CN115640828A - Vehicle-mounted digital twin cheating detection method based on antagonistic generation network - Google Patents
Vehicle-mounted digital twin cheating detection method based on antagonistic generation network Download PDFInfo
- Publication number
- CN115640828A CN115640828A CN202110817554.0A CN202110817554A CN115640828A CN 115640828 A CN115640828 A CN 115640828A CN 202110817554 A CN202110817554 A CN 202110817554A CN 115640828 A CN115640828 A CN 115640828A
- Authority
- CN
- China
- Prior art keywords
- data
- vehicle
- digital twin
- model
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention discloses a vehicle-mounted digital twin cheating detection method based on a countermeasure generation network. The scheme consists of two deep learning models, namely a fake data generator and a cheat detector, and fake data are generated and cheat is detected. The LSTM model was introduced as a spurious data generator model that utilizes global navigation satellite system/CAN/inertial measurement unit data to produce spurious data. And introducing the DenseNet as a deception detector model, and predicting according to the longitude and latitude, the speed/acceleration and the three-axis acceleration/angular velocity of the real data and the forged data. The method and the device realize the generation of the countermeasure scheme, are applied to the digital twin vehicle scene, do not need additional hardware facilities, and have low cost and portability.
Description
Technical Field
The invention belongs to the technical safety field of automatic driving, and relates to a vehicle-mounted digital twin cheating detection method and a vehicle-mounted digital twin cheating detection system based on a countermeasure generation network.
Background
With the advancement of 5G and big data technologies, vehicles are becoming increasingly interconnected and intelligent. However, network security attacks may result in life-threatening situations due to the large attack surface of the vehicle, which highlights the need for efficient security monitoring and intrusion detection systems. In a network security attack, gnss spoofing has a significant impact on vehicle security. Misleading position and speed can cause the vehicle to go to the wrong place, resulting in an accident. The spoof detection method requires computing power, while the vehicle does not deploy computing power. To address this issue, digital twin on-board networks are emerging means for future vehicle applications. The twinning of maintenance vehicle components enables the edge calculation unit to continuously monitor sensor data relating to the vehicle, detect spoofed messages, and in turn send feedback to the vehicle.
The digital twin is a full life cycle process of reflecting corresponding entity equipment by integrating multidisciplinary, multi-physical quantity, multi-scale and multi-probability simulation processes by fully utilizing data such as physical models, sensor updating, operation histories and the like and completing mapping in a virtual space. The digital twin is introduced as a map of the real driving vehicle. The edge calculation unit uses the sensor data contained in the digital twin to accomplish the anti-spoofing task, which is difficult for the calculation unit in the vehicle. This means that computing hardware needs to be added to the vehicle, which means a significant cost.
Disclosure of Invention
In order to solve the defects in the prior art, the invention aims to provide a vehicle-mounted digital twin cheating detection method based on a countermeasure generation network.
The invention provides a vehicle-mounted digital twin cheating detection method based on a countermeasure generation network. The factory does not need to add additional anti-spoofing hardware to deploy for the vehicle. The method is also not limited to vehicles with specific hardware, which can be used by most vehicles. The deep learning model applied by the invention comprises a fake data generator and a cheat detector. The spurious data generator may generate spurious data to address anomalous data source acquisition. The spoof detector is trained not only with a training data set of normal data, but also with spurious data generated by the generator. Therefore, the discrimination result of the spoof detector is not biased toward the training data set, and an unknown attack can be detected by means of the falsification data generated by the falsification data generator.
The invention provides a vehicle-mounted digital twin cheating detection method which comprises the following steps:
step 1: constructing a vehicle-mounted network system architecture based on a digital twin, broadcasting and distributing a global model by taking a Road Side Unit (RSU) as a distributed aggregator, and taking an edge server as a medium for information transmission between vehicles;
step 2: analyzing and preprocessing multidimensional data of a global navigation satellite system, a controller local area network and an inertia measurement unit;
and 3, step 3: training by using real data based on long-short term memory artificial neural network LSTM machine learning, and constructing a forged data generator model;
and 4, step 4: generating spurious data using a trained spurious data generator;
and 5: the spoof detector model was constructed using the dense convolutional network densnet and trained using supervised learning of the spurious data generated by the generator model in step 4 and the normal data from the data set comma2k19 (proposed by Harald Schafer) to enable it to determine the authenticity of the sample.
First, the present invention detects global navigation satellite system spoofing by analyzing multidimensional data from a global navigation satellite system, a controller area network, and an inertial measurement unit without requiring additional hardware devices. The multi-dimensional data comprises longitude and latitude, speed/acceleration and three-axis acceleration/angular velocity.
In addition, a generation countermeasure scheme for generating counterfeit data and detecting global navigation satellite system spoofing is also presented. The detection model is not biased towards the training data set because spurious data is introduced to train the detection model.
The system model used by the invention is a vehicle-mounted network architecture based on digital twinning. The vehicle-mounted network architecture based on the digital twin is divided into two layers: a physical layer and a digital layer.
1. Physical layer: the driving vehicle on the physical level consists of a cellular hotspot, a Global Navigation Satellite System (GNSS), a navigation device and a driver. Commands within the navigation apparatus are feedback from the road side unit, thus eliminating the need for advanced hardware within the vehicle. The cellular hotspot provides the navigation equipment with the capability of transmitting data to the RSU, and the transmission can be realized by using a 5G LTE sim card. The driven vehicle may communicate with the digital twin in this manner. The global navigation satellite system provides latitude, longitude and speed to the navigation device through an internal cable. The navigation apparatus receives feedback from the global navigation satellite system and the roadside unit and provides driving advice to the driver. The driver controls the speed and position according to the navigation device to compensate the operation of the vehicle. The driver cannot act exactly according to the recommendations of the navigation device, so there is a driver model in the edge server, simulating a real driver compensating the vehicle.
2. Digital layer: at the digital level, a digital twin is generated from the large amount of data transmitted by the vehicles. The functional modules of the digital layer comprise: the device comprises a map module, a driving module, a human behavior module and a power module. Wherein the map module comprises a map comprised of road types, road lengths, road directions and speed limits; the module will receive position information from the satellite navigation system and locate the vehicle within the map. The driving module predicts a vehicle path from a speed and a position of the vehicle. The human behavior module compares vehicle conditions to a driving schedule and compensates the vehicle. The power module analyzes the driving condition and sends a report to the navigation device.
In the vehicle-mounted network system architecture based on digital twin, the coverage range of the road side unit is c r Indicating that the edge server is used as a transmission medium for communication between vehicles, and the coverage area is c e And (4) showing. In this architecture, the vehicle, roadside unit and edge server are described as:
v i =(lat i ,lon i ,req i (t)) (1)
rsu j =(lat j ,lon j ,c r ) (2)
es k =(lat k ,lon k ,c e ) (3)
wherein v is a vehicle, i-th vehicle in formula (1), rsu is a road side unit, j-th road side unit in formula (2), es is an edge server, k-th edge server in formula (3), lat and lon represent latitude and longitude of entity, req i (t) represents a request sent by the vehicle to the road side unit at time i. In the whole vehicle-mounted network system architecture, it needs to be ensured that the road side unit is at least within the coverage range of one edge server, and the vehicle is at least within the coverage range of one road side unit.
Generally, fraud detection methods are based on supervised learning. The classifier seeks to separate normal data from abnormal data in the data set. However, it is difficult for vehicles to collect abnormal samples, usually only normal driving data, because global positioning system spoofing may lead to traffic accidents, which raises safety issues. Furthermore, even if some outlier samples are obtained in the simulation environment, the detection model trained with a particular data set may be biased towards the training data, which makes it difficult to detect unknown attacks that do not belong to the training data set.
Aiming at the defects, the global navigation satellite system deception detection method based on the generation countermeasure mechanism provided by the invention trains a detection model by utilizing forged data and real data. Since the generated falsification data is not limited to the known attack samples, the detection model can detect the abnormal data even if the abnormal data does not appear in the training data set. The vehicle-mounted digital twin cheating detection method based on the countermeasure generating network uses two neural network models which are a fake data generator and a cheating detector respectively.
In step 2, the preprocessing refers to acquiring and analyzing multidimensional data of a global navigation satellite system, a controller local area network and an inertia measurement unit, and cleaning the data; the multidimensional data comprises latitude/longitude of a global navigation satellite system, speed/acceleration of a controller area network and triaxial acceleration/angular velocity of an inertial measurement unit, so as to describe the current vehicle condition.
In step 3, the spurious data generator is pre-trained using the normal data. The spurious data generator model architecture is based on LSTM, which is a representative Recurrent Neural Network (RNNs). LSTM is an RNN architecture intended for better storage and access of information, and the network architecture of the spurious data generator is shown in figure 3.
The input layer is a state sequence with time as a sequence, and each state unit contains multidimensional data of a global navigation satellite system, a controller local area network and an inertia measurement unit at the current moment; the LSTM layer is expanded as shown in fig. 4; the dense layer converts the output of the LSTM layer into the required dimensions; the output layer outputs the prediction result.
The input to the spurious data generator model is at different times t 0 ,t 1 ,t 2 8230a state sequenceModel training based on LSTM, the generator model is trained to predict, for a given sequence of states, the most likely next state in each time stepThe value of (c). Due to the internal loading of RNNs, which contain context information for past elements, the next state can be predicted from a given sequence of states.
As in FIG. 4, X t Indicates the state at time t, h t An output representing the time at which t is present,it is shown that the operation of dot-product,represents an addition operation, sigma represents a sigmoid activation function, maps a real input to [0,1]In the scope, the sigmoid activation function and a point multiplication constitute a gate structure, and the LSTM has three gates for protecting and controlling the state of the unit. tanh represents a tanh activation function, mapping a real input to [ -1,1]Within the range. There are three LSTM cells in the figure.
The pseudo-data generator model predicts the class of the next state based on the previous state calculated by the LSTM layer and the input state for the time step. The classification cross entropy loss function is used as the probability of the output state of the dense layer (e.g., fig. 6).
Where N represents the number of states, t i Representing the next target state, theta, of a given sequence i Representing softmax activation generation vector. Theta is a value of i The calculation method of (a) is shown as follows:
wherein, prop i The ith element of the output logits vector representing the dense layer.
The softmax activation described above normalizes an N-dimensional vector prop to an N-dimensional vector θ in the range of (0, 1) and with a sum of 1, where prop represents the entire logits vector of the dense layer output.
In summary, the vehicle state sequence is fed to the LSTM layer with 256 units. The LSTM layer extracts context information for a given sequence. Then, the context information is input into the dense layer, a logits vector is output, the log-likelihood value of the next state is predicted, and the output size of the dense layer is the same as the length of the state vector in the state sequence. The prediction of the next state is obtained by sampling the probabilities of the output states from the dense layer.
In step 4, the trained spurious data generator is used to generate spurious data. Once the generator model is trained, it can generate a sequence of states that simulates the real state sequence, which can be generated by feeding back the state of the time step to the generator model. The generator model predicts the distribution of the next state based on a given sequence of states. The index of the next state is then obtained by sampling from the predicted probability distribution.
In step 5, a deception detector model is constructed, and the deception detector model is trained to detect the anomalous data. The present invention uses a dense convolutional network (DenseNet) as a detection model that contains shorter connections between layers near the input and layers near the output. DenseNet can alleviate the vanishing gradient problem, enhance feature propagation, encourage feature reuse, and reduce the number of parameters substantially.
The detection model consists of a convolutional layer, a dense block (shown in figure 7), a transition layer and a global average aggregation layer, wherein the dense block comprises a BN layer, a Leaky ReLU layer and a convolutional layer, and the transition layer comprises the BN layer, the convolutional layer and the average aggregation layer.
As in the dense block model of FIG. 7, the input is a data x 0 Neural network through layer l, where the non-linear transformation at layer i is denoted as H i ,H i It may be an accumulation of various functional operations such as Batch Normalization (BN), reLU, convolution, etc. The characteristic output of the ith layer is denoted as x i . The input to the ith layer is related not only to the output of the i-1 layer, but also to the outputs of all previous layers, denoted as:
x l =H l ([x 0 ,x 1 ,…,x l-1 ])
wherein the symbol [ ] represents a splice. The nonlinear transformation H used is a combination of BN + ReLU + convolution.
The input to the detection model contains 16 x 16 state vectors. The convolutional layers create a convolutional kernel that produces an output tensor with the input layers. The Batch Normalization (BN) layer then applies a transform to keep the average output close to 0 and the output standard deviation close to 1. The Leaky ReLU layer is a Leaky version of a rectifying linear unit, which allows small gradients when the unit is inactive. The dense block is composed of a plurality of dense layers. The transition layer is considered to be the transition between dense blocks, including the Batch Normalization (BN) layer, the convolution layer, and the average collection layer. And the average gathering layer carries out average gathering operation on the spatial data. The output of the detection model is binary information.
The detection model is trained by supervised learning using spurious data generated by the generator model and authentic data from the data set. Therefore, training the detection model is considered a binary classification problem. Samples of normal data and spurious data are labeled 0 and 1, respectively.
Similar to training the generator model using class cross entropy, the detection model is trained using class cross entropy loss, because the detection model separates the input samples into normal and abnormal classes. The class cross entropy loss is calculated by equation (4), where the number of N output classes is set to 2.
Based on the method, the invention also provides a vehicle-mounted digital twin cheating detection system based on the countermeasure generating network, which comprises the following steps: a memory and a processor; the memory has stored thereon a computer program which, when executed by the processor, carries out the above-mentioned method.
Based on the above method, the present invention also proposes a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the above method.
The invention has the beneficial effects that: 1. digital twins were introduced as a reflection of a truly driven vehicle. The edge calculation unit uses the sensor data contained in the digital twin to accomplish the anti-spoofing task, which is difficult for the calculation unit in the vehicle. The method provided by the invention is based on machine learning and sensor data in the vehicle, does not need additional hardware facilities, and is low in cost and portable. The factory does not need to add additional anti-spoofing hardware to deploy for the vehicle. The method is also not limited to vehicles with specific hardware, which can be used by most vehicles; 2. the method provided by the invention comprises a generator and a detector. The generator may generate spurious data to address anomalous data source acquisition. The detector is trained not only with the training data set, but also with spurious data. Thus, the detector is not biased towards the training data set, and unknown attacks can be detected by means of spurious data generated by the generator. Thereby, a spoofing signal in the in-vehicle network can be detected.
Drawings
Fig. 1 is a flow chart of an in-vehicle digital twin spoofing detection method based on a countering generating network of the present invention.
Figure 2 is a diagram of the digital twin based on the network architecture on board the vehicle of the present invention.
Fig. 3 is a diagram of a network architecture of a spurious data generator model of the present invention.
Fig. 4 is a diagram of a network structure of an LSTM layer model in the falsified data generator according to the present invention.
FIG. 5 is a diagram of a network architecture of the detection model of the present invention.
Fig. 6 is a diagram of a dense layer network structure of the present invention.
Fig. 7 is a diagram of a dense block network architecture in the detection model of the present invention.
Detailed Description
The invention is further described in detail with reference to the following specific examples and the accompanying drawings. The procedures, conditions, experimental methods and the like for carrying out the present invention are general knowledge and common general knowledge in the art except for the contents specifically mentioned below, and the present invention is not particularly limited.
The invention discloses a vehicle-mounted digital twin cheating detection method based on a countermeasure generation network. The scheme consists of two deep learning models, namely a fake data generator and a cheat detector, and fake data are generated and cheat is detected. The LSTM model was introduced as a spurious data generator model that utilizes global navigation satellite system/CAN/inertial measurement unit data to produce spurious data. And introducing the DenseNet as a deception detector model, and predicting according to the longitude and latitude, the speed/the acceleration and the triaxial acceleration/the angular velocity of the real data and the forged data. The method and the device realize the generation of the countermeasure scheme, are applied to the digital twin vehicle scene, do not need additional hardware facilities, and have low cost and portability.
As shown in fig. 1, an in-vehicle digital twin spoofing detecting method based on a countermeasure generating network is divided into 4 steps,
step 1: constructing a vehicle-mounted network system architecture based on digital twin, broadcasting and distributing a global model by taking a Road Side Unit (RSU) as a distributed aggregator, and taking an edge server as a medium for information transmission between vehicles;
and 2, step: and analyzing and preprocessing the multidimensional data of the global navigation satellite system, the controller local area network and the inertial measurement unit. A global navigation satellite system data set including latitude, longitude, speed, timestamp, altitude, and angle; the controller area network CAN data set consists of a CAN timestamp, a speed and a steering angle; the data set of the inertial measurement unit consists of accelerometers in three directions, and the data are preprocessed to be used as input of a fake data generator.
And step 3: training by using real data based on Long Short Term Memory artificial neural network (LSTM) machine learning, and constructing a forged data generator model; the LSTM layer in the generator model selects 256 LSTM units, and the activation function selected by the forgetting gate is f t =σ[W f ·(h t-1 ,x t )+b f ]The input gate is selected to have an activation function of i t =σ[W i ·(h t-1 ,x t )+b i ]The output gate selects the activation function as o t =σ[W o ·(h t-1 ,x t )+b o ]Wherein x is t As input to the current LSTM cell, h t-1 As a result of the output of the previous LSTM cell, W f ,W i ,W o To be applied to new input x t Connection weight of b f ,b i ,b o Corresponding deviations.
The output of the LSTM layer is used as the input of the dense layer, the output of the LSTM layer is converted into the required dimensionality, a classification cross entropy loss function is used as the probability of an output state in the dense layer, and the classification cross entropy loss function is Where N represents the number of states, t i Representing the next target state, theta, of a given sequence i Representing softmax activation generation vector.
And 4, step 4: generating fake data by using a trained fake data generator, wherein the data format generated by the fake data is consistent with the real data;
and 5: the spoof detector model is constructed using dense convolutional network DenseNet and trained using supervised learning of the spurious data generated by the generator model in step 4 and the normal data from the data set, enabling it to determine the authenticity of the sample. The input to the detection model contains 16 x 16 state vectors. The convolutional layer creates a convolution kernel, which is convolved with the input layer to produce the output tensor, the kernel size being 2 x 2 with a step size of 1. The dense block has a structure of 5-layer network, the growth rate is 4, and the input of each layer of network is all the feature maps output in the previous network. And arranging a transition layer in the middle of the dense block to realize downsampling, wherein the transition layer consists of a batch normalization layer (BN), a convolution layer (1 multiplied by 1) and an average collection layer (2 multiplied by 2). The detection model uses three dense block units, with feature map sizes of 32 × 32, 16 × 16, and 8 × 8, respectively. At the end of the last dense block, a global average gather is performed, the output of the model being binary information.
The vehicle-mounted network architecture based on the digital twin is divided into two layers: respectively, a physical layer and a digital layer. In fig. 2, there is a digital twin vehicle network architecture physical layer: the dotted line is request and the solid line is feedback. The navigation apparatus receives feedback from the roadside unit and provides driving advice to the driver. Vehicle consists of the set V = { V = 0 ,v 1 8230indicates. For each vehicle v i Digital twin vehicle dtv i Position, velocity and other data collected by the onboard sensors. The road side unit is composed of RSU = { RSU = { (RSU) o ,rsu 1 \8230; } represents receiving data and requests from vehicles and providing feedback to the relevant vehicles. Data is continuously updated from vehicle to digital twins, so that digital twins can be considered as trueAll copies of the entity. The vehicles may simultaneously send requests to the road side units, the requests being expressed as Req (t) = { Req = 0 (t),req 1 (t), \8230, wherein req i (t) represents a request from time t. The roadside unit is considered as a relay point and cannot solve the calculation task. Thus, the edge server is deployed in a specific area to address requests from vehicles with sensor data contained in the digital twin. Edge server denoted ES = { ES = { ES 0 ,es 1 8230j. The roadside unit may communicate with the vehicle and the edge server over a transmission range.
The protection content of the present invention is not limited to the above embodiments. Variations and advantages that may occur to those skilled in the art may be incorporated into the invention without departing from the spirit and scope of the inventive concept, and the scope of the appended claims is intended to be protected.
Claims (10)
1. A vehicular digital twin spoofing detection method based on a countermeasure generating network is characterized by comprising the following steps:
step 1: constructing a vehicle-mounted network system architecture based on digital twin, broadcasting and distributing a global model by taking a Road Side Unit (RSU) as a distributed aggregator, and taking an edge server as a medium for information transmission between vehicles;
step 2: analyzing and preprocessing multi-dimensional data of a global navigation satellite system GNSS, a controller area network CAN and an inertial measurement unit IMU;
and step 3: training by using real data based on long-short term memory artificial neural network LSTM machine learning, and constructing a forged data generator model;
and 4, step 4: generating spurious data using a trained spurious data generator;
and 5: the spoof detector model was constructed using the dense convolutional network DenseNet, trained using supervised learning of the spurious data generated by the generator model in step 4 and the normal data of the data set, to enable it to determine the authenticity of the sample.
2. The vehicle-mounted digital twin deception detection method based on the antagonistic generation network according to the claim 1, wherein in the step 1, the constructed vehicle-mounted digital twin network system architecture based on the digital twin is divided into two layers, namely a physical layer and a digital layer; the physical layer is a structure consisting of a cellular hotspot, a global navigation satellite system, navigation equipment and a driver; in the digital layer, the digital twin is generated from a large amount of data transmitted by the vehicle.
3. The vehicular digital twin spoofing detection method based on the countermeasure generation network according to claim 2, wherein the cellular hotspot is used for data transmission between a navigation device and a Road Side Unit (RSU); the global navigation satellite system provides latitude, longitude and speed for the navigation equipment through an internal cable; the navigation equipment is used for receiving data information of a global navigation satellite system and a Road Side Unit (RSU); the driver compensates the operation of the vehicle based on information provided by the navigation device.
4. The vehicular digital twin spoofing detection method based on the countermeasure generation network according to claim 2, characterized in that the digital layer includes the following functional modules: the device comprises a map module, a driving module, a human behavior module and a power module; wherein, the first and the second end of the pipe are connected with each other,
the map module comprises a map consisting of road types, road lengths, road directions and speed limits, and is used for receiving position information from a satellite navigation system and positioning a vehicle in the map; the driving module is used for predicting a vehicle path through the speed and the position of the vehicle; the human behavior module is used for comparing the vehicle condition with a driving plan and compensating the vehicle; the power module is used for analyzing the driving condition and sending a report to the navigation equipment.
5. The vehicle-mounted digital twin spoofing detecting method based on the antagonistic generating network according to the claim 1, characterized in that in the step 2, the preprocessing refers to acquiring and analyzing multidimensional data of a global navigation satellite system, a controller area network and an inertial measurement unit, and performing data cleaning; the multi-dimensional data includes latitude and longitude, velocity/acceleration, and three-axis acceleration/angular velocity.
6. The vehicle-mounted digital twin spoofing detection method based on antagonistic generating network as claimed in claim 1, characterized in that in step 3, the forged data generator model is based on long-short term memory artificial neural network LSTM, and the real data is used for training the generator, so that the generator produces the output of simulated real data, and the sequence of forged states.
7. The vehicular digital twin spoofing detecting method based on the antagonistic generation network according to claim 1, wherein in step 4, the fake data generator trained in step 3 is used to generate the fake state sequence data as the input of the detector in the subsequent step.
8. The vehicular digital twin spoofing detection method based on the countermeasure generation network according to claim 1, wherein in step 5, the constructed detector model is based on a dense convolution network DenseNet, and a detection model is trained by taking real data and fake data generated in step 4 as input; due to the introduction of spurious data to train the detection model, the detection model is not biased towards the training data set.
9. An on-board digital twin spoofing detection system based on a countermeasure generation network, comprising: a memory and a processor;
the memory has stored thereon a computer program which, when executed by the processor, implements the method of any of claims 1-8.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the method according to any one of claims 1-8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110817554.0A CN115640828A (en) | 2021-07-20 | 2021-07-20 | Vehicle-mounted digital twin cheating detection method based on antagonistic generation network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110817554.0A CN115640828A (en) | 2021-07-20 | 2021-07-20 | Vehicle-mounted digital twin cheating detection method based on antagonistic generation network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115640828A true CN115640828A (en) | 2023-01-24 |
Family
ID=84939902
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110817554.0A Pending CN115640828A (en) | 2021-07-20 | 2021-07-20 | Vehicle-mounted digital twin cheating detection method based on antagonistic generation network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115640828A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117407771A (en) * | 2023-12-13 | 2024-01-16 | 湘江实验室 | Bearing health state assessment method and device based on digital twin and related equipment |
-
2021
- 2021-07-20 CN CN202110817554.0A patent/CN115640828A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117407771A (en) * | 2023-12-13 | 2024-01-16 | 湘江实验室 | Bearing health state assessment method and device based on digital twin and related equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111506980B (en) | Method and device for generating traffic scene for virtual driving environment | |
CN108423006A (en) | A kind of auxiliary driving warning method and system | |
Peng et al. | Uncertainty evaluation of object detection algorithms for autonomous vehicles | |
CN105976450A (en) | Unmanned vehicle data processing method and device, and black box system | |
US11699085B2 (en) | Methods and arrangements to identify activation profile context in training data | |
CN110807924A (en) | Multi-parameter fusion method and system based on full-scale full-sample real-time traffic data | |
CN112193252A (en) | Driving risk early warning method and device, computing equipment and storage medium | |
Akowuah et al. | Physical invariant based attack detection for autonomous vehicles: Survey, vision, and challenges | |
Li et al. | Detecting stealthy cyberattacks on automated vehicles via generative adversarial networks | |
Han et al. | ADS-lead: Lifelong anomaly detection in autonomous driving systems | |
Hassan et al. | Road anomaly classification for low-cost road maintenance and route quality maps | |
Jiang et al. | Road context-aware intrusion detection system for autonomous cars | |
CN115640828A (en) | Vehicle-mounted digital twin cheating detection method based on antagonistic generation network | |
CN108932839B (en) | Method and device for judging vehicles in same-driving mode | |
CN113037750B (en) | Vehicle detection data enhancement training method and system, vehicle and storage medium | |
Peng et al. | Evaluation of emergency driving behaviour and vehicle collision risk in connected vehicle environment: A deep learning approach | |
US20230267549A1 (en) | Method of predicting the future accident risk rate of the drivers using artificial intelligence and its device | |
CN108900634A (en) | A kind of remote vehicle control method and system based on V2X | |
Sharma et al. | Towards an AI-based after-collision forensic analysis protocol for autonomous vehicles | |
CN112455460B (en) | Vehicle control method, device, equipment and storage medium | |
CN206181239U (en) | Wisdom city developments video monitor system | |
Li et al. | Research on abnormal monitoring of vehicle traffic network data based on support vector machine | |
Das et al. | DriCon: On-device Just-in-Time Context Characterization for Unexpected Driving Events | |
EP4257443A1 (en) | Method and system for automatic driving data collection and closed-loop management | |
EP4277202A1 (en) | Threat detection for a processing system of a motor vehicle |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication |