Vehicle adaptive conditioning system, method, and vehicle

Abstract

The application discloses a kind of vehicle adaptive adjustment system, method and vehicle, including user identity identification module, for identifying for identity and outputting user identification;Cabin adjustment module, for based on the user identification call personalized configuration file, control the execution device in cabin completes automatic adjustment;Multi-source information acquisition module, for real-time acquisition multi-source information in vehicle driving process;Driver behavior and intention identification module, for based on the operation characteristics of time sequence identifies driving style and / or predicts future driving operation intention;Control module, for based on the user identification, multi-source information, driving style, driving operation intention generates adaptive adjustment instruction for each subsystem of vehicle;Execution module, for executing the adaptive adjustment instruction to each subsystem of vehicle is adaptively adjusted.The application realizes the vehicle adaptive adjustment based on user identity and driving information, so as to improve personalized experience and driving comfort.

Description

Technical Field

[0001] This invention belongs to the field of vehicle control technology, specifically relating to a vehicle adaptive adjustment system, method, and vehicle. Background Technology

[0002] In current vehicle technology, drivers typically adjust the seat, rearview mirrors, and steering wheel manually before starting the vehicle to accommodate individual differences. This process is not only time-consuming but also lacks the ability to automatically recognize and configure based on user identity, resulting in insufficient personalized experience. During vehicle operation, the parameters of subsystems such as suspension, power steering, power output, and braking response are mostly fixed settings or can only be manually switched through preset modes. They cannot dynamically and adaptively adjust based on real-time multi-source data, making it difficult to achieve an optimal balance between comfort, stability, economy, and safety under different road conditions.

[0003] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of the aforementioned background technology and provide a vehicle adaptive adjustment system, method, and vehicle that can automatically adjust the cockpit based on user identity and dynamically adjust each subsystem according to multi-source information and driving intentions during vehicle operation.

[0005] The technical solution adopted in this invention is: a vehicle adaptive adjustment system, including a user identity recognition module for recognizing user identity and outputting user identifier; The cockpit adjustment module is used to retrieve a personalized configuration file based on the user identifier and control the actuators in the cockpit to complete automatic adjustment. The multi-source information acquisition module is used to collect multi-source information, including vehicle dynamic parameters, environmental perception information and positioning and navigation information, in real time during vehicle operation. The driver behavior and intent recognition module is used to identify driving style and / or predict future driving intentions based on time-series operational features. The control module is used to generate adaptive adjustment commands for each subsystem of the vehicle based on the user identifier, multi-source information, driving style, and driving operation intention using a hierarchical control architecture. The execution module includes actuators corresponding to each subsystem of the vehicle, used to execute the adaptive adjustment command to adaptively adjust each subsystem of the vehicle.

[0006] A vehicle adaptive adjustment method based on the above-mentioned vehicle adaptive adjustment system includes the following steps: When a user approaches and enters the vehicle, the system identifies the user's identity through multimodal fusion and automatically adjusts the personalized cockpit configuration based on the identified user identifier. During vehicle operation, multi-source information, including vehicle dynamic parameters, environmental perception information and positioning and navigation information, is periodically collected, and driving style and / or future driving intentions are identified based on time-series operation characteristics. Based on the user identifier, multi-source information, driving style, and driving operation intention, a hierarchical control architecture is used to generate adaptive adjustment commands for each subsystem of the vehicle. The vehicle's various subsystems are controlled to perform adaptive adjustments based on adaptive adjustment commands.

[0007] A vehicle comprising the vehicle adaptive adjustment system described above.

[0008] The beneficial effects of this invention are as follows: This invention achieves adaptive vehicle adjustment based on user identity and driving information through a user identity recognition module, a cockpit adjustment module, a multi-source information acquisition module, a driver behavior and intention recognition module, a control module, and an execution module. Specifically, it can realize automatic cockpit adjustment based on user identity and dynamically adjust each subsystem according to multi-source information and driving intention during vehicle operation, thereby improving personalized experience, driving comfort, stability, economy, and safety. Attached Figure Description

[0009] Figure 1 This is a system principle block diagram of the present invention. Detailed Implementation

[0010] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments to facilitate a clear understanding of the present invention, but these descriptions do not constitute a limitation on the present invention.

[0011] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0012] Furthermore, references to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in still other embodiments" appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0013] like Figure 1As shown, this application proposes a vehicle adaptive adjustment system, including... The user identity recognition module is used to identify the user's identity and output the user identifier. The cockpit adjustment module is used to retrieve a personalized configuration file based on the user identifier and control the actuators in the cockpit to complete automatic adjustment. The multi-source information acquisition module is used to collect multi-source information, including vehicle dynamic parameters, environmental perception information and positioning and navigation information, in real time during vehicle operation. The driver behavior and intent recognition module is used to identify driving style and / or predict future driving intentions based on time-series operational features. The control module is used to generate adaptive adjustment commands for each subsystem of the vehicle based on the user identifier, multi-source information, driving style, and driving operation intention using a hierarchical control architecture. The execution module includes actuators corresponding to each subsystem of the vehicle, used to execute the adaptive adjustment command to adaptively adjust each subsystem of the vehicle.

[0014] This embodiment provides a vehicle adaptive adjustment system that integrates multiple functional modules to achieve intelligent, personalized, and adaptive adjustment of the vehicle.

[0015] Specifically, the user identification module is responsible for determining the identity of the current vehicle user. This module verifies identity by collecting the user's biometric features or information from their carried devices, generating a unique user identifier. This identifier forms the basis for subsequent personalized configurations and adaptive adjustments. The user identification module can employ various methods to identify users. For example, it can use traditional key or remote control-based identification methods. When a user approaches the vehicle with a key, the system wirelessly identifies the key's unique code, thus confirming the user's identity. Alternatively, it can use manual input via in-vehicle buttons or a touchscreen. After entering the vehicle, the user completes identification by entering a preset password or selecting a user profile.

[0016] The cockpit adjustment module retrieves user preferences from a pre-set personalized configuration profile based on the identified user. This module then controls various actuators within the cockpit, such as the seat, steering wheel, rearview mirrors, and air conditioning, automatically adjusting them to the user's preferred settings to provide a personalized driving environment. The cockpit adjustment module employs a principle of prioritizing preset values, correcting with learned values, and constraining safety boundaries. Preset values ​​are stored in the user profile, learned values ​​are dynamically updated by the self-learning module, and safety boundaries ensure that adjustments do not obstruct the view or affect the operating space (e.g., the lowest seat position is not lower than the design baseline).

[0017] The specific adjustment objects and actuators of the cockpit adjustment module include: Seat: 12-way power adjustable mechanism (forward and backward, height, backrest angle, seat cushion tilt, lumbar support airbag, headrest), with an adjustment accuracy of 1mm or 0.5°.

[0018] Rearview mirrors: External rearview mirror horizontal / vertical angle adjustment motor, internal rearview mirror automatic anti-glare ECU.

[0019] Steering wheel: Electric column adjustment motor (front / back / up / down), steering wheel heating controller.

[0020] Air conditioning system: HVAC controller (temperature, air volume, mode, zoning), fragrance system, seat ventilation / heating controller.

[0021] Human-computer interaction system: instrument and HUD display modes, central control theme, audio DSP parameters, and voice assistant configuration.

[0022] The multi-source information acquisition module is responsible for continuously collecting information from various sensors and systems during vehicle operation. This information includes the vehicle's own dynamic parameters (such as vehicle speed and acceleration), environmental perception information surrounding the vehicle (such as road conditions and obstacles), and the vehicle's positioning and navigation information (such as current location and driving route), providing comprehensive real-time data support for the vehicle's adaptive adjustments. Specifically, the multi-source information acquisition module may include: Vehicle dynamic sensor group: six-axis IMU (sampling rate 100Hz, range ±8g), four-wheel wheel speed sensor, steering wheel angle sensor (accuracy ±0.1°), brake pressure sensor, suspension displacement sensor (travel 0~200mm).

[0023] Environmental perception unit: forward-facing three-lens camera (120° field of view, 150m detection range), four surround-view cameras, five millimeter-wave radars (forward long-range radar with a detection range of 200m, corner radar with a detection range of 80m), lidar (optional, 128 lines, 200m detection range), external temperature and humidity sensor, and rain and light sensor.

[0024] Positioning and navigation unit: dual-frequency RTK-GPS / BeiDou module (positioning accuracy ≤2cm), inertial navigation unit (deadline calculation), high-precision map engine (including lane-level geometry, slope, curvature, speed limit, road surface material, and traffic signs).

[0025] The driver behavior and intention recognition module analyzes the driver's operational behavior patterns. By processing the driver's operational characteristics over a period of time, such as accelerator, brake, and steering, this module identifies the driver's driving style (e.g., economy, comfort) and predicts their possible future driving intentions (e.g., acceleration, braking, lane change), thus providing a basis for predictive adjustments to the vehicle.

[0026] The core function of the control module is to integrate various data from user identity, multi-source information, driving style, and driving intentions. Based on this data, it generates adaptive adjustment commands for each subsystem of the vehicle through a hierarchical control architecture. These commands aim to optimize the overall performance of the vehicle to adapt to different driving scenarios and user needs. For example, the module can determine the direction and magnitude of adjustment for parameters such as the damping coefficient and vehicle height of the suspension system based on the current vehicle speed, road conditions, and the driver's driving style, using a preset lookup table or simple logical rules. Simultaneously, it adjusts the sensitivity of power output or braking response in conjunction with the driver's acceleration or braking intentions.

[0027] The execution module includes actuators (electronic control units or area controllers) corresponding to various vehicle subsystems. This module receives adaptive adjustment commands from the control module and drives the corresponding actuators to precisely adjust the vehicle's suspension, steering, powertrain, braking, and other subsystems to achieve adaptive performance optimization. For example, for the suspension system, commands can be sent to solenoid valves or motors to change the damping characteristics of shock absorbers or adjust the height of air springs. For the steering system, commands can be sent to the power steering motor to adjust the strength of the steering assist. For the powertrain system, commands can be sent to the engine control unit or transmission control unit to adjust the engine's torque output characteristics or the transmission's shift strategy. Specifically, the execution module may include a suspension controller (CDC valve actuation, air pump control), a steering controller (EPS motor actuation), a powertrain controller (VCU, MCU, TCU), a braking controller (ESP, iBooster), and a cockpit domain controller (seats, air conditioning, HUD, etc.). After receiving commands from the control module, each actuator performs low-level closed-loop control (e.g., using PID regulation within the suspension controller to achieve the target damping force) and reports the actual execution status to the control module.

[0028] The vehicle adaptive adjustment system in this embodiment achieves fully automatic, personalized, and dynamic adaptive adjustment of the vehicle before startup and during driving through user identification, multi-source information collection, driver behavior and intention recognition, and a hierarchical control architecture. This system can automatically configure the cabin environment based on the user's identity and preferences, and dynamically adjust the parameters of various vehicle subsystems based on real-time road conditions, driving behavior, and environmental changes during vehicle operation, thereby improving driving comfort, stability, fuel economy, and safety under different operating conditions.

[0029] In one embodiment, this application further proposes a user identification module, which includes an external identification unit, an internal identification unit, and a fusion decision unit.

[0030] The external identification unit integrates a BLE Bluetooth module, a UWB ultra-wideband positioning module, and an NFC near-field communication module. When a user approaches the vehicle carrying a bound device (such as a mobile phone or smart key), the external identification unit can jointly calculate the user's position, distance, and movement trajectory relative to the vehicle based on the angle of arrival (AoA) and time-of-flight (ToF) technology of the UWB ultra-wideband positioning module. When the system detects that the user has entered a preset identification area (e.g., within a certain range of the car door and moving towards the vehicle), it will actively trigger the identity recognition process and automatically unlock the car door. The BLE Bluetooth module can be used for short-range device discovery and initial communication, while the NFC near-field communication module can serve as a backup or auxiliary short-range identification method to meet specific scenario requirements.

[0031] The internal recognition unit includes an infrared binocular camera (supporting near-infrared active light source, with liveness detection capability, and resistant to photo and video attacks) deployed inside the A-pillar, a microphone array (supporting voiceprint feature extraction and noise suppression), and a capacitive fingerprint sensor in the steering wheel grip area. After a user enters the vehicle, the infrared binocular camera captures the user's facial biometrics. The microphone array captures the user's voiceprint biometrics, improving recognition accuracy through sound source localization and noise suppression technology. The capacitive fingerprint sensor captures the user's fingerprint biometrics and is typically integrated into easily accessible locations such as the steering wheel, start button, or center console.

[0032] The fusion decision unit is used to perform fusion calculations on the recognition results from the external and internal recognition units. This unit can use DS evidence theory to comprehensively analyze input information from different sensors and recognition technologies to determine a fused identity confidence level. When this identity confidence level exceeds a preset threshold, the fusion decision unit outputs the final user identifier for subsequent system use. For example, the confidence level for face recognition can be set as α, the confidence level for voiceprint recognition as β, the confidence level for external device identification as γ, and the confidence level for fingerprint recognition (if used) as δ. The formula for calculating the fused identity confidence level is: ; Among them, P i w represents the recognition probability for each channel. i Channel weights are dynamically adjusted (e.g., reducing face weight in low light). When the identity confidence level C > 0.85, the final user ID is output, and the personalized configuration file stored on the local security chip or in the cloud is retrieved.

[0033] Through the above technical solution, this application achieves multimodal, multi-stage user identification. The external identification unit begins contactless identification and door unlocking as the user approaches the vehicle, improving the convenience and experience of entering the vehicle. The internal identification unit, after the user enters the vehicle, utilizes multiple biometric features such as face, voiceprint, and fingerprint for high-precision, high-security identity verification. The fusion decision unit comprehensively judges the external and internal identification results, improving the accuracy, robustness, and security of identification, and reducing the false recognition rate. This high-precision, high-efficiency identification provides the cockpit adjustment module with an accurate user identifier, ensuring the accuracy of subsequent personalized configuration files retrieved based on the user identifier. It also provides a reliable basic input for the control module to generate adaptive adjustment commands for various vehicle subsystems, enabling the entire vehicle adaptive adjustment system to truly achieve personalized services, thereby improving the user experience and the vehicle's intelligence level.

[0034] In one embodiment, this application further proposes a driver behavior and intent recognition module based on a temporal behavior model constructed using a long short-term memory network. The input feature vector of this temporal behavior model includes: accelerator pedal opening (0~100%) and its first-order difference, brake pedal opening and its first-order difference, steering wheel angle and angular velocity, vehicle yaw rate, longitudinal acceleration, lateral acceleration, vehicle speed, turn signal status, and lane departure warning trigger status. The time window length is 3 seconds, and the sliding step is 100ms. The output is a driving style classification and intent prediction. The driving style classification maps driving behavior into three styles—economy, comfort, and sport—using K-means clustering, with cluster centers continuously updated from cloud-based group data. The intent prediction is a probability vector of operations such as acceleration, braking, left lane change, right lane change, and steering within the next 200ms, using softmax output, with a threshold of 0.85 triggering pre-adjustment.

[0035] Specifically, Long Short-Term Memory (LSTM) networks are a special type of recurrent neural network (RNN) designed to effectively process and learn long-term dependencies in time-series data. By introducing gating mechanisms (such as input gates, forget gates, and output gates), LSTM networks can selectively remember or forget information, thereby more accurately capturing the complex patterns and underlying intentions of drivers during continuous operations. This temporal behavior model uses multi-dimensional vehicle operation and dynamic parameters within a preset time window as input feature vectors. By continuously collecting and combining these features within the preset time window, the model can obtain sufficient temporal context information to comprehensively characterize the driver's immediate operational behavior and the vehicle's dynamic response. The output driving style classification aims to identify the driver's long-term driving preferences, helping the system to make more personalized long-term adjustments; the prediction of operational intentions provides a forward-looking decision-making basis for various vehicle subsystems.

[0036] Through the above technical solution, this application adopts a temporal behavior model based on a long short-term memory network, which can effectively handle the inherent temporal dependence and complexity in driver operation data. This forward-looking and high-precision driver behavior and intention recognition capability enables the control module to generate adaptive adjustment commands more timely and accurately, improving the response speed of the adjustment of various vehicle subsystems and the matching degree with the driver's intention, thereby optimizing the driving experience and driving safety.

[0037] In one embodiment, this application proposes a hierarchical control architecture, the core idea of ​​which is to decompose complex control tasks into multiple interconnected but clearly defined layers. This architecture can effectively manage system complexity, improve system modularity and maintainability, and allow the most suitable control strategies to be adopted at different layers. In a vehicle adaptive control system, the hierarchical control architecture can decouple high-level decision-making from low-level execution, enabling efficient, flexible, intelligent, safe, and highly adaptive vehicle control when facing complex and ever-changing driving environments and user needs, thereby improving the driving experience and vehicle performance.

[0038] This hierarchical control architecture consists of an upper decision-making layer, a middle coordination layer, and a lower execution layer.

[0039] The upper decision layer is the highest layer of the hierarchical control architecture. Its main function is strategy planning, that is, determining the combination of target control parameters that each subsystem of the vehicle needs to achieve based on the current state of the system. This layer receives a state vector containing user identification, multi-source information, driving style, and driving intentions as input. The state vector integrates key information such as user personalized needs, vehicle operating environment, driver subjective preferences, and future behavior predictions. The upper decision layer adopts a fusion algorithm of deep reinforcement learning (Deep Q-Network, DQN) and fuzzy logic. The deep reinforcement learning algorithm can autonomously learn the optimal decision strategy under different states through interaction with the environment to maximize long-term returns, thereby adapting to complex and ever-changing driving scenarios and user needs. The fuzzy logic fusion algorithm can handle the uncertainty and fuzziness in the input information, integrating expert experience and rules into the decision-making process, making the decision results more consistent with human intuitive judgment, and effectively integrating information from different sources. Through the fusion of these two algorithms, the upper decision layer can generate an intelligent combination of target control parameters that meets personalized needs while taking into account the environment and driving intentions.

[0040] The state vector S is defined as: ; The definitions and acquisition methods of each parameter are as follows: This application defines a comprehensive and fine-grained state vector as the input to the upper-level decision layer. This multi-dimensional, high-granularity information input enables the deep reinforcement learning and fuzzy logic fusion algorithm to more accurately understand the current complex driving scenario, predict the driver's potential operational intentions, and fully consider the driver's personalized preferences. Therefore, the generated adaptive adjustment commands can more accurately match actual needs, improving the intelligence, personalization, and safety of the vehicle's adaptive adjustment system, thereby providing the driver with a more comfortable, safer, and more habit-friendly driving experience.

[0041] The target control parameter combination is defined as follows: A=[c_damp,h_height,assist_idx,return_coeff,pedal_map,regen_level,temp_offset,bolster_coeff]; The definitions and adjustment ranges of each parameter are as follows: This application, by explicitly defining the combination of target control parameters, enables the upper-level decision-making layer to output specific and quantifiable adjustment instructions, thereby forming a comprehensive and refined adaptive adjustment framework. Upon receiving these specific target parameters, the middle-level coordination layer can more accurately apply model predictive control algorithms, decomposing them into specific control quantity sequences for each actuator while satisfying the optimization objective function and safety constraints. This not only ensures that each vehicle subsystem can perform coordinated and precise adaptive adjustments based on user identification, multi-source information, driving style, and driving intentions, but also improves the system's response speed and adjustment accuracy, thus providing drivers and passengers with a highly personalized, comfortable, and safe driving experience.

[0042] Deep reinforcement learning (DQN) is trained using a reward function that combines four metrics: ; The definitions of each sub-item are as follows: The initial values ​​of weights w1 to w4 are determined by calibration and then dynamically optimized through cloud-based group learning.

[0043] The middle coordination layer, located between the upper decision-making layer and the lower execution layer, is primarily responsible for decomposing the target control parameters determined by the upper decision-making layer into a sequence of specific control quantities for each actuator, while satisfying the preset optimization objective function and safety constraints. This layer typically employs model predictive control (MMC) algorithms. MMC algorithms establish a vehicle dynamics model, predict the vehicle's dynamic behavior over a future period within each control cycle, and optimize the control quantity sequence based on this prediction to ensure that the vehicle's actual state is as close as possible to the target control parameters, while strictly adhering to various physical limits and safety boundaries. For example, when decomposing the target control parameters, the middle coordination layer considers the interactions between subsystems such as suspension damping, vehicle height, and steering assist, ensuring that each subsystem works collaboratively, avoiding conflicts, and guaranteeing the vehicle's driving stability, comfort, and safety.

[0044] When solving the problem using the model predictive control algorithm, the objective function and constraints are transformed into a quadratic programming (QP) problem, and the interior point method is used to solve it in real time (single solution time ≤ 5ms), outputting the optimal control sequence.

[0045] The objective function to be optimized is: ; The parameters are defined as follows: This application minimizes the objective function, which includes a weighted sum of the deviations between the predictive and target control parameters, the control quantity, and the rate of change of the control quantity. This allows the control module to comprehensively consider multiple performance indicators such as tracking accuracy, control stability, and control energy consumption. This explicit and tunable objective function design ensures that the model predictive control algorithm can efficiently combine the target control parameters from the upper decision layer and, under the premise of satisfying safety constraints, decompose them into optimal and stable sequences of specific control quantities for each actuator. This enables precise, stable, and adaptive adjustment of each vehicle subsystem, improving the overall driving experience and ride comfort.

[0046] Security constraints include: Actuator physical limiting constraints are mainly used to limit the physical output range of actuators or vehicle components to prevent overload or damage: limit the maximum current of damping valves, vehicle height adjustment rate ≤5mm / s, and maximum torque of steering motors, etc.

[0047] Safety adjustment constraints focus on ensuring vehicle stability during dynamic driving and preventing loss of control: yaw rate does not exceed the adhesion limit, and roll angle does not exceed 6°.

[0048] This application introduces physical amplitude limiting constraints and safety adjustment constraints into the model predictive control algorithm of the middle coordination layer, which can effectively prevent control commands from exceeding the physical tolerance of each vehicle subsystem. This not only protects vehicle components and improves the reliability and durability of the system, but also greatly enhances the safety and comfort of the driver and passengers, ensuring that the adaptive control system prioritizes vehicle operation safety while pursuing performance optimization.

[0049] The lower execution layer is the bottom layer of the hierarchical control architecture. Its main function is to transmit the specific control quantity sequences generated by the middle coordination layer to the corresponding actuators in the execution module via the vehicle communication network (CAN FD bus (2Mbps rate) or vehicle Ethernet (100BASE-T1)). The instruction format adopts the AUTOSAR standard PDU structure, which includes the target value, timestamp, and checksum. This layer is responsible for translating abstract control instructions into actual physical actions. The lower execution layer is usually not involved in complex decision-making or optimization, but focuses on the precise execution and feedback of instructions, ensuring that the vehicle's subsystems can accurately respond to the upper-level instructions and complete adaptive adjustments.

[0050] Each actuator ECU also periodically feeds back the actual execution status (such as actual damping valve current, actual vehicle height, and actual seat position) to the upper decision-making layer via the bus, forming a closed loop. The upper decision-making layer compares the feedback value with the target value. If the deviation exceeds the threshold (such as damping force deviation > 5%), it will be corrected through MPC in the next control cycle.

[0051] In one embodiment, this application further proposes a self-learning and cloud-based collaborative module for recording user intervention behaviors and scenarios, updating personalized configuration profiles based on the recorded information, and interacting with the cloud to achieve co-evolution of the group model. This self-learning and cloud-based collaborative module aims to achieve continuous optimization and intelligent evolution of the vehicle adaptive adjustment system, including a local learning unit and a cloud-based collaborative unit.

[0052] The local learning unit records the current scenario and the user's intervention when it detects manual intervention by the user in response to the automatic adjustment results. This forms a scenario-operation sample pair, which is then updated in the locally stored personalized configuration profile using an incremental learning algorithm. Specifically, when the cockpit adjustment module or execution module completes the adjustment according to the system-generated adaptive adjustment instructions, if the user is dissatisfied with the result and makes manual adjustments (e.g., manually raising or lowering the air conditioning temperature, adjusting the seat position, changing the suspension mode, etc.), the local learning unit immediately captures this intervention. At this time, the system records the vehicle state at the time of the intervention (including multi-source information, driving style, driving intentions, etc.) as the scenario, and the user's manual adjustment as the operation, thus forming a scenario-operation sample pair. To efficiently update the personalized configuration profile, the local learning unit employs an incremental learning algorithm. This algorithm allows the system to adapt to new user preferences by making small, local adjustments to the existing model parameters when it receives a new scenario-operation sample pair and the number of user interventions in the same scenario exceeds a threshold (e.g., 3 times).

[0053] The cloud-based collaborative unit uploads the de-identified scenario-operation samples to the cloud server for training a swarm intelligence model based on a federated learning framework, and receives optimized updates to the base model parameters from the cloud server. To protect user privacy, the scenario-operation samples generated by the local learning unit undergo rigorous de-identification before being uploaded to the cloud server. De-identification can include anonymization (removing or replacing user identity information), data aggregation (statistically summarizing samples from multiple users instead of uploading the original data), or applying differential privacy techniques (adding noise to the data to obscure individual information). The de-identified data (including vehicle operation data, user intervention records, and user satisfaction feedback such as voice evaluations) is uploaded to the cloud server via an encrypted communication channel. In the cloud, these de-identified samples participate in the training of a swarm intelligence model based on a federated learning framework. Specifically, the cloud aggregates model gradient updates from millions of vehicles to form a base model (such as the optimal suspension parameter mapping for a specific vehicle model under different road conditions), which is then distributed to vehicles via OTA (Over-The-Air) updates to achieve swarm experience sharing. Federated learning is a distributed machine learning paradigm. Its core idea is that each vehicle's local model is trained locally, with only the model parameters (not the raw data) uploaded to a cloud server for aggregation. This allows for the construction of a more powerful and generalized swarm intelligence model by leveraging the learning experiences of a large number of users without disclosing the users' original data. In this way, cloud-based collaborative units can aggregate personalized learning experiences from different users to form a shared base model that captures a wider range of user preferences and driving habits. Subsequently, the cloud server updates the optimized base model parameters and distributes them to the cloud-based collaborative units of each vehicle. Upon receiving these updates, the vehicles can integrate them into their local personalized configuration profiles. This allows the local model to learn not only from its own experience but also from the collective intelligence, further improving the model's robustness and adaptability to unknown scenarios.

[0054] Through the above technical solution, this application combines local learning with cloud collaboration, enabling the vehicle adaptive adjustment system to achieve highly personalized real-time learning while also benefiting from collective intelligence, gaining wider applicability and stronger environmental adaptability, thus ensuring that the system achieves the best balance between personalization, adaptability, privacy protection and model generalization ability.

[0055] Based on the aforementioned vehicle adaptive adjustment system, this application also proposes a vehicle adaptive adjustment method. This method combines multimodal identity recognition, multi-source information acquisition, driving behavior and intention recognition, and hierarchical control architecture in a closed-loop collaborative manner, thereby solving the problems of single vehicle adjustment methods and lack of personalized automatic adaptation in the prior art.

[0056] Specifically, this method identifies the user's identity through multimodal fusion when the user approaches and enters the vehicle, and automatically adjusts the personalized cockpit configuration based on the identified user identifier.

[0057] Furthermore, during vehicle operation, the method periodically collects multi-source information, including vehicle dynamic parameters, environmental perception information, and positioning and navigation information, and identifies driving style and / or predicts future driving intentions based on time-series operational features.

[0058] Based on this, the method uses a hierarchical control architecture to generate adaptive adjustment commands for each subsystem of the vehicle, based on the user identifier, multi-source information, driving style, and driving operation intention.

[0059] Finally, the vehicle's various subsystems are adaptively adjusted according to the adaptive adjustment commands. The execution module sends the specific control sequence to the actuators through the vehicle communication network to adjust parameters such as suspension damping coefficient, vehicle height, steering assist curve index, return torque coefficient, throttle response curve index, energy recovery level, air conditioning temperature correction, and seat side wing support coefficient.

[0060] The adaptive adjustment method described above in this application is implemented through the following mechanism: Static adaptation based on identity recognition: Before a user gets into the car, the system obtains the user ID through multimodal identity recognition, retrieves the personalized profile, and automatically adjusts the seat, rearview mirror, steering wheel, air conditioning, HUD and other devices to achieve "adaptation upon getting into the car".

[0061] Scene-aware dynamic adaptation: During driving, the system uses a multi-source information acquisition module to perceive road conditions (road curvature, slope, road surface material, congestion level), environment (light, rain, snow, temperature), and vehicle status (vehicle speed, acceleration, attitude) in real time. The central decision-making unit re-solves for optimal control parameters every 20ms, enabling the suspension, steering, power, and braking systems to adjust in real time according to changes in the scene, achieving "one policy for each road and one policy for each time".

[0062] Learning and Adaptation Based on Driving Behavior: The driving behavior recognition module continuously analyzes user operation characteristics, identifies driving styles (economy / comfort / sport), and predicts intentions (such as lane changing and overtaking). The control module pre-adjusts system parameters based on style and intention. For example, when a sporty driving style is detected, it actively enhances suspension damping and improves steering response; when a lane change intention is predicted, it adjusts steering assist characteristics in advance, thus achieving "understanding the driver's intentions."

[0063] Adaptive optimization based on closed-loop feedback: The system records each manual intervention by the user and the scenario during that intervention, and corrects the personalized model through local incremental learning; at the same time, it aggregates group data in the cloud to form a better basic control strategy, and continuously evolves through OTA. This mechanism enables the system to understand users better and adapt to different regions and working conditions with increasing use.

[0064] In actual operation, the adaptive adjustment system executes the following process with a period of 20ms: The user identification module continuously monitors proximity signals and outputs the user ID upon triggering identification.

[0065] The personalized cockpit adjustment module performs static adjustments immediately after the vehicle is powered on, and makes dynamic fine adjustments based on the self-learning results while driving.

[0066] The multi-source information acquisition module synchronously collects vehicle dynamics, environmental perception, and positioning and navigation data, and stores them in a circular buffer (capacity 100 frames) after timestamp alignment.

[0067] The driving behavior and intent recognition module extracts feature sequences at 100ms intervals and runs an LSTM model to output style classification and intent prediction.

[0068] The central decision-making unit reads the current state vector S, uses the DQN policy network to output the optimal action A, and decomposes it into instructions for each executor through the MPC coordination layer.

[0069] The execution module receives and executes instructions, and then feeds back the actual status to the central unit.

[0070] The self-learning module records user interventions and scenario data, updates the local model when trigger conditions are met, and periodically uploads anonymized data to the cloud.

[0071] Through the above technical solution, this application achieves automated, personalized, and collaborative adjustment throughout the entire process from user approach to vehicle to driving. This method can automatically configure the cabin environment based on user identity and dynamically adjust the parameters of various vehicle subsystems based on real-time road conditions, driving behavior, and environmental changes, thereby improving the vehicle's ride comfort, driving stability, fuel economy, and driving safety under different operating conditions.

[0072] It should be noted that the execution process of the above methods and steps is based on the same concept as the system embodiments of this application. For details on the specific implementation process, functions, and technical effects, please refer to the system embodiments section, and they will not be repeated here. The descriptions of each embodiment have their own emphasis; parts not detailed or recorded in a certain embodiment can be referred to the relevant descriptions of other embodiments.

[0073] In one embodiment, this application also discloses a vehicle including the aforementioned vehicle adaptive adjustment system. The core innovation of this embodiment lies in achieving a unified perception, decision-making, and execution architecture for cabin adjustment and dynamic driving adjustment by integrating the vehicle adaptive adjustment system into the vehicle.

[0074] The following example demonstrates the adaptive adjustment process of this invention by showing a user (hereinafter referred to as Li) driving a smart electric SUV equipped with this system and completing a complete journey of "weekday morning rush hour commuting + weekend suburban mountain road driving".

[0075] (1) Before boarding: Identity recognition and cabin preset.

[0076] At 7:20 a.m. on Monday, Li, carrying his linked smartphone (BLE+UWB), walked from his home to his car in the underground parking garage. When he was about 1.2 meters away from the driver's side door, the system, through UWB angle of arrival and time of flight calculations, recognized that the user's approach direction and speed matched his intention to get in the car, automatically woke up the vehicle, and unlocked the door. Li opened the door, and the A-pillar infrared camera captured his facial image. This image was then fused with the phone's Bluetooth identifier using multimodal fusion (face confidence 0.92, Bluetooth identifier confidence 1.0), resulting in a combined confidence score of 0.96. This confirmed the user ID as "Li" and retrieved his personalized profile from the cloud.

[0077] Within 4 seconds, the system automatically completes the following adjustments: Seat: The fore-and-aft slide rails are adjusted to 158mm (Li's preferred position of "knees slightly bent when the pedals are fully depressed"), the backrest angle is 104°, the seat cushion height is 72mm, and the lumbar support airbag is inflated to 0.28bar. Rearview mirrors: Adjust the left and right exterior rearview mirrors to the saved angle (9° horizontally inward and 2° vertically downward), and set the anti-glare sensitivity of the interior rearview mirror to high; Steering wheel: Adjust the height to 8mm below the center and extend 20mm forward and backward; Air conditioning: Left zone 23℃, right zone 22.5℃, fan speed automatic (initially set to level 3 based on the current basement temperature of 26℃), airflow mode: face + feet; HUD: The brightness automatically adapts to the low-light environment of the underground parking lot, and the displayed content is vehicle speed + navigation abbreviation.

[0078] Mr. Li sat down and fastened his seatbelt.

[0079] (2) Urban morning rush hour: scene perception and adaptive adjustment.

[0080] Mr. Li drove out of the underground parking garage and onto a main urban road during the morning rush hour. The system identified the following scene characteristics: Road type: Urban arterial road (high-precision map labeling); Traffic condition: Congested (forward radar detects multiple vehicles moving at low speeds ahead, with an average distance between vehicles <5m); Vehicle speed: 12~25km / h, frequent starts and stops; Driving behavior recognition: The average throttle opening is 15%, the braking frequency is 0.3 times / minute, the steering wheel angle change rate is low, and the LSTM model outputs the driving style as "economy" (probability 0.78). The control module performs DQN inference based on the state vector S=[v=18km / h, frequent ax fluctuations, road curvature≈0, congested environment, style=economy], and outputs the optimal action: Suspension: CDC damping coefficient set to 0.3 (slightly soft, improving the filtering of small bumps). Steering: Assist curve index 1 (Comfort mode), self-centering torque coefficient 0.8 Powertrain: Throttle response curve index 0 (Economy mode), Energy recovery level 2 Cabin: The air conditioning fan speed automatically decreases to level 2 (to reduce blower noise). After the execution module adjusted according to the instructions, Mr. Li felt that the vehicle started smoothly, the steering was light, and the energy recovery intervention was natural, without the need for manual intervention. After driving for about 15 minutes, the system detected that Mr. Li manually lowered the air conditioning temperature from 23℃ to 22℃ and turned on the driver's seat ventilation at level 2. The self-learning and cloud collaboration module recorded this operation, with the scenario label "morning rush hour traffic jam, outside temperature 28℃, driving time 15 minutes", forming a "scenario-user correction" sample for subsequent personalized model updates.

[0081] (3) Highway: Scene switching and dynamic adjustment Mr. Li left the city roads and entered the ring expressway. The system detected that the vehicle speed increased to 80 km / h, the road type switched to expressway with a curvature of 0, and the road surface material was asphalt. The driving behavior recognition module detected that the throttle opening was stable (average 25%), the braking frequency dropped to 0.05 times / minute, and the output style remained "economy". However, the intention prediction module detected that Mr. Li had the intention to change lanes and overtake (rapid change of steering wheel angle combined with turn signal signal).

[0082] The control module is re-solved: Suspension: Damping coefficient increased to 0.6 (to improve stability), vehicle height lowered by 10mm (to reduce wind resistance); Steering: Assist curve switched to index 2 (stability mode), return torque coefficient 1.2; Powertrain: Energy recovery is reduced to level 1 (for smoother coasting), while throttle response remains in economy mode; Pre-adjustment: Upon detecting a lane change intention, the steering assist is temporarily adjusted to standard mode (index 1.5) to improve road feel during lane changes. It automatically reverts to stable mode after the lane change is completed.

[0083] While Mr. Li was cruising at high speed, the system also detected sunlight coming from the left based on the light sensor, automatically increasing the airflow of the left-side air conditioning by 10% and closing the left-side sunshade (if equipped). The self-learning module recorded that Mr. Li did not require additional intervention in the high-speed scenario, reinforcing the confidence of the current strategy.

[0084] (4) Mountain roads: complex working conditions and collaborative control Mr. Li exited the highway and entered a mountain road leading to the scenic area. The system predicted, through a high-precision map, that there would be a series of curves 2 kilometers ahead (curvature radius 60~120m, slope ±6%). The lidar and camera identified the road surface as dry asphalt with a good adhesion coefficient.

[0085] The control module enters mountain road mode: State perception: Vehicle speed 45km / h, lateral acceleration begins to build up (0.1~0.3g), and the rate of change of yaw rate increases.

[0086] Decision output: Suspension: Damping coefficient increased to 0.85 (Sport mode), vehicle height maintained at standard, to suppress body roll; Steering: Assist curve index 3 (Sports mode), steering ratio is dynamically adjusted, and hand force increases with vehicle speed and lateral acceleration; Power: Throttle response is switched to Sport mode (torque response time is reduced by 30%), energy recovery is maintained at level 2 but the recovery torque is reduced in coordination when braking in corners; Cabin: The side wing airbags of the seats are inflated to 0.5 bar to enhance lateral support; the air conditioning vent mode is adjusted to primarily blow air onto the face to avoid discomfort caused by blowing air onto the feet; Braking Coordination: When the system detects that Li is braking in a curve, the brake controller prioritizes mechanical braking and reduces energy recovery intervention according to the MPC command, so as to avoid the recovery torque disturbing the vehicle's attitude.

[0087] During the winding road, Mr. Li experienced minimal body roll, precise steering, and ample seat support, with the system automatically adjusting to a sporty mode. Throughout the journey, Mr. Li did not switch to any manual mode; the system autonomously and seamlessly transitioned from high-speed cruising to mountain road driving.

[0088] (5) Return journey and self-learning evolution In the evening, Mr. Li drove home. The system recorded that he manually lowered the air conditioning temperature to 21.5℃ again during the return trip and kept the seat ventilation at level 2. Combining the two manual temperature adjustment records of the day (lowering by 1℃ during the morning rush hour and by 0.5℃ during the return trip), the self-learning and cloud collaboration module determined that Mr. Li's temperature preference in the summer commuting scenario was lower than the preset value in his profile. Therefore, the air conditioning benchmark temperature in the personalized model for the "summer urban commuting" scenario was updated from 23℃ to 22.3℃.

[0089] After the vehicle arrives at its destination and is turned off, the system packages and uploads the day's trip data (including three driving conditions: urban congestion, highway cruising, and mountain curves), driving behavior characteristics, user intervention records, and user satisfaction (Mr. Li's voice feedback before getting out of the car: "It was very comfortable driving today") to the cloud. The cloud-based federated learning platform uses this data to incrementally train Mr. Li's personalized model, while simultaneously updating the basic model (such as the group experience of "mountain curve suspension parameters"). The updated model is then distributed to the vehicle via OTA the following morning. The next time Mr. Li drives on the same mountain road, the system will enter the curve adjustment preparation state earlier and automatically strengthen the seat side support before the curve, without requiring manual intervention.

[0090] It is understood that those skilled in the art will clearly recognize that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application.

[0091] Those skilled in the art will recognize that the units and algorithm steps of the various examples 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 implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art 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.

[0092] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application. Content not described in detail in this specification belongs to the prior art known to those skilled in the art.

Claims

1. A vehicle adaptive adjustment system, characterized in that: include The user identity recognition module is used to identify the user's identity and output the user identifier. The cockpit adjustment module is used to retrieve a personalized configuration file based on the user identifier and control the actuators in the cockpit to complete automatic adjustment. A multi-source information acquisition module is used to collect multi-source information in real time during vehicle operation; The driver behavior and intent recognition module is used to identify driving style and / or predict future driving intentions based on time-series operational features. The control module is used to generate adaptive adjustment commands for each subsystem of the vehicle based on the user identifier, multi-source information, driving style, and driving operation intention using a hierarchical control architecture. The execution module is used to execute the adaptive adjustment command to adaptively adjust each subsystem of the vehicle.

2. The vehicle adaptive adjustment system according to claim 1, characterized in that: The user identification module includes: The external identification unit is used to calculate the user's position, distance and movement trajectory relative to the vehicle when the user approaches the vehicle with the bound device, and to trigger the identity recognition process and unlock the car door when the user enters the preset identification area. An internal identification unit is used to collect the user's facial, voiceprint, and / or fingerprint biometrics after the user enters the vehicle. The fusion decision unit is used to perform fusion calculation on the recognition results from the external recognition unit and the internal recognition unit to determine the fusion identity confidence level. When the identity confidence level exceeds a preset threshold, the final user identifier is output.

3. The vehicle adaptive adjustment system according to claim 1, characterized in that: The driver behavior and intention recognition module is a temporal behavior model built on a long short-term memory network. The temporal behavior model takes the feature vectors of accelerator pedal opening, brake pedal opening, steering wheel angle, vehicle yaw rate, longitudinal acceleration, lateral acceleration and vehicle speed within a preset time window as input, and outputs the classification of driving style and the predicted probability of acceleration, braking, lane changing and steering intentions within a preset time.

4. The vehicle adaptive adjustment system according to claim 1, characterized in that: The hierarchical control architecture includes: The upper decision layer is used to determine the combination of target control parameters based on a state vector containing user identification, multi-source information, driving style, and driving operation intentions, using a fusion algorithm of deep reinforcement learning and fuzzy logic. The intermediate coordination layer is used to combine the target control parameters using model predictive control algorithms, and decompose them into a specific control quantity sequence for each actuator under the conditions of satisfying the optimization objective function and safety constraints. The lower execution layer is used to send the specific control quantity sequence to the corresponding actuator in the execution module through the vehicle communication network.

5. The vehicle adaptive adjustment system according to claim 4, characterized in that: The state vector is: Where S is the state vector; v is the vehicle speed; a x For longitudinal acceleration; a y κ represents lateral acceleration; θ represents yaw rate; κ represents road curvature; μ represents road surface adhesion coefficient; style represents driving style; env represents environment coding; and user_id represents user identifier.

6. The vehicle adaptive adjustment system according to claim 4, characterized in that: The target control parameter combination is as follows: A=[c_damp,h_height,assist_idx,return_coeff,pedal_map,regen_level,temp_offset,bolster_coeff] Where A is the target control parameter combination; c_damp is the suspension damping coefficient; h_height is the vehicle height; assist_idx is the steering assist curve index; return_coeff is the return torque coefficient; pedal_map is the throttle response curve index; regen_level is the energy recovery level; temp_offset is the air conditioning temperature correction amount; and bolster_coeff is the seat side wing support coefficient.

7. The vehicle adaptive adjustment system according to claim 4, characterized in that: The optimization objective function is: Where J is the objective function value; N p yref is the total step size in the prediction time domain; k is the step size index; yref is the target control parameter; y is the predicted control parameter; Δu is the rate of change of the control quantity; u is the control quantity; Q, R, and S are the weight matrices for output tracking, control stability, and terminal cost, respectively.

8. The vehicle adaptive adjustment system according to claim 4, characterized in that: The security constraints include physical amplitude limiting constraints and security adjustment constraints; The physical limiting constraints include the maximum current of the limiting damping valve, the vehicle height adjustment rate, and the maximum torque of the steering motor. The safety adjustment constraints include limits on yaw rate not exceeding the adhesion limit and roll angle not exceeding a set angle.

9. The vehicle adaptive adjustment system according to claim 1, characterized in that: It also includes a self-learning and cloud collaboration module, which records the user's adjustment and intervention behaviors and scenarios, updates personalized configuration profiles based on the recorded information, and interacts with the cloud for data.

10. The vehicle adaptive adjustment system according to claim 9, characterized in that: The self-learning and cloud-based collaboration module includes: The local learning unit is used to record the current scene and the user's intervention when it detects that the user has manually intervened in the result of automatic adjustment, forming a scene-operation sample pair, and using an incremental learning algorithm to update the locally stored personalized configuration profile. The cloud collaboration unit is used to upload the scene-operation samples to the cloud server after de-identification processing, participate in the training of the swarm intelligence model based on the federated learning framework, and receive the optimized basic model parameter updates from the cloud server.

11. A vehicle adaptive adjustment method based on the vehicle adaptive adjustment system according to any one of claims 1-10, characterized in that: When a user approaches and enters the vehicle, the system identifies the user's identity through multimodal fusion and automatically adjusts the personalized cockpit configuration based on the identified user identifier. During vehicle operation, multi-source information is periodically collected, and driving style and / or future driving intentions are identified based on time-series operational characteristics. Based on the user identifier, multi-source information, driving style, and driving operation intention, a hierarchical control architecture is used to generate adaptive adjustment commands for each subsystem of the vehicle. The vehicle's various subsystems are controlled to perform adaptive adjustments based on adaptive adjustment commands.

12. A vehicle, characterized in that: The vehicle adaptive adjustment system includes any one of claims 1-10.

Images