KR20200086443A - Gnss integrity monitoring method based on machine learning for satellite navigation augmentation system - Google Patents

Abstract

Disclosed is a proposal to improve anomalous behavior of a time-delay neural networks (TDNN)-based global navigation satellite system (GNSS) signal augmentation system. According to one embodiment, an integrity monitoring method comprises the steps of: generating a TDNN-based model by utilizing a calibration statistic calculated for a GNSS signal through receiver autonomous integrity monitoring; and detecting anomalous behavior of the GNSS as the GNSS signal received through the generated model is learned.

Description

GNSS INTEGRITY MONITORING METHOD BASED ON MACHINE LEARNING FOR SATELLITE NAVIGATION AUGMENTATION SYSTEM for machine learning-based GNSS integrity monitoring for satellite navigation enhancement systems

The following description relates to a machine learning-based GNSS integrity monitoring method for a satellite navigation enhancement system.

Where strict performance operations are required to use satellite navigation systems, such as civil aviation or intelligent transportation systems (ITSs) (e.g. advanced driver assistance systems, including electronic toll collection, route guidance, collision avoidance systems, and intelligent speed control systems) In, the integrity of GNSS plays an important role, and its monitoring is required. In these requisite applications, the reliability of the location extracted from the navigation data must be considered. This reliability measure is also referred to as the integrity of the navigation system, and is also a function that notifies the user when the use of the navigation system is undesirable. Unacceptable position deviation can significantly affect performance by significantly affecting positioning and navigation functions, so it is necessary to activate and monitor integrity before detecting and excluding abnormal signal sources before measuring the user's position.

GNSS provides users with basic integrity information, such as satellite status, through navigation messages. However, this integrity information is eventually delayed by not obtaining a sufficiently fast transfer rate in applications requiring a high level of security. In applications that require a high level of security, this message transfer is of little significance in integrity monitoring, because it usually takes 15 minutes to hours to deliver a message to the user that the delay is an issue. Instead, users can determine integrity using receiver automatic integrity monitoring algorithms (RAIMs) that have been used in various application fields for years, or ground-based reinforcement systems (GBAS), such as external integrity data sources from GNSS reinforcement systems, Alternatively, a satellite-based reinforcement system (SBAS) may be used to check the integrity of the measured location information through a reference receiver. The basic concept of RAIM is to detect and exclude errors through double observation, thereby increasing the reliability of the measured location information to ensure quality. The consistency of the redundant measurements provides clues as to whether the navigation satellite is operating outside its normal range and whether location information calculated from these errors is available.

However, most research related to RAIM has focused primarily on non-precision access (NPA), as it can be problematic when RAIM is used for certain services, such as access flight (APV) under vertical guidance (to NPA). The alarm limit for this is fairly moderate, but it is quite strict for APV). In the APV service, even when the system is marked as available, the positional error exceeds the alarm limit, so that false information (HMI) at the risk level may occur, so the alarm limit is low. A typical RAIM cannot be concluded that the HMI is always identified with high probability to ensure integrity requirements. In addition, recent studies have focused more on improving the availability of RAIM in multi-satellite GNSS. Rather than improving the integrity level of RAIM, it is being developed to improve progress and improve performance. Autonomous vehicles for ITS implementation do not require a large separation distance to compensate for the limitations of the accuracy of navigation, so the safety level can be greatly increased. These GNSS-based self-driving technologies are limited by stringent safety regulations, such as the required required navigation performance (RNP), which is determined by the accuracy and integrity requirements for a particular mode of operation. RNP was originally developed for aviation navigation and has since been extended to ground vehicle navigation. Therefore, in order to avoid these constraints, there is a need for an alternate integrity enhancement mechanism that detects fault sources exceeding a pre-defined strict safety limit and safely alerts unmanned/unmanned ground systems.

Modeling the test statistic calculated through RAIM using a non-linear time delay neural network, extracting useful information about defect characteristics, expressed as an irregular sequence as a residual range calculated through the observed GNSS signal, New methods and systems can be provided to more effectively capture anomalous behavior.

The integrity monitoring method generates a time-delay neural network-based model by utilizing the calculated test statistics for the Global Navigation Satellite System (GNSS) signal through Receiver Autonomous Integrity Monitoring. step; And detecting an anomalous behavior of the GNSS as learning the GNSS signal received through the generated model.

)을 통하여 계산하는 단계를 포함할 수 있다. The step of generating a model based on the time-delay neural networks comprises summing the residuals of the black statistic by quantifying self-consistency of the GNSS overlap measurement of the RAIM performing the integrity monitoring. Based on Equation 1(

).

The step of generating a model based on the time-delay neural networks includes a repetitive structure having a delay factor and a non-repetitive structure based on a feedforward direction delay element for storing time information in the model. And applying the structure.

In the step of generating a model based on the time-delay neural networks, the time-delayed neural network is used to capture anomalous behavior of the GNSS through a change in a test metric for learning a time dependency beyond a predetermined criterion. And adopting an underlying model.

The step of detecting the anomalous behavior of the GNSS may include extracting information about a binding characteristic expressed in an irregular sequence as a residual range calculated through the received GNSS signal.

The step of detecting the anomalous behavior of the GNSS is based on the sum of squares of the residual range error calculated by the standard RAIM in order to keep the time series on the input of the time delay neural network in the test statistics of the received GNSS signal. It may include the step of changing to a delayed form.

The step of detecting the anomalous behavior of the GNSS may include generating a newly proposed test statistic as a result of learning the time delay neural network.

The step of detecting anomalous behavior of the GNSS measures the difference between an offline pseudo-distance-based estimation output and an online pseudo-distance-based estimation output through a correlation of the newly proposed time series of the test statistics, and then the newly proposed And comparing the test statistic with a threshold of Equation 1 to determine whether an alarm has occurred.

The step of detecting the anomalous behavior of the GNSS is estimated based on the estimated output and the online pseudo-distance by learning the time-delayed neural network for the received GNSS signal based on the offline pseudo-distance obtained when the GNSS is in a normal state. And generating an alarm based on the difference between the outputs.

The integrity monitoring system generates a time-delay neural network-based model by using the calculated test statistics for the Global Navigation Satellite System (GNSS) signal through Receiver Autonomous Integrity Monitoring. Model generator; And a sensing unit configured to detect anomalous behavior of the GNSS by learning the GNSS signal received through the generated model.

Rather than simply operating a single RAIM, useful information about defect characteristics can be extracted using a time-delayed neural network.

It is possible to detect defects in the navigation satellites at a significantly faster rate than when operating a RAIM alone.

Through RAIM using the TDNN-based integrity monitoring algorithm, it is possible to improve the level of RAIM integrity assurance in terms of GNSS anomaly detection.

1 is a block diagram illustrating the configuration of an integrity verification system according to an embodiment.
2 is a flowchart illustrating a machine learning-based GNSS signal integrity monitoring method for a satellite navigation enhancement system in an integrity verification system according to an embodiment.
3 shows the structure of a time delay neural network (TDNN) of an integrity verification system according to an embodiment.
4 is a flowchart illustrating a method of performing integrity monitoring based on TDNN estimation measurement values in an integrity verification system according to an embodiment.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings.

1 is a block diagram for explaining the configuration of an integrity verification system according to an embodiment, and FIG. 2 is a machine learning-based GNSS signal integrity monitoring method for a satellite navigation enhancement system in an integrity verification system according to an embodiment It is a flowchart for.

The processor of the integrity monitoring system 100 may include a model generation unit 110 and a detection unit 120. The components of the processor may be representations of different functions performed by the processor according to a control instruction provided by at least one program code. The processor and its components may perform steps 210 to 220 included in the machine learning-based GNSS signal integrity monitoring method for the satellite navigation enhancement system of FIG. 2. For example, the processor and the components of the processor may be implemented to execute instructions according to at least one program code and the code of the operating system included in the memory. Here, at least one program code may correspond to the code of a program implemented to process a machine learning-based GNSS signal integrity monitoring method for a satellite navigation enhancement system. The machine learning-based GNSS signal integrity monitoring method for the satellite navigation enhancement system may not occur in the order shown, and some of the steps may be omitted or additional processes may be further included.

The processor may load program code stored in a program file for a machine learning-based GNSS signal integrity monitoring method for a satellite navigation enhancement system into a memory. At this time, each of the processor and the model generation unit 110 and the detection unit 120 included in the processor executes instructions of corresponding portions of the program codes loaded in the memory to execute subsequent steps 210 to 220 It can be different functional expressions of.

In step 210, the model generator 110 generates a time-delay neural network based model by using the calculated test statistics for the GNSS signal through receiver autonomous integrity monitoring. can do. The model generator 110 may calculate a test statistic based on the sum of squared residual distance errors by quantifying self-consistency of the GNSS overlapping measurement of the RAIM performing the integrity monitoring. The model generator 110 may apply a non-repetitive structure based on a feedforward direction delay element and a repetitive structure with delayed feedback to store time information in the model. The model generator 110 may adopt a model based on a time delay neural network to capture anomalous behavior of the GNSS through a change in a test metric for learning a time dependency beyond a predetermined criterion.

In step 220, the sensing unit 120 may detect anomalous behavior of the GNSS by learning the GNSS signal received through the generated model. The sensing unit 120 may extract information on a coupling characteristic expressed in an irregular sequence as a residual range calculated through the received GNSS signal. The detector 120 may change the test statistic based on the sum of squares of the residual range error calculated by the standard RAIM in a time delayed form so that the time series in the received statistic of the GNSS signal remains at the input of the time delay neural network. . The sensor 120 may generate a newly proposed test statistic as a result of learning a time delay neural network. The detector 120 may check whether an alarm has occurred by comparing the newly proposed test statistics with the thresholds of the test statistics calculated through the conventional method. The sensing unit 120 is based on the difference between the estimated output based on the online pseudo-range and the estimated output as learning the time-delay neural network of the received GNSS signal based on the offline pseudo-range obtained when the GNSS is in a normal state. To generate an alarm.

Hereinafter, in order to more clearly distinguish anomalous behavior from the observed GNSS signal, rather than simply operating a single RAIM, the useful information about the defect characteristics is extracted faster by using a time-delayed neural network, and the GNSS anomaly behavior is detected. In the following, a method of improving the integrity guarantee level of RAIM will be described in detail.

를 n(n은 자연수)개의 가시적 항행 위성의 내비게이션 메시지에서 획득한 의사 거리(pseudo-range) 의 측정값이라 할 때, 선형화를 통한 사용자의 위치 정보 간 기본 측정 공식과 예측 및 관측 측정값의 차이를 수학식 1과 같은 선형 방정식으로 나타낼 수 있다.The integrity verification system can model the GNSS signal.

When n is a measure of the pseudo-range obtained from navigation messages of n (n is a natural number) visible navigation satellites, the difference between the basic measurement formula and the predicted and observed measurements between the user's location information through linearization Can be represented by a linear equation such as Equation 1.

Equation 1:

는 공칭(nomimal) 사용자의 위치 정보와 시계 오차에 근거한 예측 범위와 측정 범위간의 차이로 형성된 측정 잔차(measurement residual)이다. 또한,

는 공칭 사용자의 위치 정보와 시계 오차에 대한 선형화로 형성된 관측성 행렬(observation matrix)인 반면,

는 공칭 사용자의 위치 정보에 대한 실제 위치 정보의 편차에 사용자 시계 오차를 더한 3개의 구성요소를 포함하는 내비게이션 편차 상태(navigation error state)이다. 마지막으로,

는 일반적인 수신기 잡음, 위성 위치에 대한 부정확한 정보, 위성 시계 오류 및 위성 고장으로 인한 예상치 못한 오류 등으로 발생하는 측정 오차이다. 이러한 오차는, 평균이 0이고 공분산

를 가지는, 표준 가우시안 분포의 형태를 취한다. 대부분의 경우 최소 4개 이상의 위성을 볼 수 있다는 점을 고려할 때, 중복 의사 거리 정보를 사용할 수 있으며, 위치 정보 추정은 과결정된(over-determined) 데이터의 최적으로부터 획득될 수 있다. 상수 k 상태에서의 상태 최소 제곱은 수학식 2와 같이 추정된다. here

Is a measurement residual formed by the difference between the predicted range and the measured range based on the nominal user's location information and clock error. Also,

Is an observation matrix formed by linearization of the nominal user's location information and clock error,

Is a navigation error state including three components that add the user's clock error to the deviation of the actual position information from the nominal user's position information. Finally,

Is a measurement error caused by general receiver noise, inaccurate information about satellite position, satellite clock error, and unexpected error due to satellite failure. These errors have a mean of 0 and covariance

It takes the form of a standard Gaussian distribution. Considering that at least four satellites can be viewed in most cases, redundant pseudo-distance information can be used, and location information estimation can be obtained from the optimization of over-determined data. The state least squares in the constant k state is estimated as in Equation 2.

Equation 2:

는

의 슈도-역행렬(pseudo inverse)을 나타내며(다시 말해서,

),

의 추정은

이며, 잔차 거리의 오차는

로 정의된다. RAIM은 무결성 감시를 위한 가장 쉽고 비용 효율적인 아키텍처로, 중복 측정에 대한 자기-일관성을 수령화(quantifying)한다. 이러한 검정통계량은 수학식 3과 같이 잔차 거리 오차 제곱합을 기반으로 기술된다.here,

The

Represents the pseudo-inverse of the pseudo (in other words,

),

The estimate of

And the error of the residual distance

Is defined as RAIM is the easiest and most cost-effective architecture for integrity monitoring, quantifying self-consistency for redundant measurements. This test statistic is described based on the sum of squared residual distance errors as shown in Equation (3).

Equation 3:

)은, 상수 k에서의

누적 분포 함수,

와

의 목표값으로 바로 계산할 수 있다(

). 잔차 거리 오차가

의 2차 형태이기에, 오경보의 가능성(다시 말해서,

)은 카이제곱분포 형태로 나타나고, 정상 랜덤 측정 잡음(normal random measurement noise)

에 비례한다. 변칙행동으로 손상된 위성이 발견되면, 해당 위성을 제거하여 원활한 내비게이션을 확보해야 한다. Typical RAIM thresholds (i.e.

) At constant k

Cumulative distribution function,

Wow

It can be calculated as the target value of (

). Residual distance error

Because it is a secondary form of the possibility of false alarms (in other words,

) Appears in the form of a chi-square distribution, and normal random measurement noise

Proportional to. If a satellite damaged by anomalous behavior is found, the satellite must be removed to ensure smooth navigation.

The integrity verification system will be described in detail how to perform anomalous behavior detection through a TDNN-based model. Many applications need to process time information, such as trends in a specific time window, time series up to a specific time, and so on. Common tasks dealing with time information may include prediction, classification of time series, or mapping with other time series. Long-term behavior can be modeled using neural networks applied to compensate for time delays in nonlinear processes. The neural network can apply two types of mechanisms to store time information internally (that is, a non-repetitive structure based on a feedforward direction delay element and an iterative structure with delay feedback). Unlike biological neural networks, connections between artificial neural cells are usually not added or removed after the network is created. Instead, the connection is weighted, which is adjusted by the learning algorithm. The input signal propagates through the network in the direction of the connection until it reaches the output of the network. Under supervised learning, the learning algorithm adjusts the weights to minimize discrepancies between the network output and the target values provided.

Cyclic neural network (RNN) architectures such as Long Short Term Memory (LSTM) have been shown to effectively learn the temporal dynamics of signals, but training has become more complex due to the increased complexity of the network. Another weakness of RNN is that the deeper the structure, the more difficult it is to learn because the gradient must be optimized across several hidden states. Compared to iterative structures in this respect, feedforward learning is relatively fast and more efficient in modeling simple temporal dynamics. The time-delay neural network, originally designed for speech recognition, excels when modeling dynamic systems with large time delays. These are modified feedforward networks designed to capture the dynamics of modeled processes. The feedforward network is insufficient to process the time sequence, since there is no internal memory to store past information. To overcome this limitation, past memories are introduced as a way to extend the network's input using delay lines as synapses.

는 sum function에 의해 시냅스 중량 w와 함께 포착 및 처리되며, 새로이 활성화된 뉴런의 activation potential은 도 3과 같이 활성 함수(activation function)에 의해 획득될 수 있다. 이후의 네트워크 출력은 수학식 4와 같이 주어진다. For this reason, there are various neural network structures for learning long time dependence, but we can adopt a time delay neural network to capture GNSS anomalous behavior through changes in the test metric that are less dynamic and will be explained later. . Since there was no modification to the network classification, standard learning techniques are generally based on temporal backpropagation and adjust the slope weight of the error surface on the time delay neural network. Stimulus signal

Is captured and processed together with the synaptic weight w by the sum function, and the activation potential of the newly activated neuron can be obtained by an activation function as shown in FIG. 3. Subsequent network output is given by Equation (4).

Equation 4:

, 또는 하이퍼볼릭 탄젠트,

등 다양한 미분가능함수가 사용될 수 있다. 이 두 가지 함수들의 이점은, 각각의 첫 번째 변수를 그들만의 간단한 함수로 나타낼 수 있다는 것이다.In the above equation, M is the number of inputs and L is the number of neurons on the hidden layer. The typical choice of active function is S-type function (sigmoid),

, Or hyperbolic tangent,

Various differentiable functions can be used. The advantage of these two functions is that each first variable can be represented by their own simple function.

Of the many algorithms used to train time-delay neural networks, the Levenberg-Marquadt optimization scheme is the most effective. The Levenberg-Marquadt method reduces the sum of squared errors in the equation by appropriately diversifying the parameter updates.

Equation 5:

는 k번째 값의 오차이고

는

요소를 가진 벡터이다. 이전 가중 벡터(weight vector)와 새로 계산된 가중 벡터의 차이가 작을 경우, 오차 벡터는 테일러 시리즈를 통해 첫 번째로 확장될 수 있다.In Equation 5,

Is the error of the kth value

The

It is a vector with elements. If the difference between the previous weight vector and the newly calculated weight vector is small, the error vector can be first extended through the Taylor series.

를 최소화해야 하는 경우, Newton's method는 수학식 6과 같이 나타낼 수 있다. Function with respect to weight vector w

When must be minimized, Newton's method can be expressed as Equation 6.

Equation 6:

는 Hessian matrix이며

는 기울기이다.

를 도출하는 식을 수학식 8과 같이 나타낼 수 있다.In Equation 6,

Is the Hessian matrix

Is the slope.

The equation for deriving can be expressed as Equation 8.

Equation 7:

Equation 8:

는 Jacobian matrix이며, 이는 가중치와 편차에 대한 네트워크 오차의 1차 미분으로 구성되어 있다. 수학식 8의 두 번째 항은 Gauss-Newton Method의 Jacobian matrix의 결과물과 비교하여 수학식 9와 같이 가정할 수 있다.In Equation 8,

Is the Jacobian matrix, which consists of the first derivative of the network error for weights and deviations. The second term of Equation (8) can be assumed as Equation (9) compared to the output of the Gauss-Newton Method Jacobian matrix.

Equation 9:

In order to prevent the briefly expressed Hessian matrix from being irreversible, the Levenberg-Marquadt modification of the Gauss-Newton method can be obtained as shown in Equation (10).

Equation 10:

는 단위행렬이며

는 행렬

가 양의 값을 갖도록 보장하는 매개변수로서, 행렬을 가역적으로 만드는 역할을 한다.

값을 적절히 선택하는 것이 중요하다. 왜냐하면 이는 알고리즘 함수의 안정성과 수렴 속도를 결정하기 때문이다.In Equation 10,

Is the unit matrix

The matrix

A parameter that ensures that a value is positive, which makes the matrix reversible.

It is important to choose the appropriate value. This is because it determines the stability and speed of convergence of the algorithm function.

To provide users with information about GNSS anomalous behavior, it is proposed to install an additional integrity monitoring system in the reinforcement system that identifies observations of irregular time series in the test statistic, such as residual range error. Simple approaches to the modeling process include averaging past data points, returning previous data points, and linear extrapolation. While this may be sufficient for some relatively simple time series, more sophisticated methods are required in real time series, such as navigation signals.

Time-delay neural networks use historical data for the purpose of predicting future values by finding a process model that empirically matches historical data. In addition, time-delayed neural networks with noise suppression are called finite impulse response (FIR) networks because of tapped delay lines representing past activity. Irregularities in navigation measurements, which indicate the intrinsic dynamics of a navigating satellite or nonlinearity in the navigation signal path, can be captured by mapping past periods into the future. The time-delay neural network can be combined with a time series of test statistics through general RAIM.

)이 네트워크 입력에 머물게 하기 위해서, 아래의 수학식 11과 같이 검정통계량(표준 RAIM으로 계산된 잔차 범위 오차의 제곱합을 기반으로 하는)이 시간 지연된 형태로 변경될 수 있다.Referring to Figure 4, it shows the overall schematic of the proposed approach for integrity monitoring. Navigation signal dynamics received (time series on black statistics)

In order to keep) on the network input, the test statistic (based on the sum of squares of the residual range error calculated by the standard RAIM) may be changed to a time delayed form as in Equation 11 below.

Equation 11:

는 네트워크 출력이며(

의 estimated quantity를 나타내는), M은 tapped delay lines의 개수이다. 신경 지연 신경망을 학습하면, 새로이 제안된 검정통계량이 생성되며, 이는 다음의 수학식 12와 같다.In Equation 11,

Is the network output (

M, which represents the estimated quantity of, is the number of tapped delay lines. When the neural delay neural network is learned, a newly proposed test statistic is generated, as shown in Equation 12 below.

Equation 12:

는 GNSS 위성이 정상 상태에 있을 때 획득된 오프라인 의사거리에 기초하여 전파된 입력 신호가 형성되는 TDNN의 추정 출력이고,

은 온라인 의사 거리로 추정된 출력이다. In Equation 12,

Is the estimated output of the TDNN where the propagated input signal is formed based on the offline pseudorange obtained when the GNSS satellite is in a steady state,

Is the output estimated by the online pseudorange.

(평소 상태를 나타내는)와

사이 차이에 즉각적으로 반응한다(430). 계산된 차이의 피크값은 경보를 울리는 데 사용된다. 강화된 불일치는 경보 메시지를 활성화시킨다(440). When the GNSS satellite is in a normal state, an offline pseudo-range may be acquired (410). In addition, it is possible to obtain an online pseudorange of the GNSS satellite (420). A statistical test amount may be calculated based on the obtained offline pseudo distance (411), and a statistical test amount may be calculated based on the obtained online pseudo distance (421). The estimated output 412 of the time-delayed neural network in which the propagated input signal is formed based on the obtained offline pseudo-range and the estimated output 422 of the time-delayed neural network in which the propagated input signal is formed based on the obtained on-line pseudo-distance. It can be compared (430). This approach allows the output to be contaminated by anomalous behavior,

(Indicating the usual state) and

Immediately respond to the difference between (430). The peak value of the calculated difference is used to sound the alarm. The enhanced discrepancy activates an alert message (440).

검정통계량의 평균과 표준편차를 통해 획득되며, 이는 수학식 12을 통하여 구할 수 있다. The detection threshold is usually

Obtained through the mean and standard deviation of the test statistic, which can be obtained through Equation (12).

Equation 13:

과

는 각각

의 평균과 표준편차이다.

는 요구되는 조건 하 무결성 감지 가능성을 만족시키기 위해 사용되는 multiplier(CAT-I operation상 지상기반 보강시스템(GBAS)을 지원하는 code-carrier divergence monitors의 경우, 이 값은 1 x 10^-8)이다. 표준편차의 부풀려진 값인, 팽창 인자(

)는 연속성의 위험을 피하기 위해 true error분포를 overbound한다.In Equation 13,

and

Each

Is the mean and standard deviation of.

Is 1 x 10 ^-8 for multipliers (code-carrier divergence monitors supporting ground-based reinforcement systems in CAT-I operation) that are used to satisfy the possibility of integrity detection under required conditions. Expansion factor, the inflated value of the standard deviation (

) Overbound the true error distribution to avoid the risk of continuity.

와

의 estimated quantity간 차이를 측정하는 것이다. 이러한 측정은 GNSS 변칙행동 관련 가정에 기초한 추정의 일관성을 변경한다. 이후, 수학식 13의 검정통계량을 수학식 3에서 설명한 임계값과 비교하여 경보를 발생시켜야 하는지 여부를 확인할 수 있다.In summary, in Equation 13, the role of the test statistic is

Wow

It is to measure the difference between the estimated quantities of. These measures change the consistency of the estimates based on assumptions related to GNSS anomalous behavior. Thereafter, it is possible to determine whether an alarm should be generated by comparing the test statistic of Equation 13 with the threshold described in Equation 3.

The device described above may be implemented with hardware components, software components, and/or combinations of hardware components and software components. For example, the devices and components described in the embodiments may include, for example, processors, controllers, arithmetic logic units (ALUs), digital signal processors (micro signal processors), microcomputers, field programmable gate arrays (FPGAs). , A programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions, may be implemented using one or more general purpose computers or special purpose computers. The processing device may run an operating system (OS) and one or more software applications running on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For convenience of understanding, a processing device may be described as one being used, but a person having ordinary skill in the art, the processing device may include a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that may include. For example, the processing device may include a plurality of processors or a processor and a controller. In addition, other processing configurations, such as parallel processors, are possible.

The software may include a computer program, code, instruction, or a combination of one or more of these, and configure the processing device to operate as desired, or process 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 embodied in The software may be distributed on networked computer systems, and stored or executed in a distributed manner. Software and data may be stored in 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 on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, or the like alone or in combination. The program instructions recorded on the medium may be specially designed and constructed for the embodiments or may be known and usable by 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, DVDs, and magnetic media such as floptical disks. -Hardware devices specifically configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language code that can be executed by a computer using an interpreter, etc., as well as machine language codes made by a compiler.

As described above, although the embodiments have been described by a limited embodiment and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques are performed in a different order than the described method, and/or the components of the described system, structure, device, circuit, etc. are combined or combined in a different form from the described method, or other components Alternatively, even if replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (10)

In the integrity monitoring method,
Generating a time-delay neural network-based model using a calculated test statistic for a Global Navigation Satellite System (GNSS) signal through Receiver Autonomous Integrity Monitoring; And
Detecting anomalous behavior of the GNSS as learning the GNSS signal received through the generated model
Integrity monitoring method comprising a. According to claim 1,
The step of generating a model based on the time-delay neural networks,
Calculating a test statistic according to the sum of squared residual distance errors by quantifying self-consistency for RAIM GNSS overlap measurement that performs integrity monitoring through Equation 1
Equation 1:

Integrity monitoring method comprising a. According to claim 2,
The step of generating a model based on the time-delay neural networks,
Applying a non-repetitive structure based on a feedforward direction delay element and a repetitive structure with delayed feedback to store time information in the model
Integrity monitoring method comprising a. According to claim 1,
The step of generating a model based on the time-delay neural networks,
Adopting a time-delay neural network-based model to capture anomalous behavior of GNSS through a change in the test metric to learn time dependence beyond a predetermined criterion
Integrity monitoring method comprising a. According to claim 1,
The step of detecting anomalous behavior of the GNSS,
Extracting information on the coupling characteristic expressed in an irregular sequence as a residual range calculated through the received GNSS signal.
GNSS integrity monitoring method comprising a. According to claim 1,
The step of detecting anomalous behavior of the GNSS,
Changing the test statistic based on the sum of squares of the residual range error calculated by standard RAIM in a time delayed form so that the time series in the received statistic of the GNSS signal stays at the input of the time lag neural network.
Integrity monitoring method comprising a. The method of claim 6,
The step of detecting anomalous behavior of the GNSS,
Generating a newly proposed test statistic as a result of learning the time delay neural network
Integrity monitoring method comprising a. The method of claim 6,
The step of detecting anomalous behavior of the GNSS,
After measuring the difference between the offline pseudo-distance-based estimation output and the online pseudo-distance-based estimation output through the correlation of the time series of the newly proposed test statistic, the newly proposed test statistic is compared with the threshold of Equation (1). To check whether an alarm has occurred
Integrity monitoring method comprising a. The method of claim 8,
The step of detecting anomalous behavior of the GNSS,
When the GNSS is in a steady state, an alarm is generated based on a difference between an estimated power and an estimated power based on the online pseudo distance as the received GNSS signal is trained on a time delay neural network based on the offline pseudo distance obtained. step
Integrity monitoring method comprising a. In the integrity monitoring system,
A model generator for generating a time-delay neural network-based model by using a calculated test statistic for a Global Navigation Satellite System (GNSS) signal through Receiver Autonomous Integrity Monitoring; And
As a part of learning the GNSS signal received through the generated model, a detector for detecting anomalous behavior of the GNSS
Integrity monitoring system comprising a.

Images