KR20200115710A - Vehicle Sideslip Angle Estimation Method and Apparatus Using an Integrated Deep Ensemble-Kalman Filter - Google Patents

Abstract

Provided are a method and a device for estimating a vehicle side slip angle by using a method of integrating a deep ensemble and a Kalman filter. According to an embodiment of the present invention, the device for estimating a vehicle side slip angle can comprise: a sensor information acquisition part that acquires sensor information from a vehicle; a deep ensemble based artificial neural network that receives the sensor information obtained from the sensor information acquisition part and estimates a predicted value; and a model-based observer that considers the predicted value estimated by the deep ensemble based artificial neural network as sensor information obtained from a virtual sensor, receives the predicted value, and estimates a sideslip angle value. Therefore, the reliability of the predicted value can be checked by estimating the uncertainty value of the predicted value as well as the predicted value by using the deep ensemble based artificial neural network.

Description

Vehicle Sideslip Angle Estimation Method and Apparatus Using an Integrated Deep Ensemble-Kalman Filter}

The following embodiments relate to a vehicle side slip angle estimation technology, and more particularly, to a vehicle side slip angle estimation method and apparatus in which a deep ensemble and a Kalman filter are integrated.

With the development of the automobile industry, vehicle technology is developing, and various active safety systems are being developed. At this time, it is very important to know the sideslip angle to keep the vehicle's stability.

As existing technologies, there are many methods of estimating the side slip angle using a Kalman filter based mainly on a vehicle model. In this case, in order to consider the nonlinearity of the system, an extended Kalman filter (EKF) and an unscented Kalman filter (UKF) are frequently used.

Recently, there are also many studies using artificial neural networks. There are approaches to solving this problem through a variety of networks, each showing the results of estimating side-slip angles with some success.

1A is a diagram illustrating a vehicle model used in an existing model-based observer.

As shown in FIG. 1A, in the case of a model-based observer among the prior art, there is a problem in that it is sensitive to uncertainty of a used model. First, in order to consider the nonlinearity of the system, performance is improved by using a nonlinear Kalman filter rather than a linear Kalman filter, but when model-related values such as tire cornering stiffness become uncertain, the prediction performance deteriorates.

1B is a diagram showing an existing artificial neural network-based network model.

As shown in FIG. 1B, there are a number of technologies using artificial neural networks, and this machine learning-based method has a problem that it is difficult to generalize the problem. For example, if the training dataset used to train the artificial neural network and the test dataset used to verify the trained model are not similar, it is difficult to produce a good estimation result.

Korean Patent Registration No. 10-1208369 relates to such a vehicle path prediction and method thereof, and detects the vehicle speed, steering angle, and side slip angle while the vehicle is driving, and compares the detected side slip angle with a preset reference angle. It describes a technique for a device for predicting a path.

Korean Patent Registration No. 10-1208369

The embodiments describe a method and apparatus for estimating a vehicle side slip angle in a method in which a deep ensemble and a Kalman filter are integrated, and not only estimating a predicted value using a deep ensemble-based artificial neural network, but also estimating the uncertainty value of the predicted value. By doing so, it is to provide a method and apparatus for estimating a vehicle side slip angle in a method in which a deep ensemble and a Kalman filter can be checked to determine how reliable a predicted value is.

In addition, the embodiments consider the predicted value and the uncertainty value resulting from the deep ensemble-based artificial neural network as a value obtained from a virtual new sensor and input it to the model-based observer, so that the value is determined according to the degree of uncertainty of the predicted value at every instant. The purpose of this study is to provide a method and apparatus for estimating vehicle side slip angles in a method that integrates a deep ensemble and a Kalman filter that can automatically determine how much to trust and use by a model-based observer.

According to an embodiment, an apparatus for estimating a side slip angle of a vehicle includes: a sensor information acquisition unit that acquires sensor information from a vehicle; A deep ensemble-based artificial neural network that receives the sensor information obtained from the sensor information acquisition unit and estimates a predicted value; And a model-based observer that considers the predicted value estimated by the deep ensemble-based artificial neural network as sensor information obtained from a virtual sensor, receives the predicted value and estimates a sideslip angle value. I can.

According to another embodiment, it may further include an artificial neural network learning unit for learning the deep ensemble-based artificial neural network through the acquired sensor information.

According to another embodiment, the deep ensemble-based artificial neural network estimates the predicted value by inputting the sensor information into the learned deep ensemble-based artificial neural network, estimates the uncertainty value of the predicted value, and the model-based observer The predicted value and the uncertainty value are regarded as sensor information obtained from a virtual sensor, and the predicted value and the uncertainty value and the sensor information obtained from the sensor information obtaining unit are input to a model-based observer to side-slip. You can estimate the angle value.

According to another embodiment, the model-based observer may automatically determine how much to trust and use the predicted value according to the uncertainty value of the predicted value.

According to another embodiment, the model-based observer may estimate the side slip angle using the model-based observer of a Kalman filter method based on a vehicle model.

According to another embodiment, the model-based observer may be a model-based observer of an Extended Kalman Filter (EKF) or an Unscented Kalman Filter (UKF) method.

According to another embodiment, an active safety system using the finally estimated side slip angle value in a vehicle active safety system may be further included.

A method of estimating a side slip angle of a vehicle according to an embodiment includes: acquiring sensor information from a vehicle; Estimating a predicted value by inputting the acquired sensor information into a learned deep ensemble-based artificial neural network; And estimating a sideslip angle value by considering the predicted value as sensor information obtained from a virtual sensor and inputting the predicted value to a model-based observer.

According to another embodiment, it may further include performing learning of a deep ensemble-based artificial neural network through the acquired sensor information.

According to another embodiment, the step of estimating the predicted value by inputting the sensor information into the learned deep ensemble-based artificial neural network includes inputting the sensor information into the learned deep ensemble-based artificial neural network to estimate the predicted value, The step of estimating the uncertainty value of the predicted value, and the step of estimating the side slip angle value by inputting the predicted value to a model-based observer, regards the predicted value and the uncertainty value as sensor information obtained from a virtual sensor. And estimating a side slip angle value by inputting the predicted value and the uncertainty value to a model-based observer.

According to another embodiment, the step of estimating a side slip angle value by inputting the predicted value to a model-based observer includes the sensor information obtained from the vehicle, the predicted value output from the deep ensemble-based artificial neural network, and the The uncertainty value may be input from the model-based observer and the side slip angle value may be estimated.

According to another embodiment, the step of estimating a side-slip angle value by inputting the predicted value to a model-based observer, the model-based observer, how much to trust and use the predicted value according to the uncertainty value of the predicted value. Can be determined automatically.

According to another embodiment, the step of estimating a side slip angle value by inputting the predicted value to a model-based observer includes the side slip angle using the model-based observer of a Kalman filter method based on a vehicle model. Can be estimated.

According to another embodiment, the step of estimating a side slip angle value by inputting the predicted value to a model-based observer, the model-based observer is an Extended Kalman Filter (EKF) or an Unscented Kalman Filter (Unscented Kalman Filter). Filter, UKF) type model-based observer.

According to another embodiment, the step of using the finally estimated side slip angle value in a vehicle active safety system may be further included.

According to embodiments, a deep ensemble and a Kalman filter that can determine how reliable the predicted value is by estimating the predicted value as well as the uncertainty value of the predicted value using a deep ensemble-based artificial neural network. A method and apparatus for estimating a vehicle side slip angle in an integrated manner can be provided.

In addition, according to embodiments, the predicted value and the uncertainty value resulting from the deep ensemble-based artificial neural network are regarded as values obtained from a virtual new sensor and are input to the model-based observer, so that the predicted value is determined according to the degree of uncertainty of the predicted value at every instant. It is possible to provide a method and apparatus for estimating a vehicle side slip angle in a method that integrates a deep ensemble and a Kalman filter that can automatically determine how much to trust and use a value in a model-based observer.

1A is a diagram illustrating a vehicle model used in an existing model-based observer.
1B is a diagram showing an existing artificial neural network-based network model.
2 is a flowchart illustrating a method of estimating a vehicle side slip angle according to an exemplary embodiment.
3 is a schematic diagram of an apparatus for estimating a vehicle side slip angle according to an exemplary embodiment.
4 is a diagram illustrating a structure of an artificial neural network based on a deep ensemble according to an embodiment.
5 is a diagram illustrating a result of estimating a side sleep angle of a vehicle using a deep ensemble-based artificial neural network according to an exemplary embodiment.
6 is a diagram illustrating a result of estimating a side slip angle of a vehicle using a deep ensemble-based artificial neural network and a model-based observer according to an exemplary embodiment.

Hereinafter, embodiments will be described with reference to the accompanying drawings. However, the described embodiments may be modified in various forms, and the scope of the present invention is not limited by the embodiments described below. In addition, various embodiments are provided to more completely explain the present invention to those of ordinary skill in the art. In the drawings, the shapes and sizes of elements may be exaggerated for clearer explanation.

The following embodiments relate to a technique for estimating a state variable important to an active safety system, and provide a technique for a method and apparatus for estimating a vehicle side slip angle in a method in which a deep ensemble and a Kalman filter are integrated. In order to compensate for the shortcomings of the existing model-based Kalman filter prediction and the simple artificial neural network-based estimation method, and to utilize each of the advantages, a deep ensemble method is introduced and an integrated method is proposed.

The present embodiments can be used in various vehicle active safety systems. It is essential to maintain lateral stability for safe driving of the vehicle, and the side slip angle is an important judgment index at this time. Therefore, it is very important to estimate this value well.

2 is a flowchart illustrating a method of estimating a vehicle side slip angle according to an exemplary embodiment.

Referring to FIG. 2, in the vehicle side slip angle estimation method according to an embodiment, the step of obtaining sensor information from the vehicle 210, and inputting the acquired sensor information into a learned deep ensemble-based artificial neural network to estimate a predicted value. Step 230, and a step 240 of estimating a sideslip angle value by considering the predicted value as sensor information acquired from the virtual sensor and inputting the predicted value to a model-based observer. have.

In addition, it may further include a step 220 of performing learning of a deep ensemble-based artificial neural network through the acquired sensor information.

And, it may further include the step of using the finally estimated side slip angle value in the vehicle active safety system.

The embodiments can utilize all of the advantages of a mixture of the artificial neural network technique, which has become a hot topic in recent years, and the method of an existing model-based observer. First, when learning through an artificial neural network is performed, not only the predicted value is obtained by using the deep ensemble method, but the uncertainty value of the value can also be obtained. This method is a method that is different from the prior art that only estimates the predicted value using an artificial neural network, and it is possible to know how reliable the predicted value is.

In addition, the predicted values and uncertainty values resulting from the deep ensemble-based artificial neural network can be considered as values obtained from a virtual new sensor. We can propose a new approach that utilizes this value for existing model-based observers. In this case, the uncertainty of the predicted value from the artificial neural network can be used for the measurement covariance matrix of the model-based observer (nonlinear Kalman filter). In this way, the model-based observer can automatically determine how much to trust and use the predicted value at every moment depending on the degree of uncertainty of the predicted value.

Hereinafter, a method of estimating a vehicle side slip angle according to an exemplary embodiment will be described in more detail.

The method for estimating a vehicle side slip angle according to an embodiment may be described in more detail with an example of an apparatus for estimating a vehicle side slip angle according to an embodiment.

3 is a schematic diagram of an apparatus for estimating a vehicle side slip angle according to an exemplary embodiment.

Referring to FIG. 3, the vehicle side slip angle estimation apparatus 300 according to an embodiment may include a sensor information acquisition unit 310, a deep ensemble-based artificial neural network 320, and a model-based observer 330. , Depending on the embodiment, it may further include an artificial neural network learning unit and an active safety system. The vehicle side slip angle estimation apparatus 300 according to this embodiment is performed in a manner in which a deep ensemble and a Kalman filter are integrated, and when the sensor data passes through the learned deep ensemble-based artificial neural network 320, the predicted value and the uncertainty value Can be printed together. The output predicted value and uncertainty value may be regarded as information obtained from a virtual sensor and used in the model-based observer 330. Through this, the model-based observer 330 may finally estimate the side-slip angle value in consideration of the uncertainty value transmitted from the artificial neural network at each moment.

In step 210, the sensor information acquisition unit 310 may acquire sensor information from the vehicle. For example, the sensor information acquisition unit 310 may acquire speed, yawrate, longitudinal acceleration, and lateral acceleration values from sensors such as speed, yawrate, longitudinal acceleration, and lateral acceleration configured in the vehicle.

The deep ensemble-based artificial neural network 320 is a deep learning model and may receive sensor information obtained from the sensor information acquisition unit 310 and estimate a predicted value. In particular, the deep ensemble-based artificial neural network 320 may input sensor information to the learned deep ensemble-based artificial neural network 320 to estimate a predicted value and at the same time estimate an uncertainty value of the predicted value. By estimating the uncertainty value of the predicted value in this way, it is possible to know the degree of reliability of the estimated predicted value.

The deep ensemble-based artificial neural network 320 may estimate a predicted value and an uncertainty value using a deep ensemble method when learning through the artificial neural network is performed.

Meanwhile, an artificial neural network learning unit for learning the deep ensemble-based artificial neural network 320 through the acquired sensor information according to the embodiment may be further included. The artificial neural network learning unit may be included in the deep ensemble-based artificial neural network 320.

The model-based observer 330 may regard the predicted value estimated from the deep ensemble-based artificial neural network as sensor information acquired from a virtual sensor, and receive the predicted value to estimate a sideslip angle value. In particular, the model-based observer 330 considers not only the predicted value but also the uncertainty value as sensor information obtained from a virtual sensor, and uses the predicted value and uncertainty value and the sensor information obtained from the sensor information acquisition unit 310 as a model-based observer. By entering 330, the side slip angle value can be estimated. The model-based observer 330 may automatically determine how much to trust and use the predicted value according to the uncertainty value of the predicted value.

The model-based observer 330 may estimate the side slip angle by using the model-based observer 330 of a Kalman filter method based on a vehicle model. In particular, the model-based observer 330 may be a model-based observer 330 of an extended Kalman filter (EKF) or an unscented Kalman filter (UKF) method.

Hereinafter, the model-based observers (EKF, UKF) 330 will be described in more detail with one example.

과

로 나타낼 수 있다. 그리고 이 값들을 새로운 가상의 센서로부터 얻은 측정(measurement)으로 간주하여 칼만필터에 이용할 수 있다. A new side-slip angle estimation algorithm can be proposed by integrating information obtained from the model-based observer (EKF, UKF) 330 and the deep ensemble-based artificial neural network 320 that are commonly used to estimate the side-slip angle have. The predicted value and the uncertainty value obtained from the deep ensemble-based artificial neural network 320

and

It can be expressed as In addition, these values can be regarded as measurements obtained from a new virtual sensor and used in the Kalman filter.

이 추가되어 측정 벡터(measurement vector)

를

로 표현할 수 있다. Here, as a measurement of the Kalman filter, the predicted value, which is a new measurement, is the sensor information that can be easily obtained from the current commercial vehicle, such as speed, yaw rate, longitudinal acceleration, and lateral acceleration.

Is added to the measurement vector

To

It can be expressed as

는

로 표현되어, 가상의 센서로부터 얻은 불확실도 값

에 따라 적응적으로(adaptive) 변하는 행렬(matrix)이 될 수 있다. In addition, the predicted value is a measurement noise covariance matrix

Is

Expressed as, the uncertainty value obtained from the virtual sensor

It may be a matrix that changes adaptively accordingly.

이 작을 때는 딥러닝 모델로부터 구한 예측 값

를 많이 신뢰하고, 불확실도 값

이 커질 때는 예측 값

을 덜 신뢰하고 다른 센서 측정(measurement)을 더 믿는 방식으로 최종 사이드슬립 앵글 값을 추정할 수 있다.Therefore, the uncertainty value

When is small, the predicted value obtained from the deep learning model

Trust a lot, uncertainty

Predicted value when

The final side-slip angle value can be estimated in a way that is less reliable and more reliable other sensor measurements.

An active safety system using the finally estimated side slip angle value in the vehicle active safety system may be further included.

According to embodiments, a method and apparatus for estimating a side slip angle may be designed using the deep ensemble-based artificial neural network 320 and the model-based observer 330 by mixing the advantages of conventional methods. This can be usefully used in active safety systems by estimating important state variables related to lateral safety.

4 is a diagram illustrating a structure of an artificial neural network based on a deep ensemble according to an embodiment.

Referring to FIG. 4, a structure of a deep ensemble-based artificial neural network 400 of a vehicle side slip angle estimation apparatus according to an exemplary embodiment described in FIG. 3 is shown.

만을 도출하는 것을 발전시켜, 딥 앙상블 기반 인공신경망(400)은 결과 값

뿐만 아니라 그 결과 값에 대한 불확실도 값을

의 형태로 함께 도출할 수 있다. 따라서 딥 앙상블 기반 인공신경망(400)의 현재 딥러닝 모델이 출력한

라는 값이 얼마나 믿을만한지를 함께 알려주는 것이다. 하나의 예를 들어 보다 상세히 설명한다. 여기서 사용되는 데이터셋(401)은 샘플링을 통해 샘플된 데이터셋(410)을 각 네트워크 모델(420)에 입력하여 학습 및 테스트할 수 있다. Previously, one result value when solving deep learning-based regression and classification problems

By developing the derivation of the gulf, the deep ensemble-based artificial neural network 400

Not only that, the uncertainty about the resulting value

Can be derived together in the form of. Therefore, the current deep learning model of the deep ensemble-based artificial neural network 400

Together, it tells how reliable the value of is. It will be described in more detail with one example. The data set 401 used here may be trained and tested by inputting the data set 410 sampled through sampling into each network model 420.

과 불확실도 값

을 출력할 수 있다.When designing and learning the deep ensemble-based artificial neural network (network) 400, it is assumed that a total of five network models 420 are used. At this time, when training is finished with the training dataset and the network model 420 is verified with the test dataset, a total of five network models 420 are predicted as follows:

And uncertainty values

Can be printed.

,

Network model 1:

,

,

Network model 2:

,

,

Network model 3:

,

,

Network model 4:

,

,

Network model 5:

,

The predicted value and the uncertainty value 402 of the entire deep ensemble-based artificial neural network 400 may be finally obtained by mixing the predicted value and the uncertainty value of each of the network models 420 thus derived. In this case, the predicted value and the uncertainty value 402 of the entire deep ensemble-based artificial neural network 400 may be calculated using the following equation.

[Equation 1]

[Equation 2]

과 root를 씌운

이 최종적으로 구해지는 예측 값과 표준편차 형태의 불확실도 값이 된다.Thus calculated

And rooted

This finally obtained predicted value and the uncertainty of the standard deviation form are also values.

In this way, learning of the deep ensemble-based artificial neural network 400 is performed through sensor information that can be easily obtained from a commercial vehicle, and a prediction value and an uncertainty value are estimated together from the deep ensemble-based artificial neural network 400, and this is virtualized. We can obtain the side-slip angle value by assuming that it is a sensor of and inputting it to a model-based observer. The finally estimated side slip angle value can be used in the vehicle's active safety system.

MATLAB/Simulink and Carsim were used for verification of this example. Data sets necessary for learning artificial neural networks were collected using Carsim, and deep ensemble-based artificial neural network networks were designed and learned using Python. The algorithm was designed to finally predict the side-slip angle by using this result in a model-based observer on MATLAB/Simulink.

5 is a diagram illustrating a result of estimating a side sleep angle of a vehicle using a deep ensemble-based artificial neural network according to an exemplary embodiment.

5, a side sleep angle value result obtained using a vehicle side sleep angle estimation method through a deep ensemble-based artificial neural network is shown, a dotted line (DNN) indicates a predicted value, and a solid line indicates an uncertainty value including uncertainty. .

6 is a diagram illustrating a result of estimating a side slip angle of a vehicle using a deep ensemble-based artificial neural network and a model-based observer according to an exemplary embodiment.

Referring to FIG. 6, the results of side-slip angle values obtained through two existing model-based observers (EKF and UKF), an EKF-based model-based observer (a) and a UKF-based model-based observer (b) are shown.

In addition, the results obtained through the two existing model-based observers (EKF and UKF) and the results of the method in which the method of estimating the vehicle side slip angle of the method incorporating the deep ensemble and Kalman filter proposed in this embodiment are shown together. Here, the real value represents the actual value obtained from Carsim.

[Table 1]

Table 1 shows a comparison of the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) for verifying the results. As an experimental scenario, it is assumed that the vehicle speed is accelerated from 70kph to 120kph in the middle by giving a sine wave steering in two friction coefficient situations consisting of dry asphalt and slippery road.

Looking at the results, it can be seen that performance is improved through the method according to the present embodiment in all situations. That is, in the case of EKF + DNN and UKF + DNN, it can be seen that performance is improved compared to the existing EKF and UKF methods.

In the case of embodiments, it can be applied to a vehicle active safety system. In order to apply the system, driving information of the vehicle is required, and the lateral acceleration sensor and yawrate sensor that are generally installed in commercial vehicles can be used.

The apparatus described above may be implemented as a hardware component, a software component, and/or a combination of a hardware component and a software component. For example, the devices and components described in the embodiments include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It can be implemented using one or more general purpose computers or special purpose computers, such as a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications executed on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of software. For the convenience of understanding, although it is sometimes described that one processing device is used, one of ordinary skill in the art, the processing device is a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it may include. For example, the processing device may include a plurality of processors or one processor and one controller. In addition, other processing configurations are possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to behave as desired or processed independently or collectively. You can command the device. Software and/or data may be interpreted by a processing device or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. Can be embodyed in The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of the program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

As described above, although the embodiments have been described by the limited embodiments and drawings, various modifications and variations are possible from the above description by those of ordinary skill in the art. For example, the described techniques are performed in a different order from the described method, and/or components such as a system, structure, device, circuit, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Therefore, other implementations, other embodiments, and claims and equivalents fall within the scope of the claims to be described later.

Claims (15)

A sensor information acquisition unit that acquires sensor information from a vehicle;
A deep ensemble-based artificial neural network that receives the sensor information obtained from the sensor information acquisition unit and estimates a predicted value; And
A model-based observer that considers the predicted value estimated by the deep ensemble-based artificial neural network as sensor information acquired from a virtual sensor, receives the predicted value and estimates a sideslip angle value
Including a vehicle side slip angle estimation device. The method of claim 1,
An artificial neural network learning unit that performs learning of a deep ensemble-based artificial neural network through the acquired sensor information
The vehicle side slip angle estimation apparatus further comprising. The method of claim 1,
The deep ensemble-based artificial neural network,
Input the sensor information into the learned deep ensemble-based artificial neural network to estimate the predicted value, estimate the uncertainty value of the predicted value,
The model-based observer,
The predicted value and the uncertainty value are regarded as sensor information obtained from a virtual sensor, and the predicted value, the uncertainty value, and the sensor information obtained from the sensor information obtaining unit are input to a model-based observer, and a side slip angle value To estimate
A vehicle side slip angle estimation device, characterized in that. The method of claim 3,
The model-based observer,
Automatically determining how much to trust and use the predicted value according to the uncertainty value of the predicted value
A vehicle side slip angle estimation device, characterized in that. The method of claim 1,
The model-based observer,
Estimating the side-slip angle using the model-based observer of a vehicle model-based Kalman filter method
A vehicle side slip angle estimation device, characterized in that. The method of claim 5,
The model-based observer,
Model-based observer of Extended Kalman Filter (EKF) or Unscented Kalman Filter (UKF) method
A vehicle side slip angle estimation device, characterized in that. The method of claim 1,
Active safety system using the finally estimated side slip angle value in vehicle active safety system
The vehicle side slip angle estimation apparatus further comprising. Obtaining sensor information from the vehicle;
Estimating a predicted value by inputting the acquired sensor information into a learned deep ensemble-based artificial neural network; And
Estimating a sideslip angle value by considering the predicted value as sensor information obtained from a virtual sensor and inputting the predicted value to a model-based observer
Including a vehicle side slip angle estimation method. The method of claim 8,
Learning a deep ensemble-based artificial neural network through the acquired sensor information
The vehicle side slip angle estimation method further comprising. The method of claim 8,
The step of estimating the predicted value by inputting the sensor information into the learned deep ensemble-based artificial neural network,
Estimating the predicted value by inputting the sensor information into the learned deep ensemble-based artificial neural network, and estimating an uncertainty value of the predicted value,
The step of estimating a side slip angle value by inputting the predicted value to a model-based observer,
Considering the predicted value and the uncertainty value as sensor information obtained from a virtual sensor, and estimating a side-slip angle value by inputting the predicted value and the uncertainty value to a model-based observer
A method for estimating a vehicle side slip angle, characterized in that. The method of claim 10,
The step of estimating a side slip angle value by inputting the predicted value to a model-based observer,
Estimating the side slip angle value by receiving the sensor information obtained from the vehicle and the predicted value and the uncertainty value output from the deep ensemble-based artificial neural network as input from the model-based observer
A method for estimating a vehicle side slip angle, characterized in that. The method of claim 10,
The step of estimating a side slip angle value by inputting the predicted value to a model-based observer,
The model-based observer automatically determines how much to trust and use the predicted value according to the uncertainty value of the predicted value.
A method for estimating a vehicle side slip angle, characterized in that. The method of claim 8,
The step of estimating a side slip angle value by inputting the predicted value to a model-based observer,
Estimating the side-slip angle using the model-based observer of a vehicle model-based Kalman filter method
A method for estimating a vehicle side slip angle, characterized in that. The method of claim 13,
The step of estimating a side slip angle value by inputting the predicted value to a model-based observer,
The model-based observer is a model-based observer of an Extended Kalman Filter (EKF) or an Unscented Kalman Filter (UKF) method.
A method for estimating a vehicle side slip angle, characterized in that. The method of claim 8,
Using the finally estimated side slip angle value in a vehicle active safety system
The vehicle side slip angle estimation method further comprising.

Images