CN113950113A - Hidden Markov-based Internet of vehicles switching decision algorithm - Google Patents

Hidden Markov-based Internet of vehicles switching decision algorithm Download PDF

Info

Publication number
CN113950113A
CN113950113A CN202111169755.0A CN202111169755A CN113950113A CN 113950113 A CN113950113 A CN 113950113A CN 202111169755 A CN202111169755 A CN 202111169755A CN 113950113 A CN113950113 A CN 113950113A
Authority
CN
China
Prior art keywords
vehicle
rssi
information
switching
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.)
Granted
Application number
CN202111169755.0A
Other languages
Chinese (zh)
Other versions
CN113950113B (en
Inventor
毕远国
郝晨阳
黄子烜
张东玉
王博洋
郭朋超
张娟
肖嘉池
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.)
Northeastern University China
Original Assignee
Northeastern University China
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 Northeastern University China filed Critical Northeastern University China
Priority to CN202111169755.0A priority Critical patent/CN113950113B/en
Publication of CN113950113A publication Critical patent/CN113950113A/en
Application granted granted Critical
Publication of CN113950113B publication Critical patent/CN113950113B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W36/00—Hand-off or reselection arrangements
    • H04W36/0005—Control or signalling for completing the hand-off
    • H04W36/0055—Transmission or use of information for re-establishing the radio link
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/29—Graphical models, e.g. Bayesian networks
    • G06F18/295—Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W36/00—Hand-off or reselection arrangements
    • H04W36/0005—Control or signalling for completing the hand-off
    • H04W36/0055—Transmission or use of information for re-establishing the radio link
    • H04W36/0061—Transmission or use of information for re-establishing the radio link of neighbour cell information
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W36/00—Hand-off or reselection arrangements
    • H04W36/24—Reselection being triggered by specific parameters
    • H04W36/26—Reselection being triggered by specific parameters by agreed or negotiated communication parameters
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W36/00—Hand-off or reselection arrangements
    • H04W36/24—Reselection being triggered by specific parameters
    • H04W36/32—Reselection being triggered by specific parameters by location or mobility data, e.g. speed data
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/30—Services specially adapted for particular environments, situations or purposes
    • H04W4/40—Services specially adapted for particular environments, situations or purposes for vehicles, e.g. vehicle-to-pedestrians [V2P]
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W60/00—Affiliation to network, e.g. registration; Terminating affiliation with the network, e.g. de-registration

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

The invention belongs to the technical field of vehicle networking information interaction, and provides a hidden Markov-based vehicle networking switching decision algorithm. And predicting the intensity of the receivable signals of the current connection network in a future period of time according to the intensity of the receivable signals maintained in the database, thereby judging whether network switching is needed or not. And the probability of the vehicle switching from one network to another network is obtained through elements in the hidden Markov model and the observation state information of the vehicle, and the optimal target network is determined. And after the switching is finished, modifying the hidden Markov model according to the network information before and after the switching. The problems that the switching time delay is high, the packet loss rate is high and the ping-pong effect is easy to occur in the conventional network switching algorithm are solved.

Description

Hidden Markov-based Internet of vehicles switching decision algorithm
Technical Field
The invention belongs to the field of Internet of Vehicles (IoV) bottom-layer communication protocols, and relates to a hidden Markov-based vehicle networking switching decision algorithm.
Background
With the improvement of living standard of people, the demand for real-time road safety service and information entertainment application on vehicles is continuously increased, so that good driving experience is obtained. Therefore, a high-performance internet of vehicles becomes important for various intelligent traffic applications. The mobility management mechanism can help the connection between the terminal and the network to reach the optimal state, and the switching management is a crucial process in the mobility management, so that an optimal target network can be selected when the service quality of the current connection network is poor, and the use experience of a user is ensured. However, the internet of vehicles has the characteristics of rapid movement of vehicles and frequent change of network topology, so that the network switching performance is seriously reduced.
The existing handover decision algorithms are mainly classified into five categories, namely a handover decision algorithm based on a single factor, a handover decision algorithm based on a utility function, a handover decision algorithm based on a multi-attribute decision, a handover decision algorithm based on service quality and a handover decision algorithm based on artificial intelligence. The switching decision algorithm based on single factor is single in consideration factor and is not suitable for complex environment in the Internet of vehicles. And quantifying the performance of each wireless access network by a switching decision algorithm based on the utility function, and bringing the quantified performance into the utility function, thereby judging the optimal switching target network. The switching decision algorithm based on the multi-attribute decision comprehensively considers the information of the network performance, the track of the mobile node, the habit of the user and the like, brings the factors into the commonly used multi-attribute decision algorithm to sequence the priority of the accessible network around the mobile terminal, and selects the optimal network for access. The switching decision algorithm based on artificial intelligence belongs to a new switching decision algorithm and is a popular algorithm in recent years, and the algorithm generally solves the switching problem by using artificial intelligence methods such as pattern recognition, fuzzy logic, neural network and the like. The existing handover decision algorithm is not perfect enough, certain problems exist in the aspects of handover delay and packet loss rate, and meanwhile, the phenomenon that the handover is frequently carried out back and forth between two or more networks, namely the ping-pong effect, can also occur.
Disclosure of Invention
In view of the above problems, an object of the present invention is to provide a handover decision algorithm suitable for a heterogeneous car networking environment, which effectively reduces handover delay and packet loss rate in a handover process, and avoids a ping-pong effect.
The technical scheme of the invention is as follows: firstly, a Long-Short-Term Memory network (LSTM) prediction model is used for completing the prediction of the strength of a received signal of the current connection of a vehicle, and whether the switching needs to be triggered or not and the specific time for triggering the switching are judged; secondly, introducing a Hidden Markov Model (HMM) and judging a target network which needs to be switched by the vehicle by combining the observation attribute of the vehicle; finally, in the process of system design, the compatibility of the algorithm and an LTE-V protocol stack is considered, and the whole system is divided into five modules, namely an information interaction module, an information maintenance module, a switching trigger module, a network selection module and a data forwarding module.
The information interaction module is responsible for sending and receiving information; the information maintenance module is used for storing information to be stored; the switching trigger module predicts whether switching is carried out or not; the network selection module selects an optimal switching target network; and the data forwarding module completes the encapsulation and analysis of the data.
The information interaction module is specifically used for transmitting contents such as an early registration request data packet and an update data packet to each other between the vehicle and the roadside device, and the AP also transmits an adaptive HMM element and a neighbor AP list of the AP to surrounding vehicle nodes in a broadcasting mode. The information interaction module is responsible for data interaction work among all nodes and controls the time and the type of the message to be forwarded.
The information maintenance module works in management entity DME of network layer, there are four list items of local information list, DSM service request list, application request information list and user service request message list in MIB of LTE-V protocol stack, and the list items are used to store DSM short message and message service that can be sent or need to be received. The system adds table items for realizing the switching decision algorithm, and stores and maintains the observation state of the vehicle, the neighbor network list, the HMM model and the RSSI sequence.
The switching triggering module is specifically used for helping the vehicle node to judge the Receivable Signal Strength (RSSI) of the currently connected AP in the next period in advance, so as to judge the specific time for triggering switching of the vehicle node.
In the algorithm proposed in the present invention, it is assumed that a plurality of APs exist in an area, and each AP periodically broadcasts and transmits a data packet, where the broadcast data packet includes the location and service information of the AP. The vehicle receives the data packet and estimates its RSS for the next period of time based on the information in the data packet. The position information of the vehicle itself and the position information of the AP are known, and the distance between the two is used as an important parameter for estimation. In addition, in a real internet of vehicles, obstacles such as buildings or trees affect the signal intensity of the AP, and a path loss factor is added in the calculation. The received signal strength decreases logarithmically with increasing distance, so the RSSI is expressed as:
Figure BDA0003292422800000031
wherein, dbm0Represents RSSI at 1 meter near AP, d is the distance between the vehicle node and AP, n is the path loss factor constant, d0Is a fixed value, a unit metric parameter value representing the distance between the vehicle node and the AP.
When the vehicle node is connected to the network, the RSSI list { RSSI ] of the AP currently connected with the vehicle is maintained in the window time Lt,rssit+1，…,rssit+LAnd estimating the RSSI in the next period of time according to the list, predicting in advance whether switching needs to be carried out or not so as to find and register a new care-of address, wherein a prediction model is required to obtain the next RSSI in the process. Recurrent Neural Networks (RNNs) belong to a method for time series prediction in machine learning, which uses its own output as input for the next step to retain the state from one iteration to the next. The common RNN model cannot capture the long-term dependence in the sequence, so an LSTM network is adopted to solve the problem, and the LSTM uses an effective learning algorithm based on gradient to solve the problems of exponential decay error and disappearance in the recursion process.
The LSTM network prediction model is divided into three layers, namely an input layer, a hidden layer and an output layer, wherein the input layer is an RSSI list of a vehicle node currently connected with an AP, after the list is transmitted to the hidden layer, the hidden layer updates the state of the hidden layer, and a result, namely the RSSI prediction list, is calculated according to an input value. The state of the hidden layer in the model is updated using the following formula:
Figure BDA0003292422800000041
Figure BDA0003292422800000042
where f is a nonlinear hidden layer function set to an sigmoid function and w is a bayesian vector, defining a weight matrix. The hidden layer in the LSTM network also includes memory cells and gate cells that modify the f-function in the conventional RNN model to store information, the memory cells defining some current interactions to determine which type of information to retain in memory and output hidden states.
The switching triggering algorithm based on the LSTM is divided into three steps: data preprocessing, learning, and prediction. In the algorithm, the vehicle node maintains the RSSI in the current connected AP update packet and stores it in a list in time order, trains the LSTM model in an off-line manner using the RSSI sequence as an input vector, and predicts a new RSSI sequence using the stored model parameters. In addition, when collecting RSSI values for training, it is necessary to set a window size of an input vector, which is the number of RSSI in the RSSI sequence vector required for prediction. When the window size is too small, the RSSI in the next period of time cannot be accurately predicted according to the window size, and the obtained prediction sequence has a larger difference with the actual RSSI; when the window size is too large, the training overhead is increased, so that the loss of model training is increased, and a proper window size value is selected according to the test of different window sizes so as to better adapt to the switching time and the network delay.
The network selection module is specifically the core content of a switching decision algorithm and aims to select the optimal target network to which the current vehicle is switched. The module uses hidden markov models for prediction to determine the target network to which the vehicle is to be handed off.
The HMM model of the present invention has N different states: s1，S2，…，SNEach representing a neighboring AP different in the surroundings, where a state transition occurs at each time T-1, 2, …, T, M represents the number of observed states for each state, and is a map of the vehicle motion state, xtIndicating the state at time t. An HMM model contains three parts: a state transition matrix A, an observation probability matrix B, and an initial probability distribution pi. The state transition matrix a is a matrix of probabilities that the vehicle switches between the APs, and is defined as follows:
Figure BDA0003292422800000051
wherein, ai,jRepresents a vehicle slave APiHandover to APjIs specifically expressed as
ai,j＝P(xt＝Sj|xt-1＝Si),i,j∈N. (5)
The observation probability matrix B is a matrix of probabilities that a vehicle connects to an AP in a current observation state, and is defined as follows:
Figure BDA0003292422800000052
wherein, bi,jRepresenting the vehicle being connected to the AP under current motionjIs specifically expressed as
bi,j＝P(yt＝Ok|xt＝Sj),j∈N,k∈M. (7)
O in the above formulakThe representative is the current observation state of the vehicle, which is obtained through the measured value of the vehicle and a Kalman filter. The initial probability distribution pi is the set of probabilities that a vehicle is initially connected to each AP
πi＝P(x1＝Si),1<i<N (8)
The above three elements are defined into one triple lambda,
λ＝(A,B,π) (9)
the AP to be switched to and its accompanying MAP states are hidden, while the movement attributes of the vehicle are obtained by measurement, and the probability values are trained and predicted based on the above HMM elements and the observed values of the network, mainly by two steps: adaptive learning and initial matrix estimation. An adaptive learning process is used to recursively estimate the AP probabilities, relying only on previous timestamps to update the observation matrix. The AP probability is described in two parts, based on t 1 and other time periodsInitial probability pi ═ α0(i) Get AP probability when t ═ 1:
α1(i)＝πibi(o1)ai(1) (10)
the probability of AP at time t in other time periods is
Figure BDA0003292422800000061
The initial probability distribution of AP is as follows
Figure BDA0003292422800000062
Where δ (·) is the Kronecker delta, determined by the parameter value, which is 1 if the parameter is 0 and 0 if the parameter is other values. The observation state uses w (i) instead of likelihood estimation when the switch is completed, and the weight of the ith data packet is determined by the following formula
Figure BDA0003292422800000063
Wherein, L is the serial number of the last data packet, and ensures that the weight value is not zero. The recalculation formula of the state transition matrix and the observation matrix is as follows:
Figure BDA0003292422800000064
Figure BDA0003292422800000065
wherein xpFor predicted AP nodes, xaFor an actually connected AP node, ocIs the current observed state of the vehicle at the time of the prediction.
The invention adopts the motion attributes of three vehicles to describe the observation state, namely the motion direction, the moving speed and the geographical position information of the vehicle. Under the environment of the Internet of vehicles, the vehicle acquires the geographical position information according to the vehicle-mounted equipment of the vehicle, and simultaneously acquires the position information of the AP through a data packet sent by the AP node, so that the moving direction theta of the vehicle can be calculated by the following formula
dirx＝predictx-currx (16)
diry＝predicty-curry (17)
Figure BDA0003292422800000071
Wherein the predictxAnd predictyIs the predicted coordinate information of the next position of the vehicle, currxAnd curryThe coordinate information of the current position of the vehicle is obtained, and the included angle between the motion direction of the vehicle and the x-axis direction is obtained according to the arc tangent function. After defining the angle of the moving direction of the vehicle, for convenience of use, the data is discretized into equal-width features to convert continuous values into discrete corresponding values, so that the moving direction of the vehicle is defined as 8 different directions with 45 ° as a range: n, NE, E, SE, S, SW, W, NW.
The speed s of the vehicle is obtained from a kalman filter and the current speed of the vehicle, the predicted speed is converted into a discrete data set in an observation sequence, the speed of the vehicle is divided into 6 different levels according to its magnitude, and in this model, it is assumed that the speed of the vehicle does not exceed 100 km/h.
The position information of the vehicle is the third observation sequence in the HMM model, and the two-dimensional position of the vehicle is selected for use for convenience of use, and is represented by longitude and latitude. The method comprises the steps that position information of vehicles is obtained through a vehicle-mounted GPS, estimation is carried out through a Kalman filter, the distance between the position of each vehicle and the edge of the coverage range of the currently connected AP is calculated by each vehicle, then the position of each vehicle is mapped to grid numbers on a map, and the total number of grids covered by each AP is defined by the maximum communication distance of the grid numbers.
The data forwarding module is mainly used for encapsulating and sending information to be sent layer by layer from an upper layer to a lower layer, analyzing the received information layer by layer and finally obtaining the required data; the module completes the function of information transmission between different nodes.
Drawings
Fig. 1 is a network architecture model diagram of a handover decision algorithm proposed by the present invention.
Fig. 2 is a handover flow diagram of a handover decision algorithm.
Fig. 3 is a flow chart of a handover triggering algorithm.
Fig. 4 is an overall architecture diagram of the system of the present invention.
Fig. 5 is a message format diagram of an information data packet in the information interaction module.
Fig. 6 is a flow chart of AP information packet transmission in the information interaction module.
Fig. 7 is a flowchart of transmission of an early registration request packet in the present invention.
Fig. 8 is a structural design diagram of an AP neighbor list in the present invention.
Fig. 9 is a data encapsulation flow diagram.
Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings.
As shown in fig. 1, it is a network architecture model corresponding to the handover decision algorithm proposed by the present invention. The occurrence of artificial intelligence methods such as machine learning and the like also provides a new solution for the switching problem in the Internet of vehicles, and the characteristic of learning through a large amount of data is suitable for complex and changeable scenes in the Internet of vehicles, so the invention provides a hidden Markov-based Internet of vehicles switching decision algorithm, and a neural network and an HMM are introduced into an HMIPv6 protocol. The handover in the present invention is divided into inter-domain handover for switching between networks between two different MAPs and intra-domain handover for switching between networks between the same MAPs.
Fig. 2 is a handover flow of the PRE-HMIPv6 handover decision algorithm, where the main roles are MAP, AP, and MN. The MAP is an intermediate node between the MN and its HA, and maintains a plurality of APs in a MAP domain, each of which manages HMM data and maintains a neighbor AP list. The MN will periodically use a kalman filter to track and predict the vehicle's motion attributes and be responsible for the handover prediction using HMM elements obtained from the AP. The handover decision algorithm is mainly divided into three parts: prediction of handover trigger, prediction of target network, and updating of network.
The switching triggering algorithm based on the LSTM is divided into three steps: data preprocessing, learning, and prediction. In the algorithm, the vehicle node maintains the RSSI in the current connected AP update packet and stores it in a list in time order, trains the LSTM model in an off-line manner using the RSSI sequence as an input vector, and predicts a new RSSI sequence using the stored model parameters. In addition, in collecting RSSI values for training, a window size of an input vector is set, which is the number of RSSI in the RSSI sequence vector required for prediction. When the window size is too small, the RSSI in the next period of time cannot be accurately predicted according to the window size, and the obtained prediction sequence has a larger difference with the actual RSSI; when the window size is too large, the training overhead is increased, so that the loss of model training is increased, and therefore, a value with a proper window size is selected according to tests with different window sizes to adapt to switching time and network delay.
The handover triggering algorithm is further divided into prediction of RSSI and judgment of handover triggering. Firstly, collecting RSSI observed values of a vehicle and starting a switching trigger program, if the number of the RSSIs collected by the vehicle does not reach the size of a window required by prediction, the vehicle continuously uses the current CoA to update the RSSI until the vehicle collects enough RSSI values, and then calls an RSSI prediction method based on an LSTM algorithm to obtain a predicted RSSI sequence. Then, calculating an average value through the predicted RSSI sequence and setting a threshold value for switching triggering, if the average value is lower than the threshold value, indicating that the network service quality provided by the current AP is low and network switching needs to occur, and triggering a switching algorithm to find a new CoA; otherwise, if the average value is higher than the threshold value, it indicates that the current network service quality is good, and no handover needs to occur, the vehicle will repeat the process periodically until the average value of the RSSI prediction sequence is lower than the threshold level.
The whole flow of the handover triggering algorithm is shown in fig. 3. In the algorithm provided by the invention, an LSTM network is trained in an off-line training mode, an LSTM model suitable for switching trigger prediction in the Internet of vehicles is obtained by continuously training the model and changing parameters, each vehicle node in the Internet of vehicles environment receives a copy of the trained LSTM network, the RSSI value in the next period of time is predicted through the copy to obtain an RSSI prediction sequence, and then whether switching is triggered is judged according to the average value of the obtained predicted RSSI sequence and the threshold value.
And when the switching triggering algorithm judges that switching is needed, the vehicle node needs to be switched to other networks, a new AP connection is requested to carry out normal network communication, and a switching decision stage is entered. At this stage, the vehicle node will calculate its direction, speed and position information relative to the currently connected AP and encapsulate these information in the observed state O, and then import the observed state into the HMM model, generating an AP estimate vector for the initial state in the HMM model.
When the AP switching probability reaches a preset threshold value alpha, the fact that enough learning is constructed and a proper AP node exists is indicated. The mobile node transmits an early registration request packet to the predicted new AP through the MAP. The early registration request data packet includes the current AP of the vehicle and the MAP to which the AP is attached, and the predicted AP and the MAP to which the AP is attached, and also includes the observed state of the vehicle. If the MAP domain corresponding to the predicted AP is the same as the currently connected MAP domain, i.e., intra-domain handover, then no binding update procedure by the HA is required and the MAP will forward the early registration packet to the predicted new AP. Conversely, if the predicted AP corresponds to a MAP domain that is different from the current MAP domain, i.e., an inter-domain handoff, the new MAP will add vehicle information to its cache and forward the early registration request to the vehicle's HA, while setting a life cycle timer to prevent a prediction error. When the vehicle node enters the coverage area of the new AP, a registration message is sent to the new AP. If the AP is a new predicted AP, then no message is sent to the MAP and HA, and the registration process is completed. An update message is then sent to the MAP and HA to cancel the life cycle timer. Conversely, if the HMM prediction method does not return a predicted AP, the vehicle node will use its AP broadcast cache to initiate early registration.
When the prediction is complete, the vehicle connects to the new AP and MAP, then the life cycle timer in the newly connected MAP is updated and the MAP sends an update packet to the mobile node's corresponding HA to update its time. Meanwhile, the mobile node also sends an updating data packet to the previous AP, informs the AP to delete the vehicle information according to the content of the data packet and updates the HMM model parameters of the AP. If a prediction failure occurs, the standard registration process in the protocol is switched using HMIPv6 and a new registered MAP is sent to the previous AP with an update packet containing the newly registered AP, the predicted AP, and the observed status. Upon receiving the update packet, the AP first checks whether the newly registered AP is in its neighbor list, and if not, adds a new AP in its neighbor list and updates the HMM model. In the protocol proposed by the present invention, a maximum value of the number of neighbor lists of an AP is specified, so that when a new AP is found, the lowest probability neighboring APs will be deleted and replaced with the new AP.
Fig. 4 is a general architecture diagram of the system of the present invention, and considering the compatibility of the handover decision algorithm designed by the present invention with the LTE-V protocol stack, the system is designed based on the LTE-V protocol stack, and the total of the system is the following five modules: the system comprises an information interaction module, an information maintenance module, a switching trigger module, an AP selection module and a data forwarding module.
Fig. 5 shows a message format of an information packet to be forwarded in the information interaction module, which is sent as a DSM message in the LTE-V protocol and encapsulated in the data portion. The AP information data packet is totally divided into two parts of header information and data content, wherein the header information comprises the MAC address of the node, and the data part comprises the geographical position information of the node, the generated HMM model and the neighbor AP list thereof.
The AP information data packet is sent in a timing broadcast mode, and the broadcast time interval is adjusted according to requirements. As shown in fig. 6, when the device is started, the upper layer application sends an AP information broadcast service request to the management entity DME, then sets a life cycle timer to control the transmission cycle of information, and when the life cycle timer reaches a predetermined time, acquires the latest adaptive HMM element and the neighbor list of the AP from the MIB of the management entity, and then performs integrity check of the information, and if the information is complete, encapsulates the information according to the format of an AP information packet, and invokes an interface broadcast packet of an access layer after completing the layer-by-layer encapsulation of data. And resetting the life cycle timer after each data broadcast, and waiting for the next AP information broadcast.
Fig. 7 is a flow of sending an early registration request packet, where the early registration request packet is in a unicast form, and when a vehicle node determines to start network handover, it first sends an early registration request packet to its MAP domain, and determines to select inter-domain handover or intra-domain handover according to whether the newly predicted MAP domain of the AP is the same as the currently connected MAP domain; the inter-domain handover sends a binding update to the HA. And when the vehicle node enters the communication range of the new AP, sending a registration request message to the vehicle node, if the AP is the predicted AP, finishing the registration, otherwise, sending a message to the MAP and the HA, and finishing the registration process.
Fig. 8 is a structural design diagram of an AP neighbor list that needs to be stored, and according to the characteristics of a heterogeneous internet of vehicles that the node moving speed is high and the network topology changes frequently, the vehicle node is easy to switch, so that the neighbor network list is frequently inserted, deleted, or searched, and therefore the design of the neighbor network list must be able to quickly fulfill the above requirements, so as to reduce the time of a switching decision process and improve the efficiency of a switching algorithm. Each time a vehicle node enters the coverage of one AP, the vehicle node receives a neighbor network list sent by the vehicle node, so a plurality of neighbor network lists need to be stored, each neighbor list is stored in a linked list form, each node comprises APID, Updatetime, Latitude, Longitude, Server, RSSI and next information, wherein the APID is a unique identifier for distinguishing the AP, the Updatetime is the time for updating the node information for the last time, Latitude and Longitude are position information of the node, the Server is the service type provided by the AP, and the RSSI is the intensity of receivable signals around the AP.
Fig. 9 is a frame structure change flow of a data frame when DSM transmits, where messages to be transmitted by a vehicle and an AP are stored in a data portion, and are transmitted to a DSMP entity by an application layer or a management entity DME through a DSM request primitive, the DSMP adds a DSMP header structure to the DSMP after receiving the DSMP, then transmits the DSMP header structure to an ADAPTATION layer through an ADAPTATION-layer request primitive, adds an ADAPTATION layer header to the information at the ADAPTATION layer, then transmits the DSMP header structure to an ACCESS layer through an ACCESS-layer request primitive, adds an ACCESS correlation technique to the ACCESS layer header, and ends an encapsulation process, and the ACCESS layer transmits the encapsulated messages in a broadcast form or by other methods.
The data analysis process corresponds to the data encapsulation process, when a vehicle node receives a data packet, the data packet is firstly transmitted into an ACCESS layer, the ACCESS layer indicates that a high layer receives ACCESS data by sending ACCESS-layer indication primitive, wherein the ACCESS data comprises data, address information and priority, the ADAPTATION layer analyzes the data after receiving the data, judges whether the data is legal or not by verifying the protocol type of a data head, and then sends the data to a DSMP layer through an ADAPTATION-layer indication primitive. And at the DSMP layer, identifying the type of the received message service by AID, finishing the relevant operation at the DSMP layer if the message service is the information relevant to handover, and sending the message service to the application layer through DSM.

Claims (7)

1. A hidden Markov-based vehicle networking switching decision algorithm is characterized in that firstly, a long-short term memory network (LSTM) prediction model is used for completing the prediction of the strength of a receiving signal currently connected with a vehicle, and whether switching needs to be triggered or not and the specific time for triggering switching are judged; secondly, introducing a Hidden Markov Model (HMM) and judging a target network which needs to be switched by the vehicle by combining the observation attribute of the vehicle; finally, in the process of system design, the compatibility of the algorithm and an LTE-V protocol stack is considered, and the whole system is divided into five modules, namely an information interaction module, an information maintenance module, a switching trigger module, a network selection module and a data forwarding module.
2. The hidden markov-based internet of vehicles handover decision algorithm according to claim 1, wherein the information interaction module is responsible for sending and receiving information; the information maintenance module is used for storing information to be stored; the switching trigger module predicts whether switching is carried out or not; the network selection module selects an optimal switching target network; and the data forwarding module completes the encapsulation and analysis of the data.
3. The hidden markov-based internet of vehicles handover decision algorithm according to claim 2, wherein the information interaction module is specifically configured to send an early registration request packet and an update packet to each other between the vehicle and the roadside device, and the AP also sends an adaptive HMM element and a neighbor AP list of the AP to surrounding vehicle nodes in a broadcast manner; the information interaction module is responsible for data interaction work among all nodes and controls the time and the type of the message to be forwarded.
4. The hidden markov-based internet of vehicles handover decision algorithm according to claim 3, wherein the information maintenance module specifically comprises: the module works in a management entity DME of a network layer, four table entries of a local information table, a DSM service request table, an application request information table and a user service request message table are originally stored in an MIB of an LTE-V protocol stack and used for storing DSM short messages and message services which can be sent or need to be received; the system adds table items for realizing the switching decision algorithm, and stores and maintains the observation state of the vehicle, the neighbor network list, the HMM model and the RSSI sequence.
5. The hidden markov-based internet of vehicles handover decision algorithm according to claim 4, wherein the handover triggering module is specifically: the method comprises the steps that the vehicle node is helped to judge the received signal strength RSSI of the currently connected AP in the next period of time in advance, so that the specific time for triggering switching of the vehicle node is judged;
the received signal strength decreases logarithmically with increasing distance, so the RSSI is expressed as:
Figure FDA0003292422790000021
wherein, dbm0Represents RSSI at 1 meter near AP, d is the distance between the vehicle node and AP, n is the path loss factor constant, d0A unit measurement parameter value representing a distance between the vehicle node and the AP is a fixed value;
when the vehicle node is connected to the network, the RSSI list { RSSI ] of the AP currently connected with the vehicle is maintained in the window time Lt,rssit+1，…，rssit+LEstimating the RSSI in the next period of time according to the RSSI list, and predicting in advance whether switching needs to be carried out or not so as to find and register a new care-of address; obtaining the next RSSI by adopting an LSTM network prediction model, wherein the LSTM uses an effective learning algorithm based on gradient to solve the problems of exponential decay error and disappearance in the recursion process;
the LSTM network prediction model is divided into three layers, namely an input layer, a hidden layer and an output layer, wherein the input layer is an RSSI list of a vehicle node currently connected with an AP, after the list is transmitted to the hidden layer, the hidden layer updates the state of the hidden layer, and a result, namely the RSSI prediction list, is calculated according to an input value; the state of the hidden layer in the model is updated using the following formula:
Figure FDA0003292422790000022
Figure FDA0003292422790000023
wherein f is a nonlinear hidden layer function set as an S-shaped function, and w is a Bayesian vector, and a weight matrix is defined; the hidden layer in the LSTM network prediction model also includes memory cells and gate cells that modify the f-function in the conventional RNN model to store information, the memory cells defining some current interactions to determine which type of information to retain in memory and output hidden states;
the switching triggering algorithm based on the LSTM is divided into three steps: data preprocessing, learning and prediction; in the algorithm, a vehicle node maintains the RSSI in a current connection AP update data packet and stores the RSSI in a list according to a time sequence, an LSTM network prediction model is trained in an off-line mode by using the RSSI sequence as an input vector, and a new RSSI sequence is predicted by using stored model parameters; in addition, when the RSSI values used for training are collected, the window size of an input vector is set, and the window size is the number of RSSIs in an RSSI sequence vector required for prediction; the window is too small, so that the RSSI in the next period of time cannot be accurately predicted according to the window, and the difference between the obtained predicted sequence and the actual RSSI is larger; the too large window increases the training overhead, so that the loss of model training is increased; and selecting a value with a proper window size according to the test of different window sizes.
6. The hidden Markov-based Internet of vehicles handover decision algorithm as claimed in claim 5, wherein the network selection module is the core content of the handover decision algorithm for selecting the best target network to which the current vehicle is to be handed over; the module uses a hidden Markov model for prediction to determine a target network to be switched by the vehicle;
the HMM model has N different states: s1，S2，…，SNEach of the neighboring APs represents a different neighboring AP, and a state transition can occur at each time T equal to 1,2, …, T, M represents the number of observed states for each state, and is a map of the vehicle motion state, xtRepresents the state at time t; the HMM model contains three parts: a state transition matrix A, an observation probability matrix B and an initial probability distribution pi; the state transition matrix a is a matrix of probability of switching the vehicle among the APs, and is defined as follows:
Figure FDA0003292422790000031
wherein, ai，jRepresents a vehicle slave APiHandover to APjIs specifically expressed as:
ai,j＝P(xt＝Sj|xt-1＝Si)，i,j∈N (5)
the observation probability matrix B is a matrix of probabilities that a vehicle connects to an AP in a current observation state, and is defined as follows:
Figure FDA0003292422790000041
wherein, bi,jRepresenting the vehicle being connected to the AP under current motionjIs specifically expressed as:
bi,j＝P(yt＝Ok|xt＝Sj),j∈N,k∈M. (7)
o in the above formulakThe representative is the current observation state of the vehicle, which is obtained through the measured value of the vehicle and a Kalman filter; the initial probability distribution pi is the set of probabilities that a vehicle is connected to each AP at the beginning:
πi＝P(x1＝Si),1<i<N (8)
the above three elements are defined into one triple lambda,
λ＝(A,B,π) (9)
the AP to be switched to and its accompanying MAP states are hidden, while the movement attributes of the vehicle are obtained by measurement, and the probability values are trained and predicted based on the above HMM elements and the observed values of the network, mainly by two steps: self-adaptive learning and initial matrix estimation; an adaptive learning process is used to recursively estimate the AP probabilities, updating the observation matrix only in dependence on previous timestamps; the AP probability is described in two parts, int 1 and other time periods, according to an initial probability pi ═ α0(i) Get AP probability when t ═ 1:
α1(i)＝πibi(o1)ai(1) (10)
the probability of the AP at time t in other time periods is:
Figure FDA0003292422790000042
the initial probability distribution of the APs is as follows:
Figure FDA0003292422790000043
wherein δ (·) is the Kronecker increment, determined by the parameter value, which is 1 if the parameter is 0, and 0 if the parameter is other values; the observed state uses w (i) instead of the likelihood estimate after the handoff is complete, the weight of the ith packet is determined by the following equation,
Figure FDA0003292422790000051
wherein L is the serial number of the last data packet, and the weight value is ensured not to be zero; the recalculation formula of the state transition matrix and the observation matrix is as follows:
Figure FDA0003292422790000052
Figure FDA0003292422790000053
wherein xpFor predicted AP nodes, xaFor an actually connected AP node, ocIs the current observed state of the vehicle at the time of the prediction;
describing an observation state by the motion direction, the moving speed and the geographical position information of the vehicle, and under the environment of the Internet of vehicles, acquiring the geographical position information of the vehicle according to the vehicle-mounted equipment of the vehicle, and acquiring the position information of the AP through a data packet sent by the AP node, so that the moving direction theta of the vehicle is calculated by the following formula;
dirx＝predictx-currx (16)
diry＝predicty-curry (17)
Figure FDA0003292422790000054
wherein the predictxAnd predictyIs the predicted coordinate information of the next position of the vehicle, currxAnd curryObtaining the included angle between the vehicle motion direction and the x-axis direction according to an arc tangent function; after the angle of the moving direction of the vehicle is defined, the data are dispersed into equal-width characteristics, continuous values are converted into discrete corresponding values, the range is 45 degrees, and the moving direction of the vehicle is defined as 8 different directions: n, NE, E, SE, S, SW, W, NW;
the speed s of the vehicle is obtained according to a Kalman filter and the current speed of the vehicle, the predicted speed is converted into a discrete data set in an observation sequence, the speed of the vehicle is divided into 6 different grades according to the size of the discrete data set, and the speed of the vehicle does not exceed 100 km/h;
the position information of the vehicle is a third observation sequence in the HMM model, and the two-dimensional position of the vehicle is selected for use and is represented by longitude and latitude for convenience in use; the method comprises the steps that position information of vehicles is obtained through a vehicle-mounted GPS, estimation is carried out through a Kalman filter, the distance between the position of each vehicle and the edge of the coverage range of the currently connected AP is calculated by each vehicle, then the position of each vehicle is mapped to grid numbers on a map, and the total number of grids covered by each AP is defined by the maximum communication distance of the grid numbers.
7. The hidden Markov-based Internet of vehicles switching decision algorithm as claimed in claim 6, wherein the data forwarding module is mainly used for encapsulating and sending information to be sent layer by layer from an upper layer to a lower layer, analyzing the received information layer by layer, and finally obtaining the required data; the module completes the function of information transmission between different nodes.
CN202111169755.0A 2021-10-08 2021-10-08 Internet of vehicles switching decision method based on hidden Markov Active CN113950113B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111169755.0A CN113950113B (en) 2021-10-08 2021-10-08 Internet of vehicles switching decision method based on hidden Markov

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111169755.0A CN113950113B (en) 2021-10-08 2021-10-08 Internet of vehicles switching decision method based on hidden Markov

Publications (2)

Publication Number Publication Date
CN113950113A true CN113950113A (en) 2022-01-18
CN113950113B CN113950113B (en) 2022-10-25

Family

ID=79329979

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111169755.0A Active CN113950113B (en) 2021-10-08 2021-10-08 Internet of vehicles switching decision method based on hidden Markov

Country Status (1)

Country Link
CN (1) CN113950113B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115002858A (en) * 2022-05-25 2022-09-02 中国第一汽车股份有限公司 Networking state switching method and device, vehicle and storage medium
CN116736729A (en) * 2023-08-14 2023-09-12 成都蓉奥科技有限公司 Method for generating perception error-resistant maneuvering strategy of air combat in line of sight
CN118225083A (en) * 2024-03-28 2024-06-21 苏州明泰智能装备有限公司 Robot navigation positioning method based on multi-sensor fusion
WO2024178694A1 (en) * 2023-03-01 2024-09-06 Huawei Technologies Co., Ltd. Multi-scenario prediction and proactive operations in communication networks

Citations (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009076208A2 (en) * 2007-12-05 2009-06-18 Qualcomm Incorporated Handover failure procedures in communication systems
CN103269512A (en) * 2013-04-25 2013-08-28 浙江工商大学 Mixed type DVB-H system switching decision algorithm
CN103281753A (en) * 2013-06-20 2013-09-04 重庆邮电大学 Access network selection device and method based on Hidden Markov Model
CN103731892A (en) * 2014-01-08 2014-04-16 浙江工商大学 Hidden Markov Model (HMM) based cell handover decision algorithm and device
US20160029266A1 (en) * 2009-05-01 2016-01-28 At&T Mobility Ii Llc Access control for macrocell to femtocell handover
CN107466012A (en) * 2017-08-24 2017-12-12 北京邮电大学 Network switching optimization method based on mobility prediction under home network indoors
CN109474897A (en) * 2019-01-10 2019-03-15 厦门大学 Car networking security message single-hop cooperation broadcasting method based on hidden Markov model
CN110073301A (en) * 2017-08-02 2019-07-30 强力物联网投资组合2016有限公司 The detection method and system under data collection environment in industrial Internet of Things with large data sets
CN110139331A (en) * 2019-04-22 2019-08-16 天津大学 Adaptive handover trigger strategy implementation method based on 5G network
CN110149595A (en) * 2019-05-10 2019-08-20 北京工业大学 A kind of heterogeneous network user's behavior prediction method based on HMM
CN111130853A (en) * 2019-12-04 2020-05-08 沈阳航空航天大学 Future route prediction method of software defined vehicle network based on time information
CN111629415A (en) * 2020-04-24 2020-09-04 南京理工大学 Opportunistic routing protocol based on Markov decision process model
CN111800421A (en) * 2020-07-06 2020-10-20 东北大学 Vehicle networking intrusion detection system based on hidden Markov model
CN111918348A (en) * 2020-07-10 2020-11-10 Oppo广东移动通信有限公司 Network cell switching method and device, storage medium and electronic equipment
CN112004197A (en) * 2020-08-06 2020-11-27 重庆邮电大学 Heterogeneous Internet of vehicles switching method based on vehicle track prediction
CN112703457A (en) * 2018-05-07 2021-04-23 强力物联网投资组合2016有限公司 Method and system for data collection, learning and machine signal streaming for analysis and maintenance using industrial internet of things
CN112765892A (en) * 2021-01-27 2021-05-07 东北大学 Intelligent switching judgment method in heterogeneous Internet of vehicles
WO2021118418A1 (en) * 2019-12-10 2021-06-17 Telefonaktiebolaget Lm Ericsson (Publ) Methods, ue and first network node for handling mobility information in a communications network
CN113065431A (en) * 2021-03-22 2021-07-02 浙江理工大学 Human body violation prediction method based on hidden Markov model and recurrent neural network
US11109283B1 (en) * 2020-09-22 2021-08-31 Accenture Global Solutions Limited Handover success rate prediction and management using machine learning for 5G networks

Patent Citations (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009076208A2 (en) * 2007-12-05 2009-06-18 Qualcomm Incorporated Handover failure procedures in communication systems
US20160029266A1 (en) * 2009-05-01 2016-01-28 At&T Mobility Ii Llc Access control for macrocell to femtocell handover
CN103269512A (en) * 2013-04-25 2013-08-28 浙江工商大学 Mixed type DVB-H system switching decision algorithm
CN103281753A (en) * 2013-06-20 2013-09-04 重庆邮电大学 Access network selection device and method based on Hidden Markov Model
CN103731892A (en) * 2014-01-08 2014-04-16 浙江工商大学 Hidden Markov Model (HMM) based cell handover decision algorithm and device
CN110073301A (en) * 2017-08-02 2019-07-30 强力物联网投资组合2016有限公司 The detection method and system under data collection environment in industrial Internet of Things with large data sets
CN107466012A (en) * 2017-08-24 2017-12-12 北京邮电大学 Network switching optimization method based on mobility prediction under home network indoors
CN112703457A (en) * 2018-05-07 2021-04-23 强力物联网投资组合2016有限公司 Method and system for data collection, learning and machine signal streaming for analysis and maintenance using industrial internet of things
CN109474897A (en) * 2019-01-10 2019-03-15 厦门大学 Car networking security message single-hop cooperation broadcasting method based on hidden Markov model
CN110139331A (en) * 2019-04-22 2019-08-16 天津大学 Adaptive handover trigger strategy implementation method based on 5G network
CN110149595A (en) * 2019-05-10 2019-08-20 北京工业大学 A kind of heterogeneous network user's behavior prediction method based on HMM
CN111130853A (en) * 2019-12-04 2020-05-08 沈阳航空航天大学 Future route prediction method of software defined vehicle network based on time information
WO2021118418A1 (en) * 2019-12-10 2021-06-17 Telefonaktiebolaget Lm Ericsson (Publ) Methods, ue and first network node for handling mobility information in a communications network
CN111629415A (en) * 2020-04-24 2020-09-04 南京理工大学 Opportunistic routing protocol based on Markov decision process model
CN111800421A (en) * 2020-07-06 2020-10-20 东北大学 Vehicle networking intrusion detection system based on hidden Markov model
CN111918348A (en) * 2020-07-10 2020-11-10 Oppo广东移动通信有限公司 Network cell switching method and device, storage medium and electronic equipment
CN112004197A (en) * 2020-08-06 2020-11-27 重庆邮电大学 Heterogeneous Internet of vehicles switching method based on vehicle track prediction
US11109283B1 (en) * 2020-09-22 2021-08-31 Accenture Global Solutions Limited Handover success rate prediction and management using machine learning for 5G networks
CN112765892A (en) * 2021-01-27 2021-05-07 东北大学 Intelligent switching judgment method in heterogeneous Internet of vehicles
CN113065431A (en) * 2021-03-22 2021-07-02 浙江理工大学 Human body violation prediction method based on hidden Markov model and recurrent neural network

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
CHENYANG SHEN; SHUQIN ZHANG; MICHAEL K. NG: ""A tensor-based Markov chain method for module identification from multiple networks"", 《2014 8TH INTERNATIONAL CONFERENCE ON SYSTEMS BIOLOGY (ISB)》 *
XIAOGE HUANG;SHE TANG;QIAN ZHENG;DONGYU ZHANG;QIANBIN CHEN: ""Dynamic Femtocell gNB On/Off Strategies and Seamless Dual Connectivity in 5G Heterogeneous Cellular Networks"", 《 IEEE ACCESS》 *
张娟: ""基于部分可观察马尔科夫决策过程(POMDP)的贪婪算法次优频谱接入"", 《科学技术与工程》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115002858A (en) * 2022-05-25 2022-09-02 中国第一汽车股份有限公司 Networking state switching method and device, vehicle and storage medium
CN115002858B (en) * 2022-05-25 2023-10-27 中国第一汽车股份有限公司 Networking state switching method and device, vehicle and storage medium
WO2024178694A1 (en) * 2023-03-01 2024-09-06 Huawei Technologies Co., Ltd. Multi-scenario prediction and proactive operations in communication networks
CN116736729A (en) * 2023-08-14 2023-09-12 成都蓉奥科技有限公司 Method for generating perception error-resistant maneuvering strategy of air combat in line of sight
CN116736729B (en) * 2023-08-14 2023-10-27 成都蓉奥科技有限公司 Method for generating perception error-resistant maneuvering strategy of air combat in line of sight
CN118225083A (en) * 2024-03-28 2024-06-21 苏州明泰智能装备有限公司 Robot navigation positioning method based on multi-sensor fusion

Also Published As

Publication number Publication date
CN113950113B (en) 2022-10-25

Similar Documents

Publication Publication Date Title
CN113950113B (en) Internet of vehicles switching decision method based on hidden Markov
Gao et al. V2VR: reliable hybrid-network-oriented V2V data transmission and routing considering RSUs and connectivity probability
Darwish et al. Reliable intersection-based traffic aware routing protocol for urban areas vehicular ad hoc networks
CN108632813B (en) Mobility management method and system for mobile edge computing
US20210235281A1 (en) Optimizing wireless networking using a virtual geographic information system overlay
Hawbani et al. Fuzzy-based distributed protocol for vehicle-to-vehicle communication
Saravanan et al. Routing using reinforcement learning in vehicular ad hoc networks
Yang et al. Connectivity aware routing in vehicular networks
CN105848241B (en) The cluster-dividing method and system of mobile ad hoc network
CN111510956B (en) Hybrid routing method based on clustering and reinforcement learning and ocean communication system
CN105208616A (en) Road topology based adaptive multi-copy routing method in vehicular ad hoc network
CN109743742A (en) For the motion management method and device in eV2X system based on UE behavior prediction
CN106972898A (en) Car networking data transmission scheduling method based on channel estimating
CN111356199B (en) Vehicle-mounted self-organizing network routing method in three-dimensional scene
Qiu et al. Maintaining links in the highly dynamic fanet using deep reinforcement learning
CN108513330A (en) A kind of Routing Protocol distributed self-adaption switching method based on network context perception
Cheng et al. Stable clustering for VANETs on highways
Hussain et al. Performance evaluation of vertical handover in Internet of Vehicles
Kunarak et al. Multi-criteria vertical handoff decision algorithm for overlaid heterogeneous mobile IP networks
CN110719201B (en) Distributed self-adaptive stable topology generation method based on reinforcement learning
Aboud et al. Markov chain based predictive model for efficient handover management in vehicle-to-infrastructure communications
Ghafoor et al. A novel delay-and reliability-aware inter-vehicle routing protocol.
CN114071421B (en) Open type unmanned vehicle group model in expressway scene and vehicle group forming method
CN114980185A (en) Vehicle-mounted self-organizing network routing method based on topological evolution
Hussain et al. Artificial intelligence-based network selection and optimized routing in internet of vehicles

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