Vehicle driving simulation test method and system based on digital twinning

Abstract

The invention discloses a vehicle driving simulation test method and system based on digital twinning, belongs to the technical field of vehicle driving, and realizes real interaction of a driving behavior model and a control algorithm in a virtual scene by constructing a digital twinning body highly corresponding to an actual vehicle and a driving environment. Time stamps and coordinate information of all the modules are collected, time differences and space residual errors are extracted, a space-time semantic consistency index model is constructed, and the loading process is intelligently judged and optimized; when inconsistency is detected, adjustment is performed through clock synchronization and coordinate recalibration, a closed-loop optimization mechanism is formed, and high collaboration of the model and the environment is ensured, so that the accuracy, the stability and the credibility of the simulation system are improved, and the method is suitable for complex and changeable automatic driving test scenes.

Description

Technical Field

[0001] The present invention relates to the field of vehicle driving technology, and in particular to a vehicle driving simulation test method and system based on digital twins. Background Art

[0002] Digital twin-based vehicle driving simulation testing utilizes digital twin technology to construct a highly consistent digital model of the actual vehicle and its driving scenarios in a virtual environment. The simulation system then simulates the vehicle's behavior under various road, weather, and traffic conditions. This approach can proactively identify design flaws, verify control algorithms, and optimize driving strategies without compromising actual vehicle safety, thereby accelerating R&D efforts, reducing testing costs, and enhancing the intelligence and reliability of vehicle systems.

[0003] The existing technology has the following shortcomings:

[0004] When digital twin-based vehicle driving simulation tests are performed, spatiotemporal semantic inconsistencies can arise when driving behavior models and control algorithms are loaded into the simulation platform and bound to the virtual vehicle and traffic scenario. These issues manifest themselves in time delays in perception data, asynchrony between decision execution, and coordinate system conversion errors. These issues can cause the model to misjudge traffic conditions, trigger incorrect strategies, or even lead to uncontrolled behavior. This can lead to disrupted system responses, especially in highly dynamic scenarios with multiple traffic participants, seriously compromising the accuracy and safety of simulation results. Summary of the Invention

[0005] The purpose of the present invention is to provide a vehicle driving simulation test method and system based on digital twins to address the shortcomings of the background technology.

[0006] In order to achieve the above-mentioned object, the present invention provides the following technical solution: a vehicle driving simulation test method based on digital twin, comprising:

[0007] Build a digital twin corresponding to the actual vehicle and driving environment. The digital twin includes: virtual vehicle model, virtual control system, perception scenario, road traffic environment and behavioral logic mapping rules;

[0008] Obtaining a driving behavior model and control algorithm, loading them into the digital twin, and enabling interaction between the model and the virtual traffic environment through a simulation platform;

[0009] Collect the event timestamps and coordinate information of each module in the simulation platform respectively, and extract the multi-module event timestamp differences in the time direction and the coordinate transformation residuals in the spatial direction;

[0010] Construct a spatiotemporal semantic consistency index model and determine the spatiotemporal semantic consistency of the driving behavior model and the control algorithm during loading based on the set consistency threshold;

[0011] If semantic inconsistency occurs, clock synchronization and coordinate recalibration operations are performed to adjust the inconsistent state;

[0012] After adjustment, re-evaluate spatiotemporal semantic consistency and continue to optimize.

[0013] Preferably, the driving behavior model includes: collecting human driving behavior data from actual road tests or simulation platforms, including throttle opening, braking intensity, steering wheel angle, vehicle speed, distance to the vehicle in front, road type and traffic signal information; extracting and cleaning features of the data, and constructing a driving behavior model using rule-driven, machine learning, deep learning or reinforcement learning methods.

[0014] Preferably, the control algorithm includes: a longitudinal control algorithm for outputting throttle or brake signals according to the target acceleration or speed to control the acceleration and deceleration of the vehicle; a lateral control algorithm for outputting the steering wheel angle according to the target path point to control the steering of the vehicle; the control algorithm is encapsulated as a module that can be called by the simulation platform, supporting interface calling and parameter adjustment.

[0015] Preferably, the event timestamps of the perception, decision-making, control, and vehicle feedback modules during each frame of simulation are collected; the time differences ΔT1, ΔT2, and ΔT3 between the modules are calculated, and the maximum value ΔT=max(ΔT1, ΔT2, ΔT3) is extracted; and the maximum time difference constant index MTDI is generated after analyzing the differences in the event timestamps of multiple modules. The expression is: Where, ΔT max is the maximum time difference between modules in the current frame, μ is ΔT in the benchmark operation phase max The mean of ΔT in the benchmark operation phase is max The standard deviation of .

[0016] Preferably, the local coordinates (x′, y′, z′) are transformed into global coordinates (x, y, z) using the rotation matrix R and the translation vector T: [x, y, z] = R × [x′, y′, z′] + T; the coordinate differences of the same object under different modules are compared, and the coordinate transformation residual is calculated: A is the value collected by the simulation platform, and B is the value calculated in the behavior model or control module.

[0017] Preferably, the coordinate transformation residual abnormality index is generated after the abnormal change of the coordinate transformation residual in a fixed time period, and the generation method is:

[0018] Get the coordinate transformation residual sequence within a fixed time period: E xyz (t), t=1,2,...,N; N is the total number of time points, set the window radius k, and the neighborhood of each time t is: Wt ={E xyz (tk),...,E xyz (t-1), E xyz (t), E xyz (t+1), ..., E xyz (t+k)}; k is the sliding window radius, and the window size is 2k+1;

[0019] Calculate the local median m t , the expression is: m t =median(W t ); and the local median absolute difference (MAD) t , the expression is: MAD t =median(|E xyz (i)-m t |), i∈W t ; For the current time t, calculate the standardized offset S t , the expression is: Where: λ is the scale factor, generating the coordinate transformation residual anomaly index CRAI, normalizing the offset to the anomaly index CRAI, and the output range is 0 to 1: Where: a is the steepness of the exponential function, and θ is the offset threshold.

[0020] Preferably, a spatiotemporal semantic consistency index model is constructed to judge the spatiotemporal semantic consistency of the driving behavior model and the control algorithm during loading according to a set consistency threshold, specifically including:

[0021] The maximum time difference constant index and the coordinate transformation residual constant index are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as the input of the machine learning model. The machine learning model uses each set of comprehensive feature vectors to predict the spatiotemporal semantic consistency value label of the driving behavior model and the control algorithm during loading as the prediction target, and takes minimizing the sum of the prediction errors of the spatiotemporal semantic consistency value labels of all driving behavior models and control algorithms during loading as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The spatiotemporal semantic consistency value of the driving behavior model and the control algorithm during loading is determined according to the model output results. Among them, the machine learning model is a polynomial regression model.

[0022] Preferably, the spatiotemporal semantic consistency value of the driving behavior model and the control algorithm obtained during the loading process is compared with the set consistency threshold. If the spatiotemporal semantic consistency value of the driving behavior model and the control algorithm during the loading process is greater than or equal to the set consistency threshold, it means that the loading process of the driving behavior model and the control algorithm has good spatiotemporal semantic consistency in the current scenario; if the spatiotemporal semantic consistency value of the driving behavior model and the control algorithm during the loading process is less than the set consistency threshold, it means that there is a serious timing offset or coordinate mapping anomaly in the current loading process, and the consistency requirements are not met. The model synchronization mechanism or coordinate correction process should be triggered and the simulation run should be suspended.

[0023] Preferably, if semantic inconsistency occurs, clock synchronization and coordinate recalibration operations are performed to adjust the inconsistent state, specifically including:

[0024] When there is a significant time difference ΔT between modules, the clock offset compensation parameter δ is introduced. t Synchronize: T′ i =T i +δ t Where, T i is the original timestamp; T′ i is the corrected synchronization timestamp; δ t is the clock compensation offset, which is the negative value of the time difference between the reference module and the current module;

[0025] The local coordinates in the behavior model are recalibrated based on the rotation matrix R and translation vector T to obtain a unified global coordinate.

[0026] The present invention also provides a vehicle driving simulation test system based on digital twins, which includes a model loading module, a simulation interaction module, a data acquisition and processing module, a consistency assessment module, an adjustment and correction module, and an optimization module;

[0027] Model loading module: Builds a digital twin corresponding to the actual vehicle and driving environment. The digital twin includes: virtual vehicle model, virtual control system, perception scenario, road traffic environment and behavior logic mapping rules;

[0028] Simulation interaction module: obtains the driving behavior model and control algorithm, loads them into the digital twin, and realizes the interaction between the model and the virtual traffic environment through the simulation platform;

[0029] Data acquisition and processing module: collects the event timestamps and coordinate information of each module in the simulation platform, extracts the multi-module event timestamp differences in the time direction and the coordinate transformation residuals in the spatial direction;

[0030] Consistency assessment module: Builds a spatiotemporal semantic consistency index model and determines the spatiotemporal semantic consistency of the driving behavior model and the control algorithm during loading based on the set consistency threshold;

[0031] Adjustment and correction module: If semantic inconsistency occurs, clock synchronization and coordinate recalibration operations are performed to adjust the inconsistent state;

[0032] Optimization module: re-evaluates spatiotemporal semantic consistency after adjustment and continuously optimizes.

[0033] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0034] 1. This invention constructs a spatiotemporal semantic consistency index model that includes a temporal discrepancy index and a coordinate transformation residual discrepancy index, combined with a polynomial regression algorithm, to accurately assess the spatiotemporal consistency of driving behavior models and control algorithms during loading into a simulation platform. When semantic inconsistencies are detected, the system automatically performs clock synchronization and coordinate recalibration, and continuously adjusts relevant parameters using a closed-loop optimization mechanism. This significantly improves the collaborative accuracy and response stability of the driving behavior model and control algorithm in a virtual environment.

[0035] 2. This method addresses discrepancies such as time delays and coordinate mismatches during digital twin model loading, effectively avoiding simulation misjudgments and system disruptions caused by data asynchrony or spatial deviations. This method improves the reliability and repeatability of autonomous driving model simulation testing and is suitable for complex, multi-participant, and highly dynamic driving simulation scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0037] Figure 1 This is a mind map of the method of the present invention.

[0038] Figure 2 This is a mind map of the system modules of the present invention. DETAILED DESCRIPTION

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0040] Example 1, please refer to Figure 1 As shown, the vehicle driving simulation test method based on digital twin described in this embodiment includes:

[0041] Build a digital twin corresponding to the actual vehicle and driving environment. The digital twin includes: virtual vehicle model, virtual control system, perception scenario, road traffic environment and behavioral logic mapping rules;

[0042] Obtaining a driving behavior model and control algorithm, loading them into the digital twin, and enabling interaction between the model and the virtual traffic environment through a simulation platform;

[0043] Collect the event timestamps and coordinate information of each module in the simulation platform respectively, and extract the multi-module event timestamp differences in the time direction and the coordinate transformation residuals in the spatial direction;

[0044] Construct a spatiotemporal semantic consistency index model and determine the spatiotemporal semantic consistency of the driving behavior model and the control algorithm during loading based on the set consistency threshold;

[0045] If semantic inconsistency occurs, clock synchronization and coordinate recalibration operations are performed to adjust the inconsistent state;

[0046] After adjustment, re-evaluate spatiotemporal semantic consistency and continue to optimize.

[0047] To achieve high-fidelity driving simulation testing, we first build a digital twin that corresponds to the actual vehicle and its operating environment. This digital twin includes the following key components:

[0048] Virtual vehicle model: Constructs vehicle kinematic and dynamic models based on real vehicle parameters (wheelbase, mass, dynamic characteristics, steering mechanism, etc.); supports simulated responses to control inputs such as vehicle speed, acceleration, braking, and steering; includes virtual modeling of on-board sensor layouts, such as lidar, cameras, and millimeter-wave radar.

[0049] Virtual control system: Builds virtual ECU logic to support access to driving behavior models and control algorithms; simulates decision-making, path planning, and underlying control processes; and provides an interface for interaction between the simulation platform and control logic.

[0050] Perception scenarios: Build three-dimensional simulation scenarios based on actual traffic environments such as urban roads, highways, and intersections; configure dynamic behavior models of traffic participants (such as vehicles, pedestrians, and non-motor vehicles); and simulate environmental factors such as weather, lighting, and road conditions.

[0051] Road traffic environment: Establish road structure (lane lines, intersection topology, traffic lights) based on HD maps; introduce traffic rule logic (such as priority, speed limits, and traffic light responses); and implement environmental change mechanisms that are linked to vehicle behavior.

[0052] Behavioral logic mapping rules: Build a correspondence between the actual driver's decision logic and the digital model; support mapping the driving intentions output by the model (such as "accelerate" and "turn left") into control commands and feedback to the simulated vehicle; achieve decoupling and linkage between semantic layer behavior and physical layer execution. Collect a large amount of human driver behavior data from actual road tests or simulation platforms, including throttle position, braking intensity, steering wheel angle, vehicle speed, distance to the vehicle ahead, road type, traffic signals, etc.

[0053] The raw data is cleaned and features are extracted, such as behavioral intention recognition (lane change, overtaking, following), operation style labels (aggressive, conservative), and environmental context (intersections, rainy days, nighttime, etc.).

[0054] Build a behavioral model in one of the following ways:

[0055] Rule-driven model: Constructs state-behavior mapping based on expert rules, such as determining acceleration and deceleration based on thresholds;

[0056] Machine learning models: such as decision trees and random forests;

[0057] Deep learning models: such as LSTM and Transformer networks, used for sequential behavior prediction;

[0058] Reinforcement learning model: Strategy optimization based on simulation feedback.

[0059] The trained behavior model is encapsulated into a modular software package that supports interfaces such as Python, C++, or ROS, and the output is behavior decision or motion intention.

[0060] The design and extraction of control algorithms include:

[0061] Longitudinal control algorithm (acceleration / braking): Typical algorithms include PID controller and model predictive control (MPC); the input is the target acceleration or speed, and the output is the throttle / brake signal.

[0062] Lateral control algorithm (steering control): includes pure tracking control, Stanley method, MPC steering control, etc.; the input is the target path point and the output is the steering wheel angle.

[0063] The control module is encapsulated in the form of a dynamic link library or ROS node so that it can be called in the simulation platform and has adjustable parameters and input and output interfaces.

[0064] Select a vehicle simulation platform with open interfaces, such as CARLA, PreScan, VTD, etc., and complete the vehicle dynamics model and sensor configuration.

[0065] The driving behavior model and control algorithm are mounted on the decision layer and control layer of the simulation platform respectively, and data interaction is carried out through message middleware (such as ROS, UDP, ZMQ) or API.

[0066] The driving behavior model inputs the perception data from the simulation environment (vehicle position, surrounding traffic objects, etc.) and outputs high-level driving intentions (such as acceleration, turning, and lane changing);

[0067] The control algorithm receives the intention instructions from the behavior model and outputs specific control quantities (such as steering angle and throttle value);

[0068] The instructions are applied to the virtual vehicle model through the control interface of the simulation platform.

[0069] After loading, a simulation clock synchronization and module call scheduling mechanism is established to ensure that the perception-decision-control process has the correct timing within each simulation frame, supporting real-time or accelerated simulation.

[0070] The loaded model is functionally verified in typical traffic scenarios to check the rationality of model behavior, control response smoothness and semantic consistency, and the operation log is recorded for analysis.

[0071] The timestamp collection mechanism (for different modules) includes:

[0072] Perception module: records the simulation timestamp Tperception of the data frame generated by the virtual sensor (such as lidar and camera).

[0073] Decision module: records the timestamp Tdecision when the behavior model outputs the control instruction or motion intention.

[0074] Control module: records the timestamp Tcontrol of the steering angle and acceleration commands output by the control algorithm.

[0075] Execution feedback module (such as body model): records the timestamp Tvehicle when the vehicle status is updated.

[0076] The timestamp must be based on the simulation engine's unified simulation clock to avoid errors caused by system time offset.

[0077] For each module involving spatial entities (vehicle position, path points, sensor coordinates, etc.), collect their three-dimensional coordinate data (x, y, z) in the current frame;

[0078] The coordinate sources include: absolute coordinates provided by the simulation platform (such as the world coordinate system); internal reference coordinates in the behavior model (such as the vehicle coordinate system, path-relative coordinate system); and target point coordinates used by the control module (such as the desired path point).

[0079] The time difference between each module is extracted, for example: ΔT1 = |Tperception-Tdecision|; ΔT2 = |Tdecision-Tcontrol|; ΔT3 = |Tcontrol-Tvehicle|; and is comprehensively expressed as the maximum time difference: ΔT = max(ΔT1, ΔT2, ΔT3). If ΔT is large, it means that the perception information is not transmitted to the decision module in a timely manner or the decision instruction is transmitted to the controller with a lag. This will cause the model to make decisions based on outdated data, leading to behavioral deviations (such as missing the opportunity to brake or changing lanes incorrectly).

[0080] After analyzing the differences in timestamps of multi-module events, the maximum time difference constant index MTDI is generated, and the expression is: Where, ΔT max is the maximum time difference between modules in the current frame, μ is ΔT in the benchmark operation phase max The mean of ΔT in the benchmark operation phase is max The standard deviation of .

[0081] Different modules may use different coordinate systems (e.g., the simulation platform uses the global coordinate system, while the behavioral model uses the vehicle coordinate system). The coordinate systems need to be unified. The local coordinates (x′, y′, z′) are transformed into global coordinates (x, y, z) using the rotation matrix R and the translation vector T: [x, z] = R × [x′, y′, z′] + T. The coordinate differences of the same object in different modules are compared to calculate the coordinate transformation residual: A is the value collected by the simulation platform, and B is the value calculated in the behavior model or control module.

[0082] The coordinate transformation residual abnormality index is generated by analyzing the abnormal changes of the coordinate transformation residuals within a fixed time period. The generation method is:

[0083] Get the coordinate transformation residual sequence within a fixed time period: E xyz (t), t=1,2,...,N; N is the total number of time points, set the window radius k, and the neighborhood of each time t is: W t ={E xyz (tk),...,E xyz (t-1), E xyz(t), E xyz (t+1), ..., E xyz (t+k)}; k is the sliding window radius, and the window size is 2k+1.

[0084] Calculate the local median m t , the expression is: m t =median(W t ); and the local median absolute difference (MAD) t , the expression is: MAD t =median(|E xyz (i)-m t |), i∈W t ; For the current time t, calculate the standardized offset S t , the expression is: Where: λ is the scale factor, and the recommended value is 1.4826 (to make it equivalent to the standard deviation under Gaussian distribution); generate the coordinate transformation residual anomaly index CRAI, and normalize the offset to the anomaly index CRAI, with the output range of 0 to 1: Where: a is the steepness of the exponential function, usually ranging from 2 to 5; θ is the offset threshold, with a recommended initial value of 3 (i.e., a deviation of 3 times the MAD is considered abnormal); a CRAI close to 1 indicates a strong abnormality, and a CRAI close to 0 indicates normality.

[0085] Construct a spatiotemporal semantic consistency index model and determine the spatiotemporal semantic consistency of the driving behavior model and the control algorithm during loading based on the set consistency threshold. Specifically, it includes:

[0086] The maximum time difference constant index and the coordinate transformation residual constant index are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as the input of the machine learning model. The machine learning model uses each set of comprehensive feature vectors to predict the spatiotemporal semantic consistency value label of the driving behavior model and the control algorithm during loading as the prediction target, and takes minimizing the sum of the prediction errors of the spatiotemporal semantic consistency value labels of all driving behavior models and control algorithms during loading as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The spatiotemporal semantic consistency value of the driving behavior model and the control algorithm during loading is determined according to the model output results. Among them, the machine learning model is a polynomial regression model.

[0087] Compare the spatiotemporal semantic consistency value obtained during the loading process of the driving behavior model and control algorithm with the set consistency threshold. If the spatiotemporal semantic consistency value during the loading process of the driving behavior model and control algorithm is greater than or equal to the set consistency threshold, it means that the loading process of the driving behavior model and control algorithm has good spatiotemporal semantic consistency in the current scenario, the system integration is reliable, and simulation or deployment can continue; if the spatiotemporal semantic consistency value during the loading process of the driving behavior model and control algorithm is less than the set consistency threshold, it means that the current loading process has serious timing offset or coordinate mapping anomaly, and does not meet the consistency requirements. The model synchronization mechanism or coordinate correction process should be triggered and the simulation run should be suspended.

[0088] If semantic inconsistency occurs, perform clock synchronization and coordinate recalibration to adjust the inconsistent state, including:

[0089] When there is a significant time difference ΔT between modules, the clock offset compensation parameter δ is introduced. t Synchronize: T′ i =T i +δ t Where, T i is the original timestamp (such as the local time of the perception or control module); T′ i is the corrected synchronization timestamp; δ t It is the clock compensation offset, which is the negative value of the time difference between the reference module (such as the simulation master) and the current module.

[0090] Align the event times of each module to a unified simulation clock reference, eliminating timing misalignment in the decision chain.

[0091] When the spatial semantics are inconsistent (i.e., the coordinate residuals are too large), a rigid transformation is used to recalibrate the coordinates. This involves transforming the local coordinates (x′, y′, z′) into global coordinates (x, y, z) using the rotation matrix R and the translation vector T. This ensures that the spatial points in the behavioral model and control algorithm are precisely consistent with the geographic coordinate system in the simulation platform.

[0092] After clock synchronization and coordinate recalibration, the system recollects event timestamps and coordinate data from each module and recalculates the Maximum Temporal Difference Index (MTDI) and the Coordinate Transformation Residual Difference Index (CRAI). These two indices are combined into a new comprehensive feature vector and input into the trained polynomial regression model to obtain the spatiotemporal semantic consistency value of the current driving behavior model and the control algorithm during loading.

[0093] The consistency value is compared with a pre-set consistency threshold. If the consistency value is greater than or equal to the threshold, the adjustment has achieved the expected effect, the system has good spatiotemporal coordination in the current scenario, and the simulation can proceed normally. If the consistency value is still below the threshold, the current adjustment is insufficient and the system enters the continuous optimization process.

[0094] During the continuous optimization phase, the system automatically fine-tunes time synchronization parameters (such as clock offset values) and spatial calibration parameters (such as rotation angles and translation vectors) based on the consistency assessment results, iteratively adjusting in small steps and re-evaluating the consistency value after each adjustment. This process forms a closed loop, continuously executing the "adjustment-assessment-optimization" cycle until the consistency value reaches the set threshold or the changes in the evaluation value after multiple consecutive times tend to be stable. Convergence is determined and the optimization is terminated. This method ensures that the driving behavior model and control algorithm maintain high-precision temporal and spatial semantic consistency in the simulation platform, improving the reliability of system integration and the effectiveness of simulation results.

[0095] Example 2, please refer to Figure 2 As shown, the vehicle driving simulation test system based on digital twin described in this embodiment includes a model loading module, a simulation interaction module, a data acquisition and processing module, a consistency assessment module, an adjustment and correction module, and an optimization module;

[0096] Model loading module: Builds a digital twin corresponding to the actual vehicle and driving environment. The digital twin includes: virtual vehicle model, virtual control system, perception scenario, road traffic environment and behavior logic mapping rules;

[0097] Simulation interaction module: obtains the driving behavior model and control algorithm, loads them into the digital twin, and realizes the interaction between the model and the virtual traffic environment through the simulation platform;

[0098] Data acquisition and processing module: collects the event timestamps and coordinate information of each module in the simulation platform, extracts the multi-module event timestamp differences in the time direction and the coordinate transformation residuals in the spatial direction;

[0099] Consistency assessment module: Builds a spatiotemporal semantic consistency index model and determines the spatiotemporal semantic consistency of the driving behavior model and the control algorithm during loading based on the set consistency threshold;

[0100] Adjustment and correction module: If semantic inconsistency occurs, clock synchronization and coordinate recalibration operations are performed to adjust the inconsistent state;

[0101] Optimization module: re-evaluates spatiotemporal semantic consistency after adjustment and continuously optimizes.

[0102] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0103] It should be understood that the term "and / or" as used herein simply describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the related objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0104] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0105] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A vehicle driving simulation test method based on digital twins, characterized by: include: Build a digital twin corresponding to the actual vehicle and driving environment. The digital twin includes: virtual vehicle model, virtual control system, perception scenario, road traffic environment and behavioral logic mapping rules; Obtaining a driving behavior model and control algorithm, loading them into the digital twin, and enabling interaction between the model and the virtual traffic environment through a simulation platform; Collect the event timestamps and coordinate information of each module in the simulation platform respectively, and extract the multi-module event timestamp differences in the time direction and the coordinate transformation residuals in the spatial direction; Construct a spatiotemporal semantic consistency index model and determine the spatiotemporal semantic consistency of the driving behavior model and the control algorithm during loading based on the set consistency threshold; If semantic inconsistency occurs, clock synchronization and coordinate recalibration operations are performed to adjust the inconsistent state; After adjustment, re-evaluate spatiotemporal semantic consistency and continue to optimize.

2. The vehicle driving simulation test method based on digital twin according to claim 1, characterized in that: The driving behavior model includes: collecting human driving behavior data from actual road tests or simulation platforms, including throttle opening, braking intensity, steering wheel angle, vehicle speed, distance to the vehicle in front, road type and traffic signal information; extracting and cleaning features of the data, and constructing a driving behavior model using rule-driven, machine learning, deep learning or reinforcement learning methods.

3. The vehicle driving simulation test method based on digital twin according to claim 1, characterized in that: The control algorithm includes: a longitudinal control algorithm for outputting throttle or brake signals according to the target acceleration or speed to control vehicle acceleration and deceleration; a lateral control algorithm for outputting steering wheel angles according to the target path point to control vehicle steering; the control algorithm is encapsulated as a module that can be called by the simulation platform, supporting interface calls and parameter adjustments.

4. The vehicle driving simulation test method based on digital twin according to claim 1, characterized in that: Collect event timestamps of the perception, decision-making, control, and vehicle feedback modules during each frame of simulation; calculate the time differences ΔT1, ΔT2, and ΔT3 between each module, and extract the maximum value ΔT = max(ΔT1, ΔT2, ΔT3); After analyzing the differences in timestamps of multi-module events, the maximum time difference constant index MTDI is generated, and the expression is: Where, ΔT max is the maximum time difference between modules in the current frame, μ is ΔT in the benchmark operation phase max The mean of ΔT in the benchmark operation phase is max The standard deviation of .

5. The vehicle driving simulation test method based on digital twin according to claim 4 is characterized in that: Use the rotation matrix R and translation vector T to transform the local coordinates (x′, y′, z′) into global coordinates (x, y, z): [x, y, z] = R × [x′, y′, z′] + T; compare the coordinate differences of the same object under different modules and calculate the coordinate transformation residual: A is the value collected by the simulation platform, and B is the value calculated in the behavior model or control module.

6. The vehicle driving simulation test method based on digital twin according to claim 5, characterized in that: The coordinate transformation residual abnormality index is generated by analyzing the abnormal changes of the coordinate transformation residuals within a fixed time period. The generation method is: Get the coordinate transformation residual sequence within a fixed time period: E xyz (t), t=1,2,...,N; N is the total number of time points, set the window radius k, and the neighborhood of each time t is: W t ={E xyz (tk),...,E xyz (t-1), E xyz (t), E xyz (t+1), ..., E xyz (t+k)}; k is the sliding window radius, and the window size is 2k+1; Calculate the local median m t , the expression is: m t =median(W t ); and the local median absolute difference (MAD) t , the expression is: MAD t =median(|E xyz (i)-m t |), i∈W t ; For the current time t, calculate the standardized offset S t , the expression is: Where: λ is the scale factor, generating the coordinate transformation residual anomaly index CRAI, normalizing the offset to the anomaly index CRAI, and the output range is 0 to 1: Where: a is the steepness of the exponential function, and θ is the offset threshold.

7. The vehicle driving simulation test method based on digital twin according to claim 6, characterized in that: Construct a spatiotemporal semantic consistency index model and determine the spatiotemporal semantic consistency of the driving behavior model and the control algorithm during loading based on the set consistency threshold. Specifically, it includes: The maximum time difference constant index and the coordinate transformation residual constant index are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as the input of the machine learning model. The machine learning model uses each set of comprehensive feature vectors to predict the spatiotemporal semantic consistency value label of the driving behavior model and the control algorithm during loading as the prediction target, and takes minimizing the sum of the prediction errors of the spatiotemporal semantic consistency value labels of all driving behavior models and control algorithms during loading as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The spatiotemporal semantic consistency value of the driving behavior model and the control algorithm during loading is determined according to the model output results. Among them, the machine learning model is a polynomial regression model.

8. The vehicle driving simulation test method based on digital twin according to claim 7, characterized in that: Compare the spatiotemporal semantic consistency value obtained during the loading process of the driving behavior model and the control algorithm with the set consistency threshold. If the spatiotemporal semantic consistency value during the loading process of the driving behavior model and the control algorithm is greater than or equal to the set consistency threshold, it means that the loading process of the driving behavior model and the control algorithm has good spatiotemporal semantic consistency in the current scenario; if the spatiotemporal semantic consistency value during the loading process of the driving behavior model and the control algorithm is less than the set consistency threshold, it means that the current loading process has serious timing offset or coordinate mapping anomaly, and does not meet the consistency requirements. The model synchronization mechanism or coordinate correction process should be triggered and the simulation run should be suspended.

9. The vehicle driving simulation test method based on digital twin according to claim 8, characterized in that: If semantic inconsistency occurs, perform clock synchronization and coordinate recalibration to adjust the inconsistent state, including: When there is a significant time difference ΔT between modules, the clock offset compensation parameter δ is introduced. t Synchronize: T′ i =T i +δ t Where, T i is the original timestamp; T′ i is the corrected synchronization timestamp; δ t is the clock compensation offset, which is the negative value of the time difference between the reference module and the current module; The local coordinates in the behavior model are recalibrated based on the rotation matrix R and translation vector T to obtain a unified global coordinate.

10. A vehicle driving simulation test system based on digital twins, used to implement a vehicle driving simulation test method based on digital twins according to any one of claims 1 to 9, characterized in that: It includes model loading module, simulation interaction module, data acquisition and processing module, consistency assessment module, adjustment and correction module and optimization module; Model loading module: Builds a digital twin corresponding to the actual vehicle and driving environment. The digital twin includes: virtual vehicle model, virtual control system, perception scenario, road traffic environment and behavior logic mapping rules; Simulation interaction module: obtains the driving behavior model and control algorithm, loads them into the digital twin, and realizes the interaction between the model and the virtual traffic environment through the simulation platform; Data acquisition and processing module: collects the event timestamps and coordinate information of each module in the simulation platform, extracts the multi-module event timestamp differences in the time direction and the coordinate transformation residuals in the spatial direction; Consistency assessment module: Builds a spatiotemporal semantic consistency index model and determines the spatiotemporal semantic consistency of the driving behavior model and the control algorithm during loading based on the set consistency threshold; Adjustment and correction module: If semantic inconsistency occurs, clock synchronization and coordinate recalibration operations are performed to adjust the inconsistent state; Optimization module: re-evaluates spatiotemporal semantic consistency after adjustment and continuously optimizes.

Images