Method and system for classifying passenger tendency of dedicated vehicle

Abstract

A method and system for classifying passenger tendencies of a dedicated vehicle performs the following: generating time series data by collecting driving tendencies data of passengers using the dedicated vehicle in chronological order; outputting a plurality of new features generated by learning the time series data through an artificial neural network to perform dimension reduction on the time series data; a clustering algorithm is applied to the new features generated by dimensionality reduction of the time series data to classify the driving tendency of the passenger.

Description

Technical Field

[0001] The present invention generally relates to a method and system for classifying passenger tendencies of purpose-built vehicles, and more particularly, to a method and system for classifying passenger tendencies of purpose-built vehicles capable of classifying the tendencies of various passengers using the purpose-built vehicles so as to satisfy the various passengers without providing or investing other resources or additional costs. Background Art

[0002] Recently, with the rapid development of vehicle electrification and artificial intelligence (AI) technology, vehicles are no longer simply a means of transportation, but are evolving into purpose-built vehicles (PBVs) that serve various specific purposes.

[0003] Therefore, the vehicle industry is developing specialized vehicles that provide customized services based on the preferences of passengers.

[0004] In this trend, as consumers' awareness shifts from owning a vehicle to using a dedicated vehicle, mobility services that can cater to a variety of passengers are expanding.

[0005] Like this, since multiple passengers use one dedicated vehicle, the operating time is longer than that of existing vehicles, and the driving tendencies of various passengers can be recorded.

[0006] Therefore, it can be integrated with artificial intelligence technology, which requires huge amounts of data and is easy to collect.

[0007] Based on this, if artificial intelligence technology can be used or applied to classify the driving tendencies of various passengers, customized services can be provided to each passenger, resulting in greater satisfaction and improved service quality.

[0008] Although massive amounts of data are required when applying artificial intelligence technology, dedicated vehicles are used by a variety of passengers and can therefore easily collect diverse and large amounts of data.

[0009] However, if all this data is used to learn and infer through artificial intelligence technology, it will increase the consumption of resources and time.

[0010] Therefore, it is necessary to select appropriate features and use artificial intelligence technology to reduce the dimension of the data.

[0011] In the past, dimensionality reduction techniques such as principal component analysis (PCA) and linear discriminant analysis (LDA) were used to reduce physical dimensions. However, this traditional dimensionality reduction technique has the problem of disappearing characteristic differences between features in the process of reducing physical dimensions.

[0012] At the same time, artificial intelligence technology for analyzing driver tendencies mainly uses recurrent neural networks (RNNs) that are good at processing time series data.

[0013] The input data selects the characteristics of the operating behavior of passengers who can specifically operate the steering device, accelerator, brake, etc. and the vehicle status, and explores parameters that meet the classification purpose to classify in a manner that prioritizes personal preferences.

[0014] Most existing technologies use guided learning based on this, which requires labeling work to match the correct value with the input data, so there is a problem of consuming a lot of resources and time. Summary of the invention

[0015] Problem that the invention aims to solve

[0016] Some exemplary embodiments of the present invention may provide a method and system for classifying passenger propensity of a special vehicle capable of classifying the propensity of various passengers to use the special vehicle, so as to satisfy various passengers without investing other resources or additional costs.

[0017] Means used to solve problems

[0018] To achieve the above-mentioned purpose, the passenger tendency classification method for special-purpose vehicles according to the first aspect of the present invention is characterized in that the following operations are performed: the driving tendency data of passengers using special-purpose vehicles are collected in chronological order and time series data are generated; the time series data are learned through an artificial neural network and dimension reduction is performed to output multiple new features as intermediate results; and the k-means clustering algorithm is applied to the reduced-dimensional new features to classify the driving tendencies of the passengers.

[0019] When the driving tendency data is vehicle operation data, it may include data associated with yaw rate, steering angle, vehicle speed, longitudinal acceleration, and lateral acceleration.

[0020] The time series data is learned by an autoencoder based on long short-term memory (LSTM).

[0021] A plurality of representative passenger tendency clusters may be generated by applying a k-means clustering algorithm to the new features whose dimensions have been reduced.

[0022] When classifying passenger tendency of new data after generating the representative passenger tendency clusters, the center coordinates of a plurality of the representative passenger tendency clusters may be used.

[0023] After reducing the dimension of the new data, the center coordinates of the cluster closest to the new feature whose dimension has been reduced can be obtained and the tendency of the new data can be classified into the cluster.

[0024] As an alternative, the passenger tendency classification method for special-purpose vehicles according to the present invention may include: a dimensionality reduction model generation step, collecting the driving tendency data of passengers using the special-purpose vehicles in chronological order and generating time series data, learning the time series data through an artificial neural network and performing dimensionality reduction; a representative passenger tendency cluster generation step, outputting a plurality of the new features as intermediate results of dimensionality reduction, and applying the k-means clustering algorithm to the new features to generate a plurality of representative passenger tendency clusters; and a passenger tendency classification step, calculating the distance between the new features derived from the dimensionality reduction of the new passenger's driving tendency data and the centers of the plurality of representative passenger tendency clusters, and selecting the representative passenger tendency cluster having the center closest to the new features.

[0025] The generation of the time series data is achieved in the following manner: collecting driving tendency data of multiple passengers, filtering the driving tendency data into multiple features according to the purpose, and arranging the filtered multiple features of the multiple passengers in chronological order.

[0026] When the purpose is to classify the driving tendency of the passenger, yaw rate, steering angle, vehicle speed, longitudinal acceleration, and lateral acceleration may be selected as the driving tendency data features.

[0027] The time series data is learned by an autoencoder based on long short-term memory (LSTM).

[0028] In the learned autoencoder structure, only the encoder is separated to be used as a dimensionality reduction model.

[0029] If the time series data is input into the encoder, a plurality of the new features can be output as intermediate results of dimensionality reduction.

[0030] In the step of generating representative passenger tendency clusters, if the k-means clustering algorithm is applied to the new features, a plurality of clusters and a pair of center coordinates of each cluster can be obtained.

[0031] The results generated by the plurality of clusters may be evaluated by the Silhouette Coefficient to derive the optimal number of clusters.

[0032] On the other hand, the passenger tendency classification system for special-purpose vehicles according to the second aspect of the present invention may include: a time series data generating unit, which collects driving tendency data of passengers using special-purpose vehicles in chronological order and generates time series data; a dimension reduction model generating unit, which learns the time series data through an artificial neural network and performs dimension reduction; a representative passenger tendency cluster generating unit, which outputs a plurality of the new features as intermediate results of dimension reduction, and applies the k-means clustering algorithm to the new features to generate a plurality of representative passenger tendency clusters; and a passenger tendency classification unit, which calculates the distance between the new features derived from the dimension reduction of the new passenger's driving tendency data and the centers of the plurality of representative passenger tendency clusters, and selects the representative passenger tendency cluster having the center closest to the new features.

[0033] The time series data is learned by an autoencoder based on long short-term memory (LSTM).

[0034] In the learned autoencoder structure, only the encoder is separated to be used as a dimensionality reduction model. If the time series data is input into the encoder, a plurality of the new features can be output as intermediate results of dimensionality reduction.

[0035] In the representative passenger tendency cluster generating unit, if the k-means clustering algorithm is applied to the new features, a plurality of clusters and a pair of center coordinates of each cluster can be obtained.

[0036] The results generated by the plurality of clusters may be evaluated by the Silhouette Coefficient to derive the optimal number of clusters.

[0037] Specific matters and alternatives of the exemplary embodiments are contained in the "Detailed Description" and the accompanying "Drawings".

[0038] The advantages and / or features of the present invention and methods for achieving the same will be made clear through the accompanying drawings and various embodiments described in detail below.

[0039] However, the present invention is not limited to the structures of the embodiments described below, but can be implemented through various different forms. The various embodiments described in this specification are only used to fully disclose the present invention so that ordinary technicians in the field can fully understand the scope of the present invention. The scope of the present invention will be determined by the scope of each claim in the claims of the present invention.

[0040] Effects of the Invention

[0041] According to the means for solving the aforementioned problems, some embodiments of the present invention have the following effects.

[0042] According to some embodiments of the present invention, dimensionality reduction may be performed based on an autoencoder to prevent the disappearance of characteristic differences between features and to generate new characteristic differences between features.

[0043] In addition, according to certain embodiments of the present invention, a plurality of representative passenger tendency clusters are generated by applying a k-means clustering algorithm to the new features whose dimensions have been reduced, thereby making it possible to classify the tendencies of various passengers to use dedicated vehicles, thereby having the effect of satisfying various passengers without requiring other resources or additional costs.

[0044] The above is merely illustrative and does not limit the present invention in any way. In addition to the embodiments, examples, and features described above, further embodiments, examples, and features will become apparent from the drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 1 is a flow chart showing a method for classifying passenger tendencies of a special vehicle according to the present invention in sequence.

[0046] Figure 2 is a block diagram for explaining a system for classifying passenger tendencies of a special vehicle according to an embodiment of the present invention.

[0047] Figure 3 It is a diagram for explaining a time series data generating step in a method for classifying passenger tendency of a dedicated vehicle according to an embodiment of the present invention.

[0048] Figure 4 It is a diagram for explaining a dimension reduction model generation step in a method for classifying passenger tendencies of a special vehicle according to an embodiment of the present invention.

[0049] Figure 5 1 is a diagram for explaining a representative passenger tendency cluster generation step in a method for classifying passenger tendencies of a special vehicle according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In the following detailed description, reference will be made to the accompanying drawings which are a part of the present invention. The exemplary embodiments described in the detailed description, drawings and claims are not intended to limit the present invention, and other embodiments may be used or changes may be made without departing from the spirit or scope of the present invention.

[0051] Hereinafter, the preferred embodiments of the method and system for classifying the passenger tendency of a special vehicle according to the present invention will be described in detail with reference to the accompanying drawings. For reference, the terms and words used in this specification and the claims should not be interpreted in a limited manner to the general or dictionary meanings, but should be based on the principle that the inventor can appropriately define the concepts of the terms in order to describe his invention in the best way, so as to be consistent with the meaning and concept of the technical idea of ​​the present invention. In addition, the embodiments described in this specification and the structures shown in the accompanying drawings are only the most preferred embodiments of the present invention and do not represent the technical idea of ​​the present invention. It should be understood that the present application may have various equivalents and modifications that can replace them.

[0052] Figure 1 1 is a flow chart showing a method for classifying passenger tendencies of a special vehicle according to the present invention in sequence.

[0053] The method for classifying the passenger tendencies of special-purpose vehicles according to an embodiment of the present invention is characterized in that the following operations are performed: generating time series data by collecting driving tendency data of passengers using special-purpose vehicles in chronological order (step 110); outputting a plurality of new features generated by learning the time series data through an artificial neural network to reduce the dimension of the time series data (step 130); applying a clustering algorithm to the new features generated by reducing the dimension of the time series data to classify the driving tendencies of passengers (step 150). The clustering algorithm may be a K-means clustering algorithm. In addition, the time series data may be learned by an autoencoder (step 120). Reference Figure 4 Step 120 is described in detail.

[0054] For example, the driving tendency data may be vehicle operation data that is data related to the operation of the vehicle. The driving tendency data may include, for example, but not limited to, data associated with the yaw rate, steering angle, speed, longitudinal acceleration, and lateral acceleration of the vehicle.

[0055] The k-means clustering algorithm is a vector quantization algorithm originally from signal processing that aims to partition observations into clusters, where each observation belongs to the cluster with the nearest mean (cluster center or cluster centroid), which serves as the prototype of the cluster. The k-means clustering algorithm is an algorithm that bundles the given data into k clusters, acting in a way that minimizes the dispersion of the differences in distance to each cluster.

[0056] Such a k-means clustering algorithm, as a kind of autonomous learning, performs the function of labeling unlabeled input data.

[0057] By applying the k-means clustering algorithm, resources and time can be saved compared to the existing technology that requires labeling work to match the correct value with the input data.

[0058] As an alternative, the passenger tendency classification method for special vehicles according to an embodiment of the present invention may include: a time series data generating step (step 110), a dimensionality reduction model generating step (step 130), a representative passenger tendency cluster generating step (step 160), and a passenger tendency classification step (step 170).

[0059] The time series data generation step (step 110) is performed to collect driving tendency data of passengers using special vehicles in time sequence and generate time series data. The time step data is data points indexed in time sequence.

[0060] The dimension reduction model generation step (step 130) is performed to learn the time series data generated in the time series data generation step (step 110) through an artificial neural network and perform dimension reduction.

[0061] A representative passenger tendency cluster generation step (step 160) is performed to output a plurality of new features as intermediate results of the dimension reduction in the dimension reduction model generation step (step 130), and a k-means clustering algorithm is applied to such new features (step 150) to generate a plurality of representative passenger tendency clusters (step 160).

[0062] After executing the passenger tendency classification step (step 170) to generate clusters in the representative passenger tendency cluster generation step (step 160), the distances between the new features derived by reducing the dimension of the new passenger's driving tendency data and the centers of multiple representative passenger tendency clusters are calculated, and the representative passenger tendency cluster having the center closest to the new features is selected (step 180).

[0063] Figure 21 is a block diagram for illustrating a system for classifying the passenger tendency of a special-purpose vehicle according to an embodiment of the present invention. The system for classifying the passenger tendency of a special-purpose vehicle according to an embodiment of the present invention may include a memory and a processor. The system for classifying the passenger tendency of a special-purpose vehicle according to an embodiment of the present invention may also include a time series data generating unit 210, a dimensionality reduction model generating unit 220, a representative passenger tendency cluster generating unit 230, and a passenger tendency classifying unit 240. The processor is configured to control the time series data generating unit 210, the dimensionality reduction model generating unit 220, the representative passenger tendency cluster generating unit 230, and the passenger tendency classifying unit 240 by executing computer executable instructions stored in the memory to perform the following operations.

[0064] The time series data generating section 210 may be configured to collect driving tendency data of passengers using the dedicated vehicle in chronological order and generate time series data.

[0065] The dimension reduction model generation unit 220 may be configured to learn the time series data generated in the time series data generation unit 210 through an artificial neural network and perform dimension reduction.

[0066] The representative passenger tendency cluster generation unit 230 may be configured to output a plurality of new features as intermediate results of dimension reduction in the dimension reduction model generation unit 220 , and apply a k-means clustering algorithm to such new features to generate a plurality of representative passenger tendency clusters.

[0067] The passenger tendency classification unit 240 can be configured to calculate the distance between the new features derived from the dimensionality reduction of the new passenger's driving tendency data and the centers of multiple representative passenger tendency clusters after generating clusters in the representative passenger tendency cluster generation unit 230, and select the representative passenger tendency cluster with the center closest to such new features to complete the passenger tendency classification.

[0068] The memory is configured to store a long short-term memory based autoencoder for learning the time series data.

[0069] Figure 3 It is a diagram for explaining a time series data generating step in a method for classifying passenger tendency of a dedicated vehicle according to an embodiment of the present invention.

[0070] The generation of the time series data is performed by collecting driving tendency data of a plurality of passengers, selecting a plurality of features from such driving tendency data according to a purpose, and arranging the selected plurality of features of the plurality of passengers in chronological order.

[0071] For example, when the purpose of filtering the driving tendency data is to classify the driving tendency of passengers, the features of yaw rate, steering angle, vehicle speed, longitudinal acceleration, and lateral acceleration may be filtered from the driving tendency data.

[0072] Specifically, refer to Figure 3 , collect the data of i passengers, and after screening n features according to the purpose, accumulate the screened n features of all i passengers in chronological order and generate time series data.

[0073] Figure 4 It is a diagram for explaining a dimension reduction model generation step in a method for classifying passenger tendencies of a special vehicle according to an embodiment of the present invention.

[0074] The time series data is learned by an autoencoder based on long short-term memory (LSTM).

[0075] An autoencoder is a neural network model that learns a compressed representation of its input.

[0076] Although technically training is done using a supervised learning method called self-directed learning, it can actually be an unsupervised learning method.

[0077] Typically, autoencoders are trained by partially trying a wide range of models that reproduce the input.

[0078] Autoencoder models are designed with a restricted architecture to intentionally induce this kind of problem in order to induce a bottleneck phenomenon at the midpoint of the reconstruction model of the input data.

[0079] There are many types and uses of autoencoders, but feature extraction models are often used as autoencoders.

[0080] In this case, once the model is started, the reconstructed side of the model will be discarded and the model up to the bottleneck point can be used.

[0081] The output in a bottleneck scenario is a fixed-length vector that provides a compressed representation of the input data.

[0082] After this, the input data of the domain can be fed to the model, and the output from the bottleneck phenomenon can be used to visually guide the learning of the model's feature vectors or for dimensionality reduction.

[0083] The LSTM autoencoder uses the encoder-decoder LSTM architecture to implement an autoencoder for sequence data.

[0084] Given a sequence dataset, the encoder-decoder LSTM is configured to read, encode, decode, and regenerate the input sequence.

[0085] The performance of the model is evaluated based on its ability to reproduce the input sequence.

[0086] When the desired performance level of the model is achieved, the decoder part of the model is removed, leaving only the encoder model.

[0087] This model can be used to encode an input sequence into a fixed-length vector.

[0088] As a result, the vectors can be used not only for compressed representation of sequences but also for various applications as input to other guided learning models.

[0089] In the learned autoencoder structure, if only the encoder is separated to be used as a dimensionality reduction model, the well-learned autoencoder will use the intermediate result of the reduced dimension to predict the original data. The more accurate it is, the better the characteristics of the original data are contained in the intermediate result of the reduced dimension.

[0090] like Figure 4 As shown, if time series data is input into the encoder, new features 1 to 32 are output as intermediate results of the reduced dimension of the time series data.

[0091] Therefore, the dimension of the time series data with the size of time (t) * number of features (n) is reduced to 32.

[0092] The autoencoder inputs the intermediate result of dimensionality reduction of time series data as the original input of the decoder.

[0093] Figure 5 1 is a diagram for explaining a representative passenger tendency cluster generation step in a method for classifying passenger tendencies of a special vehicle according to an embodiment of the present invention.

[0094] In the step of generating representative passenger tendency clusters, if the k-means clustering algorithm is applied to the new features, multiple clusters and a center coordinate pair of each cluster can be obtained.

[0095] That is, if the number of clusters is set to C (2 to 9) and the k-means clustering algorithm is applied to the new features, C clusters and the center coordinates of C pairs of clusters can be obtained.

[0096] The results of clustering into 2 to 9 clusters can be evaluated by the Silhouette Coefficient to derive the optimal number of clusters.

[0097] The silhouette coefficient can be evaluated by calculating the distance between each data point and the surrounding data points, and can be used as a measure to evaluate whether the data within the cluster is well clustered and whether the clusters are well distinguished.

[0098] For example, Figure 5 As shown, if the optimal silhouette coefficient is set when three clusters are formed, the center coordinates of the three clusters and three pairs of clusters can be set as representative passenger tendency clusters.

[0099] In terms of the silhouette coefficient, the better the score is, the closer the features in a cluster are to the cluster they belong to and the farther they are from other clusters.

[0100] That is, the score is high when the clusters are well classified by characteristics.

[0101] After forming a representative cluster, the center coordinates of the representative cluster are used when classifying the tendency of new data.

[0102] After reducing the dimension of the new data, the center coordinates of the cluster closest to the reduced new features are obtained to classify the clusters according to the tendency of the new data.

[0103] As such, some embodiments of the present invention use an autoencoder for dimensionality reduction, which can prevent the disappearance of characteristic differences between features and generate new characteristic differences between features. By applying the k-means clustering algorithm to the new features that have been reduced in dimension, multiple representative passenger tendency clusters are generated to classify the tendencies of various passengers using dedicated vehicles, thereby having the advantage of not requiring the provision of other resources or additional charges and being able to satisfy a variety of passengers.

[0104] The present invention is not limited to the above-mentioned embodiments and drawings, and a person skilled in the art will appreciate that various substitutions, deformations and changes may be made without departing from the technical concept of the present invention.

[0105] As described above, various embodiments herein are described for exemplary purposes, and various modifications may be made without departing from the scope and ideas of the present invention. Therefore, the various embodiments disclosed herein are not restrictive, and the scope and ideas of the present invention are represented by the appended claims.

Claims

1. A method for classifying the passenger tendencies of a special vehicle, characterized in that The steps include: Time series data are generated by collecting driving tendency data of passengers using special vehicles in chronological order; Outputting a plurality of new features generated by learning the time series data through an artificial neural network to reduce the dimension of the time series data; as well as A clustering algorithm is applied to the new features generated by reducing the dimension of the time series data to classify the driving tendency of the passengers.

2. The method for classifying the passenger tendency of a special vehicle according to claim 1, characterized in that The driving tendency data includes data associated with one or more of a yaw rate, a steering angle, a vehicle speed, a longitudinal acceleration, and a lateral acceleration of the dedicated vehicle.

3. The method for classifying the passenger tendency of a special vehicle according to claim 1, characterized in that Learning the time series data by using a long short-term memory based autoencoder; The method for classifying passenger propensity of a special vehicle further comprises the steps of: applying a clustering algorithm to the new features generated by reducing the dimension of the time series data to generate a plurality of representative passenger propensity clusters; The clustering algorithm includes a k-means clustering algorithm.

4. The method for classifying the passenger tendency of a special vehicle according to claim 3, characterized in that The following steps are also included: When new data is classified into driving tendency classes after the plurality of representative passenger tendency clusters are generated, the center coordinates of the plurality of representative passenger tendency clusters are used.

5. The method for classifying the passenger tendency of a special vehicle according to claim 4, characterized in that The following steps are also included: After reducing the dimension of the new data, the center coordinates of the cluster closest to the new feature generated by reducing the dimension of the new data are obtained, and the cluster closest to the new feature generated by reducing the dimension of the new data is classified into one or more driving tendencies for the new data.

6. A method for classifying the passenger tendencies of a special vehicle, characterized in that The steps include: collecting driving tendency data of passengers using the special vehicles in chronological order to generate time series data, and learning the time series data through an artificial neural network to reduce the dimension of the time series data; Outputting a plurality of new features as intermediate results of dimension reduction of the time series data, and applying a clustering algorithm to the new features to generate a plurality of representative passenger tendency clusters; as well as The distance between a new feature derived by dimensionality reduction of time series data generated from driving tendency data of one or more new passengers and the centers of the plurality of representative passenger tendency clusters is calculated, and a representative passenger tendency cluster having a center closest to the new feature is selected from the plurality of representative passenger tendency clusters.

7. The method for classifying the passenger tendency of a special vehicle according to claim 6, characterized in that The steps of generating the time series data include the following steps: Driving tendency data of a plurality of passengers are collected, a plurality of features are screened from the driving tendency data, and the screened plurality of features of the plurality of passengers are arranged in chronological order.

8. The method for classifying the passenger tendency of a special vehicle according to claim 6, characterized in that The following steps are also included: Learning the time series data by an autoencoder; The self-encryptor includes an autoencoder based on long short-term memory.

9. The method for classifying the passenger tendency of a special vehicle according to claim 8, characterized in that After the step of learning the time series data by the autoencoder, the method for classifying the passenger tendency of a special vehicle further comprises the following steps: The decoder is removed from the autoencoder and the encoder is retained in the autoencoder to be used as a dimensionality reduction model.

10. The method for classifying the passenger tendency of a special vehicle according to claim 9, characterized in that When the time series data is input into the encoder, a plurality of the new features are output as intermediate results of dimension reduction of the time series data.

11. The method for classifying passenger tendencies of a special vehicle according to claim 6, characterized in that The step of applying a clustering algorithm to the new feature includes the step of generating a plurality of the representative passenger tendency clusters and a center coordinate of each of the plurality of the representative passenger tendency clusters.

12. The method for classifying the passenger tendency of a special vehicle according to claim 11, characterized in that The following steps are also included: The results generated by the plurality of representative passenger tendency clusters are evaluated by means of a silhouette coefficient to optimize the number of the representative passenger tendency clusters.

13. A system for classifying the propensity of passengers in a special vehicle, characterized in that include: Memory, and processor; The processor is configured to perform the following operations by executing computer executable instructions stored in the memory: The driving tendency data of passengers using dedicated vehicles are collected in chronological order to generate time series data, Learning the time series data by an artificial neural network to reduce the dimension of the time series data, Outputting a plurality of new features as intermediate results of dimension reduction of the time series data, and applying a clustering algorithm to the new features to generate a plurality of representative passenger tendency clusters, and calculating the distance between a new feature derived by performing dimension reduction on time series data generated from driving tendency data of one or more new passengers and the centers of the plurality of representative passenger tendency clusters, and selecting a representative passenger tendency cluster having a center closest to the new feature among the plurality of representative passenger tendency clusters; The memory is configured to store a long short-term memory based autoencoder for learning the time series data.

14. The system for classifying passenger tendencies of a special vehicle according to claim 13, characterized in that The memory is configured to store an encoder having a dimensionality reduction model, The processor is configured to output a plurality of the new features as intermediate results of dimensionality reduction on the time series data when the time series data is input into the encoder.

15. The system for classifying passenger tendencies of a special vehicle according to claim 13, characterized in that The processor is configured to: When applying the clustering algorithm to the new feature, a plurality of the representative passenger tendency clusters and the center coordinates of each of the plurality of the representative passenger tendency clusters are generated, The results generated from the plurality of representative passenger tendency clusters are evaluated by the silhouette coefficient to optimize the number of the representative passenger tendency optimization clusters.

Images