An intelligent vehicle speed control method and system suitable for a mountain road

Abstract

The application discloses an intelligent vehicle speed control method and system suitable for a mountain road, and belongs to the technical field of vehicle control. A perception fusion module of the system collects road conditions, environment, vehicle states and driver intention data through a multi-source sensor, constructs a plurality of core coupling models to quantify the interaction of factors such as altitude-power and temperature-adhesion coefficient, and identifies the main / auxiliary scene labels of a composite scene based on a quantitative threshold. A forward decision module predicts the road condition and vehicle state trend by using an LSTM, constructs a multi-objective optimization function with a model predictive control algorithm as the core, dynamically adjusts the weights of safety, energy consumption and comfort targets according to the scene labels, and generates a scene-adaptive vehicle speed control strategy. An execution feedback module analyzes the instruction and drives the execution component to act, and simultaneously collects actual execution data to form a closed-loop optimization. The application realizes fine and forward-looking control of complex working conditions of the mountain road, and significantly improves the driving safety, energy efficiency and comfort.

Description

Technical Field

[0001] This invention relates to the field of vehicle control technology, specifically to an intelligent vehicle speed control method and system suitable for mountain roads. Background Technology

[0002] When vehicles travel on complex terrains such as mountain roads, they face multiple challenges, including climbing, descending, sharp bends, steep sections, altitude changes, and unpredictable weather. Existing vehicle control systems, such as cruise control or adaptive cruise control, are primarily designed for flat highways and have the following significant shortcomings: First, the control logic is relatively simplistic. Existing control logic typically considers only one or a few factors (such as distance to the vehicle ahead), failing to integrate and analyze multi-dimensional information such as road geometry, altitude, temperature, and weather, and neglecting the coupling effects between various factors. For example, high altitude leads to a decrease in engine power, and this effect is exacerbated when going uphill; rain and snow significantly reduce tire grip on downhill slopes or curves; changes in ambient temperature affect engine thermal efficiency and tire performance, etc.

[0003] Secondly, the control strategy suffers from lag. Existing control technologies are mostly reactive, meaning the control system only intervenes after abnormal conditions such as engine overheating or excessive downhill speed occur, lacking the ability to anticipate road conditions ahead. This lag not only leads to unstable vehicle handling and increased energy consumption but also poses significant safety hazards.

[0004] Furthermore, the system lacks the capability to handle special operating conditions. For uphill driving on mountain roads, continuous uphill driving can easily lead to overheating of the engine and drive system, and the existing system lacks effective predictive thermal management and load control strategies. For downhill driving on mountain roads, drivers are accustomed to frequently using friction brakes to control vehicle speed. This operation not only increases the risk of brake fade but also increases energy consumption. For electric vehicles, it also causes them to miss the opportunity for energy recovery through regenerative braking.

[0005] Therefore, there is an urgent need in this field to develop an intelligent vehicle speed control method and system suitable for driving scenarios on mountain roads. This system should have real-time perception capabilities, be able to comprehensively analyze various environmental factors, and achieve forward-looking optimal control based on the analysis results. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent vehicle speed control method and system suitable for mountain roads. It can use multi-sensor fusion technology to perceive road characteristics, altitude, temperature and weather information in real time and accurately, and intelligently handle the interaction between these factors. By using model predictive control algorithms, it can achieve forward-looking and multi-objective optimization control of vehicle speed, thereby significantly improving driving safety, reducing energy consumption, improving ride comfort, and effectively solving the problems of engine overheating uphill and excessive reliance on friction braking downhill.

[0007] To achieve the above objectives, the technical solution adopted by this invention is: an intelligent vehicle speed control method suitable for mountain roads, comprising the following steps: S1. Perception Fusion Stage: Collect multi-source data on vehicle driving environment, vehicle status and driver intention, preprocess the multi-source data, and run multiple core coupling models to quantify the coupling effect between different factors. Based on the characteristics of the mountain road scene, link and call the multiple core coupling models to output a standardized fusion information package containing road condition features, environmental parameters, vehicle status and driver intention. S2, Forward Decision-Making Stage: Based on the standardized fusion information package, the prediction model predicts the trends of road conditions, environment, and vehicle status ahead, identifies the current driving scenario, and generates composite scenario labels for the main scenario and auxiliary scenario. A multi-objective optimization function containing safety, energy consumption, and comfort objectives is constructed using a model predictive control algorithm. The weights of each objective item in the multi-objective optimization function are dynamically adjusted according to the composite scenario labels of the main scenario and auxiliary scenario to solve and generate the optimal control strategy. S3. Execution Feedback Stage: The optimal control strategy is parsed into underlying control signals and the vehicle's execution components are driven to perform actions. Real-time vehicle status feedback data after execution is collected. The feedback data is compared with the predicted value. If the deviation exceeds a preset threshold, a new decision is triggered to form a closed-loop optimization.

[0008] Furthermore, running multiple core coupling models in step S1 includes: Based on real-time altitude H and road slope θ, the engine power attenuation coefficient is calculated using an altitude-engine power coupling model. ; The tire adhesion coefficient is calculated using a temperature-tire adhesion coefficient coupling model based on temperature parameter T, road surface wetness / dryness S, and ambient humidity H. The temperature parameter T includes either the ambient temperature or the road surface temperature. Based on braking intensity B, road slope θ, and tire adhesion coefficient The rate of increase in brake temperature was calculated using a coupled model of brake temperature, brake intensity, and road condition parameters. and the predicted brake temperature within the preset distance ,in t represents the initial brake temperature, and t represents the driving time.

[0009] Furthermore, the step S2 of identifying the current driving scenario and generating a composite scenario label includes: Quantifiable core feature parameters are extracted, including road slope, slope length, curve radius of curvature, continuous curve length, precipitation intensity, and altitude. Preset judgment thresholds for each feature parameter are used to determine the primary and secondary scenes based on the principle that the more the feature parameter value deviates from the normal range, the higher the scene priority. When the road slope exceeds the preset slope threshold, uphill or downhill is determined as the primary scene and curve as the secondary scene. When the curve radius of curvature is less than the preset curvature threshold, curve is determined as the primary scene and uphill or downhill as the secondary scene. When environmental feature parameters or vehicle state feature parameters reach extreme values ​​and their impact on driving safety exceeds the road condition geometry, a temporary switch to the primary scene is triggered. The output includes a composite scene standardized label containing primary and secondary scene labels, as well as primary scene feature weight values ​​and secondary scene feature weight values.

[0010] Further, the prediction by the prediction model in step S2 includes: using an LSTM model for prediction, wherein the input layer of the LSTM model assigns a first weight coefficient to the main scene feature parameters and a second weight coefficient to the auxiliary scene feature parameters, and the first weight coefficient is greater than the second weight coefficient; configuring feature extraction priority in the hidden layer so that the model prioritizes the temporal change analysis of the main scene features; and adopting a hierarchical output form of core results plus constraint results in the output layer, wherein the core output is the key trend prediction value of the main scene, and the constraint output is the key trend prediction value of the auxiliary scene.

[0011] Furthermore, step S2, which dynamically adjusts the weights of the multi-objective optimization function based on the composite scene label, includes: establishing a preset weight matrix of the main scene label and the weights of each objective item; when the output value of the core coupled model corresponding to the main scene deviates from the normal range, triggering an increase in the weight of the core objective item of the main scene, and positively correlated the degree of deviation with the magnitude of the weight increase; in the composite scene, obtaining the basic weight value by matching the preset weight matrix with the main scene label, compressing the weight of the objective item corresponding to the auxiliary scene, and transferring the compressed weight value to the core objective item of the main scene to achieve weight redistribution; based on the measured data of the execution feedback module, when the deviation between the actual value and the predicted value of the core objective of the main scene exceeds the threshold, the weight of the core objective item of the main scene is increased again, and the adjusted weight value is sent back to the preset weight matrix to achieve iterative optimization.

[0012] Furthermore, the multi-objective optimization function is expressed as: ,in For the security objective function, Let the energy consumption objective function be... For the comfort objective function, , , For contextualized weighting coefficients; The security objective function: ,in This represents the actual brake temperature. Here, s represents the safe threshold for brake temperature, and s represents the actual slip ratio. The slip ratio safety threshold, , These are the sub-objective weight coefficients; The energy consumption objective function is adopted for fuel vehicles. Where f is the actual fuel consumption rate, This represents the optimal fuel consumption rate under current operating conditions; for electric vehicles, this is the optimal fuel consumption rate. Where e is the actual power consumption rate, The optimal power consumption rate under the current operating conditions. For regenerative braking recovery rate, , These are the weighting coefficients; The comfort objective function ,in , These are the maximum and minimum vehicle speeds within a preset control period, respectively. The target speed.

[0013] Furthermore, the step S3 of comparing the feedback data with the predicted value includes: the perception fusion module compares the actual vehicle speed, engine or motor power, braking intensity, brake temperature and battery SOC data collected in real time by the feedback acquisition unit with the predicted value in the standardized fusion information package. If the deviation of the vehicle speed fluctuation, brake temperature and energy consumption index exceeds the preset threshold, the fusion information package is updated, the prediction result is corrected and the control strategy is adjusted. The actions of the drive vehicle's actuators include: for gasoline vehicles, power control is achieved by adjusting the throttle opening and fuel injection quantity through the engine control unit, automatic gear shifting is achieved through the transmission control unit, and friction braking is achieved by adjusting the brake master cylinder pressure through the brake control unit; for electric vehicles, regenerative braking and torque control are achieved by adjusting the motor reverse drag current through the motor control unit, battery charging current is controlled through the battery management system, and friction braking is initiated through the brake control unit when regenerative braking is insufficient; when an actuator temporarily fails, the system switches to a backup control mode within a preset time.

[0014] Furthermore, the multiple core coupling models also include an altitude-motor power coupling model, expressed as follows: Where H is the real-time altitude and SOC is the remaining battery power. This is the regenerative braking efficiency coefficient.

[0015] Furthermore, in step S1, when running multiple core coupling models, the corresponding core coupling model is called and the associated auxiliary coupling model is called in conjunction with the identified main scene type. For example, in the uphill scene, the altitude-engine power model and the temperature-tire adhesion coefficient model are called in conjunction with each other, and in the downhill scene, the temperature-tire adhesion coefficient model and the brake temperature-braking intensity-road condition parameter coupling model are called in conjunction with each other.

[0016] Furthermore, the prediction range of the LSTM model is adaptively adjusted according to the vehicle speed, and the prediction distance of the main scene is greater than the prediction distance of the auxiliary scene.

[0017] This application also provides an intelligent vehicle speed control system suitable for mountain roads, used to implement the above-mentioned method. The system adopts a three-level core architecture of perception fusion, forward decision-making, and execution feedback, including: The perception fusion module is used to collect multi-source data, run multiple core coupled models, and output standardized fusion information packages. The forward-looking decision-making module is signal-connected to the perception fusion module and is used to predict trends, identify composite scene labels, construct and solve multi-objective optimization functions based on the standardized fusion information package, and generate the optimal control strategy. The execution feedback module is signal-connected to the forward decision-making module. It is used to parse the optimal control strategy into underlying control signals, drive the action of the execution components, and collect feedback data in real time to send back to the perception fusion module to form closed-loop control.

[0018] Furthermore, the perception fusion module includes: The information acquisition unit adopts a redundant design of main sensor plus auxiliary sensor, and includes road condition perception subunit, environment perception subunit, vehicle status perception subunit and driver intention perception subunit. The fusion processing unit is used to preprocess sensor data using the Kalman filter algorithm, run multiple core coupling models to quantify the coupling effects between factors, and use DS evidence theory to fuse the preprocessed data with the analysis results of the coupling models to output a standardized fusion information package. The forward-looking decision-making module includes: The forward prediction unit is equipped with an AI chip and integrates an LSTM model trained with data from mountain roads to predict trends in road conditions, environmental changes, and vehicle status evolution. The decision control unit constructs a multi-objective optimization function with model predictive control algorithm as the core, dynamically adjusts the weight of objective items according to composite scene labels, identifies uphill, downhill and curve scenes and generates corresponding control strategies; The execution feedback module includes: The instruction parsing unit is used to parse the control instructions output by the forward decision-making module into low-level control signals that can be recognized by each execution component; The execution control unit is used to drive the action of the execution components according to the underlying control signals. It includes the engine control unit, transmission control unit, and brake control unit adapted to fuel vehicles, as well as the motor control unit, battery management system, and brake control unit adapted to electric vehicles. The feedback acquisition unit integrates a dedicated feedback sensor to collect and transmit core vehicle status data in real time after execution.

[0019] The beneficial effects of the above scheme are as follows: 1. Enhanced perception accuracy and coupled analysis capabilities lead to a more reliable foundation for decision-making data. This invention quantifies the dynamic interactions between factors such as altitude and engine power, and temperature and tire adhesion coefficient by constructing multiple core coupled models, transforming raw sensor data into high-order features reflecting vehicle operating mechanisms. This approach not only enables the system to accurately perceive environmental changes but also deeply characterizes the complex influence of multi-factor coupling on vehicle status, providing precise and comprehensive data support for subsequent decision-making and improving control reliability from the source.

[0020] 2. Combining forward-looking prediction with optimized decision-making enables more proactive risk control. A long short-term memory network trained on extensive mountain road data predicts trends in road conditions, the environment, and vehicle status ahead. This prediction is then combined with model predictive control algorithms for optimization, allowing the system to generate and adjust control commands in advance. This shift from reactive to predictive control ensures smooth speed and braking adjustments on long downhill slopes, continuous uphill sections, and curves, proactively identifying and mitigating risks such as overheating, brake fade, and skidding, significantly improving driving safety and smoothness.

[0021] 3. Enhanced scene adaptability enables refined control in complex scenarios. The system identifies primary and secondary scenarios within a complex scenario based on preset quantization thresholds and generates standardized primary / secondary scenario labels. The forward-looking decision-making module dynamically adjusts the weights of safety, energy consumption, and comfort objectives in the multi-objective optimization function based on these labels, allowing the control strategy to prioritize and selectively optimize the core pain points of different scenarios. This mechanism achieves refined control under complex conditions such as curves and slopes, effectively balancing multi-dimensional performance indicators and significantly improving the vehicle's adaptability to varying road conditions.

[0022] 4. Enhanced System Robustness and Self-Optimization Capabilities. The execution feedback module collects real-time data on actual vehicle speed, brake temperature, energy consumption, and other execution performance data, comparing them with the predicted values ​​from the fused information package. When the deviation exceeds a threshold, the system uses the deviation information to iteratively correct the parameters of the core coupled model and the weights of the multi-objective optimization function, forming a dynamic closed-loop optimization cycle of "collection-fusion-prediction-decision-execution-feedback." This ensures efficient integration of functions across modules, guaranteeing adaptability to core scenarios such as uphill overheating prevention, downhill braking protection and energy recovery, and curve smooth control, balancing technological advancement with practical applicability. Attached Figure Description

[0023] Figure 1 This is an overall flowchart of the intelligent vehicle speed control method applicable to mountain roads according to the present invention. Detailed Implementation

[0024] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0025] It should be noted that, unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0026] like Figure 1 As shown, an intelligent vehicle speed control method suitable for mountain roads includes the following steps: S1. Perception Fusion Stage: Multi-source data acquisition and coupling analysis.

[0027] This phase achieves full-dimensional data collection through "high-precision map-multi-sensor array". After preprocessing and analysis of multiple core coupled models, a standardized fusion information package is output to provide accurate and real-time input support for forward-looking decision-making.

[0028] 1. Comprehensive data collection After the vehicle starts and enters the mountain road, the perception fusion module is immediately activated, and the information acquisition unit completes the acquisition of multi-source heterogeneous data in the following ways: High-precision map base data: Load the current road segment's three-dimensional geographic information, including segmental elevation, slope, and curve parameters, as well as other core road condition features.

[0029] Road condition and obstacle perception: LiDAR and millimeter-wave radar are used together to detect the distance and relative speed of obstacles ahead; high-definition cameras acquire road markings and dry / wet conditions through image recognition, and the data from the three sources are cross-validated to ensure the accuracy of road condition data collection.

[0030] Environmental parameter acquisition: Temperature, air pressure, rainfall and humidity sensors collect environmental data simultaneously, and air pressure data is converted into real-time altitude using a standard formula.

[0031] Vehicle operating status data collection: Core operating status is collected through sensors such as engine speed, torque, and brake temperature; electric vehicles additionally obtain parameters such as battery SOC through BMS.

[0032] Driver intent capture: Steering wheel angle, accelerator and brake pedal position sensors accurately capture the driver's operating intent.

[0033] 2. Data preprocessing: The fusion processing unit performs standardization preprocessing on the collected multi-source heterogeneous data, specifically including: Noise filtering: The Kalman filter algorithm is used to filter out interference data such as ranging fluctuations and pedal travel noise.

[0034] Time synchronization: The acquisition time of different sensors is unified to the system clock, and the synchronization error is controlled within a preset range.

[0035] Missing value completion: For temporarily invalid sensor data, historical trend interpolation is used to complete the missing value.

[0036] 3. Analysis of Multiple Core Coupling Models Based on the preprocessed data, the fusion processing unit runs multiple core coupling models to quantify the interaction patterns between various factors. These multiple core coupling models dynamically link the scene and various factors, and based on the real-time scene characteristics of the mountain road (downhill, uphill, curves, etc.), they link and call related single coupling models to adapt to the complex working conditions of the mountain road.

[0037] (1) Altitude-Engine Power Coupling Model It is used to quantify the coupling relationship between altitude and actual engine output power, solve the problem of reduced engine combustion efficiency and power attenuation caused by thin air at high altitudes, and adapt to the aggravated power attenuation effect of increased load under uphill conditions. It is the core basis for matching vehicle speed and torque in uphill scenarios.

[0038] Altitude increase exhibits a nonlinear negative correlation with engine power, and the steeper the slope (uphill), the smaller the power attenuation coefficient k at the same altitude. The model fits a three-dimensional nonlinear mapping curve of altitude-slope-power attenuation coefficient using measured data. The core formula is expressed as: Where H is the real-time altitude and θ is the road slope (positive for uphill and negative for downhill).

[0039] For example, after scene recognition, the core feature parameters of the uphill scene are extracted: real-time altitude H, road slope θ, and slope length L, and these are used as the core inputs of the altitude-engine power coupling model. Based on the input parameters, the model calculates the engine power attenuation coefficient k at that altitude and slope, and predicts the trend of engine power change during continuous uphill.

[0040] If the uphill scenario is accompanied by low / high temperatures (minor influencing factors), then the temperature-tire adhesion coefficient coupling model is superimposed as an aid to fine-tune the power matching strategy. For example, if the low temperature causes the tire adhesion coefficient to decrease, the power needs to be appropriately reduced to avoid slippage.

[0041] The model outputs scenario-based coupling results: the actual output power and power decay rate of the engine are adapted to the current uphill scenario, directly supporting the forward-looking decision-making module's control objectives of "preventing overheating and ensuring sufficient power".

[0042] (2) Temperature-Tire Adhesion Coefficient Coupled Model It is used to quantify the coupling relationship between ambient temperature (including road surface temperature) and tire adhesion coefficient, reflecting the impact of temperature changes on tire grip. It also adapts to the temperature coupling effect under different road surface conditions such as rain, snow, dry and wet, and is the core basis for vehicle speed control and braking intensity matching in curve and downhill scenarios.

[0043] The model fits a three-dimensional nonlinear mapping curve of temperature, road surface condition, and adhesion coefficient using measured data. The core formula is expressed as follows: Where T is the ambient and / or road surface temperature, S is the road surface dry / wet state (quantified as a 0 / 1 / 2 state value), and H is the ambient humidity.

[0044] The output includes: accurate tire adhesion coefficient value ,0< ≤1, the larger the coefficient, the stronger the tire grip.

[0045] Temperature and tire adhesion coefficient exhibit a non-linear relationship, and the dryness or wetness of the road surface is a modulating factor in the model. Dry road surface: Temperature within a reasonable range, adhesion coefficient Maintaining a high value, whether the temperature is too high or too low, A slight nonlinear decrease; On wet / snowy surfaces: coefficient of adhesion at the same temperature Significantly reduced, and the lower the temperature (e.g., close to 0°C). The rate of descent accelerated.

[0046] For example, after scene recognition, the core feature parameters of the curve scene are extracted: the curve curvature radius R and the road surface dryness / wetness state S, and combined with environmental sensor data (temperature T, humidity H) as the core input of the model; based on the input parameters, the model calculates the accurate value of the tire adhesion coefficient in the current scene. It also predicts changes in the coefficient of adhesion during cornering, such as water accumulation in the bend. A sudden drop.

[0047] If the curve is a high-altitude curve (a secondary influencing factor), then an altitude-engine / motor power coupling model is superimposed as an aid to fine-tune the power output when cornering, avoiding insufficient power when cornering due to power attenuation at high altitude.

[0048] The model outputs scenario-based coupling results: the maximum safe cornering speed and speed fluctuation threshold that are adapted to the current adhesion coefficient, directly supporting the forward-looking decision-making module's control objective of "smooth cornering and safe controllability".

[0049] (3) Altitude-motor power coupling model (electric vehicle) As an electric vehicle-adapted version of the altitude-engine power model, it quantifies the coupling relationship between altitude and actual motor output power / regenerative braking efficiency, and adapts to the impact of high altitude on motor heat dissipation and battery performance. It is the core basis for electric vehicle uphill power matching and downhill regenerative braking intensity adjustment.

[0050] The effect of altitude on motor power attenuation is much smaller than that on engine power. The model mainly fits the coupling relationship between altitude, battery SOC, and regenerative braking efficiency: as altitude increases, battery heat dissipation efficiency changes, and at the same SOC, regenerative braking efficiency exhibits a slight nonlinear decrease. The core formula can be expressed as: Where H is the real-time altitude and SOC is the remaining battery power.

[0051] The output includes: Motor power correction factor ,0< ≤1, used to correct the effect of altitude on motor output power; The regenerative braking efficiency coefficient η, 0 < η ≤ 1, is used to correct the effect of altitude on the regenerative braking energy recovery rate of electric vehicles.

[0052] (4) Derived core coupling model: Brake temperature-braking intensity-road condition parameter coupling model This model is used to quantify the coupling relationship between braking intensity, road slope, tire adhesion coefficient and brake temperature rise rate. It is the core basis for preventing brake fade and matching braking methods (engine braking / regenerative braking / friction braking) in downhill scenarios. Brake temperature prediction in downhill scenarios is based on this model.

[0053] The greater the braking intensity and the steeper the slope, the faster the brake temperature rises; the lower the tire adhesion coefficient, the greater the braking intensity is required to maintain vehicle speed control, further exacerbating the brake temperature rise. These three factors are synergistically positively correlated, as expressed by the core formula: Where t is the driving time corresponding to the preset driving time / distance; B is the braking intensity (quantized as a value of 0-1, converted by the brake pedal position sensor / regenerative braking current); θ is the road slope; and μ is the tire adhesion coefficient (output by the temperature-tire adhesion coefficient model). This is the initial brake temperature (collected by the brake temperature sensor).

[0054] The output includes: Brake temperature rise rate (℃ / s); Predicted brake temperature within the preset driving distance .

[0055] For example, after scene recognition, the core feature parameters of the downhill scene are extracted: slope θ, slope length L, and initial braking temperature, and these are used as the core input of the model. The model first calls the output of the temperature-tire adhesion coefficient coupled model (tire adhesion coefficient μ) and adds it to its own input parameters to complete the integration of multiple factors; The model calculates the rate of brake temperature rise under different braking intensities. It also predicts the maximum braking temperature within a preset slope length and outputs the regenerative braking efficiency coefficient η for the electric vehicle. The model outputs scenario-based coupling results: the maximum braking intensity within the safe threshold of brake temperature, the optimal intensity of regenerative braking (electric vehicle), and the timing of friction braking intervention, which directly support the forward-looking decision-making module's control objectives of "preventing brake fade and high energy recovery".

[0056] 4. Output: Standardized integrated information package After preprocessing and coupling analysis, the fusion processing unit outputs a standardized fusion information package containing "road condition characteristics, environmental parameters, vehicle status, and driver intent". The data update frequency meets the real-time requirements, providing accurate input for the S2 forward-looking decision-making stage.

[0057] S2, Forward-looking decision-making stage: trend prediction and multi-objective optimization decision-making.

[0058] The core of this stage is to construct a closed-loop decision-making process based on the standardized fusion information package output by S1, which consists of "scene recognition - forward prediction - multi-objective optimization - strategy generation". By accurately identifying driving scenarios, predicting multi-dimensional trends, and dynamically adjusting and optimizing target weights, the optimal control strategy adapted to the complex scenario of mountain roads is finally generated.

[0059] 1. Composite Scene Recognition and Label Generation (1) Core parameter extraction: Quantifiable parameters after Kalman filtering preprocessing are extracted as the basic data for scene determination. The core parameters are divided into two categories of road condition geometric core parameters (main determination dimensions): road surface slope θ (°), slope length L (m), curve curvature radius R (m), curve continuous length (m), precipitation intensity (mm / h), and altitude H (m). All parameters are not qualitatively described to ensure the objectivity of the determination results.

[0060] (2) Core Judgment Rules for Primary and Secondary Scenarios: For typical composite scenarios of mountain roads (the core being a combination of curves and slopes), preset quantitative judgment thresholds for slope and curve curvature are used. When the characteristic parameters of a certain scenario exceed the preset thresholds and have a decisive impact on driving, it is judged as the primary scenario: When the road surface slope is ≥8°, the main scenario is determined to be uphill or downhill, and the secondary scenario is curves. When the radius of curvature of a curve is ≤10m, the curve is determined to be the main scene, and the uphill or downhill is the auxiliary scene.

[0061] (3) Supplementary rules for temporary switching of the main scene: If environmental or vehicle status parameters reach extreme values ​​and their impact on driving safety exceeds the road condition geometry, a temporary switching of the main scene is triggered to match actual driving risks. Example 1: Combined slope and curve scenario (slope 6°, curvature 15°, no extreme values): No parameters exceed the threshold, the main scenario is determined by the driving trend (e.g., driving towards a steep slope, the downhill / uphill is the main scenario, driving towards a sharp curve, the curve is the main scenario). Example 2: Combined curve and slope scenario (slope 7°, curvature 12°, icy road surface and tire adhesion coefficient μ≤0.2): The environmental parameters are extreme values. Anti-skid (corresponding to the core pain point of curve) is temporarily determined as the main scenario, and the slope is the auxiliary scenario. Example 3: Long downhill slope and curve scenario (slope 9°, curvature 8°, brake temperature) ≥180℃): The vehicle status parameters are extreme values. The primary scenario is to temporarily determine the prevention of brake fade (corresponding to the core pain point of downhill driving), while curves are the secondary scenario.

[0062] (4) Judgment result output: After the scene judgment is completed, two types of core information are output: Standardized labels for complex scenarios, such as "mainly downhill - secondary curves", "mainly curves - secondary uphill", "mainly temporary anti-skid measures - secondary slope", etc. The weight values ​​for main / auxiliary scene features are: the weight of the main scene is 0.7~0.9, and the weight of the auxiliary scene is 0.1~0.3, which provides a quantitative basis for the subsequent LSTM model to prioritize the prediction of the main scene and adjust the weight of the multi-objective optimization function.

[0063] 2. Three-dimensional trend forecasting (based on LSTM model) Using an LSTM model trained on a large amount of mountain road data, a prediction logic of "primary scene priority and secondary scene constraint" is designed for complex scenarios to complete the three-dimensional trend prediction of road conditions, environment and vehicle status. The prediction range is adaptively adjusted with vehicle speed, and the prediction distance of the primary scene is greater than that of the secondary scene, ensuring the pertinence and sufficiency of forward-looking prediction.

[0064] (1) Input layer: Weighted enhancement of main scene feature parameters, accounting for the core weight of the input dimension. The input layer of the LSTM is a set of composite scene feature parameters (including road conditions, environment, and vehicle status) from a standardized fusion information packet. The weights of the main scene features are increased through linear weighting. Main scene feature parameters (such as slope θ, slope length L, and brake temperature when the main scene is a slope) The weight coefficients are set to 0.7~0.9, serving as the core input dimension of the LSTM model; The weight coefficients of auxiliary scene feature parameters (such as the radius of curvature R when the curve is an auxiliary scene) are set to 0.1~0.3, which serve as auxiliary input dimensions to ensure that the model prioritizes capturing the changing trends of the main scene. (2) Hidden layer: Configure feature extraction priority, with main scene features being extracted first in time sequence. Configure the feature extraction priority of the gating units (input gate, forget gate, output gate) of the LSTM hidden layer, so that the model prioritizes the temporal change analysis of the main scene features. The specific configuration logic is as follows: Input gate: Prioritize receiving time-series data of the main scene features, and only increase the weight of receiving auxiliary scene features when the main scene features do not change significantly. Forget gate: Prioritize the retention of historical time-series data of the main scene features, and weaken the invalid historical data of the auxiliary scene features to ensure that the model's memory of the main scene trend is more durable; Output gate: Prioritizes outputting the temporal extraction results of the main scene features, and the extraction results of the auxiliary scene features are superimposed on the main scene results as supplementary values.

[0065] For example, in a composite scenario of "downhill as the main feature and curves as the secondary feature", the temporal trends of slope, slope length, and brake temperature are extracted first, while the temporal changes of curve curvature are only superimposed as supplementary values ​​to ensure that the core prediction focuses on the main scenario trends such as the rise in brake temperature and regenerative braking efficiency on the downhill slope.

[0066] (3) Output layer: The prediction results are output in layers, with the main scene trend as the core output and the auxiliary scene as the constraint output. The LSTM output layer adopts a hierarchical output format of "core result - constraint result". The prediction results of the auxiliary scene are used as constraints on the prediction results of the main scene to ensure that the prediction results fit the actual working conditions of the composite scene. Core output: Key trend prediction values ​​for the main scene (e.g., brake temperature rise rate when the slope is mainly downhill). Regenerative braking efficiency η, gradient variation trend; tire adhesion coefficient μ variation trend when cornering is the main feature, and maximum safe cornering speed); Constraint output: Key trend prediction values ​​of auxiliary scenarios are superimposed on the core output as threshold constraints (e.g., when curves are auxiliary, the maximum safe speed corresponding to the radius of curvature is used as a threshold to constrain the target speed of the vehicle going downhill, so as to avoid the vehicle speed going downhill too fast and causing sideslip when cornering). Output format: Standardized "Main Scene Trend Prediction Table and Auxiliary Scene Constraint Threshold Table", which directly provides input to the multi-objective optimization function to ensure that subsequent decisions focus on the main scene.

[0067] (4) Content of three-dimensional trend prediction: Road condition trends: Predict the rate of change of slope, peak curvature of curves, and dynamic value of road surface adhesion coefficient within a preset distance ahead; Environmental trends: Predict the temperature gradient, precipitation probability, and special weather conditions in high-altitude sections of the road ahead; Vehicle status trend: Combining road conditions and environmental predictions, the engine / motor temperature evolution curve, brake temperature rise rate, and electric vehicle battery SOC recycling potential are jointly predicted by the LSTM model and the vehicle dynamics model. This allows for the early identification of risks such as temperature exceeding the threshold, and the prediction error is controlled within a preset range.

[0068] 3. Multi-objective optimization decision-making (based on MPC algorithm) The decision control unit takes "optimal safety, optimal energy consumption, and optimal comfort" as its core, constructs a multi-objective optimization function, solves it through the MPC algorithm to generate a control strategy, and outputs a frequency synchronized with a standardized fusion information packet.

[0069] (1) Definition of multi-objective optimization function: Using a weighted summation form, the expression is: in: For the safety objective function, a normalized combination of "minimizing brake temperature deviation from the threshold" and "minimizing tire slip ratio" is adopted. ,in, This represents the actual brake temperature. Here, s represents the safe threshold for brake temperature, and s represents the actual slip ratio. The slip ratio safety threshold, , These are the sub-objective weight coefficients. . The smaller the value, the higher the vehicle's driving safety.

[0070] For fuel-powered vehicles, the objective function is "minimizing fuel consumption rate". f is the actual fuel consumption rate. This represents the optimal fuel consumption rate under the current operating conditions. The smaller the value, the closer the fuel consumption rate is to the optimal value, and the better the energy performance. Electric vehicles employ a combination of "minimizing energy consumption rate and maximizing regenerative braking recovery rate." Where e is the actual power consumption rate, The optimal power consumption rate under the current operating conditions. For regenerative braking recovery rate, , These are the weighting coefficients; The smaller the value, the higher the energy utilization efficiency of the electric vehicle.

[0071] The objective function for comfort is constructed based on the principle of "minimizing vehicle speed fluctuations". ,in , These are the maximum and minimum vehicle speeds within a preset control period, respectively. The target vehicle speed is given. This function ensures smooth vehicle movement by minimizing the relative fluctuation amplitude.

[0072] (2) Dynamic adjustment of scenario-based weights: , , The weighting coefficients are based on specific scenarios; a balanced value is used during normal driving, and the value is increased during extreme weather conditions. (Safety weights) The weights are adjusted accordingly for scenarios such as long downhill slopes and continuous uphill slopes. The weights of the target items corresponding to the main scenario are given priority to be increased, while the weights of the target items in the auxiliary scenario are appropriately compressed to ensure that the optimization direction is in line with the core needs of the scenario.

[0073] (3) Control strategy generation: The MPC algorithm is used to solve the multi-objective optimization function and generate specific control commands for different scenarios, including parameters such as engine torque, braking intensity, gearbox gear (fuel vehicle) or regenerative braking intensity (electric vehicle). The output frequency is synchronized with the standardized fusion information package.

[0074] 4. Implementation process of dynamic weight adjustment for multi-objective optimization functions (1) Establish a preset weight matrix of "main scene label - target item weight". Based on engineering experience with mountain roads, a pre-defined matrix (see Table 1) was established for the main scenario and the objective items (safety J1, energy consumption J2, comfort J3) of the objective function. The matrix clearly defines the basic weight values ​​of each objective item under different main scenarios. The matrix design principle is that the basic weight of the core objective item corresponding to the main scenario is significantly higher than that of other items. For example, in the downhill scenario, which focuses on "preventing brake fade", the safety objective has the highest basic weight, providing a benchmark for subsequent dynamic adjustment of weights.

[0075] Table 1. Preset Matrix of Weights for Main Scene and Multi-Objective Tasks on the Mountain Road Note: This matrix is ​​stored in the decision control unit of the forward-looking decision module and can be iteratively optimized according to actual working conditions.

[0076] (2) Triggering conditions for weight increase The weights of the objective function are not fixed values, but are dynamically adjusted based on the quantized output value of the core coupled model. When the output value of the coupled model corresponding to the main scene deviates from the normal range, the weight of the core objective item of the main scene is increased, achieving dynamic adaptation of "the higher the risk, the higher the safety weight". The triggering rules are as follows: Downhill as the primary scenario: If the brake temperature-braking intensity-road condition parameter coupled model outputs the brake temperature rise rate... ≥1.5℃ / s, or brake temperature If the temperature is ≥180℃, the weight ω1 of safety target J1 increases from 0.45 to 0.5~0.6; For scenarios with curves as the main feature: if the tire adhesion coefficient μ output by the temperature-tire adhesion coefficient coupling model is ≤0.3 (such as in cases of icing or water accumulation), then the weight ω1 of the safety target J1 is increased from 0.5 to 0.6~0.7. For uphill scenarios: if the power attenuation coefficient k output by the altitude-engine power coupling model is less than or equal to 0.5 (high altitude steep slope), then the weight ω1 of the safety target J1 is increased from 0.4 to 0.5~0.55. The magnitude of the weight increase is positively correlated with the degree of deviation of the coupled model's output value; the greater the deviation, the greater the weight increase.

[0077] (3) Weight redistribution in composite scenarios For complex scenarios such as winding mountain roads with curves and slopes, the system achieves weight redistribution by "matching the main scenario label with a preset weight matrix and compressing the weights of the corresponding target items in the auxiliary scenario." This ensures that the optimization direction focuses on the core needs of the main scenario. The specific matching and adjustment logic is as follows: ① Tag matching: The forward-looking decision-making module matches the main scene tags of the composite scene (such as downhill as the main and curves as the secondary) with the preset weight matrix to obtain the basic weight values ​​corresponding to the main scene (e.g., downhill is the main scene with basic weights ω1=0.45, ω2=0.4, ω3=0.15). ②Weight compression and transfer: Identify the target items corresponding to the auxiliary scene, compress their weights, and transfer the compressed part to the core target items of the main scene to achieve weight redistribution (e.g., if the curve is an auxiliary scene, the weight of its corresponding comfort target J3 is compressed from 0.15 to 0.1, and the extra 0.05 weight is transferred to the safety target J1, at which point ω1=0.5, ω2=0.4, and ω3=0.1). ③ Risk-triggered enhancement: If the output value of the core coupled model corresponding to the main scenario triggers the weight enhancement condition, the weight of the core target item will be further increased based on the redistributed weights (e.g., in the downhill scenario). ≥2℃ / s, ω1 increases from 0.5 to 0.6.

[0078] (4) Closed-loop iterative optimization Weight adjustment is dynamically iterated through feedback data to ensure it closely reflects actual driving performance. The specific process is as follows: ① The execution feedback module collects actual execution data in real time (such as actual brake temperature, actual sideslip rate, and actual vehicle speed fluctuation). ② Compare the measured data with the target value of the multi-objective optimization function. If the deviation of the core objective of the main scenario exceeds the threshold (such as the actual value of the downhill braking temperature > the predicted value), then increase the weight of the core objective item of the main scenario again; if the deviation of the objective of the auxiliary scenario exceeds the threshold (such as the excessive fluctuation of vehicle speed on the curve), then appropriately increase the weight of the corresponding objective item of the auxiliary scenario (but not exceeding the weight of the main scenario). ③ The adjusted weight values ​​are sent back to the preset weight matrix to achieve iterative optimization of the matrix and provide a more accurate basis for weight allocation in similar scenarios in the future.

[0079] The following example, using an electric vehicle in a complex scenario of "primarily downhill and secondarily curves," illustrates the process of weight adjustment.

[0080] ① Basic weight matching: Downhill is the main scenario. The basic weights are obtained by matching the preset matrix: ω1=0.45 (safety, anti-brake heat fade), ω2=0.4 (energy consumption, high energy recovery), and ω3=0.15 (comfort, stable speed). ② Auxiliary scene weight compression: The curve is the auxiliary scene. The weight of comfort target J3 is compressed to 0.1, and 0.05 is transferred to safety target J1. After adjustment, the weights are ω1=0.5, ω2=0.4, and ω3=0.1. ③ Coupled model triggering: If the brake temperature-brake intensity model output =2℃ / s (deviates from the normal range), triggering the weight increase of safety target J1, ω1 increases from 0.5 to 0.6, and the final weights are ω1=0.6, ω2=0.35, ω3=0.05; ④ Function solution: The optimization function prioritizes solving the safety objective of minimizing brake temperature, while also taking into account the energy consumption objective of regenerative braking recovery. The objective of stable vehicle speed corresponding to curves is only subject to minimum constraints.

[0081] S3, Execution Feedback Phase: Instruction Execution and Closed-Loop Optimization.

[0082] The core of this stage is to transform the control strategies output by the forward-looking decision-making module into actual actions, and to achieve dynamic closed-loop correction through real-time status feedback to ensure control accuracy, safety and adaptability.

[0083] 1. Instruction parsing The instruction parsing unit of the execution feedback module receives standardized control instructions output by the forward decision-making module and converts high-level strategy instructions into low-level control signals that can be directly recognized by each execution component through a built-in parsing algorithm.

[0084] For example, the regenerative braking intensity command is converted into a current signal that the motor controller can recognize, and the target vehicle speed command is converted into a gear signal that the transmission controller can recognize; the parsing process strictly controls the delay within a preset range to ensure the real-time execution of the command and avoid the control effect being affected by parsing lag.

[0085] 2. Execution Control The execution control unit drives the corresponding execution components to perform precise actions based on the parsed underlying control signals. Different execution component configurations and control logic are adopted to address the differences in powertrain structures between gasoline and electric vehicles. (1) Execution logic of fuel-powered vehicles: The engine control unit (ECU) achieves precise control of engine power by adjusting the throttle opening and fuel injection quantity; The transmission control unit (TCU) automatically completes the transmission shifting operation based on the target vehicle speed and road conditions. The brake control unit (BCU) controls the intensity of friction braking by adjusting the pressure of the master cylinder.

[0086] (2) Electric vehicle execution logic: The motor control unit (MCU) achieves coordinated control of regenerative braking intensity and motor torque by adjusting the motor reverse drag current; The battery management system (BMS) precisely controls the battery charging current based on regenerative braking efficiency and battery status; The brake control unit (BCU) intervenes only when the regenerative braking force is insufficient and cannot meet the vehicle speed control requirements, thus avoiding over-reliance on friction braking.

[0087] Fault tolerance mechanism: The execution control unit has a complete fault tolerance capability. If a certain execution component fails temporarily, it can quickly switch to the backup control mode within a preset time to ensure that the vehicle's driving safety is not affected.

[0088] 3. Status Feedback Collection Dedicated feedback sensors are used to collect core vehicle status data in real time after execution, ensuring the comprehensiveness and accuracy of data collection. The collected data includes: Driving status data: actual vehicle speed, driving displacement; Powertrain data: Actual output power and torque of the engine / motor; Braking system data: actual braking intensity, brake temperature; Electric vehicle-specific data: Remaining battery charge (SOC) and battery charging current.

[0089] The collected data is transmitted back to the perception fusion module in real time via a high-speed communication bus, providing reliable measured data support for closed-loop correction.

[0090] 4. Closed-loop correction and optimization After receiving feedback data, the perception fusion module initiates a closed-loop correction process to dynamically optimize the control strategy. Data comparison: The actual measured data is compared one by one with the predicted values ​​in the standardized fusion information package, with a focus on verifying core control indicators such as vehicle speed fluctuation, brake temperature, and energy consumption level. Deviation judgment: If the measured value of a certain indicator deviates from the predicted value by more than a preset threshold (such as vehicle speed fluctuation exceeding the set range, abnormal increase in brake temperature, or energy consumption deviating from the optimal range), a re-decision mechanism is triggered. Strategy Adjustment: The perception fusion module updates the standardized fusion information package, and the forward-looking decision-making module re-optimizes the prediction results and control strategies based on the corrected fusion data, forming a dynamic closed loop of "collection-fusion-prediction-decision-execution-feedback" to ensure that the control strategy always fits the actual driving conditions and continuously optimizes driving safety, energy economy and ride comfort.

[0091] Example 2 This embodiment provides an intelligent vehicle speed control system suitable for mountain roads. The system adopts a three-level core architecture of "perception fusion - forward decision-making - execution feedback". Each module is connected to the signal in sequence to form a closed-loop control. The system achieves simplified and efficient control logic through functional integration.

[0092] I. Perception Fusion Module 1. Information Collection Unit A redundant design of "main sensor and auxiliary sensor" is adopted to achieve full-dimensional and high-precision data collection for key influencing factors of mountain roads. The specific sub-units are as follows: (1) Road Condition Perception Subunit: Equipped with a high-precision map, lidar, millimeter-wave radar, and high-definition camera. The high-precision map pre-stores the three-dimensional geographic information of the target mountain road, including segmental elevation data, slope change curves, and curve parameters; lidar and millimeter-wave radar jointly detect the distance and relative speed of obstacles ahead; the high-definition camera acquires road markings and road surface dryness / wetness through image recognition technology. The data from the three devices are cross-validated to ensure the accuracy of road condition feature acquisition.

[0093] (2) Environmental sensing subunit: Equipped with temperature sensor, air pressure sensor, rainfall sensor and humidity sensor. The air pressure sensor collects atmospheric pressure in real time and calculates the current altitude using the standard atmospheric pressure-altitude conversion formula; the temperature, humidity and rainfall sensors collect data synchronously to form a combination of environmental parameters.

[0094] (3) Vehicle Status Perception Subunit: Dedicated sensors are configured to address core pain points. The engine speed sensor collects the real-time engine speed and calculates the engine output power by combining it with the torque sensor data; the brake temperature sensor is installed on the inside of the brake disc to monitor the temperature of the braking system in real time; electric vehicles are additionally equipped with a battery management system (BMS) to collect battery status information, and the motor temperature sensor monitors the motor operating temperature.

[0095] (4) Driver Intention Perception Subunit: Steering wheel angle sensor collects steering angle to determine driver steering intention; accelerator pedal and brake pedal position sensors collect pedal travel respectively to quantify driver acceleration or deceleration needs.

[0096] 2. Fusion Processing Unit Equipped with a high-performance motor control unit (MCU), integrating data preprocessing and multi-factor fusion algorithms, the processing flow consists of three steps: (1) Data preprocessing: noise reduction, time synchronization and missing value filling of each sensor data are performed by Kalman filtering algorithm; (2) Construction of multiple core coupling models: Multiple core coupling models were trained based on a large amount of measured data of mountain roads, including correlation models such as altitude-engine power and temperature-tire adhesion coefficient, to quantify the interaction law between various factors; (3) Data fusion output: The DS evidence theory (Dempster-Shafer Evidence Theory) is used to fuse the preprocessed data with the results of the coupled model analysis, and output a standardized fusion information package containing "road condition features, environmental parameters, vehicle status and driver intentions" to ensure the real-time update of data.

[0097] The principle of associating composite scene models is as follows: First, identify the main scene (e.g., in a combination of curves and slopes, if the slope is ≥8°, then the uphill / downhill is the main scene; if the curvature of the curve is ≤10m, then the curve is the main scene), then match the core model, and match the auxiliary scene with the auxiliary model.

[0098] II. Forward-looking decision-making module Based on the standardized fusion information package output by the perception fusion module, the system can generate an optimal control strategy that balances safety, comfort, and energy economy by identifying potential collision risks, lane departure risks, traffic efficiency bottlenecks, and other scenario-based problems in advance.

[0099] 1. Forward-looking forecasting unit Equipped with an AI chip and integrating a Long Short-Term Memory (LSTM) network model trained on a large amount of mountain road data, it achieves three-dimensional trend prediction: (1) Road condition trend prediction: Based on the basic road segment data of the high-precision map, combined with the real-time road condition data of the perception fusion module, the LSTM model is used to predict the road condition change trend within the distance of the target ahead, including the gradient increase and decrease law, the evolution of curve parameters and the dynamic change of road surface adhesion coefficient.

[0100] (2) Environmental trend prediction: Combining real-time environmental data with local mountain road historical meteorological data, the LSTM model is used to predict environmental changes in the area ahead, including temperature gradient, precipitation probability and special weather conditions that may occur in high-altitude sections.

[0101] (3) Vehicle status trend prediction: Based on the current vehicle status data in the fusion information package, the predicted road conditions and environmental trends are used to jointly predict the evolution of vehicle status through the LSTM model and the vehicle dynamics model, including engine / motor temperature changes, brake temperature rise rate and battery SOC (electric vehicle) recycling potential, etc., and identify risks such as temperature exceeding the threshold in advance.

[0102] 2. Decision control unit Based on the Model Predictive Control (MPC) algorithm, a multi-objective optimization function is constructed. The weights can be dynamically adjusted according to the scenario (e.g., balancing safety, energy consumption, and comfort during normal driving, and increasing safety weights during extreme weather). The objective function constraints include key indicator thresholds such as brake temperature, tire slip ratio, vehicle speed fluctuation, and energy consumption. The decision-making process consists of two steps: scenario recognition and policy generation. (1) Scene recognition: By integrating road condition features (slope, curve) in the information package, the system automatically identifies three core scenes: uphill, downhill, and curve, as well as various composite scenes. The recognition is based on preset thresholds. For example, continuous uphill, continuous downhill, and sharp curve scenes are defined by thresholds for slope, slope length, or curve parameters.

[0103] (2) Strategy Generation: Generate specific control strategies for different scenarios. The strategy output is the specific control parameters: Uphill scenario strategy: The core objectives are "overheat prevention, sufficient power, and low energy consumption". Based on the predicted power decay and temperature changes, torque and vehicle speed are dynamically matched, and vehicle speed and transmission gear are adjusted in advance if necessary.

[0104] Downhill scenario strategy: The core objective is to "prevent brake fade and achieve high energy recovery". For fuel vehicles, engine braking is prioritized, and only slight friction braking is used when necessary. For electric vehicles, the intensity of regenerative braking is adjusted based on the battery SOC to balance energy recovery and braking safety.

[0105] Cornering scenario strategy: The core objective is "smooth cornering, safe and controllable". Reduce speed in advance, make minor adjustments to speed inside the corner, and gradually increase speed after exiting the corner to ensure driving stability.

[0106] III. Execution Feedback Module It is responsible for translating control strategies into actions and achieving closed-loop optimization through real-time feedback.

[0107] 1. Instruction parsing unit The unit receives standardized control commands output by the forward-looking decision-making module via a high-speed communication bus. The unit has a built-in command parsing algorithm that converts high-level strategy commands into low-level control signals that can be recognized by each execution component.

[0108] 2. Execution control unit Based on the parsed underlying control signals, the actions of each actuator are driven, and the actuator configuration is adapted to different vehicle models: (1) Components of fuel vehicle: Engine control unit (ECU) achieves power control by adjusting throttle opening and fuel injection quantity; Transmission control unit (TCU) achieves automatic gear shifting; Brake control unit (BCU) achieves friction braking by adjusting brake master cylinder pressure.

[0109] (2) Electric vehicle actuators: The motor control unit (MCU) realizes regenerative braking and torque control by adjusting the motor reverse drag current; the battery management system (BMS) controls the battery charging current; the brake control unit (BCU) is the same as that of the fuel vehicle, and only intervenes when the regenerative braking is insufficient.

[0110] The control unit is fault-tolerant; if a component fails temporarily, it can quickly switch to a backup control mode to ensure driving safety.

[0111] 3. Feedback Acquisition Unit It integrates dedicated sensors for feedback such as rotational speed, vehicle speed, and temperature, and collects core vehicle status data in real time after execution, including actual vehicle speed, engine / motor power, braking intensity, brake temperature, and battery SOC (for electric vehicles), providing data support for closed-loop correction.

[0112] Example 3 This embodiment takes the scenario of an electric vehicle going down a continuous downhill slope on a mountain road as an example to fully demonstrate the closed-loop control process of "perception fusion - forward decision-making - execution feedback".

[0113] S1. Perceptual Fusion Stage: Multi-Source Data Acquisition and Coupled Analysis After the electric vehicle enters the downhill section of the mountain road, the perception fusion module starts working. The information acquisition unit loads a high-precision map and obtains basic parameters of the continuous downhill slope (slope range, altitude change, and curve distribution) within a preset distance ahead; the lidar and camera jointly detect that the road surface is free of obstacles and dry, and initially identify the adhesion coefficient; the environmental sensors collect data on air pressure (converted to current altitude), temperature, humidity, and absence of precipitation; the vehicle status sensors collect data on current vehicle speed, brake temperature, battery SOC, motor speed, and regenerative braking intensity; the intent sensor detects that the driver has not operated the accelerator / brake, and the steering wheel returns to center. The fusion processing unit corrects the sensor data such as motor speed through Kalman filtering, corrects the adhesion coefficient to an accurate value through a temperature-tire adhesion coefficient correlation model, analyzes the effect of altitude decrease on motor performance improvement through an altitude-motor power model, and finally outputs a standardized fusion information package, with an update frequency maintained at 10Hz.

[0114] S2. Forward-looking decision-making stage: Trend prediction and multi-objective optimization decision-making After receiving the fused information packet, the forward-looking decision-making module predicts using an LSTM model: the average gradient, slope length, and road surface adhesion coefficient of the upcoming downhill section remain stable; environmental parameters show no significant changes; regarding vehicle state trends, if the current regenerative braking intensity is maintained, the vehicle speed will exceed the safe range after a preset distance, the brake temperature will rise slightly, and the battery SOC will increase significantly; if only friction braking is used for deceleration, the brake temperature will approach the safe threshold. The decision control unit identifies this as a "continuous downhill scenario," matches scenario weights ω1=0.45, ω2=0.4, and ω3=0.15, and defines the objective function: Constraints: Brake temperature ≤190℃, slip ratio ≤15%, vehicle speed fluctuation ≤5km / h, regenerative braking intensity ≤ motor safety threshold. After solving the objective function, a strategy is generated: Regenerative braking intensity and target vehicle speed are adjusted in stages; before entering a curve, the speed is reduced to an appropriate level; within the curve, the intensity and speed are fine-tuned; after exiting the curve, the speed gradually increases. A preset battery SOC threshold is used; when the threshold is reached, the regenerative braking intensity is reduced and supplemented with slight friction braking. The strategy instructions are output to the execution module.

[0115] S3. Execution Feedback Phase: Instruction Execution and Closed-Loop Optimization The instruction parsing unit of the execution feedback module translates the strategy into low-level control signals: the MCU regenerative braking current is adjusted to the corresponding intensity in stages, the TCU switches to the appropriate gear, and the vehicle speed target is set in stages. The execution control unit drives the components to perform the following actions: the MCU adjusts the current to the target value within a preset time, the TCU completes the gear switching, and the vehicle speed smoothly decreases to the target value with a fluctuation range of ≤0.5km / h; within curves, the regenerative braking intensity and vehicle speed are finely adjusted according to instructions to ensure smooth cornering.

[0116] The feedback acquisition unit collects real-time data such as actual vehicle speed, regenerative braking intensity, brake temperature, and battery SOC, and transmits this data back to the perception fusion module via the CAN bus. Closed-loop verification shows that the data deviation is ≤1%, indicating no abnormalities. After driving a preset distance, if the battery SOC approaches the threshold, the decision unit adjusts the regenerative braking intensity to the appropriate value in advance, achieving dynamic optimization.

[0117] Finally, it should be noted that any parts of this invention not described in detail are prior art. Those skilled in the art will understand that the above descriptions are merely preferred embodiments of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A smart vehicle speed control method suitable for mountain roads, characterized in that, Includes the following steps: S1. Perception Fusion Stage: Collect multi-source data on vehicle driving environment, vehicle status and driver intention, preprocess the multi-source data, and run multiple core coupling models to quantify the coupling effect between different factors. Based on the characteristics of the mountain road scene, link and call the multiple core coupling models to output a standardized fusion information package containing road condition features, environmental parameters, vehicle status and driver intention. S2, Forward Decision-Making Stage: Based on the standardized fusion information package, the prediction model predicts the trends of road conditions, environment, and vehicle status ahead, identifies the current driving scenario, and generates composite scenario labels for the main scenario and auxiliary scenario. A multi-objective optimization function containing safety, energy consumption, and comfort objectives is constructed using a model predictive control algorithm. The weights of each objective item in the multi-objective optimization function are dynamically adjusted according to the composite scenario labels of the main scenario and auxiliary scenario to solve and generate the optimal control strategy. S3. Execution Feedback Stage: The optimal control strategy is parsed into underlying control signals and the vehicle's execution components are driven to perform actions. Real-time vehicle status feedback data after execution is collected. The feedback data is compared with the predicted value. If the deviation exceeds a preset threshold, a new decision is triggered to form a closed-loop optimization.

2. The intelligent vehicle speed control method suitable for mountain roads according to claim 1, characterized in that, The running of multiple core coupling models in step S1 includes: Based on real-time altitude H and road slope θ, the engine power attenuation coefficient is calculated using an altitude-engine power coupling model. ; The tire adhesion coefficient is calculated using a temperature-tire adhesion coefficient coupling model based on temperature parameter T, road surface wetness / dryness S, and ambient humidity H. The temperature parameter T includes either the ambient temperature or the road surface temperature. Based on braking intensity B, road slope θ, and tire adhesion coefficient The rate of increase in brake temperature was calculated using a coupled model of brake temperature, brake intensity, and road condition parameters. and the predicted brake temperature within the preset distance ,in t represents the initial brake temperature, and t represents the driving time.

3. The intelligent vehicle speed control method suitable for mountain roads according to claim 1, characterized in that, The step S2, which involves identifying the current driving scenario and generating a composite scenario label, includes: Quantifiable core feature parameters are extracted, including road slope, slope length, curve radius of curvature, continuous curve length, precipitation intensity, and altitude. Preset judgment thresholds for each feature parameter are used to determine the primary and secondary scenes based on the principle that the more the feature parameter value deviates from the normal range, the higher the scene priority. When the road slope exceeds the preset slope threshold, uphill or downhill is determined as the primary scene and curve as the secondary scene. When the curve radius of curvature is less than the preset curvature threshold, curve is determined as the primary scene and uphill or downhill as the secondary scene. When environmental feature parameters or vehicle state feature parameters reach extreme values ​​and their impact on driving safety exceeds the road condition geometry, a temporary switch to the primary scene is triggered. The output includes a composite scene standardized label containing primary and secondary scene labels, as well as primary scene feature weight values ​​and secondary scene feature weight values.

4. The intelligent vehicle speed control method suitable for mountain roads according to claim 1, characterized in that, The prediction by the prediction model in step S2 includes: using an LSTM model for prediction, wherein the input layer of the LSTM model assigns a first weight coefficient to the main scene feature parameters and a second weight coefficient to the auxiliary scene feature parameters, and the first weight coefficient is greater than the second weight coefficient; the hidden layer configures the feature extraction priority so that the model prioritizes the temporal change analysis of the main scene features; the output layer adopts a hierarchical output form of core results plus constraint results, wherein the core output is the key trend prediction value of the main scene, and the constraint output is the key trend prediction value of the auxiliary scene.

5. The intelligent vehicle speed control method suitable for mountain roads according to claim 1, characterized in that, Step S2, which dynamically adjusts the weights of the multi-objective optimization function based on the composite scene label, includes: establishing a preset weight matrix of the main scene label and the weights of each objective item; when the output value of the core coupled model corresponding to the main scene deviates from the normal range, triggering an increase in the weight of the core objective item of the main scene, and positively correlated the degree of deviation with the magnitude of the weight increase; in the composite scene, obtaining the basic weight value by matching the preset weight matrix with the main scene label, compressing the weight of the objective item corresponding to the auxiliary scene, and transferring the compressed weight value to the core objective item of the main scene to achieve weight redistribution; based on the measured data of the execution feedback module, when the deviation between the actual value and the predicted value of the core objective of the main scene exceeds the threshold, the weight of the core objective item of the main scene is increased again, and the adjusted weight value is sent back to the preset weight matrix to achieve iterative optimization.

6. The intelligent vehicle speed control method suitable for mountain roads according to claim 1, characterized in that, The multi-objective optimization function is expressed as: ,in For the security objective function, Let the energy consumption objective function be... For the comfort objective function, , , For contextualized weighting coefficients; The security objective function: ,in This represents the actual brake temperature. Here, s represents the safe threshold for brake temperature, and s represents the actual slip ratio. The slip ratio safety threshold, , These are the sub-objective weight coefficients; The energy consumption objective function is adopted for fuel vehicles. Where f is the actual fuel consumption rate, This represents the optimal fuel consumption rate under current operating conditions; for electric vehicles, this is the optimal fuel consumption rate. Where e is the actual power consumption rate, The optimal power consumption rate under the current operating conditions. For regenerative braking recovery rate, , These are the weighting coefficients; The comfort objective function ,in , These are the maximum and minimum vehicle speeds within a preset control period, respectively. The target speed.

7. The intelligent vehicle speed control method suitable for mountain roads according to claim 1, characterized in that, The step S3, comparing the feedback data with the predicted value, includes: the perception fusion module comparing the actual vehicle speed, engine or motor power, braking intensity, brake temperature and battery SOC data collected in real time by the feedback acquisition unit with the predicted value in the standardized fusion information package. If the deviation of the vehicle speed fluctuation, brake temperature and energy consumption index exceeds the preset threshold, the fusion information package is updated, the prediction result is corrected and the control strategy is adjusted. The actions of the drive vehicle's actuators include: for gasoline vehicles, power control is achieved by adjusting the throttle opening and fuel injection quantity through the engine control unit, automatic gear shifting is achieved through the transmission control unit, and friction braking is achieved by adjusting the brake master cylinder pressure through the brake control unit; for electric vehicles, regenerative braking and torque control are achieved by adjusting the motor reverse drag current through the motor control unit, battery charging current is controlled through the battery management system, and friction braking is initiated through the brake control unit when regenerative braking is insufficient; when an actuator temporarily fails, the system switches to a backup control mode within a preset time.

8. The intelligent vehicle speed control method suitable for mountain roads according to claim 2, characterized in that, The multiple core coupling models also include an altitude-motor power coupling model, expressed as follows: Where H is the real-time altitude and SOC is the remaining battery power. This is the regenerative braking efficiency coefficient.

9. An intelligent vehicle speed control system suitable for mountain roads, used to implement the method described in any one of claims 1 to 8, characterized in that, The system adopts a three-tiered core architecture of perception fusion, forward-looking decision-making, and execution feedback, including: The perception fusion module is used to collect multi-source data, run multiple core coupled models, and output standardized fusion information packages. The forward-looking decision-making module is signal-connected to the perception fusion module and is used to predict trends, identify composite scene labels, construct and solve multi-objective optimization functions based on the standardized fusion information package, and generate the optimal control strategy. The execution feedback module is signal-connected to the forward decision-making module. It is used to parse the optimal control strategy into underlying control signals, drive the action of the execution components, and collect feedback data in real time to send back to the perception fusion module to form closed-loop control.

10. The intelligent vehicle speed control system suitable for mountain roads according to claim 9, characterized in that, The perception fusion module includes: The information acquisition unit adopts a redundant design of main sensor plus auxiliary sensor, and includes road condition perception subunit, environment perception subunit, vehicle status perception subunit and driver intention perception subunit. The fusion processing unit is used to preprocess sensor data using the Kalman filter algorithm, run multiple core coupling models to quantify the coupling effects between factors, and use DS evidence theory to fuse the preprocessed data with the analysis results of the coupling models to output a standardized fusion information package. The forward-looking decision-making module includes: The forward prediction unit is equipped with an AI chip and integrates an LSTM model trained with data from mountain roads to predict trends in road conditions, environmental changes, and vehicle status evolution. The decision control unit constructs a multi-objective optimization function with model predictive control algorithm as the core, dynamically adjusts the weight of objective items according to composite scene labels, identifies uphill, downhill and curve scenes and generates corresponding control strategies; The execution feedback module includes: The instruction parsing unit is used to parse the control instructions output by the forward decision-making module into low-level control signals that can be recognized by each execution component; The execution control unit is used to drive the action of the execution components according to the underlying control signals. It includes the engine control unit, transmission control unit, and brake control unit adapted to fuel vehicles, as well as the motor control unit, battery management system, and brake control unit adapted to electric vehicles. The feedback acquisition unit integrates a dedicated feedback sensor to collect and transmit core vehicle status data in real time after execution.

Images