Method for recognizing pattern in driving cycle

Abstract

The invention relates to a method for pattern recognition in a drive cycle. Here, the method comprises (a) selecting a number of samples from a plurality of sensor data sets, where each sensor data set represents a drive cycle. Thereafter, (b) unsupervised pattern recognition is performed within the number of samples to identify a pattern. Furthermore, (c) supervised pattern recognition is performed within the remaining sensor data sets of the plurality of sensor data sets to again identify a pattern; and (d) determining, for each of the plurality of sensor data sets, a degree of correspondence of the sensor data set to the pattern. In this case, steps (a) to (d) are repeatedly carried out until each correspondence degree exceeds a correspondence degree threshold value.

Description

Technical Field

[0001] The invention relates to a method for detecting patterns in driving cycles, in particular recurring patterns in sensor data sets. Background Art

[0002] Vehicle manufacturers are already able to, or at least will be able to, collect large amounts of data from customer vehicles in the near future. It is desirable to use this data to design and develop, for example, the next generation of vehicles in such a way that the vehicles are specifically tailored to customer requirements and can be produced at optimal cost. In this regard, it is necessary to be able to process large amounts of data on distributed and highly scalable computing systems (AWS, Azure, etc.).

[0003] A common method for managing large amounts of data is unsupervised pattern recognition in multivariate time series. With this method, for example, recurring traffic and driving situations in sensor data sets can be identified without necessarily having to know the actual state of these situations.

[0004] By this method, the sensor data can be broken down into recurring situations. Via the frequency of these situations (or patterns), it is possible to construct, in particular ex post, a driving cycle that is representative for all customers and is used as input for designing and testing the vehicle.

[0005] In principle, methods for carrying out pattern recognition are known.

[0006] From the scientific publication "Mobile Telematics Pattern Recognition Framework for Driving Behavior Extraction" by M. SIAMI, M. NADERPOUR and J. LU (published in IEEE Transactions on Intelligent Transportation Systems, Vol. 22, No. 3, pp. 1459-1472, March 2021, doi: 10.1109 / TITS.2020.2971214), a method for recognizing driving patterns is known.

[0007] US2022 / 0242451 A1 relates to identifying malicious events in autonomous vehicles. To this end, recorded events are compared with scenes of normal patterns to identify deviations. If the deviation is large enough, a malicious event is identified. Normal patterns can be implemented through supervised or unsupervised machine learning.

[0008] From HEINRICH, Felix's paper "Battery State Estimation of Electric Vehiclesusing Neural Networks" (Diss. Friedrich-Alexander- In Erlangen-Nürnberg (FAU), 2022), a search algorithm for unsupervised pattern recognition is known, which is based on the “Sequitur” algorithm or the “Pattern Enumeration” algorithm, for example.

[0009] US 2002 / 0194159 A1 relates to a system for parallel object-oriented data mining.

[0010] DE 10 2008 013 002 A1 describes a method and a device for observing and analyzing a plurality of real-time data sets of Internet cameras, in particular those publicly accessible via the Internet, in order to identify atypical situations and / or critical structures and report the real-time data sets thus identified to a supervisory authority. In this case, metadata in the video images are evaluated, which metadata include, for example, crowd density, crowd density distribution, and movement direction and speed of the crowd. In principle, atypical situations and / or critical structures are identified by comparing actual object group characteristics with permitted object group characteristics (for example, by querying atypical situations and / or critical structures from a database).

[0011] Disadvantages of the pattern recognition methods known to date may include the fact that cloud-based pattern recognition is only performed for known patterns and therefore unsupervised pattern recognition is not applicable. In addition, unsupervised pattern recognition methods to date may be limited to serial application (that is, only the data sets to be checked can be processed sequentially) or to local application. As a result, only smaller data sets can be checked. Finally, there are no cloud-based solutions for parallelized unsupervised pattern recognition. Summary of the invention

[0012] The object of the present invention is to provide a method which at least partially overcomes the above-mentioned disadvantages.

[0013] This object is achieved by the method according to the invention.

[0014] One aspect of the present invention relates to a method for performing pattern recognition in a driving cycle, the method comprising the following steps:

[0015] (a) selecting a number of samples from a plurality of sensor data sets, wherein each sensor data set represents a driving cycle;

[0016] (b) performing unsupervised pattern recognition within the sample size to identify patterns;

[0017] (c) performing supervised pattern recognition on remaining sensor data sets in the plurality of sensor data sets to again identify patterns; and

[0018] (d) determining, for each of the plurality of sensor data sets, a degree of correspondence between the sensor data set and the pattern;

[0019] Wherein, steps (a) to (d) are repeatedly performed until each correspondence degree exceeds a correspondence degree threshold.

[0020] The method can be implemented on a distributed and highly scalable (cloud-based) computing system (AWS, Azure, etc.).

[0021] For example, multiple sensor data sets can be stored in a so-called "Data Lake". A data lake is an information technology repository or system for storing data. The sensor data sets can exist as measurement data in a structured form, a semi-structured form or an unstructured form. The data lake can be stored on a server or in the cloud.

[0022] In this case, each sensor data set represents a driving cycle or a trip of the vehicle. The sensor data sets can originate from different vehicles.

[0023] A sensor data set comprises a stream of (possibly time-resolved) sensor data, or in other words, the values ​​of one or more sensors acquired and provided as a time series. In addition, a "sensor data set" should also include a stream of data determined by a (vehicle-side) controller based on the provided sensor data. If the time series comprises the value of a single sensor, there is a univariate time series. Conversely, if there are values ​​of multiple sensors in the time series, there is a multivariate time series. With the aid of corresponding sensors, different vehicle measurement data (e.g. speed, acceleration, temperature of the vehicle battery, voltage and / or current of the vehicle battery, state of charge of the vehicle battery, etc.) can be acquired. The sensor can be arranged on the vehicle side. Additionally or alternatively, sensors integrated in mobile terminal devices (e.g. smart phones) can also be used to acquire vehicle measurement data.

[0024] In step (a), a number of samples is selected from a plurality of sensor data sets, wherein the number of samples corresponds to the samples used for the unsupervised pattern recognition performed in step (b).

[0025] The first selection of the number of samples can be made randomly or based on a predetermined criterion. The predetermined criterion can indicate the relevance of the sensor data set to the unsupervised pattern recognition, such as the travel duration / length of the driving cycle represented by the sensor data set. In this way, the sample can include, for example, the sensor data set with the longest travel. In addition, other criteria describing the relevance of the sensor data set can be used for sample selection.

[0026] In step (b), unsupervised pattern recognition is performed within the sample number (samples) in order to identify recurring patterns. Here, a pattern comprises a time sequence of sensor signals / values ​​that recurs in the samples in the same or at least similar manner. Recurrence means that the sequence is contained at least twice in the samples.

[0027] The pattern indicates a driving situation, a traffic situation and / or an operating situation of the vehicle. A recurring pattern may be understood as a similar recurring sequence in the sensor data set.

[0028] Unsupervised pattern recognition is performed by corresponding known unsupervised pattern recognition methods, which automatically recognize patterns of this type. Such methods for unsupervised pattern recognition are based, for example, on the known Sequitur algorithm or “Pattern Enumeration” algorithm.

[0029] The algorithm for unsupervised pattern recognition is applied once and serially to the sensor data set of the sample. "Serially" means that the algorithm traverses the sample in time from the start time to the end time of the sample. The pattern recognition algorithm identifies recurring patterns in the sample and creates a (global) dictionary from the recurring patterns and their definitions.

[0030] Depending on the algorithm used for unsupervised pattern recognition, it may be necessary to process or adapt the sensor data set of the sample (sample data set), that is, to convert the sensor data set into a suitable form for pattern recognition. For example, for some algorithms, it is necessary for the sample data set to exist as a unique time series. That is, the time series of the sample data set are combined, especially connected in series, into a unique time series, so that the pattern recognition algorithm is applied to the connected time series.

[0031] In some cases, for a multivariate time series as a sample data set, it may be necessary to perform unification, that is, convert the multivariate time series into a univariate time series. In order to determine the univariate time series from the multivariate time series, for example, a unification algorithm can be used. Common unification algorithms are known to those skilled in the art.

[0032] In step (c), supervised pattern recognition is performed on the remaining sensor data sets of the plurality of sensor data sets to again identify patterns.

[0033] Therefore, unsupervised pattern recognition is performed on all sensor data sets in the plurality of sensor data sets in the data lake that are not part of the sample.

[0034] Supervised pattern recognition aims to recognize again the pattern found in step (b). For this purpose, a previously created dictionary is used. In principle, the sensor data set is compared with this dictionary.

[0035] Compared to unsupervised pattern recognition, the re-recognition of the (already found) pattern is significantly less complex in terms of computing time and memory and is therefore particularly well suited for parallelization. As a result, the re-recognition of the pattern can be carried out in parallel. That is, a corresponding worker for supervised pattern recognition can be provided for each of the other sensor data sets. Due to the lower complexity of supervised pattern recognition, the workers used for this may also be less efficient.

[0036] A worker is a program or part of a program that runs in its own single thread or process. Multiple workers can run in parallel on multiple cores or processors. Multiple workers can be distributed, for example, on multiple servers with corresponding processors.

[0037] Likewise, for supervised pattern recognition, already known methods or algorithms can be applied.

[0038] In step (d), the degree of correspondence of the sensor data set to the pattern is determined for each of the plurality of sensor data sets. The degree of correspondence indicates how much the sensor data set is covered by the pattern found in step (b). In other words, the degree of correspondence describes how high the proportion of the sensor data set can be described by the found pattern. Thus, the degree of correspondence is a quality criterion for unsupervised pattern recognition. For example, the degree of correspondence can be a percentage value. The degree of correspondence can also be understood as a degree of coverage.

[0039] Steps (a) to (d) are performed repeatedly (iteratively) until the correspondence degree of each sensor data set exceeds a correspondence degree threshold value, which may be at least 85% or at least 90%, for example. Thus, exceeding the correspondence degree threshold value is a termination condition.

[0040] Alternatively, other termination conditions may be set. For example, another termination condition may be traversing steps (a) to (d) according to a predetermined number of loop traversals. Another termination condition may include: if no new pattern is found when traversing steps (a) to (d), then the method ends.

[0041] The goal of the method according to the invention is that all trips (and thus sensor data sets) can be described or covered as completely as possible (or at the level of a corresponding degree threshold) with the identified patterns. If this goal is achieved, the method ends. However, since the existence of a pattern depends both on the driver and on the driving environment, not every pattern is present in every sensor data set. If the unsupervised pattern recognition in step (b) has not yet identified a certain pattern (or a certain situation), the supervised pattern recognition in step (c) may not recognize this pattern again.

[0042] Therefore, steps (a) to (d) are repeated as often as necessary until each of the plurality of sensor data sets can be adequately covered or described by the found pattern. Thus, a particularly efficient pattern recognition method can be provided.

[0043] In another embodiment, the number of samples (samples) can be changed when step (a) is repeatedly implemented. This means that the samples are changed when step (a) is traversed for the second, third, fourth, etc. By changing the samples, an attempt is made to recognize patterns in the current traversal that were not recognized in the previous traversal steps (a) to (d) so that the newly determined degree of correspondence of the sensor data set reaches the degree of correspondence threshold.

[0044] The change of the sample can be performed in different ways. A first way can include supplementing the sample with a further sensor data set that was not previously in the sample. The further sensor data set can have as a further condition that its degree of correspondence is less than a threshold degree of correspondence, and optionally has a significant travel length / distance. A second way can include selecting a completely new sample from a plurality of sensor data sets, wherein the sensor data set of the new sample is different from the sensor data set of the old sample.

[0045] By combining an iterative approach with varying samples, pattern recognition methods can be made particularly effective.

[0046] In a further embodiment, (as already described above) when repeating step (a), the number of samples (samples) can be supplemented with one or more sensor data sets of the plurality of sensor data sets whose degree of correspondence is less than a threshold value of the degree of correspondence. In this case, the samples are changed by supplementing sensor data sets which, after the previous run through steps (a) to (d), had a low degree of correspondence, that is, were below the threshold value of the degree of correspondence, and optionally had a certain travel distance. For example, a predetermined number of sensor data sets with the lowest degree of correspondence can be selected. Thus, an attempt is made to generate samples that fill in the gaps of unrecognized patterns in the previous samples for the repeated run through step (b).

[0047] In other specific embodiments, when step (a) is performed for the first time, the number of samples may be randomly selected from the plurality of sensor data sets.

[0048] In another embodiment, in step (c), supervised pattern recognition can be performed in parallel for the remaining sensor data sets. This means that when performing supervised pattern recognition, a plurality of the remaining sensor data sets can be checked in parallel. To this end, each available worker is associated with one of the remaining sensor data sets, so that each worker performs supervised pattern recognition individually.

[0049] Thus, less complex supervised pattern recognition can be performed more quickly by parallelization. Likewise, the available computing power (ie, the workers) is efficiently utilized.

[0050] Supervised pattern recognition can be parallelized by applying the corresponding pattern recognition algorithms to different data sets in parallel on a cloud computing platform, such as AWS, Azure, etc. Cloud-based parallelization allows the method to scale particularly well as the amount of data increases.

[0051] In another embodiment, step (a) may further include: connecting the sensor data sets (sample data sets) of the sample number in series into a unique data set. This means: merging the time series in the sensor data sets into a unique time series. The algorithm for unsupervised pattern recognition is applied once and serially to the serially connected time series of the samples. Since the sample data sets are connected in series and thus form a relatively long time series, the recurring patterns can be better distinguished compared to relatively short data sets. Thus, unsupervised pattern recognition is improved by merging the sample data sets into a unique time series. Compared with the method in which multiple data sets are examined simultaneously, the serial application of the algorithm up to a certain size of the serially connected time series is also very fast, because the coordination cost is lower, because only a unique algorithm is applied to the data set. Compared with the parallel application to multiple data sets, the serial application of the algorithm is also particularly simple. In addition, the subsequent merging of the results of the unsupervised pattern recognition performed in multiple data sets is omitted.

[0052] In another embodiment, in step (b), unsupervised pattern recognition can be implemented by a single worker. Here, an algorithm for unsupervised pattern recognition is implemented by a single worker. In some examples, the most efficient worker can be selected from a plurality of workers for this purpose. Thus, supervised pattern recognition can be performed relatively quickly.

[0053] In further embodiments, step (b) may further comprise determining a global pattern dictionary based on the unsupervised pattern recognition, wherein the global pattern dictionary defines the recognized patterns and is used for the supervised pattern recognition.

[0054] In another embodiment, in step (b), unsupervised pattern recognition can be performed in parallel for the sensor data sets in the sample number. In other words, unsupervised pattern recognition is performed separately for each sensor data set (sample data set) in the sample. Thus, the sample data sets can be checked for patterns at least partially simultaneously.

[0055] This approach is different from the implementation in which only a single worker performs unsupervised pattern recognition. That is, in this phase, all other workers are not needed, although they are activated and can be used. Therefore, all available workers are currently used to perform unsupervised pattern recognition individually for the sample data set.

[0056] Similar to parallelized supervised pattern recognition, unsupervised pattern recognition can be parallelized by applying the corresponding pattern recognition algorithms (for unsupervised pattern recognition) to different data sets in parallel on a cloud computing platform (e.g. AWS, Azure, etc.). Cloud-based parallelization makes the method particularly scalable as the amount of data increases.

[0057] Furthermore, in embodiments in which the algorithm for performing unsupervised pattern recognition is applied serially to the serially connected sample data sets, it may occur that the time series obtained from the serially connected sample data sets is too long for the algorithm. Therefore, in such cases, it is suitable to parallelize the unsupervised pattern recognition.

[0058] Therefore, in another embodiment, step (b) can be implemented by multiple workers. To this end, each available worker in the multiple workers is associated with one of the sample data sets, so that each worker performs supervised pattern recognition separately. By parallelizing unsupervised pattern recognition, the available computing power (that is, the workers) can be effectively utilized.

[0059] In another embodiment, step (b) may further include determining a specific pattern dictionary for each sensor data set of the sample number based on unsupervised pattern recognition, wherein the specific pattern dictionary defines the recognized pattern. In addition, the method may further include combining the specific pattern dictionaries into a global pattern dictionary, which is used to perform supervised pattern recognition according to step (c).

[0060] This embodiment is based on performing unsupervised pattern recognition in parallel by different workers. It follows that the individual unsupervised pattern recognition of the sample data set has a corresponding specific pattern dictionary as a result. In order to use the result in the supervised pattern recognition in step (c), the specific pattern dictionary (of the sample data set) is summarized or combined into the global dictionary mentioned above. For example, the specific dictionary can contain consistent patterns. These patterns are only adopted once in the global dictionary. Thus, the global dictionary contains all the patterns present in the specific dictionary at once. Thus, the redundancy in the specific dictionary is eliminated for the global dictionary.

[0061] In a further embodiment, the method described above may further comprise processing the sensor data set for unsupervised pattern recognition. This refers to: processing the sensor signal values / data in the sensor data set. For example, a discretization of the values ​​acquired by the sensor in a multivariate time series may be performed. In the case of a multivariate time series, the discretization is performed for each dimension of the time series. In other words, each of the different sensor signals of the (multivariate) time series is discretized. Alternatively or optionally, a frequency filtering for smoothing the sensor signals may be additionally applied.

[0062] To discretize the time series, different methods known to those skilled in the art can be chosen. For example, the multivariate (real) time series of the sensor data set can be converted into a symbolic time series (that is, a sequence of strings). For example, this conversion (or discretization) can be performed by applying the Symbolic Aggregate Approximation (SAX) algorithm.

[0063] As already mentioned above, for a multivariate time series, it may be necessary to perform unification, that is, convert the multivariate time series into a univariate time series.

[0064] In other embodiments, the processing of the sensor data sets can be performed in parallel. To this end, multiple workers can be provided so that data processing can be performed at least partially in parallel. Here, the number of existing workers determines how many sensor data sets can be processed in parallel. Parallelization can be performed on a cloud-based platform as described above with respect to pattern recognition.

[0065] Another aspect of the present disclosure relates to a method for vehicle design. The design method uses the results of the pattern recognition method (i.e., the identified patterns) for vehicle design. For example, the design method includes any of the methods described above for performing pattern recognition in a driving cycle, and further includes the following steps:

[0066] - determining a design data set based on the identified pattern; and

[0067] - Using the design data set for vehicle design.

[0068] By means of the above-described method for pattern recognition, the sensor data set can be broken down into recurring patterns (and thus recurring driving situations). These recurring patterns can then be used to determine the design data set.

[0069] In one example, a design data set can be determined (identified) from a plurality of sensor data sets, wherein the design data set is a representative sensor data set of the plurality of sensor data sets. To this end, a characterizing parameter can be determined for each sensor data set, which characterizing parameter describes the frequency of a pattern in the sensor data set. Thus, each sensor data set can be described based on the characterizing parameter, that is to say via the frequency of the occurring (and again identified) pattern.

[0070] For example, the characteristic parameter may include an absolute pattern frequency (APF), which describes the absolute frequency of a pattern in the sensor data set. The absolute pattern frequency spectrum includes the absolute pattern frequency of each pattern in the sensor data set.

[0071] For example, the characteristic parameters may also include a relative pattern coverage (RPC), which specifies how many time points (including their repetitions in the sensor data set) are covered by the pattern compared to the length of the sensor data set. It is not important here whether some of these time points are also covered by other patterns. Accordingly, the relative coverage spectrum includes the relative coverage of each pattern present in the sensor data set.

[0072] For example, the design data set can be identified from a plurality of sensor data sets in such a way that firstly a mean value of a characteristic parameter of each sensor data set is calculated and then the sensor data set which is closest to the mean value is selected as the design data set.

[0073] In another example, the design data set can be determined in other ways. For this purpose, a plurality of representative patterns are selected from the identified patterns of the global dictionary based on characteristic parameters. For example, a predetermined number of the most frequently occurring patterns and / or the patterns with the greatest relative coverage can be used to construct the design data set. However, other selection possibilities based on characteristic parameters for selecting representative patterns and their frequencies are also feasible.

[0074] The design data set thus determined (that is, identified or constructed) is used for vehicle design. For example, the component design, running stability, etc. of the vehicle can be designed and tested based on the design data set. Since the design data set is particularly representative for the driving operation of the vehicle, the vehicle design can be improved by using the design data set. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Now, embodiments of the present invention are described exemplarily and with reference to the accompanying drawings, in which:

[0076] Figure 1 A first method for identifying recurring patterns in a driving cycle is shown in a flow chart;

[0077] Figure 2 A second method for identifying recurring patterns in driving cycles is shown in a flow chart;

[0078] Figure 3 A method for identifying and using a design data set for vehicle design is shown in a flow chart; and

[0079] Figure 4 A method for constructing a design data set and using it for vehicle design is shown in a flow chart. DETAILED DESCRIPTION

[0080] refer to Figure 1 A method 100 for identifying recurring patterns in drive cycles is described.

[0081] In block 101, a plurality of sensor data sets are (data) processed. Here, the sensor data sets are stored in a data lake and each represent a driving cycle or a vehicle trip. In other words, the sensor signals in the sensor data sets are signal processed. Optionally, block 101 can also be omitted when the sensor data sets have already been processed for pattern recognition.

[0082] For processing the sensor data sets, a plurality of workers can be provided so that the data processing can be performed at least partially in parallel. In this case, the number of available workers determines how many sensor data sets can be processed in parallel.

[0083] The processing of the sensor data set can include discretization of the time series. In the case of a multivariate time series, a normalization can additionally be performed. Furthermore, the processing of the sensor data set can also be performed by applying a frequency filter for smoothing the sensor signal.

[0084] Optionally, in block 101, a preselection can also be made from the plurality of sensor data sets based on a predetermined criterion, so that only the preselection is processed. For example, the predetermined criterion can include a credibility check, which includes untrustworthy jumps in the sensor data set. Subsequently, the untrustworthy sensor data set from the plurality of sensor data sets is excluded from the pattern recognition.

[0085] For processing, the sensor data set can be retrieved from the data lake and processed by the computing unit. After processing, the sensor data set can be stored again in the data lake.

[0086] In the following, a “sensor data set” also means a processed sensor data set.

[0087] In block 102, a sample (or samples) is selected from a plurality of sensor data sets in a data lake. Unsupervised pattern recognition is provided for the sample to identify a pattern within the sample. Here, the sample may be selected randomly or based on a predetermined criterion.

[0088] In block 102a, the samples are adapted for a pattern recognition algorithm for unsupervised pattern recognition. In this case, the sample data sets are combined, in particular concatenated, into a single time series, so that the pattern recognition algorithm is applied only to the concatenated time series.

[0089] Depending on the pattern recognition algorithm used, the adaptation of the samples may include additional steps. For example, for some algorithms it is necessary to perform unification on multivariate time series.

[0090] In other examples (not shown), the adaptation according to block 102 a can also be performed in block 101 for all sensor data sets in the data lake.

[0091] In summary, in order to apply the pattern recognition algorithm for unsupervised pattern recognition, selection of samples or adaptation of samples is performed in blocks 102 and 102 a .

[0092] In block 103, unsupervised pattern recognition is performed to identify patterns in the sample (or in a time series consisting of samples). A pattern recognition algorithm is used for unsupervised pattern recognition. The corresponding pattern recognition algorithms are known to those skilled in the art. For example, a pattern recognition algorithm based on the Sequitur algorithm can be used. Alternatively, a pattern recognition algorithm based on the "Pattern-Enumeration-Algorithmus pattern enumeration algorithm" can also be used. In some examples, the pattern recognition algorithm for the unsupervised pattern can also be parameterized in addition to optimize the algorithm for the sample.

[0093] A pattern recognition algorithm is applied once and serially to the adapted samples. The pattern recognition algorithm identifies recurring patterns in the samples and creates a "global dictionary" from the recurring patterns and their definitions.

[0094] In other words, in block 103, a pattern recognition algorithm based on unsupervised pattern recognition determines recurring patterns in a plurality of samples to identify a (recurring) pattern.

[0095] In block 104, supervised pattern recognition is performed in the other (remaining) processed sensor data sets that are not part of the sample. That is, based on the previously created global dictionary, recurring patterns in the other remaining sensor data sets of the plurality of sensor data sets that are not part of the sample are identified. To perform supervised pattern recognition, corresponding pattern recognition algorithms known to those skilled in the art are used.

[0096] In block 104 a , a degree of correspondence is determined for each of the plurality of sensor data sets. The degree of correspondence is a quality criterion for the pattern recognition method 100 .

[0097] In block 105, the quality criterion is compared with a termination condition, that is, it is checked whether each of the plurality of sensor data sets has a degree of correspondence that is equal to or greater than a predetermined degree of correspondence threshold value.

[0098] For example, the predetermined threshold value of the degree of correspondence may be 80%. In other examples, the predetermined threshold value of the degree of correspondence may also be 85% or 90%.

[0099] In one example, blocks 102a, 103, 104, 104a, 105 and block 106 described later can be implemented as loopback repetitions. In this example, additional termination conditions can be set, which are checked in block 105. For example, additional termination conditions can be selected so that the loopback consisting of blocks 102a, 103, 104, 104a, 105 and 106 should traverse a predetermined number of loopback traversals. In other words, additional termination conditions can include a maximum number of loopback traversals. In this case, additional termination conditions are compared with additional quality standards indicating the number of loopback traversals. Additionally or optionally, additional termination conditions can be selected so that at least one new recurring pattern must be found in each loopback traversal. If no new pattern is found, the additional termination conditions are met.

[0100] Additional termination conditions may further include: no anomalies occurred in the pattern recognition during the unsupervised pattern recognition (block 103). For example, if too few patterns were identified during the unsupervised pattern recognition, or if the pattern recognition algorithm for the unsupervised pattern recognition terminated on its own, an anomaly may exist. In this case, the method 100 is repeated as described above.

[0101] If the termination condition (or conditions) is not met in block 105, the method proceeds to block 106. When the termination condition is met, the method ends.

[0102] In block 106, the sample is reselected or changed. In principle, in block 106, block 102 is repeatedly implemented. The change of the sample can be carried out in a variety of ways. For example, the sample can be supplemented with such a sensor data set from a plurality of sensor data sets, the correspondence degree of the sensor data set determined in block 104a is less than a predetermined correspondence degree threshold value, and optionally has a certain driving length. Alternatively, a completely new sample that was not previously part of the sample can also be selected from the plurality of sensor data sets. Next, method 100 proceeds to block 102a. If block 102a is omitted as described above, method 100 proceeds to block 103.

[0103] In the subsequent (repeated) block 103 , a new global dictionary is then created based on the changed samples, ie, the previously created global dictionary is replaced.

[0104] After the method is terminated, that is to say, when the termination condition is met, the result is a globally valid dictionary comprising a plurality of recurring patterns. Here, "globally valid" means that all data sets with respect to predetermined values ​​can be covered using the patterns stored in the dictionary.

[0105] In method 100, in block 103 (unsupervised pattern recognition), the unique time series merged by the sample data set is processed by only a unique worker. That is, no additional workers are used, although they are available.

[0106] Therefore, the following reference Figure 2 A second method 200 is described. The second method 200 uses the additional available workers to enable unsupervised pattern recognition to be performed in parallel for multiple univariate time series. Figure 2 A second method 200 for identifying recurring patterns in a driving cycle is shown in a flow chart. The second method 200 differs from the first method 100 in blocks 202a, 203 and 203a. Therefore, to avoid repetition, only the different blocks 203 and 203a are described.

[0107] In block 202a, the sensor data set of the sample is adapted for the pattern recognition algorithm for unsupervised pattern recognition (as in block 102a). To this end, a unification of the sample data sets can be performed if necessary. In contrast to block 102a, in block 202a, the sample data sets are not combined or concatenated into a single time series in any way. Instead, unsupervised pattern recognition is performed on the sample data sets in the subsequent block 203. If no adaptation of the sample data sets is required, block 202a can also be omitted.

[0108] In other examples (not shown), the adaptation according to block 202 a can also be performed in block 101 for all sensor data sets in the data lake.

[0109] In block 203, unsupervised pattern recognition is performed individually for each of the sample data sets. Here, the pattern recognition algorithm for unsupervised pattern recognition is applied individually to each of the sample data sets, so that a specific (pattern) dictionary is determined for each of the sample data sets. Since the pattern recognition algorithm runs individually on the sample data sets and is thus independent of the other sample data sets, the sample data sets can be checked in parallel in block 203. Here, the pattern recognition of the sample data sets is implemented by the workers. As already described with respect to block 103, algorithms for unsupervised pattern recognition are known to those skilled in the art. The result of performing unsupervised pattern recognition in parallel for a plurality of sample data sets is that a specific dictionary is created for each sample data set.

[0110] In block 203a, the specific dictionaries as a result of block 203 are combined into a global dictionary. For example, the specific dictionary may contain a consistent pattern. Then, the pattern is adopted only once in the global dictionary. Next, method 200 proceeds to block 104.

[0111] and Figure 1 Similar to method 100 in block 106, after changing the sample, Figure 2 Method 200 in proceeds to block 202a.

[0112] Based on the recurring patterns identified in one of the methods 100 , 200 for identifying recurring patterns and stored in the global dictionary, a design data set can be determined. This design data set can then be used for vehicle design. In the following, two methods 300 , 400 for determining and using a design data set are described.

[0113] Based on the recurring patterns identified in one of the methods 100 , 200 for identifying recurring patterns, a design data set can be determined. This design data set can then be used for vehicle design. Two methods 300 , 400 for determining and using a design data set are described below.

[0114] Figure 3 The flowchart of a method 300 for identifying a design data set from a plurality of sensor data sets and using it for vehicle design is shown in a flowchart.

[0115] In block 301, a characterizing parameter is determined for each sensor data set. Here, the characterizing parameter specifies the frequency of recurring patterns in the sensor data set. Thus, each sensor data set can be described based on the characterizing parameter, that is, via the frequency of the occurring (and again identified) patterns.

[0116] In block 302 , a design data set is determined (or identified) from a plurality of sensor data sets based on a characterizing parameter.

[0117] In block 303 , the design data set is used for vehicle design.

[0118] Figure 4 A method 400 for constructing a design data set based on re-identified patterns and using it for vehicle design is shown in a flow chart.

[0119] In block 401 , characterizing parameters are determined for each sensor data set (as in block 301 ).

[0120] In block 402, a plurality of representative patterns are determined from the identified patterns based on the characterizing parameters.

[0121] In block 403, a design data set is constructed based on a plurality of representative patterns.

[0122] In block 404 , the design data set is used for vehicle design.

[0123] This disclosure also relates to the following:

[0124] (1) A method for performing pattern recognition in a driving cycle, comprising:

[0125] (a) selecting a number of samples from a plurality of sensor data sets, wherein each sensor data set represents a driving cycle;

[0126] (b) performing unsupervised pattern recognition within the sample size to identify patterns;

[0127] (c) performing supervised pattern recognition on remaining sensor data sets in the plurality of sensor data sets to again identify patterns; and

[0128] (d) determining, for each of the plurality of sensor data sets, a degree of correspondence between the sensor data set and the pattern;

[0129] Wherein, steps (a) to (d) are repeatedly performed until each correspondence degree exceeds a correspondence degree threshold.

[0130] (2) The method according to (2), wherein the number of samples is changed when step (a) is repeated.

[0131] (3) A method according to (1) or (2), wherein, when step (a) is repeatedly implemented, the number of samples is supplemented by using one or more sensor data groups among the multiple sensor data groups whose degree of correspondence is less than a corresponding degree threshold.

[0132] (4) The method according to any one of (1) to (3), wherein when step (a) is performed for the first time, the number of samples is randomly selected from the plurality of sensor data sets.

[0133] (5) The method according to any one of (1) to (4), wherein in step (c), supervised pattern recognition is performed in parallel on the remaining sensor data sets.

[0134] (6) The method according to any one of (1) to (5), wherein step (a) further includes concatenating a sample number of sensor data sets into a single data set.

[0135] (7) The method according to any one of (1) to (5), wherein in step (b), unsupervised pattern recognition is performed on the sensor data sets in the sample quantity by a single worker.

[0136] (8) The method of (7), wherein step (b) further comprises determining a global pattern dictionary based on unsupervised pattern recognition, wherein the global pattern dictionary defines the recognized patterns and is used for supervised pattern recognition.

[0137] (9) The method according to any one of (1) to (6), wherein in step (b), unsupervised pattern recognition is performed in parallel for a sample number of sensor data sets.

[0138] (10) The method according to (6), wherein step (b) is performed by multiple workers.

[0139] (11) A method according to (6) or (7), wherein step (b) further includes determining a specific pattern dictionary for each sensor data group of the sample number based on unsupervised pattern recognition, wherein the specific pattern dictionary defines the recognized pattern, and the method further includes combining the specific pattern dictionaries into a global pattern dictionary, which is used to perform supervised pattern recognition according to step (c).

[0140] (12) The method according to any one of (1) to (11), wherein the method further comprises processing the sensor data set for unsupervised pattern recognition.

[0141] (13) The method according to (12), wherein the processing of the sensor data sets is performed in parallel.

[0142] Reference numerals list

[0143] 100 Pattern Recognition Methods

[0144] 101 Processing sensor data sets

[0145] 102 Select Sample

[0146] 102a Adaptation Sample

[0147] 103Performing unsupervised pattern recognition in samples

[0148] 104 Perform supervised pattern recognition on the remaining sensor data set

[0149] 104a Determining the degree of correspondence

[0150] 105 Check termination conditions

[0151] 106 Change Sample

[0152] 200 Pattern Recognition Methods

[0153] 203 Perform unsupervised pattern recognition separately for each data group in the sample

[0154] 203a Joining a specific dictionary

[0155] 300 First Design Method

[0156] 301 Determining Characteristic Parameters of the Sensor Data Set

[0157] 302 Identification Design Data Set

[0158] 303 Use of Design Data Sets for Vehicle Design

[0159] 400 Second Design Method

[0160] 401 Determine characteristic parameters of sensor data set

[0161] 402 Determine multiple representative patterns from the identified patterns

[0162] 403 Constructing a design data set based on multiple representative patterns

[0163] 404 Use of Design Data Sets for Vehicle Design

Claims

1. A method for performing pattern recognition in a driving cycle, comprising: a) selecting a number of samples (102) from a plurality of sensor data sets, wherein each sensor data set represents a driving cycle; b) performing unsupervised pattern recognition (103, 203) within the number of samples to identify a pattern; c) performing supervised pattern recognition (104, 204) on remaining sensor data sets in the plurality of sensor data sets to again recognize the pattern; and d) determining, for each of the plurality of sensor data sets, a degree of correspondence of the sensor data set to the pattern (105); The steps a) to d) are repeatedly performed until each degree of correspondence exceeds a corresponding degree threshold.

2. The method according to claim 1, wherein: When the step a) is repeatedly performed, the sample quantity is changed (106).

3. The method according to claim 1 or 2, wherein: When the step a) is repeatedly performed, the number of samples is supplemented by using one or more sensor data sets of the plurality of sensor data sets whose degree of correspondence is less than the threshold value of the degree of correspondence (106).

4. The method according to any one of claims 1 to 3, wherein: When step a) is carried out for the first time, the number of samples is randomly selected from the plurality of sensor data sets (102).

5. The method according to any one of claims 1 to 4, wherein: In step c), the supervised pattern recognition (104) is performed in parallel for the remaining sensor data sets.

6. The method according to any one of claims 1 to 5, wherein: In step b), the unsupervised pattern recognition (203) is performed in parallel for the sensor data sets in the sample number.

7. The method according to claim 6, wherein: The step b)(203) is implemented by a plurality of workers.

8. The method according to claim 6 or 7, wherein: The step b) further comprises determining a specific pattern dictionary (203) for each sensor data group of the sample quantity based on the unsupervised pattern recognition, wherein the specific pattern dictionary defines the recognized pattern, and the method further comprises combining the specific pattern dictionaries into a global pattern dictionary (203a), which is used to perform the supervised pattern recognition (104) according to step c).

9. The method according to any one of claims 1 to 8, further comprising processing the sensor data set (202a) for the unsupervised pattern recognition.

10. The method according to claim 9, wherein: The processing ( 202 a ) of the sensor data set takes place in parallel.

Images