A Power Control Method for Vehicle-to-Everything (V2X) Networks Based on Surrounding Node Density Prediction

Abstract

This invention discloses a power control method for vehicle-to-vehicle (V2X) networks based on surrounding node density prediction. The method specifically includes the following steps: constructing a dynamic vehicle-to-vehicle communication model; constructing an LSTM network for predicting the density of surrounding nodes; collecting vehicle location and density information on the road and training the constructed LSTM network; using the trained LSTM network to predict the surrounding density during vehicle data transmission and determine the current V2X network congestion level; and performing segmented power control based on the determination result. This invention introduces an LSTM network into C-V2X Mode 4, collects surrounding node density information within a sensing window, predicts node density during message transmission, and performs segmented power control. This effectively addresses the problem that half-duplex communication vehicles cannot be aware of environmental changes during transmission, improving packet delivery rate and network throughput.

Description

Technical Field

[0001] This invention relates to the fields of wireless communication and vehicle networking technology, and specifically to a vehicle networking power control method based on surrounding node density prediction. Background Technology

[0002] With social development and the continuous improvement of people's living standards, the increase in various means of transportation has greatly facilitated people's daily travel; on the other hand, the increasing number of vehicles on the road has also brought greater challenges to travel safety and efficiency.

[0003] The rapid development of vehicle-to-everything (V2X) communication has provided a solution to the aforementioned problems. To achieve high-quality vehicle-to-vehicle communication, 3GPP defined two communication modes in Release 14: Mode 3 and Mode 4. In Mode 3, when the vehicle is within the coverage area of ​​the cellular network, site resources are selected, allocated, and reserved by the eNodeB. In Mode 4, the vehicle can autonomously select and manage its radio resources without any cellular infrastructure support. Since full cellular coverage has not yet been achieved, Mode 4 C-V2X is considered the baseline mode for C-V2X.

[0004] Mode 4 defines a perception-based semi-persistent scheduling (SPS) scheme. In this scheme, vehicles periodically monitor channel resource usage and utilize available resources within a certain timeframe (the perception window) before sending a message. Existing research typically addresses network congestion in high-density environments by using channel information obtained during the perception period for power or rate control (e.g., SPS++) or optimizing the perception strategy to select more suitable radio resources (e.g., STS-RS). These schemes, based on information acquired during the perception window, cannot react promptly to changes in the surrounding environment during transmission, leading to reduced packet delivery rates and throughput, and making it difficult to maintain communication quality in high-density environments.

[0005] The shortcomings of a congestion control method and vehicle-mounted terminal (202110756540.2) in the prior art are as follows:

[0006] (1) The road environment during the vehicle's message transmission was not taken into account;

[0007] (2) The congestion control method is to modify the transmit power to a predicted ratio or a fixed value, which has a relatively aggressive control effect. Summary of the Invention

[0008] To overcome the shortcomings of existing congestion control technologies, which are susceptible to environmental changes when vehicles send messages, leading to reduced data packet delivery rate and throughput, this invention provides a vehicle-to-everything (V2X) power control method based on surrounding node density prediction. This method collects surrounding node density information within a sensing window and predicts the surrounding node density during transmission using an LSTM network, thus overcoming the impact of unknown environments on half-duplex communication vehicles. Simultaneously, it performs segmented and relatively stable control of transmission power, improving data packet delivery rate and network throughput.

[0009] The objective of this invention is achieved by at least one of the following technical solutions.

[0010] A power control method for vehicle-to-everything (V2X) networks based on surrounding node density prediction specifically includes the following steps:

[0011] S1. Construct a dynamic vehicle-to-vehicle communication model;

[0012] S2. Construct an LSTM network for predicting the density of nodes around a vehicle;

[0013] S3. Collect information on the location and density of vehicles on the road and train the constructed LSTM network.

[0014] S4. Use the trained LSTM network to predict the surrounding density during the vehicle data transmission period to determine the degree of vehicle network congestion in the current road segment.

[0015] S5. Perform segmented power control based on the judgment result in S4;

[0016] Further, step S1 is as follows:

[0017] S1.1 Establish a dynamic vehicle-to-vehicle communication scenario;

[0018] S1.2 Establish a vehicle-to-vehicle communication model.

[0019] Furthermore, in step S1.1, a dynamic vehicle-to-vehicle communication scenario is established, as detailed below:

[0020] N vehicles are traveling on a two-way four-lane highway segment with a total length of L and a lane width of W. The vehicle set is n = {1, 2, ..., N}. The vehicle positions follow a Poisson distribution based on the lane spacing. Assume the highway segment extends laterally from left to right, with the origin at the leftmost point of the first lane furthest from the second lane, the lane distribution direction as the x-axis, and the rightward extension direction of the lanes as the y-axis. Then the positions of the n vehicles are represented as (x... n y n ), where x n This indicates the vehicle's position on the x-axis, x n ∈[0,4*W],y nThis indicates the vehicle's position on the y-axis. n ∈[0,L];

[0021] Vehicles in lanes one and two travel in the same direction, while vehicles in lanes three and four travel in the opposite direction. All vehicles travel at a constant speed of v. Each vehicle is equipped with an LSTM prediction network module.

[0022] Furthermore, in step S1.2, a vehicle-to-vehicle communication model is established, wherein the main communication method is based on mode 4 specified in 3GP Release 14, that is, the vehicle autonomously selects radio resources and performs message reception and transmission without the aid of a base station, as detailed below:

[0023] If a vehicle sends a message to another vehicle at a certain time t, it needs to monitor radio channel resources within the perception window. The perception window is evenly divided into 10 sub-windows. During each sub-window, the vehicle monitors the usage of each resource block and selects the resource block it needs to send a message. The number of times RC is sent is a random number between [5,15]. After each message is sent, the value of RC is decremented by one. When RC returns to zero, the resource needs to be monitored again, selected, and sent. The steps are the same as before.

[0024] In addition to monitoring channel resources, the vehicle also records the number of surrounding nodes it can perceive within each sub-window, denoted as N. neighbor ={n1,n2,…,n 10}; Then, N neighbor It is fed as input to the LSTM network for training and optimization.

[0025] Further, in step S2, the input to the constructed LSTM network for predicting the density of nodes around the vehicle is the surrounding density array obtained for each perception window; a linear layer is added at the end of the LSTM network;

[0026] The entire LSTM network is trained and optimized, and the last state of the LSTM network is passed through the output of this layer to obtain a predicted surrounding density value ρ. pred This is used in the subsequent power control process.

[0027] Furthermore, in step S3, the data collection and LSTM network training process is as follows:

[0028] S3.1. Collect and preprocess data to obtain training, validation, and test sets;

[0029] S3.2 Input the surrounding density within the sensing window, and update the network weights based on the surrounding density predicted by the LSTM network at time t and the actual density value at time t.

[0030] S3.3 Repeat step S3.3 until the entire training set has been traversed;

[0031] S3.4. Evaluate the LSTM network using the validation set and optimize the relevant hyperparameters;

[0032] S3.5 Record multiple LSTM network models with similar test results on the validation set, evaluate them using the test set, and select the LSTM network with the best evaluation result for subsequent prediction.

[0033] Furthermore, in step S3.1, since it is necessary to make predictions based on the ten surrounding density values ​​of the perception window, the surrounding density obtained by each perception window is used as a set of inputs, and the surrounding density of the resource selection window is used as the output. These are then converted into array format and normalized.

[0034] The normalization process is represented as:

[0035] Seq_opt=(Seq–mean(Seq)) / std(Seq)

[0036] Where Seq_opt represents the normalized data, Seq represents the initial data obtained, mean(*) represents the mean of the variable in parentheses, and std(*) represents the variance of the variable in parentheses; then the dataset is divided into training set, validation set and test set according to the obtained data, with a ratio of 7:2:1.

[0037] Furthermore, in step S4, based on the surrounding node density ρ predicted by the LSTM network... pred The size indicates the degree of network congestion in the scenario;

[0038] The maximum number of surrounding nodes is defined as ρ. max The busy factor η can be obtained, expressed as:

[0039] η = ρ pred / ρ max

[0040] Different node density ranges are defined based on the magnitude of η, representing low density, moderate density, and high density, respectively; let η upper η represents the boundary between high-density and medium-density regions. lower This represents the boundary between medium and low density, and power control is performed based on the interval where η is located.

[0041] Furthermore, in step S5, the density factor η in different intervals and the transmission power P at time n are determined. n Set the next transmission power P n+1 , means as follows:

[0042] Pn+1 =P n +P change

[0043] Among them, P change This indicates the amount of transmission power adjusted each time.

[0044] Furthermore, depending on the different surrounding densities, P change The specific values ​​for are as follows:

[0045] When η > η upper This means that vehicles are in a high-density environment:

[0046] P change =P tx- ,

[0047] Among them, P tx- This indicates the required reduction in transmission power. When the density of surrounding nodes is high, all vehicles in the scene need to reduce their transmission power to reduce interference. Considering the good nonlinear characteristics of sigmoid, it is often used as an activation function in neural networks, expressed as:

[0048]

[0049] Therefore, using the sigmoid function to adjust the vehicle's transmit power allows for more flexible adaptation to changing node densities; the reduction in transmit power is expressed as:

[0050] P tx- =-Asigmoid((η-η) upper ) / (η upper -η lower ))

[0051] Where A is the increment value for each power change; to prevent insufficient transmission power from affecting communication quality, P is specified. min A value of 10 dBm represents the vehicle's minimum transmit power. Therefore, the power reduction is limited by:

[0052] P tx- =min(P tx- ,P n -P min );

[0053] When η lower <η<η upper If the node density on the road is moderate, then it is desirable to gradually increase the vehicle's transmission power to improve the propagation distance of vehicle messages. The increase in transmission power in this case is expressed as:

[0054]

[0055] Where θ = 0.5 is the smoothing factor;

[0056] S5.3, When η < η lower This refers to a situation where vehicles are in a low-density environment. In this case, it is necessary to increase the vehicle's transmission power to improve the propagation range of message packets. The increase in transmission power in this situation is expressed as:

[0057] P change =P tx+

[0058] Among them, P tx+ This indicates the required increase in transmission power, expressed as...

[0059] P tx+ =A-Asigmoid((η lower -η) / (η upper -η lower ))

[0060] Due to limitations in the physical performance of the actual equipment, the maximum transmission power is P. max =23dBm, therefore the increase in transmit power is limited by:

[0061] P tx+ =min(P tx+ ,P max -P n ).

[0062] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0063] (1) The present invention discloses a vehicle network power control method based on surrounding node density prediction, which uses prediction to simulate full-duplex communication, reduces the impact of vehicles not being able to obtain surrounding node information during transmission in half-duplex communication mode, and can effectively improve the data packet delivery rate and throughput of communication.

[0064] (2) The present invention adjusts the range of transmission power variation according to the actual situation, which can flexibly adapt to the vehicle network environment under different node densities, and the model has good stability. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of a vehicle-to-vehicle communication scenario in an embodiment of the present invention;

[0066] Figure 2 This is a flowchart of a vehicle-to-everything (V2X) power control method based on surrounding node density prediction disclosed in an embodiment of the present invention.

[0067] Figure 3 This is a schematic diagram of the perception-based semi-persistent scheduling algorithm process in an embodiment of the present invention;

[0068] Figure 4 This is a schematic diagram of the LSTM network structure in an embodiment of the present invention;

[0069] Figure 5 This is a comparison chart of the impact of the number of vehicles on the data packet delivery rate of different schemes in this embodiment of the invention;

[0070] Figure 6 This is a comparison chart of the impact of the number of vehicles on the system throughput of different schemes in embodiments of the present invention;

[0071] Figure 7 This is a comparison chart of the cumulative distribution function of the average network latency in embodiments of the present invention. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0073] Example 1:

[0074] like Figure 2 The illustration shows a vehicle-to-everything (V2X) power control method based on surrounding node density prediction disclosed in this embodiment. In the perception-based semi-persistent scheduling process, an LSTM network is added as an auxiliary to collect surrounding density data, predict and control the transmit power. Specific steps include:

[0075] S1, Establish as follows Figure 1 The image shows a vehicle-to-vehicle communication scenario. The steps for establishing this scenario are as follows:

[0076] S1.1 Establish a dynamic vehicle-to-vehicle communication scenario. N vehicles are traveling on a two-way four-lane highway segment with a total length of L and a lane width of W. The vehicle set is n = {1, 2, ..., N}; the vehicle positions follow a Poisson distribution based on the lane spacing. Assume the highway segment extends laterally from left to right, with the origin at the leftmost point of the first lane furthest from the second lane, the lane distribution direction as the x-axis, and the rightward extension direction of the lanes as the y-axis. Then the positions of the n vehicles are represented as (x... n y n ), where x n This indicates the vehicle's position on the x-axis, x n ∈[0,4*W],y n This indicates the vehicle's position on the y-axis. n ∈[0,L]. Vehicles in lanes one and two travel in the same direction, while vehicles in lanes three and four travel in the opposite direction to those in lanes one and two. All vehicles travel at a constant speed of v; each vehicle is equipped with an LSTM prediction network module.

[0077] S1.2 Establish a vehicle-to-vehicle communication model, in which the main communication method is based on Mode 4 specified in 3GPP Release 14, that is, the vehicle autonomously selects radio resources and performs message reception and transmission without the help of a base station, as detailed below:

[0078] In one embodiment, the sensing window is set to 1000ms, such as Figure 3 The semi-persistent scheduling process illustrated involves a vehicle sending a message to other vehicles at a certain time t. This requires monitoring radio channel resources within 1000ms prior to time t. The proposed method divides this sensing window evenly into 10 sub-windows, each lasting 100ms. During each sub-window, the vehicle monitors the usage of each resource block and selects the required resource block to send a message. The number of transmissions, RC, is a random number between [5, 15]. After each message transmission, the value of RC is decremented by one. When RC reaches zero, the resource monitoring, selection, and transmission need to be repeated, following the same steps as before.

[0079] In addition to monitoring channel resources, the vehicle also records the number of surrounding nodes it can perceive within each sub-window, denoted as N. neighbor ={n1,n2,…,n 10 Then, N neighbor It is fed as input to the LSTM network for training and optimization.

[0080] S2. Construct an LSTM network to predict the density of nodes around a vehicle.

[0081] The constructed LSTM network takes as input the surrounding density array obtained for each receptive window; a linear layer is added at the end of each LSTM unit. The entire network is trained and optimized, and the last state of the network is passed through the output of this layer to obtain a predicted surrounding density value ρ. pred This is used in the subsequent power control process.

[0082] S3. Collect the location and density information of vehicles on the road, and train the constructed LSTM network; the data collection and LSTM network training process is as follows:

[0083] S3.1 Data Collection and Preprocessing. Since prediction is required based on ten surrounding density values ​​of the sensing window, the surrounding density obtained from each sensing window is used as a set of inputs, and the surrounding density of the resource selection window is used as the output. These are then converted to array format and normalized. The normalization process is as follows:

[0084] Seq_opt=(Seq–mean(Seq)) / std(Seq)

[0085] Where Seq_opt represents the normalized data, Seq represents the initial data obtained, mean(*) represents the mean of the variable in parentheses, and std(*) represents the variance of the variable in parentheses; then the dataset is divided into training set, validation set and test set according to the obtained data, with a ratio of 7:2:1.

[0086] S3.2 Input the surrounding density within 1000ms before time t, and update the network weights based on the predicted surrounding density at time t and the actual density value at time t.

[0087] S3.3 Repeat step S3.3 until the entire training set has been traversed.

[0088] S3.4. Evaluate the network using the validation set and optimize the relevant hyperparameters.

[0089] S3.5 Record models with similar test results on multiple validation sets, evaluate them using the test set, and select the network with the best test results for subsequent prediction.

[0090] S4. Use the trained LSTM network to predict the surrounding density during vehicle data transmission to determine the congestion level of the current road segment; define the maximum number of surrounding nodes as ρ. max The busy factor η can be obtained, expressed as:

[0091] η = ρ pred / ρ max

[0092] Different node density ranges are defined based on the magnitude of η, representing low density, moderate density, and high density, respectively. Let η upper η represents the boundary between high-density and medium-density regions. lower Represents the boundary between medium and low density, and performs power control by partitioning according to the interval where η is located;

[0093] S5. Based on the density factor η in different intervals and the transmission power P at time n n Set the next transmission power P n+1 , means as follows:

[0094] P n+1 =P n +P change

[0095] Where P change This indicates the magnitude of the transmission power adjustment each time. Depending on the surrounding density, P... change They are represented as follows:

[0096] When η > η upper This means that vehicles are in a high-density environment:

[0097] P change =P tx- ,

[0098] Among them, P tx- This represents the required reduction in transmission power. When the surrounding node density is high, all vehicles in the scene need to reduce their transmission power to minimize interference. Considering the favorable non-linear characteristics of the sigmoid function, it is commonly used as an activation function in neural networks, expressed as:

[0099]

[0100] Therefore, using the sigmoid function to adjust the vehicle's transmit power allows for more flexible adaptation to varying node densities. The transmit power reduction magnitude P tx- Represented as:

[0101] P tx- =-Asigmoid((η-η) upper ) / (η upper -η lower ))

[0102] Where A is the increment value for each power change. In one embodiment, to achieve a trade-off between channel utilization and packet delivery rate, the value of A is specified as 5.0 to prevent the power adjustment process from being too aggressive; to prevent communication quality from being affected by insufficient transmit power, P is specified as... min A value of 10 dBm represents the vehicle's minimum transmit power. Therefore, the power reduction is limited by:

[0103] P tx- =min(P tx- ,P n -P min )

[0104] When η lower <η<η upper In other words, given a moderate node density on the road, we aim to gradually increase the vehicle's transmission power to extend the propagation distance of vehicle messages. The increase in transmission power, P, is then calculated. change Represented as:

[0105]

[0106] Where θ = 0.5 is the smoothing factor;

[0107] When η < η lower This refers to a situation where vehicles are in a low-density environment. In this case, it is necessary to increase the vehicle's transmission power to extend the propagation range of message packets. The increase in transmission power, P, is then... tx+ Represented as:

[0108] Pchange =P tx+

[0109] Among them, P tx+ This indicates the required increase in transmission power, expressed as...

[0110] P tx+ =A-Asigmoid((η lower -η) / (η upper -η lower) )

[0111] Due to limitations in the physical performance of the actual equipment, the maximum transmission power is P. max =23dBm, therefore the increase in transmit power is limited by:

[0112] P tx+ =min(P tx+ ,P max -P n )

[0113] The specific simulation parameters for this embodiment are shown in Table 1.

[0114] Table 1. Simulation Parameter Settings

[0115]

[0116]

[0117] In one embodiment, Figure 5 , Figure 6 , Figure 7 The charts show the results of Mode 4 communication using the present invention and three methods: the 3GPP-specified baseline method (Method 1), the method for rate control based on channel usage (Method 2), and the method for power control based on channel usage (Method 3). Method 2 determines the congestion level based on the channel busy rate and adjusts the transmission rate to a preset value or proportion according to the level of congestion. Method 3 determines the congestion level based on the channel busy rate and adjusts the transmission power to a preset value or proportion according to the level of congestion.

[0118] Figure 5 This is a graph showing the relationship between the number of vehicles and the data packet delivery rate. As the graph shows, the overall data packet delivery rate decreases with the increase in the number of vehicles on the road. Compared to Method 1, the proposed method, along with Methods 2 and 3, employs certain congestion control measures, thus achieving a higher data packet delivery rate and improving communication quality between vehicles in high-density environments. However, the proposed method flexibly controls the power variation based on node density, making it relatively more stable in high-density environments. Compared to the other three methods, the proposed method achieves a higher data packet delivery rate for vehicle nodes.

[0119] Figure 6 This graph shows the relationship between the number of vehicles and the average network throughput. As the graph shows, the average network throughput decreases with increasing vehicle density. Although Method 2 reduces channel load by lowering the transmission rate, the number of messages transmitted in the same amount of time is significantly reduced, leading to a decrease in throughput. Method 3's power control method is more aggressive, drastically reducing the message propagation range, thus significantly reducing the number of nodes that can receive messages, resulting in a significant decrease in throughput. The proposed method performs stable power control, and the algorithm is independent of the transmission rate; therefore, compared with the other three methods, the average network throughput is higher.

[0120] Figure 7 This is a cumulative distribution function graph of the inter-packet reception interval (IPG) for different methods when the number of vehicles is 250 vehicles / km. As shown in the graph, Scheme 2 increases the IPG due to the reduced transmission rate. The proposed method does not involve the data transmission rate and is not significantly different from Methods 1 and 3, thus maintaining the timeliness of vehicle-to-vehicle communication.

[0121] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A power control method for vehicle-to-everything (V2X) networks based on surrounding node density prediction, characterized in that, Specifically, the steps include the following: S1. Construct a dynamic vehicle-to-vehicle communication model; S2. Construct an LSTM network for predicting the density of nodes around a vehicle; S3. Collect information on the location and density of vehicles on the road and train the constructed LSTM network. S4. Utilize the trained LSTM network to predict the density of surrounding nodes during vehicle data transmission to determine the current level of vehicle-to-everything (V2X) congestion on the road segment; based on the surrounding node density predicted by the LSTM network... The size indicates the degree of network congestion in the scenario; The maximum value of the surrounding node density is specified as follows: The busy factor can be obtained. , is represented as: = according to The size of the surrounding nodes is used to divide different density ranges, representing low density, moderate density, and high density respectively; let... Indicates the boundary between high density and medium density. Indicates the boundary between medium and low density, according to Power control is implemented in zones within the specified area; S5. Perform segmented power control based on the judgment results in S4; based on the density factor in different intervals... and the transmission power at time n Set the next transmission power , means as follows: in, This indicates the magnitude of the transmission power adjustment each time; Depending on the density of surrounding nodes The specific values ​​for are as follows: when This means that vehicles are in a high-density environment: in, This represents the amount of emission power that needs to be reduced; using sigmoid as the activation function in a neural network, it is expressed as: The sigmoid function is used to adjust the vehicle's transmission power; the reduction in transmission power is expressed as: A (( ) / ( )) Where A is the increment value of each power change; The minimum transmission power of the vehicle is indicated, and the power reduction is limited by: = min( , ); when When the node density on the road is moderate, the increase in transmission power is expressed as: in, = 0.5, which is the smoothing factor; This is the maximum transmission power; S5.3, when In other words, when the vehicle is in a low-density environment, the increase in transmission power is expressed as follows: in, The required increase in transmission power is represented as... Therefore, the extent to which the transmission power can be increased is limited by: = min( , )。 2. The vehicle-to-everything (V2X) power control method based on surrounding node density prediction according to claim 1, characterized in that, Step S1 is as follows: S1.1 Establish a dynamic vehicle-to-vehicle communication scenario; S1.2 Establish a vehicle-to-vehicle communication model.

3. The vehicle-to-everything (V2X) power control method based on surrounding node density prediction according to claim 2, characterized in that, In step S1.1, a dynamic vehicle-to-vehicle communication scenario is established, as detailed below: N vehicles are traveling on a two-way four-lane highway segment with a total length of L and a lane width of W. The vehicle set is {1, 2, …, N}, and the vehicle positions follow a Poisson distribution based on the lane spacing. Assume the highway segment extends laterally from left to right, with the origin at the leftmost point of the first lane furthest from the second lane, the lane distribution direction as the x-axis, and the rightward extension direction of the lanes as the y-axis. Then, the… i The position of a vehicle is represented as ( ),in This indicates the vehicle's position on the x-axis. ∈[0, 4*W], Indicates the vehicle's position on the y-axis. ∈[0, L]; Vehicles in lanes one and two travel in the same direction, while vehicles in lanes three and four travel in the opposite direction. All vehicles travel at a constant speed of v. Each vehicle is equipped with an LSTM prediction network module.

4. The vehicle-to-everything (V2X) power control method based on surrounding node density prediction according to claim 3, characterized in that, In step S1.2, a vehicle-to-vehicle communication model is established, where the communication method is based on mode 4, that is, the vehicle autonomously selects radio resources and performs message reception and transmission without the aid of a base station, as detailed below: If a vehicle sends a message to another vehicle at a certain time t, it needs to monitor radio channel resources within the perception window. The perception window is evenly divided into 10 sub-windows. During each sub-window, the vehicle monitors the usage of each resource block and selects the resource block it needs to send a message. The number of times RC is sent is a random number between [5, 15]. After each message is sent, the value of RC is decremented by one. When RC returns to zero, the resources need to be monitored again, selected, and sent again. The steps are the same as before. In addition to monitoring channel resources, the vehicle also records the number of surrounding nodes it can perceive within each sub-window, represented as... ={ , , …, }; then, It is fed as input to the LSTM network for training and optimization.

5. The vehicle-to-everything (V2X) power control method based on surrounding node density prediction according to claim 1, characterized in that, In step S2, the input to the LSTM network constructed for predicting the density of nodes around the vehicle is the array of surrounding node densities obtained for each perception window; a linear layer is added at the end of the LSTM network. The entire LSTM network is trained and optimized, and the last state of the LSTM network is passed through the output of this linear layer to obtain a predicted density of surrounding nodes. .

6. The vehicle-to-everything (V2X) power control method based on surrounding node density prediction according to claim 5, characterized in that, In step S3, the data collection and LSTM network training process is as follows: S3.

1. Collect and preprocess data to obtain training, validation, and test sets; S3.2 Input the density of surrounding nodes within the perception window, and update the network weights based on the surrounding node density predicted by the LSTM network at time t and the actual density value at time t. S3.3 Repeat step S3.2 until the entire training set has been traversed; S3.

4. Evaluate the LSTM network using the validation set and optimize the relevant hyperparameters; S3.5 Record multiple LSTM network models with similar test results on the validation set, evaluate them using the test set, and select the LSTM network with the best evaluation result for subsequent prediction.

7. The vehicle-to-everything (V2X) power control method based on surrounding node density prediction according to claim 6, characterized in that, In step S3.1, the surrounding node density obtained by each perception window is used as a set of inputs, and the surrounding node density of the resource selection window is used as the output. These are converted into array format and normalized respectively. The normalization process is represented as: Where Seq_opt represents the normalized data, Seq represents the initial data obtained, mean(*) represents the mean of the variable in parentheses, and std(*) represents the variance of the variable in parentheses; then the dataset is divided into training set, validation set and test set according to the obtained data, with a ratio of 7:2:1.

Images