Vehicle-mounted network switching method and device in vehicle networking environment and electronic equipment

Abstract

The invention relates to a vehicle-mounted network switching method and device in an Internet of Vehicles environment and electronic equipment, and the method comprises the steps: determining the receiving power of a target vehicle when the target vehicle is connected with each network base station in the future based on the future trajectory information and a channel knowledge graph of the target vehicle, and the channel knowledge graph comprises the channel information of each position; according to the future trajectory information of the target vehicle, determining the residence time of the target vehicle in the coverage range of each network base station in the future; and determining a target network switching strategy of the target vehicle based on the residence time of the target vehicle in the coverage range of each network base station in the future, the receiving power when each network base station is connected in the future, the network base station switching cost and the real-time load of each network base station. The problem that network switching achieved by an existing vehicle-mounted network switching method is prone to lagging behind communication environment changes is solved.

Description

Technical Field

[0001] The present application relates to the field of Internet of Vehicles, and in particular to a method, device and electronic device for switching an in-vehicle network in an Internet of Vehicles environment. Background Art

[0002] In previous studies, various solutions suitable for 5G vehicle networking environments have been proposed to ensure that users maintain low-latency, high-quality network connections. For example, a deep recursive Q network switching strategy that considers reference signal received power (RSRP), signal-to-noise ratio (SINR), and candidate base station load is proposed to dynamically adjust and optimize the switching parameters. The Q network combines convolutional neural networks (CNN) and recursive neural networks (RNN). Deep neural networks and RNNs can extract the spatiotemporal characteristics of the wireless network environment, thereby obtaining better switching options.

[0003] The above deep recursive Q network switching strategy uses instantaneous RSRP and SINR as indicators to describe the communication environment. Based on the traditional switching algorithm, the switching parameters are changed according to the vehicle speed and the surrounding communication environment. In the Internet of Vehicles environment, due to the high-speed movement of the vehicle, the communication environment it encounters changes rapidly. The switching strategy must be proactive, otherwise it is easy for the network switching to lag behind the changes in the communication environment. If the network switching is not timely, the network connection quality will decrease in the short term.

[0004] There is currently no effective solution to the problem that the network switching implemented by the current vehicle network switching method tends to lag behind changes in the communication environment. Summary of the invention

[0005] The present invention provides a vehicle network switching method, device and electronic device in a vehicle networking environment to solve the problem that the network switching achieved by the current vehicle network switching method is prone to lag behind the changes in the communication environment.

[0006] In a first aspect, the present invention provides a method for switching an in-vehicle network in an Internet of Vehicles environment, comprising:

[0007] Based on a historical driving knowledge graph of the target vehicle, predicting future trajectory information of the target vehicle through a trained trajectory prediction model, wherein the historical driving knowledge graph includes a historical position, a historical speed, and a historical direction of the target vehicle;

[0008] Determine the receiving power of the target vehicle when it connects to each network base station in the future based on the future trajectory information of the target vehicle and the channel knowledge graph, wherein the channel knowledge graph includes channel information of each position;

[0009] Determine the future stay time of the target vehicle within the coverage area of ​​each network base station according to the future trajectory information of the target vehicle;

[0010] The target network switching strategy of the target vehicle is determined based on the future residence time of the target vehicle within the coverage of each network base station, the receiving power and network base station switching cost when connecting to each network base station in the future, and the real-time load of each network base station.

[0011] In a second aspect, the present invention provides a vehicle network switching device in a vehicle networking environment, comprising:

[0012] A trajectory prediction module, used to predict the future trajectory information of the target vehicle through a trained trajectory prediction model based on a historical driving knowledge graph of the target vehicle, wherein the historical driving knowledge graph includes a historical position, a historical speed, and a historical direction of the target vehicle;

[0013] A power prediction module, used to determine the receiving power of the target vehicle when it connects to each network base station in the future based on the future trajectory information of the target vehicle and the channel knowledge graph, wherein the channel knowledge graph includes channel information of each position;

[0014] A time prediction module, used to determine the future stay time of the target vehicle within the coverage area of ​​each network base station according to the future trajectory information of the target vehicle;

[0015] The strategy determination module is used to determine the target network switching strategy of the target vehicle based on the future residence time of the target vehicle within the coverage of each network base station, the receiving power and network base station switching cost when connecting to each network base station in the future, and the real-time load of each network base station.

[0016] In a third aspect, the present invention provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the vehicle network switching method in the vehicle networking environment described in the first aspect.

[0017] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the vehicle network switching method in the vehicle networking environment described in the first aspect are implemented.

[0018] Compared with the related art, the present invention mainly considers four factors, namely, the target vehicle's future residence time within the coverage of each network base station, the receiving power and network base station switching cost when connecting to each network base station in the future, and the real-time load of each network base station. The network switching scheme can ensure the quality of network connection and minimize the number of network switching times for determining the target vehicle's network switching strategy. At the same time, since the target vehicle's residence time within the coverage of each network base station and the receiving power when connecting to each network base station are both predicted future values, the network switching strategy is forward-looking. Compared with determining the current network switching strategy based on the current communication environment information, the target vehicle can complete the network switching before the network environment deteriorates, so that the network switching will not lag behind the changes in the communication environment. This solves the problem that the network switching implemented by the current vehicle-mounted network switching method is prone to lag behind the changes in the communication environment.

[0019] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of a vehicle network switching method in a vehicle networking environment provided in an embodiment of the present invention;

[0021] Figure 2 is an architecture diagram of a trajectory prediction model in some embodiments of the present invention;

[0022] Figure 3 is an architectural diagram of a received power prediction model in some embodiments of the present invention;

[0023] Figure 4 is a schematic diagram of the prediction accuracy of the strategy scoring prediction model in some embodiments of the present invention;

[0024] Figure 5 is a schematic diagram of changes in user achievable data rates under some embodiments of the present invention;

[0025] Figure 6 This is a schematic diagram of the change of the user's achievable data rate under the traditional 3GPP solution;

[0026] Figure 7 This is a comparison chart of the number of network switching changes under different schemes;

[0027] Figure 8 It is a comparison chart of the changes in the number of frequent switching under different schemes. DETAILED DESCRIPTION

[0028] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0029] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the", "these" and the like in this application do not represent quantitative restrictions, and they can be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether directly or indirectly. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. Usually, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0030] In the Internet of Vehicles environment, it can be assumed that there are J network base stations (APs for short) and I users (UEs for short, also called vehicles). For example, the Internet of Vehicles environment is a cellular-free network consisting of N Sub6GAPs and M millimeter wave APs, then the total number of APs is J = M + N. The distribution of APs follows the Poisson Point Process (PPP). The UE can select a target from these APs for switching.

[0031] Since large-scale fading of the channel dominates in actual handover or user association scenarios, AP j (jth AP) and UE i The channel gain (in dB) between (the i-th UE) is expressed as:

[0032] g i,j =10alog 10 (d i,j )+b+10γlog 10 (f)+χ σ

[0033] Where, di,j For AP j With UE i The distance between the two, f is the carrier frequency, a and γ are the path loss coefficients corresponding to the distance and frequency respectively, b is a floating offset value, and the parameter χ σ Represents the standard deviation of shadow fading that describes large-scale signal fluctuations.

[0034] Then UE i With AP j The SINR during connection is:

[0035]

[0036] In the formula, p j For AP j The transmission power, p j g i,j Composition UE i Connect to AP j When the received power RSRP, j' is not equal to j, σ 2 is the noise power.

[0037] During the handover process, the wireless channel usually exhibits time-varying characteristics due to the mobility of the UE. Therefore, during the entire communication period (across T time slots), the UE may be associated with different APs. Define β i,j (t) is the AP at time slot t j With UE i The correlation index is expressed as:

[0038]

[0039] From the above formula, we can see that β i,j (t) = 1 indicates UE i Select the AP in time slot t j is related, and β i,j (t) = 0 means UE i In time slot t, it will communicate with AP j Based on this, the switching index α is established i,j (t) to represent AP j and UE i The switching state corresponding to time slot t is expressed as:

[0040]

[0041] UE will be in α i,j When (t)=1, the switching operation is performed; otherwise, the current connection state is maintained.

[0042] In addition, define AP j Assigned to UE iThe bandwidth is W i,j , then UE i Connect to AP at time slot t j The achievable rate is R i,j (t) = W i,j log2(1+SINR i,j (t)).

[0043] The goal of the present invention is to minimize the number of handovers while meeting the achievable rate requirements of each UE. The optimization problem can be expressed as:

[0044]

[0045] in, Indicates the minimum rate requirement. Constraint C1 ensures that each UE i and AP j Either in the associated state or in the unassociated state. Constraint C2 indicates that each UE can only access one AP in any time slot, and constraint C3 indicates the minimum rate requirement that each UE must meet in any time slot t.

[0046] By optimizing and solving the above problem, a switching method of the vehicle network can be obtained. It should be noted that the above optimization problem can be directly optimized and solved based on mathematical methods, or some other methods can be used to solve it equivalently.

[0047] Specifically, the present invention transforms the switching problem P1 into a sequential decision problem, and then proposes a vehicle network switching method in a vehicle networking environment to solve P1.

[0048] An embodiment of the present invention provides a method for switching an in-vehicle network in a connected vehicle environment.

[0049] Figure 1 is a flow chart of a vehicle network switching method in a vehicle networking environment provided in an embodiment of the present invention. Figure 1 The process includes step S110, step S120, step S130 and step S140.

[0050] Step S110, based on the historical driving knowledge graph (DKM) of the target vehicle, predict the future trajectory information of the target vehicle through the trained trajectory prediction model, and the historical driving knowledge graph includes the historical position, historical speed and historical direction of the target vehicle.

[0051] The target vehicle is the vehicle that currently needs to determine whether to switch networks. The continuous historical movement of the vehicle can largely reflect its future driving trend. In this step, the historical trajectory of the target vehicle is regarded as a time series in the trajectory prediction model, and its future trajectory is predicted by learning its characteristics. Specifically, the continuous future movement of the vehicle can be predicted by its historical driving knowledge graph. The historical driving knowledge graph includes the historical position, historical speed and historical direction of the target vehicle (i.e., the vehicle coordinates, vehicle speed and vehicle direction at each moment in the real-time historical time window). It should be noted that in traditional vehicle trajectory prediction methods, trajectory prediction is usually only based on vehicle position. This step further considers vehicle speed and vehicle direction, thereby improving the prediction accuracy of the vehicle's future trajectory.

[0052] Preferably, since the driving trajectory of a vehicle is related to the driver's own driving habits as well as the trajectories of its neighboring vehicles, in addition to considering the historical driving knowledge graph of the target vehicle, the historical driving knowledge graph of the neighboring vehicles of the target vehicle can also be considered to further improve the prediction of the future trajectory of the target vehicle.

[0053] Accordingly, in some of the embodiments, step S110 specifically includes: predicting the future trajectory information of the target vehicle through the trained trajectory prediction model based on the historical driving knowledge graph of the target vehicle and the historical driving knowledge graph of the neighboring vehicles of the target vehicle.

[0054] The trajectory prediction model may adopt different specific models in different embodiments.

[0055] Figure 2 is a diagram of the architecture of a trajectory prediction model in some embodiments of the present invention. Figure 2 In some embodiments, the trajectory prediction model includes a feature extraction module, a multi-task prediction module and a fusion module in sequence; the input of the feature extraction module is the input of the trajectory prediction model; the multi-task prediction module is used to perform a first prediction task and a second prediction task based on the output of the feature extraction module, respectively, the first prediction task is used to predict a first future position of the target vehicle, and the second prediction task is used to predict the future speed and direction of the target vehicle; the fusion module is used to fuse the first future position and the second future position of the target vehicle to predict the future trajectory information of the target vehicle.

[0056] Among them, the feature extraction module is used to extract features from the historical driving knowledge graph of the target vehicle (in some embodiments, it can also be the historical driving knowledge graph of the target vehicle and its adjacent vehicles). The feature extraction module includes a long short-term memory network (LSTM), a bidirectional long short-term memory network (BiLSTM) and a multi-layer perceptron (MLP); the inputs of the long short-term memory network, the bidirectional long short-term memory network and the multi-layer perceptron are all inputs of the feature extraction module; the fusion result of the full connection result (connected by a fully connected network) of the output of the long short-term memory network and the bidirectional long short-term memory network and the output of the multi-layer perceptron is used as the output of the feature extraction module. The feature extraction module extracts spatiotemporal features from the historical DKM. Among them, the LSTM and BiLSTM hybrid network extracts time features, and the fully connected network extracts the spatial features of the historical DKM. In this hybrid network, LSTM has the ability to avoid gradient disappearance and can capture long-term dependencies in sequence data; BiLSTM can simultaneously perform forward and reverse processing on sequence data, can fully capture sequence features, and improve feature integrity. The advantage of this hybrid network is that it combines the long-term dependency learning advantage of LSTM and the bidirectional information capture advantage of BiLSTM to obtain a powerful feature extraction capability. In addition, the hybrid architecture is more robust to changes in data and noise, improving the generalization ability of the model.

[0057] The multi-task prediction module performs prediction tasks based on the features extracted by the feature extraction module. The first prediction task is used to predict the first future position of the target vehicle (the vehicle coordinates at each moment in the real-time future time window). The second prediction task is used to predict the future speed and direction of the target vehicle (the vehicle speed and vehicle direction at each moment in the real-time future time window). Based on the current position of the target vehicle, the second future position of the target vehicle can be calculated through the future speed and direction of the target vehicle. Since the calculated and predicted future positions of the target vehicle may be different, they are distinguished by the first and the second. Finally, the fusion module fuses the first future position and the second future position of the target vehicle to obtain the future trajectory information of the target vehicle, which is the vehicle coordinates at each moment in the future.

[0058] In the above embodiment, multi-task learning can utilize the correlation between tasks to promote the extraction of additional valuable information. At the same time, compared with single-task learning, multi-task learning has superior robustness and generalization. Considering the significant correlation between the user's future position and his driving speed and direction, these driving elements are used as auxiliary tasks to predict the future trajectory. The two tasks share common underlying model parameters and update their respective model parameters in their respective prediction layers.

[0059] It should be noted that the edge server can be used to update the trajectory prediction model in real time based on the latest data, which can reduce the difference in trajectory predictions for different intersections and lanes and improve prediction accuracy.

[0060] The trajectory prediction model and its data processing process are explained below through the same specific example.

[0061] The historical coordinates, speed, and direction sets of the target vehicle i can be defined as:

[0062] S i (t) = [s i (tT b ),s i (tT b +1),...,s i (t)] T

[0063] V i (t) = [v(tT b ),v(tT b +1),...,v(t)] T

[0064] θ i (t) = [θ(tT b ),θ(tT b +1),...,θ(t)] T

[0065] Among them, T b represents the duration of the historical trajectory, θ i (t) and V i (t) represent the driving direction vector (sequence) and speed vector (sequence) of the target vehicle i at time slot t, v(t) and θ(t) represent the driving direction and speed of the target vehicle i at time slot t, respectively. i (t) represents the position vector (sequence) of the target vehicle i at time slot t, s i (t) = [s i,x (t),s i,y (t)] is the position coordinate of target vehicle i at time slot t. Therefore, the historical DKM of target vehicle i is written as Since the motion of the target vehicle i is usually affected by neighboring vehicles, the input of MTTP includes the historical DKM of the target vehicle and neighboring vehicles. Therefore, the input of the trajectory prediction model is mathematically written as Among them, i' represents the set of vehicles around the target vehicle i, Represents the DKM of the vehicle set.

[0066] The predictor network of each task is a fully connected (FC) network architecture consisting of L linear layers, and the output of each layer is activated using a tanh function. In the multi-task prediction module, the position of the future time slots output by the predictor network of the first prediction task (the first future position) and the direction and speed prediction of the future time slots output by the predictor network of the second prediction task are:

[0067]

[0068]

[0069]

[0070] Among them, T p is the length of the prediction window.

[0071] The loss functions of the first prediction task and the second prediction task are:

[0072]

[0073]

[0074] Wherein, r is the time slot count mark.

[0075] The second future position sequence is The elements are represented as:

[0076]

[0077] Where Δt is the time interval between adjacent time slots, The first future position obtained based on the prediction and the calculated second future position As the input of the fusion module, the two prediction sequences are fused. The fusion module consists of L serially connected linear layers, where the input of each layer includes the output of all previous layers. The structure of the lth layer can be expressed as:

[0078]

[0079] where l is the count of linear layers, the total number of which is L, and f l represents the output of the lth linear layer, W l represents the weight coefficient matrix of the lth linear layer, b l Represents the bias vector of the lth linear layer, input yes and The fusion module finally outputs the trajectory prediction That is, L fThe output of the layer. The loss function of the fusion module is expressed as

[0080] Taking into account the losses of different tasks and fusion modules, the loss function of the trajectory prediction module can be expressed as:

[0081]

[0082] Among them, λ task1 , task2 , out are the weights of the loss functions for different tasks.

[0083] Step S120, based on the future trajectory information of the target vehicle and the channel knowledge graph (CKM), determine the receiving power of the target vehicle when it connects to each network base station in the future, and the channel knowledge graph includes the channel information of each position.

[0084] CKM is a database used to store channel information at a specific location, such as channel gain, shadow, path angle, delay, and RSRP. In the present invention, RSRP in the channel information is mainly used. For the considered cell-free ultra-dense HetNets scenario, the change in RSRP is mainly caused by the relative change between the UE position and the AP position. Therefore, when the position of the target vehicle at each future moment is obtained, the received power of the target vehicle when it connects to each network base station at each future moment can be determined. Assuming that CKM has been constructed, AP j Position q j , then UE i At any position S i The RSRP of (t) is expressed as in It is a function that maps the UE position and AP position to the required RSRP.

[0085] There are many methods for constructing a channel knowledge graph, such as collecting real data or using ray tracing to construct a CKM. However, the above methods have the problem of poor real-time performance. In addition, the amount of data required to build a CKM is huge, and it is difficult to collect real data. Ray tracing to generate simulation data requires an accurate physical model, and the computational overhead is extremely high. In order to overcome the above problems, in some embodiments, a new CKM construction method for a neural network with multi-modal information fusion is provided, which has high generalization and reduces the overhead of data acquisition.

[0086] In some of the embodiments, the present invention adopts a new CKM construction method. Accordingly, the channel information of each location is the receiving power when the vehicle is connected to each network base station at each location. The steps of constructing the channel knowledge graph include: based on satellite images, the system status of each network base station and the vehicle sample trajectory, the receiving power prediction model is used to predict the receiving power of the vehicle when it is connected to each network base station at different locations, thereby obtaining a channel knowledge graph. Among them, in order to comprehensively construct the channel knowledge graph, the vehicle sample trajectory should include all possible driving trajectories of the vehicle.

[0087] Figure 3 FIG. 1 is a diagram of an architecture of a received power prediction model in some embodiments of the present invention. Figure 3 The received power prediction model includes a first feature extraction network, a second feature extraction network and a prediction network. The input of the first feature extraction network is a satellite image, and the input of the second feature extraction network is the system status of each network base station and the vehicle sample trajectory. The prediction network predicts the received power of the vehicle when it connects to each network base station at different locations based on the outputs of the first feature extraction network and the second feature extraction network.

[0088] In the above embodiment, a deep learning-based receiving power prediction model is used to construct a CKM based on satellite images and the system status of each network base station, and the receiving power of the target vehicle when it is connected to each network base station in the future is given in combination with the future trajectory information of the target vehicle.

[0089] Specifically, the above embodiment creates a multi-modal information fusion neural network to learn satellite images and system status information (including the system status of each network base station and the future trajectory information of the target vehicle) To obtain the RSRP prediction of target vehicle i Then the relationship between RSRP and multi-mode input information can be expressed as Function that maps multi-mode input information to RSRP. Indicates AP j To UE i It includes AP j Location, height, transmit power, UE i Location and AP j With UE i Due to the order differences of different types of data in the system status information and the huge dimensions of satellite images, data preprocessing is required for multi-mode input information. Normalization operations are performed on the system status information of each network base station, and image segmentation operations are performed on satellite images based on the system status information of each network base station (mainly using the network base station location and all possible vehicle locations) to ensure that UEi The position is always at the bottom of the image, AP j The position of the two points is at the top. The line connecting the two points is the perpendicular bisector of the image. Images from different angles are scaled and normalized to a uniform size to meet the requirements of the model. i,j and First, after being processed by the data preprocessing module, the normalized system status information is extracted through MLP (the second feature extraction network), and the cut satellite image is extracted through CNN (the first feature extraction network). The two extracted features are fused and input into a linear RSRP prediction module (prediction network) composed of a multi-layer fully connected network to generate the required RSRP prediction value. The model is trained with the mean squared error (MSE) loss function.

[0090] Step S130, determining the future stay time of the target vehicle within the coverage area of ​​each network base station according to the future trajectory information of the target vehicle.

[0091] Since the target vehicle's future trajectory information includes the target vehicle's position at each moment in the future, the target vehicle's future stay time within the coverage of each network base station can be determined by combining the position and coverage of each network base station. For example, the coverage of network base station A is B, and when the target vehicle's trajectory from the kth to the k+nth moment is within B, it can be determined that the target vehicle is within B from the kth to the k+nth moment and stays there for n moments.

[0092] Step S140, determining the target network switching strategy of the target vehicle based on the target vehicle's future residence time within the coverage of each network base station, the receiving power and network base station switching cost when connecting to each network base station in the future, and the real-time load of each network base station.

[0093] In this step, the target network switching strategy of the target vehicle is determined based on four factors, including the target vehicle's future residence time within the coverage of each network base station, the receiving power and network base station switching cost when connecting to each network base station in the future, and the real-time load of each network base station. Among them, the receiving power of the target vehicle when connecting to each network base station in the future and the real-time load of each network base station determine the quality of network connection. The target vehicle should tend to connect to a network base station with a lower load and higher receiving power when connected; the target vehicle's future residence time within the coverage of each network base station determines the stability of network connection. The target vehicle should tend to connect to a network base station with a longer residence time, which can also reduce the number of network switches; the network base station switching cost determines the number of network switches. Each network switch will generate a certain communication cost, which includes the signaling cost from the switch and the transmission cost from the interrupted transmission. Therefore, the target vehicle should tend to switch networks less frequently. By considering the above four factors, a network switching scheme that ensures the highest quality of network connection and the least number of network switches can be determined, which is equivalent to solving the above optimization problem P. At the same time, since the target vehicle's residence time within the coverage of each network base station and the received power when connected to each network base station are both predicted future values, the network switching strategy is forward-looking. For example, a 5-second future real-time time window can be set, and the current network switching strategy can be derived based on the communication environment information for the next 5 seconds. For example, the network connected to the target vehicle is A. Based on the communication environment information for the next 5 seconds, assuming that the connection quality of network A deteriorates and the connection quality of network B is better, the current network switching strategy of switching the target vehicle from connecting to network A to connecting to network B can be obtained. Compared with determining the current network switching strategy based on the current communication environment information, the target vehicle can complete the network switching before the network environment deteriorates, so that the network switching will not lag behind the changes in the communication environment.

[0094] Specifically, a neural network model can be used to determine the best network switching solution, ie, the target network switching strategy, based on the above four factors.

[0095] Accordingly, in some of the embodiments, step S140 specifically includes: based on the target vehicle's future residence time within the coverage of each network base station, the receiving power and network base station switching cost when connecting to each network base station in the future, and the real-time load of each network base station, the scores of each network switching strategy of the target vehicle are predicted through the trained strategy score prediction model; and the network switching strategy with the highest score is used as the target network switching strategy for the target vehicle.

[0096] The strategy scoring prediction model may adopt a common neural network model. For example, the strategy scoring prediction model adopts a network model based on ResNet.

[0097] For the training of the strategy score prediction model, the TOPSIS technique can be used to generate labels for the training data. In addition, during the model use and training process, in order to normalize different factors, the above four factor indicators can be converted into utility values ​​and form feature vectors by constructing a utility function, which is convenient for the strategy score prediction model to process.

[0098] As follows, the above factors and label generation process are explained through a specific example.

[0099] The received power of the target vehicle when it connects to each network base station in the future. From the received power prediction model, we can get the UE i RSRP in each of its candidate APs. i The RSRP prediction sequence of the candidate AP at time slot t can be expressed as [rsrp i(t),1 ,rsrp i(t),2 ,...,rsrp i(t),j ,...,rsrp i(t),J ] T To normalize these values, this example defines a linear utility function, which can be expressed as:

[0100]

[0101] in, is the highest RSRP that the target vehicle i can receive at time slot t, rsrp i(t),j is the RSRP received by target vehicle i from network base station j at time slot t, ε rsrp is the RSRP threshold of the signal. Utility function value The closer it is to 1, the better the transmission quality.

[0102] For the target vehicle's future stay time within the coverage of each network base station, the AP coverage is determined by the signal reception threshold preset by the base station. The RSRP predicted in the previous section can be used to determine whether the predicted location is within the AP coverage. This example defines a linear utility function for the stay time, expressed as:

[0103]

[0104] Where, dt i(t),j UE starts at time slot t i The future residence time within the coverage area of ​​the jth AP, The maximum time that target vehicle i can stay at time slot t, ε dt Is the minimum dwell time that will cause a handover failure. The longer the dwell time, the more stable the network.

[0105] For the real-time load of each network base station, the link quality between the base station and the terminal depends not only on the SINR, but also on the available bandwidth or resource blocks. If the AP processes more requests than its resources can support, the link quality provided by the AP will degrade. For the AP load, assuming that each AP adopts a proportional fair scheduling function, the real-time load value of J candidate base stations in time slot t is expressed as:

[0106] l i(t) =[l i(t),1 ,l i(t),2 ,...,l i(t),j ,...,l i(t),J ] T

[0107]

[0108] Among them, u j(t) Indicates AP j The number of users served at time slot t. The load value decreases as the number of users served by APj increases, and the rate of decrease gradually slows down. This example defines a linear utility function of real-time load, which can be expressed as:

[0109]

[0110] Among them, ε l is the load threshold, is the minimum load value of the AP that the target vehicle i can connect to at time slot t. The lower the load, The closer it is to 1, the better the network performance.

[0111] For the network base station switching cost, the communication costs mainly considered in this example include the signaling cost from the switching and the transmission cost from the interrupted transmission. i The communication cost of connecting each candidate AP at time slot t can be expressed as:

[0112] c i(t) =[c i(t),1 ,c i(t),2 ,...,c i(t),j ,...,c i(t),J ] T

[0113]

[0114] This example uses TOPSIS technology to generate labels for model switching decision training data in the multi-factor switching decision module. First, construct the attribute matrix of all candidate BSs considering these four attribute features. For simplicity, (t) in all subsequent formulas is simplified. The attribute matrix can be written as:

[0115]

[0116] Then the positive ideal solution and the negative ideal solution are expressed as:

[0117]

[0118] in and are the best and worst attribute values ​​among all candidate APs, respectively, and x is the ordinal number of the attribute feature considered.

[0119] The distance between the attribute value of each candidate AP and the positive / negative ideal solution can be expressed as:

[0120]

[0121] Among them, ω x is the weight of each attribute feature. Then the switching score of each candidate AP can be obtained as:

[0122]

[0123] This example obtains the dataset label through the above method. On this basis, this example uses the ResNet-based network model to predict the handover score of each AP to determine whether the user needs to perform a handover and the target AP for the handover. The model input is composed of 4 normalized parameters concatenated in series, and the output is the handover score prediction corresponding to each candidate AP. In each time slot, if the network status of the currently connected AP cannot meet the UE's requirements, the UE transfers to the AP with the highest predicted handover score. Otherwise, the UE maintains the current connection.

[0124] As described above, the vehicle network switching method in the vehicle networking environment provided by the present invention has been fully described.

[0125] It can be seen from the above description that the present invention mainly considers four factors, namely, the target vehicle's future residence time within the coverage of each network base station, the future receiving power and network base station switching cost when connecting to each network base station, and the real-time load of each network base station. The network switching strategy for determining the target vehicle can ensure the quality of network connection and minimize the number of network switching. At the same time, since the target vehicle's residence time within the coverage of each network base station and the receiving power when connecting to each network base station are both predicted future values, the network switching strategy is forward-looking. Compared with determining the current network switching strategy based on the current communication environment information, the target vehicle can complete the network switching before the network environment deteriorates, so that the network switching will not lag behind the changes in the communication environment. This solves the problem that the network switching implemented by the current vehicle-mounted network switching method is prone to lag behind the changes in the communication environment.

[0126] Furthermore, in the process of determining the network switching strategy, the factors considered are relatively comprehensive, which further improves the accuracy of determining the network switching strategy.

[0127] The effectiveness and accuracy of the different technical means in the present invention are illustrated by some experimental results as follows.

[0128] In the first experiment, the accuracy of the multi-factor switching decision module (step S140) on the training set and the test set was verified. Figure 4 is a schematic diagram of the prediction accuracy of the strategy scoring prediction model in some embodiments of the present invention, referring to Figure 4 , describes the accuracy of the handover score prediction of each AP and the selection of the optimal target AP based on the ResNet network model in the multi-factor handover decision module. On the training set and test set with randomly divided data (the ratio of the present invention is training set: validation set: test set = 7:1:2), the model trained by the present invention can achieve an accuracy rate close to 100%. The feasibility of using model prediction is proved.

[0129] In the second experiment, referring to Figure 5 and Figure 6 , and compared the changes in user achievable data rate under different switching conditions using the solution proposed by the present invention and the traditional 3GPP solution under the same path. It can be seen that the solution proposed by the present invention takes into account the AP load situation. At the same time, as an active switching solution, the corresponding switching resources can be prepared before the switching is started. The user can switch to an AP with higher service quality earlier. Compared with other solutions, the switching solution designed by the present invention has higher transmission performance.

[0130] In the third experiment, referring to Figure 7 and Figure 8 , the changes in the number of user switching and frequent switching under different switching conditions are compared between the scheme proposed by the present invention and the traditional 3GPP scheme, the non-active scheme that only considers multiple factors (without trajectory prediction and receiving power prediction), and the active scheme that does not consider the load (the load of the network base station) (with trajectory prediction and receiving power prediction). It can be seen that the scheme proposed by the present invention takes into account the user's residence time and switching cost at the target AP, and as a scheme with forward-looking long-term planning, it can select a better target AP in the long term. Therefore, compared with other schemes, the switching scheme designed by the present invention greatly reduces unnecessary switching, reduces the probability of frequent switching, and reduces unnecessary overhead.

[0131] It can be seen from the above experimental results that the present invention has the following technical effects:

[0132] 1. Through experimental data, it can be observed that the switching scheme designed by the present invention performs better than the control scheme in terms of switching times and data rate, which verifies the feasibility of the scheme. At the same time, the hardware, computing, training overhead and sample requirements of the scheme are all less than those of the background technology scheme.

[0133] 2. Compared with the single-task solution that only predicts the position or driving elements, the trajectory prediction based on the multi-task learning algorithm mechanism proposed in the present invention has improved the accuracy by 84.69% and 10.48% respectively.

[0134] 3. The method of constructing CKM using satellite images proposed in the present invention avoids the need for high computational costs and detailed geographic data for ray tracing, making it suitable for optimizing large-scale real-time coverage. At the same time, compared with the statistical model that cannot fully utilize environmental characteristics, the satellite image-based solution proposed in the present invention greatly improves RSRP accuracy. Taking the precise results of ray tracing as a benchmark, the satellite image-based solution improves MSE by 60.26% and MAE by 38.80% over the results of the statistical model.

[0135] In an embodiment of the present invention, a vehicle network switching device in a vehicle networking environment is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. The terms "module", "unit", "sub-unit", etc. used below may be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0136] The vehicle network switching device in the vehicle networking environment includes:

[0137] A trajectory prediction module is used to predict the future trajectory information of the target vehicle through a trained trajectory prediction model based on the historical driving knowledge graph of the target vehicle. The historical driving knowledge graph includes the historical position, historical speed and historical direction of the target vehicle.

[0138] A power prediction module is used to determine the receiving power of the target vehicle when it connects to each network base station in the future based on the future trajectory information of the target vehicle and the channel knowledge graph, where the channel knowledge graph includes the channel information of each location;

[0139] The time prediction module is used to determine the target vehicle's future stay time within the coverage area of ​​each network base station based on the target vehicle's future trajectory information;

[0140] The strategy determination module is used to determine the target network switching strategy of the target vehicle based on the target vehicle's future residence time within the coverage of each network base station, the received power and network base station switching cost when connecting to each network base station in the future, and the real-time load of each network base station.

[0141] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0142] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the vehicle network switching method in the vehicle networking environment provided by the present invention.

[0143] In an embodiment of the present invention, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the vehicle network switching method in the vehicle networking environment provided by the present invention are implemented.

[0144] It should be understood that the specific embodiments described herein are only used to explain the application, rather than to limit it. Based on the embodiments provided in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of this application.

[0145] Obviously, the drawings are only some examples or embodiments of the present application. For ordinary technicians in the field, the present application can also be applied to other similar situations based on these drawings without creative work. In addition, it is understandable that although the work done in this development process may be complicated and lengthy, for ordinary technicians in the field, certain changes in design, manufacturing or production based on the technical content disclosed in this application are only conventional technical means and should not be regarded as insufficient content disclosed in this application.

Claims

1. A vehicle network switching method in a vehicle networking environment, characterized in that: include: Based on a historical driving knowledge graph of the target vehicle, predicting future trajectory information of the target vehicle through a trained trajectory prediction model, wherein the historical driving knowledge graph includes a historical position, a historical speed, and a historical direction of the target vehicle; Determine the receiving power of the target vehicle when it connects to each network base station in the future based on the future trajectory information of the target vehicle and the channel knowledge graph, wherein the channel knowledge graph includes channel information of each position; Determine the future stay time of the target vehicle within the coverage area of ​​each network base station according to the future trajectory information of the target vehicle; The target network switching strategy of the target vehicle is determined based on the future residence time of the target vehicle within the coverage of each network base station, the receiving power and network base station switching cost when connecting to each network base station in the future, and the real-time load of each network base station.

2. The vehicle network switching method in the vehicle networking environment according to claim 1 is characterized in that: Based on the historical driving knowledge graph of the target vehicle, the future trajectory information of the target vehicle is predicted by the trained trajectory prediction model, including: Based on the historical driving knowledge graph of the target vehicle and the historical driving knowledge graph of the neighboring vehicles of the target vehicle, the future trajectory information of the target vehicle is predicted by the trained trajectory prediction model.

3. The vehicle network switching method in a vehicle networking environment according to claim 1, characterized in that: The trajectory prediction model includes a feature extraction module, a multi-task prediction module and a fusion module in sequence; The input of the feature extraction module is the input of the trajectory prediction model; The multi-task prediction module is used to perform a first prediction task and a second prediction task based on the output of the feature extraction module, respectively, the first prediction task is used to predict the first future position of the target vehicle, and the second prediction task is used to predict the future speed and direction of the target vehicle; The fusion module is used to fuse the first future position and the second future position of the target vehicle to predict the future trajectory information of the target vehicle.

4. The vehicle network switching method in the vehicle networking environment according to claim 3 is characterized in that: The feature extraction module includes a long short-term memory network, a bidirectional long short-term memory network and a multi-layer perceptron; The inputs of the long short-term memory network, the bidirectional long short-term memory network and the multi-layer perceptron are all inputs of the feature extraction module; The fusion result of the full connection result of the output of the long short-term memory network and the bidirectional long short-term memory network and the output of the multi-layer perceptron is used as the output of the feature extraction module.

5. The vehicle network switching method in a vehicle networking environment according to claim 3, characterized in that: The channel information at each location is the received power when the vehicle is connected to each network base station at each location; the steps of constructing the channel knowledge graph include: Based on satellite images, system states of each network base station and vehicle sample trajectories, the received power when the vehicle is connected to each network base station at different locations is predicted by a received power prediction model, thereby obtaining the channel knowledge graph; Among them, the receiving power prediction model includes a first feature extraction network, a second feature extraction network and a prediction network, the input of the first feature extraction network is the satellite image, the input of the second feature extraction network is the system status of each network base station and the vehicle sample trajectory, and the prediction network predicts the receiving power of the vehicle when it is connected to each network base station at different locations based on the output of the first feature extraction network and the second feature extraction network.

6. The vehicle network switching method in a vehicle networking environment according to claim 1, characterized in that: Based on the future residence time of the target vehicle within the coverage of each network base station, the future receiving power and network base station switching cost when connecting to each network base station, and the real-time load of each network base station, the target network switching strategy of the target vehicle is determined, including: Based on the target vehicle's future residence time within the coverage of each network base station, the received power and network base station switching cost when connecting to each network base station in the future, and the real-time load of each network base station, the trained strategy score prediction model is used to predict the scores of each network switching strategy of the target vehicle; The network switching strategy with the highest score is used as the target network switching strategy for the target vehicle.

7. The vehicle network switching method in a vehicle networking environment according to claim 6, characterized in that: The labels of the training data of the strategy scoring prediction model are generated by the TOPSIS technique.

8. A vehicle network switching device in a vehicle networking environment, characterized in that: include: A trajectory prediction module, used to predict the future trajectory information of the target vehicle through a trained trajectory prediction model based on a historical driving knowledge graph of the target vehicle, wherein the historical driving knowledge graph includes a historical position, a historical speed, and a historical direction of the target vehicle; A power prediction module, used to determine the receiving power of the target vehicle when it connects to each network base station in the future based on the future trajectory information of the target vehicle and the channel knowledge graph, wherein the channel knowledge graph includes channel information of each position; A time prediction module, used to determine the future stay time of the target vehicle within the coverage area of ​​each network base station according to the future trajectory information of the target vehicle; The strategy determination module is used to determine the target network switching strategy of the target vehicle based on the future residence time of the target vehicle within the coverage of each network base station, the receiving power and network base station switching cost when connecting to each network base station in the future, and the real-time load of each network base station.

9. An electronic device, comprising a memory and a processor, characterized in that: The memory stores a computer program, and the processor is configured to run the computer program to execute the vehicle network switching method in a vehicle networking environment according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the vehicle network switching method in a vehicle networking environment described in any one of claims 1 to 7 are implemented.

Images