Intelligent control method and system for bidirectional running wheeled sightseeing train

Abstract

The application relates to the field of intelligent traffic control, and discloses an intelligent control method and system for a bidirectional driving wheeled sightseeing train, aiming to solve the problems of low efficiency, poor safety and weak adaptability caused by the dependence of traditional sightseeing trains on fixed tracks, path solidification and control lag. The method fuses laser radar, vision, millimeter wave radar and inertial data by deploying a front-rear symmetrical multi-modal sensor array to construct a dynamic environment map; an improved hybrid A-star algorithm is used to plan a global collision-free trajectory, and model predictive control is combined to realize high-precision tracking, thereby realizing bidirectional autonomous operation under trackless conditions, multi-vehicle intelligent cooperation and all-scene adaptive control, and significantly improving the traffic efficiency, safety and intelligent level of the scenic area.

Description

Technical Field

[0001] This invention belongs to the field of intelligent traffic control, specifically relating to an intelligent control method and system for a bidirectional wheeled sightseeing train. Background Technology

[0002] With the rapid development of urban cultural tourism and theme park economies, wheeled sightseeing trains, as mobile carriers integrating transportation, guided tours, and immersive experiences, have been widely used in scenic spots, commercial districts, and large parks. These vehicles typically operate on tracks or semi-constrained paths, their core function being to safely and efficiently complete multi-station loop connections while ensuring tourist comfort and operational flexibility. Traditional sightseeing trains generally employ a one-way design, relying on fixed routes and manual scheduling. This presents significant limitations in handling sudden passenger flows, temporary detours, or bidirectional demand, especially during peak hours, easily leading to station congestion, low turnaround efficiency, and energy waste.

[0003] To address the aforementioned issues, existing technologies have proposed an improved form: the bidirectional wheeled sightseeing train. Through redesigning the vehicle structure and drive system, it enables direction switching without requiring a U-turn. However, bidirectional control often relies on driver experience or simple sensor thresholds to trigger steering commands, failing to establish an intelligent collaborative mechanism between vehicles, roads, and the environment. This leads to misjudgments of direction-changing timing, frequent path conflicts, and uneven energy consumption distribution. Furthermore, the lack of a dynamic perception module for passenger boarding and alighting status, station queuing density, and weather conditions makes the control strategy difficult to adapt to the changing needs of real-world operating scenarios, especially at night, in rain or fog, or in environments with high pedestrian traffic.

[0004] Therefore, it is urgent to build an intelligent control system that integrates environmental perception, intention reasoning and adaptive decision-making to achieve stable, efficient and humanized operation of the two-way wheeled sightseeing train in all scenarios. Summary of the Invention

[0005] This invention provides an intelligent control method and system for a bidirectional wheeled sightseeing train. By constructing a global perception fusion architecture and a multi-body collaborative decision-making mechanism, it achieves high-precision trajectory tracking, dynamic obstacle avoidance, seamless bidirectional switching, and multi-car collaborative control of the bidirectional wheeled sightseeing train in complex scenic road environments. This system breaks through the traditional unidirectional track dependence and manual driving mode. Under conditions without fixed physical track constraints, it establishes an autonomous navigation system based on the coupling of environmental semantic understanding and vehicle dynamics, solving the fundamental technical contradictions of existing sightseeing trains, such as low operating efficiency, insufficient safety redundancy, and poor scene adaptability caused by structural rigidity, fixed paths, and control lag.

[0006] Firstly, an intelligent control method for a bidirectional wheeled sightseeing train includes:

[0007] The lidar array, millimeter-wave radar group, panoramic vision sensor group and inertial measurement unit deployed at the front and rear of the vehicle body simultaneously collect three-dimensional point cloud data in front and behind, relative speed and distance information of obstacles, road texture and marking semantic features, and the six degrees of freedom motion state of the vehicle body.

[0008] The above multi-source heterogeneous sensor data is input into the vehicle edge computing platform. The spatiotemporal alignment module completes the unification of sampling time of each sensor and coordinate system transformation, generating a global environmental grid map with the center of the vehicle body as the origin and a dynamic obstacle trajectory prediction sequence.

[0009] Based on the preset electronic fence boundary of the scenic area and the real-time updated global map, the improved hybrid A-star path planning algorithm is called to generate the globally optimal collision-free trajectory from the current position to the target station, taking into account the minimum turning radius of the vehicle body, the maximum side tilt angle limit, and the constraints of the two-way driving switching area.

[0010] Meanwhile, a model predictive control framework is introduced, with the trajectory curvature change rate, lateral offset, and longitudinal acceleration within the look-ahead window as optimization targets. Combined with the current road surface friction coefficient estimate and load distribution status, the independent torque distribution command for each drive wheel and the steering wheel angle correction are calculated in real time.

[0011] During the bidirectional driving mode switching phase, the redundant braking coordination module is activated, the front and rear drive shaft clutches are locked synchronously, the hydraulic accumulator pressure of the steering mechanism is released, and after the vehicle body comes to a complete stop, the master-slave communication relationship of the main control unit is switched, the zero point of the inertial navigation system is reset, and the coordinates of the starting point and ending point of the path planning are reinitialized.

[0012] For multi-car train operation scenarios, a queue maintenance controller based on a distributed consensus protocol is established. The position and attitude of the rear of the leading car serve as a virtual navigator, and the relative position and attitude deviation of the front of the following car is calculated in real time. By adjusting the speed difference of its own driving wheels and the active articulation angle, the preset formation spacing and heading synchronization are maintained.

[0013] Secondly, an intelligent control system for a bidirectional wheeled sightseeing train includes an environmental perception fusion module, a global path planning module, a local trajectory tracking module, a bidirectional mode switching control module, a multi-carriage collaborative formation module, an actuator drive module, and a central decision-making and arbitration module.

[0014] The environmental perception fusion module is located at the front and rear ends of the vehicle body and includes three sets of sixteen-line lidar, four sets of 77GHz millimeter-wave radar, six sets of two-megapixel wide-angle cameras and one set of six-axis microelectromechanical inertial measurement unit. All sensor output signals are transmitted to the on-board industrial control computer via gigabit Ethernet bus.

[0015] The global path planning module runs on an embedded processor with a real-time operating system. It has a built-in high-precision map database of scenic spots and a dynamic obstacle occupancy grid update engine. It adopts a hierarchical state space search strategy, superimposes real-time traffic flow density weights on the static road network topology, and outputs a global reference trajectory that meets the two-way traffic rules.

[0016] The local trajectory tracking module uses a nonlinear model predictive controller. Its prediction time domain is set to five seconds, the control cycle is twenty milliseconds, and the state variables include lateral position error, heading angle deviation, yaw rate, and longitudinal vehicle speed. The control input is the difference between the front wheel steering angle and the torque of the left and right drive wheels. The constraints cover the tire adhesion ellipse limit and the suspension travel boundary.

[0017] The bidirectional mode switching control module is equipped with dual redundant brake air circuits and an electro-hydraulic steering valve group. After receiving the mode switching command, it first triggers the electronic parking brake of all wheels. After the vehicle speed returns to zero, it disconnects the main drive motor enable signal, releases the mechanical lock pin of the steering system after a delay of 500 milliseconds, and simultaneously restarts the inertial navigation system and loads the reverse driving parameter set.

[0018] The multi-carriage collaborative formation module is deployed on an independent controller in each carriage. It exchanges relative position and speed information through a dedicated wireless local area network and uses a sliding mode variable structure control law to suppress articulation angle oscillation, ensuring that the overlap error of the center of gravity trajectory of each carriage is less than ten centimeters when driving on curves.

[0019] The actuator drive module includes wheel-side permanent magnet synchronous motors and their vector control drivers, electro-hydraulic proportional steering valves, electromagnetic brake calipers, and air suspension height adjustment valves. It receives standardized control commands from the central decision-making and arbitration module to complete torque output, steering angle execution, braking force application, and vehicle attitude adjustment.

[0020] The central decision-making and arbitration module, as the core scheduling unit of the system, is responsible for coordinating the operating priorities of various functional modules, handling abnormal conditions such as sensor failure, communication interruption, and path blockage, and ensuring that the system still has basic operating capabilities in degraded mode according to the preset safety state machine switching control strategy.

[0021] As one embodiment of the present invention, in the environmental perception fusion module:

[0022] After voxel filtering and ground segmentation, the lidar point cloud data is input into a 3D target detection network based on an attention mechanism, which outputs obstacle category, size, position, and motion vector.

[0023] Visual sensor images are processed by a semantic segmentation neural network to identify lane line types, traffic signs, pedestrian areas, and drivable road surface boundaries.

[0024] The raw echo signal from the millimeter-wave radar is subjected to constant false alarm rate detection and cluster analysis to generate a list of moving obstacles;

[0025] The inertial measurement unit outputs wheel speed pulse signals that are fused by a Kalman filter to calculate the vehicle's absolute position and attitude angles.

[0026] The above four types of data are fused at the feature level under a unified time benchmark to construct a dynamic environmental situation map.

[0027] As one embodiment of the present invention, in the global path planning module:

[0028] The electronic fence data of the scenic area is stored in the format of Geographic Information System (GIS) and includes restricted areas, one-way streets, slope restriction areas, and minimum turning radius constraint areas.

[0029] The path planning algorithm adopts a hierarchical state space search strategy, with the upper layer being a road network topology search and the lower layer being a local grid map search.

[0030] Introducing a bidirectional driving penalty factor into the traditional A-star heuristic function, additional cost weights are applied to path nodes that need to switch driving directions;

[0031] The generated global trajectory is smoothed using cubic B-spline curves to ensure that the curvature continuity meets the vehicle dynamics constraints.

[0032] As one embodiment of the present invention, in the local trajectory tracking module:

[0033] The objective function of the model predictive controller consists of three terms: the square integral of the lateral tracking error, the penalty term for the rate of change of control input, and the energy consumption minimization term.

[0034] In addition to the vehicle dynamics boundary, the constraints include a comfort index constraint, namely, the absolute value of longitudinal acceleration does not exceed 0.5 m / s².

[0035] The solver uses a sequential quadratic programming algorithm to complete rolling optimization calculations within twenty milliseconds.

[0036] In one embodiment of the present invention, the bidirectional mode switching control module includes:

[0037] The braking system adopts a dual-circuit air pressure braking architecture. The front axle and the rear axle are pressurized by independent air tanks. During the switching process, the spring energy storage brake is activated first. After confirming that the braking torque of the four wheels reaches 80% of the rated value, the power output of the drive system is cut off.

[0038] The steering system is equipped with a double-acting hydraulic cylinder. When switching, the direction of the oil circuit is switched through a three-position four-way solenoid valve, so that the steering wheel rotates 180 degrees from the neutral position to the reverse driving preparation angle.

[0039] The inertial navigation system reset process consists of two stages: gyroscope zero-bias calibration and accelerometer gravity vector alignment, with a total time of no more than three seconds.

[0040] As one embodiment of the present invention, in the multi-carriage cooperative platooning module:

[0041] A point-to-point communication link is established between adjacent carriages via a directional antenna. The transmission period is fifty milliseconds. The data packet contains the vehicle's position, speed, heading angle, articulation angle, and fault status code.

[0042] The queue controller adopts a navigator-follower architecture. The following vehicle calculates the desired relative pose based on the motion state broadcast by the navigator vehicle, and eliminates the deviation between the actual relative pose and the desired value by adjusting the differential ratio of its own drive wheels and the extension and retraction of the active articulated hydraulic cylinder.

[0043] The articulation angle control introduces a feedforward compensation term, which predicts the required articulation angle based on the current vehicle speed and radius of curvature, thereby reducing tracking lag.

[0044] In one embodiment of the present invention, the actuator driving module includes:

[0045] The wheel-side motor is a water-cooled permanent magnet synchronous motor with a rated power of 15 kilowatts and a peak torque of 300 Nm. It is equipped with a rotary transformer and a temperature sensor.

[0046] The steering mechanism is a rack and pinion electric power steering system with a maximum steering angle of ±40 degrees and a response bandwidth of 10 Hz.

[0047] The braking system uses disc brakes with an electronic control unit, which can realize independent braking pressure adjustment for each wheel, with a minimum pressure build-up time of two hundred milliseconds;

[0048] The air suspension system uses feedback from the vehicle height sensor to adjust the inflation and deflation of the airbags via a proportional solenoid valve to maintain a constant ground clearance.

[0049] As one embodiment of the present invention, in the central decision-making arbitration module:

[0050] Built-in multi-fault diagnosis logic: when any LiDAR data is lost for more than one second, the weight of millimeter-wave radar and vision fusion is automatically increased; when wireless communication interruption lasts for more than three seconds, local path replanning is initiated and the speed is reduced to walking speed.

[0051] When an unavoidable obstacle is detected ahead of the path, the emergency brakes should be activated immediately and the incident should be reported to the scenic area dispatch center.

[0052] The system operation log is stored on the solid-state drive at minute-by-minute granularity, containing all raw sensor data, control command sequences, and state machine transition records.

[0053] As one embodiment of the present invention, during peak hours in the scenic area, the central decision-making and arbitration module receives a heat map of passenger flow density from the ticketing system and dynamically adjusts the departure interval and stop order of each sightseeing train to avoid congestion of the train convoy in narrow sections.

[0054] In rainy or snowy weather conditions, the maximum driving speed limit is automatically reduced, the following distance is increased, and the tire anti-skid control strategy is activated in advance.

[0055] In nighttime operation mode, the intensity of near-infrared illumination is enhanced, the visual recognition threshold is reduced, and the safety margin of braking distance is extended.

[0056] In summary, this application includes at least one of the following beneficial technical effects:

[0057] 1. This invention constructs a comprehensive environmental perception capability by using a symmetrically deployed multimodal sensor array and spatiotemporal alignment fusion technology. Combined with an improved hybrid A-star algorithm and model predictive control, it can plan and track the globally optimal path in real time and with high precision in complex scenic road environments without fixed physical tracks. It also enables safe, smooth, and rapid switching between two-way driving modes, completely eliminating the dependence on fixed tracks and significantly improving the flexibility of path planning and the maneuverability of vehicle operation.

[0058] 2. By establishing a queue-keeping controller based on a distributed consensus protocol and introducing sliding mode variable structure control and feedforward compensation mechanism, the oscillations during the articulation of multiple carriages can be effectively suppressed, ensuring that each carriage can maintain accurate formation spacing and heading synchronization when traveling on straight lines and curves. This significantly reduces the tracking error of long train formations on complex paths and improves the overall smoothness and safety of operation.

[0059] 3. The central decision-making arbitration module incorporates multiple fault diagnosis logics and degradation operation strategies, which can intelligently respond to various abnormal situations such as sensor failure, communication interruption, and path blockage. By dynamically adjusting the fusion weights and switching control strategies, it ensures that the system still has basic operational capabilities after some functions are degraded, thereby significantly improving the robustness and reliability of the system in all weather and multi-scenario environments. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention.

[0061] Figure 2 This is a schematic diagram of the core principle framework of the present invention. Detailed Implementation

[0062] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following description is provided in conjunction with the appendix. Figure 1 and 2The following is a detailed description of specific embodiments based on the present invention, including preferred embodiments.

[0063] Example 1:

[0064] This invention provides an intelligent control method and system for a bidirectional wheeled sightseeing train. By constructing a full-domain perception fusion architecture and a multi-body collaborative decision-making mechanism, it achieves high-precision trajectory tracking, dynamic obstacle avoidance, seamless bidirectional switching, and multi-car collaborative control of the bidirectional wheeled sightseeing train in complex scenic road environments. It breaks through the traditional unidirectional track dependence and manual driving mode, and establishes an autonomous navigation system based on the coupling of environmental semantic understanding and vehicle dynamics under the condition of no fixed physical track constraints. This solves the fundamental technical contradictions of existing sightseeing trains, such as low operating efficiency, insufficient safety redundancy, and poor scene adaptability caused by structural rigidity, fixed path, and control lag.

[0065] A method for intelligent control of a bidirectional wheeled sightseeing train includes the following steps:

[0066] S1 is a multimodal sensor array deployed at the front and rear of the vehicle body. The multimodal sensor array includes a lidar array, a millimeter-wave radar group, a panoramic vision sensor group, and an inertial measurement unit. It simultaneously collects three-dimensional point cloud data of the front and rear, relative speed and distance information of obstacles, road texture and marking semantic features, and the six-degree-of-freedom motion state of the vehicle body.

[0067] S2, input the above multi-source heterogeneous sensor data into the vehicle edge computing platform, and complete the unification of the sampling time of each sensor and coordinate system transformation through the spatiotemporal alignment module to generate a global environmental grid map with the center of the vehicle body as the origin and a dynamic obstacle trajectory prediction sequence.

[0068] S3, based on the preset scenic area electronic fence boundary and the real-time updated global map, calls the improved hybrid A-star path planning algorithm to generate the globally optimal collision-free trajectory from the current position to the target station, taking into account the minimum turning radius of the vehicle body, the maximum side tilt angle limit, and the constraints of the two-way driving switching area.

[0069] S4 introduces a model predictive control framework, with the trajectory curvature change rate, lateral offset, and longitudinal acceleration within the look-ahead window as optimization targets. Combined with the current road surface friction coefficient estimate and load distribution status, it calculates the independent torque distribution command for each drive wheel and the steering wheel angle correction in real time.

[0070] S5, during the bidirectional driving mode switching phase, activates the redundant braking coordination module, synchronously locks the front and rear drive shaft clutches, releases the hydraulic accumulator pressure of the steering mechanism, and after the vehicle body comes to a complete stop, switches the master-slave communication relationship of the main control unit, resets the zero point of the inertial navigation system, and re-initializes the coordinates of the starting point and ending point of the path planning.

[0071] S6, for multi-car train operation scenarios, establishes a queue maintenance controller based on a distributed consensus protocol. The position and attitude of the rear of the leading car serve as a virtual navigator, and the relative position and attitude deviation of the front of the following car is calculated in real time. By adjusting the speed difference of its own driving wheels and the active articulation angle, the preset formation spacing and heading synchronization are maintained.

[0072] In step S1, the lidar array, millimeter-wave radar group, panoramic vision sensor group and inertial measurement unit deployed at the front and rear of the vehicle body simultaneously collect three-dimensional point cloud data of the front and rear, relative speed and distance information of obstacles, road texture and marking semantic features, and the six-degree-of-freedom motion state of the vehicle body.

[0073] The S101 lidar array consists of three sets of sixteen-line lidars, installed on the left and right sides of the front and rear of the vehicle and in the center, respectively. The scanning frequency is twenty frames per second, the horizontal field of view is 360 degrees, the vertical field of view is 30 degrees, and the ranging accuracy is ±2 centimeters. It is used to generate point cloud data.

[0074] The S102 millimeter-wave radar group consists of four 77GHz millimeter-wave radars, arranged at the four corners of the vehicle body. The detection range is 0.5 meters to 150 meters, the speed resolution is 0.1 meters per second, and the angular resolution is 1 degree. It is used to generate radar echo signals.

[0075] The S103 panoramic vision sensor group includes six 2-megapixel wide-angle cameras that cover the front, rear, left, right, and up / down views of the vehicle, with a frame rate of 30 Hz and a dynamic range of 120 decibels, used to generate image data.

[0076] The S104 is a six-axis microelectromechanical system (MEMS) containing a three-axis gyroscope and a three-axis accelerometer. It has a sampling frequency of 200 Hz and a zero-bias stability better than 0.1 degree per hour. It is used to generate vehicle motion state data.

[0077] S105: All sensors trigger synchronous acquisition upon startup, and the data packets carry hardware timestamps, with time synchronization accuracy better than one millisecond.

[0078] The lidar outputs raw point cloud data, including the three-dimensional coordinates, reflection intensity, and timestamp of each point; the millimeter-wave radar outputs raw radar echo signals, including the distance, radial velocity, azimuth, elevation angle, and signal-to-noise ratio of each target; the vision sensor outputs raw image data, including the pixel matrix and exposure parameters; the inertial measurement unit outputs raw vehicle motion state data, including raw values ​​of angular velocity and acceleration. All raw data are transmitted to the onboard industrial control computer via a gigabit Ethernet bus for subsequent processing.

[0079] In step S2, the above-mentioned multi-source heterogeneous sensor data is input into the vehicle edge computing platform. The spatiotemporal alignment module completes the unification of sampling time of each sensor and coordinate system transformation, generating a global environmental grid map with the center of the vehicle body as the origin and a dynamic obstacle trajectory prediction sequence.

[0080] S201, the spatiotemporal alignment module first interpolates all data to a unified time reference based on the hardware timestamps of each sensor. The interpolation method uses cubic spline interpolation, and the time reference is set to the sampling time of the inertial measurement unit.

[0081] The coordinate system transformation process involves two steps:

[0082] The first step is to transform the data from the local coordinate system of each sensor to the vehicle coordinate system. The transformation matrix is ​​obtained through factory calibration and includes rotation matrix and translation vector.

[0083] The second step is to transform the data in the vehicle coordinate system to the global navigation coordinate system. The transformation parameters are provided by the vehicle pose calculated by the fusion of the inertial measurement unit and the wheel speed sensor.

[0084] S202, the LiDAR point cloud data is voxel filtered in the vehicle coordinate system. The voxel size is 0.1 meters. After filtering out outliers, ground segmentation is performed. A plane fitting method based on normal vectors is used to divide the point cloud into ground points and non-ground points. The non-ground point cloud is input into a 3D target detection network based on an attention mechanism. The 3D target detection network includes a point cloud encoder, an attention feature extractor, and a bounding box regression head. It outputs obstacle category, size, position, and motion vector. The categories include four types: pedestrians, vehicles, fixed obstacles, and movable objects.

[0085] S203: The panoramic vision sensor image stream is input to the semantic segmentation neural network. The network adopts an encoder-decoder structure, with the encoder being a residual network and the decoder being a transposed convolutional layer. It outputs pixel-level semantic labels, and the label categories include six types: drivable areas, lane lines, pedestrian crossings, traffic signs, green belts, and buildings.

[0086] S204: The raw echo signal from the millimeter-wave radar is processed by a constant false alarm rate detector. The detection threshold is dynamically adjusted according to the background noise. The detected target points are grouped by a density clustering algorithm. The centroid position and average velocity of each group of targets are calculated to generate a list of moving obstacles.

[0087] S205, the inertial measurement unit outputs wheel speed pulse signals fused by a Kalman filter, the state vector includes position, velocity and attitude angle, the observation vector is wheel speed and angular velocity, the process noise covariance matrix is ​​adaptively adjusted according to road conditions, and the absolute position and attitude angle of the vehicle body in the global navigation coordinate system are calculated and output.

[0088] The above four types of processed data are fused at the feature level under a unified time reference to construct a dynamic environmental situation map. The fusion process adopts a weighted average method, and the weights are dynamically allocated according to the sensor confidence. The confidence is calculated from the historical detection accuracy and the current environmental conditions. The final output is a global environmental grid map with a grid size of 0.5 meters. Each grid contains three fields: occupancy probability, semantic category, and dynamic attribute. At the same time, a dynamic obstacle trajectory prediction sequence is output with a prediction time domain of 5 seconds and a step size of 0.5 seconds. Each prediction point contains four attributes: position, velocity, acceleration, and category confidence.

[0089] In step S3, based on the preset scenic area electronic fence boundary and the real-time updated global map, the improved hybrid A-star path planning algorithm is called to generate the globally optimal collision-free trajectory from the current position to the target station, taking into account the minimum turning radius of the vehicle body, the maximum side tilt angle limit, and the constraints of the bidirectional driving switching area.

[0090] S301, the scenic area electronic fence data is stored in the vehicle solid-state drive in geographic information system format. It includes polygon boundaries, attribute fields and topological relationships. The attribute fields include four categories: restricted areas, one-way streets, slope restriction areas, and minimum turning radius constraint areas.

[0091] The S302 path planning algorithm runs on an embedded processor with a real-time operating system, a processor clock speed of 2 GHz, and 8 GB of memory.

[0092] The path planning algorithm adopts a hierarchical state-space search strategy, that is, it uses a path search algorithm. Specifically, the upper layer is a road network topology map search, where nodes are intersections and key landmarks, edges are connecting paths, and edge weights are the weighted sum of path length and traffic flow density; the lower layer is a local grid map search, where the state space is the position and heading angle, and the control inputs are the turning angle and forward distance.

[0093] S303, the path search algorithm introduces a bidirectional driving penalty factor into the traditional A-star heuristic function, and applies an additional cost weight to path nodes that need to switch driving directions. The weight value is calculated based on the length of the switching area and the historical switching success rate.

[0094] During the search process, the state expansion must meet vehicle dynamics constraints, including a minimum turning radius of five meters, a maximum roll angle of eight degrees, and a maximum longitudinal acceleration of two meters per second squared. The generated path node sequence is smoothed using cubic B-spline curves. The control points are path nodes, and the node weights are dynamically adjusted according to the rate of curvature change to ensure that the curvature continuity meets the vehicle's lateral acceleration constraints. Finally, a global reference trajectory is output, with a trajectory point interval of 0.5 meters. Each point contains four attributes: global coordinates, desired velocity, desired heading angle, and radius of curvature. The trajectory is stored in a circular buffer with a length of one thousand points, supporting dynamic updates and overwriting.

[0095] In step S4, a model predictive control framework is introduced, with the trajectory curvature change rate, lateral offset, and longitudinal acceleration within the look-ahead window as optimization targets. Combined with the current road surface friction coefficient estimate and load distribution status, the independent torque distribution command for each drive wheel and the steering wheel angle correction are calculated in real time.

[0096] The S401 model predictive controller has a cycle of 20 milliseconds and a prediction time domain of 5 seconds, corresponding to 100 control steps. The state variables include four quantities: lateral position error, heading angle deviation, yaw rate, and longitudinal speed. The state equation is based on a two-degree-of-freedom bicycle model, and the model parameters include the vehicle mass, moment of inertia, wheelbase, and tire lateral stiffness.

[0097] S402, the control input is the front wheel steering angle and the torque difference between the left and right drive wheels. The torque difference is defined as the left wheel torque minus the right wheel torque.

[0098] S403, the objective function contains three terms: the first term is the square integral of the lateral tracking error of the trajectory, with a weight coefficient of one; the second term is the penalty term for the rate of change of the control input, with a weight coefficient of zero to one; the third term is the energy consumption minimization term, defined as the sum of squares of the drive wheel torques, with a weight coefficient of zero to one.

[0099] S404, the constraints include four categories: the first category is vehicle dynamics boundaries, including the tire adhesion elliptic limit, where longitudinal and lateral forces satisfy the elliptic equation; the second category is suspension travel boundaries, where the body roll angle does not exceed eight degrees; the third category is comfort index constraints, where the absolute value of longitudinal acceleration does not exceed 0.5 m / s³; the fourth category is actuator physical limitations, where the front wheel steering angle range is ±40 degrees and the drive wheel torque range is from -300 Nm to +300 Nm.

[0100] S405, the estimated road surface friction coefficient is calculated by combining the road surface type identified by the visual sensor and the road surface reflectivity measured by the millimeter-wave radar. The road surface types include four categories: dry asphalt, wet asphalt, gravel, and ice and snow, with corresponding friction coefficients of 0.8, 0.4, 0.3, and 0.1, respectively.

[0101] S406, the load distribution is estimated by the seat pressure sensor and the air suspension height sensor, and the front and rear axle load ratio and the left and right side load difference are calculated.

[0102] S407, the optimization problem is solved using a sequential quadratic programming algorithm. The algorithm completes the rolling optimization calculation within 20 milliseconds and outputs the optimal control sequence.

[0103] The controller executes only the first control quantity in the sequence, which is the difference between the front wheel steering angle and the torque of the left and right drive wheels in the next cycle. The torque difference is converted into independent torque commands for the left and right drive wheels by the allocation algorithm. The allocation principle is to keep the total driving force unchanged. The torque difference is used to generate yaw moment. The steering wheel steering angle correction is superimposed on the desired heading angle output by the path planning to form the final steering command.

[0104] In step S5, during the bidirectional driving mode switching phase, the redundant braking coordination module is activated, the front and rear drive shaft clutches are locked synchronously, the hydraulic accumulator pressure of the steering mechanism is released, and after the vehicle body comes to a complete stop, the master-slave communication relationship of the main control unit is switched, the zero point of the inertial navigation system is reset, and the coordinates of the starting point and ending point of the path planning are reinitialized.

[0105] S501, the mode switching command is issued by the central decision-making arbitration module. The trigger condition is that the preset switching area is reached and there are no dynamic obstacles in front. The redundant braking coordination module first activates the electronic parking brake of all wheels. The braking command is sent to the brake electronic control unit of each wheel through the controller local area network bus. The electronic control unit adjusts the opening of the solenoid valve so that the brake caliper clamps the brake disc and the braking torque gradually increases to 80% of the rated value. The increase rate is 20% of the rated value per second.

[0106] S502, at the same time, the front and rear drive shaft clutches receive a lock-up command, the clutch solenoid valve is energized, and the hydraulic oil pressure pushes the piston, so that the clutch plates are fully engaged, and the torque transmission of the drive shaft is interrupted.

[0107] S503, the hydraulic accumulator pressure release command of the steering mechanism is sent to the three-position four-way solenoid valve. The solenoid valve switches to the neutral position, the accumulator is connected to the oil tank, and the pressure drops to zero within 0.5 seconds.

[0108] S504: The vehicle speed is monitored by wheel speed sensors. When the speed of all four wheels is less than 0.1 meters per second and lasts for 0.5 seconds, the vehicle is determined to be completely stationary.

[0109] S505, at this time, the master-slave relationship of the master control unit is switched. The original master control unit is downgraded to a slave and the original slave is upgraded to a master. The switching process is confirmed by the heartbeat mechanism and the switching time is less than one hundred milliseconds.

[0110] The S506 inertial navigation system reset process consists of two stages: the first stage is gyroscope zero-bias calibration, which lasts for one second, during which the vehicle body remains stationary and the average output value of the gyroscope is used as the zero-bias compensation value; the second stage is accelerometer gravity vector alignment, which lasts for two seconds, using a rotation matrix to align the accelerometer measurements to the local horizontal coordinate system with an alignment accuracy better than one degree zero.

[0111] S507, the path planning module is reinitialized, the starting coordinates are set to the current vehicle position, the ending coordinates are recalculated according to the driving direction after switching, the electronic fence data is loaded with the reverse driving parameter set, including the reverse one-way rule and the reverse slope limit, the entire mode switching process takes no more than three seconds, and the system automatically enters the reverse driving control mode after the switch is completed.

[0112] In step S6, for multi-car train formation operation scenarios, a queue maintenance controller based on a distributed consensus protocol is established. The position and attitude of the rear of the front car are used as a virtual navigator, and the relative position and attitude deviation of the front of the rear car is calculated in real time. By adjusting the speed difference of its own drive wheels and the active articulation angle, the preset formation spacing and heading synchronization are maintained.

[0113] The S601 queue holding controller is deployed in an independent controller in each car. The controllers exchange data through a dedicated wireless local area network with a communication cycle of fifty milliseconds. The data packet contains the car's position, speed, heading angle, articulation angle, and fault status code. The data packet is verified using a cyclic redundancy check code, and the packet loss rate is less than one in a thousand.

[0114] In the S602, navigator-follower architecture, the first carriage is the physical navigator, and its tail position and attitude are broadcast to the subsequent carriages as a virtual navigator. The desired position of the i-th carriage is defined as the tail position of the i-th minus one carriage moving backward along its heading by a preset distance, which is ten meters. The desired heading angle is equal to the heading angle of the i-th minus one carriage.

[0115] S603, the deviation between the actual relative position and the desired value of the i-th car is calculated in real time. The position deviation is the Euclidean distance between the actual position and the desired position, and the heading deviation is the difference between the actual heading angle and the desired heading angle. The driving wheel speed difference adjustment is calculated by the position deviation through a proportional-integral-derivative controller. The proportional gain is 0.5, the integral time is 5 seconds, and the derivative time is 0.1 seconds.

[0116] S604, the active articulation angle adjustment is calculated by the sliding mode variable structure controller based on the heading deviation. The sliding surface is a linear combination of the heading deviation and the articulation angular velocity. The control law includes an equivalent control term and a switching control term. The switching gain is adaptively adjusted according to the road surface friction coefficient.

[0117] S605 introduces a feedforward compensation term for articulation angle control. The feedforward amount is calculated based on the current vehicle speed and the radius of curvature. The radius of curvature is provided by the path planning module. The feedforward formula is that the articulation angle equals the wheelbase divided by the radius of curvature.

[0118] S606, the controller outputs drive wheel speed difference command and active articulated hydraulic cylinder extension and retraction command. The drive wheel speed difference command is sent to the drive motor controller, and the active articulated hydraulic cylinder extension and retraction command is sent to the electro-hydraulic proportional valve, ensuring that the overlap error of the center of gravity trajectory of each carriage is less than 10 centimeters when driving on a curve, and the formation spacing error is less than 0.5 meters when driving on a straight line.

[0119] As one embodiment of the present invention, the intelligent control method also includes an energy management subsystem, which monitors the state of charge of the power battery and the voltage level of the supercapacitor in real time, and dynamically allocates driving power and regenerative braking intensity based on the remaining mileage and gradient prediction results; on long downhill sections, it prioritizes the use of motor braking and stores the regenerative energy in the supercapacitor; on uphill sections, it increases the battery output power in advance and shuts down unnecessary on-board equipment to reduce the load.

[0120] As one embodiment of the present invention, before the intelligent control method is put into operation for the first time, it is necessary to complete the collection and annotation of a high-precision map of the entire scenic area, including all road centerlines, intersection topology, slope change points, and the location of visual obstructions; the data collection vehicle travels at a constant speed of five kilometers per hour, simultaneously recording the lidar point cloud, visual images, and differential global positioning system trajectory; the subsequent map updates are carried out through crowdsourcing, with the environmental change detection results uploaded by the daily operating vehicles, which are then manually reviewed in the cloud and merged into the main map database.

[0121] As one embodiment of the present invention, the intelligent control method activates a three-level emergency response mechanism when encountering a sudden emergency: Level 1 is local obstacle avoidance, which avoids obstacles by adjusting the trajectory and decelerating; Level 2 is area parking, which finds the nearest safe stopping point and evacuates passengers; Level 3 is full-line braking, which cuts off all power output and applies maximum braking force, while broadcasting an emergency stop signal to surrounding vehicles.

[0122] As one embodiment of the present invention, the intelligent control method also includes an audio-visual warning subsystem, which automatically plays pre-recorded voice prompts and flashes amber warning lights when turning, reversing, or approaching densely populated pedestrian areas; the warning intensity and frequency are adaptively adjusted according to the ambient noise level to ensure that the warning effect is not drowned out by background noise; the system records the time, location and duration of each warning trigger for later safety auditing.

[0123] As one embodiment of the present invention, the intelligent control method supports multilingual voice command recognition. Passengers can issue commands such as "speed up", "pause the explanation", and "turn on the air conditioner" through the microphone in the carriage. The system will execute the corresponding operation after voiceprint verification. The voice recognition model is specially optimized for common accents and background noise in scenic areas, and the recognition accuracy is not less than 95%.

[0124] As one embodiment of the present invention, the intelligent control method automatically links with the platform screen door control system when the vehicle enters or leaves the station. After the vehicle has come to a complete stop and the doors are aligned, an opening permission signal is sent. After the platform detects that the doors are open, the corresponding screen doors are opened simultaneously. The closing process uses both pressure sensing strips and laser curtains for detection. The vehicle is only allowed to start after confirming that no people or objects are trapped.

[0125] As one embodiment of the present invention, the intelligent control method also includes a passenger comfort optimization subsystem, which automatically adjusts the air conditioning air volume, fresh air ratio and seat heating intensity based on the temperature and humidity sensor, carbon dioxide concentration detector and seat pressure distribution data in the passenger compartment; and reduces the vehicle speed and adjusts the suspension damping coefficient in advance on bumpy road sections to reduce the amplitude of vehicle body sway.

[0126] As one embodiment of the present invention, the intelligent control method automatically enters a low-power sleep mode during long-term vehicle downtime, shutting down the power supply of unnecessary sensors and actuators, and keeping only the core monitoring module running; when a dispatching command or passenger card swipe signal is detected, the system is quickly woken up and completes self-check, and returns to standby state within 30 seconds.

[0127] As one embodiment of the present invention, the intelligent control method also includes a carbon emission metering function, which calculates the energy consumption and equivalent carbon dioxide emissions of the trip in real time based on the actual mileage, load weight, road conditions, gradient, and ambient temperature; after data aggregation, a green travel report for the scenic area is generated for management departments to refer to in formulating energy conservation and emission reduction strategies.

[0128] As one embodiment of the present invention, the intelligent control method supports multi-vehicle collaborative scheduling. In the scenario of multiple loop lines operating in parallel in a scenic area, the central server dynamically adjusts the departure interval and intersection avoidance strategy according to the location, speed and passenger volume of each vehicle, so as to avoid oncoming conflicts or rear-end collisions in the same direction on the shared road section and maximize the traffic efficiency of the road network.

[0129] As one embodiment of the present invention, the intelligent control method also includes an accessibility service function, which automatically deploys the ramp, locks the special fixing device, and adjusts the interior lighting and broadcast volume of the carriage after recognizing the boarding needs of wheelchair passengers; the system records each accessibility service usage to optimize facility layout and service process.

[0130] As one embodiment of the present invention, the intelligent control method automatically switches to a low center of gravity mode when the vehicle passes through special structures such as bridges, tunnels, and culverts, reduces the air suspension height, tightens the seat belt pretensioners, closes the windows, and activates structural stress monitoring to ensure the safety of the vehicle and infrastructure.

[0131] As one embodiment of the present invention, the intelligent control method supports an immersive guided tour service, automatically playing historical stories, ecological knowledge, and cultural anecdotes of the corresponding scenic spots based on the vehicle's real-time location and orientation; the audio content is recorded by professional announcers, supports multilingual switching, and the volume is adaptively adjusted according to ambient noise to ensure that every passenger can hear clearly.

[0132] As one embodiment of the present invention, the intelligent control method also includes a privacy protection mechanism. All collected passenger facial images, voice clips, and location trajectories are anonymized. The original data is encrypted and stored locally for no more than 24 hours. After being desensitized, it is uploaded to the cloud for service optimization, strictly complying with personal information protection regulations.

[0133] As one embodiment of the present invention, the intelligent control method automatically establishes a handshake agreement with the charging pile during vehicle charging, negotiates the optimal charging current and voltage curve, prioritizes charging during off-peak hours, and balances the grid load; during the charging process, it monitors the temperature and voltage balance of individual battery cells in real time, and immediately terminates charging and alarms if any abnormality is detected.

[0134] As one embodiment of the present invention, the intelligent control method supports customized theme services. During holidays or special events, the color of the carriage lights, the style of the background music, and the version of the tour guide script can be switched with one click to create a festive atmosphere. The system has multiple theme schemes pre-stored, and maintenance personnel can deploy them remotely through the backend.

[0135] As one embodiment of the present invention, the intelligent control method also includes a disaster emergency evacuation function. When a major disaster signal such as an earthquake, fire, or flood is detected, all vehicle doors are immediately unlocked, an escape guidance broadcast is played, the shortest evacuation route is planned, and a direct connection is established with the scenic area's emergency command center to report the number and location of people in the vehicle in real time.

[0136] As one embodiment of the present invention, the intelligent control method automatically enters maintenance mode during vehicle cleaning and maintenance, closes all external communication interfaces, locks control permissions, and only allows authorized engineers to access the diagnostic port via a physical key; the system records the content and duration of each maintenance operation and generates an electronic maintenance history.

[0137] As one embodiment of the present invention, the intelligent control method supports linkage with augmented reality glasses. After passengers wear special glasses, virtual tour guide images, historical scene restoration animations, and animal and plant identification tags can be superimposed on the car windows to enhance the tour experience. The system dynamically adjusts the position and level of detail of the AR content based on the passenger's gaze focus.

[0138] As one embodiment of the present invention, the intelligent control method also includes social interaction functions, allowing passengers to participate in scenic spot knowledge quizzes, vote on the duration of their next stop, and share tour photos to the scenic area's public screens via the touchscreen in the carriage; the system dynamically adjusts the depth of the explanation content and the recommendations of entertainment projects based on the level of interaction.

[0139] As one embodiment of the present invention, the intelligent control method automatically executes a data erasure procedure when the vehicle is decommissioned, thoroughly erasing all operating logs, passenger information, and map data. The hardware modules are disassembled and classified according to environmental protection standards, and the core control board is handed over to a professional organization for safe destruction to prevent the leakage of sensitive information.

[0140] Example 2:

[0141] The present invention also provides an intelligent control system for a bidirectional wheeled sightseeing train, including an environmental perception fusion module, a global path planning module, a local trajectory tracking module, a bidirectional mode switching control module, a multi-carriage collaborative formation module, an actuator drive module, and a central decision-making and arbitration module.

[0142] The environmental perception fusion module is located at the front and rear ends of the vehicle body and includes three sets of 16-line LiDAR, four sets of 77GHz millimeter-wave radar, six sets of two-megapixel wide-angle cameras, and one set of six-axis microelectromechanical inertial measurement unit. All sensor output signals are transmitted to the on-board industrial control computer via gigabit Ethernet bus.

[0143] In the environmental perception fusion module, LiDAR point cloud data, after voxel filtering and ground segmentation, is input into a 3D target detection network based on an attention mechanism, outputting obstacle categories, sizes, positions, and motion vectors; visual sensor image streams are processed by a semantic segmentation neural network to identify lane line types, traffic signs, pedestrian areas, and drivable road surface boundaries; millimeter-wave radar raw echo signals are processed by constant false alarm rate detection and cluster analysis to generate a list of moving obstacles; the output of the inertial measurement unit is fused with wheel speed pulse signals through a Kalman filter to calculate the vehicle's absolute position and attitude angles; the above four types of data are fused at the feature level under a unified time reference to construct a dynamic environmental situation map.

[0144] The global path planning module runs on an embedded processor with a real-time operating system. It has a built-in high-precision map database of scenic spots and a dynamic obstacle occupancy grid update engine. It adopts a hierarchical state space search strategy, superimposes real-time traffic flow density weights on the static road network topology, and outputs a global reference trajectory that meets the two-way traffic rules.

[0145] In the global path planning module, the scenic area's electronic fence data is stored in a geographic information system format, including restricted areas, one-way streets, slope restriction areas, and minimum turning radius constraint areas. The path search algorithm introduces a bidirectional driving penalty factor into the traditional A-star heuristic function, applying additional cost weights to path nodes that need to switch driving directions. The generated global trajectory is smoothed by cubic B-spline curves to ensure that the curvature continuity meets the vehicle dynamics constraints.

[0146] The local trajectory tracking module uses a nonlinear model predictive controller with a prediction time domain set to five seconds and a control cycle of twenty milliseconds. The state variables include lateral position error, heading angle deviation, yaw rate, and longitudinal vehicle speed. The control input is the difference between the front wheel steering angle and the torque of the left and right drive wheels. The constraints cover the tire adhesion ellipse limit and the suspension travel boundary.

[0147] In the local trajectory tracking module, the objective function of the model predictive controller includes three terms: the integral term of the square of the lateral tracking error, the penalty term for the rate of change of control input, and the energy consumption minimization term. In addition to the vehicle dynamics boundary, the constraints include a comfort index constraint, namely, the absolute value of the longitudinal jerk does not exceed 0.5 m / s³. The solver uses a sequential quadratic programming algorithm to complete the rolling optimization calculation within 20 milliseconds.

[0148] The bidirectional mode switching control module is equipped with dual redundant brake air circuits and an electro-hydraulic steering valve group. After receiving the mode switching command, it first triggers the electronic parking brake of all wheels. After the vehicle speed returns to zero, it disconnects the main drive motor enable signal, releases the mechanical lock pin of the steering system after a delay of 500 milliseconds, and simultaneously restarts the inertial navigation system and loads the reverse driving parameter set.

[0149] In the bidirectional mode switching control module, the braking system adopts a dual-circuit air pressure braking architecture, with the front and rear axles each supplied with pressure by independent air tanks. During the switching process, the spring energy storage brake is activated first, and the drive system power output is cut off after confirming that the braking torque of the four wheels reaches 80% of the rated value. The steering system is equipped with a double-acting hydraulic cylinder, which switches the oil circuit direction through a three-position four-way solenoid valve during switching, causing the steering wheel to rotate 180 degrees from the neutral position to the reverse driving preparation angle. The inertial navigation system reset process includes two stages: gyroscope zero bias calibration and accelerometer gravity vector alignment, with a total time of no more than three seconds.

[0150] The multi-car cooperative formation module is deployed on an independent controller in each car. It exchanges relative position and speed information through a dedicated wireless local area network and uses a sliding mode variable structure control law to suppress articulation angle oscillation, ensuring that the overlap error of the center of gravity trajectory of each car is less than ten centimeters when driving on curves.

[0151] In the multi-car cooperative formation module, adjacent cars establish point-to-point communication links through directional antennas, with a transmission period of fifty milliseconds. The data packets include the car's position, speed, heading angle, articulation angle, and fault status code. The platoon controller adopts a navigator-follower architecture. The following car calculates the desired relative pose based on the motion status broadcast by the navigator car and eliminates the deviation between the actual relative pose and the desired value by adjusting the differential ratio of its own drive wheels and the extension and retraction of the active articulation hydraulic cylinder. The articulation angle control introduces a feedforward compensation term to predict the required articulation angle based on the current vehicle speed and radius of curvature, reducing tracking lag.

[0152] The actuator drive module includes wheel-side permanent magnet synchronous motors and their vector control drivers, electro-hydraulic proportional steering valves, electromagnetic brake calipers, and air suspension height adjustment valves. It receives standardized control commands from the central decision-making arbitration module to complete torque output, steering angle execution, braking force application, and vehicle attitude adjustment.

[0153] In the actuator drive module, the wheel-side motor is a water-cooled permanent magnet synchronous motor with a rated power of 15 kW and a peak torque of 300 Nm, equipped with a rotary transformer and a temperature sensor; the steering actuator is a rack and pinion electric power steering with a maximum steering angle of ±40 degrees and a response bandwidth of 10 Hz; the braking system uses disc brakes with an electronic control unit, which can realize independent braking pressure adjustment of each wheel, with a minimum pressure build-up time of 200 milliseconds; the air suspension system adjusts the airbag inflation and deflation volume through a proportional solenoid valve based on feedback from the vehicle height sensor to maintain a constant vehicle ground clearance.

[0154] The central decision-making and arbitration module, as the core scheduling unit of the system, is responsible for coordinating the operating priorities of various functional modules, handling abnormal conditions such as sensor failure, communication interruption, and path blockage, and ensuring that the system still has basic operating capabilities in degraded mode according to the preset safety state machine switching control strategy.

[0155] The central decision-making and arbitration module incorporates multiple fault diagnosis logics. When any lidar data is lost for more than one second, the weight of the millimeter-wave radar and vision fusion is automatically increased. When wireless communication is interrupted for more than three seconds, local path replanning is initiated and the speed is reduced to walking speed. When an unavoidable obstacle is detected ahead of the path, emergency braking is immediately triggered and the event is reported to the scenic area dispatch center. The system operation log is stored on a solid-state drive with minute-level granularity, containing all raw sensor data, control command sequences, and state machine jump records.

[0156] As one embodiment of the present invention, the intelligent control system supports remote OTA upgrade function, and scenic area operation and maintenance personnel can push new path planning algorithm parameters, obstacle recognition model weights or control law gain coefficients through encrypted wireless channels.

[0157] The system automatically downloads the update package during non-operational periods, completes the integrity verification, and loads the new version of the software on the next startup; the upgrade process does not affect the stored operation logs and fault records.

[0158] As one embodiment of the present invention, the intelligent control system is equipped with a human-machine interaction terminal, which is set in the driver's cabin and passenger compartment. The driver can select the operating mode, view the system status, and manually take over control through the touch screen; the passenger terminal displays the current vehicle speed, remaining mileage, scenic spot introduction, and emergency call button; all operation commands must be verified by authorization and double confirmation to prevent accidental triggering that could cause system abnormalities.

[0159] As one embodiment of the present invention, the intelligent control system connects to the scenic area's smart transportation platform through a vehicle-road cooperative interface, receiving traffic light phase information, temporary construction zone notices, and special activity route control instructions; the system dynamically adjusts its route planning strategy accordingly, avoiding controlled areas or waiting for the green light; at the same time, it uploads the vehicle's real-time location and estimated arrival time to the platform for tourists to query and optimize scheduling.

[0160] As one embodiment of the present invention, the intelligent control system has a built-in digital twin simulation engine. Before each major parameter adjustment or the opening of a new route, a full-scenario stress test is first carried out in a virtual environment to simulate various extreme weather, sensor failure, communication delay, sudden passenger behavior and other abnormal conditions to verify the robustness of the control strategy. Only after the simulation results meet the standards can it be deployed to the real vehicle.

[0161] As one embodiment of the present invention, the intelligent control system is equipped with a health status monitoring module, which collects the vibration spectrum, temperature gradient and current harmonic characteristics of each key component in real time, and uses a support vector machine classifier to determine whether there are early signs of failure; when the predicted remaining life is lower than a preset threshold, a maintenance work order is automatically generated and pushed to the mobile terminal of the maintenance personnel to realize predictive maintenance.

[0162] As one embodiment of the present invention, the intelligent control system supports dynamic adjustment of the formation size, automatically combining two to five carriages into flexible formations based on real-time passenger flow and the number of available vehicles; when a new carriage is added to the formation, it automatically completes identity registration, parameter synchronization and control transfer; when leaving the formation, it smoothly decelerates and enters the waiting area, without any manual intervention throughout the entire process.

[0163] As one embodiment of the present invention, the intelligent control system is equipped with an anti-theft and anti-vandalism monitoring device. Through vibration sensors in the carriage and video analysis algorithms, it identifies abnormal collisions, illegal intrusions, and items left behind. Once an alarm is triggered, it immediately locks the doors, starts recording video inside the vehicle for evidence collection, and sends location information and on-site footage to the security center.

[0164] As one embodiment of the present invention, the intelligent control system supports deep integration with the scenic area ticketing system. When passengers enter the station by swiping their ID cards or QR codes, the system automatically assigns the best carriage seats and guides them to the designated waiting area. After the vehicle arrives at the station, it pushes boarding reminders and carriage number guidance to passengers through Bluetooth beacons and mobile APP, reducing the time spent on the platform.

[0165] As one embodiment of the present invention, the intelligent control system is equipped with a teaching demonstration mode, which allows technicians to learn the system architecture and control logic during non-operational periods. In this mode, some sensor inputs can be frozen, simulated fault signals can be injected, and the system response behavior can be observed to help newcomers quickly master the fault diagnosis and emergency response procedures.

[0166] As one embodiment of the present invention, the intelligent control system has a built-in legal compliance check module to ensure that all control decisions comply with local road traffic safety regulations and scenic area management regulations; for example, it automatically limits the speed to five kilometers per hour near children's playgrounds, prohibits honking and strong light exposure in the core area of ​​cultural relic protection, and permanently saves system logs for supervision and review.

[0167] As one embodiment of the present invention, the intelligent control system supports connection with the meteorological early warning platform. When it receives an alarm for extreme weather such as rainstorm, strong wind, or hail, it automatically activates the emergency plan, including reducing speed, shortening the interval between trains, returning early, turning on the defogging and defrosting device, and pushing safety reminder information to passengers.

[0168] As one embodiment of the present invention, the intelligent control system is equipped with a biometric identification module. Drivers need to verify their identity by fingerprint or iris before starting work. The system records operation logs and binds them to personal accounts. Passengers can choose facial payment or vein recognition to quickly enter and exit the station, improving traffic efficiency while ensuring transaction security.

[0169] As one embodiment of the present invention, the intelligent control system has a built-in anti-interference communication protocol, which can maintain reliable transmission of control commands and status data even in the complex electromagnetic environment of the scenic area; it adopts frequency hopping spread spectrum technology to combat co-channel interference, forward error correction coding to resist data packet loss, and a heartbeat packet mechanism to detect link connectivity, with a disconnection reconnection time of less than 500 milliseconds.

[0170] As one embodiment of the present invention, the intelligent control system supports seamless integration with the urban public transportation system, and links with subway, bus and shared bicycle stations at the scenic area exit to provide passengers with next-trip travel suggestions and transfer navigation; the system predicts the peak departure passenger flow based on historical data and dispatches shuttle vehicles in advance to avoid congestion.

[0171] As one embodiment of the present invention, the intelligent control system is equipped with a fatigue driving monitoring device, which judges the operator's mental state by analyzing the driver's facial expression and detecting the steering wheel grip force; when distraction or slow reaction is detected, a graded alarm is triggered, and in severe cases, the system is forcibly switched to the automatic driving mode and a replacement is notified to take over.

[0172] As one embodiment of the present invention, the intelligent control system has a built-in intellectual property protection module. All core algorithms and control parameters are digitally watermarked and access permissions are controlled to prevent unauthorized copying and tampering. When the system starts up, it automatically verifies the integrity of the software. If an anomaly is detected, it immediately locks the system and reports it to the security center.

[0173] As one embodiment of the present invention, the intelligent control system supports collaborative work with the drone inspection system. The drone regularly flies along the sightseeing route to photograph potential hazards such as road damage, fallen trees, and blurred signs. After AI analysis, the images generate maintenance work orders, and the system dynamically adjusts the vehicle's driving route to avoid dangerous areas.

[0174] As one embodiment of the present invention, the intelligent control system is equipped with an acoustic environment optimization module. Through microphone arrays arranged at the four corners of the carriage, the internal noise spectrum is collected in real time, and the active noise-canceling speakers are driven to emit anti-phase sound waves to cancel low-frequency components such as engine noise, wind noise, and tire noise, thereby controlling the background noise in the carriage to below 45 decibels.

[0175] As one embodiment of the present invention, the intelligent control system has a built-in blockchain evidence storage module, which generates hash values ​​for key events of each trip, such as start and stop times, route changes, emergency braking, and passenger complaints, and stores them on the blockchain to ensure that the operational data is tamper-proof and to provide a reliable basis for accident liability determination and service quality assessment.

[0176] As one embodiment of the present invention, the intelligent control system supports linkage with IoT devices in scenic areas such as smart trash cans, smart streetlights, and smart seats. When a vehicle approaches, the system can turn on the path lighting in advance, open the trash can lid, and preheat the seat temperature to achieve seamless smart service. When the equipment is in abnormal condition, it can automatically report a maintenance request.

[0177] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An intelligent control method for a bidirectional wheeled sightseeing train, characterized in that, include: Environmental perception data is collected synchronously by multimodal sensor arrays deployed at the front and rear of the vehicle body; Multi-source heterogeneous sensor data are spatiotemporally aligned and coordinate system transformed to generate a global environmental grid map with the vehicle center as the origin and a dynamic obstacle trajectory prediction sequence. Based on the preset electronic fence boundary of the scenic area and the real-time updated global map, the improved hybrid A-star path planning algorithm is called to generate the globally optimal collision-free trajectory. The scenic area's electronic fence data is stored in Geographic Information System (GIS) format, including restricted areas, one-way streets, slope restriction areas, and minimum turning radius constraint areas. The path planning algorithm adopts a hierarchical state space search strategy, with the upper layer being a road network topology search and the lower layer being a local grid map search. A bidirectional driving penalty factor is introduced into the traditional A-star heuristic function, applying additional cost weights to path nodes that need to switch driving directions. The state extension must meet vehicle dynamics constraints, including a minimum turning radius of five meters, a maximum roll angle of eight degrees, and a maximum longitudinal acceleration of two meters per second squared. The generated path node sequence is smoothed by cubic B-spline curves to ensure that the curvature continuity meets the vehicle's lateral acceleration constraints, and a global reference trajectory is output. The trajectory points are spaced at 0.5 meters apart, and each point contains four attributes: global coordinates, desired velocity, desired heading angle, and radius of curvature. A model predictive control framework is adopted, with trajectory tracking error and energy consumption within the look-ahead window as optimization targets. Combined with road conditions and load distribution, the drive wheel torque distribution and steering wheel angle commands are calculated in real time. During the bidirectional driving mode switching phase, the redundant braking coordination module is activated and executes the following steps in sequence: triggering the electronic parking brake for all wheels, gradually increasing the braking torque to 80% of the rated value, with an increase rate of 20% of the rated value per second; causing the front and rear drive shaft clutches to receive lock-up commands, and hydraulic oil pressure pushes the pistons to fully engage the clutch plates; sending the hydraulic accumulator pressure release command of the steering mechanism to the three-position four-way solenoid valve, causing the pressure to drop to zero within 0.5 seconds, and causing the steering wheels to rotate 180 degrees from the neutral position to the reverse driving preparation angle; after the vehicle speed is monitored by the wheel speed sensors, when the wheel speeds of all four wheels are less than 0.1 meters per second and remain below 0.5 seconds for 0.5 seconds, and the vehicle is determined to be completely stationary, the master-slave relationship of the main control unit communication is switched, and the inertial navigation system and path planning module are reset and re-initialized. For multi-car formation scenarios, a distributed queue maintenance controller is established. Based on the virtual navigator, the relative pose deviation is calculated, and the formation spacing and heading synchronization are maintained by adjusting the speed difference of the drive wheels and the active articulation angle. The controller is deployed in an independent controller in each carriage and exchanges data through a dedicated wireless local area network with a communication cycle of fifty milliseconds. The desired position of the i-th carriage is defined as the position of the rear of the i-th minus one carriage moving backward by a preset distance of ten meters along its heading, and the desired heading angle is equal to the heading angle of the i-th minus one carriage. Position deviation is calculated by a proportional-integral-derivative (PID) controller to adjust the speed difference of the drive wheels. The proportional gain is 0.5, the integral time is 5 seconds, and the derivative time is 0.1 seconds. Heading deviation is calculated by a sliding mode variable structure controller to adjust the active articulation angle. The sliding surface is a linear combination of heading deviation and articulation angular velocity, and its switching gain is adaptively adjusted according to the road friction coefficient. The articulation angle control introduces a feedforward compensation term. The feedforward amount is calculated based on the current vehicle speed and the radius of curvature. The formula is that the articulation angle equals the wheelbase divided by the radius of curvature. The controller outputs drive wheel speed difference commands and active articulation hydraulic cylinder extension commands to ensure that the overlap error of the center of gravity trajectory of each compartment is less than 10 centimeters when driving on curves.

2. The intelligent control method for a bidirectional wheeled sightseeing train according to claim 1, characterized in that, The multimodal sensor array includes: Three sets of 16-line lidar are installed on the left and right sides and the center of the front and rear of the vehicle, respectively. The scanning frequency is 20 frames per second, the horizontal field of view is 360 degrees, the vertical field of view is 30 degrees, and the ranging accuracy is ±2 centimeters. Four 77GHz millimeter-wave radars are arranged at the four corners of the vehicle, with a detection range of 0.5 meters to 150 meters, a speed resolution of 0.1 meters per second, and an angular resolution of one degree. Six sets of two-megapixel wide-angle cameras cover the front, rear, left, right, and up and down views of the vehicle, with a frame rate of 30 Hz and a dynamic range of 120 decibels. The inertial measurement unit is a six-axis microelectromechanical system, which includes a three-axis gyroscope and a three-axis accelerometer. The sampling frequency is 200 Hz, and the zero-bias stability is better than 0.1 degree per hour. All sensors trigger synchronous acquisition upon startup, with data packets carrying hardware timestamps, ensuring time synchronization accuracy better than one millisecond.

3. The intelligent control method for a bidirectional wheeled sightseeing train according to claim 2, characterized in that, Generate a global environment grid map with the vehicle body center as the origin and a dynamic obstacle trajectory prediction sequence, including: Based on the hardware timestamps of each sensor, all data are interpolated to a unified time base. The interpolation method is cubic spline interpolation, and the time base is set to the sampling time of the inertial measurement unit. The data from the local coordinate system of each sensor is transformed to the vehicle coordinate system. The transformation matrix is ​​obtained through factory calibration and includes a rotation matrix and a translation vector. The data in the vehicle coordinate system is transformed to the global navigation coordinate system. The transformation parameters are provided by the vehicle pose calculated by the fusion of the inertial measurement unit and the wheel speed sensor. The lidar point cloud data is subjected to voxel filtering in the vehicle coordinate system. The voxel size is 0.1 meters. After filtering out outliers, ground segmentation is performed. A plane fitting method based on normal vectors is used to divide the point cloud into ground points and non-ground points. Input non-ground point clouds into an attention-based 3D target detection network, and output obstacle category, size, position, and motion vector; The visual sensor image stream is input into the semantic segmentation neural network, which outputs pixel-level semantic labels. The raw echo signal from the millimeter-wave radar is processed by a constant false alarm rate detector and then grouped by a density clustering algorithm to generate a list of moving obstacles. The output of the inertial measurement unit is fused with the wheel speed pulse signal through a Kalman filter to calculate the absolute position and attitude angle of the vehicle body in the global navigation coordinate system; The above four types of processed data are fused at the feature level under a unified time reference to construct a dynamic environmental situation map. The fusion process adopts a weighted average method, and the weights are dynamically allocated according to the sensor confidence level.

4. The intelligent control method for a bidirectional wheeled sightseeing train according to claim 3, characterized in that, The model predictive control framework includes: The model predictive controller has a cycle of 20 milliseconds, a prediction time domain of 5 seconds, and state variables including lateral position error, heading angle deviation, yaw rate, and longitudinal speed. The control input is the difference between the front wheel steering angle and the torque of the left and right drive wheels; The objective function includes a squared integral term for the lateral tracking error, a penalty term for the rate of change of control input, and an energy consumption minimization term; The constraints include tire adhesion elliptic limits, suspension travel boundaries, absolute value of longitudinal acceleration not exceeding 0.5 m / s³, and physical limitations of the actuators. The road surface friction coefficient estimate is calculated by combining the road surface type identified by the visual sensor with the road surface reflectivity measured by the millimeter-wave radar; The load distribution is estimated jointly by the seat pressure sensor and the air suspension height sensor; The optimization problem is solved using a sequential quadratic programming algorithm, which outputs the optimal control sequence and executes only the first control variable in the sequence.

5. An intelligent control system for a bidirectional wheeled sightseeing train, applied to the intelligent control method for the bidirectional wheeled sightseeing train as described in any one of claims 1 to 4, characterized in that, include: The environmental perception fusion module is used to synchronously collect environmental perception data through multimodal sensor arrays deployed at the front and rear of the vehicle body; The global path planning module is used to generate the globally optimal collision-free trajectory by calling the improved hybrid A-star path planning algorithm based on the preset scenic area electronic fence boundary and the real-time updated global map. The local trajectory tracking module is used to introduce a model predictive control framework to calculate the independent torque distribution commands for each drive wheel and the steering wheel angle correction in real time; The bidirectional mode switching control module is used to activate the redundant braking coordination module during the bidirectional driving mode switching phase, perform clutch lock-up and steering pressure release operations, and switch the master-slave relationship of the main control unit, reset the navigation system and reinitialize the path planning coordinates after the vehicle body stops. The multi-car cooperative formation module establishes a distributed queue maintenance controller for multi-car formation scenarios. It calculates the relative pose deviation based on the virtual navigator and maintains the formation spacing and heading synchronization by adjusting the speed difference of the drive wheels and the active articulation angle. The actuator drive module is used to receive control commands and complete torque output, steering angle execution, braking force application, and vehicle attitude adjustment; The central decision-making and arbitration module is used to coordinate the operating priorities of various functional modules, handle abnormal operating conditions, and switch control strategies according to the preset safety state machine.

6. The intelligent control system for the bidirectional wheeled sightseeing train according to claim 5, characterized in that, In the environmental perception fusion module: The lidar point cloud data is input into an attention-based 3D target detection network after voxel filtering and ground segmentation. Visual sensor images are processed by a semantic segmentation neural network; The raw echo signal from the millimeter-wave radar is subjected to constant false alarm rate detection and cluster analysis; The inertial measurement unit output is fused with wheel speed pulse signals via a Kalman filter to calculate the absolute position and attitude angle of the vehicle body; The above four types of data are fused at the feature level under a unified time benchmark to construct a dynamic environmental situation map.

7. The intelligent control system for the bidirectional wheeled sightseeing train according to claim 6, characterized in that, The central decision-making arbitration module incorporates multiple fault diagnosis logics for: If any LiDAR data is lost for more than a set time, the weight of the millimeter-wave radar and vision fusion will be automatically increased. If the wireless communication interruption continues for more than a set time, local route replanning will be initiated and the speed will be reduced to walking speed. When an unavoidable obstacle is detected ahead of the path, emergency braking is immediately triggered and the incident is reported. The system operation log is stored at a set granularity and includes raw sensor data, control command sequences, and state machine transition records.

Images