Self-adaptive intelligent regulation and control system for digital twin-driven two-wheeled self-balancing vehicle

Abstract

The invention relates to the technical field of vehicle control, and discloses a self-adaptive intelligent regulation and control system for a digital twin-driven two-wheeled self-balancing vehicle. The system comprises a perception and data acquisition module, a driver portrait and state identification module, a real-time digital twinning module, a prospective emergency regulation and control module, a dynamic coupling decision module, a safety coverage and underlying execution module, a data optimization and evolution module and a cloud platform. The real-time twinning module interacts with the portrait module in a two-way mode, portrait parameters are corrected in real time through model mismatch signals, and a human-vehicle coupling dynamics model is corrected; the prospective emergency regulation and control module analyzes a future state and calculates a risk index and an emergency level; the dynamic decision-making module dynamically adjusts and controls the optimization target and constraint according to the risk index and the emergency level; the data optimization and evolution module captures a data packet when a key event is triggered and uploads the data packet to the cloud, and the cloud pushes a strategy update packet to the vehicle-mounted system after offline optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle control, in particular to a two-wheeled self-balancing vehicle adaptive intelligent control system driven by digital twin. BACKGROUND

[0002] Two-wheeled self-balancing vehicles, especially self-balancing vehicles with front and rear double wheels in longitudinal layout, as a kind of intelligent transportation tool, its core technology relies on inertial measurement unit (IMU) to obtain vehicle attitude in real time, and through self-balancing control algorithm based on control moment gyro self-balancing mechanism to adjust pitch angle and roll angle to maintain stability. The existing technology mainly focuses on improving the sensor fusion accuracy and the robustness of the control algorithm. In order to improve safety, some vehicles have been equipped with anti-lock braking system (ABS) or traction control system (TCS). At the same time, some solutions also begin to introduce over-the-air technology (OTA) to realize firmware update and parameter adjustment.

[0003] Although the existing technology has achieved certain results in basic balance control, it still faces bottlenecks. First of all, the current control strategy mostly adopts fixed parameter model, which cannot dynamically adapt to the individualized driving style (such as aggressive or conservative) of the driver, resulting in poor driving experience, and may hide safety risks under certain operations. Secondly, this kind of control based on fixed parameters has limited ability to deal with complex working conditions. When the vehicle encounters sudden crosswind, uneven road or low adhesion road, the system often faces instability risk due to response lag or insufficient control margin.

[0004] In addition, the existing active safety functions are mostly limited to passive response, lacking the ability of forward-looking prediction of future risks, and cannot take active intervention before instability or collision occurs. At the same time, a large amount of data generated by the system during operation cannot be fully utilized; these data are usually only used for instantaneous control, and cannot form a closed-loop data optimization process. The mechanism of uploading key event data to the cloud for deep analysis, simulation and strategy iteration is lacking, resulting in the fact that the control logic cannot evolve continuously with the accumulation of operation experience, and it is difficult to cope with increasingly complex driving scenarios.

[0005] Therefore, the present application proposes a two-wheeled self-balancing vehicle adaptive intelligent control system driven by digital twin to solve the deficiencies of the prior art. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a two-wheeled self-balancing vehicle adaptive intelligent control system driven by digital twin, which solves the problems of poor adaptability, lack of risk prediction ability and inability to utilize data for continuous evolution of the existing self-balancing vehicle control system under complex working conditions.

[0007] In order to achieve the above object, the present application is realized by the following technical scheme: A digital twin driven two-wheeled self-balancing vehicle adaptive intelligent regulation and control system, comprising: a perception and data acquisition module, configured to acquire real-time running data of the two-wheeled self-balancing vehicle; a driver portrait and state recognition module, configured to construct a driver personalized style portrait parameter based on the real-time running data; a real-time digital twin module, configured to construct a human-vehicle coupled dynamics model based on a vehicle dynamics model and in combination with the driver personalized style portrait parameter, and predict a future state sequence of the vehicle; a forward-looking emergency regulation and control module, configured to analyze the future state sequence, calculate a real-time risk index and divide an emergency state level; a dynamic coupling decision module, configured to adjust a control optimization target and a control constraint in real time according to the real-time risk index and the emergency state level, and solve and generate an optimal control instruction sequence; a safety coverage and bottom layer execution module, configured to perform safety arbitration on the optimal control instruction sequence according to the emergency state level, and determine a final control instruction; a data optimization and evolution module, configured to capture and package a high-value emergency data packet in response to a key event trigger signal of the forward-looking emergency regulation and control module; a cloud platform, configured to receive the high-value emergency data packet and generate a strategy update packet; wherein the real-time digital twin module is further configured to generate a model mismatch signal and send it to the driver portrait and state recognition module, and the driver portrait and state recognition module is configured to receive the model mismatch signal and correct the driver portrait parameter, and the corrected driver portrait parameter is fed back to the real-time digital twin module to correct the human-vehicle coupled dynamics model; and the data optimization and evolution module is further configured to receive the strategy update packet generated by the cloud platform and update the regulation and control logic of the vehicle-mounted system.

[0008] Preferably, the real-time digital twin module is specifically configured to: compare a measurement predicted value based on the human-vehicle coupled dynamics model with actual measurement data from the perception and data acquisition module, generate a prediction residual, and generate the model mismatch signal based on statistical characteristics of the prediction residual.

[0009] Preferably, the driver portrait and state recognition module is specifically configured to: when receiving the model mismatch signal, fuse driver direct operation data features from the perception and data acquisition module to identify fatigue or distraction abnormal states of the driver, and correct the driver portrait parameter.

[0010] Preferably, the forward-looking emergency regulation and control module is specifically configured to: dynamically adjust a weight coefficient used to calculate the real-time risk index based on the driver portrait parameter and the current vehicle speed, and dynamically adjust a group of dynamic threshold values used to divide the emergency state level.

[0011] Preferably, the dynamic coupling decision module is implemented as a model predictive controller, and the dynamic coupling decision module is specifically configured to nonlinearly increase a state weight corresponding to vehicle posture stability in an internal cost function to adjust the control optimization objective when the real-time risk index increases.

[0012] Preferably, the dynamic coupling decision module is further specifically configured to dynamically tighten a state constraint boundary in an internal optimization problem to adjust the control constraint when the real-time risk index increases.

[0013] Preferably, the safety override and underlying execution module is specifically configured to adopt an optimal control instruction sequence when the emergency state level is lower than an override trigger threshold, and ignore the optimal control instruction sequence and invoke a preset emergency control action vector as the final control instruction when the emergency state level is greater than or equal to the override trigger threshold.

[0014] Preferably, the data optimization and evolution module is specifically configured to access an internal circular buffer to intercept full-dimensional data within a preset time window before and after a time corresponding to the critical event trigger signal, and the full-dimensional data includes internal state data and I / O data from a perception and data acquisition module to a safety override and underlying execution module, and a high-precision time stamp is retained.

[0015] Preferably, the cloud platform is specifically configured to perform deep rollback and counterfactual simulation using the high-value emergency data package, and perform iterative optimization of a dynamic threshold of the forward-looking emergency regulation module or a dynamic weight function of the dynamic coupling decision module using the simulation results and an offline reinforcement learning algorithm to generate the strategy update package.

[0016] Preferably, the human-vehicle coupling dynamics model is a state transition function, and the driver individualized style portrait parameters explicitly determine the dynamic characteristics of the state transition function, so that when the driver portrait parameters are updated, the dynamic characteristics of the state transition function are correspondingly changed.

[0017] The application provides a digital twin driven two-wheeled self-balancing vehicle adaptive intelligent regulation system.

[0018] 1、The application realizes accurate perception and self-adaptation of the system to the driver state through the bidirectional closed-loop correction mechanism of the real-time digital twin module and the driver portrait and state recognition module. When the driver appears fatigue, distraction or abnormal driving behavior, the system can quickly recognize this abnormal state through the model mismatch signal and correct the human-vehicle coupling dynamics model in real time, ensuring that the regulation strategy always matches the instantaneous state of the driver and improving the safety of human-machine cooperation.

[0019] 2, The application realizes the active prediction of instability risk through the analysis of the future state sequence by the forward-looking emergency regulation module. Instead of passively responding to deviations, the system adjusts the control optimization target and control constraints in real time through the dynamic coupling decision module according to the calculated real-time risk index. This enables the regulation system to automatically tighten the safety margin when the risk increases, prioritizing attitude stability, effectively suppressing potential instability risk, and improving the active safety performance of the vehicle under complex working conditions.

[0020] 3, The application builds a closed-loop evolution system from vehicle edge data triggering to cloud strategy optimization through the collaborative work of the data optimization and evolution module and the cloud platform. After experiencing key events, the vehicle system can upload high-value emergency data packets to the cloud; the cloud platform iteratively optimizes the strategy through algorithms such as counterfactual simulation and offline reinforcement learning, and pushes the updated strategy model to the vehicle. This enables the regulation system to learn from historical experience, achieving continuous improvement and iteration of safety performance. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is the system overall function and hardware block diagram of the application;

[0022] Figure 2 is a schematic diagram of the twin-image portrait bidirectional iteration mechanism of the application;

[0023] Figure 3 is a schematic diagram of the dynamic risk coupling decision mechanism of the application;

[0024] Figure 4 is a schematic diagram of the vehicle-cloud collaborative evolution closed loop of the application.

[0025] Among them, 100, perception and data acquisition module; 200, driver portrait and state recognition module; 300, real-time digital twin module; 400, forward-looking emergency regulation module; 500, dynamic coupling decision module; 600, safety coverage and bottom execution module; 700, data optimization and evolution module; 800, cloud platform. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the application will be described in detail below with reference to the drawings of the application specification. Obviously, the described embodiments are only a part of the embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the application.

[0027] Referring to the drawings Figure 1 The digital twin driven two-wheeled self-balancing car adaptive intelligent regulation system provided by the application can include:

[0028] The perception and data acquisition module 100 is responsible for collecting real-time running data of the two-wheeled self-balancing vehicle, including inertial measurement unit (IMU) data of the vehicle, wheel encoder data, and driver's steering and acceleration / deceleration operation input.

[0029] The driver portrait and state recognition module 200 constructs the personalized style portrait parameters of the driver based on the input data; the real-time digital twin module 300 constructs a human-vehicle coupled dynamics model based on the vehicle dynamics model and in combination with the driver portrait parameters, and predicts the state sequence of the vehicle in a future time window using the model.

[0030] The real-time digital twin module 300 and the driver portrait and state recognition module 200 perform bidirectional data interaction to realize online adaptive correction of the model.

[0031] In one direction, the real-time digital twin module 300 compares its predicted state with the measured data from the perception and data acquisition module 100, generates a residual or mismatch signal of the model prediction, and sends the mismatch signal to the driver portrait and state recognition module 200; the driver portrait and state recognition module 200, after receiving the mismatch signal, identifies the current state of the driver (such as fatigue or distraction abnormal state, dangerous driving behavior or abnormal acceleration / deceleration, etc.) in combination with the direct operation data features of the driver, and makes real-time correction to the driver portrait parameters.

[0032] In the other direction, the driver portrait and state recognition module 200 feeds back the corrected driver portrait parameters to the real-time digital twin module 300, and the real-time digital twin module 300 uses the parameters to correct the internal dynamics model online to improve the accuracy of subsequent predictions.

[0033] The forward-looking emergency regulation module 400 analyzes the future state sequence, evaluates the instability risk or collision risk of the vehicle in the future time domain, and calculates a continuous real-time risk index and classifies the current emergency state level (such as normal, attention, warning, emergency) accordingly.

[0034] The dynamic coupling decision module 500 (such as a model predictive controller) adjusts its internal control optimization target (such as the weight matrix in the cost function) and control constraints (such as the allowed roll angle safety boundary) in real time according to the received risk index and emergency level; the dynamic coupling decision module 500 solves and generates the optimal control instruction sequence under the adjusted target and constraints, and sends it to the safety coverage and underlying execution module 600.

[0035] The safety cover and bottom layer execution module 600 performs safety arbitration according to the emergency state level output by the forward-looking emergency regulation module 400; in a non-emergency state, the safety cover and bottom layer execution module 600 adopts the instructions of the dynamic coupling decision module 500; in a high-level emergency state, the safety cover and bottom layer execution module 600 uses a preset emergency action (such as starting a self-balancing mode or activating an emergency brake) to cover or modify the instructions; the safety cover and bottom layer execution module 600 finally sends the control instructions determined by arbitration to the bottom layer actuators (such as motor drivers) of the vehicle, completing the closed-loop regulation and control of the vehicle posture and operation.

[0036] In the above process, when and only when the forward-looking emergency regulation module 400 identifies a state of attention, warning or emergency level, it will also send a key event trigger signal to the data optimization and evolution module 700; the data optimization and evolution module 700 acts as a vehicle-mounted edge unit, and after receiving the trigger signal, immediately performs data slicing operation, captures the full-module data with high-precision time synchronization within the time window before and after the key event occurs, and packs it into a high-value emergency data package; the data optimization and evolution module 700 then preferentially uploads the data package to the cloud platform 800.

[0037] After the cloud platform 800 receives the data package, it performs root cause analysis, counterfactual simulation and offline reinforcement learning algorithm to optimize the judgment threshold of the forward-looking emergency regulation module 400 or the dynamic weight function of the dynamic coupling decision module 500, and generates a new strategy model.

[0038] The new strategy model is pushed to the vehicle-mounted system by the data optimization and evolution module 700 through OTA (over-the-air technology) to update the corresponding regulation logic, thereby realizing closed-loop evolution of system safety performance.

[0039] The present application provides a kind of digital twin driven two-wheel self-balancing car adaptive intelligent regulation method, comprising the following steps:

[0040] S100, the real-time running data of two-wheel self-balancing car is collected by perception and data acquisition module, data includes inertial measurement unit data, wheel encoder data and driver's steering and driving operation input, and the preprocessed data is output to real-time digital twin module and driver portrait and state identification module simultaneously;

[0041] S200, constructing a driver personalized style portrait parameter based on the input data by the driver portrait and state recognition module, and constructing a human-vehicle coupled dynamics model based on the vehicle dynamics and the portrait parameter by the real-time digital twin module to predict the future state sequence of the vehicle; in this process, the real-time digital twin module and the driver portrait and state recognition module perform bidirectional data interaction: the real-time digital twin module calculates the residual of the predicted state and the measured data and generates a model mismatch signal feedback to the driver portrait and state recognition module, and the driver portrait and state recognition module identifies the driver state and corrects the portrait parameter accordingly, and the corrected parameter is fed back to the real-time digital twin module to correct the dynamics model online;

[0042] S300, analyzing the future state prediction sequence output by the real-time digital twin module by the forward-looking emergency regulation module, evaluating the instability risk and collision risk of the vehicle in the future time domain, calculating a continuous real-time risk index and dividing an emergency state level, and synchronously outputting the real-time risk index and the emergency state level to the dynamic coupling decision module and the safety coverage and bottom layer execution module;

[0043] S400, adjusting the weight matrix and state constraint boundary of the optimization objective function inside the control algorithm in real time by the dynamic coupling decision module according to the received real-time risk index and emergency state level, and solving and generating an optimal control instruction sequence under the adjusted objective function and constraint boundary;

[0044] S500, executing safety arbitration according to the emergency state level by the safety coverage and bottom layer execution module, adopting the instruction of the dynamic coupling decision module in the non-emergency state, covering or correcting the instruction by using the preset emergency action in the high-level emergency state, and sending the finally determined control instruction to the vehicle bottom layer actuator to complete the closed-loop control;

[0045] S600, when the forward-looking emergency regulation module triggers a specific level of emergency state, responding to the key event trigger signal by the data optimization and evolution module, performing edge-side data slicing operation, capturing and packaging full-module data containing high-precision time synchronization information to generate high-value emergency data package, and uploading the data package to the cloud platform in priority; the cloud platform performs root cause analysis, counterfactual simulation and strategy optimization based on the data package, and the generated updated strategy model is pushed to the vehicle by over-the-air technology to update the regulation logic.

[0046] The specific implementation principles of each step and each logical module involved therein will be described in detail below with reference to the accompanying drawings.

[0047] Referring to the accompanying Figure 1The system provided by the application can include a central control unit (ECU) as the calculation and control core of the system, a sensor component for state sensing, an actuator for performing control, a self-balancing module, and a vehicle networking communication module.

[0048] In one embodiment, the sensor component can specifically include an inertial measurement unit (IMU) for measuring the real-time attitude angle (such as roll angle , pitch angle ) and corresponding angular velocity (such as ) of the two-wheeled self-balancing vehicle, at least a wheel encoder for measuring the wheel rotation speed to calculate the vehicle speed , and a driver input sensor.

[0049] The driver input sensor can include a steering sensor (such as an angle or torque sensor installed on the steering mechanism) to detect the steering operation of the driver , and a drive sensor (such as a Hall sensor, a pressure sensor) to detect the acceleration or braking intention of the driver ; the actuator can include one or more motor drivers and drive motors controlled thereby; the self-balancing module can be integrated in the central control unit (ECU) or the motor driver, and its main function is to perform basic self-balancing PID or LQR control, maintain the static and low-speed dynamic stability of the vehicle, and receive safety coverage instructions from the underlying execution module 600 in the high-level emergency state, forcibly intervene and output enhanced stabilizing torque; the vehicle networking communication module (OBU) is used to realize wireless data communication between the vehicle-mounted system and the cloud platform 800.

[0050] In one embodiment of the application, the functions of the perception and data acquisition module 100 are realized by a software program running on the central control unit, which is responsible for performing the S100 step and can specifically include the following sub-steps:

[0051] S100a, the perception and data acquisition module 100 periodically reads the raw accelerometer and gyroscope data from the inertial measurement unit (IMU), reads the pulse count or angle data from the wheel encoder, and reads the voltage, current or angle signal from the driver input sensor.

[0052] S100b, the perception and data acquisition module 100 performs preprocessing on the obtained raw data; for example, fusion calculation is performed on the IMU raw data to obtain the attitude angle and angular velocity For the fusion of IMU data, one skilled in the art can use, for example, complementary filtering or Kalman filtering algorithms, the implementation of which belongs to the known technology in the art, which will not be repeated here; at the same time, the perception and data acquisition module 100 processes the wheel encoder data to calculate the vehicle speed , and calibrates and normalizes the driver input signals .

[0053] S100c, the perception and data acquisition module 100 adds high-precision timestamps to all pre-processed valid data (e.g. ); the timestamp is derived from the unified system clock of the central control unit, ensuring the alignment of different sensor data in time. This high-precision time synchronization mechanism is the basis for the accurate prediction of the subsequent real-time digital twin module 300 and the accurate backtracking of the data optimization and evolution module 700.

[0054] S100d, the perception and data acquisition module 100 sends the processed and timestamped data to the driver portrait and state recognition module 200 and the real-time digital twin module 300 through the internal data bus or shared memory.

[0055] The driver portrait and state recognition module 200 is responsible for performing the portrait construction and state recognition part in S200; the driver portrait and state recognition module 200 receives real-time data from the perception and data acquisition module 100, and receives model mismatch signals from the real-time digital twin module 300; the functions of the driver portrait and state recognition module 200 can specifically include the following sub-steps:

[0056] S200a, the driver portrait and state recognition module 200 constructs and maintains a personalized portrait parameter set based on the driver's historical operation data and real-time operation data; the portrait parameter is a set of parameters that characterize the driver's specific driving style; in one embodiment, the parameter set can be represented as: ; wherein, characterizes the driver's gain or aggressiveness of driving or braking operation; characterizes the driver's average reaction delay after receiving external stimuli; characterizes the driver's steering operation smoothness or high-frequency noise level; the driver portrait and state recognition module 200 can initialize through statistical analysis of the driver's long-term historical operation data, and iteratively update in real time through online parameter identification algorithms (e.g., recursive least squares); the specific identification algorithm belongs to the known technology in the art, which will not be repeated here.

[0057] S200b, the driver profile and state recognition module 200 continuously receives a model mismatch signal from the real-time digital twin module 300 ; the signal (whose generation is detailed in section 2.3 later) is used to quantify the degree of deviation between the driver's actual operation and the personalized profile parameters currently maintained by the module 200.

[0058] S200c, when the model mismatch signal exceeds a preset threshold in terms of its amplitude or its integral value within a certain time window, indicating that the driver's instantaneous behavior deviates significantly from his regular driving style profile, the driver profile and state recognition module 200 determines that the driver may be in an abnormal state and immediately performs multi-source fusion recognition of the abnormal state.

[0059] S200d, in performing multi-source fusion recognition, the driver profile and state recognition module 200 comprehensively analyzes two types of input data: the first type of input is the driver's direct operation data features from the perception and data acquisition module 100, such as the dithering frequency of steering operation, long-time steering fine-tuning, or the rate of change of acceleration / brake pedal force; the second type of input is the model mismatch signal that triggers recognition in S200b; the classifier (e.g., a pre-trained support vector machine or decision tree) inside the driver profile and state recognition module 200 fuses and analyzes these two types of input to distinguish whether the abnormality is caused by driver fatigue, distraction, or other occasional misoperation.

[0060] S200e, the driver profile and state recognition module 200 generates updated profile parameters according to the recognition result of S200d; if the recognition result is "normal" state (e.g., only occasional misoperation), the driver profile and state recognition module 200 can perform regular adaptive fine-tuning of the profile parameters ; if the recognition result is "fatigue" or "distraction" or "abnormal driving", the driver profile and state recognition module 200 can switch the profile parameters to a preset, corresponding to the abnormal state, safety parameter model (e.g., driver fatigue parameter or driver distraction parameter or abnormal driving parameter ), which usually has a lower degree of aggressiveness or more conservative control characteristics.

[0061] S200f, the driver profile and state recognition module 200 finally updates the profile parameters (whether fine-tuned or switched ​) to the real-time digital twin module 300 and the dynamic coupling decision module 500 for their adjustment of the dynamic model and control decision in the subsequent steps.

[0062] The real-time digital twin module 300 is responsible for performing the model construction and state prediction part in S200; it receives real-time data from the perception and data acquisition module 100 and personalized profile parameters from the driver profile and state recognition module 200 .

[0063] S200g, the real-time digital twin module 300 constructs a driver-vehicle coupling dynamic model based on vehicle dynamics principles and the received profile parameters ; the model is represented in the form of a nonlinear state space equation: ; where: is the state vector of the system, which in one embodiment can include the attitude angle, angular velocity and speed of the vehicle, for example ; is the time derivative of the state vector; is the measurement input vector of the system, which in one embodiment can include the driver operation input from the perception and data acquisition module 100, for example ; is the set of profile parameters from the driver profile and state recognition module 200; is a nonlinear state transition function, which describes the vehicle dynamics. The key is that the mathematical structure or internal parameters of the function are explicitly determined by the profile parameters (i.e. is a function of ); therefore, when is updated from the driver profile and state recognition module 200 (e.g., from to ), the dynamic characteristics of the state transition function also change accordingly, thereby realizing the coupling and online adaptation of the driver-vehicle model; is the process noise vector, which is used to represent the uncertainty of the model.

[0064] S200h, the real-time digital twin module 300 uses the driver-vehicle coupling model to perform real-time state estimation in combination with high-precision timestamp data from the perception and data acquisition module 100; the real-time digital twin module 300 establishes the observation equation: ; where: is the sensor measurement vector at time ; is the true state vector at time ; The observation function describes the mapping relationship between the state vector and the sensor measurements; To measure the noise vector.

[0065] The real-time digital twin module 300 employs a state estimation algorithm, such as Extended Kalman Filter (EKF) or Unscented Kalman Filter (UKF), to estimate the system model in the S200g. Observation model in S200h Combine and integrate Perform recursive calculations to obtain the current... The optimal estimate of the system state at time t The specific implementations of EKF or UKF are well-known technologies in this field and will not be elaborated upon here.

[0066] S200i, the real-time digital twin module 300 estimates the current state based on S200h. The human-vehicle coupling model built in S200g Perform forward prediction; the real-time digital twin module 300 performs forward prediction in the preset future prediction time domain. Inside, for the state equation Perform forward integration or discrete recursion; during the prediction process, for the driver input in the future time domain... Zero-order preservation (i.e., assuming) can be used. (Keep current value) or based on profile parameters Other forecasting strategies.

[0067] S200j, the real-time digital twin module 300, ultimately outputs future... The state prediction sequence of the step is denoted as The future state sequence is sent to the forward-looking emergency control module 400 as the data basis for its forward-looking risk assessment. At the same time, the prediction residuals (new information) generated by the real-time digital twin module 300 during the state estimation process of S200h will be used to generate model mismatch signals and fed back to the driver profile and state identification module 200.

[0068] See attached document Figure 2 Step S200 further includes a twin-portrait bidirectional iterative correction mechanism, which constitutes a closed-loop information interaction between the real-time digital twin module 300 and the driver portrait and state recognition module 200.

[0069] S200k, Generation of Model Mismatch Signal: When the real-time digital twin module 300 executes the state estimation algorithm (e.g., EKF / UKF) of S200h, it calculates the prediction residual (information) in the update step. : ;in: is the predicted residual error vector at time instant t; is the actual sensor measurement vector at time instant t; is the predicted residual error vector at time instant t; is the actual sensor measurement vector at time instant t; is based on the predicted residual error vector at time instant t; is the state estimate at time instant t; is the system model is the predicted measurement at time instant t. S200l, the real-time digital twin module 300 monitors the statistical properties of the predicted residual error

[0070] in real time; in one embodiment, the real-time digital twin module 300 can compute the sliding variance of the residual error or its normalized square where is the covariance matrix of .

[0071] S200m, the real-time digital twin module 300 compares the residual error statistics computed in S200l with a dynamic threshold ; the dynamic threshold is not a fixed value, but a function related to the current vehicle state (e.g. vehicle speed ) and the current driver profile parameters , i.e. .

[0072] S200n, when the residual error statistics continuously exceed the dynamic threshold within a pre-set time window, it indicates that the model inside the real-time digital twin module 300 has failed to accurately describe the current behavior of the driver, the real-time digital twin module 300 determines that a “model mismatch” has occurred, and generates a model mismatch signal ; this signal is sent to the driver profile and state recognition module 200. S200o, online correction of profile parameters: the driver profile and state recognition module 200 receives this

[0073] signal, and uses it as one of the multi-source fusion inputs to recognize abnormal states of the driver (e.g. fatigue, distraction) as in S200d. S200p, after the recognition and correction in S200d and S200e, the driver profile and state recognition module 200 outputs the updated profile parameters

[0074] (e.g. ) as in S200f.

[0075] S200q, the real-time digital twin module 300 receives the updated profile parameters ​​Afterwards, in the next control cycle, it is immediately used to update the human-vehicle coupled dynamics model defined in S200g; specifically, the real-time digital twin module 300 updates its internal state transition function to .

[0076] S200r, through the steps of S200k to S200q, the system realizes online, bidirectional, closed-loop correction between the portrait parameters and the twin model, ensuring that the real-time digital twin module 300 can maintain the accuracy of its prediction even when the driver's state changes.

[0077] The forward-looking emergency control module 400 is responsible for performing S300 steps; the forward-looking emergency control module 400 receives the future state prediction sequence output from the real-time digital twin module 300; including the following sub-steps:

[0078] S300a, the forward-looking emergency control module 400 performs a forward-looking risk assessment on the received future state sequence; in one embodiment, the assessment includes: the forward-looking emergency control module 400 traverses each future predicted state in the sequence (where ), and checks whether the key safety state quantities (e.g., the predicted roll angle and roll angular velocity ) exceed the preset safety boundaries (e.g., , ); the forward-looking emergency control module 400 combines the covariance information obtained in the state estimation (S200h) to assess the probability that the key safety state quantities touch or cross the safety boundaries within the prediction time domain , denoted as ; at the same time, in another embodiment, the forward-looking emergency control module 400 can also combine the obstacle information from other perception systems to calculate the estimated collision time .

[0079] S300b, the forward-looking emergency control module 400 then calculates a continuous real-time risk index based on the evaluation results of S300a; in one embodiment, the index is quantified by the following function form:

[0080] ;

[0081] where: is the real-time risk index; is the future instability probability calculated in S300a; is the estimated collision time, isthe inverse of or a non-linear mapping function of increases as decreases; and are dynamic weight coefficients, which are functions of the driver profile parameters and the current vehicle speed ; for example, when the driver profile is “novice” or the vehicle speed is high, the weight coefficients and increase accordingly to increase the sensitivity of the system to risk.

[0082] S300c, the prospective emergency control module 400 divides the current emergency state level according to the calculated in S300b; this division is achieved by comparing with a set of dynamic thresholds ; in one embodiment, the emergency state level can be divided into four levels (e.g., 1 = normal, 2 = attention, 3 = warning, 4 = emergency), and the division logic is as follows:

[0083] ;

[0084] wherein the thresholds are all components of the set of dynamic thresholds , which are functions of the vehicle speed and the profile parameters ; for example, at high speed or in a “fatigue” state, the thresholds will decrease accordingly, so that the system enters a higher emergency state level earlier.

[0085] S300d, the prospective emergency control module 400 finally outputs the calculated real-time risk index and the emergency state level to the dynamic coupling decision module 500 and the safety cover and underlying execution module 600 as the basis for adjusting the control strategy.

[0086] S300e, in addition, when is greater than or equal to a preset trigger level (e.g., when , i.e., any non-“normal” state), the prospective emergency control module 400 immediately sends a key event trigger signal to the data optimization and evolution module 700 to start the data slicing process in S600.

[0087] Refer to the attached Figure 3, the dynamic coupling decision module 500 is responsible for performing S400; the dynamic coupling decision module 500 receives the real-time risk index and emergency state level from the forward-looking emergency regulation module 400, and receives the current state estimation from the real-time digital twin module 300 and the portrait parameters from the driver portrait and state recognition module 200 ; the dynamic coupling decision module 500 adopts a model predictive control (MPC) algorithm as its core control strategy, and its specific functions can include the following sub-steps:

[0088] S400a, the dynamic coupling decision module 500 first constructs an optimal control problem in a finite time domain at each control period; the basic goal of the problem is to calculate a set of optimal control input sequences under the premise of meeting vehicle physical constraints and safety constraints, so that the state of the vehicle can track the reference trajectory (such as the speed and steering path expected by the driver) and maintain its own balance stability. The optimization problem is usually formalized as minimizing a quadratic cost function .

[0089] S400b, the dynamic coupling decision module 500 dynamically reconstructs the weight matrix in the cost function according to the received real-time risk index; unlike traditional MPC using fixed weights, the cost function in this application explicitly contains a risk index variable, which is expressed as:

[0090] ;

[0091] where: is the prediction time domain length; is the step system state vector predicted at time ; is the reference state vector, usually including the target speed and zero attitude angle (i.e. upright state); is the control input vector of the step; is the dynamic state weight matrix, which is used to punish state deviation; is the dynamic control input weight matrix, which is used to punish control energy or control increment.

[0092] S400c, the specific adjustment logic of the dynamic weight matrix is that the matrix is a diagonal matrix, and the diagonal elements correspond to the punishment weights of each state quantity (such as roll angle , pitch angle , speed ); the dynamic coupling decision module 500 internally stores a weight adjustment function or lookup table, and when the real-time risk index As the elevation increases, the function non-linearly increases the matrix corresponding to the vehicle stability state (i.e. roll angle and pitch angle ), while keeping or relatively reducing the weight element values corresponding to the tracking performance (such as speed ); for example, the change of the weight can be subject to an exponential growth relationship ( with a positive coefficient); through this mechanism, when the system perceives an increased risk, the optimization focus of the controller will automatically shift from "responding to the driver's acceleration instruction" to "maintaining the absolute stability of the vehicle posture", thereby realizing the flexible switching of the control strategy.

[0093] S400d, in addition to the reconstruction of the target function, the dynamic coupling decision module 500 also dynamically tightens the state constraint boundary of the optimization problem according to the real-time risk index ; when solving the MPC problem, it is usually necessary to satisfy the constraint condition ; in the present application, the state constraint boundary (such as the maximum allowed roll angle ) is no longer a physical limit constant, but a function of , denoted as ; ; wherein is the physical limit boundary of the vehicle, is the tightening coefficient; when increases, decreases accordingly, meaning that the controller is forced to find the optimal solution within a narrower safety channel, thereby reserving a larger physical safety margin for potential uncertain disturbances.

[0094] S400e, the dynamic coupling decision module 500 will convert the constructed optimization problem containing dynamic weights , and dynamic constraints into a standard quadratic programming (QP) problem for solving; for the numerical solution of the QP problem, those skilled in the art can use mature algorithms such as the active set method or the interior point method, which will not be described here. After the solution is completed, the dynamic coupling decision module 500 outputs the first item of the optimal control input sequence as the current primary control instruction, and sends it to the safety override and underlying execution module 600.

[0095] The safety override and underlying execution module 600 is responsible for executing S500; the safety override and underlying execution module 600 receives the emergency state level from the prospective emergency regulation module 400, and the primary control instruction .

[0096] S500a, the Security Overlay and Low-Level Execution Module 600 executes the Security Overlay (SOM) arbitration logic in each control cycle; the core function of this arbitration logic is to... The value determines whether to adopt the dynamic optimization instructions of the dynamic coupling decision module 500 or execute the preset emergency safety strategy.

[0097] S500b, in one embodiment, the arbitration logic will... With one or more preset coverage trigger thresholds To achieve this, comparison is used; when Below At times (for example, (At the "Normal" or "Caution" level), the security coverage and underlying execution module 600 determines that the system risk is within the controllable range of the dynamically coupled decision module 500, and therefore selects to adopt the primary control command. And set it as the final control command. ,Right now .

[0098] S500c, when Greater than or equal to At times (for example, When the system is at a "warning" or "emergency" level, and the security coverage and underlying execution module 600 determines that the risks faced by the system have exceeded the scope of normal dynamic adjustment, the security coverage and underlying execution module 600 will ignore or discard the primary control commands from the dynamic coupling decision module 500. .

[0099] In response to the situation in S500c where the emergency status level is determined to be greater than or equal to the coverage trigger threshold, the security coverage and underlying execution module 600 will retrieve the preset emergency control action vector from its internal security storage unit. And set it as the final control command. ,Right now Emergency control action vector It is independent of the dynamic coupling decision module 500, which is pre-calibrated and stored. Its goal is to enable the vehicle to quickly and stably transition to a defined safe state in an emergency. For example, it may include a maximum braking torque command or a specific stability command to maintain self-balancing and upright posture.

[0100] In S500e, the security coverage and low-level execution module 600 is further responsible for distributing low-level control commands; the security coverage and low-level execution module 600 will arbitrate the final control commands determined in S500b or S500d. (This instruction is a high-level logic instruction, such as a target torque value or a target current value) and is sent to the vehicle's underlying actuator.

[0101] S500f, the underlying actuator (e.g., a motor driver) receives this Instructions are given and converted into corresponding physical drive signals; in one embodiment, the motor driver will... The target torque command is converted into the duty cycle of the pulse width modulation signal required to drive the motor; the conversion of this command into a physical signal and the implementation of the drive circuit are well-known technologies in this field and will not be described in detail here.

[0102] See attached document Figure 4 The data optimization and evolution module 700 is responsible for implementing the vehicle-side implementation of step S600; in one embodiment, the data optimization and evolution module 700 is implemented as a vehicle-side edge computing unit with independent data processing and storage capabilities.

[0103] During system operation, the S600a data optimization and evolution module 700 monitors key event trigger signals from the proactive emergency control module 400 in real time. .

[0104] S600b, if and only if the data optimization and evolution module 700 receives When a signal is received (e.g., in Time, corresponding The Data Optimization and Evolution Module 700 immediately executes critical event-driven edge intelligent slicing operations.

[0105] In S600c, during the slicing operation, the data optimization and evolution module 700 accesses its internal circular buffer, which continuously caches internal state and I / O data from all other logic modules within the system; the data optimization and evolution module 700, based on... At any given moment, capture and solidify the preset time window. The full-dimensional data within; including To trace back the time before the event, Record the duration after the event.

[0106] S600d, the full-dimensional data specifically includes, but is not limited to:

[0107] All preprocessed sensor data from the sensing and data acquisition module 100 within this time window;

[0108] Image parameters from the driver profile and status recognition module 200 in this window The evolution trajectory of the data, and the abnormal state identification results output at any given time;

[0109] all future state prediction sequences generated by the real-time digital twin module 300 within this window, as well as model mismatch signals of time-series data;

[0110] real-time risk indices computed by the forward-looking emergency control module 400 within this window and emergency state levels of time-series data;

[0111] primary control instructions solved by the dynamically coupled decision module 500 within this window sequence, as well as its internal dynamic adjustment of weight matrix and constraint boundary change records;

[0112] final control instructions arbitrated by the safety overlay and underlying execution module 600 within this window sequence.

[0113] S600e, the data optimization and evolution module 700 subsequently packages the full-dimension data captured in S600d, generating a high-value emergency data package (HVDP); in the process of generation, the data optimization and evolution module 700 strictly checks and retains the high-precision timestamps of all data (originating from S100c) to ensure the absolute time alignment of all time-series data within the data package.

[0114] S600f, the data optimization and evolution module 700 performs encryption processing (for example, using an asymmetric encryption algorithm) on the generated high-value emergency data package to ensure the security of the data during transmission.

[0115] S600g, the data optimization and evolution module 700 subsequently assigns the encrypted high-value emergency data package the highest upload priority through the on-board communication module (OBU) and sends it to the cloud platform 800 for subsequent offline analysis and strategy evolution.

[0116] The cloud platform 800 is responsible for implementing the cloud-side implementation of S600 steps; the cloud platform 800 receives and decrypts the high-value emergency data package (HVDP) uploaded from the data optimization and evolution module 700.

[0117] S600h, the cloud platform 800 performs deep backtracking and counterfactual simulation using the HVDP; in one embodiment, the cloud platform 800 first performs root cause analysis (RCA) on the full-dimension time-series data in the HVDP; this analysis compares the predicted state of the real-time digital twin module 300 at the time of the key event with the measured state of the perception and data acquisition module 100 the deviation between the actual and the expected value, and the forward emergency control module 400 computing logic to trace back the root cause of the critical event.

[0118] S600i, after completing the root cause analysis, the cloud platform 800 starts a high-precision cloud digital twin model; the cloud model has higher fidelity than the on-board real-time digital twin module 300; the cloud platform 800 takes the sensor input data recorded in the HVDP as the input of the cloud model, and reproduces the critical event scenario.

[0119] S600j, on the basis of reproducing the scenario, the cloud platform 800 performs counterfactual simulation; specifically, the cloud platform 800 systematically modifies a specific parameter in the on-board control strategy and re-runs the simulation; for example, modify a set of coefficient values of the dynamic threshold used to divide the state space in the forward emergency control module 400, or modify the specific function form (for example, the coefficient in S400c) of the dynamic weight matrix of the cost function in the dynamic coupling decision module 500; the cloud platform 800 observes whether the system safety indicators (for example, the maximum roll angle or the safety margin ) obtained by the simulation under the modified parameter are improved.

[0120] S600k, the cloud platform 800 collects a large number of HVDPs and their counterfactual simulation results from the vehicle and other vehicles, and constructs an offline data set; the data set is used to perform policy iteration based on offline reinforcement learning.

[0121] S600l, in one embodiment, the goal of policy iteration is to optimize the dynamic threshold in S300c or the weight function in S400c; the cloud platform 800 parameterizes these threshold values or weight functions; the offline reinforcement learning algorithm (for example, the conservative Q-learning (CQL) algorithm) learns a policy model using the offline data set; the goal of the policy model is to find a new set of parameters (for example or ) that maximizes the simulation safety indicator (as the reward function ) defined in S600j, given the (state, action) distribution represented by the HVDP; for the specific implementation of the offline reinforcement learning algorithm, it belongs to the public technical knowledge in the field, and will not be repeated here.

[0122] ​​The S600m cloud platform 800 will use the optimal policy model trained and validated in the S600l (i.e., a set of updated...) or The parameter list or function is packaged into a policy update package.

[0123] The S600n cloud platform 800 pushes the policy update package to the vehicle-mounted data optimization and evolution module 700 via the vehicle-to-everything (V2X) communication module through the over-the-air (OTA) communication link.

[0124] S600o, after receiving and verifying the strategy update package, the data optimization and evolution module 700 is responsible for writing the updated parameters into the corresponding storage area of ​​the forward-looking emergency control module 400 or the dynamic coupling decision module 500 to replace the original control logic.

[0125] Through the steps from S600a to S600o, the system forms a closed loop of vehicle-cloud collaboration, from edge-triggered operation in the vehicle to offline optimization in the cloud and OTA updates in the vehicle, thus realizing the continuous evolution of the system's safety performance.

[0126] A specific application embodiment of the present invention demonstrates how the system, through the collaborative work of its modules, achieves a complete closed loop from personalized adaptation to proactive emergency response in a continuous dynamic scenario containing multiple risks.

[0127] In this scenario, suppose a driver starts a two-wheeled self-balancing vehicle; during the initial stage of operation, the sensing and data acquisition module 100 collects the driver's operational data. The driver profile and status recognition module 200 analyzes the data and constructs personalized profile parameters based on S200a. (For example, characterized as a "radical" style); the real-time digital twin module 300 adopts this... Parameter construction of human-vehicle coupling model At this point, the vehicle control response is quite sensitive, meeting the driver's expectations.

[0128] Subsequently, the vehicle enters a complex urban road section containing a series of manhole cover depressions or speed bumps; traditional reactive controllers would experience severe turbulence here due to lag. In this invention, the real-time digital twin module 300 (which can incorporate high-precision map data) anticipates the impending turbulence in its future state prediction sequence; the forward-looking emergency control module 400 analyzes this sequence based on S300a, determines that there is a short-term instability risk, and then calculates the instantaneously increased risk index in S300b. The dynamic coupling decision module 500, based on S400c, immediately and dynamically increases the cost function. The weights corresponding to attitude stability It outputs the optimal control command in advance, so that the motor torque is adjusted before the vehicle reaches the depression. This control command works in conjunction with the self-balancing module to intervene in stability control in advance and smoothly pass through the area.

[0129] Next, the vehicle enters a medium-to-high-speed section (such as a bridge, at a speed of...). (Increased to 60km / h); at this time, the vehicle encountered a sudden crosswind disturbance; the forward-looking emergency control module 400 in the S300c, due to the high vehicle speed... and portrait (Radical style), its dynamic threshold The level was already low; sudden crosswind disturbances increased the predicted probability of instability. The dramatic increase made Rapidly exceeding preset thresholds; dynamically coupled decision module 500 The weights were significantly increased, and the state constraints in S400d (e.g.) The system is tightened; the safety coverage and underlying execution module 600 may trigger S500d at this time, sending an enhanced stability control command to the self-balancing module; the system control switches from responding to the driver's steering mode to the self-balancing mode that forcibly maintains stability, effectively suppressing the roll attitude deviation caused by high-speed crosswinds.

[0130] After driving on this medium-high speed section for a period of time, the driver became fatigued, and his steering operation... Slight, abnormal jitter began to appear; within the real-time digital twin module 300 The model is built based on a "normal" profile and cannot predict this abnormal jitter; therefore, the prediction residuals calculated in S200k... (i.e., the difference between model prediction and actual jitter) began to consistently exceed the dynamic threshold in S200m. The real-time digital twin module 300 then sends a model mismatch signal to the driver profile and state recognition module 200. (S200n).

[0131] Driver profile and status recognition module 200 received Based on the signal and the steering vibration characteristics fused from the sensing and data acquisition module 100, the S200d determines that the driver's state has switched to "fatigue"; the driver profile and state recognition module 200 immediately sends the profile parameters Revised to It is then broadcast to the real-time digital twin module 300 and the dynamic coupling decision module 500 (S200q).

[0132] Real-time digital twin module 300 immediately adopted Update it The model recovers the prediction accuracy; meanwhile, the forward-looking emergency control module 400 receives the parameters, whose dynamic thresholds are greatly reduced (i.e. more sensitive to risks); the dynamic coupling decision module 500 also receives the and thus the elevated , which further increases the stable weights in the matrix, causing the vehicle control style to automatically switch to the "conservative" and "high stability" modes to compensate for the risks brought by the driver fatigue.

[0133] At the same time, the signal generated by the forward-looking emergency control module 400 triggers the data optimization and evolution module 700, which captures the high-value emergency data packets before and after the "fatigue driving" event according to S600c and uploads them to the cloud platform 800. The cloud platform 800 performs offline reinforcement learning on this event according to S600l, optimizes the threshold parameters, and pushes them to the vehicle through OTA updates of S600n, so that the system's recognition of the fatigue state of this type of driver and the coping strategies evolve in the future.

Claims

1. A digital twin-driven adaptive intelligent control system for a two-wheeled self-balancing vehicle, characterized in that, include: The sensing and data acquisition module is used to collect real-time operating data of the two-wheeled self-balancing vehicle; The driver profiling and status recognition module is used to construct personalized style profile parameters for the driver based on the real-time operation data. The real-time digital twin module is used to construct a human-vehicle coupled dynamic model based on the vehicle dynamics model and combined with the driver's personalized style profile parameters, and to predict the future state sequence of the vehicle. The forward-looking emergency control module is used to analyze the future state sequence, calculate the real-time risk index, and classify emergency state levels. The dynamic coupling decision module is used to adjust the control optimization objective and control constraints in real time based on the real-time risk index and the emergency state level, and solve to generate the optimal control command sequence. The security coverage and underlying execution module is used to perform security arbitration on the optimal control command sequence based on the emergency status level, and determine the final control command; The data optimization and evolution module is used to capture and package high-value emergency data packets in response to key event triggering signals from the forward-looking emergency control module. The cloud platform is used to receive the high-value emergency data packets and generate policy update packets; The real-time digital twin module is also used to generate a model mismatch signal and send it to the driver profile and state identification module. The driver profile and state identification module is used to receive the model mismatch signal and correct the driver profile parameters. The corrected driver profile parameters are fed back to the real-time digital twin module to correct the dynamic model of human-vehicle coupling. The data optimization and evolution module is also used to receive the strategy update package generated by the cloud platform and update the control logic of the vehicle system.

2. The adaptive intelligent control system for a two-wheeled self-balancing vehicle driven by a digital twin according to claim 1, characterized in that, The real-time digital twin module is specifically used for: The predicted measurement values ​​based on the human-vehicle coupling dynamic model are compared with the measured data from the sensing and data acquisition module to generate prediction residuals, and the model mismatch signal is generated based on the statistical characteristics of the prediction residuals.

3. The adaptive intelligent control system for a two-wheeled self-balancing vehicle driven by a digital twin according to claim 1, characterized in that, The driver profiling and status recognition module is specifically used for: Upon receiving a model mismatch signal, the driver's direct operation data features from the perception and data acquisition module are fused to identify the driver's fatigue or distraction, dangerous driving behavior, or abnormal acceleration or deceleration behavior, and the driver profile parameters are corrected accordingly.

4. The adaptive intelligent control system for a two-wheeled self-balancing vehicle driven by a digital twin according to claim 1, characterized in that, The aforementioned forward-looking emergency control module is specifically used for: Based on the driver profile parameters and the current vehicle speed, the weighting coefficients used to calculate the real-time risk index are dynamically adjusted, as are a set of dynamic thresholds used to classify emergency status levels.

5. The adaptive intelligent control system for a two-wheeled self-balancing vehicle driven by a digital twin according to claim 1, characterized in that, The dynamically coupled decision module is implemented as a model prediction controller, and the dynamically coupled decision module is specifically used for: When the real-time risk index increases, the state weights corresponding to vehicle attitude stability in the internal cost function are nonlinearly increased to adjust the control optimization objective.

6. The adaptive intelligent control system for a two-wheeled self-balancing vehicle driven by a digital twin according to claim 1, characterized in that, The dynamic coupling decision module is also specifically used for: When the real-time risk index increases, the state constraint boundary in the internal optimization problem is dynamically tightened to adjust the control constraints.

7. The adaptive intelligent control system for a two-wheeled self-balancing vehicle driven by a digital twin according to claim 1, characterized in that, The security overlay and underlying execution module is specifically used for: When the emergency state level is lower than the coverage trigger threshold, the optimal control command sequence shall be adopted. When the emergency status level is greater than or equal to the coverage trigger threshold, the optimal control instruction sequence is ignored, and a preset emergency control action vector is retrieved as the final control instruction.

8. The adaptive intelligent control system for a two-wheeled self-balancing vehicle driven by a digital twin according to claim 1, characterized in that, The data optimization and evolution module is specifically used for: Access the internal circular buffer and extract full-dimensional data within a preset time window before and after the time corresponding to the trigger signal of the key event; The full-dimensional data includes internal state data and I / O data from the perception and data acquisition module to the security coverage and underlying execution module, and retains high-precision timestamps.

9. The adaptive intelligent control system for a two-wheeled self-balancing vehicle driven by a digital twin according to claim 1, characterized in that, The cloud platform is specifically used for: Utilize the aforementioned high-value emergency data package to perform deep backtracking and counterfactual simulation; Using the simulation results and offline reinforcement learning algorithm, the dynamic threshold of the forward-looking emergency control module or the dynamic weight function of the dynamically coupled decision module is iteratively optimized to generate the policy update package.

10. The adaptive intelligent control system for a two-wheeled self-balancing vehicle driven by a digital twin according to claim 1, characterized in that, The dynamic model of the human-vehicle coupling is a state transition function. The driver's personalized style profile parameters explicitly determine the dynamic characteristics of the state transition function, so that when the driver profile parameters are updated, the dynamic characteristics of the state transition function change accordingly.

Images