WO2022176093A1

WO2022176093A1 - Abnormality detection system, vehicle-mounted device, abnormality detection method, and program

Info

Publication number: WO2022176093A1
Application number: PCT/JP2021/006068
Authority: WO
Inventors: 剛志是永; 健司 ▲高▼尾; 昌弘山田; 陽平知識
Original assignee: 三菱重工機械システム株式会社
Priority date: 2021-02-18
Filing date: 2021-02-18
Publication date: 2022-08-25
Also published as: JP7467756B2; JPWO2022176093A1

Abstract

This abnormality detection system comprises: a state observation unit for collecting, via a GNSS receiver mounted in a vehicle, GNSS positioning data including information relating to satellite signals received from GNSS satellites and information about the position of the vehicle estimated from the satellite signals; a generation unit for generating pseudo GNSS positioning data using a generation model trained using the GNSS positioning data; and an abnormality determination unit for, when new GNSS positioning data is acquired, determining whether the GNSS positioning data is abnormal on the basis of a comparison of the same with pseudo GNSS positioning data generated by the generation unit.

Description

Anomaly detection system, vehicle-mounted device, anomaly detection method, and program

The present disclosure relates to an anomaly detection system, an onboard device, an anomaly detection method, and a program.

The Global Navigation Satellite System (hereinafter also referred to as GNSS) is known as a technology for identifying the traveling position of mobile objects such as vehicles. For example, Patent Document 1 describes a technique in which a GNSS receiver (vehicle device) mounted on a vehicle receives satellite signals transmitted from GNSS satellites and estimates the position of the vehicle based on the received satellite signals. ing.

In addition, GNSS charging that identifies the travel route of the vehicle from time-series changes in GNSS positioning data, including information on received satellite signals and self-location information estimated from satellite signals, and charges a toll according to the travel route. system is considered.

Japanese Patent Application Laid-Open No. 9-5092

Applications used in GNSS billing systems, etc., check GNSS positioning data for anomalies in order to verify operation during the development stage, detect malfunctions of on-board units after operation, and detect fraud such as spoofing and jamming of GNSS positioning data. required throughout the life cycle.

However, the quality of GNSS positioning data such as the size of position error and signal strength depends on the characteristics of each model of the onboard unit, the characteristics of the vehicle in which the onboard unit is installed, and the usage environment of the onboard unit (existing around the driving route). (presence or absence of buildings, etc.). Therefore, it has been difficult to detect anomalies in GNSS positioning data by rule-based determination processing that combines simple conditional branching such as comparison with a threshold value.

The present disclosure has been made in view of such problems, and an anomaly detection system capable of detecting anomalies in actually measured GNSS positioning data using pseudo GNSS positioning data generated by a trained generative model. , an onboard device, an anomaly detection method, and a program.

According to one aspect of the present disclosure, an anomaly detection system provides information about satellite signals received from GNSS satellites and information about the position of the vehicle estimated from the satellite signals via a GNSS receiver mounted on a vehicle. A state observation unit that collects GNSS positioning data including a generation unit that generates pseudo GNSS positioning data by a generation model learned using the GNSS positioning data, and a generation unit that generates new GNSS positioning data when new GNSS positioning data is acquired and an abnormality determination unit that determines whether or not the newly acquired GNSS positioning data is abnormal based on comparison with the pseudo GNSS positioning data obtained.

According to one aspect of the present disclosure, the vehicle-mounted device includes information on satellite signals received from GNSS satellites via a GNSS receiver mounted on the vehicle, and information on the position of the vehicle estimated from the satellite signals. A state observation unit that collects GNSS positioning data; and an abnormality determination unit that determines whether or not the GNSS positioning data is abnormal based on comparison with the pseudo GNSS positioning data.

According to one aspect of the present disclosure, an anomaly detection method includes information about satellite signals received from GNSS satellites and information about the position of the vehicle estimated from the satellite signals via a GNSS receiver mounted on a vehicle. generating pseudo GNSS positioning data by a generative model trained using the GNSS positioning data; and generating the generated GNSS positioning data when new GNSS positioning data is acquired. and determining whether the GNSS positioning data is abnormal based on comparison with the pseudo GNSS positioning data.

According to one aspect of the present disclosure, a program includes information regarding satellite signals received from GNSS satellites via a GNSS receiver onboard a vehicle, and information regarding the position of the vehicle estimated from the satellite signals. collecting positioning data; generating pseudo-GNSS positioning data by a generative model trained using the GNSS positioning data; and generating the pseudo-GNSS positioning data when new GNSS positioning data is acquired. A step of determining whether or not the GNSS positioning data is abnormal based on the comparison with the positioning data is caused to be executed by the computer of the vehicle-mounted device.

According to the anomaly detection system, vehicle-mounted device, anomaly detection method, and program according to the present disclosure, an anomaly in actually measured GNSS positioning data is detected using pseudo GNSS positioning data generated by a learned generation model. can be done.

1 is a diagram showing the overall configuration of an anomaly detection system according to a first embodiment of the present disclosure; FIG. 1 is a diagram showing a functional configuration of an anomaly detection system according to a first embodiment of the present disclosure; FIG. FIG. 3 is a diagram illustrating an example of a functional configuration of a learning unit and a learning process according to the first embodiment of the present disclosure; FIG. It is a figure which shows an example of the abnormality detection process which concerns on 1st Embodiment of this indication. FIG. 10 is a diagram illustrating an example of a functional configuration of a learning unit and a learning process according to the second embodiment of the present disclosure; FIG. 11 is a diagram illustrating an example of anomaly detection processing according to the second embodiment of the present disclosure; FIG. FIG. 12 is a diagram showing the functional configuration of an anomaly detection system according to a third embodiment of the present disclosure; FIG. It is a figure showing an example of hardware constitutions of onboard equipment concerning one embodiment of this indication, and a center server.

<First embodiment>
An abnormality detection system 1 according to a first embodiment of the present disclosure will be described below with reference to FIGS. 1 to 4. FIG.

(Overall configuration of anomaly detection system)
FIG. 1 is a diagram showing the overall configuration of an anomaly detection system according to the first embodiment of the present disclosure.
As shown in FIG. 1 , the anomaly detection system 1 includes a plurality of vehicle-mounted devices 10 and a center server 20 .

A plurality of vehicle-mounted devices 10 are mounted on vehicle A, respectively. The vehicle-mounted device 10 receives satellite signals emitted by the GNSS satellites 40 and estimates the position of the vehicle A on which the device is mounted based on the signals. The vehicle-mounted device 10 also collects and records GNSS positioning data including information about the received satellite signals, information about the estimated position of the vehicle A, and the like.

The center server 20 is communicably connected to the vehicle-mounted device 10 via a wireless communication network N (such as a cellular network). The center server 20 collects GNSS positioning data from each of the plurality of vehicle-mounted devices 10 and learns a generation model that generates pseudo GNSS data simulating normal GNSS data in each vehicle-mounted device 10 . Moreover, the center server 20 detects abnormality of the GNSS positioning data actually measured in each vehicle-mounted device 10 using the learned generation model.

(Functional configuration of anomaly detection system)
FIG. 2 is a diagram showing the functional configuration of the anomaly detection system according to the first embodiment of the present disclosure.
As shown in FIG. 2 , the vehicle-mounted device 10 includes a GNSS receiver 100 , a sensor information acquisition section 101 , a state observation section 102 , an output section 103 and a storage medium 104 .

The GNSS receiver 100 receives satellite signals transmitted from the GNSS satellites 40 and estimates the position of the vehicle A. The GNSS receiver 100 also generates GNSS positioning data including signal reception information about the received satellite signals and estimated position information about the position of the vehicle A, and outputs the data to the state observation unit 102 .

The signal reception information includes the signal strength of the satellite signals received by the GNSS receiver 100, the number of satellites, the elevation angle, the azimuth angle, and the like. The estimated position information includes the estimated position (latitude, longitude, altitude) of vehicle A at the time of signal reception, estimated speed, and the like.

The sensor information acquisition unit 101 acquires sensor information from the in-vehicle sensor 30 of the vehicle A. The vehicle-mounted sensor 30 is a receiver that receives a beacon signal capable of detecting the position of the vehicle A, for example, from a position information distribution beacon provided on the roadside. The in-vehicle sensor 30 may be a receiver that receives a signal capable of detecting the vehicle speed of the vehicle A at the installation position of the speed measuring device from the speed measuring device provided on the roadside. In addition, the in-vehicle sensor 30 may be an in-vehicle camera or LiDAR (Light Detection and Ranging) capable of detecting the surrounding environment of the vehicle A (presence or absence of surrounding buildings, etc.). Furthermore, the vehicle-mounted sensor 30 may be a vehicle speed sensor, an acceleration sensor, an angular velocity sensor, or the like of the vehicle A. The sensor information acquisition unit 101 outputs various sensor information acquired via the in-vehicle sensor 30 to the state observation unit 102 .

The state observation unit 102 collects GNSS positioning data via the GNSS receiver 100 .

In addition, the state observation unit 102 further collects positioning errors of the GNSS positioning data. The positioning error is the difference (position error) between the estimated position of vehicle A based on satellite signals and the actual position. The positioning error may also include the difference (speed error) between the estimated speed of vehicle A based on the satellite signal and the actual speed.

For example, state observation unit 102 identifies the location of vehicle A on the link in the map data based on map data stored in advance in storage medium 104, and calculates the distance between the estimated location of vehicle A and the location on the link. is obtained as the position error. Note that the state observation unit 102 may obtain, as a position error, the difference between the estimated position of the vehicle A and the position of the vehicle A specified based on the beacon signal of the position information distribution beacon acquired by the sensor information acquisition unit 101. good. In addition, the state observation unit 102 uses the difference between the estimated speed of the vehicle A and the speed of the vehicle A obtained from a roadside speed measuring device or the vehicle speed sensor of the vehicle A via the sensor information acquisition unit 101 as a speed error. Ask.

In addition, the state observation unit 102 further collects receiver information indicating the characteristics of the GNSS receiver 100, mounting environment information indicating the mounting environment of the GNSS receiver 100, and surrounding environment information indicating the presence or absence of shields around the vehicle A. may be collected.

The characteristics of the GNSS receiver 100 are antenna sensitivity, directivity, frequency characteristics, and the like. Receiver information includes, for example, the type of GNSS receiver. Moreover, receiver information may also contain the information which shows the manufacturing time of the GNSS receiver 100, and the attachment period to the vehicle A. FIG. From the date of manufacture and the period of installation, the aging effects of the GNSS receiver 100 can be detected.

The installation environment information includes information on the vehicle on which the GNSS receiver 100 is mounted (model, etc.), antenna installation position, and antenna installation angle.

The surrounding environment information is information that enables detection of the presence or absence of obstacles that block the satellite signals of the GNSS satellites 40, such as buildings that exist around the vehicle A. The surrounding environment information is, for example, an image captured by an in-vehicle camera acquired by the sensor information acquiring unit 101 and a LiDAR measurement result.

The state observation unit 102 transmits the generated GNSS positioning data to the center server 20 each time the GNSS receiver 100 generates GNSS positioning data. In another embodiment, the state observation unit 102 collects a plurality of accumulated GNSS positioning data to the center server 20 every predetermined period or every time a predetermined number of GNSS positioning data is accumulated in the storage medium 104. You may send. Moreover, the state observation part 102 transmits these to the center server 20 with GNSS positioning data, when the positioning error, receiver information, installation environment information, and surrounding environment information are collected.

The output unit 103 is a display device such as a liquid crystal display, and outputs (displays) anomaly detection information regarding anomalies in GNSS positioning data. Note that the output unit 103 may further include a speaker for outputting the abnormality detection information by voice or the like.

The storage medium 104 stores various data acquired and generated by each part of the vehicle-mounted device 10 . For example, the storage medium 104 stores sensor information acquired by the sensor information acquisition unit 101, GNSS positioning data collected by the state observation unit 102, and the like.

The center server 20 also includes a learning unit 200 , a generation unit 201 , an abnormality determination unit 202 and a storage medium 203 .

The learning unit 200 learns a generation model for generating pseudo GNSS positioning data using the learning data received from the vehicle-mounted device 10 . The learning data includes GNSS positioning data collected by the state observation unit 102 of the vehicle-mounted device 10 . The learning data may further include receiver information of the vehicle-mounted device 10, installation environment information, and information about the surrounding environment of the vehicle A in which the vehicle-mounted device 10 is mounted.

The generation unit 201 generates pseudo GNSS positioning data using the generation model learned by the learning unit 200 .

When new GNSS positioning data is acquired from the vehicle-mounted device 10, the abnormality determination unit 202 determines whether the GNSS positioning data is abnormal based on comparison with the pseudo GNSS positioning data generated by the generation unit 201. determine whether

Various data acquired and generated by each part of the center server 20 are stored in the storage medium 203 . For example, the storage medium 203 stores GNSS positioning data collected from each vehicle-mounted device 10, a generation model learned by the learning unit 200, and the like.

(learning process)
FIG. 3 is a diagram illustrating an example of a functional configuration of a learning unit and a learning process according to the first embodiment of the present disclosure; FIG.
The learning unit 200 according to the present embodiment learns a generative model for generating pseudo-GNSS positioning data using the technology of generative adversarial networks. As shown in FIG. 3, learning section 200 has generator 2001 and discriminator 2002 .

The generator 2001 is a generation model that generates pseudo GNSS positioning data x' simulating normal GNSS positioning data from the input latent variable z. That is, the generator 2001 maps the input latent variable z to an observation space composed of a plurality of actually measured GNSS positioning data. In the storage medium 203 of the center server 20, GNSS positioning data in a normal state actually measured by the vehicle-mounted device 10 is stored in advance as learning data. Using this learning data, the generator 2001 performs learning so as to generate pseudo GNSS positioning data close to the genuine GNSS positioning data acquired during normal operation. The normal GNSS positioning data is, for example, GNSS positioning data determined to be normal by an engineer or the like. Also, the normal GNSS positioning data may be GNSS positioning data collected by the GNSS receiver 100 that has been confirmed to be operating normally without failure or the like.

The discriminator 2002 identifies whether the input GNSS positioning data is GNSS positioning data x actually measured by the vehicle-mounted device 10 or pseudo GNSS positioning data x' generated by the generator 2001 . The discriminator 2002 learns based on the learning data accumulated in the storage medium 203 so as to distinguish between the GNSS positioning data that is genuine data and the pseudo GNSS positioning data that is fake.

It should be noted that, as described above, the learning data may further include positioning errors. By using such training data, the generator 2001 is trained to generate pseudo GNSS positioning data with positioning errors that are closer to real GNSS positioning data. Also, the discriminator 2002 is learned so that even if the GNSS positioning data contains a positioning error, the authenticity of the data can be discriminated more accurately.

In addition, the learning data includes receiver information (model of GNSS receiver 100, manufacturing time, specification period, etc.), installation environment information (GNSS receiver 100 installation position, etc.), surrounding environment information (presence or absence of a shield that shields satellite signals, etc.) may further include The signal reception information (signal strength, number of satellites, etc.) of the GNSS satellites 40 and the estimated position information of the vehicle A obtained from the received satellite signals are based on the characteristics of the receiver model, the installation environment, or the vehicle A (the GNSS receiver 100 ) changes depending on conditions such as the presence or absence of a shield around. For example, receiver characteristics (antenna sensitivity, directivity, frequency characteristics) differ for each model of GNSS receiver 100 . For this reason, when the two vehicle-mounted devices 10 have GNSS receivers 100 of different models, even if satellite signals are received at the same position at the same time, they are included in the GNSS positioning data output by each vehicle-mounted device 10. Signal reception information and estimated location information may differ. Similarly, the GNSS positioning data may differ depending on the surrounding environment such as the mounting environment of the GNSS receiver 100 and the presence or absence of a shield. For this reason, the generator 2001 further learns the receiver information, the mounting environment information, and the surrounding environment information contained in the learning data so that it can generate pseudo GNSS positioning data that is closer to the real GNSS positioning data according to these information. do. Similarly, the discriminator 2002 is learned so that even if the GNSS positioning data is affected by the receiver characteristics, the installation environment, and the presence or absence of obstructions, the authenticity of the data can be discriminated more accurately.

In addition, the learning unit 200 performs processing for further learning the generator 2001 and the discriminator 2002 by repeating training for the discriminator 2002 to discriminate whether the pseudo GNSS positioning data generated by the generator 2001 is genuine or fake. The flow of this learning process will be described below with reference to FIG.

First, the generator 2001 generates pseudo GNSS positioning data x' based on the input latent variable z (step S101). Here, for example, a random number is used as the latent variable z.

Next, the pseudo GNSS positioning data x' generated by the generator 2001 or the GNSS positioning data x accumulated in the storage medium 203 is input to the discriminator 2002 (step S102).

The discriminator 2002 identifies whether the input data is GNSS positioning data x (genuine) or pseudo GNSS positioning data x' (fake), and outputs identification result P (step S103).

In the phase of learning the discriminator 2002, learning (parameter adjustment, etc.) is performed so that the discriminator 2002 can more accurately discriminate between the fake and the genuine article based on the correctness of the identification result P (step S104). At this time, the generator 2001 remains unchanged and fixed.

Also, in the phase of learning the generator 2001, the generator 2001 learns based on the identification result P so that it can generate pseudo GNSS positioning data that is closer to the real thing (such that the discriminator 2002 misidentifies it as real). (step S105). At this time, the discriminator 2002 remains fixed without being changed.

The learning unit 200 alternately learns the generator 2001 and the discriminator 2002 by repeatedly executing the above steps. Thereby, the learning unit 200 can learn a generative model capable of generating the pseudo GNSS positioning data x′ closer to the real thing. The learned generator 2001 (generative model) is stored in the storage medium 203 and used by the generator 201 when the center server 20 performs anomaly detection processing, which will be described later.

(abnormality detection processing)
FIG. 4 is a diagram illustrating an example of anomaly detection processing according to the first embodiment of the present disclosure.
Hereinafter, the flow of anomaly detection processing of GNSS positioning data in the anomaly detection system 1 will be described with reference to FIG. 4 . This abnormality detection process is performed after the learning process (FIG. 3) described above is completed.

First, the state observation unit 102 of the vehicle-mounted device 10 acquires GNSS positioning data via the GNSS receiver 100 (step S111). Also, the state observation unit 102 acquires sensor information via the sensor information acquisition unit 101 (step S112).

Also, the state observation unit 102 transmits evaluation data for detecting anomalies in the GNSS positioning data to the center server 20 (step S113). The evaluation data includes GNSS positioning data. Also, the evaluation data may include at least one of the positioning error, receiver information, installation environment information, and surrounding environment information collected by the state observation unit 102 .

Next, the generation unit 201 of the center server 20 generates pseudo GNSS positioning data x' using the generation model that has been learned by the learning unit 200 (step S114). Note that in this embodiment, the latent variable z corresponding to normal GNSS positioning data is unknown. Therefore, the generation unit 201 sequentially inputs all possible values of the latent variable z, and generates a plurality of pseudo GNSS positioning data x' based on each value. Then, one of the generated pseudo GNSS positioning data is close to the real GNSS positioning data.

Next, the abnormality determination unit 202 of the center server 20 compares the GNSS positioning data x, which is the actual measurement data acquired from the vehicle-mounted device 10, with the pseudo GNSS positioning data x′ generated by the generation unit 201, and determines the GNSS positioning data. It is determined whether there is an abnormality in x (step S115).

The pseudo GNSS positioning data x' simulates GNSS positioning data estimated to be obtained when the vehicle-mounted device 10 is in a normal state. Therefore, if the state of the vehicle-mounted device 10 is normal, the GNSS positioning data x actually measured by the vehicle-mounted device 10 is data close to any of the plurality of pseudo GNSS positioning data x' generated by the generation unit 201. should be. For this reason, the abnormality determination unit 202 calculates, for example, the degree of matching between the GNSS positioning data x and each of the plurality of pseudo GNSS positioning data x', and the pseudo GNSS positioning data x' having a value equal to or higher than a predetermined value of the degree of matching is If there is, the GNSS positioning data x is determined to be normal. On the other hand, the abnormality determination unit 202 determines that the GNSS positioning data x is abnormal when there is no pseudo GNSS positioning data x' having a degree of matching equal to or greater than a predetermined value.

When the determination is completed, the abnormality determination unit 202 transmits abnormality detection information including the determination result indicating the presence or absence of abnormality to the vehicle-mounted device 10 (step S116). Note that the abnormality detection information may include the degree of matching calculated by the abnormality determination unit 202 .

Also, the output unit 103 of the vehicle-mounted device 10 outputs the abnormality detection information received from the center server 20 and presents it to the passengers of the vehicle A (step S117). Note that the output unit 103 may output the abnormality detection information as a warning only when the abnormality detection information includes information indicating that the GNSS positioning data x is abnormal.

The anomaly detection system 1 executes the series of processes shown in FIG. 4 each time the state observation unit 102 acquires GNSS positioning data or at predetermined time intervals to determine whether there is an anomaly in the GNSS positioning data.

(Effect)
As described above, the anomaly detection system 1 according to the present embodiment includes the GNSS positioning data x acquired by the state observation unit 102 and the pseudo GNSS positioning data x′ created by the generation unit 201 using the learned generation model. , it is determined whether or not the acquired GNSS positioning data x is abnormal.
By doing so, the abnormality detection system 1 detects that the GNSS positioning data x is abnormal when the GNSS positioning data x actually measured by the GNSS receiver 100 deviates from the pseudo GNSS positioning data x'. be able to.

The center server 20 of the anomaly detection system 1 further includes a learning unit 200 that learns a generation model for generating pseudo GNSS positioning data using normal GNSS positioning data collected by the state observation unit 102 .
By doing so, the anomaly detection system 1 can learn a generation model capable of generating pseudo GNSS positioning data simulating data estimated to be acquired when the state of the GNSS receiver 100 is normal. can. Therefore, by using the generative model learned in this way, the anomaly detection system 1, when the actually measured GNSS positioning data x deviates from the pseudo GNSS positioning data x′ simulating the normal data, , the GNSS positioning data x can be determined to be abnormal.

In addition, the state observation unit 102 further collects positioning errors of the GNSS positioning data, and the learning unit 200 further uses the positioning errors to learn the generative model.
The actually measured GNSS positioning data may contain positioning errors. Therefore, the anomaly detection system 1 according to the present embodiment learns the generative model including the positioning error. Thereby, since the abnormality detection system 1 can also simulate the positioning error that occurs in the actual vehicle-mounted device 10 (GNSS receiver 100), it is possible to generate pseudo GNSS positioning data including such a positioning error. Become. Then, since the anomaly detection system 1 can generate pseudo GNSS positioning data that is closer to the real thing, it is possible to improve the accuracy of anomaly detection.

In addition, the state observation unit 102 further collects receiver information capable of detecting the characteristics of the GNSS receiver 100, and the learning unit 200 further uses the receiver information to learn the generative model.
Depending on the type of GNSS receiver 100, receiver characteristics (antenna sensitivity, directivity, frequency characteristics, etc.) may differ. That is, using different GNSS receivers 100 may result in different positioning results (contents of GNSS positioning data) even at the same place at the same time. For this reason, the anomaly detection system 1 according to the present embodiment learns a generative model including the receiver information (type) of the GNSS receiver 100 . Thereby, the anomaly detection system 1 can generate more realistic pseudo GNSS positioning data according to the characteristics of the GNSS receiver 100 . As a result, it is possible to perform anomaly detection in consideration of the difference in characteristics of each GNSS receiver 100, and improve the accuracy of anomaly detection.

Moreover, the receiver information may include information capable of specifying the manufacturing time and usage period of the GNSS receiver 100 .
Thereby, the anomaly detection system 1 can further learn changes in the characteristics of the GNSS receiver 100 due to deterioration over time. As a result, it is possible to perform anomaly detection in consideration of aging deterioration of each GNSS receiver 100, and improve the accuracy of anomaly detection.

In addition, the state observation unit 102 further collects mounting environment information indicating the mounting environment of the GNSS receiver 100, and the learning unit 200 further uses the mounting environment information to learn the generation model.
For example, if the vehicle in which the GNSS receiver 100 is mounted, the mounting position, or the like in the vehicle A changes, the signal strength of the satellite signal received by the GNSS receiver 100 may be affected. If it does so, even if it is the GNSS receiver 100 of the same model, a positioning result may differ. For this reason, the anomaly detection system 1 according to the present embodiment learns the generation model including the installation environment information of the GNSS receiver 100 . Thereby, the anomaly detection system 1 can generate more realistic pseudo GNSS positioning data according to the mounting environment of the GNSS receiver 100 . As a result, it is possible to perform anomaly detection in consideration of the installation environment of the GNSS receiver 100, and improve the accuracy of anomaly detection.

In addition, the state observation unit 102 further collects surrounding environment information regarding the surrounding environment of the vehicle A, and the learning unit 200 further uses the surrounding environment information to learn the generative model.
If there is a building around the vehicle A that blocks satellite signals, the signal strength of the satellite signals received by the GNSS receiver 100 may be affected. If it does so, even if the state of the onboard equipment 10 (GNSS receiver 100) is normal, there exists a possibility that a positioning error etc. may change a lot with the driving|running|working position of the vehicle A. For this reason, the anomaly detection system 1 according to the present embodiment learns the generative model including the surrounding environment information of the vehicle A. FIG. As a result, the anomaly detection system 1 can generate more realistic pseudo GNSS positioning data according to the presence or absence of a shield around the vehicle A. As a result, it is possible to perform anomaly detection in consideration of the shielding situation around the vehicle A, thereby improving the accuracy of anomaly detection.

<Second embodiment>
Next, an abnormality detection system 1 according to a second embodiment of the present disclosure will be described with reference to FIGS. 5 and 6. FIG.
Components common to those of the first embodiment are denoted by the same reference numerals, and detailed description thereof is omitted.

(learning process)
FIG. 5 is a diagram illustrating an example of a functional configuration of a learning unit and a learning process according to the second embodiment of the present disclosure;
As shown in FIG. 5, the learning section 200 according to this embodiment further has an encoder 2003 .

The encoder 2003 is a model (latent variable model) for obtaining the optimum latent variable z' to be input to the generator 2001. The encoder 2003 maps the GNSS positioning data accumulated in the storage medium 203 to the latent space, and learns so that when the GNSS positioning data is input, the latent variable z' corresponding to the input data can be output.

In addition, the learning unit 200 performs processing for further learning the generator 2001, the discriminator 2002, and the encoder 2003. The flow of this learning process will be described below with reference to FIG.

First, the generator 2001 generates pseudo GNSS positioning data x' based on the input latent variable z, as in the first embodiment (step S201).

Also, the encoder 2003 obtains the latent variable z' corresponding to the input GNSS positioning data x (step S202).

Next, data consisting of a combination of GNSS positioning data and latent variables is input to the discriminator 2002 (step S202). The data input here is a set (x', z) consisting of the pseudo GNSS positioning data x' generated by the generator 2001 and the latent variable z used for its generation, or the GNSS positioning data input to the encoder 2003 It is a tuple (x,z') consisting of x and its corresponding latent variable z'.

The discriminator 2002 identifies whether the input data is GNSS positioning data x (genuine) or pseudo GNSS positioning data x', and outputs identification result P (step S204).

The discriminator 2002 and the generator 2001 perform learning based on the identification result P in each learning phase, as in the first embodiment (steps S205 and S206).

Also, in the learning phase of the encoder 2003, the encoder 2003 maps the GNSS positioning data x into the latent space and appropriate latent variables Learning is performed so that z' can be output.

The learning unit 200 alternately learns the generator 2001, the discriminator 2002, and the encoder 2003 by repeatedly executing the above steps. As a result, the learning unit 200 can learn a generative model capable of generating more realistic pseudo GNSS positioning data x', and learn a latent variable model capable of outputting a more appropriate latent variable z'. The learned generator 2001 (generative model) and encoder 2003 (latent variable model) are stored in the storage medium 203 and used by the generator 201 when the center server 20 performs anomaly detection processing, which will be described later.

(abnormality detection processing)
FIG. 6 is a diagram illustrating an example of anomaly detection processing according to the second embodiment of the present disclosure.
Hereinafter, the flow of anomaly detection processing of GNSS positioning data in the anomaly detection system 1 will be described with reference to FIG. This abnormality detection process is performed after the learning process (FIG. 5) described above is completed.

First, the state observation unit 102 of the vehicle-mounted device 10 acquires GNSS positioning data (step S211) and acquires sensor information (step S212). Also, the state observation unit 102 transmits evaluation data for detecting an abnormality in the GNSS positioning data to the center server 20 (step S213). These processes are the same as steps S111 to S113 of the first embodiment (FIG. 4).

Next, the generation unit 201 of the center server 20 inputs the GNSS positioning data acquired from the vehicle-mounted device 10 to the latent variable model learned by the learning unit 200, and generates a latent variable z' corresponding to this GNSS positioning data. is obtained (step S214).

The generation unit 201 also inputs the latent variable z' output from the latent variable model to the generation model, and generates pseudo GNSS positioning data x' from this latent variable z' (step S215).

Next, the abnormality determination unit 202 of the center server 20 compares the GNSS positioning data x, which is the actual measurement data acquired from the vehicle-mounted device 10, with the pseudo GNSS positioning data x′ generated by the generation unit 201, and determines the GNSS positioning data. It is determined whether there is an abnormality in x (step S216). In this embodiment, the generator 201 generates only one piece of pseudo GNSS positioning data based on the latent variable z'. For this reason, the abnormality determination unit 202 only needs to compare the GNSS positioning data x with one generated pseudo GNSS positioning data x'. When the determination is completed, the abnormality determination unit 202 transmits abnormality detection information including the determination result indicating the presence or absence of abnormality to the vehicle-mounted device 10 (step S217). Upon receiving the abnormality detection information from the center server 20, the output unit 103 of the vehicle-mounted device 10 outputs this information and presents it to the passengers of the vehicle A (step S218). These specific processes are the same as steps S115 to S117 of the first embodiment (FIG. 4).

The anomaly detection system 1 executes the series of processes shown in FIG. 6 each time the state observation unit 102 acquires GNSS positioning data or at predetermined time intervals to determine whether there is an anomaly in the GNSS positioning data.

(Effect)
As described above, in the anomaly detection system 1 according to this embodiment, the learning unit 200 further learns the latent variable model using GNSS positioning data. The generation unit 201 also inputs the latent variable z' output from the latent variable model to the generation model to generate the pseudo GNSS positioning data x'.
Since the anomaly detection system 1 according to the present embodiment can identify the latent variable z′ from the GNSS positioning data using a learned latent variable model, each of the plurality of latent variables There is no need to generate a plurality of pseudo-GNSS positioning data based on and compare all pseudo-GNSS positioning data with the GNSS positioning data. Therefore, the anomaly detection system 1 can significantly reduce the processing load on the generation unit 201 and the anomaly determination unit 202 .

<Third Embodiment>
Next, an abnormality detection system 1 according to a third embodiment of the present disclosure will be described with reference to FIG.
Components common to those of the first and second embodiments are denoted by the same reference numerals, and detailed description thereof is omitted.

In the anomaly detection system 1 according to the first and second embodiments, the center server 20 determines whether or not there is an anomaly in the GNSS positioning data collected by each of the plurality of vehicle-mounted devices 10 . On the other hand, in the abnormality detection system 1 according to the present embodiment, each of the plurality of vehicle-mounted devices 10 determines whether there is an abnormality in the GNSS positioning data acquired by itself.

Also, the learning unit 200 of the center server 20 according to this embodiment has the same functional configuration as the learning unit 200 according to the first or second embodiment. When the learning unit 200 completes learning of the generative model and the latent variable model, these learned models are distributed to each vehicle-mounted device 10 and stored in the storage medium 104 of the vehicle-mounted device 10 .

(Functional configuration of anomaly detection system)
FIG. 7 is a diagram showing the functional configuration of an anomaly detection system according to the third embodiment of the present disclosure.
As shown in FIG. 7 , the vehicle-mounted device 10 according to the present embodiment further includes a generation section 111 and an abnormality determination section 112 .

The generation unit 111 and the abnormality determination unit 112 have the same functions as the generation unit 201 and the abnormality determination unit 202 of the center server 20 according to the first or second embodiment.

That is, when only the generation model is distributed from the center server 20, the generation unit 111 performs the same processing as in step S114 of the first embodiment (FIG. 4) to generate a plurality of pseudo GNSS positioning data x'. . In addition, the abnormality determination unit 112 performs the same processing as in step S115 of the first embodiment (FIG. 4), and determines the GNSS positioning data x acquired by the state observation unit 102 and each of the plurality of pseudo GNSS positioning data x′. By comparison, the presence or absence of abnormality is determined.

Further, when the generation model and the latent variable model are distributed from the center server 20, the generation unit 111 performs the same processing as in step S214 of the second embodiment (FIG. 6), and transforms the GNSS positioning data into the latent variable model. Input to obtain the latent variable z'. The generation unit 111 also inputs this latent variable z' to the generation model to generate one piece of pseudo GNSS positioning data x'. The abnormality determination unit 112 performs the same processing as in step S216 of the second embodiment ( FIG. 6 ), compares the GNSS positioning data x acquired by the state observation unit 102 with the pseudo GNSS positioning data x′, and determines whether there is an abnormality. Determine the presence or absence of

(Effect)
As described above, the vehicle-mounted device 10 according to the present embodiment includes the state observation unit 102 that collects GNSS positioning data, the generation unit 111 that generates pseudo GNSS positioning data using a learned generation model, and a newly acquired and an abnormality determination unit 112 that determines whether or not the GNSS positioning data x is abnormal based on a comparison between the GNSS positioning data x and the pseudo GNSS positioning data x'.
By doing in this way, the onboard equipment 10 mounted in each vehicle A can detect abnormality of the GNSS positioning data which self collected. This eliminates the need for the abnormality detection system 1 to frequently transmit and receive evaluation data (GNSS positioning data) between the vehicle-mounted device 10 and the center server 20, thereby significantly reducing the amount of communication. In addition, compared to the mode in which the center server 20 performs abnormality detection processing for a plurality of vehicle-mounted devices 10, the load of the abnormality detection processing can be distributed to each vehicle-mounted device 10, so the processing load of the center server 20 can be greatly reduced. can be reduced.

<Hardware configuration>
FIG. 8 is a diagram illustrating an example of a hardware configuration of a vehicle-mounted device and a center server according to an embodiment of the present disclosure;
As shown in FIG. 8, computer 900 includes CPU 901 , main memory 902 , auxiliary memory 903 , and interface 904 .

The vehicle-mounted device 10 and the center server 20 according to each of the above-described embodiments are each implemented in a computer 900. The operation of each processing unit described above is stored in the auxiliary storage device 903 in the form of a program. The CPU 901 reads out the program from the auxiliary storage device 903, develops it in the main storage device 902, and executes the above processing according to the program. In addition, the CPU 901 secures storage areas corresponding to the storage units described above in the main storage device 902 according to the program.

The program may be for realizing a part of the functions to be exhibited by the computer 900. For example, the program may function in combination with another program already stored in the auxiliary storage device 903 or in combination with another program installed in another device. In other embodiments, the computer 900 may include a custom LSI (Large Scale Integrated Circuit) such as a PLD (Programmable Logic Device) in addition to or instead of the above configuration. Examples of PLDs include PAL (Programmable Array Logic), GAL (Generic Array Logic), CPLD (Complex Programmable Logic Device), and FPGA (Field Programmable Gate Array). In this case, part or all of the functions implemented by the CPU 901 may be implemented by the integrated circuit.

Examples of the auxiliary storage device 903 include magnetic disks, magneto-optical disks, optical disks, and semiconductor memories. Auxiliary storage device 903 may be an internal medium directly connected to the bus of computer 900, or an external storage device 910 connected to computer 900 via interface 904 or a communication line. Further, when this program is delivered to the computer 900 via a communication line, the computer 900 receiving the delivery may develop the program in the main storage device 902 and execute the above process. In at least one embodiment, secondary storage 903 is a non-transitory, tangible storage medium.

Also, the program may be for realizing part of the functions described above.
Furthermore, the program may be a so-called difference file (difference program) that implements the above-described functions in combination with another program already stored in the auxiliary storage device 903 .

As described above, several embodiments according to the present invention have been described, but all these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and modifications can be made without departing from the scope of the invention. These embodiments and their modifications are included in the gist and technical scope of the invention.

For example, in each of the above-described embodiments, the example in which the learning unit 200 uses normal GNSS positioning data as learning data has been described, but the present invention is not limited to this. For example, the state observation unit 102 may further collect GNSS positioning data at the time of abnormality. A label indicating "abnormal" is added to the GNSS positioning data in the abnormal state. Also, the learning unit 200 gives a penalty when the generator 2001 generates pseudo GNSS positioning data close to the GNSS positioning data labeled "abnormal". Thereby, the generator 2001 (generative model) can be learned to generate only pseudo GNSS positioning data close to the GNSS positioning data in the normal state.

Further, in each of the above-described embodiments, an example in which the

abnormality determination units

202 and 112 determine the presence or absence of an abnormality based on the degree of matching between the GNSS positioning data x and the pseudo GNSS positioning data x' has been described, but the present invention is limited to this. never. In another embodiment, the

abnormality determination units

202 and 112 may determine the presence or absence of an abnormality using a discriminator (discriminator 2002) learned by the learning unit 200. FIG. For example, the discriminator (discriminator 2002) outputs as the discrimination result P the probability of discriminating the input data as authentic. At this time, when the probability P1 that the GNSS positioning data x is genuine is lower than the probability P2 that the pseudo GNSS positioning data x' is genuine, the

abnormality determination units

202 and 112 determine that the GNSS positioning data x is abnormal. do. Moreover, in order to allow some error, the

abnormality determination units

202 and 112 may determine that the GNSS positioning data x is abnormal when the probability P1 is lower than the probability P2 by a predetermined value or more.

<Appendix>
The anomaly detection system, vehicle-mounted device, anomaly detection method, and program described in the above-described embodiments are grasped as follows, for example.

According to the first aspect of the present disclosure, an anomaly detection system (1) includes information about satellite signals received from GNSS satellites via a GNSS receiver (100) mounted on a vehicle, and estimated from the satellite signals. A state observation unit (102) that collects GNSS positioning data including information about the position of the vehicle, and a generation unit (111, 201) that generates pseudo GNSS positioning data by a generation model trained using the GNSS positioning data. ) and, when new GNSS positioning data is acquired, based on the comparison with the generated pseudo GNSS positioning data, abnormality to determine whether the newly acquired GNSS positioning data is abnormal and a determination unit (112, 202).

By doing so, the anomaly detection system can detect that the GNSS positioning data is anomalous when the GNSS positioning data actually measured by the GNSS receiver deviates from the pseudo GNSS positioning data.

According to the second aspect of the present disclosure, the anomaly detection system (1) according to the first aspect learns the generative model using the normal GNSS positioning data collected by the state observation unit (102). It further comprises a learning unit (200).

By doing so, the anomaly detection system can learn a generation model that can generate pseudo GNSS positioning data simulating the data estimated to be obtained when the GNSS receiver is in a normal state. Therefore, the anomaly detection system, by using the generative model learned in this way, from the pseudo GNSS positioning data simulating the normal data, when the actually measured GNSS positioning data deviates, GNSS positioning data can be accurately determined as abnormal.

According to the third aspect of the present disclosure, in the anomaly detection system (1) according to the second aspect, the state observation unit (102) further collects positioning errors of the GNSS positioning data, and the learning unit ( 200) further uses the positioning error to learn the generative model.

By doing so, the anomaly detection system can accurately simulate the positioning error of the actual GNSS receiver. Then, the anomaly detection system can generate pseudo GNSS positioning data that is closer to the real thing, so that the accuracy of anomaly detection can be improved.

According to a fourth aspect of the present disclosure, in the anomaly detection system (1) according to the second or third aspect, the state observation unit (102) is a receiver capable of detecting characteristics of the GNSS receiver (100) Collecting further information, the learning unit (200) further uses the receiver information to learn the generative model.

By doing so, the anomaly detection system can generate more realistic pseudo GNSS positioning data according to the characteristics of the GNSS receiver. As a result, it is possible to perform anomaly detection in consideration of the difference in characteristics of each GNSS receiver, and improve the accuracy of anomaly detection.

According to a fifth aspect of the present disclosure, in the anomaly detection system (1) according to any one of the second to fourth aspects, the state observation unit (102) is installed in the installation environment of the GNSS receiver (100). and the learning unit (200) further uses the mounting environment information to learn the generative model.

By doing so, the anomaly detection system can generate more realistic pseudo GNSS positioning data according to the installation environment of the GNSS receiver. As a result, it is possible to perform anomaly detection in consideration of the installation environment of the GNSS receiver, and improve the accuracy of anomaly detection.

According to a sixth aspect of the present disclosure, in the anomaly detection system (1) according to any one of the second to fifth aspects, the state observation unit (102) provides surrounding environment information about the surrounding environment of the vehicle is further collected, and the learning unit (200) further uses the surrounding environment information to learn the generative model.

By doing so, the anomaly detection system can generate more realistic pseudo GNSS positioning data that is more realistic according to the surrounding environment of the vehicle. As a result, it is possible to perform anomaly detection in consideration of the shielding situation around the vehicle, thereby improving the accuracy of anomaly detection.

According to the seventh aspect of the present disclosure, in the anomaly detection system (1) according to any one of the second to sixth aspects, the learning unit (200) uses the GNSS positioning data to use the pseudo A latent variable model that outputs latent variables for generating GNSS positioning data is further learned, and the generation unit (111, 201) inputs the latent variables output from the latent variable model to the generation model. , to generate the pseudo-GNSS positioning data.

If the latent variables cannot be identified, the generation unit changes the latent variables one by one to generate a plurality of pseudo-GNSS positioning data based on each latent variable, and separately generates the GNSS positioning data and the plurality of pseudo-GNSS positioning data. I had to compare. However, the anomaly detection system according to the above aspect can identify latent variables using a learned latent variable model, so it is possible to greatly reduce the processing load on the generation unit and the anomaly determination unit.

According to the eighth aspect of the present disclosure, the vehicle-mounted device (10) is estimated from the information about the satellite signal received from the GNSS satellite and the satellite signal via the GNSS receiver (100) mounted on the vehicle A state observation unit (102) that collects GNSS positioning data including information about the position of the vehicle, a generation unit (111) that generates pseudo GNSS positioning data by a generation model trained using the GNSS positioning data, an abnormality determination unit (112) that determines, when new GNSS positioning data is acquired, whether or not the GNSS positioning data is abnormal based on comparison with the generated pseudo GNSS positioning data; Prepare.

By doing so, it is possible for the on-board equipment installed in each vehicle to detect anomalies in the GNSS positioning data it has collected. This eliminates the need for the anomaly detection system to frequently transmit and receive evaluation data (GNSS positioning data) between the vehicle-mounted device and the center server, thereby significantly reducing the amount of communication. Moreover, the load of abnormality detection processing can be distributed to each vehicle-mounted device.

According to a ninth aspect of the present disclosure, an anomaly detection method includes information about satellite signals received from GNSS satellites via a GNSS receiver (100) mounted on a vehicle, and the vehicle estimated from the satellite signals. a step of collecting GNSS positioning data including information about the position of the GNSS positioning data; generating pseudo-GNSS positioning data by a generative model trained using the GNSS positioning data; and determining whether the GNSS positioning data is abnormal based on comparison with the generated pseudo GNSS positioning data.

According to a tenth aspect of the present disclosure, the program, via a GNSS receiver (100) mounted on a vehicle, provides information about satellite signals received from GNSS satellites and the position of the vehicle estimated from the satellite signals. generating pseudo-GNSS positioning data by a generative model trained using the GNSS positioning data; generating, when new GNSS positioning data is obtained, generating The computer (900) of the vehicle-mounted device (10) is made to perform a step of determining whether or not the GNSS positioning data is abnormal based on the comparison with the pseudo GNSS positioning data thus obtained.

1 Anomaly detection system 10 Onboard device 100 GNSS receiver 101 Sensor information acquisition unit 102 State observation unit 103 Output unit 104 Storage medium 111 Generation unit 112 Abnormality determination unit 20 Center server 200 Learning unit 2001 Generator 2002 Discriminator 2003 Encoder 201 Generation unit 202 Abnormality determination unit 203 Storage medium 30 In-vehicle sensor 40 GNSS satellite 900 Computer 901 CPU
902 Main storage device 903 Auxiliary storage device 904 Interface 910 External storage device

Claims

a state observer that collects GNSS positioning data, including information about satellite signals received from GNSS satellites and information about the position of the vehicle deduced from the satellite signals, via a GNSS receiver mounted on the vehicle;
A generation unit that generates pseudo GNSS positioning data by a generation model learned using the GNSS positioning data;
an abnormality determination unit that, when new GNSS positioning data is acquired, determines whether or not the newly acquired GNSS positioning data is abnormal based on comparison with the generated pseudo GNSS positioning data; ,
anomaly detection system.
Further comprising a learning unit that learns the generative model using the normal GNSS positioning data collected by the state observation unit,
The anomaly detection system according to claim 1.
The state observation unit further collects positioning errors of the GNSS positioning data,
The learning unit further uses the positioning error to learn the generative model.
The anomaly detection system according to claim 2.
The state observation unit further collects receiver information capable of detecting characteristics of the GNSS receiver,
The learning unit learns the generative model further using the receiver information.
The anomaly detection system according to claim 2 or 3.
The state observation unit further collects mounting environment information indicating the mounting environment of the GNSS receiver,
The learning unit learns the generative model further using the mounting environment information.
The anomaly detection system according to any one of claims 2 to 4.
The state observation unit further collects surrounding environment information about the surrounding environment of the vehicle,
The anomaly detection system according to any one of claims 2 to 5, wherein the learning unit learns the generative model further using the surrounding environment information.
The learning unit uses the GNSS positioning data to further learn a latent variable model that outputs latent variables for generating the pseudo GNSS positioning data,
The generation unit inputs the latent variables output from the latent variable model to the generation model to generate the pseudo GNSS positioning data.
The anomaly detection system according to any one of claims 2 to 6.
a state observer that collects GNSS positioning data, including information about satellite signals received from GNSS satellites and information about the position of the vehicle deduced from the satellite signals, via a GNSS receiver mounted on the vehicle;
A generation unit that generates pseudo GNSS positioning data by a generation model learned using the GNSS positioning data;
When new GNSS positioning data is acquired, an abnormality determination unit that determines whether or not the GNSS positioning data is abnormal based on comparison with the generated pseudo GNSS positioning data;
In-vehicle device with
collecting GNSS positioning data, via a GNSS receiver onboard a vehicle, including information about satellite signals received from GNSS satellites and information about the position of the vehicle deduced from the satellite signals;
generating pseudo GNSS positioning data by a generative model trained using the GNSS positioning data;
When new GNSS positioning data is acquired, determining whether the GNSS positioning data is abnormal based on comparison with the generated pseudo GNSS positioning data;
An anomaly detection method comprising:
collecting GNSS positioning data, via a GNSS receiver onboard a vehicle, including information about satellite signals received from GNSS satellites and information about the position of the vehicle deduced from the satellite signals;
generating pseudo GNSS positioning data by a generative model trained using the GNSS positioning data;
When new GNSS positioning data is acquired, determining whether the GNSS positioning data is abnormal based on comparison with the generated pseudo GNSS positioning data;
A program that causes the on-board computer to execute