Dynamic beam adaptive optimization method for Internet of Vehicles

Abstract

The invention discloses a dynamic beam adaptive optimization method for the Internet of Vehicles, and the method comprises the steps: firstly obtaining the current position, speed and driving direction of a vehicle, and carrying out the prediction of a future trajectory through combining historical trajectory information and employing a Kalman filtering technology and the like; and selecting a candidate beam direction from the predefined beam codebook. The system receives channel quality feedback information of a communication link in real time, and dynamically updates a beam forming weight based on an improved recursive least square algorithm. According to the algorithm, a composite adaptive adjustment mechanism of an error size and an error change trend is introduced, a multi-objective optimization function is constructed, and performance indexes such as link quality maximization, interference power minimization and beam switching frequency minimization are effectively balanced in a dynamic environment, so that an optimal beam direction is obtained. Finally, the real-time switching of the antenna array beams is realized through a low-overhead and low-delay control protocol. According to the method, track perception and a self-adaptive feedback mechanism are fused, and good real-time performance, stability and expandability are achieved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of wireless communication and intelligent transportation, and particularly relates to a dynamic beam adaptive optimization method for vehicle networking. Specifically, the present application belongs to the technical field of beam management and resource scheduling in vehicle-to-everything (V2X) communication, millimeter wave communication, and massive multiple-input multiple-output (MIMO) systems, and is suitable for antenna beam direction prediction and adaptive control in high-speed mobile scenarios in 5G / 6G communication systems, and has a wide engineering application prospect. BACKGROUND

[0002] With the rapid development of intelligent transportation systems (ITS) and vehicle networking (V2X) technologies, vehicles have higher requirements for wireless communication with high reliability, low delay, and large bandwidth. In particular, in 5G and future 6G communication systems, millimeter wave frequency bands and massive MIMO antenna arrays are widely used to provide higher data rates and spectrum utilization efficiency. In vehicle networking communication systems for high-speed mobile scenarios, beamforming technology has become one of the key means to improve link stability and throughput performance.

[0003] However, due to the high-speed mobility of vehicles and the complex dynamic characteristics of the communication environment, existing beam management mechanisms still face many challenges in practical applications, mainly in the following aspects:

[0004] (1) Large beam switching delay: Existing systems mostly use fixed periods or simple triggering mechanisms for beam switching, lack real-time adaptive ability, and cause switching response lag, which easily leads to link interruption or communication quality degradation.

[0005] (2) Strong unpredictability of channel state: In vehicle networking environments, strong multipath propagation, moving obstacles, and rapid changes in interference sources often occur, causing the channel state to fluctuate dramatically, and traditional static or periodic estimation methods are difficult to accurately capture.

[0006] (3) System resource waste caused by high-frequency switching: Frequent beam switching will significantly increase control signaling overhead, increase system load and energy consumption, especially in massive antenna array systems, which puts significant pressure on physical resource scheduling.

[0007] (4) Lack of trajectory awareness mechanism: Most existing beam control algorithms do not fully consider vehicle motion trends and trajectory prior information, resulting in lack of foresight in beam adjustment and affecting the sustained tracking ability of beam pointing.

[0008] To solve the above problems, the existing beam management method has obvious shortcomings in the face of high-speed movement, unstable channel and resource limited scenes. Therefore, it is urgent to propose a dynamic beam adaptive optimization method with vehicle trajectory prediction ability, real-time channel feedback optimization mechanism and low overhead beam control scheme to adapt to high-speed dynamic communication environment and ensure the link stability and communication quality of vehicle networking system. SUMMARY

[0009] The purpose of the present application is to provide a dynamic beam adaptive optimization method for vehicle networking. It aims to solve the problems of beam tracking lag, channel change, and interference suppression in high-speed mobile scenarios. Based on real-time vehicle motion information, the method integrates trajectory prediction, channel feedback, beam selection and weight optimization modules, and builds a dynamic closed-loop beam adaptive control mechanism.

[0010] TECHNICAL SCHEME: The dynamic beam adaptive optimization method for vehicle networking of the present application comprises the following steps:

[0011] S1, obtaining the current position, speed and direction information of the target vehicle;

[0012] S2, based on Kalman filtering algorithm, combining historical trajectory and current state, predicting the future motion trajectory of the vehicle in a predetermined time period;

[0013] S3, according to the predicted trajectory direction, selecting multiple candidate beam directions from the beam codebook to construct a candidate beam set;

[0014] S4, real-time receiving communication link feedback information, including signal to noise ratio SINR, received signal strength indication RSSI and channel quality indication CQI, and updating at fixed time interval or event triggered mode;

[0015] S5, constructing a multi-objective beam optimization objective function, the target including maximizing SINR, minimizing interference power and beam switching frequency;

[0016] S6, based on the improved RLS algorithm, combining channel feedback and candidate beam information, adaptively and dynamically adjusting the beam forming weight, and outputting multiple optimized weight solutions;

[0017] S7, according to the preset beam performance index, evaluating and sorting multiple beam schemes;

[0018] S8, outputting the optimal beam direction and its weight as the beam control scheme;

[0019] S9, through the low delay signaling interface, controlling the antenna array to complete the real-time switching of beam direction, and realizing the cooperative scheduling between vehicle terminal and roadside unit.

[0020] Further, step S1 specifically includes the following steps:

[0021] S1.1, position acquisition: the current position of the vehicle can be acquired in real time by a high-precision GNSS module, and the inertial navigation system INS is used for short-time position interpolation and occlusion compensation;

[0022] S1.2, speed acquisition: the vehicle speed can be obtained by reading the vehicle CAN bus, or calculated by the difference between continuous positions of GNSS, or obtained by fusing IMU accelerometer and wheel speed sensor;

[0023] S1.3, driving direction acquisition: the driving direction angle can be calculated by the trajectory solution of the GNSS module, or output in real time by the magnetic sensor and gyroscope sensor of the vehicle; the angle information is used to construct a direction vector in the trajectory prediction module, to determine whether the beam main lobe pointing direction is aligned with the vehicle forward trend;

[0024] S1.4, data fusion and reporting: all collected state parameters will be filtered and time-aligned locally, and then sent to the roadside unit or edge computing node through the V2X communication link within a certain period; the state data format includes: Position(x, y), Speed(v), Heading(θ), Timestamp(t).

[0025] S1.5, abnormal state processing: when the positioning signal is missing or the vehicle speed suddenly changes beyond the set threshold, the system will automatically start the abnormal detection mechanism, suspend the call of the beam prediction module, and prevent the input error from being transmitted to the subsequent module.

[0026] Further, step S2 specifically includes the following steps:

[0027] S2.1, state modeling: the motion state of the vehicle is modeled as a linear dynamic system, and its state vector includes position, speed and heading direction; the state variable is set as:

[0028] X k =[x k ,y k ,v k ,θ k ] T

[0029] Where: x k , y k are the current position coordinates, v k is the speed, and θ k is the heading angle;

[0030] S2.1, state transition model: the transition of vehicle state between adjacent time can be modeled as:

[0031] X k = A · X k + ω k

[0032] X k = A · X k + ω k

[0033] where: A is state transition matrix, based on vehicle kinematics modeling; ω k process noise, subject to zero mean Gaussian distribution, covariance is Q;

[0034] State transition matrix A is:

[0035]

[0036] Time interval is Δt;

[0037] S2.1, definition and role of covariance matrix P:

[0038] In Kalman filtering, covariance matrix P k represents the degree of uncertainty of the current state estimation of the system, quantifies the range of prediction error, and is used to calculate Kalman gain K k , which determines the fusion ratio of prediction and observation, and the coefficient determines the enhancement of prediction dependence when the observation data is unstable and the correction effect when the observation is accurate; that is:

[0039]

[0040] where, X k is the true state value of vehicle operation, is the predicted state value, and E[·] is the expected value between the true value and the predicted value;

[0041] S2.4, Kalman prediction and update process: Kalman filtering includes: prediction stage and update stage;

[0042] Prediction stage:

[0043]

[0044] P k|k-1 = A · P k-1|k-1 · A T + Q

[0045] where: K k is the Kalman gain, used to weigh the contribution of the predicted value and the observed value to the final estimate; H is the observation matrix, which maps the state space to the observation space; R is the observation noise covariance matrix, which reflects the uncertainty of observation. If the value is small, the gain is large, and the observation is biased; P k|k-1State covariance matrix in prediction stage, used to measure prediction uncertainty, if the value is small, the gain is small, and the prediction is biased;

[0046] Update stage combines actual observation value Z k :

[0047] K k = P k|k-1 ·H T ·(H·P k|k-1 ·H T +R) -1

[0048]

[0049] P k|k =(I-K k ·H)P k|k-1

[0050] Wherein: K k is the Kalman gain, used to weigh the contribution of prediction value and observation value to the final estimate; H is the observation matrix, which maps the state space to the observation space; R is the observation noise covariance matrix, which reflects the uncertainty of observation, if the value is small, the gain is large, and the observation is biased; P k|k-1 State covariance matrix in prediction stage, used to measure prediction uncertainty, if the value is small, the gain is small, and the prediction is biased; I is the unit matrix.

[0051] S2.5, trajectory output in time window design:

[0052] The system finally outputs the predicted position sequence in a period of time in the future:

[0053] {(x k+1 ,y k+1 ), (x k+2 ,y k+2 ), …, (x k+n ,y k+n )}

[0054] Time window T pred Adjust dynamically according to vehicle speed to cover a beam update cycle.

[0055] Further, step S3 specifically comprises the following steps:

[0056] S3.1, beam codebook definition and formatting: the system pre-constructs a beam direction codebook Containing N standard beam directions, each direction corresponding to a set of fixed array beam forming vectors W i , the codebook format is:

[0057]

[0058] where θ i denotes the i-th beam lobe center direction; beam codebook can be uniformly discretized, such as every 5° or 10°;

[0059] S3.2, trajectory angle extraction and angular mapping: extract the vehicle's future average heading angle from the S2 trajectory prediction module output defined as:

[0060]

[0061] where θ k is the predicted heading angle in each time window; combined with the trajectory curvature prediction trend to determine whether to pre-select a wide beam or multiple directions, after mapping the predicted angle to the codebook space angular domain, the system calculates the angle difference between each beam direction and

[0062] S3.3, candidate beam screening strategy: set the angle tolerance threshold δ, and screen the beam directions that meet the conditions to form the candidate set:

[0063]

[0064] Optional expansion strategy: ① bandwidth matching strategy: dynamically adjust δ combined with vehicle speed / prediction uncertainty, expand the tolerance angle in high-speed scenarios, and use narrow angle in low-speed scenarios; ② bilateral balance strategy: ensure that the candidate beams are distributed symmetrically around the predicted main direction, improving the matching robustness; ③ probability weight initialization: each candidate direction can be initially assigned a probability weight based on the matching degree, which is used for subsequent reinforcement learning or multi-arm Bandit beam selection algorithm;

[0065] S3.4, multi-resolution beam set design

[0066] To adapt to the prediction accuracy and direction distribution density in different scenarios, the codebook adopts a multi-resolution structure: coarse-grained beam Low-Res and fine-grained beam High-Res;

[0067] S3.5, collaborative screening mechanism

[0068] If the candidate beam screening task is completed by the RSU or edge computing node in collaboration, send the vehicle's predicted angle information to the edge platform through the low-overhead control link, and complete the screening in the codebook by the platform, and then send the candidate beam set to the vehicle;

[0069] S3.6, structured candidate beam output

[0070] ​After the completion of the candidate beam direction screening, the system will output a structured candidate beam set results, and pass it to the subsequent channel feedback evaluation module and beam weight optimization module.

[0071] Further, step S5 is specifically:

[0072] The constructed objective function includes the following three main optimization objectives:

[0073] Maximize the communication quality, i.e., maximize the SINR

[0074] The goal is to improve the effective communication signal quality of the receiving end, enhance the robustness and modulation capacity of the channel, and use the average SINR to represent:

[0075]

[0076] Where: W is the current beamforming weight vector; h is the target channel vector; h j The channel vector corresponding to the interference direction; σ 2 System thermal noise power;

[0077] Minimize the interference power Interference Suppression

[0078] The goal is to reduce the interference energy of the beam direction leaking to other users or directions, and define the interference power term as:

[0079]

[0080] It can be constrained by constructing null steering or minimum leakage beam form to ensure beam directivity;

[0081] Minimize the beam switching cost Stability-aware Smoothness

[0082] In order to prevent the link instability and system load caused by frequent switching, a beam change penalty term is added:

[0083] f3(W t ,W t-1 )=||W t -W t-1 ||

[0084] Where: W t is the current frame beam weight; W t-1 is the beam weight of the last frame; This term measures the beam change intensity, which is used to control the energy consumption and delay caused by frequent direction changes;

[0085] Multi-objective optimization function construction

[0086] The final objective function is constructed in the form of weighted linear combination as:

[0087] J(W) = a f1(W) + β f2(W) + γ f3(W) t ,W t-1 )

[0088] Wherein: a, β, γ ∈ [0, 1] are weight factors of each objective, used to balance the optimization priority between different indicators; each weight can be statically set, or dynamically adjusted according to the scene or channel state.

[0089] Further, step S6 specifically includes the following steps:

[0090] S6.1, the traditional RLS is used to minimize the following cost function:

[0091]

[0092] Wherein: W is the beamforming weight vector; x(k) is the kth sampling input; d(k) is the expected signal; λ is the forgetting factor, which controls the degree of attenuation of historical sample weight (0 < λ ≤ 1);

[0093] The weight update formula of RLS is:

[0094]

[0095] W(t) = W(t-1) + K(t) e * (t)

[0096] P(t) = λ -1 [P(t-1) - K(t) x H (t) P(t-1)]

[0097] Wherein: P(t) is the covariance matrix, e(t) = d(t) - W H (t-1) x(t) is the current error;

[0098] S6.2, adaptive adjustment of the forgetting factor λ:

[0099] A compound mechanism based on error size and error trend is proposed to dynamically adjust the forgetting factor, thereby controlling the balance between learning rate and stability;

[0100] λ(t) = λ min + (λ max - λ min ) · (1 - tanh (a r · |e(t)| + β r · Δe(t)))

[0101] Wherein: |e(t)| is the current error amplitude, reflecting the estimation bias; Δe(t) = e(t) - e(t-1) is the error trend; α r and β r are the adjustment weight factors, used to control the sensitivity of each component; λ min , λ max control the adjustable range of λ, usually set as [0.90, 0.9999];

[0102] S6.3, coupling update strategy with multi-objective function

[0103] The system regards the target function value error term as the equivalent expected error in RLS, dynamically drives the weight iteration process, so that RLS becomes a gradient-guided target minimization process controller;

[0104]

[0105] S6.4, algorithm execution flow

[0106] Initialize the weight W(0), the covariance matrix P(0) = δ -1 I;

[0107] For each time t, get the input x(t), the expectation d(t);

[0108] Calculate the error e(t) and the change amount Δe(t);

[0109] Calculate the current forgetting factor λ(t);

[0110] Iterate K(t), W(t), P(t) according to the update formula;

[0111] Output the latest weight W(t) to the beam control module.

[0112] Further, step S7 specifically includes the following steps:

[0113] S7.1, candidate scheme structure definition

[0114] Each candidate beam scheme B i contains: beam direction angle θ i , weight vector W i and performance index set m i , wherein m i is expressed as follows:

[0115] m i = [SINR i , RSSI i , CQI i , ΔW i ]

[0116] Wherein: SINR i is the signal-to-noise ratio of the scheme in this direction; RSSI i is the received signal strength; CQI i is the channel quality indicator; ΔW i =||W i -W prev || 2 is the difference between the previous frame scheme weight;

[0117] S7.2, comprehensive performance score mechanism

[0118] The system calculates a weighted score S for each candidate scheme i for ranking comparison:

[0119] S i =α s ·SINR i +β s ·CQI i -γ s ·ΔW i

[0120] Wherein: α s , β s , γ s are performance index weight factors, satisfying α s +β s +γ s =1; the system can dynamically adjust the weight configuration according to different communication scenarios;

[0121] S7.3, ranking and optimal selection strategy

[0122] After scoring all candidate schemes, the system performs ranking operations:

[0123]

[0124] Optional ranking strategies include: weighted score sorting: direct linear sorting; Pareto optimal method: suitable for multi-objective balanced cases, selecting solutions that are not suppressed by other schemes at the same time; threshold pruning method: first eliminate schemes with SINR or CQI below the set threshold, and then sort the remaining set;

[0125] S7.4, output and control interface

[0126] The final optimal beam scheme B opt =(θ opt , W opt ) will be sent to the antenna control module to drive the array to perform beam direction switching operations. This information can also be returned to the vehicle terminal simultaneously to achieve local control closed loop.

[0127] Further, step S8 specifically includes the following steps:

[0128] S8.1, output instruction generation

[0129] Once the system selects the optimal beam scheme B opt = (θ opt , W opt ), a control instruction package containing the direction number and array weight will be immediately constructed, with the following format:

[0130] I beam = {beam_id, W opt , T exec}

[0131] Where: beam_id corresponds to the number in the beam codebook; W opt is the array weight vector required to be loaded for the current frame; T exec is the timestamp for the control instruction to take effect, used for synchronization with other modules;

[0132] The instruction will be issued to the antenna array module, RSU beam controller or vehicle-mounted beam transmitter through the control bus or virtual control channel;

[0133] S8.2, antenna array control response

[0134] After receiving the beam switching control instruction, the antenna array module will perform the following operations:

[0135] Load weight vector: write W opt to the digital precoding unit;

[0136] Adjust beam direction: adjust the phase and amplitude coefficients of each array element to make the beam main lobe point to θ opt ;

[0137] Trigger calibration mechanism: if the direction changes exceed the set angle threshold, trigger a beam calibration process once;

[0138] Control period: the entire beam switching process control delay should be controlled within 1-5 ms to meet the communication continuity requirements of high-speed moving vehicles;

[0139] S8.3, dynamic tracking and synchronization compensation mechanism

[0140] The following mechanisms are introduced:

[0141] Tracking correction: if the vehicle has acceleration or steering trajectory deviation behavior in a short period of time, the system will make fine adjustments based on the current beam direction;

[0142] Time synchronization: synchronize the beam switching frame number with the vehicle terminal through the control channel to ensure the consistency of the switching process;

[0143] Switching confirmation mechanism: the array can inform the control module through a feedback signal after completing the switching, to confirm that the switching is completed.

[0144] S8.4, multi-antenna system coordination

[0145] For a scenario supporting dual-transmission and dual-reception or multi-RSU cooperation, the system supports a parallel beam instruction distribution mechanism, which can simultaneously distribute instructions in different directions to multiple antenna modules.

[0146] The application also discloses a computer device, including a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to realize the steps of the method.

[0147] The application also discloses a computer readable storage medium, which stores a computer program / instruction, and the computer program / instruction is executed by a processor to realize the steps of the method.

[0148] Beneficial effects: compared with the prior art, the application has the following remarkable advantages:

[0149] In the application, firstly, the current position, speed and direction of the vehicle and other basic state parameters are acquired through the vehicle-mounted GNSS and the inertial navigation unit. Then, the future motion trajectory in a short time is predicted by using the Kalman filtering algorithm to obtain the predicted heading angle. According to the prediction result, the system screens a plurality of candidate beam directions from the preset beam codebook, and realizes the dual fusion mechanism of trajectory perception and beam optimization. The method firstly introduces the motion path prediction into the beamforming strategy, so that the beam design has stronger foresight, and the continuity and anti-interference ability of the link can be effectively improved.

[0150] Next, the system receives the channel feedback information of the communication link in real time, including the signal-to-noise ratio (SINR), the received signal strength (RSSI) and the channel quality indicator (CQI) and other indexes. On this basis, a composite multi-objective optimization function is constructed, which takes the maximization of link quality, the minimization of interference power and the minimization of beam switching frequency as the target, and comprehensively balances various performance indexes, which is superior to the traditional single-target beam control algorithm. The system uses an improved recursive least square (RLS) algorithm to adaptively update the beamforming weight, introduces a joint evaluation mechanism of error amplitude and change trend, dynamically adjusts the forgetting factor, and significantly enhances the robustness and adaptability of the system in the high-speed moving or sudden interference scene.

[0151] In terms of beam selection, the application designs an intelligent beam candidate mechanism: the preset codebook is screened according to the prediction result, and the candidate set with the potential optimal performance is preferentially selected, so that the full-codebook exhaustive search is avoided, the system calculation overhead is greatly reduced, and the processing efficiency is improved.

[0152] Finally, the system ranks and selects the optimal scheme from multiple candidate beam schemes, and quickly issues the optimal scheme to the vehicle terminal or road side unit through a low-delay, lightweight control protocol, realizing beam coordination control and real-time switching of the antenna array, with good real-time performance and low-delay control capability. In addition, this method has low dependence on existing communication hardware, good integration and scalability, and can be deployed as an independent module in V2X, C-V2X or 5G NR-V2X system, and can flexibly adapt to different vehicle networking communication architectures and network environments, with broad application prospects. The present application is particularly suitable for 5G / 6G, millimeter wave communication, large-scale MIMO and other communication scenarios, especially in intelligent highway, unmanned driving and vehicle-road cooperative (V2X) integrated systems, with important engineering promotion value and practical application significance. BRIEF DESCRIPTION OF DRAWINGS

[0153] Figure 1 The overall architecture diagram of the vehicle networking dynamic beam optimization system;

[0154] Figure 2 Flow chart of the dynamic beam adaptive optimization method;

[0155] Figure 3 Beam weight optimization process block diagram based on improved RLS algorithm

[0156] Figure 4 Control signaling interaction flow chart of RSU and OBU cooperative scheduling. DETAILED DESCRIPTION

[0157] In order to overcome the problems of large beam switching delay, unpredictable channel state, high system overhead and lack of trajectory awareness in existing vehicle networking communication systems, the present application proposes a dynamic beam adaptive optimization method for vehicle networking, which combines vehicle motion trajectory prediction, real-time channel quality feedback and adaptive adjustment mechanism of beam forming weight, aiming to improve the beam tracking capability and communication link stability in high-speed dynamic scenarios.

[0158] The technical scheme of the present application includes the following steps:

[0159] S1, vehicle state information acquisition: acquiring the current position, speed and direction information of the target vehicle;

[0160] In the implementation process of the present application, the vehicle state information acquisition module is used to collect the current motion state parameters of the on-board unit (OBU), as an important input for subsequent trajectory prediction and beam candidate pre-selection. This module can be deployed in the vehicle communication unit, or can be implemented cooperatively with the road side unit (RSU). Specifically, the following steps are included:

[0161] (1) Position acquisition: The current position of the vehicle can be obtained in real time by a high-precision GNSS module (such as a dual-system positioning terminal supporting Beidou + GPS), and the positioning frequency is recommended to be no less than 10 Hz to meet the continuous trajectory requirement in high-speed mobile scenarios. To improve robustness, an inertial navigation system (INS) can be used for short-time position interpolation and occlusion compensation.

[0162] (2) Speed acquisition: The vehicle speed can be directly read from the vehicle CAN bus or calculated by GNSS continuous position difference. In some implementations, speed information can also be obtained by fusing IMU (accelerometer) and wheel speed sensor to ensure fast response to acceleration / deceleration process.

[0163] (3) Direction acquisition: The driving direction angle can be calculated by the trajectory solution of the GNSS module, or output in real time by the vehicle's built-in magnetometer and gyroscope sensor. This angle information is used to construct the direction vector in the trajectory prediction module, and is an important basis for judging whether the beam main lobe direction is aligned with the vehicle's forward trend.

[0164] (4) Data fusion and reporting: All collected state parameters will be filtered and time-aligned locally, and then sent to the roadside unit or edge computing node through the V2X communication link (such as DSRC or C-V2X) within a certain period. The state data format includes: Position(x, y), Speed(v), Heading(θ), Timestamp(t).

[0165] (5) Abnormal state handling: When the positioning signal is missing or the vehicle speed suddenly changes beyond the set threshold, the system will automatically start the abnormal detection mechanism, suspend the invocation of the beam prediction module, prevent input error from being transmitted to the subsequent module, and ensure the robustness and stability of the system.

[0166] S2, trajectory prediction module based on Kalman filter: After completing the collection of vehicle state information, the system enters the trajectory prediction stage. In this step, the Kalman filter algorithm is introduced to realize the dynamic prediction of the target vehicle's motion trajectory in the short-term future.

[0167] (1) State modeling: The motion state of the vehicle is modeled as a linear dynamic system, and its state vector includes position, speed and heading direction. Set the state variable as:

[0168] X k =[x k ,y k ,v k ,θ k ] T ,

[0169] where: xk y k current position coordinate, v k is velocity, θ k is heading angle.

[0170] (2) State transition model: The transition of vehicle state between adjacent time instants can be modeled as: k = A · X k + ω k

[0171] X k = A · X k + ω k

[0172] where: A state transition matrix, based on vehicle kinematics modeling; ω k process noise, subject to zero-mean Gaussian distribution, with covariance Q;

[0173] A common state transition matrix A (assuming time interval as Δt) is:

[0174]

[0175] (3) Definition and role of covariance matrix P:

[0176] In Kalman filtering, covariance matrix P k represents the degree of uncertainty of the current state estimation of the system, quantifies the range of prediction error, and is used to calculate Kalman gain K k , which determines the fusion ratio of prediction and observation. At the same time, this coefficient determines the enhancement of prediction dependence when the observation data is unstable, and the enhancement of correction effect when the observation is accurate; that is:

[0177]

[0178] (4) Kalman prediction and update process: Kalman filtering contains two core steps: prediction stage and update stage.

[0179] Prediction stage:

[0180]

[0181] P k|k-1 = A · P k-1|k-1 · A T + Q

[0182] The update stage combines the actual observation value Z k :

[0183] K k = P k|k-1 · H T · (H · P k|k-1·H T +R) -1

[0184]

[0185] P k|k =(I-K k ·H)P k|k-1

[0186] Where: H is the observation matrix, which maps the state space to the observation space; R is the observation noise covariance matrix; K is the Kalman gain, which is used to balance the prediction and observation weights.

[0187] (5) Trajectory output on time window design:

[0188] The system finally outputs the predicted position sequence in the future period of time: {(x k+1 ,y k+1 ), (x k+2 ,y k+2 ), …, (x k+n ,y k+n )} time window T pred It can be dynamically adjusted according to the vehicle speed, and is generally set to 1-3 seconds to cover a beam update period.

[0189] S3, candidate beam direction screening: after completing the vehicle future motion trajectory prediction, the system needs to screen a plurality of candidate beam directions with spatial alignment potential in the pre-defined beam codebook according to the predicted direction information, for subsequent channel quality evaluation and weight optimization. For this purpose, the application designs a candidate beam direction screening module, and the specific process includes the following parts:

[0190] (1) Beam codebook definition and formatting: the system pre-constructs a beam direction codebook containing N standard beam directions, each direction corresponding to a set of fixed array beam forming vectors W i , the codebook format is:

[0191]

[0192] Where, θ i represents the center direction of the i-th beam main lobe; the beam codebook can be uniformly and discretely divided, such as 5° or 10° distribution points; the beam width (3dB angle) of each direction can be set by the antenna array structure, for example: 8 elements with a bandwidth of about 30°, 16 elements with a bandwidth of about 15°.

[0193] (2) Trajectory angle extraction and angular domain mapping: extract the vehicle future average driving direction angle from the output of S2 trajectory prediction module, defined as:

[0194]

[0195] where θ k is the predicted heading angle within the prediction time window; further combined with trajectory curvature prediction to determine whether a wide beam or multiple directions should be pre-selected. After mapping the predicted angle to the codebook space angular domain, the system calculates the angle difference between each beam direction and

[0196] (3) Candidate beam screening strategy: set an angle tolerance threshold δ (e.g. 10°-15°), and screen the beam directions that meet the condition to form a candidate set:

[0197]

[0198] Optional extension strategies: ① Bandwidth matching strategy: dynamically adjust δ combined with vehicle speed / prediction uncertainty, expand the tolerance angle in high-speed scenarios, and use narrow angles in low-speed scenarios; ② Bilateral balance strategy: ensure that the candidate beams are distributed symmetrically around the predicted main direction, improving matching robustness; ③ Probability weight initialization: each candidate direction can be initially assigned a probability weight based on matching degree, which is used for subsequent reinforcement learning or multi-arm Bandit beam selection algorithm.

[0199] (4) Multi-resolution beam set design

[0200] To adapt to the prediction accuracy and direction distribution density in different scenarios, the codebook can adopt a multi-resolution structure: ① Low-resolution beam (Low-Res): used for initial detection or rapid scanning, wide bandwidth, and large coverage; ② High-resolution beam (High-Res): used for precise beam locking and tracking, small angle interval, and strong resolution capability. The system can dynamically select the current resolution level according to the real-time trajectory stability index (such as prediction error covariance).

[0201] (5) Collaborative screening mechanism (vehicle-roadside collaborative deployment)

[0202] If the candidate beam screening task is completed by the RSU or edge computing node in collaboration, the vehicle prediction angle information can be sent to the edge platform through a low-overhead control link, and the platform can complete the screening in the codebook and issue the candidate beam set to the vehicle. The screening process in this mode supports: ① Multi-vehicle parallel screening; ② Unified coordination of candidate beam directions to reduce neighbor vehicle interference; ③ Utilize the storage and computing advantages of edge nodes to quickly generate multi-user candidate beam sets.

[0203] (6) Structured candidate beam output

[0204] ​After the completion of the candidate beam direction screening, the system will output a structured candidate beam set result and pass it to the subsequent channel feedback evaluation module and beam weight optimization module. This output not only contains the direction angle information of each candidate beam, but also can be accompanied by its index number in the beam codebook, direction deviation amount, and initial score value, etc. metadata for subsequent beam sorting, optimization and optimization. The candidate beam set usually contains 3 to 7 directions, whose main lobe center angle has a high spatial alignment with the predicted driving direction of the vehicle, so as to ensure the effective establishment and stable maintenance of the subsequent communication link.

[0205] S4, communication link quality feedback module

[0206] After the completion of the candidate beam direction screening, the system enters the communication link feedback phase. The purpose of this phase is to perceive the current channel state in real time to provide necessary environmental information support for subsequent beam forming weight optimization. The present application adopts a multi-dimensional link quality index feedback mechanism to obtain multiple key physical layer parameters from the receiving end to more comprehensively evaluate the communication performance of each candidate beam direction in the current environment.

[0207] Specifically, the received link feedback information includes the following three types of indicators:

[0208] (1) Signal-to-Interference-plus-Noise Ratio (SINR):

[0209] SINR is a core indicator for measuring received signal quality, representing the ratio of useful signal power to the sum of background interference and noise power. Higher SINR usually means higher modulation rate and lower bit error rate, which is suitable for main performance reference in beam direction selection.

[0210] (2) Received Signal Strength Indicator (RSSI):

[0211] RSSI is a physical layer indicator that characterizes the overall received signal strength, without distinguishing between signal and interference components. Although it cannot directly reflect the communication quality, it can be used to roughly judge the beam coverage or link reachability.

[0212] (3) Channel Quality Indicator (CQI):

[0213] CQI is estimated and fed back by the receiving device according to the current channel condition, and is usually used to guide the selection of link adaptive modulation and coding (AMC) parameters, which is a bridge indicator between the transmission layer and the physical layer.

[0214] The above indexes can be measured locally by the vehicle terminal after receiving each candidate beam direction signal, and transmitted back to the beam control module in real time through the uplink control channel, or collected by the roadside unit and calculated in conjunction with the trajectory prediction module.

[0215] To realize multi-direction performance evaluation, the system supports periodic or triggered parallel measurement mechanism, that is, multiple candidate beam directions are evaluated simultaneously in one feedback period, and the corresponding link indexes are packaged and uploaded in a structured data format, providing quantitative support for subsequent weight adjustment and optimal beam selection.

[0216] S5, multi-target beam optimization objective function construction module

[0217] In order to realize the optimal matching of beam forming direction and communication link quality in high-speed dynamic scene, the application constructs a multi-objective optimization objective function based on channel feedback, comprehensively considers multiple aspects such as link performance, interference suppression and system stability, and ensures the accuracy, stability and resource efficiency of beam selection.

[0218] The objective function constructed by the application includes the following three main optimization objectives:

[0219] (1) Maximizing communication quality (maximizing SINR)

[0220] The goal is to improve the effective communication signal quality of the receiving end and enhance the robustness and modulation capability of the channel. Specifically, the average SINR can be used to represent it: (take its negative as one of the minimized objective functions)

[0221]

[0222] Where: W is the current beam forming weight vector; h is the target channel vector; h j The channel vector corresponding to the interference direction; σ 2 System thermal noise power.

[0223] (2) Minimizing interference power (Interference Suppression)

[0224] The goal is to reduce the interference energy leaked by the beam direction to other users or directions. Define the interference power term as:

[0225]

[0226] It can be constrained by constructing a null steering or minimum leakage beam form to ensure beam directivity.

[0227] (3) Minimizing beam switching cost (Stability-aware Smoothness)

[0228] To prevent link instability and system load caused by frequent switching, the application adds a beam change penalty term:

[0229] f3(W t ,W t-1 )=||W t -W t-1 ||

[0230] Where: W t is the current frame beam weight; W t-1 is the last frame beam weight; this term measures the beam change intensity, used to control the energy consumption and delay caused by frequent direction jumps.

[0231] (4) Multi-objective optimization function construction

[0232] The final objective function is constructed in the form of weighted linear combination as:

[0233] J(W)=α·f1(W)+β·f2(W)+γ·f3(W t ,W t-1 )

[0234] Where: α, β, γ ∈ [0, 1] are the weight factors of each target, used to balance the optimization priority between different indicators; each weight can be statically set, or dynamically adjusted according to the scene or channel state, for example, increase γ to control the switching frequency in high-speed environment, and increase β in interference-intensive scenarios.

[0235] (5) Optional target item expansion

[0236] In addition to the above three core optimization indicators, the system can also introduce the following target items: energy minimization term: constraint the total amount of beam output power; QoS equivalent score term: fusion CQI index for quality evaluation; distributed cooperation gain term: in the multi- RSU or multi-vehicle cooperation, add global optimization factor.

[0237] S6, improved RLS beam forming weight adaptive adjustment module

[0238] After completing the multi-objective optimization function construction, in order to realize the fast convergence and adaptive adjustment of the beam forming weight vector corresponding to the optimal beam direction, the application introduces an improved recursive least squares algorithm. Compared with the traditional RLS, this algorithm has stronger robustness in dynamic environment, and is particularly suitable for real-time updating of beam weight in the high-speed changing channel scene of Internet of Vehicles.

[0239] (1) The traditional RLS is used to minimize the following cost function:

[0240]

[0241] Wherein: W is a beam forming weight vector (to be estimated parameter); x(k) is the kth sampling input (received signal vector); d(k) is an expected signal; λ is a forgetting factor, controlling the attenuation degree of historical sample weight (0 < λ ≤ 1).

[0242] The weight updating formula of RLS is:

[0243]

[0244] W(t) = W(t-1) + K(t) * e * (t)

[0245] P(t) = λ -1 [P(t-1) - K(t) * x H (t) * P(t-1)]

[0246] Wherein: P(t) is a covariance matrix, e(t) = d(t) - W H (t-1) * x(t) is a current error.

[0247] (2) Adaptive adjustment of forgetting factor λ:

[0248] In order to improve the real-time adaptive ability of RLS in high-speed changing channel, a compound mechanism based on error size and error trend is proposed to dynamically adjust the forgetting factor, so as to control the balance between learning rate and stability.

[0249] λ(t) = λ min + (λ max - λ min ) * (1 - tanh (α r * |e(t)| + β r * Δe(t))

[0250] Wherein: |e(t)| is the current error amplitude, reflecting the estimation deviation; Δe(t) = e(t) - e(t-1) is the error change trend; α r and β r are adjustment weight factors, used to control the sensitivity of each component; λ min , λ max control the adjustable range of λ, usually set to [0.90, 0.9999]. The mechanism can realize the following two behaviors: if the error is large or the error is increasing, the system automatically reduces λ to speed up the learning of new samples; if the error tends to be stable, the system gradually increases λ to enhance the smoothness and reduce the fluctuation.

[0251] (3) Coupling update strategy with multi-objective function

[0252] In the present invention, the RLS weight update not only converges to a single objective (such as MSE), but also solves the multi-objective function (see S5) in real time. The system takes the error term of the objective function as the equivalent expected error in RLS, dynamically drives the weight iteration process, so that RLS becomes a gradient-guided objective minimization process controller.

[0253]

[0254] (4) Algorithm execution flow

[0255] ① Initialize weight W(0), covariance matrix P(0) = δ -1 I;

[0256] ② For each time t, get input x(t), expectation d(t);

[0257] ③ Calculate error e(t) and change Δe(t);

[0258] ④ Calculate the current forgetting factor λ(t)

[0259] ⑤ Update K(t), W(t), P(t) according to the update formula;

[0260] ⑥ Output the latest weight W(t) to the beam control module.

[0261] S7, candidate beam scheme sorting and optimal selection

[0262] After completing the beamforming weight adaptive optimization based on the improved RLS algorithm, the system usually obtains multiple candidate beam schemes with different performance. The present invention introduces a candidate scheme sorting and optimal selection mechanism based on multi-dimensional performance indicators at this stage, which is used to select the beam direction with the best comprehensive communication effect under the current environment, and issues it to the antenna array control module as the final execution scheme.

[0263] (1) Candidate scheme structure definition

[0264] Each candidate beam scheme B i contains: beam direction angle θ i , weight vector W i and performance indicator set m i , where m i is expressed as follows:

[0265] m i = [SINR i , RSSI i , CQI i , ΔW i ]

[0266] Where: SINRi is the signal-to-noise ratio of the scheme in this direction; RSSI i is the received signal strength; CQI i is the channel quality indicator; AW i = ||W i -W prev | 2 is the difference of the previous frame scheme weight (representing switching cost).

[0267] (2) Comprehensive performance scoring mechanism

[0268] The system calculates the weighted score S for each candidate scheme i for ranking comparison:

[0269] S i = a s · SINR i + b s · CQI i - g s · AW i

[0270] wherein: a s , b s , g s are performance index weight factors, satisfying a s + b s + g s = 1; the system can dynamically adjust the weight configuration according to different communication scenarios (high-speed traffic, complex interference, low delay, etc.).

[0271] (3) Ranking and optimal selection strategy

[0272] After scoring all candidate schemes, the system performs ranking operation:

[0273]

[0274] The optional ranking strategies include: ① Weighted Score Sorting: as shown above, direct linear ranking; ② Pareto Optimal Method: suitable for multi-objective balanced cases, selecting solutions that are not simultaneously suppressed by other schemes; ③ Threshold Pruning Method: first eliminate schemes with SINR or CQI below the set threshold, and then rank in the remaining set.

[0275] (4) Output and control interface

[0276] The final optimal beam scheme B opt = (q opt , W opt) will be sent to the antenna control module to drive the array to perform beam direction switching operation. This information can also be returned to the vehicle terminal synchronously to achieve local control closed loop.

[0277] S8, output the optimal beam direction and control the antenna array to complete beam switching

[0278] After completing the sorting and optimal selection of S7 candidate beam schemes, the system will finally determine the optimal beam direction θ opt and its corresponding beam forming weight vector W opt output, used to control the antenna array to complete beam switching and direction pointing update at the physical layer. The beam execution module proposed in the present application has low delay response capability and software and hardware cooperative scheduling mechanism to adapt to the rapid link switching demand in high-speed vehicle networking scene.

[0279] (1) output instruction generation

[0280] Once the system selects the optimal beam scheme B opt = (θ opt , W opt ), a control instruction package containing direction number and array weight will be immediately constructed, with the following format:

[0281] I beam = {beam_id, W opt , T exec}

[0282] Where: beam_id corresponds to the number in the beam codebook (such as θ = 85° corresponding to the 11th beam); W opt is the array weight vector required to be loaded in the current frame; T exec is the timestamp of the control instruction taking effect, used for synchronization with other modules.

[0283] The instruction will be issued to the antenna array module, RSU beam controller or vehicle-mounted beam transmitter through the control bus or virtual control channel.

[0284] (2) antenna array control response

[0285] After receiving the beam switching control instruction, the antenna array module will perform the following operations:

[0286] ① load weight vector: write W opt into the digital precoding unit;

[0287] ② adjust beam direction: adjust the phase and amplitude coefficients of each array element to make the beam main lobe point to θ opt ;

[0288] ③ trigger calibration mechanism: if the direction changes more than a set angle threshold (such as > 15°), trigger a beam calibration process once;

[0289] ④Control cycle: The entire beam switching process control delay should be controlled within 1-5ms to meet the communication continuity requirements of high-speed moving vehicles.

[0290] (3) Dynamic tracking and synchronization compensation mechanism

[0291] In order to maintain the stability of the beam in multiple consecutive frame transmission, the present application introduces the following mechanism:

[0292] Tracking correction: If the vehicle has trajectory deviation behavior such as acceleration and turning in a short time, the system can make fine adjustment on the basis of the current beam direction;

[0293] Time synchronization: Synchronize the beam switching frame number through the control channel with the vehicle terminal to ensure the consistency of the switching process;

[0294] Switching confirmation mechanism: After the array completes the switching, it can inform the control module through the feedback signal to confirm the completion of the switching.

[0295] (4) Multi-antenna system coordination

[0296] For the scene supporting double-transmitting and double-receiving or multi-RSU cooperation, the system supports parallel beam instruction distribution mechanism, which can simultaneously distribute different direction instructions to multiple antenna modules (such as front, rear or side of the vehicle), realize multi-directional cooperative beam transmission, and improve network coverage and redundancy fault tolerance capability.

[0297] S9, through low delay signaling to realize the cooperative scheduling of vehicle terminal and road side unit

[0298] In order to guarantee the real-time, accuracy and system coordination of the beam switching process, the present application introduces the low delay signaling interaction mechanism between the vehicle terminal (OBU) and the road side unit (RSU) after the beam direction output and control execution, realizes the cooperative scheduling and beam synchronous switching among multiple modules.

[0299] (1) Cooperative scheduling architecture

[0300] The present application adopts distributed / semi-centralized control architecture, which is composed of the following three functional entities:

[0301] ① RSU side beam scheduling controller: responsible for unified prediction result receiving, candidate direction screening, RLS optimization, beam optimization and instruction issuing.

[0302] ② Vehicle OBU execution module: real-time receive beam instruction, complete local array loading and synchronization confirmation, responsible for feedback communication quality index.

[0303] ③Control signaling interface: built on top of C-V2X, DSRC or 5G NR-V2X basic control channel, using compressed encoding to transmit low-bit control instructions.

[0304] (2) Control signaling design and format

[0305] At the control execution level, to achieve low-delay beam switching control for vehicle-road cooperation, the system designs a compact and easy-to-quickly-process control signaling format. The signaling data packet is suggested to include the following key fields in Table 1:

[0306] Table 1

[0307] Field Name Description beam_id Selected beam number exec_time Instruction effective time (frame number or absolute timestamp) W_opt Weight vector (optional field, if non-codebook numerology) validity_flag Flag to indicate whether the current instruction is valid or a retry instruction ACK / NACK Handover execution acknowledgement signal

[0308] (3) Switching synchronization mechanism

[0309] To ensure that beam switching is synchronized on the terminal and RSU side, the invention uses the following two ways: clock synchronization mode: all terminal devices share GPS or edge node clock, and the instruction takes effect based on the "unified frame number"; handshake confirmation mechanism: using a two-way confirmation process, the RSU waits for the OBU to return ACK after issuing an instruction, and then updates the local array.

[0310] (4) Low-delay guarantee measures

[0311] To ensure that signaling exchange is completed within milliseconds, the system uses the following strategies:

[0312] ① Control signaling bandwidth isolation and priority scheduling;

[0313] ② Light protocol stack transmission based on UDP-Lite or C-V2X sidelink;

[0314] ③ Instruction packing and compression, carrying at most 1-3 switching candidates each time to reduce bit load;

[0315] ④ Support "predicted switching" and "quick fallback" mechanisms in poor signal coverage scenarios to enhance robustness.

[0316] (5) Joint state maintenance and policy synchronization

[0317] The system maintains a vehicle beam state table (BeamAssignment Table) in the background, recording the current beam number, last switching time, and last feedback channel quality of each terminal, and implements scheduling memory and adaptive optimization strategy adjustment for the next period.

Claims

1. A dynamic beam adaptive optimization method for Internet of Vehicles, characterized in that, The method comprises the following steps: S1, obtaining the current position, speed and direction information of the target vehicle; S2, based on the Kalman filter algorithm, combining historical trajectory and current state, predicting the future trajectory of the vehicle in a predetermined time period; S3, according to the predicted trajectory direction, selecting a plurality of candidate beam directions from the beam codebook to construct a candidate beam set; S4, real-time receiving communication link feedback information, including signal-to-noise ratio SINR, received signal strength indication RSSI and channel quality indication CQI, and updating at fixed time intervals or event triggered mode; S5, constructing a multi-target beam optimization objective function, the target including maximizing SINR, minimizing interference power and beam switching frequency; S6, based on the improved RLS algorithm, combining channel feedback and candidate beam information, adaptively and dynamically adjusting the beam forming weight, and outputting a plurality of optimized weight solutions; S7, according to the preset beam performance index, evaluating and sorting a plurality of beam schemes; S8, outputting the optimal beam direction and its weight as the beam control scheme; S9, through the low delay signaling interface, controlling the antenna array to complete the real-time switching of the beam direction, and realizing the cooperative scheduling between the vehicle terminal and the road side unit.

2. The dynamic beam adaptive optimization method for vehicle networking according to claim 1, characterized in that, Step S1 specifically comprises the following steps: S1.1, position acquisition: the current position of the vehicle can be obtained in real time by a high-precision GNSS module, and the inertial navigation system INS is used for short-time position interpolation and shielding compensation; S1.2, speed acquisition: the vehicle speed can be obtained from the vehicle CAN bus, or calculated by the difference between continuous positions of GNSS, or obtained by fusing IMU accelerometer and wheel speed sensor; S1.3, driving direction acquisition: the driving direction angle can be calculated by the trajectory solution of the GNSS module, or output in real time by the magnetic sensor and gyroscope sensor of the vehicle; The angle information is used to build a direction vector in the trajectory prediction module, to judge whether the beam main lobe direction is aligned with the vehicle forward trend; S1.4, data fusion and reporting: after all the collected state parameters are filtered and time-aligned, they are sent to the road side unit or edge computing node through the V2X communication link within a certain period; The state data format includes: Position(x,y), Speed(v), Heading(θ), Timestamp(t). S1.5, abnormal state processing: when the positioning signal is missing or the vehicle speed suddenly changes beyond the set threshold, the system will automatically start the abnormal detection mechanism, suspend the call of the beam prediction module, and prevent the input error from being transmitted to the subsequent module.

3. The dynamic beam adaptive optimization method for vehicle networking according to claim 1, characterized in that, Step S2 specifically comprises the following steps: S2.1, state modeling: the motion state of the vehicle is modeled as a linear dynamic system, and the state vector includes position, speed and heading direction; Set the state variable as: X k = [x k , y k , v k , θ k ] T where: x k y k current position coordinates, v k is the velocity, θ k is the heading angle; S2.1, state transition model: the transition of vehicle state between adjacent time can be modeled as: X k = A - X k + ω k where: A is the state transition matrix, based on vehicle kinematic modeling; ω k Process noise, subject to zero-mean Gaussian distribution, covariance is Q; The state transition matrix A is: to describe the change of system state between time k and k+1; The time interval is Δt; S2.1, definition and role of covariance matrix P: In Kalman filtering, the covariance matrix P k represents the degree of uncertainty of the current state estimation of the system, quantifies the range of prediction error, and is used to calculate the Kalman gain K k , which determines the fusion ratio of prediction and observation. At the same time, the coefficient determines the enhancement of prediction dependence when the observation data is unstable, and the enhancement of correction effect when the observation is accurate; that is: wherein X k is the true value of the state of the vehicle, is the predicted value of the state, and E[•] is the expectation between the true and predicted values. S2.4, Kalman prediction and update process: Kalman filtering includes: prediction stage and update stage; Prediction stage: P k|k-1 = A - P k-1|k-1 · A T + Q wherein, a predicted current state at time k, based on data at time k-1; a previous state estimate at time k-1, obtained after updating based on observations at time k-1; P k|k-1 a covariance matrix at time k, indicating a degree of confidence in the current state estimate; P k-1|k-1 a covariance after updating based on observations at time k-1; A T a transpose of the state matrix; The update phase incorporates actual observations Z k : K k = P k|k-1 · H T · (H·P k|k-1 · H T + R) -1 P k|k = (I - K k · H)P k|k-1 where: K k is the Kalman gain, used to weigh the contribution of the predicted value and the observed value to the final estimate; H is the observation matrix, mapping the state space to the observation space; R is the observation noise covariance matrix, reflecting the uncertainty of the observation, if the value is small, the gain is large, and the observation is biased; P k|k-1 is the state covariance matrix in the prediction stage, used to measure the prediction uncertainty, if the value is small, the gain is small, and the prediction is biased; I is the unit matrix; S2.5, trajectory output and time window design: The system finally outputs the predicted position sequence in the future period of time: {(x k+1 ,y k+1 ), (x k+2 ,y k+2 ),..., (x k+n ,y k+n )} Time window T pred Dynamically adjusted according to vehicle speed to cover a beam update period.

4. The dynamic beam adaptive optimization method for vehicle networking according to claim 1, characterized in that, Step S3 specifically includes the following steps: S3.1, Beam codebook definition and formatting: System pre-constructs a beam direction codebook contains N standard beam directions, each direction corresponds to a set of fixed array beamforming vectors W i Codebook format: where θ i represents the i-th beam lobe center direction; the beam codebook can be uniformly and discretely divided, such as being arranged at intervals of 5° or 10°; the beam width of each direction is set by the antenna array structure; S3.2, Trajectory angle extraction and angular domain mapping: Extract the vehicle future average heading angle from the S2 trajectory prediction module output is defined as: where θ k is the predicted heading angle for each time window; in conjunction with trajectory curvature prediction, the trend determines whether a wide beam or multiple directions should be pre-selected, after mapping the predicted angle to the codebook spatial angular domain, the system calculates the angle difference between each beam direction and ​ S3.3, candidate beam screening strategy: set the angle tolerance threshold δ, and screen the beam directions that meet the conditions to form the candidate set: Optional extension strategy: ① Bandwidth matching strategy: dynamically adjust δ combined with vehicle speed / predicted uncertainty, expand the tolerance angle in high-speed scenarios, and use narrow angle in low-speed scenarios; ② Double-side balance strategy: ensure that the candidate beams are distributed symmetrically around the predicted main direction, and improve the matching robustness; ③ Probability weight initialization: each candidate direction can be initially assigned a probability weight based on the matching degree, which is used for subsequent reinforcement learning or multi-arm Bandit beam selection algorithm; S3.4, multi-resolution beam set design To adapt to the prediction accuracy and direction distribution density in different scenarios, the codebook A multi-resolution structure is adopted: coarse-grained beam Low-Res and fine-grained beam High-Res; S3.5, cooperative screening mechanism If the candidate beam screening task is completed by the RSU or edge computing node, send the vehicle prediction angle information to the edge platform through the low-overhead control link, complete the screening in the codebook by the platform, and deliver the candidate beam set to the vehicle; S3.6, structured candidate beam output After completing the candidate beam direction screening, the system will output a structured candidate beam set result, and pass it to the subsequent channel feedback evaluation module and beam weight optimization module.

5. The dynamic beam adaptive optimization method for vehicle networking according to claim 1, characterized in that, Step S5 specifically includes: The constructed objective function includes the following three main optimization objectives: Maximize the communication quality, i.e., maximize the SINR The goal is to improve the effective communication signal quality of the receiving end and enhance the robustness and modulation capability of the channel, represented by the average SINR: where: W is the current beamforming weight vector; htargetis the target channel vector; h j the channel vector corresponding to the interference direction; σ 2 the system thermal noise power; Minimize the interference power Interference Suppression The goal is to reduce the interference energy leaked to other users or directions by the beam direction, and define the interference power term as: It can be constrained by constructing null steering or minimum leakage beam form to ensure beam directivity; Minimize the beam switching cost Stabilization-aware Smoothness To prevent link instability and system load caused by frequent switching, add a beam change penalty term: f3(W t ,W t-1 ) = ||W t -W t-1 || wherein: W t is the current frame beam weight; W t-1 is the last frame beam weight; this term measures the beam change strength, used to control the energy consumption and delay caused by frequent direction hopping; Multi-objective optimization function construction The final objective function is constructed in the form of weighted linear combination: J(W) = a - f1(W) + β - f2(W) + γ - f3(W) t ,W t-1 ) Where: α, β, γ ∈ [0, 1] are the weight factors of each target, used to balance the optimization priority between different indicators; Each weight can be statically set, or dynamically adjusted according to the scene or channel state.

6. The dynamic beam adaptive optimization method for vehicle networking according to claim 1, characterized in that, Step S6 specifically includes the following steps: S6.1, traditional RLS is used to minimize the following cost function: Where: W is the beamforming weight vector; x(k) is the kth sampling input; d(k) is the desired signal; λ is the forgetting factor, which controls the decay degree of historical sample weight (0 < λ ≤ 1); The weight update formula of RLS is: W(t) = W(t - 1) + K(t) - e * (t) P(t) = λ -1 [P(t-1) - K(t)x H (t)P(t-1)] where: P(t) is the covariance matrix, e(t) = d(t) - W H (t-1)x(t) is the current error; S6.2, adaptive adjustment of forgetting factor λ: A compound mechanism based on error size and error trend is proposed to dynamically adjust the forgetting factor, thereby controlling the balance between learning rate and stability; λ(t)=λ min +(λ max -l min )·(1-tanh(α r ·|e(t)|+β r ·D(t))) wherein: |e(t)| is the current error amplitude, reflecting the estimation bias; Δe(t) = e(t) - e(t-1) is the error change trend; α r and β r are the adjustment weight factors, used to control the sensitivity of each component; λ min , λ max controls the adjustable range of λ, which is usually set as [0.90, 0.9999]; S6.3, Coupling update strategy of multi-objective function The system takes the objective function value error term as the equivalent expected error in RLS, dynamically drives the weight iteration process, so that RLS becomes a gradient-guided target minimization process controller; S6.4, Algorithm execution flow Initialize the weight W(0), the covariance matrix P(0) = δ -1 I; For each time t, get the input x(t), the expectation d(t); Calculate the error e(t) and the change amount Δe(t); Calculate the current forgetting factor λ(t); Iterate K(t), W(t), P(t) according to the update formula; Output the latest weight W(t) to the beam control module.

7. The dynamic beam adaptive optimization method for vehicle networking according to claim 1, characterized in that, Step S7 specifically includes the following steps: S7.1, Candidate scheme structure definition Each candidate beam scheme B i Comprises: beam direction angle θ i , weight vector W i And performance index set m i Where m i The expression is as follows: m i = [SINR i , RSSI i , CQI i , ΔW i ] where: SINR i is the signal to noise ratio of the scheme in that direction; RSSI i is the received signal strength; CQI i is the channel quality indicator; AW i = ||W i - W prev || AW 2 is the difference in the previous frame scheme weight. S7.2, Comprehensive performance scoring mechanism The system computes a weighted score S for each candidate solution i for ranking comparison: S i = a s · SINR i + β s · CQI i - γ s · ΔW i Wherein: α s , β s , γ s are performance index weight factors, satisfying α s + β s + γ s = 1; the system can dynamically adjust the weight configuration according to different communication scenarios; S7.3, Ranking and optimal selection strategy After scoring all candidate schemes, the system performs ranking operations: Optional ranking strategies include: weighted scoring method Weighted Score Sorting: direct linear ranking; Pareto optimal method: suitable for multi-objective balancing, select solutions that are not suppressed by other schemes at the same time; threshold pruning method: first remove schemes with SINR or CQI below the set threshold, then sort the remaining set; S7.4, Output and control interface The final selected optimal beam scheme B opt = (θ opt ,W opt ) will be sent to the antenna control module to drive the array to switch the beam direction operation, and this information can also be returned to the vehicle terminal synchronously to achieve local control closed loop.

8. The dynamic beam adaptive optimization method for vehicle networking according to claim 1, characterized in that, Step S8 specifically includes the following steps: S8.1, Output instruction generation Once the system has selected the optimal beam solution B opt = (θ opt , W opt ), the control instruction packet containing the direction number and array weight will be immediately constructed, in the following format: I beam = {beam_id, W opt , T exec} Wherein: beam_id corresponds to the number in the beam codebook; W opt Array weighting vectors required to load the current frame; T exec Timestamp of the time when the control instruction takes effect, used for synchronization with other modules; The instruction will be issued to the antenna array module, RSU beam controller or vehicle-mounted beam transmitter through the control bus or virtual control channel; S8.2, Antenna array control response After receiving the beam switching control instruction, the antenna array module will perform the following operations: Load weight vector: W opt Write digital precoding unit; Adjusting beam direction: by adjusting the phase and amplitude coefficients of each array element, the main lobe of the beam is directed towards θ opt ; Trigger the calibration mechanism: if the direction changes exceed the set angle threshold, trigger a beam calibration process; Control period: the entire beam switching process control delay should be controlled within 1-5ms to meet the communication continuity requirements of high-speed moving vehicles; S8.3, Dynamic tracking and synchronization compensation mechanism The following mechanisms are introduced: Tracking correction: if the vehicle has acceleration or steering trajectory deviation behavior in a short period of time, the system makes fine adjustments based on the current beam direction; Time synchronization: synchronize the beam switching frame number with the vehicle terminal through the control channel to ensure consistency during the switching process; Switching confirmation mechanism: the array can notify the control module of the completion of the switching through a feedback signal after completing the switching; S8.4, Multi-antenna system coordination For scenarios that support dual-transmit dual-receive or multi-RSU cooperation, the system supports parallel beam instruction distribution mechanisms that can simultaneously distribute different direction instructions to multiple antenna modules.

9. A computer apparatus comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program, when executed by the processor, causes the processor to perform the method of any one of claims 1 to 8. The processor executes the computer program to implement the steps of the method of claim 1.

10. A computer readable storage medium having stored thereon computer programs / instructions, characterized in that, The computer program / instructions are executed by the processor to implement the steps of the method of claim 1.

Images