus embody a program executable by the processor 32, which implements the method according to the invention. The device 30 has an input 33 for receiving information, in particular signals. Data generated by the processor 32 is provided via an output 34. Furthermore, it can be stored in the memory 31.Input 33 and output 34 can be combined into a bidirectional interface.

[0033] The processor 32 can comprise one or more processor units, such as microprocessors, digital signal processors, or combinations thereof.

[0034] The memory elements 26, 31 of the described embodiments can have both volatile and non-volatile memory areas and can include a wide variety of storage devices and storage media, for example hard disks, optical storage media or semiconductor memory.

[0035] Further details of the invention will be presented below with reference to Fig. 4 bis Fig. 11 This will be described. In this embodiment, signals from a means of transportation are considered. Of course, the solution according to the invention is not limited to this application. It can also be used, among other things, in mobile production machines.

[0036] Fig. 4 Figure 40 schematically represents a means of transport 40 in which a solution according to the invention is implemented. In this example, the means of transport 40 is a motor vehicle. The motor vehicle has a plurality of sensors 41i, some of which are shown by way of example, and from which sensor signals can be provided with respect to a number of components of the motor vehicle. Furthermore, the motor vehicle has a device 20 according to the invention for processing the signals. Other components of the motor vehicle are a navigation system 42, a data transmission unit 43, and a number of assistance systems 44, one of which is shown by way of example. By means of the data transmission unit 43, a connection to service providers can be established, for example, for further processing of the signals. A memory 45 is provided for storing data.Data exchange between the various components of the motor vehicle takes place via a network 46, e.g. via a CAN bus.

[0037] Fig. 5 Figure 1 schematically shows a series of signals Si that are to undergo preprocessing. There are n signals Si, of which three signals S1, S2, Sn are shown as examples. The signals can be, for example, sensor signals, modeled quantities, or control variables. The signals Si can be transmitted, for example, on the CAN bus of a motor vehicle. The signals Si sometimes contain gaps or time periods Ti in which no usable data is present. These time periods Ti are preferably removed from all signals Si during preprocessing; that is, the corresponding time periods Ti are extracted from the signals Si. The signals Si after preprocessing are in Fig. 6 shown.

[0038] Fig. 7 The diagram schematically shows a subdivision of the preprocessed signals Si into sections Ai_n. In the example shown, the signals Si are subdivided into m sections Ai_n, each with the same length L. A time series interpretation is then performed based on these sections Ai_n, in which a feature vector is extracted for each signal Si and each of the sections Ai_n.

[0039] Fig. 8 This illustrates the extraction of feature vectors from the segments Ai_n. After extraction, m arrays of features are available. The dimensions of the m arrays are determined by the number n of signals and the length L of the individual signal segments Ai_n. Statistical features are then determined based on the individual feature vectors. The length L of the individual signal segments Ai_n can, for example, be determined empirically. Evaluations have shown that for determining aging that occurs over a usage period of several hundred hours, aggregation in the range of one hour yields good results.

[0040] Fig. 9 This illustrates a transformation of the feature vectors into a statistical feature space. After determining the statistical features, m arrays of statistical features are available. The dimensions of the m arrays are determined, firstly, by the number n of signals, and secondly, by the number A of statistical features determined for each feature vector. Assuming that high-resolution time series in the vehicle are recorded at a frequency of 10 Hz, and that these time series are now summarized into one hour using a statistical feature, the amount of data is reduced from 1 × 60 × 60 × 10 = 36,000 measurements to a single value.

[0041] Fig. 10 This illustrates a transformation of the feature space of statistical features into a one-dimensional representation. For this purpose, the statistical features are subjected to principal component analysis. In this example, only a single principal component (PC) is retained. After the principal component analysis, a single array of principal components (PC) is obtained. The dimensions of the array are determined, on the one hand, by the number n of signals and, on the other hand, by the number m of segments. This array serves as the basis for a clustering algorithm.

[0042] Fig. 11 The diagram illustrates clusters Ci generated based on the one-dimensional representation of the statistical characteristics. In the example shown, three clusters C1, C2, and C3 are visible. Each cluster Ci comprises a plurality of signals Si. Additionally, there exists a signal Sn that is not assigned to any cluster Ci. From each cluster Ci, one signal Si can be selected as a representative Ri. This could be, for example, the first participant found in the respective cluster Ci, or the participant within the cluster Ci who is closest to the center of the cluster Ci. The representatives Ri and the signal Sn, which is not assigned to any cluster Ci, ultimately constitute the resulting set of signals, indicated by the dashed ellipses.

[0043] The first cluster C 1 can, for example, include the following signals S i: S 1 : Speed ​​of front left wheel S 2 : Speed ​​of front right wheel S 24 : Speed ​​of rear left wheel S 15 : Speed ​​of rear right wheel S 5 : Speed ​​of wheel S 28 : Speed ​​of vehicle

[0044] The signal S 28, i.e. the speed of the vehicle, serves as representative R 1 of the first cluster C 1.

[0045] The second cluster C 2 can, for example, include the following signals S i: S 7 : Calculated gear S 8 : Gear S 76 : Target gear S 19 : Gear 2

[0046] The signal S 8, i.e. the walk, serves as representative R 2 of the second cluster C 2.

[0047] The third cluster C 3 can, for example, include the following signals S i: S 3 : Time 1 S 33 : Time 2 S 21 : Time 3 S 14 : Time 4 S 120 : Time 5 S 6 : Time 6 S 41 : Time 7

[0048] The signal S 3, i.e. a first time signal, serves as representative R 3 of the third cluster C 3.

[0049] Other clusters can result, for example, from signals indicating pedal position and engine power, or from signals indicating oil temperature and coolant temperature.

[0050] Consider a vehicle following a route with varying available bandwidths and continuously transmitting data. The route includes sections with high bandwidth, e.g., due to 5G availability in urban areas; sections with medium bandwidth, e.g., 4G in outlying areas; and sections with low bandwidth, e.g., 2G in smaller villages or on rural roads. Based on the available bandwidth, the clustering algorithm is parameterized such that in high-bandwidth sections, many clusters are formed and thus many representatives are transmitted; in medium-bandwidth sections, a medium number of clusters are formed and thus a smaller number of representatives are transmitted; and in low-bandwidth sections, few clusters are formed and thus few representatives are transmitted.The amount of data to be transferred can therefore be adapted to the available bandwidth solely through clustering. Fig. 12 illustrates an example of clustering with many clusters ( Fig. 12a )), a medium number of clusters ( Fig. 12b )) and a few clusters ( Fig. 12c )).

[0051] The following example illustrates how to configure the hyperparameter settings for the various available bandwidths. The clustering algorithm has configurable settings, called hyperparameters, which influence the result. Different hyperparameter settings result in varying numbers of clusters and cluster quality. Cluster quality is described by the silhouette index. Fig. 13 This shows a corresponding silhouette index for different numbers of clusters resulting from the clustering. Each star represents a configuration of the clustering algorithm and thus corresponding hyperparameter settings. The x-axis plots the number of clusters resulting from the setting parameters, and the y-axis shows the corresponding quality or performance of the clustering algorithm when the corresponding setting parameters are defined. The data in this example comes from a vehicle that provides approximately 400 different signals. The best result is achieved with approximately 190 clusters. The corresponding silhouette index is approximately 0.5. Now, for example, three categories can be defined and assigned to the different bandwidths. The categories are in Fig. 13 The dashed horizontal lines indicate the different categories. The first category includes configurations with a silhouette index between approximately 0.4 and the maximum. The second category includes configurations with a silhouette index between approximately 0.3 and 0.4. The third category includes configurations with a silhouette index of less than approximately 0.3. Within each category, the best available configuration is selected during clustering. For high bandwidth, this results in approximately 190 clusters and a silhouette index of approximately 0.5. For medium bandwidth, the best configuration provides approximately 102 clusters and a silhouette index of approximately 0.4. Compared to the 190 clusters, this represents a reduction in data transmission of approximately 46%. For low bandwidth, the best configuration provides approximately 54 clusters and a silhouette index of approximately 0.3. Compared to the 190 clusters, this represents a reduction in data transmission of approximately 70%.The respective configurations are marked by the arrows shown. Bezugszeichenliste

[0052] 10. Sequencing the signals 11. Determining statistical characteristics 12. Transforming a feature space 13. Clustering the signals 14. Determining signals as representatives for the clusters 15. Providing the representatives for transmission 20 Device 21 Input 22 Sequencing module 23 Analysis module 24 Cluster module 25 Control module 26 Memory 27 Output 28 User interface 30 Device 31 Memory 32 Processor 33 Input 34 Output 40 Means of transport 41 Sensor 42 Navigation system 43 Data transmission unit 44 Assistance system 45 Storage 46 Network A Number of defined statistical characteristics A i_n Section C i Cluster HK Principal component L Length of sections m Number of sections n Number of signals R i Representative S i Signal T i Time period

Claims

1. Method for processing signals (Si) in a process of continuous data provision, comprising the steps of: - sequencing (10) the signals (Si) into portions (Ai_n); - determining (11) at least one statistical feature for each of the portions (Ai_n); - clustering (13) the signals (Si) on the basis of the determined statistical features by means of a clustering algorithm; - determining (14) representatives (Ri) for the clusters (Ci); - providing (15) the representatives (Ri) for a transfer; and - transferring the provided representatives (Ri); wherein the number of clusters (Ci) is automatically adapted to a changing available bandwidth (B) in that many clusters (Ci) are formed (13), and thus many representatives (Ri) are transferred, in the case of a high bandwidth, a medium number of clusters (Ci) are formed (13), and thus a smaller number of representatives (Ri) are transferred, in the case of a medium bandwidth, and few clusters (Ci) are formed (13), and thus few representatives (Ri) are transferred, in the case of a low bandwidth.

2. Method according to claim 1, wherein, for the adaptation of the number of clusters (Ci) to the available bandwidth (B), hyperparameters of the clustering algorithm are set.

3. Method according to claim 1 or 2, wherein, before the clustering (13), a feature space of the determined statistical features is transformed (12) into a space of lower dimension.

4. Method according to claim 3, wherein, for the transformation (12) of the feature space, a principal component analysis is applied to the determined statistical features or at least one determined statistical feature is selected.

5. Method according to any of the preceding claims, wherein the at least one statistical feature is an average value, a maximum value, a minimum value or a quantile.

6. Method according to any of the preceding claims, wherein, for the clustering (13) of the signals (Si), a density-based clustering method, a partitioning clustering method, or a hierarchical clustering method is used.

7. Computer program comprising instructions which, upon execution by a computer, cause the computer to execute the steps of a method according to any of claims 1 to 6 for processing signals (Si) in a process of continuous data provision.

8. Device (30) for processing signals (Si) in a process of continuous data provision, comprising a memory (31) in which instructions are stored and a processor (32), wherein the processor (32) is configured, upon execution of the instructions, to execute the steps of a method according to any of claims 1 to 6 for processing signals (Si) in a process of continuous data provision9. Means of transport (40), characterized in that the means of transport (40) has a device (30) according to claim 8 for processing signals (Si) in a process of continuous data provision or is configured to execute a method according to any of claims 1 to 6 for processing signals (Si) in a process of continuous data provision.