KR102482968B1 - Method and Apparatus for Positioning Train Using Deep Kalman Filter - Google Patents

Abstract

A train positioning method and apparatus using a deep Kalman filter are disclosed.
According to one aspect of the present invention, in a train positioning method implemented by a computer and using a modified extended Kalman filter, the process of selecting two base stations; Observing a first Time Difference of Arrival (TDoA) for two radio signals received from a train through the two base stations; predicting a current state vector of the train from a previous state vector of the train; obtaining a second TDoA having a non-linear relationship with the predicted state vector; updating the predicted state vector according to a difference between the first TDoA and the second TDoA; and determining the position of the train from the updated state vector.

Description

Method and Apparatus for Positioning Train Using Deep Kalman Filter

Embodiments of the present invention relate to a train positioning method and apparatus using a modified Kalman filter and deep learning.

The information described in this section simply provides background information on the present invention and does not constitute prior art.

High Speed Train (HST) systems with a maximum travel speed of 300 km/h or more are widely used in many countries. Because people can travel to their destination faster, people can greatly improve their quality of life through the HST system. For example, with the introduction of the HST system in Korea in 2004, people can enjoy the 'half-day living zone', which allows them to move within half a day regardless of the origin and destination in Korea.

All trains generally depend on the distance information between the front train and the rear train, and this distance information is measured taking into account the safety margin. However, there is a problem in that if the distance between trains is excessively extended for safety, the number of available trains is reduced and operational efficiency is lowered. Accurate and reliable train positioning systems are needed to ensure the maximum number of trains available, i.e. to balance safety and efficiency.

Existing train positioning methods include a method of estimating the position of a train using a track circuit, a radio frequency tag (RF Tag), and a tachometer. In addition, there is a method of estimating the location of a train using a wireless communication system. Recently, a positioning method using a wireless communication system has been studied to be applied both indoors and outdoors. Examples include Bluetooth, Wi-Fi and Beacon for indoor use and Global Positioning System (GPS) or cellular systems for outdoor use.

While research on indoor positioning methods is actively conducted, studies on other outdoor positioning methods except GPS are insufficient. As a conventional outdoor positioning method, the LTE (Long-Term Evolution) system using OTDOA (Observed Time Difference Of Arrival) is difficult to guarantee a straight path (Line Of Sight; LOS), so errors occurring when estimating the train position is big On the other hand, since the 5G NR system not only satisfies a much higher LOS probability than the 4G LTE system, but also satisfies a much higher sampling rate, 5G-based train positioning methods are being actively studied.

When estimating the location of a train in a 5G system as well as a 4G system, base stations may use a Time Difference of Arrival (TDoA) of signals received from a train. In addition, when estimating the position of a train from TDoA, the positioning accuracy can be increased by using a Kalman filter. Here, the Kalman filter observes the position and speed of the train, predicts the position and speed of the train, and updates the current position and speed of the train using the observed and predicted data, that is, It is an estimation filter. Such a Kalman filter can be divided into a general Kalman filter assuming a linear-Gaussian system and an extended Kalman filter assuming a nonlinear-Gaussian system.

However, the actual environment is mostly a nonlinear-Gaussian environment, and in this nonlinear-non-Gaussian environment, there is a problem that the performance of the general Kalman filter and the extended Kalman filter is poor.

In particular, when using an extended Kalman filter in a nonlinear system, the existing positioning method linearizes the nonlinear relationship between observation information and prediction information about the train through first-order Taylor Series approximation, so positioning according to approximation error could not be reduced. In addition, when using the Kalman filter, the conventional positioning method assumes a Gaussian environment and does not consider a non-Gaussian environment, so there is a problem in that positioning accuracy is lowered.

Therefore, it is necessary to study a method for accurately estimating the train position even in a nonlinear-non-Gaussian system.

Embodiments of the present invention do not apply an approximation method for linearizing a non-linear relationship between observation information and prediction information of a train, but use a modified extended Kalman filter that uses the non-linear relationship as it is, so that the non-linear-non-Gaussian The main object is to provide a train positioning method and apparatus for increasing train positioning accuracy in a system.

Other embodiments of the present invention apply deep learning on the premise of a non-Gaussian environment in the update step of the modified extended Kalman filter, thereby improving train positioning accuracy in a nonlinear-non-Gaussian system (in particular, a non-Gaussian system). One object is to provide a method and apparatus.

According to one aspect of the present invention, in a train positioning method implemented by a computer and using a modified extended Kalman filter, the process of selecting two base stations; Observing a first Time Difference of Arrival (TDoA) for two radio signals received from a train through the two base stations; predicting a current state vector of the train from a previous state vector of the train; obtaining a second TDoA having a non-linear relationship with the predicted state vector; updating the predicted state vector according to a difference between the first TDoA and the second TDoA; and determining the position of the train from the updated state vector.

According to another aspect of the present embodiment, a train positioning device using a modified extended Kalman filter, comprising: a memory for storing instructions; and a processor that executes the instructions, wherein the processor selects two base stations by executing the instructions, and sets a first arrival time difference (Time Difference) for two radio signals received from a train through the two base stations. of Arrival (TDoA) is observed, the current state vector of the train is predicted from the previous state vector of the train, a second TDoA having a non-linear relationship with the predicted state vector is obtained, and the first TDoA and the second TDoA are obtained. Provided is a train positioning device that updates the predicted state vector according to the difference between and determines the position of the train from the updated state vector.

As described above, according to an embodiment of the present invention, a modified extended Kalman filter that uses the nonlinear relationship as it is, without applying an approximation method for linearizing the nonlinear relationship between observation information and prediction information of a train, By using it, it is possible to increase train positioning accuracy in a non-linear-non-Gaussian system.

According to another embodiment of the present invention, by applying deep learning on the premise of a non-Gaussian environment in the updating step of the modified extended Kalman filter, train positioning accuracy can be increased in a nonlinear-non-Gaussian system (in particular, a non-Gaussian system). .

1 is a configuration diagram illustrating components of a train positioning device according to an embodiment of the present invention.
2 is a diagram illustrating a process of selecting two base stations according to an embodiment of the present invention.
3A and 3B are diagrams illustrated to explain deep learning according to an embodiment of the present invention.
4 is a flowchart illustrating a train positioning method according to an embodiment of the present invention.

Hereinafter, some embodiments of the present invention will be described in detail through exemplary drawings. In adding reference numerals to components of each drawing, it should be noted that the same components have the same numerals as much as possible even if they are displayed on different drawings. In addition, in describing the present invention, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present invention, the detailed description will be omitted.

Also, terms such as first, second, A, B, (a), and (b) may be used in describing the components of the present invention. These terms are only used to distinguish the component from other components, and the nature, order, or order of the corresponding component is not limited by the term. Throughout the specification, when a part 'includes' or 'includes' a certain component, it means that it may further include other components without excluding other components unless otherwise stated. . In addition, terms such as '~unit' and 'module' described in the specification refer to a unit that processes at least one function or operation, and may be implemented by hardware, software, or a combination of hardware and software.

Hereinafter, the positioning device is preferably implemented as a server located outside the train, but is not limited thereto and may be implemented by a device, terminal, etc. located inside the train. The positioning device may store in advance at least one of a speed profile, base station identification information, location coordinates corresponding to the base station identification information, or train identification information. Here, the location coordinates refer to latitude and longitude or 2-dimensional or 3-dimensional coordinates based on a specific point.

1 is a configuration diagram illustrating components of a train positioning device according to an embodiment of the present invention.

Referring to FIG. 1 , a train positioning device 10 according to an embodiment of the present invention includes a memory 100, a processor 110, and a communication unit 120. However, not all of the components shown in FIG. 1 are essential components of the positioning device 10 . The positioning device 10 may be implemented with more components than those shown in FIG. 1, or the positioning device 10 may be implemented with fewer components than those shown in FIG.

For example, the positioning device 10 according to an embodiment may include a processor 110, a communication unit 120, and a memory 100, and the positioning device 10 according to an embodiment may include the processor 110 ), the communication unit 120, and the memory 100 may further include at least one of a user input unit and an output unit.

To estimate the position of the train, the positioning device 10 first selects two base stations. According to the embodiment of the present invention, since the train is a high speed train (HST), the train track can be treated as a straight track. Therefore, unlike existing positioning methods that require at least three base stations for triangulation, the positioning device 10 according to an embodiment of the present invention can estimate the position of a train using only two base stations.

A method for the positioning device 10 to select two base stations according to an embodiment of the present invention will be described in detail with reference to FIG. 2 .

The positioning device 10 observes a first time difference of arrival (TDoA) for two radio signals received from a train through two base stations. However, the positioning device 10 may use a received signal strength indicator (RSSI) in addition to TDoA.

Meanwhile, the positioning device 10 predicts the current state vector of the train from the previous state vector of the train. Here, the state vector of the train means information about the position and speed of the train. For example, when a train is located on a two-dimensional plane, the state vector of the train may mean the position and speed of the train on the x-axis. In addition, the previous state vector of the train means the state vector of the train updated by the modified extended Kalman filter in the previous iteration process. Meanwhile, the initial value of the state vector may be set by the user.

Specifically, the positioning device 10 may predict the current state vector of the train by applying a state transition matrix to the previous state vector and adding noise.

The positioning device 10 obtains a second TDoA having a non-linear relationship with the predicted state vector. The second TDoA is not an accurately known value, but a value estimated from the predicted state vector of the train. The second TDoA is used to update the train's predicted state vector.

At this time, the conventional positioning method linearizes the second TDoA having a non-linear relationship with the predicted state vector. The second TDoA is converted into a linear relationship with the predicted state vector through first-order Taylor series approximation. In detail, after converting the second TDoA into a Taylor series, all derivative terms of 3rd or higher order are removed and approximated with a maximum 2nd order derivative term. The approximated TDoA is applied to the extended Kalman filter to estimate the position of the train. At this time, due to approximation, the conventional positioning method has a problem in that positioning accuracy is lowered in a nonlinear environment.

On the other hand, according to an embodiment of the present invention, the second TDoA is not linearized from a non-linear relationship with the predicted state vector. The positioning device 10 according to an embodiment of the present invention uses the second TDoA that has a non-linear relationship with the predicted state vector without using approximation. Specifically, the second TDoA is converted into a Taylor series, but the higher order derivative is not removed and used to update the predicted state vector of the train.

Meanwhile, the positioning device 10 may associate the first TDoA with the second TDoA by including noise in the first TDoA and the second TDoA. In other words, the first TDoA is expressed as the sum of the second TDoA and the noise.

The positioning device 10 updates the predicted state vector according to the difference between the first TDoA and the second TDoA. Specifically, the positioning device 10 obtains a difference between the first TDoA and the second TDoA, and further obtains a Kalman gain. The positioning device 10 generates a correction vector including a difference between the first TDoA and the second TDoA and a Kalman gain. The positioning device 10 updates the predicted state vector using the correction vector. The updated change amount of the predicted state vector increases as the Kalman gain or the difference between the first TDoA and the second TDoA increases.

The positioning device 10 according to an embodiment of the present invention may update the predicted state vector using deep learning. First, the positioning device 10 generates a correction vector by multiplying the difference between the first TDoA and the second TDoA by the Kalman gain. The positioning device 10 obtains an output vector by inputting the correction vector and the predicted state vector to the neural network on which learning has been completed. The positioning device 10 determines the output vector as an updated state vector for the predicted state vector.

Here, the trained neural network is trained to output the first output vector according to the speed profile of the train when receiving the first correction vector and the first predicted state vector. The first correction vector and the first predicted state vector are data used in the learning step, and are distinguished from the correction vector and the predicted state vector input to the neural network after learning has been completed. On the other hand, the speed profile is information representing the relationship between the speed and position of the train. In the case of a train, when the speed of a train located between a departure point and a destination is a predetermined value, location information corresponding to the predetermined value may be stored in advance. This will be described in detail with reference to FIG. 3A.

According to an embodiment of the present invention, when the positioning device 10 uses a neural network, each of the predicted state vector and the first TDoA may include non-Gaussian noise. If the predicted state vector and the first TDoA include non-Gaussian noise, the correction vector also includes non-Gaussian noise. At this time, when the positioning device 10 generates an updated state vector by simply adding the predicted state vector and the correction vector, positioning accuracy may be degraded due to non-Gaussian noise. However, when the positioning device 10 uses a neural network, even if the predicted state vector and correction vector are affected by non-Gaussian noise, it can be minimized. That is, the positioning device 10 can prevent deterioration in positioning accuracy even in a non-Gaussian environment.

Finally, the positioning device 10 determines the position of the train from the updated state vector.

Through the above operation, the positioning device 10 can accurately estimate the position of the train even in a nonlinear system and a non-Gaussian system. In addition, the positioning device 10 can improve the accuracy of the position even in a linear system or a Gaussian system through the above-described operation.

Hereinafter, a process in which the positioning device 10 uses the modified extended Kalman filter will be described through equations. To describe the process of applying the modified extended Kalman filter, it can be divided into a prediction step and an update step. The prediction step includes predicting the current state of the train from the previous state and observing the current state of the train. The update step is a step of estimating the next state of the train using prediction information and observation information.

The train's state vector includes the train's position and speed. The state vector of the train can be expressed as Equation 1.

In this case, in the case of a high-speed train, all values related to the y-axis may be 0. In Equation 1, n is a discontinuous sampling time, and is used to indicate the nth data of the train.

Meanwhile, in the prediction step, the state vector predicted from the previous state vector of the train and the covariance matrix predicted from the previous error covariance matrix can be expressed as Equation 2.

는 예측된 상태벡터,

는 이전 상태벡터, F는 상태이전행렬,

은 예측된 오차 공분산행렬,

은 이전 오차 공분산행렬, Q는 예측 노이즈에 대한 오차 공분산 행렬을 의미한다. 예측된 오차 공분산 행렬은 수학식 4에서

의 제곱의 평균으로 정의될 수 있다. 예측 노이즈는 평균이 0인 비가우시안 노이즈(또는 가우시안 노이즈)이고, 오차 공분산행렬은 예측 노이즈의 제곱 평균을 의미한다. 이전 상태벡터와 이전 공분산 행렬은 이전 단계에서 갱신된 정보들이다.in Equation 2

is the predicted state vector,

is a previous state vector, F is a previous state matrix,

is the predicted error covariance matrix,

is the previous error covariance matrix, and Q is the error covariance matrix for prediction noise. The predicted error covariance matrix is

can be defined as the average of the squares of The prediction noise is non-Gaussian noise (or Gaussian noise) with an average of 0, and the error covariance matrix means the square mean of the prediction noise. The previous state vector and the previous covariance matrix are information updated in the previous step.

In the prediction step, the relationship between the first TDoA and the second TDoA can be expressed as Equation 3.

은 기지국을 통해 구한 제1 TDoA,

는 예측된 상태벡터로부터 추정한 제2 TDoA,

은 관측 노이즈(measurement noise)이다.

는 예측 단계와 갱신 단계 사이의 오차이고, H는 비선형 측정 함수의 야코비안 행렬(Jacobian matrix)이다. N은 기지국의 인덱스다.in Equation 3

is the first TDoA obtained through the base station,

Is a second TDoA estimated from the predicted state vector,

is the measurement noise.

is the error between the prediction step and the update step, and H is the Jacobian matrix of the nonlinear measurement function. N is the index of the base station.

In Equation 3, the average of the observation noise is zero. Also, h is a non-linear measurement function. Existing positioning methods linearize or linearly approximate a non-linear measurement function by ignoring terms in which the value of k is 3 or greater, whereas the positioning method of the present invention does not linearize or approximate.

The second TDoA can be expressed as Equation 4.

는 열차의 위치 좌표,

는 상호상관(cross-correlation)이 두 번째로 큰 기지국의 위치 좌표,

는 기준 기지국의 위치 좌표다. 제2 TDoA는 수학식 4에서 시간의 역수로 표현되지만, 시간으로 표현될 수도 있다.in Equation 4

is the location coordinate of the train,

Is the location coordinate of the base station with the second largest cross-correlation,

Is the location coordinate of the reference base station. The second TDoA is expressed as the reciprocal of time in Equation 4, but may also be expressed as time.

Meanwhile, in the updating step, the updated state vector of the train can be expressed as Equation 5.

는 갱신된 상태벡터,

는 칼만 이득(Kalman gain),

는 예측 단계와 갱신 단계 사이의 오차이고, H는 비선형 측정 함수의 야코비안 행렬(Jacobian matrix)이다. 칼만 이득과 갱신된 오차 공분산행렬은 수학식 6과 같이 표현될 수 있다.in Equation 5

is the updated state vector,

is the Kalman gain,

is the error between the prediction step and the update step, and H is the Jacobian matrix of the nonlinear measurement function. The Kalman gain and the updated error covariance matrix can be expressed as Equation 6.

Referring back to FIG. 1 , the processor 110 typically controls overall operations of the positioning device 10 and processes data. For example, the processor 110 may generally control the communication unit 120, the user input unit, and the output unit by executing programs stored in the memory 100. Also, the processor 110 may perform the functions of the positioning device 10 by executing programs stored in the memory 100 . The processor 110 may include at least one processor. Depending on its function and role, the processor 110 may include a plurality of processors or may include a single integrated processor. In one embodiment, the processor 110 may include at least one processor that provides a notification message by executing at least one program stored in the memory 100 . In one embodiment, the processor 110 may obtain a signal transmitted by the train through communication between the communication unit 120 and the base station.

The communication unit 120 may include one or more components that enable the positioning device 10 to communicate with other electronic devices (not shown). Other electronic devices (not shown) may be computing devices or sensing devices, but are not limited thereto. Also, for example, other electronic devices may be modules included in the train, such as the positioning device 10 . For example, the communication unit 120 may include a short-distance communication unit, a mobile communication unit, and a broadcast reception unit.

The short-range wireless communication unit includes a Bluetooth communication unit, a Bluetooth Low Energy (BLE) communication unit, a Near Field Communication unit (WLAN) communication unit, a Zigbee communication unit, and an infrared data (IrDA) communication unit. Association) communication unit, WFD (Wi-Fi Direct) communication unit, UWB (ultra wideband) communication unit, Ant+ communication unit, etc. may be included, but is not limited thereto.

2 is a diagram illustrating a process of selecting two base stations according to an embodiment of the present invention.

Referring to FIG. 2 , a train 200 , a plurality of base stations 210 , 211 , 212 , 213 , and 214 , and a candidate base station 220 are shown. The train 200 is running in the northeast direction, and since the track is treated as a straight line in the High Speed Train (HST) system, it will be described that the train 200 is running in a straight line. The plurality of base stations 210, 211, 212, 213, and 214 have the same vertical distance from the straight line and may be arranged in a zigzag pattern. However, this is only one embodiment, and base stations may be variously arranged.

The candidate base station 220 means at least three base stations located before and after the reference base station (reference RRH) having the largest cross-correlation in the previous iteration. Cross-correlation means a correlation between radio signals received by base stations from the train 200 and respective known Sounding Reference Signals (SRSs). In FIG. 2, the second base station 211 is the reference base station having the largest cross-correlation in the previous iteration, and the candidate base stations 220 are the first base station 210, the second base station 211, and the third base station 212. . At this time, the candidate base station 220 is not necessarily limited to three base stations, and may include at least three base stations.

In the current iteration, a positioning device (not shown) obtains a cross-correlation for a candidate base station 220 among a plurality of base stations 210, 211, 212, 213, and 214.

Then, the positioning device selects two base stations having the largest cross-correlation within the candidate base station 220 . Signals received by the two selected base stations are used to estimate the location of the train 200. A base station having the largest cross-correlation within the candidate base station 220 becomes a reference base station.

According to an embodiment of the present invention, the positioning device may select two base stations without considering the candidate base station 220 . Specifically, the positioning device receives SRS radio signals received from the train 200 by the plurality of base stations 210, 211, 212, 213, and 214. The positioning device may select a base station having the largest cross-correlation among the plurality of base stations 210, 211, 212, 213, and 214 as a reference base station, and further select a base station having the second highest cross-correlation. The positioning device may estimate the position of the train 200 using signals received by the two selected base stations.

3A and 3B are diagrams illustrated to explain deep learning according to an embodiment of the present invention.

Referring to Figure 3a, the speed profile of the train is shown. The speed profile is information that records the speed of the train and the corresponding time. Velocity profiles are required to be used as label data for deep learning.

In general, trains, unlike other means of transportation, have a fixed starting point and a destination, and a fixed speed at each instant of movement. Through the speed profile, it is possible to know at what speed the train is and how long it has passed since the departure time. For example, if a train is running at 450 km/h, the train is about 2300 time units past its departure time. Here, the time unit means Estimated Time Samples in FIG. 3A, which may vary in various ways according to a sampling rate. However, the unit of time may be replaced by the location of the train. That is, the speed profile may record the position and speed of the train in advance.

Referring to FIG. 3B, a first predicted state vector 300, a first correction vector 302, a label vector 304, a neural network 310, and a first output vector 320 are shown.

In a non-Gaussian environment, since the predicted state vector or correction vector of a train includes non-Gaussian distribution noise, when an updated state vector is generated by simply calculating the predicted state vector and the correction vector by four arithmetic operations, the positioning accuracy deteriorates. Therefore, the positioning device according to an embodiment of the present invention inputs the predicted state vector and the correction vector to the trained neural network and obtains the updated state vector, thereby preventing the positioning accuracy from deteriorating even in a non-Gaussian environment. there is.

The positioning device may train the neural network 310 in advance before estimating the location of the train. The neural network 310 may be supervised through deep learning.

The positioning device uses the first predicted state vector 300, the first correction vector 302, and the label vector 304 as learning data. The first predicted state vector 300 means a state vector predicted through a state transition matrix from a previous state vector of the train. The first correction vector 302 means a vector obtained by multiplying the difference between the first TDoA and the second TDoA by the Kalman gain according to the modified extended Kalman filter. The first correction vector 302 is used to update the first predicted state vector 300 . On the other hand, the label vector 304 means the position and speed of the train according to the speed profile as shown in FIG. 3A.

The positioning device labels the first predicted state vector 300 and the first correction vector 302 with the label vector 304 and then inputs the label to the neural network 310 . The positioning device obtains the first output vector 320 output from the neural network 310 and compares the first output vector 320 with the label vector 304 . The positioning device indirectly adjusts parameters of the neural network 310 through a loss function so that the first output vector 320 approaches the label vector 304 .

When the similarity between the first output vector 320 and the label vector 304 satisfies a predetermined condition, learning of the neural network 310 is completed.

Later, the positioning device may obtain an updated state vector of the train by inputting the predicted state vector and the correction vector of the train to the neural network 310 on which learning has been completed.

4 is a flowchart illustrating a train positioning method according to an embodiment of the present invention.

To estimate the position of the train, the positioning device first selects two base stations (S400). According to an embodiment of the present invention, the positioning device may select two base stations through the operation according to FIG. 2 .

The positioning device observes a first Time Difference of Arrival (TDoA) for the two radio signals received from the train through the two base stations (S402). However, the positioning device may use a received signal strength indicator (RSSI) in addition to TDoA.

Meanwhile, the positioning device predicts the current state vector of the train from the previous state vector of the train (S404). Here, the state vector of the train means information about the position and speed of the train. The previous state vector of the train means the state vector of the train updated by the modified extended Kalman filter in the previous iteration process. Specifically, the positioning device may predict the current state vector of the train by applying a state transition matrix to the previous state vector and adding noise.

The positioning device obtains a second TDoA having a non-linear relationship with the predicted state vector (S406). The second TDoA is not an accurately known value, but a value estimated from the predicted state vector of the train. The second TDoA is used to update the train's predicted state vector.

At this time, the conventional positioning method linearizes the second TDoA having a non-linear relationship with the predicted state vector. The second TDoA is converted into a linear relationship with the predicted state vector through first-order Taylor series approximation. At this time, due to approximation, the conventional positioning method has a problem in that positioning accuracy is lowered in a nonlinear environment. On the other hand, according to an embodiment of the present invention, the second TDoA is not linearized from a non-linear relationship with the predicted state vector. The positioning device according to an embodiment of the present invention uses the second TDoA, which has a non-linear relationship with the predicted state vector, without using approximation. Specifically, the second TDoA is converted into a Taylor series, but the higher order derivative is not removed and used to update the predicted state vector of the train.

The positioning device updates the predicted state vector according to the difference between the first TDoA and the second TDoA (S408). Specifically, the positioning device obtains a difference between the first TDoA and the second TDoA, and further obtains a Kalman gain. The positioning device generates a correction vector including a difference between the first TDoA and the second TDoA and a Kalman gain. The positioning device updates the predicted state vector using the correction vector. The updated change amount of the predicted state vector increases as the Kalman gain or the difference between the first TDoA and the second TDoA increases.

The positioning device according to an embodiment of the present invention may update a predicted state vector using deep learning. First, the positioning device generates a correction vector by multiplying the difference between the first TDoA and the second TDoA by a Kalman gain. The positioning device acquires an output vector by inputting a correction vector and a predicted state vector to a neural network on which learning has been completed. The positioning device determines the output vector as an updated state vector for the predicted state vector. The neural network may be trained through a learning process according to FIGS. 3A and 3B.

According to an embodiment of the present invention, when the positioning device uses a neural network, each of the predicted state vector and the first TDoA may include non-Gaussian noise. If the predicted state vector and the first TDoA include non-Gaussian noise, the correction vector also includes non-Gaussian noise. In this case, when the positioning device simply adds the predicted state vector and the correction vector to generate an updated state vector, positioning accuracy may be degraded due to non-Gaussian noise. However, when the positioning device uses a neural network, even if the predicted state vector and correction vector are affected by non-Gaussian noise, it can be minimized. That is, the positioning device can prevent deterioration in positioning accuracy even in a non-Gaussian environment.

Finally, the positioning device determines the position of the train from the updated state vector (S410).

Through the above operation, the positioning device can accurately estimate the position of the train even in a nonlinear system and a non-Gaussian system. In addition, the positioning device can improve the accuracy of the position even in a linear system or a Gaussian system through the above-described operation.

Although it is described in FIG. 4 that steps S400 to S410 are sequentially executed, this is merely an example of the technical idea of an embodiment of the present invention. In other words, those skilled in the art to which an embodiment of the present invention belongs may change and execute the sequence described in FIG. 4 without departing from the essential characteristics of the embodiment of the present invention, or one of steps S400 to S410. Since it will be possible to apply various modifications and variations by executing the above process in parallel, FIG. 4 is not limited to a time-series sequence.

Meanwhile, the processes shown in FIG. 4 can be implemented as computer readable codes on a computer readable recording medium. A computer-readable recording medium includes all types of recording devices in which data that can be read by a computer system is stored. That is, such a computer-readable recording medium may be a non-transitory medium such as ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc., and may also be a carrier wave (e.g. , transmission over the Internet) and a transitory medium such as a data transmission medium. In addition, the computer-readable recording medium may be distributed to computer systems connected through a network to store and execute computer-readable codes in a distributed manner.

In addition, components of the present invention may use an integrated circuit structure such as a memory, a processor, a logic circuit, a look-up table, and the like. These integrated circuit structures execute each of the functions described herein through the control of one or more microprocessors or other control devices. In addition, the components of the present invention may be specifically implemented by a program or part of code that includes one or more executable instructions for performing a specific logical function and is executed by one or more microprocessors or other control devices. In addition, the components of the present invention may include or be implemented by a central processing unit (CPU), a microprocessor, etc. that perform each function. Also, components of the present invention may store instructions executed by one or more processors in one or more memories.

The above description is merely an example of the technical idea of the present embodiment, and various modifications and variations can be made to those skilled in the art without departing from the essential characteristics of the present embodiment. Therefore, the present embodiments are not intended to limit the technical idea of the present embodiment, but to explain, and the scope of the technical idea of the present embodiment is not limited by these embodiments. The scope of protection of this embodiment should be construed according to the claims below, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of rights of this embodiment.

100: memory 110: processor
120: communication department

Claims (15)

In the train positioning method implemented by a computer and using a modified extended Kalman filter,
process of selecting two base stations;
Observing a first Time Difference of Arrival (TDoA) for two radio signals received from a train through the two base stations;
predicting a current state vector of the train from a previous state vector of the train;
obtaining a second TDoA having a non-linear relationship with the predicted state vector;
updating the predicted state vector according to a difference between the first TDoA and the second TDoA; and
Process of determining the position of the train from the updated state vector
A train positioning method comprising a. According to claim 1,
The selection process is
obtaining a cross-correlation between a received signal and a known reference signal for each of at least three base stations; and
Selecting two base stations having the largest cross-correlation among the at least three base stations
A train positioning method comprising a. According to claim 2,
The at least three base stations,
A train positioning method that is at least three base stations located before and after the reference base station with the largest cross-correlation in the previous iteration. According to claim 1,
The updating process is
generating a correction vector by multiplying a difference between the first TDoA and the second TDoA by a Kalman gain;
obtaining an output vector by inputting the correction vector and the predicted state vector to the neural network after learning has been completed; and
Determining the output vector as the updated state vector for the predicted state vector
A train positioning method comprising a. According to claim 4,
The trained neural network,
The train positioning method which is learned to output a first output vector according to a speed profile of the train when the first correction vector and the first predicted state vector are received. According to claim 5,
The predicted state vector and the first TDoA each include non-Gaussian noise. According to claim 1,
The second TDoA,
A train positioning method that is not linearized from a non-linear relationship with the predicted state vector. In the train positioning device using the modified extended Kalman filter,
memory for storing instructions; and
a processor that executes the instructions;
By executing the instructions, the processor
Choose two base stations,
Observing a first Time Difference of Arrival (TDoA) for two radio signals received from a train through the two base stations;
Predicting the current state vector of the train from the previous state vector of the train;
Obtaining a second TDoA in a non-linear relationship with the predicted state vector;
Updating the predicted state vector according to the difference between the first TDoA and the second TDoA;
A train positioning device for determining the position of the train from the updated state vector. According to claim 8,
By executing the instructions, the processor
Obtaining a cross correlation between a received signal and a known reference signal for each of the at least three base stations;
and selecting two base stations having the largest cross-correlation among the at least three base stations as the two base stations. According to claim 9,
The at least three base stations,
A train positioning device that is at least three base stations located before and after the reference base station with the highest cross-correlation in the previous iteration. According to claim 8,
By executing the instructions, the processor
A correction vector is generated by multiplying a difference between the first TDoA and the second TDoA by a Kalman gain;
An output vector is obtained by inputting the correction vector and the predicted state vector to the neural network after learning has been completed,
and determining the output vector as the updated state vector for the predicted state vector. According to claim 11,
The trained neural network,
The train positioning device is learned to output a first output vector according to a speed profile of the train when the first correction vector and the first predicted state vector are received. According to claim 12,
Each of the predicted state vector and the first TDoA includes non-Gaussian noise. According to claim 13,
The second TDoA,
A train positioning device that is not linearized from a non-linear relationship with the predicted state vector. A computer program stored in a medium to execute the train positioning method according to any one of claims 1 to 7.

Images