CN115116213A - Internet of vehicles trust management method based on improved fully-connected neural network - Google Patents

Internet of vehicles trust management method based on improved fully-connected neural network Download PDF

Info

Publication number
CN115116213A
CN115116213A CN202210549002.0A CN202210549002A CN115116213A CN 115116213 A CN115116213 A CN 115116213A CN 202210549002 A CN202210549002 A CN 202210549002A CN 115116213 A CN115116213 A CN 115116213A
Authority
CN
China
Prior art keywords
trust
message
vehicle
neural network
data
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.)
Granted
Application number
CN202210549002.0A
Other languages
Chinese (zh)
Other versions
CN115116213B (en
Inventor
徐红云
旷勇
蒋睿
刘志康
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN202210549002.0A priority Critical patent/CN115116213B/en
Publication of CN115116213A publication Critical patent/CN115116213A/en
Application granted granted Critical
Publication of CN115116213B publication Critical patent/CN115116213B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0108—Measuring and analyzing of parameters relative to traffic conditions based on the source of data
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W84/00—Network topologies
    • H04W84/18—Self-organising networks, e.g. ad-hoc networks or sensor networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention discloses a vehicle networking trust management method based on an improved fully-connected neural network, which comprises the steps of firstly collecting road condition information by using a traffic simulation platform to generate a training data set, dividing characteristic information in the data set into two training characteristics of data trust and entity trust, then deploying a trained network model into a vehicle-mounted unit, judging interactive nodes by the vehicle-mounted unit, changing the trust value of the other side according to the judgment result, and enabling the nodes to be incapable of normally communicating when the trust value is lower than a set threshold value, thereby realizing the trust management of the vehicle networking. The vehicle networking trust management method based on the improved fully-connected neural network carries out vehicle networking trust management, simulates the moving and communication conditions of the vehicle in the road network environment as truly as possible, can extract the vehicle characteristics from the vehicle interaction messages, and has wide application scenes.

Description

Vehicle networking trust management method based on improved full-connection neural network
Technical Field
The invention relates to the technical field of a vehicle networking trust mechanism, in particular to a vehicle networking trust management method based on an improved fully-connected neural network.
Background
In recent years, the Internet of things (Internet of Vehicle) has been a new derivative of the traditional Ad-hoc Network of vehicles (Vehicle) as an important branch of the rapid development of the Internet of things. In such a heterogeneous network, ensuring the security of communication between communicating entities is a prerequisite for inter-node interaction. In previous research, researchers have generally used cryptographic methods to solve security and privacy concerns. However, there are some security issues, such as trust and reputation, which are still difficult to be realized by the conventional encryption method, and the conventional privacy protection means cannot resist the attack from the inside. Meanwhile, the car networking topological structure has high expandability, and the interaction between the car nodes is frequent and short. Therefore, in combination with entity trust and content trust, establishing an accurate and lightweight trust management mechanism is an effective way to deal with the problem of secure communication of the internet of vehicles. With the increasing expression and learning capabilities of the deep neural network and the fact that the vehicle-mounted unit has certain storage and calculation capabilities, it becomes possible to deploy the trained improved fully-connected neural network to the vehicle-mounted unit for calculating the vehicle trust value.
Disclosure of Invention
The invention aims to overcome the defects and shortcomings of the prior art, and provides a vehicle networking trust management method based on an improved fully-connected neural network, which can extract the characteristics in road condition information, truly simulate the movement and communication conditions of vehicles in a road network environment, improve the detection of the vehicles and messages, and further realize the trust management of the vehicle networking.
In order to realize the purpose, the technical scheme provided by the invention is as follows: a vehicle networking trust management method based on an improved fully-connected neural network comprises the following steps:
1) carrying out real-time traffic simulation by using an urban traffic simulation simulator SUMO and a discrete event simulation platform OMNet + +, collecting road condition information during simulation operation, and arranging the road condition information into a data set, wherein the data set comprises road condition information received by vehicles and road side units RSUs at each communication moment and corresponding trust value variation;
2) preprocessing a data set, dividing feature information in the data set into two training features of data trust and entity trust according to the requirements of a mixed trust management mechanism, taking a trust value as a training label of an improved fully-connected neural network, and then respectively normalizing the two training features and the trust value variation; the improvement of the fully-connected neural network is that an input layer of the fully-connected neural network is divided into a data trust characteristic input layer and an entity trust characteristic input layer according to the two training characteristics;
3) sending the preprocessed data set into an improved fully-connected neural network for training, converting two types of training characteristics into high-dimensional vectors after being input into respective corresponding input layers in the training process, performing characteristic fusion on the high-dimensional vectors after the two types of training characteristics are converted by a characteristic fusion module of the fully-connected neural network, inputting an output vector sequence of the characteristic fusion module into a prediction layer of the fully-connected neural network to obtain a prediction result, and mapping the prediction result into a trust value variable quantity corresponding to the input characteristics;
4) the trained improved fully-connected neural network is deployed in the simulation platform, the vehicle collects road condition information sent by other vehicles or RSUs and processes the road condition information into characteristics capable of being input by the improved fully-connected neural network, then the variable quantity of the trust value of the information source at the time of road condition information collection is predicted, the trust value is dynamically adjusted by applying a reward and punishment mechanism, and malicious or untrustworthy information sources are screened out.
Further, in step 1), the communication time is defined as the time when the nodes receive messages from other nodes, the nodes acquire and share the traffic information by sending and receiving messages, and a content Message mainly carried by one piece of traffic information is represented as: message ═ id { (id) s ,t s ,pos s ,v s ,dir s ,dis s_RSU ,R s Denotes the number id s At t s Time of day, pos s A piece of road condition information sent by a position is the main content of a data set, wherein v s 、dir s And dis s_RSU Respectively indicating that the message sender is at t s The moving speed, moving direction and distance to the nearest RSU at the moment; r s Representing road condition information, defined as R s ＝(pos a ,acc type ,dis s_a ) Wherein pos a Target position, acc, indicating the traffic status decision of the message sender type For pos decided by the sender of the message or by other nodes a Grade of road conditions, dis s_a For the message sender to pos a The distance of (d);
in the training data set, each Message corresponds to a training label, and the meaning is as follows: after the Message sender sends the ith Message, the variable quantity generated by the trust value of the Message sender is y i 。
Further, in step 2), the spatially related feature information related to the position and distance of the vehicle or the RSU is used as a training feature for measuring the trust of the entity and is expressed as a feature vector f e ，f e ＝(pos s ,pos r ,pos a ,dis s_a ,dis s_RSU ,dis r_RSU ) Wherein pos s 、pos r 、pos a Respectively indicating the location of the message sender, the location of the message receiver and the target location, dis, at which the message sender makes a road condition decision s_a For the message sender to pos a Distance of (dis) s_RSU And dis r_RSU Respectively representing the distance between the message sender and the message receiver to the respective nearest RSU;
the related information related to the message content and the characteristic information with small granularity, such as the vehicle moving speed and the vehicle moving angle, are taken as training characteristics for measuring the data trust and are expressed as a characteristic vector f d ，f d ＝(N total_msg ,N acci_msg ,Trust sender ,acc type ,v s ,dir s ,v r ,dir r ,t last_rcv ) In which N is total_msg Is the total number of messages received from the message sender, N acc_msg Is the number of received accident messages sent from the message sender, Trust sender Is a historical trust value, t, of the sender of the message last_rcv The time when the node last received the message from the message sender;
for n messages, a data set D can be generated n ，D n ＝(F entity ,F data Y), wherein the matrix F is composed of eigenvectors measuring the trust of the entity entity Is shown as
Figure BDA0003653705530000031
Figure BDA0003653705530000032
Representation matrix F entity The nth eigenvector in the matrix, T represents the transposition operation of the eigenvector or the matrix, and the matrix F formed by the eigenvectors for measuring the data trust data Is shown as
Figure BDA0003653705530000033
Figure BDA0003653705530000034
Representation matrix F data The set of trust value variation amounts corresponding to the n messages is y, and y is (y ═ y) 1 ,y 2 ,y 3 ,...,y n ) T ，y i E { -0.02,0.01}, wherein i ∈ [1, n }, and]that is, the variation of the trust value corresponding to the ith Message is expressed, and in order to improve the training effect, the variation of the trust value as the training label is mapped to 0 or 1, if y is i Negative values, map to 0; if y i Positive values map to 1.
Further, in step 3), after the feature vectors in the data set are normalized according to columns, the feature vectors are converted into high-dimensional vectors e through a full-connection hidden layer of a full-connection neural network entity And e data I.e. feature vectors f measuring trust of an entity e And a feature vector f that measures data trust d In a high-dimensional representation of (b), wherein e entity ＝F(f e ,W entity )，e data ＝F(f d ,W data ) F is the fully-connected hidden layer with activation function ReLu, W entity And W data Respectively are weight parameters of two fully-connected hidden layers;
comprehensively considering the influence of entity trust and data trust on the final trust value, converting the characteristic vector for measuring entity trust and the characteristic vector for measuring data trustAfter the training is high-dimensional, feature fusion is carried out on the two types of training features, and then the probability t of mapping the variable quantity of the trust value to be 1 is obtained through a full-connection hidden layer and a full-connection prediction layer pre ，t pre ＝σ{μ[e entity ||e data ,W hidden ],W predict Where μ is a hidden layer with an activation function of ReLu, | | represents a feature fusion concat operation, W hidden And W predict Weight parameters for the hidden layer and the predicted layer, respectively, sigma is a predicted layer with an activation function sigmoid,
Figure BDA0003653705530000041
z is an output value of the hidden layer, and sigma (z) is a value activated by the sigmoid activation function;
using the bipartite cross entropy as a loss function for evaluating the difference between the predicted value and the training label after the forward propagation of the improved fully-connected neural network, namely calculating a bipartite cross entropy loss value BCE of the improved fully-connected neural network,
Figure BDA0003653705530000042
y i represents the trust value variation, i.e. the predicted value, p (y) corresponding to the ith Message i ) Is the predicted value y i A probability of 1, n is the number of training passes.
Further, in step 4), the improved fully-connected neural network trained in step 3) is deployed in an on-board unit (OBU) of the vehicle, and the vehicle or a road test unit (RSU) evaluates a message sender by adopting the following steps after receiving a piece of road condition information:
4.1) extracting the characteristics of one road condition information, inputting the road condition information into an improved fully-connected neural network loaded by an OBU (on-board unit), if the output result is 1, giving a positive evaluation to a message sender, otherwise, regarding the message sender as that a malicious behavior is generated, and feeding the result back to a nearby RSU after evaluation;
4.2) the RSU records the evaluation summary of vehicles in the jurisdiction, dynamically adjusts the trust value of the vehicle by utilizing a reward and punishment mechanism according to the feedback condition of the vehicle, namely when the evaluation of the vehicle feedback is positive, a message sender is rewarded for adjusting the trust value upwards, otherwise, the trust value of the message sender is adjusted downwards to serve as a penalty, if the trust value of a certain vehicle node is lower than a set threshold value, the vehicle is marked as a dishonest node, meanwhile, a node identification of the vehicle is broadcasted to the vehicles in the jurisdiction, messages from dishonest nodes cannot be adopted by a message receiver, but the message receiver still evaluates the messages from the dishonest nodes and feeds back the messages to the RSU, and when the trust value of the dishonest nodes is higher than the set threshold value again, the nodes can carry out normal communication again.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. the invention closely combines a method for processing vehicle data and feature extraction in deep learning and a vehicle networking trust hybrid trust management mechanism, and deploys an improved fully-connected neural network to a vehicle-mounted unit for calculating a vehicle trust value.
2. Compared with the traditional statistical model, the method can better cope with the situation that the ratio of malicious nodes in the road network is more than 50%, expand the application situation of a trust management mechanism and improve the precision.
3. The method has wide application prospect, can be well applied to the reliability judgment and calculation of the interaction node of the vehicle in the vehicle networking environment, further realizes the detection of global malicious or dishonest nodes, and has wide application prospect.
Drawings
FIG. 1 is an architectural diagram of the method of the present invention; in the figure, RSU is a road side unit, vehicle is a vehicle, and FCNN is an improved fully-connected neural network.
FIG. 2 is a block diagram of an improved fully-connected neural network; in the figure, Data Feature is a Data trust related Feature, Entity Feature is an Entity trust related Feature, Hidden Layer, Concat represents a Feature connection operation, Prediction Layer is a Hidden Layer, and Prediction trust variation is predicted by an improved fully-connected neural network.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the embodiments of the present invention are not limited thereto.
Referring to fig. 1, the present embodiment provides a vehicle networking trust management method based on an improved fully-connected neural network, which is a method developed by combining an improved fully-connected neural network in deep learning and a hybrid trust evaluation, and can be used for vehicle networking trust management on a vehicle networking data set, and the method includes the following steps:
1) the method comprises the following steps of carrying out real-time traffic simulation by using an urban traffic simulation simulator SUMO and a discrete event simulation platform OMNet + +, collecting road condition information during simulation operation, and arranging the road condition information into a data set, wherein the data set comprises road condition information received by vehicles and road side units RSUs at each communication moment and corresponding variable quantities of trust values, and the specific conditions are as follows:
the nodes acquire and share the traffic information by receiving and sending messages, and the information Message mainly carried by one traffic report Message can be represented as follows: message ═ id { (id) s ,t s ,pos s ,v s ,dir s ,dis s_RSU ,R s Denotes the number id s At t s Time of day, pos s A road condition message sent by a position is the main content forming a training data set, wherein v s 、dir s And dis s_RSU Respectively indicating that the message sender is at t s The moving speed, moving direction and distance to the nearest RSU at the moment; r s ＝{pos acci ,acc type ,dis s_a Denotes traffic information, pos acci Target position, acc, indicating the traffic status decision of the message sender type Is pos judged by a sender or other nodes acci Grade of road conditions, e.g. acc type 0 indicates that the road condition is good, acc type A value of 1 indicates that congestion, dis, has occurred there s_a For the message sender to pos acci Of the distance of (c).
In the training data set, each Message corresponds to a label, and the meaning is as follows: after the message sender sends the message, the variable quantity generated by the trust value of the message sender is y i When y is i A value of 1 indicates that a positive rating is to be generated and that the vehicle, upon receiving this message, would increase the confidence level of the sender.
2) Preprocessing a data set, dividing feature information in the data set into two training features of data trust and entity trust according to the requirements of a mixed trust management mechanism, taking a trust value as a training label of an improved fully-connected neural network, and then respectively carrying out normalization processing on the two training features and the trust value variation; referring to fig. 2, the improvement of the fully-connected neural network is that an input layer of the fully-connected neural network is divided into a data trust characteristic input layer and an entity trust characteristic input layer according to the two types of training characteristics, and firstly, characteristic information in a message is extracted from the following dimensions respectively:
the characteristic information related to the vehicle or the space such as the position, the distance and the like of the RSU is used as a training characteristic for measuring the trust of the entity and is expressed as a characteristic vector f e ，f e ＝(pos s ,pos r ,pos a ,dis s_a ,dis s_RSU ,dis r_RSU ) Wherein pos s 、pos r 、pos a Respectively representing the position of the message sender, the position of the message receiver and the target position, dis, for judging the road condition of the message sender s_a For the message sender to pos a Distance of (dis) s_RSU And dis r_RSU Respectively, the distance of the message sender and the message receiver to the respective closest RSU.
The related information related to the message content and the characteristic information with smaller granularity, such as the vehicle moving speed, the vehicle moving angle and the like, are used as training characteristics for measuring the data trust and are expressed as a characteristic vector f d ，f d ＝(N total_msg ,N acci_msg ,Trust sender ,acc type ,v s ,dir s ,v r ,dir r ,t last_rcv ) In which N is total_msg Is the total number of messages received from the message sender, N acc_msg Is the number of received accident messages sent from the message sender, Trust sender Is a historical trust value, t, of the sender of the message last_rcv For the last time the node received a message fromThe time of the message sender message.
For n messages, a data set D can be generated n ，D n ＝(F entity ,F data Y), wherein the matrix F is composed of eigenvectors measuring the trust of the entity entity Is shown as
Figure BDA0003653705530000071
Representation matrix F entity The nth eigenvector in the matrix, T represents the transposition operation of the eigenvector or the matrix, and the matrix F formed by the eigenvectors for measuring the data trust data Is shown as
Figure BDA0003653705530000072
Representation matrix F data The set of trust value variation amounts corresponding to the n messages is y, and y is (y ═ y) 1 ,y 2 ,y 3 ,...,y n ) T ，y i E { -0.02,0.01}, wherein i ∈ [1, n }, and]that is, the variation of the trust value corresponding to the ith Message is expressed, and in order to improve the training effect, the variation of the trust value as the training label is mapped to 0 or 1, if y is i Negative values, map to 0; if y i Positive values map to 1.
3) Converting the two types of preprocessed training features into high-dimensional vectors in respective input layers, performing feature fusion on the obtained high-dimensional vectors, inputting an output vector sequence of a feature fusion module into a prediction layer, and mapping the output vector sequence into a trust value variable quantity corresponding to the input features, wherein the method comprises the following steps:
3.1) after normalizing the characteristic vectors in the data set according to columns, converting the characteristic vectors into high-dimensional vectors e through a full-connection hidden layer of a full-connection neural network entity And e data I.e. feature vectors f measuring trust of an entity e And a feature vector f for measuring data trust d In a high dimension of (a), wherein, e entity ＝F(f e ,W entity )，e data ＝F(f d ,W data ) F is the fully-connected hidden layer with activation function ReLu, W entity And W data Are respectively two full-connected hidden layersThe weight parameter of (2).
3.2) comprehensively considering the influence of entity trust and data trust on the final trust value, converting the characteristic vector for measuring entity trust and the characteristic vector for measuring data trust into high dimension, performing characteristic fusion on the two types of training characteristics, and then obtaining the probability t of mapping the trust value variable quantity to 1 through a full-connection hidden layer and a full-connection prediction layer pre ，t pre ＝σ{μ[e entity ||e data ,W hidden ],W predict Where μ is a hidden layer with an activation function of ReLu, | | represents a feature fusion concat operation, W hidden And W predict Weight parameters for the hidden layer and the predicted layer, respectively, sigma is a predicted layer with an activation function sigmoid,
Figure BDA0003653705530000081
z is the output value of the hidden layer, and σ (z) is the value after sigmoid activation function activation.
3.3) using the dichotomous cross entropy as a loss function for evaluating the difference between the predicted value and the training label after the forward propagation of the improved fully-connected neural network, namely calculating a dichotomous cross entropy loss value BCE of the improved fully-connected neural network,
Figure BDA0003653705530000082
y i represents the trust value variation (i.e. predicted value) corresponding to the ith Message, p (y) i ) Is the predicted value y i A probability of 1, n is the number of training passes.
4) The trained improved fully-connected neural network is deployed in a simulation platform, the vehicle collects information sent by other vehicles or RSUs and processes the information into characteristics which can be input by the improved fully-connected neural network, then the variable quantity of the trust value of the information source at the information acquisition time is predicted, and malicious or dishonest information sources are further screened out, wherein the specific conditions are as follows:
4.1) extracting the characteristics of a road condition message according to the method in the step 3), inputting the road condition message into an improved full-connection neural network loaded by the OBU, if the output result is 1, giving a positive evaluation to the message sender, otherwise, regarding the message sender as generating a malicious behavior, and feeding the result back to a nearby RSU after evaluation.
4.2) the RSU records the evaluation summary of vehicles in the jurisdiction, and dynamically adjusts the trust value of the vehicle by utilizing a reward and punishment mechanism according to the feedback condition of the vehicle, namely when the evaluation fed back by the vehicle is positive, a message sending party is rewarded for adjusting the trust value upwards, otherwise, the trust value of the message sending party is adjusted downwards to serve as a penalty, for example, if the feedback is positive, the trust value is increased by 0.01, and if the feedback is reverse, the trust value is subtracted by 0.02. If the trust value of a certain vehicle node is lower than the set threshold value, the vehicle is marked as a dishonest node, meanwhile, the node identification of the vehicle is broadcasted to vehicles in the jurisdiction, the message from the dishonest node cannot be adopted by a message receiver, but the message receiver still evaluates the message from the dishonest node and feeds the message back to the RSU, and when the trust value of the dishonest node is higher than the set threshold value again, the node can carry out normal communication again. The threshold may be set to 0.5 in general.
In conclusion, the invention mainly realizes the real-time traffic simulation by utilizing the urban traffic simulation simulator SUMO and the discrete event simulation platform OMNet + + to form a data set; carrying out feature extraction on the target vehicle and the message content by utilizing an improved full-connection neural network; training the improved fully-connected neural network by using simulation data, and deploying the trained improved neural network into a vehicle-mounted unit; and carrying out credibility evaluation on the received message by utilizing the improved fully-connected neural network, and realizing dynamic trust management in the car networking environment. The method is based on a deep learning and Internet of vehicles mixed trust mechanism, can be well applied to the credibility judgment and calculation of the vehicle interaction node in the Internet of vehicles environment, further realizes the detection of global malicious or dishonest nodes, and has wide application prospect.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims (5)

1. A vehicle networking trust management method based on an improved fully-connected neural network is characterized by comprising the following steps:
1) carrying out real-time traffic simulation by using an urban traffic simulation simulator SUMO and a discrete event simulation platform OMNet + +, collecting road condition information during simulation operation, and arranging the road condition information into a data set, wherein the data set comprises road condition information received by vehicles and road side units RSUs at each communication moment and corresponding trust value variation;
2) preprocessing a data set, dividing feature information in the data set into two training features of data trust and entity trust according to the requirements of a mixed trust management mechanism, taking a trust value as a training label of an improved fully-connected neural network, and then respectively normalizing the two training features and the trust value variation; the improvement of the fully-connected neural network is that an input layer of the fully-connected neural network is divided into a data trust characteristic input layer and an entity trust characteristic input layer according to the two training characteristics;
3) sending the preprocessed data set into an improved fully-connected neural network for training, converting two types of training characteristics into high-dimensional vectors after being input into respective corresponding input layers in the training process, performing characteristic fusion on the high-dimensional vectors after the two types of training characteristics are converted by a characteristic fusion module of the fully-connected neural network, inputting an output vector sequence of the characteristic fusion module into a prediction layer of the fully-connected neural network to obtain a prediction result, and mapping the prediction result into a trust value variable quantity corresponding to the input characteristics;
4) the trained improved fully-connected neural network is deployed in the simulation platform, the vehicle collects road condition information sent by other vehicles or RSUs and processes the road condition information into characteristics capable of being input by the improved fully-connected neural network, then the variable quantity of the trust value of the information source at the time of road condition information collection is predicted, the trust value is dynamically adjusted by applying a reward and punishment mechanism, and malicious or untrustworthy information sources are screened out.
2. According to claimThe internet of vehicles trust management method based on the improved fully-connected neural network of claim 1, characterized in that, in step 1), the communication time is defined as the time when the nodes receive messages from other nodes, the nodes acquire and share traffic information by sending and receiving messages, and a content Message mainly carried by a piece of traffic information is represented as: message ═ id { (id) s ,t s ,pos s ,v s ,dir s ,dis s_RSU ,R s Denotes the number id s At t s Time of day, pos s A piece of road condition information sent by a position is the main content of a data set, wherein v s 、dir s And dis s_RSU Respectively indicating that the message sender is at t s The moving speed, moving direction and distance to the nearest RSU at the moment; r s Representing road condition information, defined as R s ＝(pos a ,acc type ,dis s_a ) Wherein pos a Target position, acc, indicating the traffic status decision made by the sender of the message type For pos decided by the sender of the message or by other nodes a Grade of road conditions, dis s_a For the message sender to pos a The distance of (a);
in the training data set, each Message corresponds to a training label, and the meaning is as follows: after the Message sender sends the ith Message, the variable quantity generated by the trust value of the Message sender is y i 。
3. The Internet of vehicles trust management method based on improved fully-connected neural network as claimed in claim 1, characterized in that in step 2), the spatially-related feature information related to the position and distance of the vehicle or RSU is used as a training feature for measuring entity trust and is expressed as a feature vector f e ，f e ＝(pos s ,pos r ,pos a ,dis s_a ,dis s_RSU ,dis r_RSU ) Wherein pos s 、pos r 、pos a Respectively indicating the location of the message sender, the location of the message receiver and the target location, dis, at which the message sender makes a road condition decision s_a For the message sender to pos a Distance of (dis) s_RSU And dis r_RSU Respectively representing the distance between the message sender and the message receiver to the respective nearest RSU;
the related information related to the message content and the characteristic information with small granularity, such as the vehicle moving speed and the vehicle moving angle, are taken as training characteristics for measuring the data trust and are expressed as a characteristic vector f d ，f d ＝(N total_msg ,N acci_msg ,Trust sender ,acc type ,v s ,dir s ,v r ,dir r ,t last_rcv ) In which N is total_msg Is the total number of messages received from the message sender, N acc_msg Is the number of received accident messages sent from the message sender, Trust sender Historical trust value, t, for the sender of the message last_rcv The time when the node last received the message from the message sender;
for n messages, a data set D can be generated n ，D n ＝(F entity ,F data Y), wherein the matrix F is composed of eigenvectors measuring the trust of the entity entity Is shown as
Figure FDA0003653705520000021
Figure FDA0003653705520000022
Representation matrix F entity The nth eigenvector in the matrix, T represents the transposition operation of the eigenvector or the matrix, and the matrix F formed by the eigenvectors for measuring the data trust data Is shown as
Figure FDA0003653705520000031
Figure FDA0003653705520000032
Representation matrix F data The set of trust value variation amounts corresponding to the n messages is y, and y is (y ═ y) 1 ,y 2 ,y 3 ,...,y n ) T ，y i E { -0.02,0.01}, wherein i ∈ [1, n }, and]that is, the variation of the trust value corresponding to the ith Message is expressed, and in order to improve the training effect, the variation of the trust value as the training label is mapped to 0 or 1, if y is i Negative values, map to 0; if y i Positive values map to 1.
4. The vehicle networking trust management method based on the improved fully-connected neural network as claimed in claim 1, wherein in step 3), after the feature vectors in the data set are normalized by columns, the feature vectors are converted into high-dimensional vectors e through a fully-connected hidden layer of the fully-connected neural network entity And e data I.e. feature vectors f measuring trust of an entity e And a feature vector f for measuring data trust d In a high-dimensional representation of (b), wherein e entity ＝F(f e ,W entity )，e data ＝F(f d ,W data ) F is the fully-connected hidden layer with activation function ReLu, W entity And W data Respectively are weight parameters of two fully-connected hidden layers;
comprehensively considering the influence of entity trust and data trust on the final trust value, converting the characteristic vector for measuring entity trust and the characteristic vector for measuring data trust into high dimension, performing characteristic fusion on the two types of training characteristics, and then obtaining the probability t of mapping the trust value variable quantity to 1 through a full-connection hidden layer and a full-connection prediction layer pre ，t pre ＝σ{μ[e entity ||e data ,W hidden ],W predict Where μ is a hidden layer with an activation function of ReLu, | | represents a feature fusion concat operation, W hidden And W predict Weight parameters for the hidden layer and the predicted layer, respectively, sigma is a predicted layer with an activation function sigmoid,
Figure FDA0003653705520000033
z is an output value of the hidden layer, and sigma (z) is a value activated by the sigmoid activation function;
using bisection cross entropy as an evaluation improvementA loss function of the difference between the predicted value and the training label after the forward propagation in the fully-connected neural network is obtained, namely a binary cross entropy loss value BCE of the improved fully-connected neural network is calculated,
Figure FDA0003653705520000034
y i represents the trust value variation, i.e. the predicted value, p (y) corresponding to the ith Message i ) Is the predicted value y i A probability of 1, n is the number of training passes.
5. The vehicle networking trust management method based on the improved fully-connected neural network as claimed in claim 1, wherein in step 4), the improved fully-connected neural network trained in step 3) is deployed into an on-board unit (OBU) of a vehicle, and the vehicle or a road test unit (RSU) evaluates a message sender by following steps after receiving a piece of road condition information:
4.1) extracting the characteristics of one road condition information, inputting the road condition information into an improved fully-connected neural network loaded by an OBU (on-board unit), if the output result is 1, giving a positive evaluation to a message sender, otherwise, regarding the message sender as that a malicious behavior is generated, and feeding the result back to a nearby RSU after evaluation;
4.2) the RSU records the evaluation summary of vehicles in the jurisdiction, dynamically adjusts the trust value of the vehicle by utilizing a reward and punishment mechanism according to the feedback condition of the vehicle, namely when the evaluation of the vehicle feedback is positive, a message sender is rewarded for adjusting the trust value upwards, otherwise, the trust value of the message sender is adjusted downwards to serve as a penalty, if the trust value of a certain vehicle node is lower than a set threshold value, the vehicle is marked as a dishonest node, meanwhile, a node identification of the vehicle is broadcasted to the vehicles in the jurisdiction, messages from dishonest nodes cannot be adopted by a message receiver, but the message receiver still evaluates the messages from the dishonest nodes and feeds back the messages to the RSU, and when the trust value of the dishonest nodes is higher than the set threshold value again, the nodes can carry out normal communication again.
CN202210549002.0A 2022-05-20 2022-05-20 Internet of vehicles trust management method based on improved fully-connected neural network Active CN115116213B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210549002.0A CN115116213B (en) 2022-05-20 2022-05-20 Internet of vehicles trust management method based on improved fully-connected neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210549002.0A CN115116213B (en) 2022-05-20 2022-05-20 Internet of vehicles trust management method based on improved fully-connected neural network

Publications (2)

Publication Number Publication Date
CN115116213A true CN115116213A (en) 2022-09-27
CN115116213B CN115116213B (en) 2023-08-22

Family

ID=83326478

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210549002.0A Active CN115116213B (en) 2022-05-20 2022-05-20 Internet of vehicles trust management method based on improved fully-connected neural network

Country Status (1)

Country Link
CN (1) CN115116213B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110177370A (en) * 2019-05-31 2019-08-27 长安大学 A kind of collusion malice vehicle node detection method towards car networking
WO2020258060A2 (en) * 2019-06-25 2020-12-30 南京邮电大学 Blockchain-based privacy protection trust model for internet of vehicles
CN113301530A (en) * 2021-04-08 2021-08-24 西安电子科技大学 Method for improving vehicle networking trust based on road side unit

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110177370A (en) * 2019-05-31 2019-08-27 长安大学 A kind of collusion malice vehicle node detection method towards car networking
WO2020258060A2 (en) * 2019-06-25 2020-12-30 南京邮电大学 Blockchain-based privacy protection trust model for internet of vehicles
CN113301530A (en) * 2021-04-08 2021-08-24 西安电子科技大学 Method for improving vehicle networking trust based on road side unit

Also Published As

Publication number Publication date
CN115116213B (en) 2023-08-22

Similar Documents

Publication Publication Date Title
Alladi et al. Artificial intelligence (AI)-empowered intrusion detection architecture for the internet of vehicles
US11270579B2 (en) Transportation network speed foreeasting method using deep capsule networks with nested LSTM models
Hawlader et al. Intelligent misbehavior detection system for detecting false position attacks in vehicular networks
Zhang et al. Toward crowdsourced transportation mode identification: A semisupervised federated learning approach
CN114202120A (en) Urban traffic travel time prediction method aiming at multi-source heterogeneous data
CN114912719B (en) Heterogeneous traffic individual trajectory collaborative prediction method based on graph neural network
Khot et al. Position falsification misbehavior detection in vanets
CN114780619B (en) Abnormity early warning method for automatic engineering audit data
Zhou et al. Support vector machine and back propagation neutral network approaches for trip mode prediction using mobile phone data
Joseph et al. A novel hybrid deep learning algorithm for smart city traffic congestion predictions
Wu et al. A deep learning approach to secure vehicle to road side unit communications in intelligent transportation system
Mohanty et al. Identification and evaluation of the effective criteria for detection of congestion in a smart city
JABBAR et al. Predictive intelligence: A neural network learning system for traffic condition prediction and monitoring on freeways
CN117610734A (en) Deep learning-based user behavior prediction method, system and electronic equipment
Radi et al. Enhanced Implementation of Intelligent Transportation Systems (ITS) based on Machine Learning Approaches
CN115116213A (en) Internet of vehicles trust management method based on improved fully-connected neural network
CN110213741B (en) Method for detecting authenticity of vehicle sending information in real time based on width learning
CN111723997A (en) Automatic generation method of urban major traffic accident data sample based on GAN
Hou et al. A vehicle alarm network for high-temperature fault diagnosis of electric vehicles
Srivastava et al. Web survey data and commuter mode choice analysis using artificial neural network
Hei et al. ConvCatb: An attention-based CNN-CATBOOST risk prediction model for driving safety
Liu et al. Time-Series Misalignment Aware DNN Adversarial Attacks for Connected Autonomous Vehicles
Marouane et al. A review and a tutorial of ML-based MDS technology within a VANET context: From data collection to trained model deployment
Tkachev et al. On a Problem of the Monitoring Devices Placement on Transport Networks
Wowo et al. Using Vehicle Data to Enhance Prediction of Accident-Prone Areas

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant