Internet of vehicles vehicle information interaction system based on YTS engine

Abstract

The invention discloses an Internet of Vehicles vehicle information interaction system based on a YTS engine, and relates to the technical field of vehicle information interaction, a data acquisition module fuses multi-source sensing data, and an AI computing power optimization module adopts a dynamic computing power scheduling strategy; the 3D modeling and physical simulation module ensures the real-time performance of modeling based on an AI-driven 3D environment reconstruction and physical simulation technology; the cloud cooperative computing module processes multi-vehicle data in a distributed manner by using a cloud computing architecture of a YTS engine, optimizes a local model and improves the accuracy of an automatic driving decision; and the intelligent interaction module adaptively adjusts a modal man-machine interaction mode according to the dynamic change of the automatic driving decision, so that the driving experience is improved. The technical bottlenecks of information interaction delay, insufficient 3D modeling precision, poor decision stability and the like of the existing vehicle networking system in a high-speed dynamic environment are broken through; efficient energy consumption management, intelligent cloud optimization and accurate interaction are realized, and the safety and reliability of automatic driving are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle information interaction, and particularly relates to a vehicle networking vehicle information interaction system based on the YTS engine. Background Art

[0002] With the rapid development of vehicle networking technology, vehicle information interaction systems are gradually evolving towards intelligence and automation. Existing vehicle networking interaction systems mainly rely on traditional sensor data collection and basic computing models, suffering from problems such as low computing efficiency, large data processing latency, and limited intelligent decision-making capabilities. To address these issues, artificial intelligence (AI) technology has been gradually introduced into vehicle networking systems to optimize data analysis and enhance information interaction capabilities.

[0003] The existing technologies have the following deficiencies:

[0004] During high-speed driving, in complex road scenarios (such as intersections, tunnels, bridges, etc.), and in the process of multi-vehicle collaborative interaction, vehicles need to obtain, render, and calculate external environment information in real time. However, due to the high dynamicity and data complexity of vehicle networking systems, existing visualization systems are difficult to simultaneously meet the requirements of low-latency information interaction, accurate 3D modeling, and physical environment consistency. For example, in an autonomous driving scenario, if the system fails to generate an environment model with physical accuracy in real time, it may lead to misjudgment of information such as obstacles and road boundaries by the vehicle, thereby affecting driving safety. In addition, remote rendering and physical simulation under the traditional cloud computing mode often result in information lag due to data transmission latency, reducing the accuracy of in-vehicle AI decision-making. Summary of the Invention

[0005] The purpose of the present invention is to provide a vehicle networking vehicle information interaction system based on the YTS engine to solve the deficiencies in the background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A vehicle networking vehicle information interaction system based on the YTS engine, including a data collection module, an AI computing power optimization module, a 3D modeling and physical simulation module, a cloud collaborative computing module, and an intelligent interaction module;

[0007] The data collection module is used to collect vehicle sensor data, including camera, radar, lidar, GPS, and V2X communication data, and send the data to the AI computing power optimization module;

[0008] The AI computing power optimization module, based on the AI computing power scheduling mechanism of the YTS engine, dynamically allocates computing resources, and performs fusion calculation and deep learning inference on the collected vehicle sensor data to judge the fluctuation of real-time data processing efficiency;

[0009] 3D Modeling and Physical Simulation Module: An AI-driven 3D modeling component and physical simulation system based on the YTS engine to model the vehicle's surrounding environment, perform dynamic simulation under physical rule constraints, and judge the real-time nature of the modeling results;

[0010] Cloud Collaborative Computing Module: Used to evaluate the accuracy of autonomous driving decisions of the vehicle networking system in a high-speed dynamic environment according to the real-time nature of the modeling results and the fluctuations in real-time data processing efficiency. Based on the evaluation results, utilize the cloud computing architecture of the YTS engine to distributively process real-time data from multiple vehicles and optimize the local model through cloud AI training;

[0011] Intelligent Interaction Module: Used to dynamically adjust the multi-modal human-machine interaction method in the cockpit according to the dynamic changes in the accuracy of autonomous driving decisions within a fixed time period.

[0012] Preferably, in the AI computing power optimization module, after analyzing the fluctuations in energy consumption during the AI inference process, an inference energy consumption fluctuation index is generated. The method for obtaining the inference energy consumption fluctuation index is as follows:

[0013] Set a fixed time window T. During the inference process, sample the energy consumption data E(t) every Δt to form N sampling points: {E0, E1, E2,..., E N-1}; where: E N-1 represents the energy consumption value at the (N - 1)-th sampling. Perform discrete Fourier transform on the collected energy consumption data. The expression is: where: X k represents the complex value of the k-th frequency component, e -j2πkn / N is the kernel function of the discrete Fourier transform. Calculate the power spectral density. The expression is: P k =|X k | 2 ; where: P k represents the energy intensity of the k-th frequency component. Calculate the total energy P total , and the expression is: And calculate the high-frequency energy P high , and the expression is: where, set the threshold frequency f thres =0.1Hz. Calculate the inference energy consumption fluctuation index ECFI. The expression is:

[0014] Preferably, in the 3D modeling and physical simulation module, after analyzing the complete time consumption from sensor data input to result output, a modeling time consumption anomaly index is generated. The method for obtaining the modeling time consumption anomaly index is as follows:

[0015] Define a time window \(W\) and collect the time consumption of each modeling task: \(W=\{W_1, W_2, \cdots, W_M\}\); where: \(W_M\) represents the time consumption of the \(M\)th modeling task, and \(M\) is the total number of sampled modeling tasks; use the elbow method to determine the optimal number of clusters \(K\), calculate the within-cluster sum of squared errors \(WCSS\) for different values of \(K\), and the expression is: M}; where: M represents the time consumption of the \(M\)th modeling task, and \(M\) is the total number of sampled modeling tasks; use the elbow method to determine the optimal number of clusters \(K\), calculate the within-cluster sum of squared errors \(WCSS\) for different values of \(K\), and the expression is: where: \(C_k\) k is the set of samples in the \(k\)th cluster, and \(\mu_k\) k is the central mean of the \(k\)th cluster; select the value of \(K\) at the inflection point, randomly initialize \(K\) cluster centers \(\mu_1, \mu_2, \cdots, \mu_K\), calculate the Euclidean distance from each sample to the cluster center, and assign it to the nearest cluster, calculate the new cluster center \(\mu_k'\), and the expression is: K ,calculate the Euclidean distance from each sample to the cluster center, and assign it to the nearest cluster, calculate the new cluster center \(\mu_k'\), and the expression is: k ,the expression is: Divide the task categories according to the clustering results: the low time consumption group represents normal tasks, the medium time consumption group represents normal tasks, and the high time consumption group represents abnormal tasks; calculate the modeling time consumption anomaly index \(MTCAI\), and the expression is: where: is the average time consumption of abnormal modeling tasks, is the average time consumption of normal modeling tasks.

[0016] Preferably, in the cloud collaborative computing module, normalize the inference energy consumption fluctuation index and the modeling time consumption anomaly index so that they are both within \([0, 1]\), and calculate the accuracy value of the vehicle networking system's autonomous driving decision in a high-speed dynamic environment according to the normalized inference energy consumption fluctuation index and the modeling time consumption anomaly index.

[0017] Preferably, compare the obtained accuracy value of the vehicle networking system's autonomous driving decision in a high-speed dynamic environment with a pre-set accuracy threshold. If the accuracy value of the vehicle networking system's autonomous driving decision in a high-speed dynamic environment is greater than or equal to the pre-set accuracy threshold, it means that the accuracy of the vehicle networking system's autonomous driving decision in a high-speed dynamic environment is high, and no warning signal is generated at this time; if the accuracy value of the vehicle networking system's autonomous driving decision in a high-speed dynamic environment is less than the pre-set accuracy threshold, it means that the accuracy of the vehicle networking system's autonomous driving decision in a high-speed dynamic environment is low, and a warning signal is generated at this time. Use the cloud computing architecture of the YTS engine to distributively process the real-time data from multiple vehicles and optimize the local model through cloud AI training.

[0018] Preferably, in the intelligent interaction module, calculate the autonomous driving decision status index \(S\) AD , and the formula is: \(S\) AD =\(\gamma_1\cdot A\) AD -\(\gamma_2\cdot(ECFI\)norm +MTCAI norm ); where: ECFI norm , MTCAI norm is the normalized inference energy consumption fluctuation index, γ1 and γ2 are weight coefficients, satisfying γ1 + γ2 = 1, and A AD is the accuracy value of the vehicle networking system's autonomous driving decision-making in a high-speed dynamic environment;

[0019] Based on the calculated S AD , the system dynamically adjusts the human-computer interaction mode and classifies the states:

[0020] Full autonomous driving mode, that is, S AD > 0.85: Enhance voice interaction, HUD, and AR displays;

[0021] Semi-autonomous driving mode, that is, 0.5 ≤ S AD ≤ 0.85: Reduce interaction information and retain key reminders for the driver;

[0022] Manual driving mode, that is, S AD < 0.5: Provide a warning signal to remind the driver to take over the vehicle.

[0023] Preferably, if S AD continuously drops below 0.5, then trigger the YTS engine cloud computing optimization to optimize the inference task and improve the stability of autonomous driving.

[0024] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0025] 1. Through the five major modules of data collection, AI computing power optimization, 3D modeling and physical simulation, cloud collaborative computing, and intelligent interaction, the present invention solves the problem that the existing vehicle networking system is difficult to balance low-latency information interaction, precise 3D modeling, and physical environment consistency in a high-speed dynamic environment. The innovation points of the present invention include AI computing power scheduling optimization, multi-sensor fusion computing, high-precision 3D modeling and simulation, cloud distributed computing, and intelligent interaction dynamic adjustment. By introducing the inference energy consumption fluctuation index (ECFI) and the modeling time consumption anomaly index (MTCAI), this system can accurately evaluate the stability of autonomous driving calculations, dynamically optimize the computing power allocation, and improve the reliability of environmental perception and decision-making.

[0026] 2. The cloud collaborative computing module of the present invention ensures that tasks can be dynamically offloaded to the cloud when the computing power pressure is too high or the calculation is abnormal, improving the computing efficiency. In addition, the intelligent interaction module combines the autonomous driving decision-making status indicators to adaptively adjust multi-modal interaction methods such as voice interaction, HUD, and AR, providing a more intuitive and intelligent driving experience. This system not only improves the safety and stability of autonomous driving, but also reduces energy consumption and improves the utilization rate of computing resources, providing strong support for the application of vehicle networking technology in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0028] Figure 1 It is a system module diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0030] Embodiment, please refer to Figure 1 As shown, a vehicle networking vehicle information interaction system based on the YTS engine in this embodiment includes a data acquisition module, an AI computing power optimization module, a 3D modeling and physical simulation module, a cloud collaborative computing module, and an intelligent interaction module;

[0031] The data acquisition module is used to collect vehicle sensor data, including camera, radar, lidar, GPS, and V2X communication data, and send the data to the AI computing power optimization module;

[0032] The AI computing power optimization module, based on the AI computing power scheduling mechanism of the YTS engine, dynamically allocates computing resources, and performs fusion calculation and deep learning inference on the collected vehicle sensor data to judge the fluctuation of the real-time data processing efficiency;

[0033] The 3D modeling and physical simulation module, based on the AI-driven 3D modeling component and physical simulation system of the YTS engine, models the vehicle surrounding environment, performs dynamic simulation under the constraint of physical rules, and judges the real-time performance of the modeling result;

[0034] A cloud collaboration computing module, which is used to evaluate the accuracy of autonomous driving decisions of the vehicle networking system in a high-speed dynamic environment according to the real-time nature of the modeling results and the fluctuations in the real-time data processing efficiency. According to the evaluation results, it uses the cloud computing architecture of the YTS engine to distributively process the real-time data from multiple vehicles and optimize the local model through cloud AI training;

[0035] An intelligent interaction module, which is used to dynamically adjust the multi-modal human-computer interaction method in the cockpit according to the dynamic changes in the accuracy of autonomous driving decisions within a fixed time period.

[0036] The specific steps of the data acquisition module include:

[0037] Step 1: Initialize the sensor system:

[0038] 1.1. After the vehicle starts, the data acquisition module is activated to check the working status of each sensor (camera, radar, lidar, GPS, V2X communication module).

[0039] 1.2. Conduct a self-check of the sensors to ensure that each sensor is operating normally. If any abnormalities are found, trigger a system warning or a backup plan.

[0040] 1.3. Adjust the data acquisition mode and sampling frequency according to the current driving mode (such as manual driving, assisted driving, autonomous driving).

[0041] Step 2: Camera data acquisition:

[0042] 2.1. Multiple cameras installed on the front, rear, left, and right of the vehicle start to collect road video streams and image data in real time.

[0043] 2.2. Conduct preliminary image processing through computer vision algorithms, including denoising, color enhancement, edge detection, etc.

[0044] 2.3. Encode the extracted image features (such as lane markings, pedestrians, traffic signs, etc.) to reduce data redundancy and improve transmission efficiency.

[0045] 2.4. The processed data is transmitted to the AI computing power optimization module through the bus interface.

[0046] Step 3: Radar data acquisition:

[0047] 3.1. Activate the on-vehicle millimeter-wave radar to start detecting the distance, speed, and relative angle information of obstacles in front and around.

[0048] 3.2. Combine the analysis of the echo signals to track moving targets and predict their movement trajectories.

[0049] 3.3. Filter out environmental noise (such as interference from rain and snow), and extract target information with high confidence.

[0050] 3.4. Package the structured data and transmit it to the AI computing power optimization module.

[0051] Step 4: LiDAR data acquisition:

[0052] 4.1. The LiDAR starts a 360° scan of the surrounding environment to generate point cloud data.

[0053] 4.2. Through preprocessing of the point cloud data, remove low-confidence points (such as reflection interference, duplicate points).

[0054] 4.3. Perform point cloud clustering to preliminarily classify categories such as vehicles, pedestrians, and obstacles, and improve the calculation efficiency.

[0055] 4.4. Combine historical point cloud data to perform time series analysis on the scene and predict the movement trends of dynamic targets.

[0056] 4.5. Encode the processed point cloud data and transmit it to the AI computing power optimization module.

[0057] Step 5: GPS data acquisition

[0058] 5.1. The in-vehicle GPS module obtains information such as the longitude, latitude, altitude, speed, and heading of the current vehicle.

[0059] 5.2. Combine with the inertial navigation system (INS) for data fusion to improve the stability and accuracy of the position information.

[0060] 5.3. Perform differential GPS (RTK) correction to ensure that the vehicle positioning error is within centimeters.

[0061] 5.4. Encode the GPS data and transmit it to the AI computing power optimization module.

[0062] Step 6: V2X communication data acquisition:

[0063] 6.1. The vehicle activates the V2X (Vehicle-to-Everything) communication module to establish data connections with other vehicles (V2V), infrastructure (V2I), pedestrians (V2P), and the cloud (V2N).

[0064] 6.2. Receive the driving states (position, speed, acceleration, intention prediction, etc.) of surrounding vehicles.

[0065] 6.3. Obtain road infrastructure information such as traffic lights, road closures, and sudden accidents to improve the safety of driving decisions.

[0066] 6.4. Synchronize V2X communication data to the in-vehicle system through protocols such as 5G / DSRC and transmit it to the AI computing power optimization module.

[0067] Start the AI computing power optimization module and load the AI computing power scheduling mechanism of the YTS engine. Detect the current hardware resources (CPU, GPU, NPU, etc.) and available computing power, and establish a computing power status monitoring model. Set a dynamic computing power allocation strategy according to system load, network latency, and task priority.

[0068] Receive camera, radar, lidar, GPS, and V2X data transmitted by the data acquisition module. Perform timestamp synchronization to ensure that all data is aligned under the same time reference. Use data fusion algorithms such as Kalman filtering and Bayesian inference to jointly calculate the data from different sensors to improve the stability and accuracy of environmental perception. Analyze data trends through deep learning models (such as Transformer, LSTM, etc.) and extract key features, such as vehicle motion state, obstacle trajectory, traffic flow changes, etc.

[0069] Load pre-trained AI models (such as object detection, path prediction, traffic behavior analysis, etc.). Under the computing power scheduling of the YTS engine, dynamically allocate tasks to different computing units (local terminal, edge computing node, cloud). Adopt model compression and inference acceleration technologies (such as TensorRT, ONNX optimization) to improve inference speed and reduce computing latency. Combine reinforcement learning algorithms to optimize the decision-making ability of the AI model in different driving scenarios and improve system adaptability.

[0070] Monitor indicators such as computing load, processing latency, and energy consumption during the AI inference process. Evaluate the fluctuations in real-time data processing efficiency to determine whether there are computing power bottlenecks or resource waste. Dynamically adjust the computing power scheduling strategy according to changes in computing load, such as reducing low-priority tasks and increasing edge computing to share the pressure. Adopt an adaptive computing power optimization algorithm to automatically adjust the complexity of the inference model and balance computing power usage under different working conditions to ensure the stability and efficiency of data processing.

[0071] Package the final results of fusion calculation and AI inference to form structured information (such as object detection results, environmental modeling data, path planning suggestions, etc.). According to task requirements, select data storage, upload to the cloud, or transmit to the 3D modeling and physical simulation module. Record the key parameters during the AI computing power optimization process to provide data support for subsequent system optimization and model update.

[0072] After analyzing the fluctuations in energy consumption during the AI inference process, generate an inference energy consumption fluctuation index. The method for obtaining the inference energy consumption fluctuation index is as follows:

[0073] Set a fixed time window T (e.g. 60 seconds), and during the inference process, sample the energy consumption data E(t) every Δt to form N sampling points: {E0, E1, E2, ..., E N-1}; where: E N-1 Represents the energy consumption value at the N-1th sampling time (unit: W). The sampling interval Δt should ensure that the Nyquist sampling theorem is satisfied (10ms to 100ms is recommended). The total number of samples N = T / Δt (for example, if Δt = 0.1s, then N = 600).

[0074] Perform discrete Fourier transform (DFT) on the collected energy consumption data, and the expression is: Where: X k represents the complex value of the kth frequency component, e -j2πkn / N is the kernel function of discrete Fourier transform, and the power spectral density (PSD) is calculated. The expression is: P k =|X k | 2 ; Among them: P k Represents the energy intensity of the kth frequency component and calculates the total energy P total , the expression is: Low-frequency components (0-0.1 Hz) represent slow trend fluctuations and are not included in the volatility index. Calculate high-frequency energy P high (the part with frequency higher than 0.1Hz), the expression is: Among them, the threshold frequency f is set thres =0.1Hz, exclude the low-frequency change trend, and only calculate the high-frequency fluctuation energy. Calculate the inference energy consumption fluctuation index ECFI, the expression is: A higher ECFI indicates more severe high-frequency fluctuations and more unstable inference energy consumption. When ECFI is below 10%, inference energy consumption is relatively stable; when it exceeds 30%, computing power allocation may be abnormal.

[0075] The 3D modeling and physical simulation module specifically collects raw environmental perception data from cameras, radar, LiDAR, GPS, and V2X, ensuring data timestamp alignment and time synchronization to avoid lags or mismatches in multi-sensor information.

[0076] Convert different sensor data formats into a unified three-dimensional point cloud coordinate system, such as LiDAR point cloud (.pcd / .ply), image depth map (.png / .exr). Perform data denoising (removing mismeasured points, reflection noise), and data completion (using AI to predict missing point clouds). Adopt geographical coordinate transformation (WGS-84 → UTM → vehicle local coordinate system) to ensure the unity of GPS, LiDAR, and camera data. Align the high-precision map (HD Map), calibrate the environmental modeling coordinates, and improve the accuracy of the static environment.

[0077] Use AI models (such as YOLO, PointNet++, BEVFormer) to identify targets such as vehicles, pedestrians, road signs, and obstacles. Combine multi-view fusion to ensure the consistency of detection results and improve the accuracy of target recognition.

[0078] Adopt VoxelNet, SuperPoint+SuperGlue for point cloud feature extraction to generate a sparse point cloud structure. Combine the Marching Cubes algorithm to convert the point cloud into a continuous 3D mesh model (Mesh).

[0079] Use Transformer-based 3D segmentation models (such as RangeNet++, Cylinder3D) to classify the point cloud: road areas (lanes, non-motor vehicle lanes, sidewalks), passable areas (dynamic obstacles, parking spaces), non-passable areas (guardrails, buildings, pedestrians, other obstacles), to form a high-precision 3D semantic map for autonomous driving perception and planning.

[0080] Based on the YTS engine physical simulation system, load the environmental modeling results to construct an interactive 3D physical scene. Select a suitable physical engine (such as PhysX, Bullet, ODE), and set the dynamic parameters: gravity (9.81m / s 2 ), friction coefficient (tire-road surface contact), collision detection (polygon approximation, BVH acceleration structure).

[0081] Perform real-time physical simulation to predict the future motion trajectories of vehicles, pedestrians, and dynamic objects. Adopt rigid body dynamics to calculate factors such as inertia, collision, and road adhesion during vehicle driving. Combine computational fluid dynamics (CFD) to simulate the impact of environments such as rain and water accumulation on vehicle handling.

[0082] Adopt algorithms such as AABB (axis-aligned bounding box), GJK (Gilbert-Johnson-Keerthi) for fast collision detection. Calculate the collision point, impact force, and reaction force, and feedback them to the AI decision-making system to avoid autonomous driving accidents.

[0083] Record the complete time consumption from sensor data input → 3D modeling → simulation calculation → result output. After analyzing the complete time consumption from sensor data input to result output, generate a modeling time consumption anomaly index. The method for obtaining the modeling time consumption anomaly index is as follows:

[0084] Define a time window W (such as the past 1000 modeling tasks), and collect the time consumption of each modeling task (unit: ms): W = {W1, W2,..., W M}; where: W M represents the time consumption of the Mth modeling task (unit: milliseconds ms). M is the total number of sampled modeling tasks (such as M = 1000).

[0085] Use the elbow method to determine the optimal number of clusters K, and calculate the within-cluster sum of squared errors WCSS for different K values. The expression is: where: C k is the set of samples in the kth cluster, and μ k is the central mean of the kth cluster; select the K value at the inflection point. Usually, K = 3 or K = 4 is the best.

[0086] Randomly initialize K cluster centers μ1, μ2,..., μ K , calculate the Euclidean distance from each sample to the cluster center, and assign it to the nearest cluster. Calculate the new cluster center μ k . The expression is: Iteratively update until the center is stable or the maximum number of iterations (such as 100 times) is reached.

[0087] Divide the task categories according to the clustering results. Assume K = 3, and the clustering results may be: low time consumption group (normal tasks), medium time consumption group (normal tasks), high time consumption group (abnormal tasks);

[0088] Calculate the modeling time consumption anomaly index MTCAI. The expression is: where: is the average time consumption of abnormal modeling tasks (unit: ms), is the average time consumption of normal modeling tasks (unit: ms). The higher the MTCAI, the greater the modeling time fluctuation, and the need to optimize the computing power allocation and computing resource scheduling.

[0089] A cloud collaborative computing module is used to evaluate the accuracy of autonomous driving decisions of the vehicle networking system in a high-speed dynamic environment according to the real-time nature of the modeling results and the fluctuations in real-time data processing efficiency. According to the evaluation results, use the cloud computing architecture of the YTS engine to distributively process the real-time data from multiple vehicles and optimize the local model through cloud AI training.

[0090] Normalize the inference energy consumption fluctuation index and the modeling time consumption anomaly index so that they are both within [0, 1]. Calculate the accuracy value of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment based on the normalized inference energy consumption fluctuation index and the modeling time consumption anomaly index.

[0091] For example, the present invention can use the following formula to calculate the accuracy value of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment. The calculation expression is: In the formula, A AD is the accuracy value of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment, ECFI is the inference energy consumption fluctuation index, MTCAI is the modeling time consumption anomaly index, a1 and a2 are the weight coefficients of the inference energy consumption fluctuation index and the modeling time consumption anomaly index (which can be optimized according to experimental experience or machine learning), and both a1 and a2 are greater than 0.

[0092] Compare the obtained accuracy value of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment with a pre-set accuracy threshold. If the accuracy value of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment is greater than or equal to the pre-set accuracy threshold, it indicates that the accuracy of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment is high, and no warning signal is generated at this time; if the accuracy value of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment is less than the pre-set accuracy threshold, it indicates that the accuracy of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment is low, and a warning signal is generated at this time. Utilize the cloud computing architecture of the YTS engine to distributively process the real-time data from multiple vehicles and optimize the local model through cloud AI training.

[0093] Allocate computing tasks among multiple vehicles using cloud computing resources, reduce the local computing power load, optimize the inference and modeling processes, and improve the stability of autonomous driving decisions.

[0094] Hierarchical calculation tasks: Local end (Edge): Process emergency inference tasks (such as pedestrian detection, collision prediction). Cloud: Process high-computation tasks (such as 3D environment reconstruction, deep learning model training). Roadside edge computing (MEC, Multi-access Edge Computing): Conduct regional-level vehicle data aggregation and intelligent optimization to reduce cloud data transmission latency.

[0095] Monitor the local computing power pressure and network bandwidth to determine whether some tasks need to be offloaded to the cloud. Offloading strategy: If A AD is lower than 0.5, transfer large-scale inference tasks (such as semantic segmentation, trajectory prediction) to the cloud for processing. If A ADBelow 0.3, all 3D modeling tasks are transferred to the cloud, and MEC accelerated computing is preferentially adopted.

[0096] Through cloud AI training, local inference tasks are optimized, reducing computing resource occupancy and improving energy consumption efficiency. Methods such as model pruning and knowledge distillation are used to optimize the local AI inference model and reduce computing overhead. If the vehicle AI inference model is too large and occupies too much computing power, 30% of redundant neurons are pruned to reduce the GPU computing load. Knowledge distillation is adopted to convert complex cloud models into lightweight local models to improve inference speed.

[0097] Adopt policy optimization based on reinforcement learning (RL): Simulate different driving scenarios in the cloud to optimize the autonomous driving strategy and adjust the vehicle's control parameters (such as acceleration and brake response time). The optimized strategy is sent to the local terminal via OTA (Over-the-Air) to improve decision-making accuracy.

[0098] The goal of the intelligent interaction module is to adjust the multi-modal human-machine interaction method in the cockpit according to the dynamic change of the autonomous driving decision accuracy A AD to provide the best driving experience.

[0099] This module evaluates the system state by calculating the inference energy consumption fluctuation index ECFI and the modeling time consumption anomaly index MTCAI, and dynamically adjusts the human-machine interaction method, such as voice interaction, touch interface, HUD (head-up display), augmented reality (AR), etc.

[0100] The core of intelligent interaction adjustment is the autonomous driving decision state index S AD , which is calculated based on the inference energy consumption fluctuation ECFI and the modeling time anomaly index MTCAI. The formula is: S AD =γ1·A AD -γ2·(ECFI norm +MTCAI norm ); where: ECFI norm , MTCAI norm are the normalized inference energy consumption fluctuation indices, and γ1 and γ2 are weight coefficients satisfying γ1 + γ2 = 1, which can be optimized according to experiments: If the interaction system is sensitive to decision accuracy, γ1 = 0.7 and γ2 = 0.3 can be set; if the interaction system is sensitive to system load fluctuations, γ1 = 0.5 and γ2 = 0.5 can be set.

[0101] According to the calculated S AD , the system dynamically adjusts the human-machine interaction mode to ensure that the driver or passenger obtains the best interaction experience and conducts state classification:

[0102] Status 1: S AD > 0.85, full autonomous driving mode (enhanced interaction) is required; Voice interaction: The cockpit provides more detailed autonomous driving status information, such as road condition analysis and navigation instructions. AR - HUD (Augmented Reality Head - Up Display): Displays information such as the real - time driving path and autonomous driving prediction trajectory. Simplified touch interface: Reduces the need for driver intervention and only provides high - priority interaction options.

[0103] Status 2: 0.5 ≤ S AD ≤ 0.8, semi - autonomous driving mode (standard interaction) is required. Voice interaction: Reminds only when necessary, such as lane change and obstacles ahead. Reduced HUD display, only basic navigation information is retained to reduce driver information load. Driver Monitoring System (DMS): Enhances the monitoring of driver attention and provides additional reminders if attention drops.

[0104] Status 3: S AD <0.5, manual driving mode (safety - enhanced interaction) is required. Warning signal: Reminds the driver to take over the vehicle through sound / vibration. Reduces the AR - HUD display information and only retains key driving information to reduce interference. Increases driver manual control prompts, such as steering wheel vibration feedback.

[0105] If S AD continues to be below 0.5, the system triggers the YTS engine cloud computing optimization to optimize the inference task and improve the stability of autonomous driving.

[0106] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0107] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0108] As described above, the specific implementation manners of the present application are only provided, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application.

Claims

1. A vehicle information interaction system for the Internet of Vehicles based on the YTS engine, characterized in that: It includes a data acquisition module, an AI computing power optimization module, a 3D modeling and physical simulation module, a cloud collaborative computing module, and an intelligent interaction module; The data acquisition module is used to collect vehicle sensor data, including camera, radar, lidar, GPS, and V2X communication data, and send the data to the AI computing power optimization module; The AI computing power optimization module, based on the AI computing power scheduling mechanism of the YTS engine, dynamically allocates computing resources, and performs fusion computing and deep learning inference on the collected vehicle sensor data to judge the fluctuation of the real-time data processing efficiency; The 3D modeling and physical simulation module, based on the AI-driven 3D modeling component and physical simulation system of the YTS engine, models the vehicle surrounding environment, performs dynamic simulation under the constraint of physical rules, and judges the real-time performance of the modeling result; The cloud collaborative computing module is used to evaluate the accuracy of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment according to the real-time performance of the modeling result and the fluctuation of the real-time data processing efficiency. According to the evaluation result, using the cloud computing architecture of the YTS engine, it distributes and processes the real-time data from multiple vehicles, and optimizes the local model through cloud AI training; The intelligent interaction module is used to dynamically adjust the multi-modal human-computer interaction method in the cockpit according to the dynamic change of the accuracy of the autonomous driving decision within a fixed time period.

2. The vehicle information interaction system for the Internet of Vehicles based on the YTS engine according to claim 1, wherein: In the AI computing power optimization module, after analyzing the fluctuation of the energy consumption during the AI inference process, an inference energy consumption fluctuation index is generated. The acquisition method of the inference energy consumption fluctuation index is: Set a fixed time window T, and during the inference process, sample the energy consumption data E(t) every Δt to form N sampling points: {E0, E1, E2, ..., E N-1 }; where: E N-1 Represents the energy consumption value at the N-1th sampling time. The collected energy consumption data is subjected to discrete Fourier transform, and the expression is: Where: X k represents the complex value of the kth frequency component, e -j2πkn / N is the kernel function of discrete Fourier transform, and the power spectrum density is calculated. The expression is: P k =|X k | 2 ; Among them: P k Represents the energy intensity of the kth frequency component and calculates the total energy P total , the expression is: And calculate the high frequency energy P high , the expression is: Among them, the threshold frequency f is set thres =0.1Hz, calculate the inference energy consumption fluctuation index ECFI, the expression is:

3. The vehicle information interaction system based on the YTS engine according to claim 2, wherein: In the 3D modeling and physical simulation module, after analyzing the complete time consumption from sensor data input to result output, a modeling time consumption anomaly index is generated. The acquisition method of the modeling time consumption anomaly index is: Define a time window W and collect the time consumption of each modeling task: W = {W1, W2,..., W M}; where: W M represents the time consumption of the Mth modeling task, and M is the total number of sampled modeling tasks; Use the elbow method to determine the optimal number of clusters K, and calculate the within-cluster sum of squared errors WCSS for different values of K. The expression is: where: C k is the set of samples in the kth cluster, and μ k is the central mean of the kth cluster; Select the value of K at the inflection point, randomly initialize K cluster centers μ1, μ2,..., μ K , calculate the Euclidean distance from each sample to the cluster center, and assign it to the nearest cluster. Calculate the new cluster center μ k , and the expression is: Divide the task categories according to the clustering results: The low time consumption group represents normal tasks, the medium time consumption group represents normal tasks, and the high time consumption group represents abnormal tasks; Calculate the modeling time consumption anomaly index MTCAI, and the expression is: where: is the average time consumption of abnormal modeling tasks, is the average time consumption of normal modeling tasks.

4. The vehicle information interaction system based on the YTS engine according to claim 3, wherein: In the cloud collaborative computing module, the inference energy consumption fluctuation index and the modeling time consumption anomaly index are normalized so that they are both within [0,1]. According to the normalized inference energy consumption fluctuation index and modeling time consumption anomaly index, the accuracy value of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment is calculated.

5. The vehicle information interaction system based on the YTS engine according to claim 4, characterized in that: Compare the obtained accuracy value of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment with the pre-set accuracy threshold. If the accuracy value of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment is greater than or equal to the pre-set accuracy threshold, it indicates that the accuracy of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment is high, and no warning signal is generated at this time; if the accuracy value of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment is less than the pre-set accuracy threshold, it indicates that the accuracy of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment is low, and a warning signal is generated at this time. Using the cloud computing architecture of the YTS engine, it distributes and processes the real-time data from multiple vehicles, and optimizes the local model through cloud AI training.

6. The vehicle information interaction system based on the YTS engine according to claim 5, characterized in that: In the intelligent interaction module, calculate the autonomous driving decision state index S AD , and the formula is: S AD =γ1·A AD -γ2·(ECFI norm +MTCAI norm ); where: ECFI norm , MTCAI norm are the normalized inference energy consumption fluctuation indexes, γ1 and γ2 are weight coefficients, satisfying γ1 + γ2 = 1, and A AD is the accuracy value of the autonomous driving decision of the vehicle networking system in a high-speed dynamic environment; According to the calculated S AD , the system dynamically adjusts the human-computer interaction mode and classifies the states: Full self-driving mode, i.e., S AD > 0.85: Enhanced voice interaction, HUD, and AR display; Semi-automatic driving mode, i.e., 0.5 ≤ S AD ≤ 0.85: Reduce interaction information and retain key driver alerts; Manual driving mode, i.e., S AD <0.5: Provide a warning signal to remind the driver to take over the vehicle.

7. An in-vehicle network vehicle information interaction system based on the YTS engine according to claim 6, characterized in that: If S AD continues to be below 0.5, the YTS engine cloud computing optimization is triggered to optimize the inference task and improve the stability of autonomous driving.

Images