Large new energy vehicle steep slope slow descending control method and system based on multiple sensors

Abstract

The application discloses a large new energy vehicle steep slope slow descending control method and system based on multiple sensors, aiming at a multi-sensor fusion perception system of a mine scene, through space-time registration and data fusion of multiple source information such as laser radar, millimeter wave radar, visual camera, ultrasonic sensor and vehicle body posture sensor, accurate identification of the mine steep slope scene and high-precision calculation of the slope are realized, and the slope calculation error can be controlled within ±0.5°. An adaptive hierarchical decision mechanism based on scene characteristics is proposed, control parameters are dynamically adjusted according to factors such as the closedness of the mine scene, slope change and road surface state, and adaptive optimization of the slow descending control strategy is realized. A collaborative optimization mechanism of brake control and energy recovery is designed, the energy recovery intensity is dynamically adjusted according to brake demand and battery state, and the unification of safe braking and efficient energy recovery is realized.

Description

Technical Field

[0001] This invention belongs to the field of new energy vehicle control technology. It relates to a method and system for controlling steep slope descent of large new energy vehicles based on multiple sensors, which is particularly suitable for special scenarios such as underground parking garages. Background Technology

[0002] Large new energy vehicles (such as large SUVs and new energy commercial vehicles) have physical characteristics such as large body size, heavy curb weight, and large rotational inertia, which pose unique challenges when driving on steep slopes in underground parking garages. As an important parking space in cities, underground parking garages typically have characteristics such as strong spatial enclosure, dim ambient light, limited visibility, and large differences in slope flatness. These factors together constitute a severe test of the vehicle's descent control capabilities.

[0003] On one hand, existing automotive hill descent control technologies are mostly designed for open ground environments and rely primarily on a single type of sensor for environmental perception. Typical existing solutions employ a combination of vehicle speed and slope sensors, triggering the braking system to control vehicle speed by detecting changes in vehicle speed and road gradient. For example, some existing technologies monitor the rate of change of vehicle speed using wheel speed sensors; when the speed exceeds a preset threshold, the hydraulic braking system is activated to apply braking force. Other existing technologies use gravity acceleration sensors to estimate the slope and combine this with a fixed braking torque output strategy to achieve the descent function.

[0004] On the other hand, some high-end models have begun to introduce multi-sensor fusion technology, but it is mainly applied to the global perception level of autonomous driving. These technologies typically fuse data from sensors such as cameras and radar to build an environmental model of the vehicle's surroundings, but they are not specifically optimized for the steep slopes of underground parking garages. In terms of sensor configuration, existing technologies mostly use a combination of forward-facing monocular cameras or millimeter-wave radar, and the detection range and accuracy are significantly limited by the enclosed environment of underground parking garages.

[0005] Furthermore, in terms of braking control strategies, existing technologies generally employ fixed control parameters and braking logic. For example, some systems set a fixed target vehicle speed (e.g., 5 km / h) and a fixed braking torque output curve, without considering the specific characteristics of underground parking scenarios. Regarding energy recovery, although new energy vehicles possess braking energy recovery capabilities, existing technologies typically design descent control and energy recovery as two independent subsystems, lacking a collaborative optimization mechanism.

[0006] Existing technologies have many significant drawbacks in the application of steep slope descent control in large-scale new energy vehicle parking garages, which severely restrict the practicality and safety of the technology.

[0007] First, in terms of perception accuracy, single-sensor configurations exhibit severe limitations in the complex environment of underground parking garages. Underground parking garages typically suffer from insufficient light intensity (often below 50 lux), strong electromagnetic interference, and complex multipath reflections, causing systems relying on a single sensor to fail to accurately acquire environmental information. For example, visual cameras experience a sharp decline in image quality in low-light conditions, making it difficult to accurately identify road markings and obstacles; millimeter-wave radar is susceptible to multipath reflections in enclosed spaces, easily generating false targets; and ultrasonic sensors have limited detection range, failing to meet the long-distance perception requirements of steep slope scenarios. More critically, large new energy vehicles, due to their large size and wide blind spots, exhibit even more pronounced blind spot problems with single sensors, easily leading to serious consequences such as inaccurate slope calculations and missed obstacle detections.

[0008] Secondly, regarding scenario adaptability, existing descent control functions fail to fully consider the unique characteristics of underground parking garages. The steep slopes of underground parking garages differ fundamentally from those of surface slopes in terms of environmental characteristics: underground parking garages exhibit significant spatial enclosure, constant low-light conditions, specific road surface materials (mostly cement or epoxy resin flooring), and unique slope geometry. However, existing technologies employ general descent control strategies, failing to optimize for these specific characteristics of underground parking garages. For example, fixed-parameter descent strategies cannot adapt to variations in slope gradient and road surface adhesion coefficients, leading to tire slippage on wet surfaces and potentially affecting ride comfort on dry surfaces due to excessive braking.

[0009] Secondly, in terms of braking control precision, existing technologies are incompatible with the dynamic characteristics of large new energy vehicles. Large new energy vehicles typically have a curb weight exceeding 2.5 tons and a much greater moment of inertia than ordinary passenger cars. This means that under the same gradient conditions, these vehicles experience greater downhill acceleration and longer braking distances. Existing technologies often cannot meet the braking torque output requirements of large vehicles, easily leading to insufficient braking capacity and subsequent rollover. Furthermore, the braking systems of large vehicles experience higher thermal loads, and existing technologies lack mechanisms to prevent and compensate for brake fade, posing a safety hazard of brake failure.

[0010] Finally, in terms of energy utilization efficiency, existing technologies have failed to achieve synergistic optimization between braking control and energy recovery. Existing energy recovery systems typically employ fixed recovery strategies, neglecting dynamic changes in factors such as battery state and braking demand, resulting in low energy recovery efficiency. Summary of the Invention

[0011] The purpose of this invention is to provide a method and system for hill descent control of large new energy vehicles based on multiple sensors. It achieves accurate identification of special scenarios and accurate calculation of slope through multi-sensor fusion, adaptive adjustment of control parameters through a hierarchical decision-making mechanism, and unification of safe braking and efficient energy recovery through a collaborative optimization mechanism.

[0012] The technical solution to achieve the purpose of this invention is as follows: A method for hill descent control of large new energy vehicles based on multiple sensors includes the following steps: S01: Acquire data from multiple sensors on the vehicle, and obtain multi-sensor data after preprocessing; S02: Extract scene features based on multi-sensor data, match the extracted scene features with the scene features in the preset underground parking scene feature library, determine whether it is an underground parking steep slope scene, and classify the underground parking scene. S03: After identifying a steep slope scene in the underground parking lot, the slope is calculated to obtain the slope parameters; S04: Generate an adaptive descent control strategy based on the parking lot scene classification results and slope parameters. The control strategy includes: The control is decomposed into different layers for adaptive hierarchical decision-making. The scenario adaptation layer is used to determine the basic control parameters based on the scenario characteristics of the underground parking lot; the slope adaptation layer is used to adjust the control parameters according to the slope changes; and the vehicle status adaptation layer is used to make dynamic adjustments based on the real-time status of the vehicle. Establish vehicle dynamics and braking system models to predict vehicle status over a future period of time. A unified cost function is established to coordinate the optimization of energy recovery and braking control. In each control cycle, a multi-objective optimization problem is solved based on the vehicle state, battery state, and road surface state to obtain the optimal distribution of motor braking torque and mechanical braking torque.

[0013] In the preferred technical solution, the preprocessing in step S01 includes: Remove data noise; The PTP protocol is used to synchronize the clocks of each sensor, and an interpolation algorithm is used to align the data of the low-frequency sensor to the timestamp of the high-frequency sensor. The rigid body transformation method is used to transform the local coordinate system of each sensor to a unified vehicle coordinate system. The transformation parameters are obtained through sensor calibration.

[0014] In the preferred technical solution, scene features extracted based on multi-sensor data include: Extract light intensity and texture features from visual images, and calculate the average brightness value of the image; Spatial enclosure features are extracted from lidar point cloud data, and the existence of lateral occlusion is determined by analyzing the spatial distribution of the point cloud data. Road surface material features are extracted from the texture features of visual images and the spatial closure features of LiDAR data, and road surface types are identified based on texture features and reflectivity. The slope change features are extracted from the vehicle attitude sensor data, and the rate of change of pitch angle is used to determine whether the vehicle is going downhill.

[0015] In the preferred technical solution, a multi-dimensional comprehensive evaluation method is used to determine whether it is a steep slope scenario in a parking garage, including: Calculate the similarity between the features of the scene to be identified and the features of each scene in the basement scene feature database. The similarity is calculated using weighted Euclidean distance, and the weights of different scene features are determined according to their contribution to scene identification. When the similarity exceeds a preset threshold, the scene is determined to be a steep slope scene in a parking garage.

[0016] In the preferred technical solution, step S03 involves slope calculation to obtain slope parameters including: Slope point cloud is extracted from lidar point cloud data, and the slope plane is fitted using the random sampling consensus algorithm to obtain the slope estimate based on lidar. Pitch angle data is read from the vehicle attitude sensor and smoothed by Kalman filtering to obtain the slope estimate based on the vehicle attitude sensor. An adaptive weighted fusion algorithm is used to fuse the two slope estimates and output the fused slope value. By analyzing historical data, a mapping relationship between slope measurement error and vehicle status and environmental conditions is established, resulting in an error model. The slope value is compensated in real time. The error includes systematic error, random error and state-related error. Systematic error is eliminated through calibration and compensation, random error is suppressed through filtering algorithm, and state-related error is corrected in real time through adaptive compensation mechanism.

[0017] In the preferred technical solution, after obtaining the slope parameters, the assessment of slope smoothness and road surface condition is also included: The slope flatness assessment is based on lidar point cloud data. The flatness of the slope is calculated by selecting multiple sampling areas on the slope and calculating the standard deviation of the distance from the point cloud data in each sampling area to the fitted plane. The magnitude of the standard deviation is used to characterize the flatness of the slope. The road surface condition assessment is based on visual images and environmental sensor data. The specific method is as follows: extract road surface texture features from visual images and use deep learning algorithms to identify the dry and wet state of the road surface; read temperature and humidity data from environmental sensors to determine whether there is a risk of icing; and calculate the estimated value of the road surface adhesion coefficient.

[0018] In the preferred technical solution, a unified cost function is established, and the synergistic optimization of energy recovery and braking control includes: Collect the vehicle state vector x_k = [v_k, a_k, θ_k, μ_k, SOC_k, T_batt_k] at the current time k, where v_k is the vehicle speed, a_k is the acceleration, θ_k is the slope, μ_k is the road adhesion coefficient, SOC_k is the battery state of charge, and T_batt_k is the battery temperature; based on the vehicle dynamics model, predict the vehicle state trajectory in the next N time steps. Calculate the braking force F_brake_req required to maintain the target vehicle speed based on the target vehicle speed v_target and the current vehicle speed v_k: F_brake_req = m·g·sinθ_k - F_roll - F_aero Where m is the vehicle mass, g is the gravitational acceleration, F_roll is the rolling resistance, and F_aero is the air resistance; Based on the current SOC and battery temperature, calculate the battery's maximum charging power P_charge_max and maximum charging current I_charge_max; Establish a unified cost function J: J = w1·J_brake + w2·J_energy + w3·J_comfort + w4·J_stability Among them, w1, w2, w3, and w4 are weights, J_brake is the braking error term, J_energy is the energy recovery term, J_comfort is the ride comfort term, and J_stability is the stability term; Constraints: T_motor_min ≤ T_motor ≤ T_motor_max T_mech_min ≤ T_mech ≤ T_mech_max T_motor + T_mech = T_brake_req I_charge ≤ I_charge_max P_charge ≤ P_charge_max 0 ≤ λ_i ≤ λ_critical Where T_motor is the motor braking torque, T_motor_min and T_motor_max are the minimum and maximum motor braking torques respectively, T_mech is the mechanical braking torque, T_mech_min and T_mech_max are the minimum and maximum mechanical braking torques respectively, T_brake_req is the total braking torque demand, I_charge is the charging current, I_charge_max is the maximum charging current, P_charge is the charging power, P_charge_max is the maximum charging probability, λ_i is the wheel slip ratio, and λ_critical is the critical slip ratio.

[0019] In the preferred technical solution, the braking error term J_brake is: J_brake = Σᵢ₌1ᴺ (v_target - v_i)² v_i represents the actual vehicle speed; Energy recovery item J_energy: J_energy = -α·η·P_recover Where P_recover is the actual recovered power, η is the energy conversion efficiency, α is the energy recovery weighting coefficient, and the negative sign indicates maximizing energy recovery. P_recover is calculated as follows: P_recover = T_motor·ω_motor·η_motor Where T_motor is the motor braking torque, ω_motor is the motor speed, and η_motor is the motor efficiency; Ride comfort item J_comfort: J_comfort = Σᵢ₌1ᴺ (da_i / dt)² + Σᵢ₌1ᴺ (ΔT_i)² a_i represents the acceleration, and ΔT_i represents the change in braking torque; Stability term J_stability: J_stability = Σᵢ₌1ᴺ (λ_i - λ_opt)² Where λ_i is the slip ratio of each wheel, and λ_opt is the optimal slip ratio.

[0020] This invention also discloses a multi-sensor-based hill descent control system for large new energy vehicles, used to implement the aforementioned multi-sensor-based hill descent control method for large new energy vehicles, comprising: The multi-sensor perception module acquires data collected by multiple sensors on the vehicle and preprocesses it to obtain multi-sensor data. The scene recognition module extracts scene features based on multi-sensor data, matches the extracted scene features with each scene feature in the preset underground parking scene feature library, determines whether it is an underground parking steep slope scene, and classifies the underground parking scene. The slope calculation module, when identified as a steep slope scene in a parking garage, performs slope calculation to obtain slope parameters; The descent control decision module generates an adaptive descent control strategy based on the parking garage scenario classification results and slope parameters. The control strategy includes: The control is decomposed into different layers for adaptive hierarchical decision-making. The scenario adaptation layer is used to determine the basic control parameters based on the scenario characteristics of the underground parking lot; the slope adaptation layer is used to adjust the control parameters according to the slope changes; and the vehicle status adaptation layer is used to make dynamic adjustments based on the real-time status of the vehicle. Establish vehicle dynamics and braking system models to predict vehicle status over a future period of time. A unified cost function is established to coordinate the optimization of energy recovery and braking control. In each control cycle, a multi-objective optimization problem is solved based on the vehicle state, battery state, and road surface state to obtain the optimal distribution of motor braking torque and mechanical braking torque.

[0021] The present invention also discloses a computer storage medium storing a computer program, which, when executed, implements the above-mentioned multi-sensor-based steep slope descent control method for large new energy vehicles.

[0022] Compared with the prior art, the significant advantages of this invention are: 1. In terms of safety, this invention significantly improves the perception accuracy and reliability in steep underground parking garage scenarios through multi-sensor fusion sensing technology. The multi-sensor configuration eliminates the blind spots of a single sensor, enabling accurate obstacle identification, slope calculation, and road condition assessment even in dimly lit and highly interference-prone underground parking garage environments. High-precision perception information lays the foundation for precise braking control, effectively preventing safety accidents such as vehicle rollback, loss of speed control, and tire slippage caused by perception errors. Simultaneously, the fault diagnosis and emergency handling module provides multiple safety safeguards; even in the event of sensor or braking system malfunctions, the system can promptly detect and take emergency measures to ensure the safety of the vehicle and its occupants.

[0023] 2. Regarding adaptability, this invention is specifically optimized for the unique characteristics of underground parking garage scenarios. By constructing a feature library of underground parking garage scenarios and employing a self-learning mechanism, the system can identify different types of steep slopes in underground parking garages (old underground parking garages, newly built underground parking garages, shopping mall underground parking garages, residential underground parking garages, etc.) and dynamically adjust the control strategy according to the scenario characteristics. The hierarchical decision-making mechanism enables the system to adapt to various dynamic factors such as slope changes, road surface condition changes, and battery condition changes, achieving true adaptive control. This strong adaptability allows this invention to be widely applied to various underground parking garage scenarios without requiring special calibration for specific underground parking garages, thus reducing application costs.

[0024] 3. In terms of comfort, this invention significantly improves ride comfort through a refined control strategy. The segmented braking strategy avoids vehicle pitching and skidding caused by excessive braking, making the descent process smoother. Adaptive speed control dynamically adjusts the target speed according to the slope and road conditions, avoiding panic caused by excessive speed or inefficiency caused by excessive speed.

[0025] 4. In terms of economics, this invention significantly improves energy recovery efficiency through the synergistic optimization of braking control and energy recovery. Simultaneously, energy recovery reduces the frequency of mechanical braking use, decreases wear on the braking system, extends the maintenance cycle of the braking system, and further reduces operating costs.

[0026] 5. Regarding reliability, this invention significantly improves system reliability through multiple redundancy designs and fault-tolerance mechanisms. The multi-sensor configuration provides perceptual redundancy; even if one sensor fails, the system can still obtain necessary information through other sensors. The dual configuration of motor braking and mechanical braking provides execution redundancy; even if one braking method fails, the other can still guarantee basic braking function. The graded fault diagnosis and emergency handling mechanism provides safety redundancy, ensuring safe operation of the system under various abnormal conditions. Attached Figure Description

[0027] Figure 1 This is a flowchart of the hill descent control method for large new energy vehicles based on multiple sensors in this embodiment; Figure 2 This is a schematic diagram of the overall system architecture of the present invention; Figure 3 This is a schematic diagram of the multi-sensor fusion sensing process; Figure 4 For adaptive hierarchical decision control flowchart; Figure 5 A flowchart illustrating the collaborative workflow of braking control and energy recovery; Figure 6 This is a schematic diagram illustrating the principle of the slope calculation algorithm. Detailed Implementation

[0028] The principle of this invention is as follows: Based on multi-source information fusion theory and adaptive control theory, this invention achieves accurate perception of steep slope scenarios in underground parking lots through spatiotemporal registration and feature-level fusion of multi-sensor data. On this basis, a hierarchical decision-making mechanism is adopted to map multi-dimensional information such as scene recognition results, slope parameters, and vehicle status into optimal control commands. Finally, through the coordinated work of the braking execution module and the energy recovery coordination module, precise speed control and efficient recovery of braking energy are achieved.

[0029] Example 1: like Figure 1 As shown, a method for hill descent control of large new energy vehicles based on multiple sensors includes the following steps: S01: Acquire data from multiple sensors on the vehicle, and obtain multi-sensor data after preprocessing; S02: Extract scene features based on multi-sensor data, match the extracted scene features with the scene features in the preset underground parking scene feature library, determine whether it is an underground parking steep slope scene, and classify the underground parking scene. S03: After identifying a steep slope scene in the underground parking lot, the slope is calculated to obtain the slope parameters; S04: Generate an adaptive descent control strategy based on the parking lot scene classification results and slope parameters. The control strategy includes: The control is decomposed into different layers for adaptive hierarchical decision-making. The scenario adaptation layer is used to determine the basic control parameters based on the scenario characteristics of the underground parking lot; the slope adaptation layer is used to adjust the control parameters according to the slope changes; and the vehicle status adaptation layer is used to make dynamic adjustments based on the real-time status of the vehicle. Establish vehicle dynamics and braking system models to predict vehicle status over a future period of time. A unified cost function is established to coordinate the optimization of energy recovery and braking control. In each control cycle, a multi-objective optimization problem is solved based on the vehicle state, battery state, and road surface state to obtain the optimal distribution of motor braking torque and mechanical braking torque.

[0030] In another embodiment, a computer storage medium stores a computer program that, when executed, implements the aforementioned multi-sensor-based hill descent control method for large new energy vehicles. The method described above will not be elaborated further here.

[0031] In another embodiment, such as Figure 2As shown, a multi-sensor-based hill descent control system for large new energy vehicles is presented to implement the aforementioned multi-sensor-based hill descent control method for large new energy vehicles. This system adopts a modular design, consisting of a multi-sensor perception module, a scene recognition and slope calculation module, and a hill descent control decision module. The hill descent control decision module further includes a braking execution module, an energy recovery coordination module, and a fault diagnosis and emergency handling module. All modules interact and collaborate with the vehicle control unit (VCU) via a CAN bus, forming a complete perception-decision-execution closed-loop control system.

[0032] At the technical principle level, this invention is based on multi-source information fusion theory and adaptive control theory. Through spatiotemporal registration and feature-level fusion of multi-sensor data, it achieves accurate perception of steep slope scenarios in underground parking lots. Building upon this, a hierarchical decision-making mechanism is employed to map multi-dimensional information such as scene recognition results, slope parameters, and vehicle status into optimal control commands. Ultimately, through the coordinated operation of the braking execution module and the energy recovery coordination module, precise speed control and efficient recovery of braking energy are achieved.

[0033] In terms of architecture, this invention adopts a distributed architecture design. The multi-sensor perception module serves as the data acquisition layer, responsible for acquiring raw environmental data and vehicle status data; the scene recognition and slope calculation module serves as the perception processing layer, completing data preprocessing, feature extraction, and scene understanding; the descent control decision module serves as the decision layer, generating control commands based on the perception results; the braking execution module and energy recovery coordination module serve as the execution layer, responsible for the physical implementation of the control commands; and the fault diagnosis and emergency handling module serves as the safety assurance layer, monitoring the system status in real time and handling abnormal situations. This layered architecture design ensures the system's modularity, scalability, and maintainability.

[0034] In terms of key technologies, this invention achieves breakthroughs in three core areas. First, multi-sensor spatiotemporal registration and data fusion technology, using algorithms such as Kalman filtering and Bayesian estimation to achieve accurate fusion of heterogeneous sensor data, eliminating time synchronization and spatial registration errors. Second, underground parking scene feature extraction and recognition technology, constructing an underground parking scene feature database and employing machine learning algorithms to automatically identify and classify steep slope scenarios in underground parking lots. Third, adaptive hierarchical decision control technology, achieving dynamic optimization of control parameters through the collaborative work of scene adaptation layers, slope adaptation layers, and vehicle body state adaptation layers.

[0035] Specific Implementation Example 1: Accurate Identification and Slope Calculation of Steep Slope Scenarios in Underground Parking Lots Implementation scenario description This embodiment addresses the identification and slope calculation of large new energy vehicles entering underground parking garages on steep slopes. The specific scenario is as follows: a vehicle drives from a surface road into an underground parking garage, needing to traverse a steep slope of 8-15 degrees and 30-50 meters in length. This section exhibits typical characteristics of an underground parking garage: strong spatial enclosure (walls or railings on both sides), dim ambient lighting (illuminance typically below 50 lux), road surface material of cement or epoxy resin, and potential variations in slope smoothness and localized slipperiness.

[0036] Technical Configuration Description To achieve accurate identification and slope calculation in this scenario, the following sensor system is configured in this embodiment: The lidar is a solid-state lidar, installed in the center of the vehicle's front bumper. It has a detection range of 0.5-50 meters, a horizontal field of view of 120 degrees, a vertical field of view of 25 degrees, a scanning frequency of 10Hz, and an angular resolution of 0.1 degrees. This sensor is responsible for collecting 3D point cloud data of the steep slope of the parking garage for slope terrain reconstruction and obstacle detection.

[0037] The millimeter-wave radar operates in the 77GHz band and is installed on both sides of the vehicle's front bumper. It has a detection range of 0.1-100 meters, a horizontal field of view of ±45 degrees, a distance resolution of 0.1 meters, and a speed resolution of 0.1 km / h. This sensor is responsible for monitoring dynamic obstacles above and below steep slopes, providing information on the distance, speed, and orientation of these obstacles.

[0038] The visual camera uses a 2-megapixel CMOS sensor, mounted above the rearview mirror on the windshield. It features low-light enhancement, a minimum illumination of 0.1 lux, and a frame rate of 30fps. This sensor is responsible for acquiring image information of the steep slope of the parking garage, used for road marking recognition, slope change detection, and road condition assessment.

[0039] The vehicle attitude sensor employs a six-axis inertial measurement unit (IMU), mounted at the vehicle's center of gravity (typically under the front seats). It includes a three-axis accelerometer and a three-axis gyroscope. The accelerometer has a range of ±2g (g is the acceleration due to gravity, approximately 9.8 m / s²), the gyroscope has a range of ±500 degrees / second, and the sampling frequency is 100Hz. This sensor is responsible for collecting the vehicle's pitch, roll, and acceleration data for gradient-assisted calculations and vehicle stability assessments.

[0040] The ultrasonic sensor employs eight probes, four of which are mounted on the front bumper and four on the rear bumper. It has a detection range of 0.02-5 meters and a detection frequency of 40kHz. This sensor is responsible for detecting obstacles at close range and assists in low-speed obstacle avoidance.

[0041] The environmental sensor integrates a temperature sensor, a humidity sensor, and a road condition sensor, and is installed on the left side of the front bumper. It measures temperature from -40 to 125 degrees Celsius and humidity from 0 to 100% RH. This sensor is responsible for collecting environmental parameters within the parking garage, providing environmental information support for braking control.

[0042] Detailed Explanation of Implementation Steps like Figure 3 As shown, the implementation process of this embodiment is carried out according to the following steps: Step 1: Multi-sensor data acquisition and preprocessing. Each sensor synchronously acquires data according to a preset sampling frequency. The LiDAR outputs 3D point cloud data, the millimeter-wave radar outputs target list data, the vision camera outputs image data, the vehicle attitude sensor outputs IMU data, the ultrasonic sensor outputs distance data, and the environmental sensor outputs environmental parameter data. The acquired raw data needs to be preprocessed, including data denoising, time synchronization calibration, and spatial coordinate unification.

[0043] Data denoising employs a median filtering algorithm to process the lidar point cloud data, eliminating isolated noise points. A Kalman filter (a recursive filtering algorithm used to estimate system state) is used to smooth the millimeter-wave radar data, reducing measurement random errors. Time synchronization calibration uses the Precision Time Protocol (PTP) to ensure that the timestamp error of each sensor is less than 1ms. Spatial coordinates are unified using the vehicle coordinate system as the reference coordinate system, transforming the measurement data from each sensor to a unified vehicle coordinate system to eliminate spatial deviations caused by sensor installation positions.

[0044] Step 2: Basement Scene Feature Extraction. Based on the preprocessed multi-sensor data, key features of the basement scene are extracted. Specifically, this includes: extracting light intensity features from visual images and calculating the average brightness value of the image; extracting spatial closure features from LiDAR point cloud data and determining the presence of lateral occlusion by analyzing the spatial distribution of the point cloud; extracting road surface material features from visual images and LiDAR data, identifying the road surface type based on texture analysis and reflectivity analysis; and extracting slope change features from vehicle attitude sensor data, determining whether the vehicle is going downhill by analyzing the rate of change of the pitch angle.

[0045] Feature extraction employs a multimodal feature fusion method, correlating and complementing feature information from different sensors. For example, combining the texture features of visual images with the geometric features of LiDAR (LiDAR) point clouds improves the accuracy of road surface material identification; combining the Doppler velocity information of millimeter-wave radar (typically 30-300 GHz) with the motion information of visual images improves the reliability of dynamic obstacle detection.

[0046] Step 3: Basement Scene Recognition and Classification. The extracted scene features are matched against a pre-defined basement scene feature library, and a Support Vector Machine (SVM) classification algorithm is used for scene classification. The basement scene feature library contains a large number of feature samples from real basement scenes, covering different types of basements (old basements, newly built basements, shopping mall basements, residential basements, etc.) and different environmental conditions (daytime, nighttime, dry, humid, etc.).

[0047] The scene recognition criteria employ a multi-dimensional comprehensive evaluation method. First, the similarity between the features of the scene to be recognized and various scene features in the feature library is calculated. The similarity is calculated using weighted Euclidean distance, with the weights of different features determined based on their contribution to scene recognition. When the similarity exceeds a preset threshold (set to 85% in this embodiment), the scene is determined to be a steep slope in a parking garage. Simultaneously, the system also outputs the scene's confidence parameter for reference in subsequent control decisions.

[0048] Step 4: Accurately calculate the slope gradient of steep slopes. For example... Figure 6 As shown, when a steep slope in a parking garage is identified, the system initiates a slope calculation procedure. The slope calculation employs a multi-sensor fusion method, combining pitch angle data from the vehicle's attitude sensor and 3D terrain data from the lidar for joint estimation.

[0049] The specific calculation process is as follows: First, slope point clouds are extracted from the LiDAR point cloud data, and the Random Sample Consensus (RANSAC) algorithm is used to fit the slope plane to obtain a LiDAR-based slope estimate. Second, pitch angle data is read from the vehicle attitude sensor and smoothed using Kalman filtering to obtain an IMU-based slope estimate. Then, an adaptive weighted fusion algorithm is used to fuse the two slope estimates, with the weights dynamically adjusted according to the uncertainties of each sensor. Finally, the fused slope value is output as the final calculation result.

[0050] The core formula for slope calculation is as follows: LiDAR slope estimation: Let the slope point cloud data acquired by lidar be P = {p1, p2, ..., p n}, where each point pᵢ = (xᵢ, yᵢ, zᵢ) represents three-dimensional spatial coordinates. The RANSAC algorithm is used to fit the plane equation: ax + by + cz + d = 0 Where (a, b, c) are the plane normal vectors. d represents the directed distance from the plane to the origin of the coordinate system. d > 0: the plane is located on the side of the positive direction of the normal vector; d < 0: the plane is located on the side of the negative direction of the normal vector; d = 0: the plane passes through the origin of the coordinate system.

[0051] The slope angle θ_LiDAR estimated by the lidar is: θ_LiDAR = arctan IMU slope estimation: The pitch angle measured by the vehicle attitude sensor is θ_IMU, which is obtained after Kalman filtering and smoothing: θ_IMU_filtered = K·θ_IMU_prev + (1-K)·θ_IMU_current Where K is the Kalman gain, which is dynamically adjusted based on measurement noise and process noise.

[0052] θ_IMU_prev is the filtered estimate from the previous moment, which can be obtained based on the prior estimate of the pitch angle from historical data, reflecting the continuity of the state. θ_IMU_current is the measurement value at the current moment, which is the pitch angle observation value directly measured by the IMU sensor, containing measurement noise but reflecting the latest state.

[0053] The final slope estimate θ_fused is: θ_fused = w 11 ·θ_LiDAR + w 12 ·θ_IMU_filtered Wherein, weight w 11 and w 12 Dynamically determined based on the measurement uncertainties of each sensor: , Wherein, σ_LiDAR and σ_IMU are the measurement standard deviations of the lidar and IMU, respectively, reflecting the measurement uncertainty.

[0054] To ensure measurement accuracy, this embodiment also employs an error compensation mechanism. By analyzing historical data, a mapping relationship is established between slope measurement error and vehicle status (such as speed and acceleration) and environmental conditions (such as temperature and humidity), and the measurement results are compensated in real time. After compensation, the slope measurement error can be controlled within ±0.5°, meeting the requirements of high-precision control.

[0055] Introducing an error compensation term: θ_final = θ_fused + Δθ_comp Δθ_ Where v is vehicle speed, a_x is longitudinal acceleration, T is temperature, and H is humidity. Error compensation function It was obtained by fitting experimental data.

[0056] Step 5: Slope smoothness and road surface condition assessment. In addition to slope calculation, this embodiment also assesses the slope smoothness and road surface condition to provide more comprehensive environmental information for subsequent braking control.

[0057] Slope smoothness assessment is based on lidar point cloud data, calculating the flatness error of the slope. The specific method involves selecting multiple sampling areas on the slope and calculating the standard deviation of the distance from the point cloud data within each sampling area to the fitted plane. The magnitude of the standard deviation characterizes the smoothness of the slope. A larger standard deviation indicates a more uneven slope, resulting in a stronger sense of bumpiness when vehicles are driving on it.

[0058] Road surface condition assessment is based on visual images and environmental sensor data. Road surface texture features are extracted from visual images, and deep learning algorithms are used to identify the dryness or wetness of the road surface. Temperature and humidity data are read from environmental sensors to determine the risk of icing. By comprehensively analyzing multi-source information, an estimated coefficient of adhesion is output, ranging from 0.1 to 0.9, with smaller values ​​indicating a wetter and more slippery road surface.

[0059] Technical details In this embodiment, several key technical details need to be explained in detail.

[0060] Firstly, there's the multi-sensor time synchronization technology. Because different sensors have different sampling frequencies and data processing delays, directly using data from each sensor would lead to time inconsistencies. This embodiment employs a combination of hardware and software synchronization. For hardware synchronization, the PTP protocol is used to synchronize the clocks of each sensor, achieving sub-microsecond accuracy. For software synchronization, an interpolation algorithm is used to align the data from low-frequency sensors to the timestamps of high-frequency sensors. This dual synchronization mechanism ensures the time consistency of multi-sensor data.

[0061] Secondly, there's the coordinate system technology. Since each sensor is installed in a different location and uses a different coordinate system, directly using this data can lead to spatial inconsistencies. This embodiment uses a rigid body transformation method to transform the local coordinate systems of each sensor to a unified vehicle coordinate system. The transformation parameters (rotation matrix and translation vector) are obtained through sensor calibration, and the calibration accuracy directly affects the final fusion effect. This embodiment uses a high-precision calibration field for sensor calibration, controlling the calibration error to the millimeter level.

[0062] Secondly, the construction method of the feature library for underground parking scenes is discussed. The quality of the feature library directly affects the accuracy of scene recognition. This embodiment uses a combination of real data acquisition and simulation generation to construct the feature library. For real data acquisition, multi-sensor data is collected by driving a vehicle in an actual underground parking scene, covering different types of parking lots and various environmental conditions. For simulation generation, data for various virtual scenes are generated based on the CAD model and sensor model of the real parking lot, supplementing boundary situations that are difficult to cover with real data. Through this combined real and virtual approach, a comprehensive and robust feature library is constructed.

[0063] Finally, error analysis and compensation for slope measurement are addressed. The main sources of error in slope measurement include sensor measurement errors, installation errors, and algorithm errors. This embodiment analyzes the characteristics of various error sources through extensive experiments and establishes an error model. For systematic errors (such as installation errors), calibration and compensation are used to eliminate them; for random errors (such as sensor noise), filtering algorithms are used to suppress them; and for state-related errors (such as errors caused by vehicle dynamics), an adaptive compensation mechanism is used for real-time correction. This comprehensive error processing method significantly improves the accuracy and robustness of slope measurement.

[0064] Data Flow Description The data flow in this embodiment is clear and unambiguous. The raw data stream flows from each sensor to the scene recognition and slope calculation module. After processing steps such as preprocessing, feature extraction, scene matching, and slope calculation, the module finally outputs parameters such as scene recognition results, slope values, slope smoothness, and road surface condition. These parameters are transmitted to the descent control decision module via the CAN bus as input for subsequent control decisions.

[0065] Several key feedback mechanisms exist throughout the data stream. First, scene recognition results are fed back to the feature extraction module to dynamically adjust feature extraction parameters and strategies, improving recognition accuracy. Second, slope calculation results are fed back to the sensor data preprocessing module to optimize the weight configuration of the data fusion algorithm, improving calculation accuracy. Third, road surface condition assessment results are fed back to the energy recovery coordination module to adjust the energy recovery strategy, ensuring braking safety is prioritized under low-adhesion conditions such as wet and slippery roads.

[0066] Specific Implementation Example 2: Adaptive Hierarchical Decision Control and Energy Recovery Synergy Implementation scenario description This embodiment addresses the adaptive control and energy recovery issues of large new energy vehicles during steep slope descent in underground parking garages. The specific scenario is as follows: the vehicle is already traveling on a steep slope in the garage and needs to descend smoothly at a constant low speed (e.g., 3-5 km / h) while recovering as much braking energy as possible. The challenge in this scenario is that the slope angle may vary between 8-15 degrees, the road surface adhesion coefficient may vary between 0.3-0.8, and the battery SOC may vary between 20%-90%. The control strategy needs to be adaptively adjusted while ensuring braking safety.

[0067] Technical Configuration Description The control system in this embodiment is based on the vehicle control unit (VCU, the core control unit of a new energy vehicle), and uses a 32-bit high-performance processor with a main frequency of 200MHz (megahertz, a unit of frequency, 1MHz=10). 6 The system features floating-point arithmetic capabilities and hardware acceleration units. The control software is developed using a model-based design approach, employing MATLAB / Simulink (MATLAB for Matrix Laboratory, a high-level computing language and interactive environment for algorithm development, data analysis, and numerical computation; Simulink is a component of MATLAB used for multi-domain simulation and model-based design of dynamic and embedded systems) to automatically generate C code, ensuring the reliability and real-time performance of the control algorithm.

[0068] The braking system employs a hybrid braking architecture that combines motor braking and mechanical braking. Motor braking is achieved by a drive motor; in this embodiment, a permanent magnet synchronous motor (PMSM) is used, with a peak power of 200 kW and a peak torque of 400 N·m, featuring brake regenerative braking with a maximum regenerative torque of 200 N·m. Mechanical braking utilizes an electro-hydraulic braking system (EHB), where a motor-driven hydraulic pump establishes braking pressure. All four wheels are independently controlled, with a braking pressure range of 0-10 MPa (megapascals, a unit of pressure, 1 MPa = 10 N·m). 6 Pa), response time less than 100ms (milliseconds).

[0069] The energy recovery system works in conjunction with the power battery system. The power battery uses lithium-ion batteries with a rated voltage of 400V (volts) and a capacity of 100kWh (kilowatt-hours, energy unit, 1kWh = 3.6 × 10⁻⁶). 6(Joules), with fast charging capability. The Battery Management System (BMS) monitors battery status, protects the battery, and manages battery energy in real time, monitoring parameters such as battery voltage, current, and temperature, and calculating the battery's SOC (State of Charge, representing the percentage of remaining battery capacity; SOC=100% indicates the battery is fully charged) value and health status, providing a basis for energy recovery strategies.

[0070] Detailed Explanation of Implementation Steps The implementation process of this embodiment is carried out according to the following steps: Step 1: Control Parameter Initialization. After the system identifies a steep slope scenario in the parking garage and completes the slope calculation, the descent control decision module initializes the control parameters based on the scenario identification results and slope parameters. Initial parameters include: target vehicle speed, braking torque distribution ratio, energy recovery intensity, etc.

[0071] The initialization of the target vehicle speed adopts a scenario-adaptive strategy. For steep slopes in older underground parking garages with strong enclosure and poor visibility, the initial target vehicle speed is set to a lower value (e.g., 2-3 km / h); for steep slopes in newly built underground parking garages with better visibility, the initial target vehicle speed can be set to a higher value (e.g., 5-8 km / h). This differentiated setting fully considers the safety requirements of different underground parking garage scenarios.

[0072] The initialization of the braking torque distribution ratio adopts a slope adaptation strategy. Based on the measured slope value, the component of the vehicle's weight along the slope direction (sliding force) is calculated to determine the required braking force. For steep slopes with large gradients (such as above 12 degrees), the mechanical braking distribution ratio is increased to ensure sufficient braking capacity; for steep slopes with small gradients (such as 8-10 degrees), electric motor braking is mainly used to improve energy recovery efficiency.

[0073] The initialization of energy recovery intensity adopts a battery state adaptation strategy. It reads the SOC value and battery temperature provided by the battery management system to determine the battery's charging acceptance capability. When the SOC value is below 80% and the battery temperature is within the normal operating range (0-45 degrees Celsius), the energy recovery intensity is set to the maximum value; when the SOC value is high or the battery temperature is abnormal, the energy recovery intensity is appropriately reduced to avoid overcharging or damage to the battery.

[0074] Step 2: Real-time Status Monitoring and Adaptive Parameter Adjustment. During the descent, the system continuously monitors the vehicle and environmental conditions and dynamically adjusts the control parameters.

[0075] Vehicle speed monitoring employs a fusion method combining wheel speed sensors and vehicle attitude sensors. Wheel speed sensors provide the rotational speeds of the four wheels, and the average speed is calculated to obtain the vehicle's longitudinal velocity. Vehicle attitude sensors provide acceleration information, and the speed estimate is obtained through integration. The fusion of these two methods yields a more accurate vehicle speed estimate, which is then used for closed-loop control.

[0076] Tire slippage detection is achieved by comparing the wheel speeds of the four wheels. When the wheel speed of a particular wheel is significantly lower than that of the other wheels (wheel speed difference greater than 10%), it is determined that the wheel is showing signs of slippage. At this point, the system immediately reduces the braking force on that wheel and, if necessary, cuts off energy recovery to prioritize vehicle stability.

[0077] Road condition monitoring relies on data from visual cameras and environmental sensors. The visual cameras continuously capture images of the road surface and use image recognition algorithms to detect changes in the road's wetness or dryness. Environmental sensors monitor temperature and humidity changes to determine the risk of icing. When a decrease in the road surface adhesion coefficient is detected, the system automatically reduces the target vehicle speed to increase braking safety margin.

[0078] Step 3: Dynamic distribution and execution of braking torque. Based on real-time monitored status information, the system dynamically adjusts the torque distribution between motor braking and mechanical braking.

[0079] Motor braking priority strategy: Under the premise of ensuring braking safety, motor braking is used first. Motor braking has the advantages of fast response, precise control, and energy recovery. The system calculates the required braking force in real time. When the motor braking capacity meets the demand, motor braking is used alone; when the motor braking capacity is insufficient (such as when the gradient is too large or the vehicle speed is too high), mechanical braking automatically intervenes to supplement the braking force.

[0080] Segmented Mechanical Braking Strategy: To avoid excessive mechanical braking that could cause vehicle skidding or passenger discomfort, a segmented application strategy is adopted for mechanical braking. Based on the required braking force, the braking pressure is divided into multiple levels (e.g., 0.1MPa, 0.3MPa, 0.6MPa, 1.0MPa), and applied progressively. Simultaneously, a low-pass filter is used to smooth the braking pressure command, preventing sudden pressure changes.

[0081] Braking Coordination Control: Precise coordination is required between electric motor braking and mechanical braking. The system monitors the actual output torque of the electric motor braking in real time. When the electric motor braking torque reaches its upper limit or a malfunction occurs, the mechanical braking immediately takes over to ensure continuous braking. Simultaneously, the current braking mode is displayed to the driver via the instrument panel, improving system transparency.

[0082] Step 4: Adaptive Energy Recovery Control. The energy recovery strategy is closely coupled with braking control and needs to be dynamically adjusted according to braking demand and battery status.

[0083] Maximum energy recovery control: When the battery SOC value is low (e.g., below 60%) and the temperature is normal, the system operates at maximum energy recovery intensity to recover as much braking energy as possible. At this time, the motor braking torque is mainly determined by the energy recovery demand, and the braking control strategy needs to adapt to this torque characteristic.

[0084] Energy recovery limitation control: When the battery SOC value is high (e.g., above 80%) or the temperature is abnormal, the battery's charging acceptance capability decreases, and the system needs to limit the energy recovery intensity. A linear reduction strategy is adopted, with the energy recovery intensity decreasing by 20% for every 5% increase in SOC value, until energy recovery is completely stopped.

[0085] Energy recovery cut-off control: When tire slippage or emergency braking demand is detected, the system immediately cuts off energy recovery and switches the motor braking to pure braking mode to prioritize braking safety. This reflects the principle of safety first.

[0086] Step 5: Coordinated Optimization of Braking Control and Energy Recovery. This is the core innovative step of this embodiment. By establishing a unified optimization framework, it achieves coordinated optimization of braking safety and energy recovery efficiency, such as... Figure 5 As shown, this demonstrates the adaptive capability of the invention under different operating conditions. When the system detects a dry road surface and a low battery SOC, it automatically increases the motor braking ratio to improve energy recovery efficiency; when it detects a slippery road surface or a high battery SOC, it automatically increases the mechanical braking ratio to prioritize braking safety. This ability to dynamically adjust the strategy based on real-time operating conditions enables the system to adapt to complex and ever-changing underground parking environments, providing optimal descent control performance under various conditions.

[0087] The basic idea of ​​collaborative optimization is to solve a multi-objective optimization problem within each control cycle based on the current vehicle state (vehicle speed, acceleration, slope, etc.), battery state (SOC, temperature, etc.) and road surface state (adhesion coefficient, etc.) to obtain the optimal distribution scheme of motor braking torque and mechanical braking torque.

[0088] Specific steps for collaborative optimization: 5.1 State Acquisition and Prediction. The vehicle state vector x_k = [v_k, a_k, θ_k, μ_k, SOC_k, T_batt_k] at the current time k is acquired, where v_k is the vehicle speed, a_k is the acceleration, θ_k is the slope, μ_k is the road adhesion coefficient, SOC_k is the battery state of charge, and T_batt_k is the battery temperature. Based on the vehicle dynamics model, the vehicle's state trajectory over the next N time steps is predicted.

[0089] 5.2 Braking Demand Calculation. Based on the target vehicle speed v_target and the current vehicle speed v_k, calculate the braking force F_brake_req required to maintain the target vehicle speed: F_brake_req = m·g·sinθ - F_roll - F_aero Where m is the vehicle mass, g is the gravitational acceleration (approximately 9.8 m / s²), F_roll is the rolling resistance, and F_aero is the air resistance.

[0090] 5.3 Battery State Assessment. Based on the current SOC and battery temperature, calculate the battery's maximum charging power P_charge_max and maximum charging current I_charge_max: P_charge_max = f(SOC, T_batt) I_charge_max = g(SOC, T_batt) The functions f(·) and g(·) were obtained by fitting experimental data on battery characteristics.

[0091] 5.4 Cooperative Optimization Solution. A unified cost function J is established, comprehensively considering multiple objectives such as braking error, energy recovery, and ride comfort: Detailed formula for the cost function: J = w1·J_brake + w2·J_energy + w3·J_comfort + w4·J_stability The definitions of each item are as follows: Braking error term J_brake: J_brake = Σᵢ₌1ᴺ (v_target - v_i)² This ensures that the actual vehicle speed tracks the target vehicle speed, guaranteeing braking safety.

[0092] Energy recovery item J_energy: J_energy = -α·η·P_recover Where P_recover is the actual recovered power, η is the energy conversion efficiency, and α is the energy recovery weighting coefficient. The negative sign indicates maximizing energy recovery. P_recover is calculated as follows: P_recover = T_motor·ω_motor·η_motor Where T_motor is the motor braking torque, ω_motor is the motor speed, and η_motor is the motor efficiency.

[0093] Ride comfort item J_comfort: J_comfort = Σᵢ₌1ᴺ (da_i / dt)² + Σᵢ₌1ᴺ (ΔT_i)² The first term represents the rate of change of acceleration (jerk, the rate of change of acceleration), reflecting longitudinal impact; the second term represents the change in braking torque, reflecting braking smoothness.

[0094] Stability term J_stability: J_stability = Σᵢ₌1ᴺ (λ_i - λ_opt)² Where λ_i is the slip ratio of each wheel, and λ_opt is the optimal slip ratio (usually 0.15-0.20). This term ensures that the wheels do not slip, thus guaranteeing vehicle stability.

[0095] Constraints: The optimization problem also needs to satisfy the following constraints: T_motor_min ≤ T_motor ≤ T_motor_max T_mech_min ≤ T_mech ≤ T_mech_max T_motor + T_mech = T_brake_req I_charge ≤ I_charge_max P_charge ≤ P_charge_max 0 ≤ λ_i ≤ λ_critical Where T_motor is the motor braking torque, T_motor_min and T_motor_max are the minimum and maximum motor braking torques respectively, T_mech is the mechanical braking torque, T_mech_min and T_mech_max are the minimum and maximum mechanical braking torques respectively, T_brake_req is the total braking torque demand, I_charge is the charging current, I_charge_max is the maximum charging current, P_charge is the charging power, P_charge_max is the maximum charging probability, λ_i is the wheel slip ratio, and λ_critical is the critical slip ratio.

[0096] It should be noted that the total braking demand torque T_brake_req (N·m) is the rotational torque acting on the wheel (brake actuator).

[0097] The required braking force F_brake_req (N) is the translational force acting on the vehicle's center of gravity (vehicle dynamics control). Conversion relationship: F_brake_req = T_brake_req / r_wheel (r_wheel is the wheel radius) F_brake_req: Obtained based on vehicle dynamics balance calculations (considering gravity components, rolling resistance, and air resistance). T_brake_req: Obtained based on braking force demand conversion, used for braking system control.

[0098] 5.5 Adaptive Weight Adjustment. The weights w1, w2, w3, and w4 of each objective item are dynamically adjusted based on the current operating conditions. When a slippery road surface is detected (μ < 0.4), increase w4 (stability weight) and decrease w2 (energy recovery weight).

[0099] When the battery SOC is high (SOC > 80%), reduce w2 (energy recovery weight).

[0100] Increase w1 (braking accuracy weight) when approaching the bottom of the slope or when stopping is required.

[0101] 5.6 Optimization Solution and Execution. The above-mentioned constrained optimization problem is solved using either Sequential Quadratic Programming (SQP) or Particle Swarm Optimization (PSO), yielding the optimal control sequence u* = {T_motor}. , T_mech Apply the first control command and repeat the above process in the next control cycle.

[0102] Step 6: System Status Monitoring and Fault Handling. The system continuously monitors the working status of each module and promptly detects and handles faults.

[0103] Sensor fault detection: Sensor faults are detected through methods such as data validity checks and cross-validation of data between sensors. When a sensor fault is detected, the system uses data from redundant sensors or an estimated value to replace it, ensuring continuous system operation.

[0104] Braking system fault detection: Monitors key parameters such as braking pressure and braking torque. When braking failure or a severe decrease in braking capacity is detected, the system immediately activates the emergency braking mode, alerts the driver with audible and visual alarms, and attempts to restore braking function.

[0105] Energy recovery fault detection: Monitor parameters such as energy recovery current and voltage. When an abnormality in energy recovery is detected (such as low recovery efficiency or battery overvoltage), the system cuts off energy recovery and uses pure mechanical braking to ensure braking safety.

[0106] Technical details Several innovative technical details in this embodiment need to be highlighted.

[0107] First is the adaptive hierarchical decision-making mechanism. For example... Figure 4As shown, this mechanism decomposes the complex control problem into three relatively independent layers: the scene adaptation layer, the slope adaptation layer, and the vehicle status adaptation layer. The scene adaptation layer is responsible for determining the basic control parameters based on the macroscopic characteristics of the parking garage scene (such as enclosure and visibility conditions); the slope adaptation layer is responsible for adjusting the control parameters based on the microscopic changes in slope; and the vehicle status adaptation layer is responsible for fine-grained control based on the real-time status of the vehicle (such as vehicle speed and wheel speed). This layered structure ensures both the comprehensiveness and real-time performance of the control.

[0108] exist Figure 3 In this hierarchical decision-making framework, comfort control is integrated across all levels. The scenario adaptation layer presets basic comfort parameters based on the type of parking garage; the slope adaptation layer adjusts the smoothness of braking application based on the rate of change of slope; and the vehicle status adaptation layer reduces longitudinal impact by limiting the rate of change of acceleration (jerk). This multi-layered comfort control strategy makes the descent process both smooth and seamless, significantly improving the passenger experience.

[0109] Secondly, there is the dynamic braking torque distribution algorithm. This algorithm is based on Model Predictive Control (MPC) theory, establishing a vehicle dynamics model and a braking system model to predict the vehicle's state over a future period and optimize the braking torque distribution. The algorithm considers various constraints, such as braking capacity constraints, tire adhesion constraints, and comfort constraints, optimizing energy recovery efficiency while ensuring safety. The algorithm employs a rolling optimization strategy, recalculating the optimal control sequence in each control cycle, thus improving the robustness of the control (robustness refers to the system's ability to maintain stable performance under parameter perturbations or external disturbances).

[0110] Secondly, there is the coordinated optimization of energy recovery and braking control. In traditional designs, the energy recovery system and braking system are independent; the energy recovery strategy does not consider braking demand, and the braking strategy does not consider energy recovery efficiency. This embodiment unifies both under a single optimization framework, simultaneously optimizing braking safety and energy recovery efficiency. By establishing a unified cost function and comprehensively considering multiple objectives such as braking error, energy recovery, and ride comfort, a multi-objective optimization algorithm is used to solve for the optimal control strategy. This coordinated design significantly improves the overall performance of the system.

[0111] Specifically, the cost function comprehensively considers multiple conflicting objectives. The braking error term ensures accurate speed tracking of the target value, guaranteeing safety; the energy recovery term maximizes braking energy recovery, improving fuel economy; the comfort term limits the rate of change of acceleration and torque, enhancing the ride experience; and the stability term prevents wheel slippage, ensuring vehicle stability. Through an adaptive weight adjustment mechanism, the system can dynamically balance the importance of each objective according to different operating conditions. For example, on slippery roads, the stability weight increases, while the energy recovery weight decreases; when the battery is fully charged, the energy recovery weight decreases, while the braking accuracy weight increases. This dynamic balancing mechanism ensures that the system achieves optimal overall performance under various operating conditions.

[0112] Finally, there's fault diagnosis and fault-tolerant control. The system employs a hierarchical fault diagnosis strategy, combining sensor-level fault detection, module-level fault detection, and system-level fault detection to form a comprehensive fault detection network. Once a fault is detected, the system immediately activates the fault-tolerant control mode, adopting different fault-tolerant strategies based on the severity of the fault. For minor faults, sensor data fusion or estimated values ​​are used as substitutes; for severe faults, the system switches to emergency mode or prompts the driver to take over. This hierarchical fault-tolerant mechanism significantly improves the system's reliability.

[0113] Data Flow Description In this embodiment, the data flow forms a complete closed-loop control circuit. Sensing data flows from multiple sensors to the scene recognition and slope calculation module; the recognition results and calculation parameters flow to the descent control decision module. The decision module generates control commands, which flow to the braking execution module and the energy recovery coordination module, respectively. Response data from the execution modules (such as actual braking torque and actual vehicle speed) is fed back to the decision module, forming a closed-loop control. Simultaneously, the fault diagnosis module monitors the status data of all modules and triggers an emergency handling procedure when an anomaly is detected.

[0114] Within the data stream, there are several key feedback loops. First, the vehicle speed feedback loop: the actual vehicle speed is fed back to the decision-making module, compared with the target speed, forming a closed-loop control to ensure stable vehicle speed. Second, the wheel speed feedback loop: the wheel speeds of all four wheels are fed back to the decision-making module for slippage detection and braking force distribution, improving vehicle stability. Third, the battery status feedback loop: battery SOC and temperature are fed back to the energy recovery module for adjusting the energy recovery strategy and optimizing energy recovery efficiency.

[0115] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for hill descent control of large new energy vehicles based on multiple sensors, characterized in that, Includes the following steps: S01: Acquire data from multiple sensors on the vehicle, and obtain multi-sensor data after preprocessing; S02: Extract scene features based on multi-sensor data, match the extracted scene features with the scene features in the preset underground parking scene feature library, determine whether it is an underground parking steep slope scene, and classify the underground parking scene. S03: After identifying a steep slope scene in the underground parking lot, the slope is calculated to obtain the slope parameters; S04: Generate an adaptive descent control strategy based on the parking lot scene classification results and slope parameters. The control strategy includes: The control is decomposed into different layers for adaptive hierarchical decision-making. The scenario adaptation layer is used to determine the basic control parameters based on the scenario characteristics of the underground parking lot; the slope adaptation layer is used to adjust the control parameters according to the slope changes; and the vehicle status adaptation layer is used to make dynamic adjustments based on the real-time status of the vehicle. Establish vehicle dynamics and braking system models to predict vehicle status over a future period of time. A unified cost function is established to coordinate the optimization of energy recovery and braking control. In each control cycle, a multi-objective optimization problem is solved based on the vehicle state, battery state, and road surface state to obtain the optimal distribution of motor braking torque and mechanical braking torque.

2. The method for hill descent control of large new energy vehicles based on multiple sensors according to claim 1, characterized in that, The preprocessing in step S01 includes: Remove data noise; The PTP protocol is used to synchronize the clocks of each sensor, and an interpolation algorithm is used to align the data of the low-frequency sensor to the timestamp of the high-frequency sensor. The rigid body transformation method is used to transform the local coordinate system of each sensor to a unified vehicle coordinate system. The transformation parameters are obtained through sensor calibration.

3. The method for hill descent control of large new energy vehicles based on multiple sensors according to claim 1, characterized in that, Scene features extracted from multi-sensor data include: Extract light intensity and texture features from visual images, and calculate the average brightness value of the image; Spatial enclosure features are extracted from lidar point cloud data, and the existence of lateral occlusion is determined by analyzing the spatial distribution of the point cloud data. Road surface material features are extracted from the texture features of visual images and the spatial closure features of LiDAR data, and road surface types are identified based on texture features and reflectivity. The slope change features are extracted from the vehicle attitude sensor data, and the rate of change of pitch angle is used to determine whether the vehicle is going downhill.

4. The method for hill descent control of large new energy vehicles based on multiple sensors according to claim 1, characterized in that, A multi-dimensional comprehensive evaluation method is used to determine whether a parking garage has a steep slope, including: Calculate the similarity between the features of the scene to be identified and the features of each scene in the basement scene feature database. The similarity is calculated using weighted Euclidean distance, and the weights of different scene features are determined according to their contribution to scene identification. When the similarity exceeds a preset threshold, the scene is determined to be a steep slope scene in a parking garage.

5. The method for hill descent control of large new energy vehicles based on multiple sensors according to claim 1, characterized in that, In step S03, slope calculation is performed to obtain slope parameters including: Slope point cloud is extracted from lidar point cloud data, and the slope plane is fitted using the random sampling consensus algorithm to obtain the slope estimate based on lidar. Pitch angle data is read from the vehicle attitude sensor and smoothed by Kalman filtering to obtain the slope estimate based on the vehicle attitude sensor. An adaptive weighted fusion algorithm is used to fuse the two slope estimates and output the fused slope value. By analyzing historical data, a mapping relationship between slope measurement error and vehicle status and environmental conditions is established, resulting in an error model. The slope value is compensated in real time. The error includes systematic error, random error and state-related error. Systematic error is eliminated through calibration and compensation, random error is suppressed through filtering algorithm, and state-related error is corrected in real time through adaptive compensation mechanism.

6. The method for hill descent control of large new energy vehicles based on multiple sensors according to claim 1, characterized in that, After obtaining the slope parameters, the assessment also includes slope smoothness and road surface condition evaluation: The slope flatness assessment is based on lidar point cloud data. The flatness of the slope is calculated by selecting multiple sampling areas on the slope and calculating the standard deviation of the distance from the point cloud data in each sampling area to the fitted plane. The magnitude of the standard deviation is used to characterize the flatness of the slope. The road surface condition assessment is based on visual images and environmental sensor data. The specific method is as follows: extract road surface texture features from visual images and use deep learning algorithms to identify the dry and wet state of the road surface; read temperature and humidity data from environmental sensors to determine whether there is a risk of icing; and calculate the estimated value of the road surface adhesion coefficient.

7. The method for hill descent control of large new energy vehicles based on multiple sensors according to claim 1, characterized in that, Establishing a unified cost function and co-optimizing energy recovery and braking control includes: Collect the vehicle state vector x_k = [v_k, a_k, θ_k, μ_k, SOC_k, T_batt_k] at the current time k, where v_k is the vehicle speed, a_k is the acceleration, θ_k is the slope, μ_k is the road adhesion coefficient, SOC_k is the battery state of charge, and T_batt_k is the battery temperature; based on the vehicle dynamics model, predict the vehicle state trajectory in the next N time steps. Calculate the braking force F_brake_req required to maintain the target vehicle speed based on the target vehicle speed v_target and the current vehicle speed v_k: F_brake_req = m·g·sinθ_k - F_roll - F_aero Where m is the vehicle mass, g is the gravitational acceleration, F_roll is the rolling resistance, and F_aero is the air resistance; Based on the current SOC and battery temperature, calculate the battery's maximum charging power P_charge_max and maximum charging current I_charge_max; Establish a unified cost function J: J = w1·J_brake + w2·J_energy + w3·J_comfort + w4·J_stability Among them, w1, w2, w3, and w4 are weights, J_brake is the braking error term, J_energy is the energy recovery term, J_comfort is the ride comfort term, and J_stability is the stability term; Constraints: T_motor_min ≤ T_motor ≤ T_motor_max T_mech_min ≤ T_mech ≤ T_mech_max T_motor + T_mech = T_brake_req I_charge ≤ I_charge_max P_charge ≤ P_charge_max 0 ≤ λ_i ≤ λ_critical Where T_motor is the motor braking torque, T_motor_min and T_motor_max are the minimum and maximum motor braking torques respectively, T_mech is the mechanical braking torque, T_mech_min and T_mech_max are the minimum and maximum mechanical braking torques respectively, T_brake_req is the total braking torque demand, I_charge is the charging current, I_charge_max is the maximum charging current, P_charge is the charging power, P_charge_max is the maximum charging probability, λ_i is the wheel slip ratio, and λ_critical is the critical slip ratio.

8. The method for hill descent control of large new energy vehicles based on multiple sensors according to claim 7, characterized in that, Braking error term J_brake: J_brake = Σᵢ₌1ᴺ (v_target - v_i)² v_i represents the actual vehicle speed; Energy recovery item J_energy: J_energy = -α·η·P_recover Where P_recover is the actual recovered power, η is the energy conversion efficiency, α is the energy recovery weighting coefficient, and the negative sign indicates maximizing energy recovery. P_recover is calculated as follows: P_recover = T_motor·ω_motor·η_motor Where T_motor is the motor braking torque, ω_motor is the motor speed, and η_motor is the motor efficiency; Ride comfort item J_comfort: J_comfort = Σᵢ₌1ᴺ (da_i / dt)² + Σᵢ₌1ᴺ (ΔT_i)² a_i represents the acceleration, and ΔT_i represents the change in braking torque; Stability term J_stability: J_stability = Σᵢ₌1ᴺ (λ_i - λ_opt)² Where λ_i is the slip ratio of each wheel, and λ_opt is the optimal slip ratio.

9. A multi-sensor-based hill descent control system for large new energy vehicles, used to implement the multi-sensor-based hill descent control method for large new energy vehicles as described in any one of claims 1-8, characterized in that, include: The multi-sensor perception module acquires data collected by multiple sensors on the vehicle and preprocesses it to obtain multi-sensor data. The scene recognition module extracts scene features based on multi-sensor data, matches the extracted scene features with each scene feature in the preset underground parking scene feature library, determines whether it is an underground parking steep slope scene, and classifies the underground parking scene. The slope calculation module, when identified as a steep slope scene in a parking garage, performs slope calculation to obtain slope parameters; The descent control decision module generates an adaptive descent control strategy based on the parking garage scenario classification results and slope parameters. The control strategy includes: The control is decomposed into different layers for adaptive hierarchical decision-making. The scenario adaptation layer is used to determine the basic control parameters based on the scenario characteristics of the underground parking lot; the slope adaptation layer is used to adjust the control parameters according to the slope changes; and the vehicle status adaptation layer is used to make dynamic adjustments based on the real-time status of the vehicle. Establish vehicle dynamics and braking system models to predict vehicle status over a future period of time. A unified cost function is established to coordinate the optimization of energy recovery and braking control. In each control cycle, a multi-objective optimization problem is solved based on the vehicle state, battery state, and road surface state to obtain the optimal distribution of motor braking torque and mechanical braking torque.

10. A computer storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the multi-sensor-based hill descent control method for large new energy vehicles as described in any one of claims 1-8.

Images