Intelligent heavy truck steer-by-wire chassis parameter manual adjustment and self-learning system

Abstract

The application discloses an intelligent heavy truck chassis parameter manual adjustment and self-learning system and belongs to the field of intelligent heavy truck control. The system comprises the following steps: after obtaining manual adjustment instructions and working condition formulas, the system completes parameter configuration, chassis control and driving execution, and integrates multiple sensors to realize synchronous acquisition of driving data and self-optimization of parameters; by generating a grid data acquisition path matched with driving conditions, the chassis is controlled to drive along the preset path and acquire multi-dimensional driving data, the data is mapped to a unified working condition plane coordinate system in combination with chassis posture parameters and sensor calibration information, and model training and parameter fine-tuning are carried out based on working condition characteristic constraints, so that the optimal parameter combination is obtained and effectiveness is determined; and the integration of the parameter manual adjustment and self-optimization process is realized.

Description

Technical Field

[0001] This invention belongs to the field of intelligent heavy-duty truck control, specifically an intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system. Background Technology

[0002] Existing intelligent heavy-duty truck drive-by-wire chassis parameter adjustments typically rely on manual input of fixed parameters based on experience, resulting in low parameter matching accuracy and poor adaptability. Independently set-up parameter verification stations require additional testing grounds and equipment, leading to a complex and inefficient process. Traditional parameter calibration methods require downtime for adjustments, impacting operational efficiency and failing to adapt to complex and changing driving conditions. Fixed parameters are prone to issues such as insufficient driving stability and high energy consumption under different road conditions, resulting in a mismatch between chassis performance and operating requirements, affecting driving safety and operational economy. To address these problems, this application designs a manual adjustment and self-learning system for intelligent heavy-duty truck drive-by-wire chassis parameters. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes an intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system. After receiving manual adjustment instructions and operating condition formulas, the system completes parameter configuration, chassis control, and driving execution, and integrates multiple sensors to achieve synchronous acquisition of driving data and parameter self-optimization. By generating a gridded data acquisition path that matches the driving conditions, the system controls the chassis to drive along the preset path and collect multi-dimensional driving data. Combining chassis attitude parameters and sensor calibration information, the data is mapped to a unified operating condition plane coordinate system. Then, based on operating condition feature constraints, model training and parameter fine-tuning are performed to obtain the optimal parameter combination and determine its effectiveness. This achieves the integration of manual parameter adjustment and self-optimization processes.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] The intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system includes: instruction acquisition module, chassis control module, model training module, and parameter verification module;

[0006] The instruction acquisition module acquires manual adjustment instructions and operating condition formulas, and controls the main controller to load initial parameters and driving strategies that match the operating condition formulas according to the manual adjustment instructions; the initial parameters include steering gain, braking response time, and drive torque distribution ratio;

[0007] Under the control of the main controller, the chassis control module completes the initial parameter distribution and executes chassis driving control to obtain a drive-by-wire chassis that operates according to the target parameters. At the same time, it collects driving data through a multi-sensor group. The driving data includes vehicle status, steering angle, yaw rate, braking distance, and energy consumption.

[0008] The model training module receives driving data, combines chassis attitude parameters and sensor calibration information, maps the driving data to the working condition plane coordinate system, fuses the data according to the gridded data acquisition path and trains the model to obtain an optimized parameter model.

[0009] The parameter verification module determines the validity of the parameters output by the optimized parameter model. If the parameters are valid, they are saved to the chassis control library and a parameter version number is generated and synchronized to the background. If the parameters are invalid, the parameter deviation information is recorded and returned to the manual adjustment stage.

[0010] Specifically, the chassis control module includes: a parameter loading unit, a driving control unit, a working condition switching unit, and a parameter temporary storage unit;

[0011] The parameter loading unit controls the main controller to read the manually input adjustment parameters and working condition formulas, parse them to obtain the target parameter values ​​of the steering, braking and drive systems, and load them into the actuator control channel;

[0012] The driving control unit controls the chassis to travel along a preset route according to the road condition type and vehicle status of the working condition formula. During the driving process, the current target parameter values ​​are sent to the execution terminal in real time. The road condition type includes uphill and downhill, unpaved, rain, snow and dust. The vehicle status includes load and attitude. The execution terminal includes a steering system, a braking system and a drive system.

[0013] The working condition switching unit continuously monitors the real-time road condition data fed back by the vehicle sensors. When it detects that the real-time road condition data has changed from the preset road condition type in the working condition formula, the working condition switching unit dynamically adjusts the execution weight of each target parameter value, generates a dynamic parameter sequence, and provides the adjusted target parameter values ​​to the driving control unit in real time.

[0014] After a single driving cycle ends, the parameter storage unit associates the dynamic parameter sequence of this driving cycle, the corresponding real-time driving data, and the working condition recipe identifier, and packages and stores the associated data in the local database to obtain a drive-by-wire chassis that operates according to the target parameters.

[0015] Specifically, the system further includes a parameter correction module, which is configured with correction logic. This correction logic is used to correct parameter deviations in real time during parameter execution, including:

[0016] After the main controller sends the target parameter values ​​to the actuator, it collects the feedback data of the actuator in real time, and at the same time, it collects the attitude sensor data of the vehicle; the feedback data includes at least the actual torque of the steering motor and the actual pressure of the brake master cylinder; the attitude sensor data includes at least the yaw rate and the roll angle.

[0017] The collected feedback data is compared with the corresponding target parameter value to calculate the deviation value. Based on the magnitude and trend of the deviation value, the adjustment direction and adjustment amount are determined. Then, based on the adjustment direction and adjustment amount, the control command output by the main controller to the execution end is corrected, and the process is continuously iterated until the deviation value is reduced and stabilized within the preset deviation threshold range.

[0018] After the deviation value stabilizes, the system maintains the current parameter output gain. At the same time, the parameter correction mechanism combines the collected attitude sensor data and works with the main controller to switch to dynamic correction mode. In dynamic correction mode, the attitude sensor data and execution deviation are comprehensively analyzed by the PID control algorithm to calculate the compensation coefficient, and the compensation coefficient is used to perform real-time dynamic compensation on the parameters output by the main controller.

[0019] Specifically, the calculation process of the compensation coefficient includes:

[0020] The main controller, while keeping the basic parameter output unchanged, collects attitude sensor data and actuator response delay in real time, adjusts the parameter execution sensitivity according to the attitude sensor data, and compares the actuator response delay with a preset delay threshold to determine whether the current parameter setting is in an appropriate state.

[0021] If the system is determined to be in a suitable state, the main controller stops coarse sensitivity adjustment and switches to fine adjustment mode based on road condition characteristics. During the switching process, the continuity of parameter output is maintained through a transition band smoothing algorithm. The road condition characteristics are collected in real time by on-board sensors, including road surface adhesion coefficient, slope, or from pre-stored working condition formulas.

[0022] In fine-tuning mode, the system collects multi-dimensional feedback signals and compares the feedback signals with preset operating condition adaptation requirements. When the feedback signals meet the adaptation requirements, the system determines that the parameters match the current road conditions. The feedback signals include driving stability indicators and energy consumption data.

[0023] Based on the matching results, combined with real-time attitude sensor data and execution deviation, the main controller performs comprehensive analysis using a PID control algorithm to calculate compensation coefficients for core parameters, which are used to correct parameter outputs to meet expected operating conditions; the core parameters include steering gain and braking response time.

[0024] Specifically, the model training module includes: a data acquisition path generation unit, a data acquisition unit, an initial data fusion unit, and a parameter optimization unit;

[0025] The data acquisition path generation unit obtains a preset working condition type and generates a rasterized data acquisition path that matches the working condition type based on the boundary coordinates of the driving range and the working condition accuracy requirements.

[0026] The data acquisition unit receives the gridded data acquisition path and controls the chassis to drive along the gridded data acquisition path. The chassis travels at equal distances along the gridded data acquisition path as the trigger condition, and multi-dimensional driving data is collected through a multi-sensor group.

[0027] The initial data fusion unit receives driving data, combines it with chassis attitude parameters and preset sensor calibration information, maps the driving data to a unified working plane coordinate system, and then integrates the mapped driving data according to the grid position to obtain the initial dataset; the sensor calibration information includes camera intrinsic parameters and radar installation angle calibration data.

[0028] The parameter optimization unit extracts working condition feature anchors from the initial dataset using a feature extraction algorithm. Using these anchors as constraints, it trains and fine-tunes the parameters of a preset self-learning model to generate an optimized parameter model. The working condition feature anchors include at least one of the following: vehicle state threshold, peak steering angle, road surface adhesion coefficient range, and standard braking distance, determined by statistical data features. The self-learning model is trained using a BP neural network.

[0029] Specifically, the model training module is configured with a self-learning strategy. The self-learning strategy trains and optimizes the parameter model based on the driving data collected by the multi-sensor group and outputs the optimal parameter combination. The self-learning strategy includes path planning logic, data mapping logic, and model training logic. The path planning logic is configured in the path generation unit, the data mapping logic is configured in the initial data fusion unit, and the model training logic is configured in the parameter optimization unit.

[0030] Specifically, the path planning logic includes:

[0031] The main acquisition direction is determined by the typical driving trajectory and vehicle state range of the working condition type, and the vertical direction of the main acquisition direction is determined by the road condition feature points and the effective detection range of the sensor. Based on the main acquisition direction and the vertical direction, a working condition plane coordinate system matching the working condition is established and the corresponding acquisition boundary is determined; the road condition feature points include curves, slopes and intersections.

[0032] The established data acquisition boundary is divided into multiple sub-regions according to the working condition type, and grid parameters are set for each sub-region. The types of sub-regions include paved roads in port areas, mine ramps, mine gravel roads, and mining areas. The grid parameters include the distance between acquisition points, the data overlap between adjacent acquisition points, and the boundary redundancy. The distance between acquisition points is determined according to the sensor data update frequency and the preset sampling accuracy.

[0033] Based on the determined main acquisition direction and combined with the set grid parameters for each sub-region, multiple parallel main acquisition lines are generated in each sub-region to form a set of main acquisition lines.

[0034] Within the acquisition boundary, cross-check lines perpendicular to the main acquisition direction are inserted at preset intervals, and the intersection of the cross-check lines with each main acquisition line in the main acquisition line set is set as the data alignment node.

[0035] For each main acquisition line in the set of main acquisition lines, its start and end points are determined in conjunction with the data alignment nodes. At the same time, all main acquisition lines are arranged in sequence so that the end point of an adjacent main acquisition line is adjacent to or coincides with the start point of the next main acquisition line, forming an initial acquisition grid.

[0036] Based on the deviation between the actual road conditions and the typical driving trajectory, the initial data acquisition grid is translated and fine-tuned, and the origin of the initial data acquisition grid is anchored at the starting point of the working condition.

[0037] Output a rasterized data acquisition path; the rasterized data acquisition path includes the main acquisition line sequence, the start and end coordinates of each main acquisition line, and the coordinates of all data alignment nodes.

[0038] Specifically, the data mapping logic includes:

[0039] The chassis is controlled to travel along a gridded data acquisition path. At each acquisition point along the path, the chassis attitude parameters are collected by the onboard inertial measurement unit. These parameters include the chassis's spatial position, yaw rate, roll angle, and pitch angle. Sensor calibration information is retrieved from a pre-set database. For each acquisition point, the chassis attitude parameters are combined with the corresponding sensor extrinsic parameters to obtain a mapping matrix of the sensor data in the working plane coordinate system. The raw driving data collected by each sensor at the corresponding acquisition point is acquired. Combining the sensor intrinsic parameters and the mapping matrix, the raw driving data is converted into standardized data in the working plane coordinate system. Based on the grid index of the gridded data acquisition path, i.e., the grid position information of each acquisition point, the standardized data of each acquisition point is associated with the corresponding grid position. Simultaneously, the standardized data from multiple sensors within the same grid are fused to obtain the initial dataset for grid association.

[0040] Specifically, the model training logic includes:

[0041] Based on the operating condition type, feature anchor point constraint relationships are established, including vehicle state stability constraints, steering response timeliness constraints, heavy-load braking distance rationality constraints, energy consumption economy constraints, and driving comfort constraints. The raster index and corresponding data acquisition location of each sub-region in the rasterized data acquisition path are obtained. The initial dataset is called, and for each sub-region, the operating condition feature anchor points in the driving data are extracted and compared with the feature anchor point constraint relationships to calculate the operating condition feature anchor point deviation for each constraint item. While keeping the raster index unchanged, with the sub-region driving data satisfying the feature anchor point constraint relationships as the objective, and combined with the operating condition feature anchor point deviation, the training parameters of the self-learning model are calculated using the error backpropagation algorithm. Using the output training parameters as the initial configuration, the initial dataset is called as the training sample to train the preset self-learning model, generating an optimized parameter model.

[0042] Specifically, the parameter verification module includes: a data parsing unit, a validity detection unit, and a judgment output unit;

[0043] The data parsing unit receives the optimized parameter model and extracts the initial parameters and corresponding driving performance indicators;

[0044] The validity detection unit uses driving performance indicators as the detection object and performs multi-dimensional validity judgment in combination with the system's preset judgment thresholds. The validity judgment includes at least the detection of driving stability, braking safety, steering control, energy consumption economy and parameter adaptability.

[0045] The judgment output unit receives the detection result. If the detection is qualified, a qualified signal is generated, and the optimal parameter combination in the current initial parameters is fixed to the chassis control library. A unique parameter version number is generated and synchronized to the background archive. If the detection is unqualified, an unqualified signal is generated, and parameter deviation information and corresponding working condition adaptation problems are recorded and then fed back to the manual adjustment stage.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] The intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system proposed in this invention integrates multiple sensor groups on the intelligent heavy-duty truck drive-by-wire chassis and combines gridded data acquisition with a self-learning strategy based on operating condition feature constraints. This achieves the integration of manual parameter adjustment and self-optimization processes, eliminating reliance on independent parameter verification stations and reducing additional investment in testing sites and equipment. By introducing chassis attitude parameters and sensor calibration information in the data fusion stage, it effectively eliminates data deviations and adaptation errors caused by changes in road conditions, improving the accuracy of parameter optimization. Furthermore, through a model training process constrained by operating condition feature anchor points, the optimized parameters are deeply bound to the geometric features and performance requirements of different driving conditions, ensuring more reliable parameter validity judgment results. The overall solution not only shortens the parameter adjustment cycle and avoids operational losses caused by secondary adjustments due to downtime, but also stably completes manual adjustment and self-learning optimization of drive-by-wire chassis parameters under complex and changing driving conditions, improving chassis driving safety, handling, and operational economy. Attached Figure Description

[0048] Figure 1 This is a system architecture diagram of the intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system of the present invention;

[0049] Figure 2 This is a flowchart illustrating the implementation of the compensation coefficient of the intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system of the present invention.

[0050] Figure 3 This is a flowchart illustrating the path planning logic of the intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system of the present invention. Detailed Implementation

[0051] Example 1:

[0052] Please see Figure 1 The present invention provides an embodiment of an intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system, comprising: an instruction acquisition module, a chassis control module, a model training module, and a parameter verification module;

[0053] The instruction acquisition module acquires manual adjustment instructions and operating condition formulas, and controls the main controller to load initial parameters and driving strategies that match the operating condition formulas according to the manual adjustment instructions; the initial parameters include steering gain, braking response time, and drive torque distribution ratio;

[0054] Furthermore, manual adjustment commands can be input through the interactive interface in the cab or remotely issued from the background. Based on this, the main controller reads the initial parameters corresponding to the working condition, load, and driving speed from the local parameter library, and issues them to the parameter execution and chassis control channels under the same task token.

[0055] Furthermore, the main controller first analyzes the manually input adjustment parameters and verifies their rationality, such as whether they are within the safety threshold range. Then, it loads the corresponding driving strategy, such as prioritizing stability and energy consumption under high-speed conditions and prioritizing maneuverability and passability under construction site conditions. During parameter execution, the parameter correction mechanism intervenes: using the actuator feedback deviation as an adaptation signal, it makes small adjustments to the main controller's output parameters to make the parameter execution accuracy more consistent. Then, it records the real-time compensation amount for subsequent model training and parameter optimization. The purpose of this is to incorporate the parameter correction behavior into the first parameter execution process, avoid downtime for secondary adjustments, shorten the optimization cycle, and improve operational efficiency. At the same time, the task token and the working condition recipe are bound one by one to ensure the consistency and traceability of parameters during the self-learning process.

[0056] Those skilled in the art will understand that the source of the manual adjustment command can be the driver's cab interface or an external management system. The formula field can include road condition type, load level, speed range, performance priority, etc. It is only necessary to meet the minimum requirement of retrieving the corresponding initial parameters by token. This application does not impose any further limitations.

[0057] Under the control of the main controller, the chassis control module completes the initial parameter distribution and executes chassis driving control to obtain a drive-by-wire chassis that operates according to the target parameters. At the same time, it collects driving data through a multi-sensor group. The driving data includes vehicle status, steering angle, yaw rate, braking distance, and energy consumption.

[0058] The model training module receives driving data, combines chassis attitude parameters and sensor calibration information, maps the driving data to the working condition plane coordinate system, fuses the data according to the gridded data acquisition path and trains the model to obtain an optimized parameter model.

[0059] The parameter verification module determines the validity of the parameters output by the optimized parameter model. If the parameters are valid, they are saved to the chassis control library and a parameter version number is generated and synchronized to the background. If the parameters are invalid, the parameter deviation information is recorded and returned to the manual adjustment stage.

[0060] Furthermore, this application is particularly applicable to several common and challenging scenarios in low-speed environments in ports and mines, including: the need for precise low-speed control in dense container transshipment at ports, where fixed parameters can easily lead to alignment deviations; the stability requirements for heavy-load transportation on mine ramps, where unpaved / gravel roads can easily cause vehicle bumps; the parameter adaptation requirements for frequent switching between empty / full containers and empty / heavy loads in port / mine operations, where matching accuracy directly affects operational efficiency and driving safety; and the need for precise low-speed steering in densely packed port yards and narrow mine tunnels, requiring rapid parameter adaptation for operations in confined spaces. Even on complex mine ramps, muddy port roads, when sensors are affected by dust or after a temporary shutdown and restart, the system can still restore a consistent parameter optimization coordinate system by using the gridded acquisition path and alignment nodes, ensuring the comparability and traceability of parameters between batches.

[0061] The chassis control module includes: a parameter loading unit, a driving control unit, a working condition switching unit, and a parameter temporary storage unit;

[0062] The parameter loading unit controls the main controller to read the manually input adjustment parameters and working condition formulas, parse them to obtain the target parameter values ​​of the steering, braking and drive systems, and load them into the actuator control channel;

[0063] The driving control unit controls the chassis to travel along a preset route according to the road condition type and vehicle status of the working condition formula. During the driving process, the current target parameter values ​​are sent to the execution terminal in real time. The road condition type includes uphill and downhill, unpaved, rain, snow and dust, etc., and the vehicle status includes load and attitude, etc. The execution terminal includes a steering system, a braking system and a drive system.

[0064] Furthermore, the driving control unit controls the chassis to travel along a preset route according to the road condition type and vehicle status of the working condition formula. During the driving process, it sends the current target parameter values ​​to the execution terminal in real time, including:

[0065] (1) The driving control unit receives the complete operating condition recipe, preset driving route, and target parameter values ​​of the steering system, braking system, and drive system transmitted by the system;

[0066] (2) Conduct in-depth analysis and classification, including: distinguishing different scenarios such as paved roads in port areas, mine ramps, gravel roads, and densely packed yard passages based on the road condition types in the working condition formula, and matching the corresponding basic driving strategies with the low-to-medium speed operation characteristics of each scenario. For example, paved roads in port areas focus on low-speed precise positioning and smooth start-stop, mine ramps focus on heavy-load anti-slip slope and braking stability, and densely packed yard passages focus on precise steering and narrow space avoidance; at the same time, the vehicle status is decomposed into three dimensions: load level, vehicle status range, and attitude safety range. Among them, the load level is... The levels include light load or heavy load, and the vehicle status range is set with low-speed operation 0-15km / h and medium-speed transfer 15-30km / h, providing clear standards for the control of different road sections; then the preset driving route is broken down into continuous short-distance driving segments according to road condition change nodes, such as port loading and unloading points, transfer channels, storage yards, mining points, ramps, and unloading points. Each segment corresponds to the matched and analyzed road condition type and vehicle status adaptation standard. At the same time, the target parameter values ​​are classified and organized according to steering, braking, and drive systems, and the initial control parameter standards of each system are clarified.

[0067] (3) Before starting the driving state initialization verification, first use the vehicle positioning system to confirm whether the current actual position of the chassis is consistent with the starting point of the preset driving route. If there is a deviation, generate a small position correction command to control the chassis to slowly move to the starting position. Then check the working status of the steering, braking and drive actuators one by one to confirm whether each actuator is in normal response mode and there is no fault alarm information. Finally, temporarily load the initial values ​​of the classified target parameters into the control channels of each actuator to complete the parameter pre-configuration before driving.

[0068] (4) After initialization, the driving control unit sends a start command to the drive system. According to the vehicle state adaptation standard corresponding to the current road segment, the drive motor is controlled to gradually output torque so that the chassis can accelerate smoothly from a stationary state. During the acceleration process, the actual vehicle state is monitored in real time to ensure that the vehicle state rise rate meets the smoothness requirements of the current road conditions and avoid instability caused by rapid acceleration. At the same time, according to the first segment path information of the preset driving route, the initial steering command is sent to the steering system to guide the chassis to drive along the starting direction of the preset driving route.

[0069] (5) During the chassis driving process, the driving control unit maintains high-frequency real-time monitoring and continuously collects actual road condition information such as port ground flatness, container stacking density, mine slope, and road gravel distribution through road condition sensors. It compares the actual road condition with the preset road condition type in the working condition formula to determine whether the actual road condition is consistent with the preset road condition. It also obtains load data, actual driving vehicle status, yaw rate, roll angle, pitch angle, and other data in real time through vehicle status sensors to determine the driving stability and vehicle status adaptability. At the same time, it receives the execution feedback information of each actuator on the target parameters in real time through the actuator feedback channel to confirm whether the parameters are executed accurately as required.

[0070] (6) Based on the real-time monitored data, the driving control unit dynamically adjusts the target parameter value. If the actual road conditions are consistent with the preset and the vehicle status is within a safe range, the current target parameter value remains unchanged. If the actual road conditions are detected to change, such as driving from the paved road in the port area to the gravel road in the mine, and the road surface becomes more bumpy, the steering gain is increased and the braking response time is shortened accordingly to improve the chassis's handling stability and anti-bumping ability. If the load is detected to switch from empty to full, the steering gain is reduced and the drive torque distribution ratio is optimized to ensure steering accuracy and power sufficiency under low speed and heavy load. If the pitch angle is detected to exceed 10 degrees when driving on the mine slope, the braking system execution weight is immediately increased and the drive system is switched to anti-slip mode to avoid the vehicle body sliding. If the execution terminal feedback parameter execution has a deviation, the output amplitude of the target parameter is specifically corrected to ensure that the execution effect meets expectations.

[0071] (7) After the parameter adjustment is completed, the driving control unit sends the adjusted target parameter values ​​to the corresponding execution terminals in real time at a frequency of milliseconds. The steering parameters are sent to the steering motor execution terminal, the braking parameters are sent to the brake master cylinder execution terminal, and the drive parameters are sent to the drive motor controller. The parameter effective timestamp is attached during the sending process to ensure that each execution terminal receives and executes synchronously. At the same time, the parameter reception confirmation signal of each execution terminal is received. If no confirmation signal is received, the parameter is immediately resent until the execution terminal reports successful reception to avoid failure of medium and low speed control due to parameter transmission omission.

[0072] (8) When the chassis reaches the end of the current road segment, such as the end of the transfer channel or the top of the slope, the driving control unit confirms the location information through the positioning system and automatically switches to the preset parameters of the next driving segment, including the corresponding road condition type, vehicle status adaptation standard and target parameter initial value. The above process of real-time monitoring, dynamic adjustment and parameter distribution is repeated until the chassis completes all segments of the preset driving route, such as from the loading and unloading position to the storage yard, or from the mining point to the unloading point.

[0073] The working condition switching unit continuously monitors the real-time road condition data fed back by the vehicle sensors. When it detects that the real-time road condition data has changed from the preset road condition type in the working condition formula, the working condition switching unit dynamically adjusts the execution weight of each target parameter value, generates a dynamic parameter sequence, and provides the adjusted target parameter values ​​to the driving control unit in real time.

[0074] Furthermore, the formation process of the dynamic parameter sequence includes:

[0075] (1) The original road condition data is collected by combining multiple on-board sensors. The camera and lidar work together to collect features such as the gap between port containers, the steps of the mine slope, the smoothness of the road surface, cracks and potholes. The millimeter-wave radar and ultrasonic sensor collect data such as the density of port operating vehicles, the passing distance of mining transport vehicles, and the distance to obstacles to determine the degree of congestion in the traffic environment. The wheel speed sensor and inertial measurement unit collect data such as vehicle speed fluctuation, vehicle body bump frequency, yaw rate, side tilt angle, and pitch angle to assist in sensing the road surface adhesion conditions and vehicle body stability. The load sensor collects real-time load data, and the tire pressure sensor and brake feedback sensor indirectly reflect the changes in road resistance. The collected original road condition data is processed by noise reduction filtering and time synchronization to generate standardized real-time road condition data containing three dimensions: road surface level, obstacle level, and vehicle state adaptability.

[0076] (2) The working condition switching unit dynamically compares the standardized real-time road condition data with the preset road condition type and vehicle status adaptation standard in the working condition recipe: First, it extracts key feature values ​​from the real-time road condition data, such as comparing the road surface smoothness index with the preset port area paved road smoothness threshold, the slope gradient with the preset mine slope gradient threshold, and the load data with the preset light / heavy load threshold. At the same time, it calculates the overall matching degree between the real-time status and the preset standard. The matching degree calculation adopts the weighted scoring method, in which the obstacle density accounts for 40% and the road surface smoothness accounts for 40% in the port scenario. 30% for vehicle status adaptability; 40% for road surface adhesion conditions in mining scenarios, 30% for load adaptability, and 30% for vehicle body stability. A comprehensive score above 70% is considered a match, while a score below 70% triggers a status change warning. If the matching degree is between 50% and 70%, it is considered a gradual change, such as from a port transfer channel to a storage yard. If the matching degree is below 50%, it is considered a sudden change, such as suddenly entering a muddy mining road or switching to a heavy load. The judgment process combines the data change trends of three consecutive collection cycles to avoid misjudgment due to sensor false detection in a single cycle.

[0077] (3) After confirming that the road conditions have changed, the working condition switching unit starts the dynamic adjustment process of the target parameter execution weight: First, it calls the pre-stored weight adjustment rule library. This rule library is based on the port mine real vehicle test data and contains weight adjustment strategies under different road condition change scenarios. For example, when it is determined that the port area paved road changes suddenly to the mine slope heavy load working condition, the weight adjustment rule library clearly increases the steering system execution weight from 30% to 45%, the braking system execution weight from 20% to 30%, and the drive system execution weight from 50% to 25%. If it is determined that the port area transfer channel changes gradually to the yard dense channel, the weight adjustment adopts the gradient adjustment strategy. Each collection cycle is adjusted by 5% step by step to avoid the sudden change of weight causing the vehicle body attitude fluctuation. During the adjustment process, the yaw rate, roll angle and other data fed back by the vehicle attitude sensor will be received in real time as the feedback basis for weight adjustment. For example, when the steering weight is increased, if the yaw rate fluctuation is detected to exceed the threshold, the weight increase is paused and 2% is reverted to ensure the driving stability during the adjustment process.

[0078] (4) After the weight adjustment is completed, the working condition switching unit will perform a weighted summation operation on the adjusted system execution weights and the corresponding target parameter benchmark values ​​to generate a dynamic parameter sequence;

[0079] (5) The working condition switching unit sends the generated dynamic parameter sequence to the driving control unit in real time, and simultaneously synchronizes it to the parameter temporary storage unit for associated storage. The stored content includes the timestamp of the road condition change event corresponding to the dynamic parameter sequence, real-time road condition data, adjustment basis and other information. The transmission process adopts a breakpoint resume mechanism. If the transmission is interrupted, only the sequence nodes that were not successfully sent are transmitted after reconnection, ensuring the integrity of the dynamic parameter sequence transmission. After receiving the dynamic parameter sequence, the driving control unit will switch the parameter execution step by step according to the effective time interval in the sequence. The working condition switching unit continuously monitors the driving status data after the parameter switching. If it is detected that the parameter execution effect does not meet expectations, such as the vehicle stability index still exceeding the standard, the weight adjustment and parameter generation process is repeated to ensure that the chassis driving status is always adapted to the real-time road conditions.

[0080] After a single driving cycle ends, the parameter storage unit associates the dynamic parameter sequence of this driving cycle, the corresponding real-time driving data, and the working condition recipe identifier, and packages and stores the associated data in the local database to obtain a drive-by-wire chassis that operates according to the target parameters.

[0081] The system also includes a parameter correction module, which is configured with correction logic. This correction logic is used to correct parameter deviations in real time during parameter execution, including:

[0082] A1: After the main controller sends the target parameter value to the actuator, the feedback data of the actuator is collected in real time. At the same time, the attitude sensor data of the vehicle is collected. The feedback data includes at least the actual torque of the steering motor and the actual pressure of the brake master cylinder. The attitude sensor data includes at least the yaw rate and the roll angle.

[0083] A2: Compare the collected feedback data with the corresponding target parameter values, calculate the deviation value, and determine the adjustment direction and adjustment amount based on the magnitude and trend of the deviation value. Then, based on the adjustment direction and adjustment amount, correct the control command output by the main controller to the execution end, and continue to iterate this process until the deviation value is reduced and stabilized within the preset deviation threshold range.

[0084] Furthermore, the deviation value is calculated by subtracting the target parameter value from the actual value of the feedback data to obtain the absolute value and sign of the deviation. A positive deviation indicates that the actual value is higher than the target value, and a negative deviation indicates that the actual value is lower than the target value. For example, if the target value of the steering gain is 5.0 and the actual feedback value is 5.5, then the deviation value is positive 0.5.

[0085] Furthermore, the deviation change trend is evaluated by combining the preset trend analysis rules: if the absolute value of the deviation gradually decreases for three consecutive periods, it is determined to be a deviation convergence trend; if the absolute value of the deviation increases for three consecutive periods and the rate of change exceeds 5%, it is determined to be a deviation divergence trend; if the absolute value of the deviation fluctuates within a small range and the fluctuation amplitude is less than 30% of the threshold, it is determined to be a deviation stable fluctuation trend.

[0086] Furthermore, the determination of the adjustment direction is directly based on the sign of the deviation and the system characteristics, including: in the steering system, a positive deviation corresponds to a lowering adjustment direction, and a negative deviation corresponds to an increasing adjustment direction; in the braking system, a positive deviation corresponds to a shortening adjustment direction, and a negative deviation corresponds to an extending adjustment direction; in the drive system, a positive deviation corresponds to a balanced distribution adjustment direction, and a negative deviation corresponds to a biased distribution adjustment direction. The adjustment amount is determined by a calculation method that combines a basic adjustment amount with a dynamic correction amount. This includes: the basic adjustment amount is determined based on the ratio of the absolute value of the deviation to the threshold. For example, when the absolute value of the deviation accounts for 50% of the threshold, the basic adjustment amount is set to 5% of the target parameter value; the dynamic correction amount is adjusted in combination with the deviation change trend and related factors. When the deviation shows a diverging trend, the correction amount is increased by 30% on the basis of the basic adjustment amount; when the deviation shows a converging trend, the correction amount is reduced by 20% on the basis of the basic adjustment amount; if the deviation is related to road conditions, a road condition adaptation coefficient is also added. For example, the braking adjustment amount for a mining gravel road is increased by 30%, the steering adjustment amount for a port muddy road is increased by 25%, and the driving adjustment amount for a mining heavy-duty ramp is increased by 40%. Finally, a combination scheme of adjustment direction and adjustment amount for each system is generated. For example, if the steering system's adjustment direction is to decrease, the adjustment amount is 3%.

[0087] Furthermore, the control commands output from the main controller to the actuators are modified: the steering system adjusts the control voltage of the steering motor to increase or decrease the steering gain; the braking system adjusts the pressure supply rate of the master cylinder to shorten or lengthen the braking response time; the drive system adjusts the control current of each drive motor to change the torque distribution ratio. The modified control commands are then sent to the corresponding actuators, along with a timestamp of the command taking effect and an adjustment identifier. After receiving the control commands, the actuators provide real-time feedback of the command reception confirmation signal and the preliminary result of the command execution. If the main controller does not receive the confirmation signal within 50 milliseconds, it immediately resends the command to avoid transmission omissions. If the preliminary result of the execution shows an abnormal command execution, such as the adjustment amount exceeding the maximum adjustment range of the actuator, the current adjustment is paused, and a backup adjustment scheme is switched, such as reducing the adjustment amount to the actuator's standard range.

[0088] A3: After the deviation value stabilizes, the system maintains the current parameter output gain. At the same time, the parameter correction mechanism combines the collected attitude sensor data and works with the main controller to switch to dynamic correction mode. In dynamic correction mode, the attitude sensor data and execution deviation are comprehensively analyzed through the PID control algorithm to calculate the compensation coefficient. The compensation coefficient is then used to perform real-time dynamic compensation on the parameters output by the main controller to adapt to the stability requirements of medium and low speed heavy load scenarios. The PID control algorithm is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0089] Furthermore, after the deviation value stabilizes, the system maintains the current parameter output gain. Simultaneously, the parameter correction mechanism, in conjunction with the acquired attitude sensor data and the main controller, switches to dynamic correction mode, including:

[0090] (1) After the deviation value stabilizes, the main controller immediately triggers the parameter output gain locking mechanism: the parameter output gain values ​​of the current steering, braking and drive systems, such as steering gain amplification coefficient, braking pressure output gain and torque distribution gain, are cached in a dedicated register. At the same time, a gain locking command is sent to the execution end of each system to prevent the triggering of the gain adjustment command in non-dynamic correction mode. The register adopts dual backup storage to prevent the loss of gain due to the failure of a single storage unit. After the command is sent, the locking confirmation signal of the execution end must be received. If it is not received, it is sent again until confirmation is received to ensure that the gain locking is effective.

[0091] (2) The parameter calibration mechanism starts the acquisition and preprocessing of attitude sensor data, including: The parameter calibration mechanism first sends a data acquisition start command to the vehicle attitude sensor cluster. The vehicle attitude sensor cluster includes an inertial measurement unit, a fiber optic gyroscope, a longitudinal accelerometer, and a lateral accelerometer, which are responsible for acquiring the spatial attitude parameters of the chassis respectively: The inertial measurement unit acquires the real-time values ​​of yaw rate, roll angle, and pitch angle; the fiber optic gyroscope assists in correcting the cumulative error of yaw rate; and the longitudinal and lateral accelerometers supplement the acceleration change data of low-speed start-stop in the port and driving on the mine slope. The acquired raw data needs to be preprocessed in three stages: The first stage is through Kalman oscillator. The filtering algorithm removes instantaneous noise, such as sudden fluctuations in the roll angle caused by the bumps of the mine's gravel road and fluctuations in the pitch angle caused by the uneven ground in the port's container loading and unloading area. The second stage synchronizes the data from multiple sensors through timestamp alignment. The third stage converts the data into standardized data consistent with the current working condition plane coordinate system through working condition adaptation calibration. For example, the pitch angle value is corrected according to the slope parameter of the mine ramp, and the yaw rate threshold is corrected according to the width of the port yard passage. Finally, a four-dimensional attitude data sequence containing yaw rate, roll angle, pitch angle and acceleration is generated. The Kalman filter algorithm is prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0092] (3) The parameter calibration mechanism sends a dynamic calibration mode switching request to the main controller through the internal CAN bus of the system. The request information includes a deviation stability confirmation certificate and a summary of the attitude data sequence. After receiving the request, the main controller returns a cooperative ready signal to the parameter calibration mechanism and calls the preset dynamic calibration mode configuration parameter library. The dynamic calibration mode configuration parameter library includes basic configurations such as calibration cycle, data acquisition priority, and fault emergency strategy under different working conditions.

[0093] (4) Coordinated switching and parameter connection of dynamic correction mode: The main controller first sends a mode switching warning instruction to each system execution terminal. The mode switching warning instruction specifies the switching countdown, usually set to two hundred milliseconds, and the parameter maintenance rules during the switching period. That is, the execution terminal is required to maintain the output parameters corresponding to the current gain and there should be no parameter mutation. At the same time, the parameter correction mechanism loads the working parameters specific to the dynamic correction mode: including the acquisition period of attitude sensor data, data fusion algorithm parameters, and feedback format of correction results. After the countdown ends, the main controller sends a mode switching execution instruction to the parameter correction mechanism. The parameter correction mechanism immediately sends back a confirmation of receipt. Both parties trigger the mode switching synchronously: the main controller switches the system control logic from the deviation correction mode to the dynamic correction mode, closes the original deviation iteration adjustment channel, and opens the dedicated data interaction channel with the parameter correction mechanism; the parameter correction mechanism starts the real-time acquisition and transmission link of attitude sensor data and continuously sends the preprocessed attitude sensor data to the main controller according to the correction cycle.

[0094] Furthermore, in the parameter execution chain of the online control chassis, the actual execution effect of parameters always has unavoidable uncertainties. For example, in the parameter issuance stage, the actuator may experience slight execution deviations due to wear from mine dust, corrosion from the humid environment of the port, temperature changes, or power supply fluctuations. During travel, changes in road conditions such as sudden changes in the slope of mine ramps and a sharp drop in the adhesion coefficient of muddy roads in ports may also lead to a decrease in parameter adaptability. If not corrected, these small cumulative deviations will be amplified during low-to-medium speed heavy-load travel, manifesting as port alignment deviations, risks of slippage on mine ramps, and increased vehicle body vibration, thereby directly affecting operational efficiency and driving safety. Traditional methods usually rely on manual readjustment after shutdown or parameter calibration at an independent test site, but this not only prolongs the operating cycle but also increases the complexity of the main controller's path planning and operating costs. This embodiment integrates a parameter correction mechanism at the execution end and directly introduces feedback and dynamic correction modes during parameter execution, achieving rapid correction of parameter deviations and thus avoiding additional shutdown adjustments.

[0095] Example 2:

[0096] Please see Figure 2 The calculation process of the compensation coefficient in this embodiment includes:

[0097] B1: While keeping the basic parameter output unchanged, the main controller collects attitude sensor data and actuator response delay in real time, adjusts the parameter execution sensitivity according to the attitude sensor data, and compares the actuator response delay with the preset delay threshold to determine whether the current parameter setting is in an adapted state.

[0098] Furthermore, attitude sensor data is collected through an on-board attitude sensing cluster to generate an attitude data sequence, and actuator response data is synchronously captured through the actuator feedback channel to generate an actuator response delay sequence.

[0099] Furthermore, the sensitivity of parameter execution is adjusted based on attitude sensor data. This includes: the main controller performing feature analysis on the attitude data sequence: extracting the fluctuation amplitude of yaw rate, the rate of change of roll angle, and the correlation between pitch angle and vehicle state. The feature values ​​are compared with preset sensitivity adaptation standards. For example, if the fluctuation amplitude of yaw rate exceeds 5 degrees per second, it indicates that the current steering sensitivity is insufficient, and the execution sensitivity of steering parameters needs to be increased. For example, the voltage amplification factor of the steering motor is adjusted from 1.0 times the base value to 1.2 times. If the rate of change of roll angle is too fast, such as exceeding 1 degree per millisecond, it indicates that the chassis attitude adjustment is lagging, and the steering sensitivity needs to be reduced to avoid overcorrection. For example, the voltage amplification factor is adjusted to 0.9 times. Sensitivity adjustment adopts a gradient fine-tuning strategy: each adjustment does not exceed 5% of the base value. After adjustment, the new sensitivity parameter is immediately written into the control buffer of the execution end, but the locking state of the parameter base output value is not changed, ensuring that the sensitivity adjustment only affects the execution response speed of the parameter and does not change the target value of the core parameter.

[0100] Furthermore, the fluctuation range of the yaw rate is calculated by the difference between the maximum and minimum values ​​over N consecutive acquisition cycles; the rate of change of the roll angle is calculated by the absolute value of the difference between adjacent cycles; and the correlation between the pitch angle and the vehicle state is determined by whether the pitch angle changes linearly when the vehicle state improves.

[0101] Furthermore, the main controller compares the real-time delay values ​​in the actuator response delay sequence with the condition-specific delay thresholds set during the initialization phase: in the port scenario, the steering system response delay is less than or equal to 0.3 seconds, the braking system less than or equal to 0.4 seconds, and the drive system less than or equal to 0.3 seconds; in the mining scenario, the steering system response delay is less than or equal to 0.35 seconds, the braking system less than or equal to 0.3 seconds, and the drive system less than or equal to 0.25 seconds; and this condition is met for three consecutive acquisition cycles, while the attitude data feature values ​​are all within the preset stable range, such as small yaw rate fluctuations in the port. If the pitch angle fluctuation is less than or equal to 0.5 degrees per second and the mine pitch angle fluctuation is less than or equal to 0.3 degrees per second, the current parameter settings are determined to be in an adapted state. An adaptation confirmation signal is generated and uploaded to the system monitoring module, while the current sensitivity parameters remain unchanged. If the response delay of any system exceeds the threshold, or the attitude data characteristic value exceeds the stable range, it is determined to be in an unadapted state: the deviation data is immediately recorded, an adaptation adjustment command is generated and fed back to the parameter correction mechanism, triggering the subsequent parameter optimization process; at the same time, sensitivity fine-tuning is paused and the sensitivity parameters are restored to the state before adjustment to avoid chassis attitude instability due to unadapted adjustment.

[0102] B2: If the system is determined to be in an adaptation state, the main controller stops coarse sensitivity adjustment and switches to fine adjustment mode based on road condition characteristics. During the switching process, the continuity of parameter output is maintained through a transition band smoothing algorithm. The road condition characteristics are collected in real time by on-board sensors, including road surface smoothness, container gaps, and operating vehicle density in port scenarios, and road surface adhesion coefficient, slope, and gravel distribution in mining scenarios, or from pre-stored working condition formulas. The transition band smoothing algorithm is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0103] Furthermore, once the main controller confirms, through coarse sensitivity adjustment and adaptation status determination, that the current parameter settings are in an adapted state—that is, the attitude data characteristics are stable and the actuator response delay is within a preset delay threshold—the system immediately initiates the switch from coarse sensitivity adjustment to fine-tuning mode based on road condition characteristics, including:

[0104] (1) The main controller performs secondary confirmation of the adaptation state and generates the coarse adjustment stop command: The main controller retrieves the last 5 consecutive acquisition cycles of data that are determined to be in the adaptation state, and compares them again with the preset adaptation standard. For example, it confirms that the yaw rate fluctuation is always less than or equal to 0.5 degrees per second and the steering response delay is stable at 0.18 seconds to ensure the stability of the adaptation state and avoid false switching due to instantaneous data meeting the standard. After confirming that there are no errors, the main controller generates a coarse adjustment stop command, which includes the timestamp of stopping coarse adjustment, the currently locked sensitivity parameters after coarse adjustment, and sends it to the parameter adjustment module through a dedicated control channel.

[0105] (2) After parsing the coarse adjustment stop command, the parameter adjustment module first shuts down the attitude data-sensitivity correlation adjustment algorithm in the coarse adjustment stage, that is, stops the logic of dynamically modifying the sensitivity parameters according to the real-time changes of yaw rate and roll angle; then, it stores the currently effective sensitivity parameters into the initial parameter cache area of ​​fine adjustment mode. At the same time, the parameter adjustment module sends a confirmation signal to the main controller that the coarse adjustment has stopped, along with a copy of the cached sensitivity parameters.

[0106] (3) The main controller synchronously collects road condition feature data of the current driving section through the vehicle-mounted multi-sensor fusion system, and generates a standardized road condition feature vector after preprocessing;

[0107] (4) The main controller matches the dedicated parameter configuration library of the fine-tuning mode based on the standardized road condition feature vector. The system has a pre-stored fine-tuning parameter configuration library covering typical sub-scenarios of ports and mines. Each configuration contains information such as road condition feature matching conditions, fine-tuning adjustment cycle, parameter adjustment range limit, priority weight, etc. The main controller compares the currently generated road condition feature vector with the matching conditions in the configuration library one by one. For example, if the road condition feature vector shows a dense channel in the port yard, such as a channel width of 5 meters and a container gap of 0.8 meters, then the fine-tuning configuration corresponding to the dense channel in the port area is matched to shorten the adjustment cycle and reduce the steering adjustment range; if it shows a steep slope in the mine, such as a slope of 15 degrees and heavy load, then the configuration of the heavy load slope in the mine is matched to extend the braking adjustment cycle and increase the driving torque adjustment weight. The adjustment cycle, range limit, priority and other parameters in the configuration are extracted to generate the current road condition fine-tuning configuration table.

[0108] (5) The main controller sends a command to start fine-tuning mode, along with the address of the current road condition fine-tuning configuration table and the fine-tuning initial parameter cache area. After receiving the fine-tuning mode command, the parameter adjustment module first reads the fine-tuning initial parameters from the specified address, that is, the sensitivity parameters after coarse-tuning, as the starting value for fine-tuning; then it loads the current road condition fine-tuning configuration table and initializes it.

[0109] (6) After initialization, a fine-tuning mode ready signal is sent to the main controller. The main controller then starts the fine-tuning mode. The parameter adjustment module fine-tunes the configuration table according to the current road conditions and continuously receives the real-time road condition feature vector pushed by the main controller.

[0110] (7) The main controller uploads a status report that has been switched to fine-tuning mode to the system monitoring module to complete the mode switch. The status report includes the switching timestamp, current road condition feature vector, current road condition fine-tuning configuration table summary, and fine-tuning initial parameter values.

[0111] B3: In fine-tuning mode, the system collects multi-dimensional feedback signals and compares the feedback signals with preset operating condition adaptation requirements. When the feedback signals meet the adaptation requirements, the system determines that the parameters match the current road conditions. The feedback signals include driving stability indicators and energy consumption data.

[0112] B4: Based on the matching results, combined with real-time attitude sensor data and execution deviation, the main controller performs comprehensive analysis through a PID control algorithm to calculate compensation coefficients for core parameters, which are used to correct parameter outputs to meet expected operating conditions; the core parameters include steering gain and braking response time.

[0113] The model training module includes: a data acquisition path generation unit, a data acquisition unit, an initial data fusion unit, and a parameter optimization unit;

[0114] The data acquisition path generation unit obtains a preset working condition type and generates a rasterized data acquisition path that matches the working condition type based on the boundary coordinates of the driving range and the working condition accuracy requirements.

[0115] Furthermore, the boundary coordinates of the driving range are defined by Cartesian coordinate points, but the definition of Cartesian coordinate points is prior art in this field and is not an inventive solution of this application, so it will not be elaborated here.

[0116] Furthermore, the standards for setting the accuracy requirements under different operating conditions are as follows: In medium- and low-speed operating scenarios such as dense passageways in port yards and narrow mine tunnels, due to the limited operating space and the need for frequent and precise turning and alignment, such as container loading and unloading and vehicle passing in mine tunnels, excessively large sampling intervals may result in the loss of key control data such as steering angles and vehicle posture. Therefore, higher accuracy is required, i.e., smaller grid cells are used to ensure the accuracy of parameter adjustments. In contrast, in medium- and low-speed scenarios such as mining areas and open-pit port transfer yards, the operating area has no dense obstacles and the driving path is relatively regular. The core focus is on the regional coverage integrity of driving data rather than high-frequency detail capture. The accuracy requirements can be appropriately reduced, and larger grid cells are used to reduce data redundancy and improve the efficiency of data acquisition and model training.

[0117] Furthermore, after clarifying the accuracy requirements for the working conditions, the acquisition path generation unit begins to calculate the grid parameters: based on the size and shape of the driving range and the determined accuracy, it calculates how many rows and columns of grids the entire area needs to be divided into. Specifically, a boundary alignment algorithm is used to fine-tune the starting point and direction of the grid so that it is parallel to or coincides with the main boundary of the driving range, thereby ensuring that the vast majority of grids are standard rectangles or squares. The boundary alignment algorithm is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0118] Furthermore, after the number of rows and columns of the grid is determined, the acquisition path generation unit begins to generate the coordinate information of each grid one by one: for each grid, the precise coordinates of its four vertices and the coordinates of the grid center point are calculated, where the center point is the target acquisition point; the acquisition path generation unit arranges these points from top to bottom or from left to right to form a continuous, non-repeating main acquisition line sequence, i.e., the gridded data acquisition path. The main acquisition line sequence ensures that the autonomous vehicle can travel along a preset, efficient S-shaped or zigzag route, passing through all acquisition points in sequence, avoiding repeated acquisition or omissions that may be caused by random driving.

[0119] The data acquisition unit receives the gridded data acquisition path and controls the chassis to drive along the path. The chassis travels at equal distances along the gridded data acquisition path as the trigger condition, and multi-dimensional driving data is collected through the vehicle-mounted multi-sensor group.

[0120] Furthermore, the data acquisition unit receives the gridded data acquisition path and controls the chassis to travel along the path. Using the equidistant travel distance of the chassis along the gridded data acquisition path as a trigger condition, it collects multi-dimensional driving data through the onboard multi-sensor group, including:

[0121] (1) Receive the rasterized data acquisition path and perform in-depth analysis to convert it into a path point sequence in the local coordinate system of the port / mine that the chassis control system can understand and execute, such as the port area UTM coordinate system and the mine local operation coordinate system;

[0122] (2) The parsed path point sequence is converted into continuous control commands for the chassis drive system and steering system according to the pure tracking algorithm. The pure tracking algorithm is the prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0123] (3) While controlling the chassis to travel, the time when the chassis passes the first collection point, such as the starting point of the container loading and unloading position, the starting point of the mining point, or the starting point of the path, is used as the benchmark. The wheel speed pulse signal and the odometer data of the vehicle inertial measurement unit are used to calculate the cumulative travel distance of the chassis relative to the starting point. The cumulative travel distance is then compared periodically with the preset equidistant interval, such as every 5 meters.

[0124] (4) When the cumulative driving distance is detected to reach or exceed the preset equal interval, the triggering condition is met. At this time, a synchronous data acquisition trigger signal is generated and sent to all sensors in the vehicle multi-sensor group at the same time.

[0125] (5) After receiving the data acquisition trigger signal, each sensor synchronously starts a data acquisition once within a millisecond time to obtain multi-dimensional driving data, including the alignment deviation in the port scene, the slope of the mine scene, heavy load torque, etc.

[0126] (6) The collected multi-dimensional raw data is temporarily stored in the local buffer of the sensor and then transmitted to the main storage system of the data acquisition unit through the high-speed data bus, prioritizing the storage of key scenario data such as mine slope braking data and port alignment and turning data;

[0127] (7) After completing a data acquisition, reset the internal distance counter or record the current trigger position, such as the port channel inflection point or the midpoint of the mine slope, and start calculating the next equidistant interval, waiting for the next trigger condition to be met. At the same time, the chassis driving control is not interrupted and continues to move along the preset path, such as the yard transfer route or the mine unloading route.

[0128] The initial data fusion unit receives multi-dimensional driving data, combines it with chassis attitude parameters and preset sensor calibration information, maps the driving data to a unified working plane coordinate system, and then integrates the mapped driving data according to the grid position to obtain the initial dataset; the sensor calibration information includes camera intrinsic parameters and radar installation angle calibration data.

[0129] Specifically, the chassis's spatial position, yaw rate, roll angle, and pitch angle at each collection point are acquired in real time using onboard inertial measurement units and wheel speed sensors. These attitude parameters are combined with pre-calibrated sensor extrinsic parameters to establish a mapping matrix of sensor data in the working plane coordinate system, thereby associating the sensor data with the physical space of port / mining operations. Combined with sensor intrinsic parameters, the raw data from each collection point can be mapped one by one to the working plane coordinate system, forming standardized data corresponding to the actual working scenario. Subsequently, the standardized data is fused according to the rasterized data acquisition path, enabling seamless connection of data from each collection point within a unified reference system, thus obtaining the initial dataset. Because this process uses a combination of multiple attitude acquisitions and coordinate transformations, it can effectively offset the error accumulation caused by attitude deviations due to chassis movement, making the fused dataset consistent with the actual working conditions in terms of characteristics.

[0130] Those skilled in the art will understand that the acquisition of internal and external parameters can be achieved through conventional sensor calibration experiments, and the acquisition of attitude parameters can be obtained by relying on conventional sensor groups and attitude calculation algorithms of the wire-controlled chassis. Therefore, this mapping and conversion method does not depend on specific hardware. It only needs to provide attitude information and calibration data to the minimum to complete the coordinate unification. This application does not impose any further limitations here.

[0131] The parameter optimization unit extracts working condition feature anchor points from the initial dataset using a feature extraction algorithm. Using these anchor points as constraints, it trains and fine-tunes the parameters of a pre-defined self-learning model to generate an optimized parameter model. The working condition feature anchor points include at least one of the following: port alignment accuracy threshold, mine heavy-load vehicle speed threshold, peak steering angle, road surface adhesion coefficient range, and standard value of heavy-load braking distance, determined by statistical characteristics of actual vehicles in ports / mines. The self-learning model is trained using a BP neural network. The feature extraction algorithm employs principal component analysis (PCA). Both PCA and BP neural networks are existing technologies in this field and are not inventive solutions for this application; therefore, they will not be elaborated upon here.

[0132] The model training module is equipped with a self-learning strategy. The self-learning strategy trains and optimizes the parameter model based on the multi-dimensional driving data collected by the multi-sensor group and outputs the optimal parameter combination. The self-learning strategy includes path planning logic, data mapping logic and model training logic. The path planning logic is configured in the path generation unit, the data mapping logic is configured in the initial data fusion unit, and the model training logic is configured in the parameter optimization unit.

[0133] Furthermore, the multi-dimensional driving data is first mapped to the working condition plane through chassis attitude and sensor calibration, forming an initial dataset aligned with port / mine working conditions. Subsequently, the self-learning model is trained and its parameters are fine-tuned using a set of working condition feature anchor points consisting of port alignment accuracy threshold, mine heavy-load vehicle speed threshold, peak steering angle, road surface adhesion coefficient range, and heavy-load braking distance standard value as constraints. This ensures that the parameter optimization results converge consistently across the entire working condition range in terms of port precision alignment, mine slope stability, and heavy-load energy consumption. This sequential change is not simply an algorithm replacement; it aims to use working condition knowledge as the guiding principle for parameter optimization, avoiding optimization deviations caused by purely data-driven approaches in complex scenarios or areas with data noise, such as mine gravel roads or port muddy roads.

[0134] In this embodiment, the multi-sensor group is directly integrated into the drive-by-wire chassis. The chassis can simultaneously complete multi-dimensional driving data collection while performing normal operating trajectories such as port container transfer and heavy-duty mining transportation, thus eliminating the need for separate test sites and verification stations. This approach avoids additional hardware and site layout. In traditional single-sensor systems, due to the complex and ever-changing operating scenarios in ports and mines, such as dense passageways in port areas and steep, gravel roads in mines, the data collected in a single trip often fails to cover all scene features, resulting only in local scene data. Furthermore, the chassis's attitude and position constantly change in different scenarios, such as inclined mine slopes and turning port passageways. Direct data fusion often leads to error accumulation and feature distortion. Without processing, the final dataset cannot fully reflect the needs of port and mine operations, nor can it accurately support parameter optimization indicators such as precise alignment, slope anti-slippage, and anti-bump stability. However, in this application, the application of a multi-sensor group allows for direct integration into key locations on the chassis without significant modifications to the drive-by-wire chassis structure. Compared to a single sensor, the multi-sensor group can collect multi-dimensional data such as port alignment deviation, mine slope, heavy-load torque, and vehicle posture, improving the ability to perceive complex working conditions. At the same time, it can cover all working scenarios as the chassis operates, realizing an integrated process of data collection during operation. This avoids the complexity of additionally setting up testing equipment and simulating port and mine road conditions in independent test sites, and also reduces the downtime risk caused by equipment updates or maintenance, thereby improving the overall continuity and flexibility of operation.

[0135] Example 3:

[0136] Please see Figure 3 The path planning logic described in this embodiment includes:

[0137] C1: The main data collection direction is determined based on typical driving trajectories of different working conditions, such as port loading / unloading-stockyard transfer trajectories and mine excavation-unloading ramp trajectories, as well as vehicle status intervals. The vertical direction of the main data collection direction is determined based on road condition feature points and the effective detection range of sensors. Based on the main data collection direction and the vertical direction, a working condition plane coordinate system matching the working conditions is established and the corresponding data collection boundaries are determined. The road condition feature points include container loading / unloading positions in ports, turning points of stockyard passages, starting / ending points of mine ramps, and boundary points of gravel road sections.

[0138] C2: Divide the determined data acquisition boundary into multiple sub-regions according to the working condition type, and set grid parameters for each sub-region; the types of the sub-regions include paved roads in port areas, dense passages in port areas, mine ramps, gravel roads in mines, and mining areas; the grid parameters include the spacing between acquisition points, the data overlap between adjacent acquisition points, and the boundary redundancy. The spacing between acquisition points is determined according to the sensor data update frequency and the preset sampling accuracy, such as 1 meter for dense passages in port areas and 3 meters for mining areas;

[0139] Furthermore, the division of each sub-working condition zone can be completed by pre-setting working condition classification rules, such as logical partitioning according to work type, road surface type, or work speed; or it can be dynamically determined by real-time identification of scene features through sensor data during driving. Both are conventional methods that can be understood and implemented by those skilled in the art. This application is not limited to specific partitioning methods, as long as it can ensure that the generation process of the rasterized acquisition path adopts differentiated raster parameters in different sub-working condition zones.

[0140] Furthermore, the grid parameters are calculated using a sensor performance model. This involves using the data update frequency, detection accuracy, and calibrated mapping scale of the multi-sensor group, combined with target sampling accuracy (e.g., 1 cm for port alignment and 3 cm for mine slopes), to estimate the spacing between sampling points within each sub-region. Simultaneously, boundary redundancy is determined through analysis of the edge features of the operating range or statistical analysis of historical driving data. This ensures the data acquisition range is slightly larger than the actual operating area, thus offsetting minor offsets or sensor detection errors during driving and preventing feature breaks during data fusion.

[0141] C3: Based on the determined main acquisition direction and combined with the set grid parameters of each sub-region, multiple parallel main acquisition lines are generated in each sub-region to form a set of main acquisition lines;

[0142] Furthermore, the specific steps of C3 include:

[0143] (1) Obtain the main acquisition direction and the grid parameters set for each sub-region;

[0144] (2) For each sub-region, define its spatial boundary. The spatial boundary is usually defined by a set of coordinate points, forming a polygon.

[0145] (3) Calculate an edge of the sub-region perpendicular to the main acquisition direction, and then generate a straight line parallel to the main acquisition direction inside this edge as the first main acquisition line. The length of the first main acquisition line is usually slightly shorter than the length of the sub-region in the main acquisition direction to ensure that it falls completely inside the sub-region and does not overlap with or exceed the boundary.

[0146] (4) After generating the first main acquisition line, the position of the remaining acquisition lines is calculated according to the set grid spacing. The grid spacing refers to the vertical distance between two adjacent main acquisition lines. The system takes the first main acquisition line as the reference and moves it one grid spacing distance along the direction perpendicular to the main acquisition direction to generate the second, third and so on until the last main acquisition line. Each time a main acquisition line is generated, the system will perform a boundary check to ensure that the newly generated line is still completely inside the sub-region.

[0147] (5) When the last main acquisition line is generated, the main acquisition line set in a sub-region is formed. The system will automatically switch to the next sub-region. When the main acquisition lines of all sub-regions are generated, the system will summarize the main acquisition lines of all sub-regions, remove any possible duplicate or invalid lines, sort them according to the sub-region number, and finally package them into a main acquisition line set.

[0148] C4: Within the acquisition boundary, insert cross-check lines perpendicular to the main acquisition direction at preset intervals, and set the intersection of the cross-check lines with each main acquisition line in the main acquisition line set as the data alignment node;

[0149] Specifically, cross-validation lines are used to interrupt the propagation of data errors across regions and provide rigid anchor points between regions during the data fusion phase; by using the intersection point as a data alignment node, data collection can be restarted from the nearest node when a driving interruption or sudden change in operating conditions occurs, without compromising the consistency of the dataset.

[0150] Furthermore, the specific steps of C4 include:

[0151] (1) Calculate the direction of the cross-check line based on the direction vector of the main acquisition: Since the cross-check line needs to be perpendicular to the main acquisition direction, the direction of the cross-check line is set to vertically upward by vector perpendicular operation. For example, when the main acquisition direction is horizontal to the right, the cross-check line direction is set to vertical upward. Determine the standard direction vector of the cross-check line to ensure that all cross-check lines are parallel and strictly perpendicular to the main acquisition line. Then locate the insertion reference starting point of the cross-check line: Take the starting edge boundary of the acquisition boundary along the main acquisition direction as the reference edge. Select the point closest to the boundary vertex and aligned with the starting edge of the main acquisition line as the first insertion reference point. For example, when the main acquisition lines all start from the left side of the boundary, the reference point is selected at the midpoint of the left boundary to ensure that the first cross-check line can cover the starting segment of all main acquisition lines. At the same time, record the coordinates of the reference starting point and the direction information of the boundary where it is located.

[0152] (2) Taking the reference starting point as the origin, calculate the initial insertion position of each cross-check line according to the preset interval along the direction vector of the cross-check line: calculate the reference point coordinates of the second, third and other cross-check lines in sequence according to the interval parameter to form the reference point sequence of the cross-check line. For each reference point, generate a complete initial line segment of the cross-check line: starting from the reference point, extend along the direction vector to the opposite side of the acquisition boundary until both endpoints of the line segment exceed the outer edge of the acquisition boundary; then perform the boundary trimming of the cross-check line: calculate the two intersection points of the initial line segment and the acquisition boundary through the intersection operation of the line segment and the polygon, and use these two intersection points as the new starting point and ending point to trim the cross-check line completely inside the acquisition boundary, ensuring that the cross-check line neither exceeds the boundary nor breaks. After each cross-check line is trimmed, immediately mark its sequence number, starting / ending coordinates and corresponding interval position to form a candidate set of cross-check lines.

[0153] (3) Perform pairing operations on the candidate set of cross-validation lines and the set of main acquisition lines one by one: For each cross-validation line, extract all main acquisition lines in the set of main acquisition lines that are located in the same sub-region and perpendicular to the direction of the cross-validation line. Solve the theoretical intersection coordinates of the two lines by solving the equations of the lines. After the calculation is completed, perform validity screening on each theoretical intersection point: determine whether the intersection point falls within the actual line segment range of the cross-validation line, that is, between the start and end points of the cross-validation line after trimming, and within the actual line segment range of the main acquisition line, such as between the start and end points of the main acquisition line. The equations of the lines are the prior art in this field and are not the inventive solution of this application. They will not be described in detail here.

[0154] (4) Standardize the selected valid intersections: Assign a unique node identifier to each intersection, which includes the cross-check line number, the main acquisition line number and the sub-region identifier to ensure that the node can be traced to the corresponding line and region. At the same time, collect the association information of each node, including the node's precise coordinates, the interval position of the cross-check line to which it belongs, the driving direction of the main acquisition line to which it belongs, etc., to form a three-dimensional data structure of node identifier-coordinate-line association. Sort all data alignment nodes according to the cross-check line number and the main acquisition line number to generate a data alignment node set, store it synchronously in the path data buffer area, and send an alignment node ready signal to the data acquisition unit.

[0155] C5: For each main acquisition line in the set of main acquisition lines, determine its start and end points in conjunction with the data alignment nodes. At the same time, arrange all main acquisition lines in sequence so that the end point of an adjacent main acquisition line is adjacent to or coincides with the start point of the next main acquisition line, forming an initial acquisition grid.

[0156] Specifically, sequential arrangement can reduce non-collection mileage, such as port detours and mine turnarounds, and keep the connection positions of adjacent collection lines fixed, which facilitates the formation of stable start and end references near the cross-check line; the end point being adjacent to the next start point can shorten the travel path and improve collection efficiency.

[0157] C6: Based on the deviation between the actual road conditions and the typical driving trajectory, the initial data acquisition grid is translated and fine-tuned, and the origin of the initial data acquisition grid is anchored at the starting mark point of the working condition.

[0158] Specifically, while the initial dataset already satisfies overall feature coverage, slight feature distortion still occurs locally due to driving errors and sensor noise. During model training, the dataset is not modified globally; instead, minor adjustments and weights are made to the data features on a grid-by-grid basis. The scope of these adjustments is limited by sensor accuracy and driving stability tolerance, avoiding over-correction that could disrupt real-world characteristics. Constraints are derived from the set of anchor points, with the goal of optimizing driving stability metrics globally, ensuring braking distance meets safety standards, providing steering response that meets handling requirements, and achieving economical energy consumption.

[0159] C7: Output rasterized data acquisition path; the rasterized data acquisition path includes the main acquisition line sequence, the start and end coordinates of each main acquisition line, and the coordinates of all data alignment nodes.

[0160] The data mapping logic includes:

[0161] D1: Control the chassis to travel along the gridded data acquisition path, and collect chassis attitude parameters at each acquisition point of the gridded data acquisition path through the vehicle-mounted inertial measurement unit; the chassis attitude parameters include the chassis's spatial position, yaw rate, roll angle and pitch angle;

[0162] D2: Retrieve sensor calibration information from the system's preset database;

[0163] D3: For each acquisition point, the chassis attitude parameters of that acquisition point are combined with the corresponding sensor extrinsic parameters to obtain the mapping matrix of the sensor data in the working plane coordinate system at that acquisition point.

[0164] Furthermore, for the chassis attitude parameters, a transformation matrix is ​​constructed from the chassis coordinate system to the working condition plane coordinate system, that is, the chassis-to-working condition transformation matrix. It is a 4×4 homogeneous transformation matrix that integrates translation and rotation. One part of the matrix is ​​used to represent translating the origin of the chassis coordinate system to a specified position in the working condition plane coordinate system, and the other part is a 3×3 rotation matrix used to represent rotating the chassis coordinate system to the direction aligned with the working condition plane coordinate system according to the roll, pitch and yaw angles.

[0165] Furthermore, for the sensor extrinsic parameters, a transformation matrix is ​​constructed from the sensor coordinate system to the chassis coordinate system, i.e., the sensor-to-chassis transformation matrix. Similarly, it is also a 4×4 homogeneous transformation matrix. Its translation part comes from the sensor's installation position offset relative to the chassis, and the rotation part comes from the sensor's installation attitude angle relative to the chassis. The corresponding rotation matrix is ​​calculated through trigonometric functions. The trigonometric functions are existing technology in this field and are not the inventive solution of this application, so they will not be described in detail here.

[0166] Furthermore, the combination calculation of the chassis attitude parameters at the acquisition point with the corresponding sensor extrinsic parameters is achieved by multiplying the transformation matrix from the sensor to the chassis with the transformation matrix representing the chassis to the working condition.

[0167] D4: Acquire the raw driving data collected by each sensor at the corresponding acquisition point, and combine the sensor intrinsic parameters and the corresponding acquisition point mapping matrix to convert the raw driving data into standardized data in the working condition plane coordinate system.

[0168] Furthermore, the specific conversion process includes: acquiring raw driving data, performing preliminary coordinate transformation in conjunction with sensor intrinsic parameters, so that all raw driving data are converted into physical quantity data in the sensor's own coordinate system, and obtaining preliminary converted sensor data; performing matrix operations on the preliminary converted sensor data and the mapping matrix of the corresponding acquisition points, so that all preliminary converted sensor data are unified to the working condition plane coordinate system. Among these, coordinate transformation and matrix operations are existing technologies in the field and are not inventive solutions of this application, and will not be described in detail here.

[0169] D5: Based on the raster index of the raster data acquisition path, i.e. the raster position information of each acquisition point, the standardized data of each acquisition point is associated with the corresponding raster position. At the same time, the standardized data of multiple sensors within the same raster are fused to finally obtain the initial dataset of raster association.

[0170] Those skilled in the art will understand that the acquisition of intrinsic and extrinsic parameters is achieved through sensor calibration experiments specific to port / mining scenarios, such as radar calibration in dusty environments and camera calibration in humid environments. The acquisition of chassis attitude parameters can be obtained by relying on conventional sensor groups and attitude calculation algorithms of drive-by-wire chassis. Therefore, this mapping and conversion method does not depend on specific hardware. It only needs to provide the minimum attitude information and calibration data to complete the coordinate unification. This application does not impose any further limitations here.

[0171] The model training logic includes:

[0172] E1: Establish feature anchor point constraint relationships based on the working condition type. The feature anchor point constraint relationships include vehicle state stability constraints, steering response timeliness constraints, heavy-load braking distance rationality constraints, energy consumption economy constraints, and driving smoothness constraints.

[0173] In this invention, the vehicle stability constraint is that the vehicle speed fluctuation range is within ±2 kilometers per hour; the steering response time constraint is that the response time from the steering command to the actual steering angle is less than or equal to 0.3 seconds; the heavy-load braking distance rationality constraint is that the braking distance under heavy-load conditions in mining is less than or equal to a preset standard value, such as less than or equal to 15 meters at a vehicle speed of 30 kilometers per hour; the energy consumption economy constraint is that the energy consumption per unit mileage is less than or equal to the energy consumption threshold; and the driving smoothness constraint is that the vertical acceleration of the vehicle body is less than or equal to 0.5 times the gravitational acceleration.

[0174] E2: Obtain the raster index and corresponding data acquisition position of each sub-region in the rasterized data acquisition path, call the initial dataset, and for each sub-region, extract the working condition feature anchor points in the driving data and compare them with the feature anchor point constraint relationship, and calculate the working condition feature anchor point deviation of each constraint item.

[0175] E3: While keeping the grid index unchanged, with the goal of satisfying the feature anchor point constraint relationship of the sub-region driving data, and combined with the deviation of the working condition feature anchor point, the training parameters of the self-learning model are calculated by the error backpropagation algorithm. The backpropagation algorithm is the prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0176] E4: Using the output training parameters as the initial configuration, the initial dataset is called as the training sample to train the preset self-learning model and generate an optimized parameter model. The training process of the self-learning model is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0177] The parameter verification module includes: a data parsing unit, a validity detection unit, and a judgment output unit;

[0178] The data parsing unit receives the optimized parameter model and extracts the initial parameters and corresponding driving performance indicators; the initial parameters include steering gain, braking response time, and drive torque distribution ratio.

[0179] The validity detection unit uses driving performance indicators as the detection object and performs multi-dimensional validity judgment in combination with the system's preset judgment thresholds. The validity judgment includes at least the detection of driving stability, braking safety, steering control, energy consumption economy and parameter adaptability.

[0180] Specifically, the validity determination focuses on the parameters output by the optimized parameter model, and uses a combined performance and adaptation index to evaluate parameter quality, avoiding judgment bias caused by single operating conditions or inconsistent standards. The evaluation content covers key aspects such as driving stability, braking safety, steering handling, energy economy, and parameter adaptation. The judgment criteria are derived from operating condition safety standards or operational optimization goals, and can be adaptively fine-tuned on a batch-by-batch basis, which will not be elaborated here.

[0181] In this embodiment, driving stability is evaluated by the yaw rate fluctuation amplitude and roll angle threshold, and braking safety is determined by braking distance and braking response time; steering handling is measured by steering response delay and steering accuracy, and the consistency of steering return torque is checked; energy economy is calculated by statistically analyzing energy consumption per unit mileage in different driving speed ranges, and high energy consumption caused by poor parameter matching is detected; parameter adaptability is evaluated by measuring the fluctuation range of performance indicators under multiple operating conditions to assess the versatility of parameters.

[0182] The judgment output unit receives the detection result. If the detection is qualified, a qualified signal is generated, and the optimal parameter combination in the current initial parameters is fixed to the chassis control library. A unique parameter version number is generated and synchronized to the background archive. If the detection is unqualified, an unqualified signal is generated, and parameter deviation information and corresponding working condition adaptation problems are recorded and then fed back to the manual adjustment stage.

[0183] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. A system for manual adjustment and self-learning of parameters for intelligent heavy-duty truck drive-by-wire chassis, characterized in that: include: Command acquisition module, chassis control module, model training module, parameter verification module; The instruction acquisition module acquires manual adjustment instructions and operating condition formulas, and controls the main controller to load initial parameters and driving strategies that match the operating condition formulas according to the manual adjustment instructions; the initial parameters include steering gain, braking response time, and drive torque distribution ratio; Under the control of the main controller, the chassis control module completes the initial parameter distribution and executes chassis driving control to obtain a drive-by-wire chassis that operates according to the target parameters. At the same time, it collects driving data through a multi-sensor group. The driving data includes vehicle status, steering angle, yaw rate, braking distance, and energy consumption. The model training module receives driving data, combines chassis attitude parameters and sensor calibration information, maps the driving data to the working condition plane coordinate system, fuses the data according to the gridded data acquisition path and trains the model to obtain an optimized parameter model. The parameter verification module determines the validity of the parameters output by the optimized parameter model. If the parameters are valid, they are saved to the chassis control library and a parameter version number is generated and synchronized to the background. If the parameters are invalid, the parameter deviation information is recorded and returned to the manual adjustment stage. The model training module includes: a data acquisition path generation unit, a data acquisition unit, an initial data fusion unit, and a parameter optimization unit; The data acquisition path generation unit obtains a preset working condition type and generates a rasterized data acquisition path that matches the working condition type based on the boundary coordinates of the driving range and the working condition accuracy requirements. The data acquisition unit receives the gridded data acquisition path and controls the chassis to drive along the gridded data acquisition path. The chassis travels at equal distances along the gridded data acquisition path as the trigger condition, and multi-dimensional driving data is collected through a multi-sensor group. The initial data fusion unit receives driving data, combines it with chassis attitude parameters and preset sensor calibration information, maps the driving data to a unified working plane coordinate system, and then integrates the mapped driving data according to the grid position to obtain the initial dataset; the sensor calibration information includes camera intrinsic parameters and radar installation angle calibration data. The parameter optimization unit extracts working condition feature anchors from the initial dataset using a feature extraction algorithm. Using these anchors as constraints, it trains and fine-tunes the parameters of a preset self-learning model to generate an optimized parameter model. The working condition feature anchors include at least one of the following: vehicle state threshold, peak steering angle, road surface adhesion coefficient range, and standard braking distance, determined by statistical data features. The self-learning model is trained using a BP neural network.

2. The intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system as described in claim 1, characterized in that, The chassis control module includes: a parameter loading unit, a driving control unit, a working condition switching unit, and a parameter temporary storage unit; The parameter loading unit controls the main controller to read the manually input adjustment parameters and working condition formulas, parse them to obtain the target parameter values ​​of the steering, braking and drive systems, and load them into the actuator control channel; The driving control unit controls the chassis to travel along a preset route according to the road condition type and vehicle status of the working condition formula. During the driving process, the current target parameter values ​​are sent to the execution terminal in real time. The road condition type includes uphill and downhill, unpaved, rain, snow and dust. The vehicle status includes load and attitude. The execution terminal includes a steering system, a braking system and a drive system. The working condition switching unit continuously monitors the real-time road condition data fed back by the vehicle sensors. When it detects that the real-time road condition data has changed from the preset road condition type in the working condition formula, the working condition switching unit dynamically adjusts the execution weight of each target parameter value, generates a dynamic parameter sequence, and provides the adjusted target parameter values ​​to the driving control unit in real time. After a single driving cycle ends, the parameter storage unit associates the dynamic parameter sequence of this driving cycle, the corresponding real-time driving data, and the working condition recipe identifier, and packages and stores the associated data in the local database to obtain a drive-by-wire chassis that operates according to the target parameters.

3. The intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system as described in claim 1, characterized in that, The system also includes a parameter correction module, which is configured with correction logic. This correction logic is used to correct parameter deviations in real time during parameter execution, including: After the main controller sends the target parameter values ​​to the actuator, it collects the feedback data of the actuator in real time, and at the same time, it collects the attitude sensor data of the vehicle; the feedback data includes at least the actual torque of the steering motor and the actual pressure of the brake master cylinder; the attitude sensor data includes at least the yaw rate and the roll angle. The collected feedback data is compared with the corresponding target parameter value to calculate the deviation value. Based on the magnitude and trend of the deviation value, the adjustment direction and adjustment amount are determined. Then, based on the adjustment direction and adjustment amount, the control command output by the main controller to the execution end is corrected, and the process is continuously iterated until the deviation value is reduced and stabilized within the preset deviation threshold range. After the deviation value stabilizes, the system maintains the current parameter output gain. At the same time, the parameter correction module combines the collected attitude sensor data and works with the main controller to switch to dynamic correction mode. In dynamic correction mode, the attitude sensor data and execution deviation are comprehensively analyzed by the PID control algorithm to calculate the compensation coefficient, and the compensation coefficient is used to perform real-time dynamic compensation on the parameters output by the main controller.

4. The intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system as described in claim 3, characterized in that, The calculation process of the compensation coefficient includes: The main controller, while keeping the basic parameter output unchanged, collects attitude sensor data and actuator response delay in real time, adjusts the parameter execution sensitivity according to the attitude sensor data, and compares the actuator response delay with a preset delay threshold to determine whether the current parameter setting is in an appropriate state. If the system is determined to be in a suitable state, the main controller stops coarse sensitivity adjustment and switches to fine adjustment mode based on road condition characteristics. During the switching process, the continuity of parameter output is maintained through a transition band smoothing algorithm. The road condition characteristics are collected in real time by on-board sensors, including road surface adhesion coefficient, slope, or from pre-stored working condition formulas. In fine-tuning mode, the system collects multi-dimensional feedback signals and compares the feedback signals with preset operating condition adaptation requirements. When the feedback signals meet the adaptation requirements, the system determines that the parameters match the current road conditions. The feedback signals include driving stability indicators and energy consumption data. Based on the matching results, combined with real-time attitude sensor data and execution deviation, the main controller performs a comprehensive analysis using a PID control algorithm to calculate compensation coefficients for core parameters, including steering gain and braking response time.

5. The intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system as described in claim 4, characterized in that, The model training module is equipped with a self-learning strategy. The self-learning strategy trains and optimizes the parameter model based on the driving data collected by the multi-sensor group and outputs the optimal parameter combination. The self-learning strategy includes path planning logic, data mapping logic and model training logic. The path planning logic is configured in the path generation unit, the data mapping logic is configured in the initial data fusion unit, and the model training logic is configured in the parameter optimization unit.

6. The intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system as described in claim 5, characterized in that, The path planning logic includes: The main acquisition direction is determined by the typical driving trajectory and vehicle state range of the working condition type, and the vertical direction of the main acquisition direction is determined by the road condition feature points and the effective detection range of the sensor. Based on the main acquisition direction and the vertical direction, a working condition plane coordinate system matching the working condition is established and the corresponding acquisition boundary is determined; the road condition feature points include curves, slopes and intersections. The established data acquisition boundary is divided into multiple sub-regions according to the working condition type, and grid parameters are set for each sub-region. The types of sub-regions include paved roads in port areas, mine ramps, mine gravel roads, and mining areas. The grid parameters include the distance between acquisition points, the data overlap between adjacent acquisition points, and the boundary redundancy. The distance between acquisition points is determined according to the sensor data update frequency and the preset sampling accuracy. Based on the determined main acquisition direction and combined with the set grid parameters for each sub-region, multiple parallel main acquisition lines are generated in each sub-region to form a set of main acquisition lines. Within the acquisition boundary, cross-check lines perpendicular to the main acquisition direction are inserted at preset intervals, and the intersection of the cross-check lines with each main acquisition line in the main acquisition line set is set as the data alignment node. For each main acquisition line in the set of main acquisition lines, its start and end points are determined in conjunction with the data alignment nodes. At the same time, all main acquisition lines are arranged in sequence so that the end point of an adjacent main acquisition line is adjacent to or coincides with the start point of the next main acquisition line, forming an initial acquisition grid. Based on the deviation between the actual road conditions and the typical driving trajectory, the initial data acquisition grid is translated and fine-tuned, and the origin of the initial data acquisition grid is anchored at the starting point of the working condition. Output a rasterized data acquisition path; the rasterized data acquisition path includes the main acquisition line sequence, the start and end coordinates of each main acquisition line, and the coordinates of all data alignment nodes.

7. The intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system as described in claim 6, characterized in that, The data mapping logic includes: The chassis is controlled to travel along a gridded data acquisition path. At each acquisition point along the path, the chassis attitude parameters are collected by the onboard inertial measurement unit. These parameters include the chassis's spatial position, yaw rate, roll angle, and pitch angle. Sensor calibration information is retrieved from a pre-set database. For each acquisition point, the chassis attitude parameters are combined with the corresponding sensor extrinsic parameters to obtain a mapping matrix of the sensor data in the working plane coordinate system. The raw driving data collected by each sensor at the corresponding acquisition point is acquired. Combining the sensor intrinsic parameters and the mapping matrix, the raw driving data is converted into standardized data in the working plane coordinate system. Based on the grid index of the gridded data acquisition path, i.e., the grid position information of each acquisition point, the standardized data of each acquisition point is associated with the corresponding grid position. Simultaneously, the standardized data from multiple sensors within the same grid are fused to obtain the initial dataset for grid association.

8. The intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system as described in claim 7, characterized in that, The model training logic includes: Based on the operating condition type, feature anchor point constraint relationships are established, including vehicle state stability constraints, steering response timeliness constraints, heavy-load braking distance rationality constraints, energy consumption economy constraints, and driving comfort constraints. The raster index and corresponding data acquisition location of each sub-region in the rasterized data acquisition path are obtained. The initial dataset is called, and for each sub-region, the operating condition feature anchor points in the driving data are extracted and compared with the feature anchor point constraint relationships to calculate the operating condition feature anchor point deviation for each constraint item. While keeping the raster index unchanged, with the sub-region driving data satisfying the feature anchor point constraint relationships as the objective, and combined with the operating condition feature anchor point deviation, the training parameters of the self-learning model are calculated using the error backpropagation algorithm. Using the output training parameters as the initial configuration, the initial dataset is called as the training sample to train the preset self-learning model, generating an optimized parameter model.

9. The intelligent heavy-duty truck drive-by-wire chassis parameter manual adjustment and self-learning system as described in claim 1, characterized in that, The parameter verification module includes: a data parsing unit, a validity detection unit, and a judgment output unit; The data parsing unit receives the optimized parameter model and extracts the initial parameters and corresponding driving performance indicators; The validity detection unit uses driving performance indicators as the detection object and performs multi-dimensional validity judgment in combination with the system's preset judgment thresholds. The validity judgment includes at least the detection of driving stability, braking safety, steering control, energy consumption economy and parameter adaptability. The judgment output unit receives the detection result. If the detection is qualified, a qualified signal is generated, and the optimal parameter combination in the current initial parameters is fixed to the chassis control library. A unique parameter version number is generated and synchronized to the background archive. If the detection is unqualified, an unqualified signal is generated, and parameter deviation information and corresponding working condition adaptation problems are recorded and then fed back to the manual adjustment stage.

Images