Method and apparatus for vehicle velocity prediction based on can data
The vehicle speed prediction model analyzes past driving and road data to predict future speed, addressing the limitations of conventional methods by incorporating a sequence-to-sequence architecture with advanced neural networks, resulting in improved prediction accuracy and energy management.
Patent Information
- Application Number
- PCT/KR2024/005457
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-30
- Filing Date
- 2024-04-23
- Publication Date
- 2024-12-05
AI Technical Summary
Conventional vehicle speed prediction methods fail to accurately predict future speed changes due to their focus on past speed trends and neglect of driving patterns, especially how drivers react to road conditions, leading to suboptimal energy management and fuel efficiency.
A vehicle speed prediction model that analyzes past driving data and rear road data using an encoder to determine driving patterns and driver profiles, and predicts future speed based on these patterns and front road data, incorporating a sequence-to-sequence architecture with a Variable Selection Network, Long Short-Term Memory encoder, and Feature-wise Linear Modulation layers to account for road slope changes.
Improves prediction performance by considering driving patterns and road conditions, reducing mean absolute error and root-mean-square error compared to conventional models, and provides insights into driver behavior for better energy management.
Smart Images

Figure KR2024005457_05122024_PF_FP_ABST
Abstract
Description
Method and device for predicting vehicle speed based on CAN data
[0001] The present disclosure relates to a method and device for predicting vehicle speed based on CAN data. More specifically, the present disclosure relates to a method and device for predicting vehicle speed based on CAN data, which can improve vehicle speed prediction performance by analyzing driving patterns in an encoder based on CAN data, which is past driving data, and rearward road data, and predicting future vehicle speed based on the analyzed driving patterns and forward road data.
[0002]
[0003] In light of the climate and environmental crisis, global interest in reducing vehicle emissions is growing. Consequently, automotive research is making significant efforts to improve vehicle fuel efficiency and reduce exhaust gas emissions. One example of a method for improving fuel efficiency is vehicle velocity prediction (VVP).
[0004] The primary purpose of vehicle speed prediction is to predict energy needs at a certain point in the future and develop energy management strategies that reduce energy consumption. To develop energy management strategies, near-future velocity prediction is required.
[0005] Predicting vehicle speed primarily requires analyzing vehicle characteristics, driving patterns, and road conditions. However, previous studies have focused on analyzing past speed trends, making it impossible to predict changes in speed that deviate from these trends.
[0006] Recent advances in deep learning have led to extensive research on vehicle speed prediction based on various neural networks. While this has significantly improved prediction performance compared to previous studies, these studies still have limitations. Specifically, vehicle speed is significantly influenced by the driver's accelerator and brake pedal application patterns, making driving pattern analysis a crucial element in vehicle speed prediction. However, existing deep learning-based research focuses solely on accurately modeling road conditions and fails to account for driving patterns, such as how drivers react to changing road conditions. Furthermore, existing deep learning-based research lacks the ability to explain the model's output, making it difficult to analyze or debug the results.
[0007]
[0008] The problem to be solved by the present disclosure is to provide a vehicle speed prediction method and device based on CAN data that can improve vehicle speed prediction performance by analyzing a driving pattern in an encoder based on CAN data, which is past driving data, and rear road data, and predicting future vehicle speed based on the analyzed driving pattern and front road data.
[0009] The problems to be solved by the present disclosure are not limited to the problems described above, and problems not mentioned can be clearly understood by a person having ordinary skill in the technical field to which the present disclosure pertains from this specification and the attached drawings.
[0010]
[0011] In order to solve the above-described problem, a method for predicting a future speed of a vehicle using a vehicle speed prediction model according to one embodiment of the present disclosure includes a step in which an encoder analyzes past driving data of a vehicle and rear road data of the vehicle to output a driving pattern of a driver and a profile of the driver; and a step in which a decoder predicts a future speed of the vehicle based on the driving pattern, the past driving data, and the front road data of the vehicle, wherein the past driving data includes CAN (Controller Area Network) data, the rear road data includes a rear road slope, and the front road data includes a front road slope.
[0012] In order to solve the above-described problem, a device for predicting a future speed of a vehicle using a vehicle speed prediction model according to one embodiment of the present disclosure includes: an encoder for analyzing past driving data of the vehicle and rear road data of the vehicle to output a driving pattern of a driver and a profile of the driver; and a decoder for predicting a future speed of the vehicle based on the driving pattern, the past driving data, and the front road data of the vehicle, wherein the past driving data includes CAN (Controller Area Network) data, the rear road data includes a rear road slope, and the front road data includes a front road slope.
[0013] The solutions of the present disclosure are not limited to the solutions described above, and solutions that are not mentioned will be clearly understood by a person having ordinary skill in the art to which the present disclosure pertains from this specification and the attached drawings.
[0014]
[0015] According to embodiments of the present disclosure, vehicle speed prediction performance can be improved by analyzing driving patterns as well as vehicle characteristics and road conditions, and utilizing the analyzed driving patterns to predict vehicle speed.
[0016] The effects of the present disclosure are not limited to the effects described above, and effects not mentioned can be clearly understood by a person having ordinary skill in the art to which the present disclosure pertains from this specification and the attached drawings.
[0017]
[0018] FIG. 1 is a schematic diagram illustrating the architecture of a vehicle speed prediction model based on CAN data according to one embodiment of the present disclosure.
[0019] FIG. 2 is a diagram illustrating in more detail the architecture of a vehicle speed prediction model according to one embodiment of the present disclosure.
[0020] FIG. 3 is a diagram illustrating a driving path of a dataset for training a vehicle speed prediction model according to one embodiment of the present disclosure.
[0021] FIG. 4 is a table illustrating input types selected from among data collected for training a vehicle speed prediction model according to one embodiment of the present disclosure.
[0022] FIG. 5 is a table comparing the performance of a vehicle speed prediction model according to one embodiment of the present disclosure and conventional vehicle speed prediction models.
[0023] Figure 6 is a graph showing vehicle speed prediction results for driving data randomly selected from a test set on different driving routes.
[0024] Figure 7 is a diagram illustrating the results of analyzing variable weights for acceleration and deceleration situations.
[0025] FIG. 8 is a hardware configuration diagram of a computing device capable of implementing a vehicle speed prediction model according to one embodiment of the present disclosure.
[0026] FIG. 9 is a flowchart illustrating a vehicle speed prediction method based on CAN data according to one embodiment of the present disclosure.
[0027]
[0028] The advantages and features of the present disclosure, and methods for achieving them, will become clearer with reference to the embodiments described below in detail, along with the accompanying drawings. However, the present disclosure is not limited to the embodiments disclosed below and may be implemented in various different forms. These embodiments are provided solely to ensure the completeness of the present disclosure and to fully inform those skilled in the art of the scope of the invention. The present disclosure is defined solely by the scope of the claims.
[0029] Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used in their common sense to those of ordinary skill in the art to which this disclosure pertains. Furthermore, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.
[0030] The terminology used herein is for the purpose of describing embodiments and is not intended to limit the present disclosure. In this specification, singular forms also include plural forms, unless specifically stated otherwise. As used herein, the terms "comprises" and / or "comprising" do not exclude the presence or addition of one or more other components in addition to the components mentioned.
[0031] Hereinafter, embodiments of the present disclosure will be described with reference to the attached drawings. In the drawings, like reference numerals represent like components.
[0032] FIG. 1 is a schematic diagram illustrating the architecture of a vehicle speed prediction model based on CAN data according to one embodiment of the present disclosure. For convenience of explanation, the "CAN data-based vehicle speed prediction model" will be referred to as the "vehicle speed prediction model."
[0033] Referring to Figure 1, the vehicle speed prediction model can be implemented as a sequence-to-sequence (seq2seq) model. The sequence-to-sequence model can include an encoder that analyzes past data and a decoder that predicts the future speed of the vehicle.
[0034] The encoder analyzes CAN (Controller Area Network) data and rear road data to model the driver's driving pattern and driver profile. CAN data is time-domain data that includes a vehicle's historical drive record. CAN data can be collected from the OBD-II (on-board diagnostics 2) protocol via the CAN bus. CAN data can consist of a total of 30 types of sensor data transmitted to an electronic control unit (ECU), including accelerator pedal values, fuel consumption, wheel velocity, and transmission oil temperature. Rear road data can include rear road slope. In the present disclosure, rear road slope is can be expressed as, is a scalar indicating the road slope.
[0035] The decoder predicts the future speed of the vehicle based on the driving pattern modeled by the encoder, past driving data, and forward road data. The forward road data may include the forward road slope. In an embodiment, the slope of the forward 200 m may be indexed at 20 m intervals and input to the decoder. In the present disclosure, the future speed is can be expressed as, represents the time it takes to run 200m.
[0036] For a more detailed description of the encoder and decoder, refer to Fig. 2.
[0037] FIG. 2 is a diagram illustrating in more detail the architecture of a vehicle speed prediction model according to one embodiment of the present disclosure.
[0038] Referring to FIG. 2, the encoder may include a variable selection network (VSN), a long short-term memory (LSTM) encoder layer, and a feature-wise linear modulation (FiLM) layer.
[0039] VSNs weight crucial variables at each time point and remove noise, or features that unnecessarily impact performance. To achieve this, VSNs can be composed of Gated Linear Units (GLU) layers and Gated Residual Network (GRN) layers.
[0040] The GLU layer acts as a gate that controls whether a linear or nonlinear operation is performed. The GLU layer has the form of Equation 1.
[0041]
[0042] In mathematical expression 1, σ(·) represents the sigmoid activation function, and X' represents the input vector of the GLU layer. W (·) represents the weight, and b (·) indicates bias. ⊙ indicates element-wise product.
[0043] The GRN layer provides the flexibility to provide non-linearity (non) as needed. The GRN layer has the form shown in Equation 2.
[0044]
[0045]
[0046]
[0047] In mathematical expression 2, ELU stands for exponential linear unit activation function. means the input (primary input) of GRN. denotes an additional condition vector. η1 and η2 denote the two hidden layers of the GRN. If there is no condition in the GRN layer, is equal to 0. The GRN layer increases the flexibility of the model by performing nonlinear processing only when necessary through a gate structure and residual connection.
[0048] The output of the softmax layer following the GRN layer has the format of mathematical expression 3.
[0049]
[0050] In mathematical expression 3 represents the nth input feature and the variable selection weight for time t. represents a vector of static variables. represents a flattened vector of all past inputs. When CAN data is input, the VSN calculates variable selection weights by passing the flattened input vector on the time axis to the GRN layer and the softmax layer. Then, by multiplying the input variables by the variable selection weights, it highlights crucial variables and excludes noise variables.
[0051]
[0052] In Equation 4, C represents a past driving record. And ξ represents the output of a weighted VSN. In one embodiment of the present disclosure, past CAN data is used as an input feature. Then, the VSN analyzes the importance of each variable and selects variables that have a high contribution to predicting future speed. For example, the VSN selects a predetermined number of variables in descending order of ξ value. In another example, the VSN selects all variables whose ξ value is greater than a threshold value among the variables.
[0053] The FiLM layer combines historical driving data with rear-end road data. In one embodiment, the FiLM layer applies feature-wise linear modulation to input features based on conditioning information to effectively combine the input feature map with additional information. This process is expressed mathematically as shown in Equations 5 and 6.
[0054]
[0055] In Equation 5, f(·) and h(·) represent two single linear layers. And Xpast represents the rear road gradient vector through 1-D convolution.
[0056]
[0057] In Equation 6, γ and β represent modulations of neural network activations H, and H represents the hidden representation of LSTM.
[0058] The purpose of using the combination of Equations 5 and 6 is to create features that can represent how drivers react to changing road conditions. The FiLM layer masks periods of no change in road conditions, allowing the model to focus on areas where driving patterns can be captured.
[0059] Referring to FIG. 2, the decoder may include a FiLM layer, an LSTM decoder layer, and a GRN layer.
[0060] The decoder's FiLM layer combines driving pattern vectors with forward road data, enabling focused analysis on sections where gradient changes occur. Using the results from the FiLM layer, the decoder can predict how the driver will change speed based on the road gradient ahead.
[0061] The architecture of a vehicle speed prediction model according to an embodiment of the present disclosure has been described with reference to FIGS. 1 and 2. Below, a dataset for training a vehicle speed prediction model, training of a vehicle speed prediction model using the dataset, and experimental results for a trained vehicle speed prediction model will be described in sequence.
[0062] Datasets
[0063] The vehicle speed prediction model described above can be trained using a dataset. In one embodiment, the dataset may consist of multiple drivers who drove twice, once from the starting point to the destination, and once from the destination to the starting point. For example, as illustrated in Figure 3, the dataset may consist of 10 drivers who drove round trips from Korea University to Sangam World Cup Stadium. The driving route illustrated in Figure 3 consists of city roads (green line), highways (red line), and parking lots (blue line), and the total distance is approximately 18 km, taking an average of 35 minutes one way.
[0064] Meanwhile, data can be collected at one-second intervals using OBD-II (On Board Diagnostics-II), which consists of 51 types of CAN data. According to an embodiment, 30 input types can be selected from the collected data, excluding inputs with overlapping meanings. The selected input types are listed in Figure 4.
[0065] A total of eight driving datasets were used to train the vehicle speed prediction model, while two driving datasets were used to evaluate the model's performance. Forward road data can be used, including the 200-meter slope ahead, obtained from Google Earth.
[0066] Training
[0067] Multi-task learning can be performed on the encoder to analyze driving patterns based on road conditions. Subsequently, an end-to-end classifier can be added, classifying drivers at the encoder output and predicting speed at the decoder's final output. The loss function for training the vehicle speed prediction model is given in Equation 7.
[0068]
[0069] In mathematical formula 7 stands for weighting factor. stands for the mean squared error (MSE) loss for vehicle speed prediction. And stands for the cross entropy (CE) loss for driver recognition. By changing the vehicle speed prediction results, the results can be compared by model, and the comparison results will be described later with reference to Fig. 5.
[0070] To ensure fair comparison, all models used in the performance evaluation are trained under the same settings. The Adam optimizer is used with an initial learning rate of 0.001 for 100 epochs (where an epoch is the number of times the entire dataset has been learned). The model is evaluated after every epoch. Training samples are divided into sliding time windows with a duration of 10 seconds and a step size of 1 second.
[0071] Results and Analyses
[0072] The vehicle speed prediction model according to one embodiment of the present disclosure can be compared with existing vehicle speed prediction models. Examples of existing vehicle speed prediction models include the LSTM model, an LSTM model with added attention, and the TFT (Temporal Fusion Transformers) model.
[0073] The test set consists of driving data from two unknown drivers for a total driving time of 6.28 hours. The performance of vehicle speed prediction can be evaluated based on the mean absolute error (MAE) and root-mean-square error (RMSE) caused by the unknown drivers on two different driving routes. The experimental results of the evaluation set are illustrated in Figures 5 and 6.
[0074] FIG. 5 is a table comparing the performance of a vehicle speed prediction model according to one embodiment of the present disclosure and conventional vehicle speed prediction models.
[0075] In FIG. 5, a vehicle speed prediction model according to an embodiment of the present disclosure is denoted as 'CAN2V'. Furthermore, 'CAN2V(w=1)' means that 1 is applied as the weighting coefficient value in Equation 7. This means that driver classification is not performed through an encoder. Furthermore, 'CAN2V(w=0.001)' means that 0.01 is applied as the weighting coefficient value in Equation 7. This means that driver classification is performed through an encoder.
[0076] As illustrated in FIG. 5, the vehicle speed prediction model according to an embodiment of the present disclosure can be seen to have the lowest MAE and RMSE in all situations compared to conventional vehicle speed prediction models. In addition, when comparing the MAE (2.89 and 3.01) when w = 0.01 with the MAE (3.21 and 3.44) when w = 1, the MAE when w = 0.01 is 0.32 and 0.43 lower than the MAE when w = 1, respectively. In other words, the MAE and RMSE are lower when the driver is classified than when the driver is not classified. These results demonstrate that the CAN2V encoder appropriately represents driver characteristics and improves vehicle speed prediction performance.
[0077] Figure 6 is a graph showing vehicle speed prediction results for driving data randomly selected from a test set on different driving routes.
[0078] Figure 6(a) is a graph depicting the predicted driving route from Korea University to Sangam World Cup Stadium. Figure 6(b) is a graph depicting the predicted driving route from Sangam World Cup Stadium to Korea University. In Figures 6(a) and 6(b), the gray solid line represents the target speed, the black dotted line represents the speed predicted by the LSTM model, and the red dotted line represents the speed predicted by the CAN2V model.
[0079] Referring to Figures 6(a) and 6(b), we can see that the LSTM model also predicts the target speed well, but the CAN2V model predicts the target speed better than the LSTM model. Specifically, several peaks and valleys are marked with blue solid circles in Figures 6(a) and 6(b), and examining these parts, we can see that the CAN2V model is closer to the target speed than the LSTM model. In other words, we can see that the CAN2V model predicts the target speed better than the LSTM model. In particular, we can see that the prediction performance of the CAN2V model for highways is superior to that of the LSTM model.
[0080] Meanwhile, according to the CAN2V model, driving paths can be divided into acceleration and deceleration situations, allowing analysis of which variables have higher weights in each situation. For a more detailed explanation, refer to Figure 7.
[0081] Figure 7 is a diagram illustrating the results of analyzing variable weights for acceleration and deceleration situations.
[0082] Figure 7(a) illustrates variable selection weights in acceleration situations from 40 km / h to 65 km / h. Figure 7(b) illustrates variable selection weights in deceleration situations from 54 km / h to 24 km / h. In Figures 7(a) and 7(b), the x-axis represents the input time window, and the y-axis represents the input features of the CAN data.
[0083] Referring to Figure 7(a), we can see that front wheel velocity (rows 10 and 11) and current gear (row 26) are important in acceleration situations. Referring to Figure 7(b), we can see that lateral acceleration speed (row 18) and load value (row 2) are important in deceleration situations. These variables are related to acceleration and deceleration in terms of vehicle dynamics. Thus, the CAN2V model is proven to be effective in CAN data analysis.
[0084] Above, a vehicle speed prediction model according to one embodiment of the present disclosure has been described. The vehicle speed prediction model described above can be implemented as a device. Below, with reference to FIG. 8, an exemplary computing device capable of implementing the vehicle speed prediction model according to one embodiment of the present disclosure will be described.
[0085] Referring to FIG. 8, a computing device (800) may include one or more processors (810), storage (850) for storing a computer program (851), memory (820) for loading a computer program (851) executed by the processor (810), a bus (830), and a network interface (840). However, only components related to the embodiments of the present disclosure are illustrated in FIG. 8. Therefore, a person skilled in the art to which the present disclosure pertains may understand that other general components may be included in addition to the components illustrated in FIG. 8.
[0086] The processor (810) controls the overall operation of each component of the computing device (800). The processor (810) may include a central processing unit (CPU), a microprocessor unit (MPU), a microcontroller unit (MCU), a graphics processing unit (GPU), or any other type of processor well known in the art of the present disclosure. In addition, the processor (810) may perform calculations for at least one computer program for executing a vehicle speed prediction model according to embodiments of the present disclosure. The computing device (800) may include one or more processors.
[0087] The memory (820) stores data that supports various functions of the computing device (800). The memory (820) stores one or more of a plurality of computer programs (apps, application programs, or application software) running on the computing device (800), data, instructions, and information for the operation of the computing device (800). At least some of the computer programs may be downloaded from an external device (not shown). In addition, at least some of the computer programs may exist on the computing device (800) from the time of shipment for the basic functions of the computing device (800) (e.g., message reception, message transmission).
[0088] Meanwhile, the memory (820) can load one or more computer programs (851) from the storage (850) to execute a vehicle speed prediction model according to embodiments of the present disclosure. In FIG. 8, a random access memory (RAM) is illustrated as an example of the memory (820).
[0089] The bus (830) provides communication between components of the computing device (800). The bus (830) may be implemented as various types of buses, such as an address bus, a data bus, and a control bus.
[0090] The network interface (840) supports wired and wireless Internet communication of the computing device (800). Furthermore, the network interface (840) may support various communication methods other than Internet communication. To this end, the network interface (840) may be configured to include a communication module well known in the technical field of the present disclosure.
[0091] Storage (850) can non-temporarily store one or more computer programs (851). Storage (850) can be configured to include non-volatile memory such as Read Only Memory (ROM), Erasable Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), flash memory, a hard disk, a removable disk, or any form of computer-readable recording medium well known in the art to which the present disclosure pertains.
[0092] Meanwhile, the components described above can be implemented as modules. A module can be a software or hardware component, such as a Field Programmable Gate Array (FPGA) or an Application Specific Integrated Circuit (ASIC), that performs certain functions. However, a module is not limited to software or hardware. A module can be configured to reside on an addressable storage medium or be configured to execute one or more processors.
[0093] Thus, as an example, a module includes components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided by components and modules may be combined into a smaller number of components and modules or further separated into additional components and modules.
[0094] FIG. 9 is a flowchart illustrating a vehicle speed prediction method based on CAN data according to one embodiment of the present disclosure.
[0095] First, the encoder of the vehicle speed prediction model analyzes the vehicle's past driving data and the vehicle's rear road data to output the driver's driving pattern and driver's profile (S910).
[0096] The above step S910 may include a step of selecting variables that have a high contribution to predicting the future speed of the vehicle among input variables included in past driving data, and a step of capturing how the driver reacts to changes in road conditions by combining the past driving data and changes in the rear road slope.
[0097] The step of selecting variables among input variables included in past driving data that have a high contribution to predicting the future speed of a vehicle may include the step of calculating variable selection weights for each input variable included in the past driving data, and the step of multiplying each input variable by the variable selection weights and selecting variables based on the weighted values. In this case, the variable selection weights may be calculated based on the mathematical expressions 1 to 3 described above.
[0098] The step of capturing how a driver reacts to changing road conditions by combining past driving data with changes in rearward road slope may include combining the input feature map and the rearward road slope by applying feature-wise linear modulation to the input features based on conditioning information. This step may be calculated based on Equations 5 and 6 described above.
[0099] After step S910, the decoder of the vehicle speed prediction model predicts the future speed of the vehicle based on the driving pattern, past driving data, and road data ahead of the vehicle (S920).
[0100] The above step S920 may include a step of combining driving patterns and forward road data to intensively analyze a section where a change in the forward road slope occurs, and a step of predicting how the driver will change the speed of the vehicle according to the forward road slope.
[0101] Meanwhile, the disclosed embodiments may be implemented in the form of a recording medium storing computer-executable instructions. The instructions may be stored in the form of program code, and when executed by a processor, may generate program modules to perform the operations of the disclosed embodiments. The recording medium may be implemented as a computer-readable recording medium.
[0102] Computer-readable storage media include all types of storage media that store instructions that can be deciphered by a computer. Examples include read-only memory (ROM), random access memory (RAM), magnetic tape, magnetic disks, flash memory, and optical data storage devices.
[0103] Additionally, a computer-readable recording medium may be provided in the form of a non-transitory storage medium. Here, the term "non-transitory storage medium" simply means a tangible device that does not contain signals (e.g., electromagnetic waves). This term does not distinguish between cases where data is permanently stored in the storage medium and cases where data is temporarily stored. For example, a "non-transitory storage medium" may include a buffer in which data is temporarily stored.
[0104] According to one embodiment, the method according to various embodiments disclosed in the present document may be provided as included in a computer program product. The computer program product may be traded as a product between a seller and a buyer. The computer program product may be distributed in the form of a machine-readable recording medium (e.g., compact disc read only memory (CD-ROM)), distributed through an application store (e.g., Play Store™) or directly between two user devices (e.g., smartphones), or distributed online (e.g., downloaded or uploaded). In the case of online distribution, at least a portion of the computer program product (e.g., a downloadable app) may be temporarily stored or temporarily generated on a machine-readable recording medium, such as the memory of a manufacturer's server, an application store's server, or an intermediary server.
[0105] The features, structures, effects, etc. described in the above embodiments are included in at least one embodiment of the present disclosure, and are not necessarily limited to just one embodiment. Furthermore, the features, structures, effects, etc. exemplified in each embodiment can be combined or modified in other embodiments by a person skilled in the art to which the embodiments pertain. Therefore, the contents related to such combinations and modifications should be construed as being included within the scope of the present disclosure.
[0106] In addition, although the above description focuses on the embodiments, these are merely examples and do not limit the present disclosure. Those skilled in the art to which the present disclosure pertains will appreciate that various modifications and applications not exemplified above are possible without departing from the essential characteristics of the present embodiments. In other words, each component specifically shown in the embodiments can be modified and implemented. In addition, differences related to such modifications and applications should be interpreted as being included within the scope of the present disclosure defined in the appended claims.
Claims
1. A method for predicting the future speed of a vehicle using a vehicle speed prediction model, A step in which the encoder analyzes the vehicle's past driving data and the vehicle's rear road data to output the driver's driving pattern and driver's profile; and A decoder comprising a step of predicting the future speed of the vehicle based on the driving pattern, the past driving data and the road data ahead of the vehicle, The above past driving data includes CAN (Controller Area Network) data, The above rear road data includes the rear road slope, The above road ahead data includes the road ahead slope, A method for predicting vehicle speed based on CAN data.
2. In paragraph 1, The above outputting step is: A step in which a variable selection network selects variables that have a high contribution to predicting the future speed of a vehicle among input variables included in the above-mentioned past driving data; and The FiLM layer comprises a step of combining the above-described past driving data with the changes in the rear road slope to capture how the driver reacts to changes in road conditions. A method for predicting vehicle speed based on CAN data.
3. In paragraph 2, The steps for selecting the above variables are: A step of calculating variable selection weights for each input variable included in the above past driving data; and A step of selecting variables based on the weighted values by multiplying the variable selection weights by the input variables, A method for predicting vehicle speed based on CAN data.
4. In paragraph 3, The above variable selection weights are calculated based on [Mathematical Formula 1] to [Mathematical Formula 3], The above weighted value is calculated based on [Mathematical Formula 4]. A method for predicting vehicle speed based on CAN data. [Mathematical Formula 1] σ(·): sigmoid activation function X': Input vector of the GLU layer W (·) : Weight b (·) : bias ⊙ : Element-wise product [Equation 2] ELU: exponential linear unit activation function : Additional condition vector η1 and η2: two hidden layers of the GRN [Equation 3] : Variable selection weight for the nth input feature and time t : Vector of static variables : a flattened vector of all past inputs [Equation 4] C: Past driving data ξ: weighted value 5. In paragraph 2, The above capturing steps are: A step of combining an input feature map and the rear road slope by applying feature-wise linear modulation to the input features based on conditioning information, A method for predicting vehicle speed based on CAN data.
6. In paragraph 5, The above combining step is calculated based on [Mathematical Formula 5] and [Mathematical Formula 6]. A method for predicting vehicle speed based on CAN data. [Equation 5] f(·) and h(·): two single linear layers X past : Rear road slope vector through 1D convolution [Equation 6] γ and β: Modulations of neural network activations H H: LSTM's hidden representation 7. In paragraph 1, The step of predicting the future speed of the above vehicle is: A step of intensively analyzing a section where a change in the slope of the road ahead occurs by combining the driving pattern and the road ahead data; and A step of predicting how the driver will change the speed of the vehicle according to the forward road slope, A method for predicting vehicle speed based on CAN data.
8. In paragraph 1, The loss function in learning the above vehicle speed prediction model is determined by [Mathematical Formula 7]. A method for predicting vehicle speed based on CAN data. [Equation 7] : weighting factor : Mean Squared Error (MSE) loss for vehicle speed prediction : Cross entropy (CE) loss for driver recognition 9. In a device for predicting the future speed of a vehicle using a vehicle speed prediction model, An encoder that analyzes the vehicle's past driving data and the vehicle's rear road data to output the driver's driving pattern and driver profile; and A decoder for predicting the future speed of the vehicle based on the driving pattern, the past driving data and the road data ahead of the vehicle, The above past driving data includes CAN (Controller Area Network) data, The above rear road data includes the rear road slope, The above road ahead data includes the road ahead slope, Vehicle speed prediction device based on CAN data.
10. In paragraph 9, The above encoder, A variable selection network that selects variables that have a high contribution to predicting the future speed of a vehicle among input variables included in the above past driving data; and A FiLM layer that captures how a driver reacts to changing road conditions by combining the above-mentioned past driving data with the changes in the rear road slope, Vehicle speed prediction device based on CAN data.
11. In paragraph 10, The steps for selecting the above variables are: A step of calculating variable selection weights for each input variable included in the above past driving data; and A step of selecting variables based on the weighted values by multiplying the variable selection weights by the input variables, Vehicle speed prediction device based on CAN data.
12. In paragraph 11, The above variable selection weights are calculated based on [Mathematical Formula 1] to [Mathematical Formula 3], The above weighted value is calculated based on [Mathematical Formula 4]. Vehicle speed prediction device based on CAN data. [Mathematical Formula 1] σ(·): sigmoid activation function X': Input vector of the GLU layer W (·) : Weight b (·) : bias ⊙ : Element-wise product [Equation 2] ELU: exponential linear unit activation function : Additional condition vector η1 and η2: two hidden layers of the GRN [Equation 3] : Variable selection weight for the nth input feature and time t : Vector of static variables : a flattened vector of all past inputs [Equation 4] C: Past driving data ξ: weighted value 13. In paragraph 10, The above capturing steps are: A step of combining an input feature map and the rear road slope by applying feature-wise linear modulation to the input features based on conditioning information, A method for predicting vehicle speed based on CAN data.
14. In paragraph 13, The above combining step is calculated based on [Mathematical Formula 5] and [Mathematical Formula 6]. A method for predicting vehicle speed based on CAN data. [Equation 5] f(·) and h(·): two single linear layers X past : Rear road slope vector through 1D convolution [Equation 6] γ and β: Modulations of neural network activations H H: LSTM's hidden representation 15. A non-transitory computer-readable recording medium having recorded thereon a program including commands for executing each step of a vehicle speed prediction method based on CAN data described in any one of claims 1 to 8.
Citation Information
Patent Citations
Pure electric automobile remaining mileage model predication method based on path information
CN109733248A
Driving support method, and driving support device, automatic driving control device, vehicle and program utilizing the same
JP2016216021A
Control system and method of hybrid vehicle
KR1020170016706A
Wagom that can be transformed into chair
KR1020230168530A
KR20220036200A
