Method and system for speed control of unmanned vehicle based on fuzzy control

Abstract

The present application relates to the technical field of unmanned vehicle control, in particular to a speed control method and system for unmanned vehicle based on fuzzy control, comprising: obtaining road environment information and traffic participant information in real time through a vehicle-mounted sensing system, and obtaining a comprehensive risk index through fusion processing; inputting the comprehensive risk index into a fuzzy controller to output a preliminary expected acceleration adjustment amount; performing dynamic limiting on the adjustment amount according to a road adhesion coefficient to generate a constrained expected acceleration value; calling an improved model predictive control algorithm, pre-compensating vehicle states in a prediction time domain in combination with a curvature radius, taking the constrained expected acceleration value as a core target, and solving a final execution control instruction in combination with a vehicle dynamics model to regulate and control a vehicle driving and braking system. The method can improve the accuracy and safety of unmanned vehicle speed control and adapt to complex roads and following scenes.

Description

Technical Field

[0001] This invention relates to the field of autonomous vehicle control technology, and in particular to an autonomous vehicle speed control method and system based on fuzzy control. Background Technology

[0002] Speed ​​control of autonomous vehicles is a key technology for ensuring vehicle safety and ride comfort. Currently, conventional speed control methods for autonomous vehicles mostly rely on onboard perception systems to acquire information about the road environment and traffic participants, and then combine this information with algorithms such as fuzzy control and model predictive control to generate control commands to regulate the vehicle's drive and braking systems. In practical applications, these methods typically rely solely on a fuzzy controller to output the desired acceleration, and then use conventional model predictive control algorithms to solve for the control commands to adjust the vehicle's speed.

[0003] In conventional technical solutions, the initial expected acceleration adjustment output by the fuzzy controller does not consider the influence of the road adhesion coefficient and lacks dynamic amplitude limiting, which can easily lead to the output acceleration exceeding the actual dynamic limits of the vehicle. On roads with low adhesion coefficients, this can easily result in slippage, braking failure, and other problems, affecting driving safety. Furthermore, conventional model predictive control algorithms do not compensate for the vehicle state in the prediction time domain by incorporating the road curvature radius when solving for control commands. In complex road scenarios such as curves, the prediction accuracy is low, the control response is lagging, and it is difficult to adapt to the speed control requirements of roads with different curvatures, making it impossible to achieve precise speed control in car-following scenarios. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a speed control method and system for unmanned vehicles based on fuzzy control.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a speed control method for unmanned vehicles based on fuzzy control, comprising:

[0006] The vehicle-mounted perception system acquires real-time information on the road environment and traffic participants in front of the autonomous vehicle. The road environment information includes the radius of curvature, the road surface adhesion coefficient, and the slope angle. The traffic participant information includes the relative distance, relative speed, and movement trend of the target object.

[0007] The road environment information and traffic participant information are fused and processed to calculate a comprehensive risk index that characterizes the urgency of the current car-following scenario;

[0008] The comprehensive risk index is input into a preset fuzzy controller, which outputs an initial expected acceleration adjustment amount based on a preset membership function and fuzzy inference rules.

[0009] The desired acceleration adjustment amount is dynamically limited based on the road surface adhesion coefficient to generate a constrained desired acceleration value;

[0010] An improved model predictive control algorithm is invoked, with the constrained desired acceleration value as the core control objective. Combined with the vehicle dynamics model, the final execution control command is obtained. The execution control command is used to directly regulate the vehicle's drive system and braking system. The improved model predictive control algorithm performs pre-compensation on the vehicle state in the prediction time domain based on the radius of curvature.

[0011] As a further aspect of the present invention, the road environment information and traffic participant information are fused and processed to calculate a comprehensive risk index characterizing the urgency of the current car-following scenario, including:

[0012] Extract the relative distance and relative speed between the vehicle and the vehicle in front from the traffic participant information;

[0013] Based on the vehicle braking model, the minimum safe distance is calculated according to the relative speed and the current road surface adhesion coefficient.

[0014] Calculate the ratio of the relative distance to the minimum safe distance to obtain the first-level risk factor;

[0015] Analyze the movement trend of the target object. If the target object has the intention to cut in, decelerate, or stop, then calculate the second-level risk factor based on the intensity of the intention and the remaining reaction time.

[0016] Combining the radius of curvature and slope angle, the third-level risk factor is obtained by querying the pre-set curve and slope risk mapping table;

[0017] A weighted fusion strategy is adopted to fuse the first-level risk factor, the second-level risk factor, and the third-level risk factor. The weight of each factor in the weighted fusion strategy is dynamically adjusted according to the current vehicle speed and the road surface adhesion coefficient. The fusion result is the comprehensive risk index.

[0018] As a further aspect of the present invention, the improved model predictive control algorithm is invoked, with the constrained desired acceleration value as the core control objective, and combined with the vehicle dynamics model to obtain the final execution control command, including:

[0019] A simplified vehicle dynamics model incorporating longitudinal kinematics and tire force constraints is established, and the constrained desired acceleration value is used as the reference acceleration of the simplified vehicle dynamics model at the start of the prediction time domain.

[0020] Within each control cycle of the improved model predictive control algorithm, based on the current vehicle state, the vehicle dynamics simplified model is used to predict the vehicle position and speed sequence at multiple future sampling times.

[0021] The predicted vehicle position and velocity sequence is compared with the reference state sequence generated based on the expected acceleration value after the constraints, and an optimization problem is constructed with the state deviation and the rate of change of control quantity as the objective function.

[0022] In the process of solving the optimization problem, the predicted lateral displacement is constrained based on the current radius of curvature, and the vehicle gravity component is compensated based on the current slope angle.

[0023] Solve the optimization problem to obtain the optimal future control input sequence, and use the first control quantity in the optimal future control input sequence as the execution control command at the current moment, and output it to the vehicle's drive system and braking system.

[0024] As a further aspect of the present invention, the improved model predictive control algorithm pre-compensates the vehicle state in the prediction time domain based on the radius of curvature, including:

[0025] In the state prediction module of the improved model predictive control algorithm, the prediction time domain is divided into multiple sub-intervals;

[0026] Based on the radius of curvature, calculate the theoretical change in centripetal acceleration caused by the curve in each sub-interval when the vehicle travels along the predetermined trajectory.

[0027] The theoretical centripetal acceleration change is converted into a compensation amount for the longitudinal desired acceleration. A negative compensation amount indicates that deceleration needs to be done in advance, while a positive compensation amount indicates that appropriate acceleration is allowed.

[0028] In the rolling optimization calculation, the compensation amount is superimposed on the reference acceleration sequence generated based on the desired acceleration value after the constraint to form a pre-compensated reference state sequence.

[0029] The pre-compensated reference state sequence and the state sequence predicted by the vehicle dynamics model are used for optimization, so that the generated control commands can actively adapt to changes in the curvature of the road ahead.

[0030] As a further aspect of the present invention, the desired acceleration adjustment amount is dynamically limited based on the road surface adhesion coefficient to generate a constrained desired acceleration value, including:

[0031] Based on the real-time detected or estimated road adhesion coefficient, the preset maximum allowable acceleration and maximum allowable deceleration mapping table is queried to obtain the acceleration limit and deceleration limit under the current road conditions.

[0032] The desired acceleration adjustment output by the fuzzy controller is superimposed with the actual acceleration of the current vehicle to obtain the unconstrained original value of the desired acceleration.

[0033] Compare the original value of the desired acceleration with the acceleration limit value and the deceleration limit value;

[0034] If the original value of the desired acceleration is greater than the acceleration limit value, then the constrained desired acceleration value is set as the acceleration limit value;

[0035] If the original value of the desired acceleration is less than the deceleration limit value, then the constrained desired acceleration value is set as the deceleration limit value;

[0036] If the original value of the desired acceleration is between the acceleration limit value and the deceleration limit value, then the constrained desired acceleration value is equal to the original value of the desired acceleration.

[0037] As a further aspect of the present invention, the method further includes a step of real-time estimation of the road surface adhesion coefficient:

[0038] Monitor the vehicle's wheel speed signal, longitudinal acceleration signal, and output torque signal of the drive motor or braking system;

[0039] Based on the vehicle's dynamics, the wheel slip ratio is calculated using the wheel speed signal;

[0040] By combining the longitudinal acceleration signal and the output torque signal, the longitudinal force between the tire and the road surface is calculated using a reverse calculation method.

[0041] Based on the wheel slip ratio and the calculated longitudinal force, the tire characteristic curve can be queried or the tire model can be used to estimate the current road adhesion coefficient in real time.

[0042] The estimated road surface adhesion coefficient is subjected to low-pass filtering to eliminate noise interference and obtain a stable estimate for control decision-making.

[0043] As a further aspect of the present invention, the predicted vehicle position and velocity sequence is compared with a reference state sequence generated based on the desired acceleration value after the constraints, and an optimization problem is constructed with the state deviation and the rate of change of control quantity as objective functions, including:

[0044] The position deviation and speed deviation are obtained by subtracting the vehicle position and speed predicted at each sampling time in the prediction time domain from the reference position and reference speed in the reference state sequence at the corresponding time.

[0045] Weighting coefficients are assigned to the position deviation, speed deviation, and rate of change of the control quantity, and the weighting coefficients are adaptively adjusted according to the current vehicle speed and the comprehensive risk index.

[0046] Construct an objective function, which is the sum of the weighted squared position deviation, the weighted squared velocity deviation, and the squared rate of change of the weighted control quantity at all sampling times in the prediction time domain;

[0047] The constraints of the optimization problem include the allowable fluctuation range of the desired acceleration value after the constraints, the physical limits of the vehicle drive system and braking system, and the lower limit constraint of the position calculated based on the safe distance model.

[0048] As a further aspect of the present invention, the weighting coefficient is adaptively adjusted based on the current vehicle speed and the comprehensive risk index, including:

[0049] Define a set of basic weight coefficients, including basic weights for position deviation, velocity deviation, and control rate of change;

[0050] Establish a weight adjustment coefficient table, which is indexed by vehicle speed range and comprehensive risk index level;

[0051] Based on the current real-time vehicle speed and the calculated comprehensive risk index, the weight adjustment coefficient table is consulted to obtain the position weight adjustment factor, speed weight adjustment factor and control rate of change weight adjustment factor.

[0052] Multiply the base weight of the position deviation by the position weight adjustment factor to obtain the final applied position deviation weight coefficient;

[0053] Multiply the base weight of the speed deviation by the speed weight adjustment factor to obtain the final applied speed deviation weight coefficient;

[0054] Multiply the basic weight of the control change rate by the control change rate weight adjustment factor to obtain the final applied control change rate weight coefficient.

[0055] As a further aspect of the present invention, the method further includes an online fine-tuning step of fuzzy rules based on historical control data:

[0056] Record the input of the comprehensive risk index, the output of the expected acceleration adjustment amount, and the final actual acceleration response and following stability evaluation of the vehicle in each control cycle;

[0057] When the actual following distance is below the comfort threshold for an extended period or the rate of change of acceleration exceeds the smoothness threshold, the current control effect is deemed unsatisfactory.

[0058] For the fuzzy set where the comprehensive risk index is located when the control effect is not ideal, the conclusion part of the corresponding fuzzy inference rule in the fuzzy controller is slightly modified.

[0059] The direction of the correction is to make the desired acceleration adjustment in the output more smooth or conservative, and the magnitude of the correction is positively correlated with the duration of the non-ideal degree.

[0060] The revised fuzzy rules are updated in the rule base of the fuzzy controller for subsequent control decisions.

[0061] As a further aspect of the present invention, the present invention also includes a speed control system for unmanned vehicles based on fuzzy control, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the speed control method for unmanned vehicles based on fuzzy control as described above.

[0062] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0063] The initial desired acceleration adjustment amount output by the fuzzy controller is dynamically limited based on the road surface adhesion coefficient to generate a constrained desired acceleration value. This dynamic limiting process adapts to different road surface adhesion conditions, ensuring the desired acceleration value remains within the vehicle's actual dynamic load range. This avoids abnormal situations such as vehicle slippage and brake failure caused by excessively high or low acceleration exceeding the road surface adhesion capacity. It makes the acceleration output more closely match the vehicle's actual operating state, adapting to the speed control requirements of different road environments and improving the stability and adaptability of speed control.

[0064] In the improved model predictive control algorithm, the vehicle state in the prediction time domain is pre-compensated based on the radius of curvature. The constrained expected acceleration value is used as the core control objective, and the final control command is obtained by combining the vehicle dynamics model. The radius of curvature pre-compensation can adapt to road scenarios with different curvatures, such as curves, in advance, correct the deviation of the vehicle state in the prediction time domain, improve the accuracy of vehicle state prediction, reduce control response lag, and enable the execution of control commands to match the current road curvature and car-following requirements more quickly and accurately. This adapts to car-following scenarios in complex road environments, making vehicle speed adjustments smoother and more in line with changes in road morphology and the state of traffic participants. Attached Figure Description

[0065] Figure 1 This is a flowchart of the speed control method for unmanned vehicles based on fuzzy control according to the present invention;

[0066] Figure 2 A flowchart for calculating the comprehensive risk index through fusion processing;

[0067] Figure 3 A flowchart for executing control instructions to improve model predictive control solutions. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0069] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0070] See Figure 1 This invention provides a speed control method for autonomous vehicles based on fuzzy control, the overall implementation of which is as follows:

[0071] The vehicle-mounted perception system acquires real-time road environment information and traffic participant information ahead of the autonomous vehicle. Road environment information includes the radius of curvature, road surface adhesion coefficient, and slope angle; traffic participant information includes the relative distance, relative speed, and movement trend of target objects. This road environment information and traffic participant information are fused to calculate a comprehensive risk index characterizing the urgency of the current car-following scenario. This comprehensive risk index is input to a pre-defined fuzzy controller, which outputs a preliminary expected acceleration adjustment based on a pre-defined membership function and fuzzy inference rules. The fuzzy controller consists of an input fuzzification unit, a fuzzy inference unit, a rule storage unit, and an output defuzzification unit connected sequentially. The input fuzzification unit converts the continuous numerical form of the comprehensive risk index into a corresponding fuzzy set, which is divided into three categories: low risk, medium risk, and high risk. Each category of fuzzy set is described using a pre-defined continuous membership function. The rule storage unit contains fuzzy inference rules matched to the car-following scenario, with each rule corresponding to a mapping relationship between the input fuzzy set and the output expected acceleration adjustment. The fuzzy inference unit performs fuzzy logic operations based on the input fuzzy set and the inference rules in the rule storage unit and outputs the fuzzy result. The output defuzzification unit converts the fuzzy results output by the fuzzy inference unit into a preliminary expected acceleration adjustment in continuous numerical form. This output can be directly used for subsequent dynamic limiting processing. Based on the road adhesion coefficient, this expected acceleration adjustment is dynamically limited to generate a constrained expected acceleration value. An improved model predictive control algorithm is then invoked, using this constrained expected acceleration value as the core control objective. Combined with the vehicle dynamics model, the final execution control command is obtained. This command directly regulates the vehicle's drive and braking systems. Furthermore, the improved model predictive control algorithm pre-compensates for the vehicle state in the prediction time domain based on the radius of curvature. The improved model predictive control algorithm consists of a state prediction unit, a reference trajectory generation unit, a rolling optimization unit, and a constraint execution unit. The state prediction unit, based on a simplified vehicle dynamics model, calculates the vehicle's future position and velocity point-by-point in the prediction time domain according to the current vehicle state and the road curvature radius. The reference trajectory generation unit, based on the constrained expected acceleration value and combined with the state pre-compensation amount from the radius of curvature, generates a continuous reference state sequence in the prediction time domain. The rolling optimization unit aims to minimize the deviation between the predicted state and the reference state and ensure smooth changes in the control input, completing the optimization calculation cycle by cycle. The constraint execution unit transforms the physical limits of the vehicle's drive and braking systems, the dynamic limits corresponding to the road adhesion coefficient, and the road geometric constraints into boundary conditions for the optimization problem, ensuring that the optimized output control commands are within a safe and feasible range. In each control cycle, the algorithm outputs only the first control input of the optimal control sequence, which is used to directly drive the vehicle's drive and braking systems.

[0072] In one embodiment of the present invention, see [reference] Figure 2 The system extracts the relative distance and speed between the vehicle and the vehicle in front from traffic participant information. Based on the vehicle braking model, it calculates the minimum safe distance according to the relative speed and the current road surface adhesion coefficient. The ratio of the relative distance to the minimum safe distance is then calculated to obtain the first-level risk factor. The system analyzes the movement trend of the target object. If the target object intends to cut in, decelerate, or stop, the second-level risk factor is calculated based on the intensity of the intention and the remaining reaction time. Combining the radius of curvature and the slope angle, the system queries a pre-set curve and slope risk mapping table to obtain the third-level risk factor. A weighted fusion strategy is used to fuse the first, second, and third-level risk factors. The weight of each factor in this weighted fusion strategy is dynamically adjusted according to the current vehicle speed and the road surface adhesion coefficient. The fusion result is the comprehensive risk index.

[0073] In practical implementation, the comprehensive risk index calculation of the speed control method for autonomous vehicles based on fuzzy control involves information fusion at multiple levels. The relative distance and relative speed between the vehicle and the vehicle in front are extracted from traffic participant information. In a specific scenario, if the vehicle's speed is 60 km / h, the vehicle in front's speed is 50 km / h, the relative speed is 10 km / h, the relative distance is 40 meters, and the current road surface adhesion coefficient is 0.8, the following formula can be used to calculate the minimum safe distance based on the vehicle braking model:

[0074]

[0075] in: For minimum safe distance, For relative velocity, The road surface adhesion coefficient, The value is gravitational acceleration. Substituting the values, the minimum safe distance is approximately 25 meters. Then, the ratio of the relative distance to the minimum safe distance, 40 / 25 = 1.6, is calculated to obtain the first-level risk factor. In some embodiments, if the target object intends to cut in and the remaining reaction time is 2 seconds, the second-level risk factor is calculated based on the product of the intent intensity coefficient and the reaction time. For example, when the intensity coefficient is 0.3, the second-level risk factor is 0.6. Combining the radius of curvature of 200 meters and the slope angle of 3 degrees, a pre-set curve and slope risk mapping table is consulted to obtain the third-level risk factor of 0.4. When using a weighted fusion strategy, the weights are dynamically adjusted based on the current vehicle speed and the road surface adhesion coefficient. If a vehicle speed of 60 km / h corresponds to a position weight of 0.5, a speed weight of 0.3, and a risk weight of 0.2, then the comprehensive risk index is 1.6 × 0.5 + 0.6 × 0.3 + 0.4 × 0.2 = 1.06. It is understandable that, in different scenarios, if the relative distance becomes 30 meters and the minimum safe distance remains 25 meters, then the first-level risk factor is 1.2. When other factors remain unchanged, the comprehensive risk index is adjusted to 1.2×0.5+0.6×0.3+0.4×0.2=0.86, which reflects the change in the urgency of the scenario.

[0076] In some embodiments, the motion trend analysis of the target object involves tracking multi-frame perception data. If the brake lights of the vehicle in front are detected to be on and the relative speed increases for three consecutive frames, it is determined that the intention to decelerate is strong. The remaining reaction time is calculated as 3 seconds based on the ratio of relative distance to relative speed, and the second-level risk factor is adjusted to 0.9. The construction of the curve and slope risk mapping table depends on the combined influence of road curvature and slope. For example, when the curvature radius is 150 meters and the slope angle is 5 degrees, the third-level risk factor is set to 0.7. The dynamic adjustment of the weighted fusion strategy is based on vehicle speed segments. When the vehicle speed is higher than 80 kilometers per hour, the weight of the first-level risk factor is increased to 0.6 to strengthen the influence of the safe distance. Optionally, when the road surface adhesion coefficient decreases from 0.8 to 0.5, the minimum safe distance increases to 40 meters, the first-level risk factor decreases from 1.6 to 1.0, and the comprehensive risk index is adjusted accordingly. It can be understood that the calculation results under different parameter combinations reflect the differences in scenarios. For example, under low-adhesion conditions at night, the weight allocation of each factor is further tilted towards the safety side.

[0077] In one embodiment of the present invention, see [reference] Figure 3A simplified vehicle dynamics model incorporating longitudinal kinematics and tire force constraints is established, and the constrained desired acceleration value is used as the reference acceleration at the start of the prediction time domain for the simplified vehicle dynamics model. Within each control cycle of the improved model predictive control algorithm, based on the current vehicle state, the simplified vehicle dynamics model is used to predict the vehicle position and velocity sequences at multiple future sampling times. In the state prediction module of the improved model predictive control algorithm, the prediction time domain is divided into multiple sub-intervals. The theoretical centripetal acceleration change caused by the curve in each sub-interval is calculated based on the radius of curvature. This theoretical centripetal acceleration change is converted into a compensation amount for the longitudinal desired acceleration. A negative compensation amount indicates that deceleration needs to be initiated in advance, while a positive value indicates that appropriate acceleration is permissible. After receiving the road curvature radius, the state prediction unit first determines the pre-compensation intensity within the prediction time domain based on the curvature radius, and then superimposes the pre-compensation amount point by point onto the reference acceleration sequence, allowing the reference trajectory to adapt to the curve driving requirements in advance. The rolling optimization unit compares the pre-compensated reference state sequence with the predicted state sequence output by the vehicle dynamics model to ensure that the optimized execution control command can respond in advance to changes in road curvature, avoiding speed fluctuations and driving risks caused by control lag. In the rolling optimization calculation, the compensation amount is superimposed on the reference acceleration sequence generated based on the constrained expected acceleration value to form a pre-compensated reference state sequence. The predicted vehicle position and velocity sequence is compared with this pre-compensated reference state sequence to construct an optimization problem with the state deviation and the rate of change of control quantity as the objective function. In solving this optimization problem, the predicted lateral displacement is constrained according to the current radius of curvature, and the vehicle gravity component is compensated according to the current slope angle. Solving this optimization problem yields the optimal future control input sequence, and the first control quantity in this sequence is used as the execution control command at the current moment and output to the vehicle's drive system and braking system.

[0078] In practical implementation, the improved model predictive control algorithm relies on a simplified vehicle dynamics model, which includes longitudinal kinematic equations and tire force constraints. The constrained desired acceleration value is used as the reference acceleration at the start of the prediction time domain. Within one control cycle, based on the current vehicle state (e.g., speed 20 m / s, longitudinal position 100 m), the simplified vehicle dynamics model is used to predict the vehicle position and velocity sequence at multiple future sampling moments. For example, with a prediction time domain of 3 seconds and a sampling interval of 0.1 seconds, 30 sets of predicted position and velocity values ​​are obtained. In the state prediction module of the improved model predictive control algorithm, the prediction time domain is divided into multiple sub-intervals, for example, one sub-interval per second. Based on a radius of curvature of 300 meters, the theoretical centripetal acceleration change caused by the curve in each sub-interval is calculated as the vehicle travels along the predetermined trajectory. The calculation formula is as follows:

[0079]

[0080] in: This is the theoretical change in centripetal acceleration. Current vehicle speed Let be the radius of curvature. Substituting the values, the theoretical change in centripetal acceleration is approximately 1.33 m / s². This is converted into a compensation for the desired longitudinal acceleration. A negative compensation value indicates that deceleration needs to be initiated earlier. In the rolling optimization calculation, the compensation is superimposed on the reference acceleration sequence generated based on the constrained desired acceleration values. For example, if the original reference acceleration sequence is [-1.0, -0.8, -0.6] m / s², the superimposed compensation results in a pre-compensated reference state sequence of [-2.33, -2.13, -1.93] m / s².

[0081] In some embodiments, the predicted vehicle position and velocity sequence is compared with the pre-compensated reference state sequence. When constructing the objective function, the predicted vehicle position and velocity at each sampling moment are subtracted from the reference position and reference velocity in the corresponding reference state sequence to obtain the position deviation and velocity deviation. For example, the predicted position of 105 meters differs from the reference position of 103 meters by 2 meters, and the predicted velocity of 19 meters per second differs from the reference velocity of 18.5 meters per second by 0.5 meters per second. In solving the optimization problem, the predicted lateral displacement is constrained based on the current radius of curvature of 300 meters to ensure that the lateral deviation of the vehicle in the curve does not exceed half the lane width. At the same time, the vehicle gravity component is compensated based on the current slope angle of 2 degrees, and the longitudinal component of gravity is added to the dynamic equation. When solving the optimization problem, the weighted sum of squares of the state deviation and the rate of change of the control quantity is minimized. Under the conditions of satisfying the allowable fluctuation range of the expected acceleration value after constraints and the physical limits of the drive system and braking system, the optimal future control input sequence is obtained. For example, the control sequence is [-2000 N, -1800 N, -1600 N], and the first control quantity -2000 N is used as the output of the control command at the current moment.

[0082] Optionally, the tire force constraint in the simplified vehicle dynamics model adopts a linearized model, assuming that the tire force is proportional to the slip ratio, and the proportionality coefficient is determined by the current road surface adhesion coefficient. During state prediction, if the current vehicle speed changes, for example, from 20 m / s to 18 m / s, the theoretical centripetal acceleration change is recalculated, and the compensation amount is adjusted to update the pre-compensated reference state sequence. It can be understood that the constraints of the optimization problem include the lower limit of the position calculated by the safe distance model, for example, the distance to the vehicle in front is not less than 20 meters. If the predicted position is lower than this lower limit, a penalty term is applied in the optimization. In some embodiments, the weight of the rate of change of the control variable is set to 0.1 to balance the control response speed and smoothness. When the radius of curvature decreases to 150 meters, the theoretical centripetal acceleration change increases to 2.67 m / s², and the absolute value of the compensation amount increases, making the pre-compensated reference acceleration sequence further biased towards deceleration, thereby adapting to the curve in advance. Optionally, the number of partitions in the prediction time domain can be adjusted according to the frequency of curvature changes. If the radius of curvature varies significantly within the prediction time domain, the number of sub-intervals can be increased to improve pre-compensation accuracy. It can be understood that by superimposing the compensation amount onto the reference sequence, the improved model predictive control algorithm can proactively consider the influence of the road curvature ahead during optimization, making the generated control commands more closely match actual driving needs.

[0083] In one embodiment of the present invention, based on the real-time detected or estimated road adhesion coefficient, a preset maximum allowable acceleration and maximum allowable deceleration mapping table is consulted to obtain the acceleration limit value and deceleration limit value under the current road conditions; the expected acceleration adjustment amount output by the fuzzy controller is superimposed with the actual acceleration of the current vehicle to obtain the unconstrained original value of expected acceleration; this original value of expected acceleration is compared with the acceleration limit value and the deceleration limit value. If the original value of expected acceleration is greater than the acceleration limit value, the constrained expected acceleration value is set as the acceleration limit value; if the original value of expected acceleration is smaller than the acceleration limit value, the constrained expected acceleration value is set as the acceleration limit value. If the deceleration limit is reached, the constrained desired acceleration value is set as the deceleration limit. If the original desired acceleration value is between the acceleration limit and the deceleration limit, the constrained desired acceleration value is equal to the original desired acceleration value. The constrained desired acceleration value is directly used as the core control target input of the improved model predictive control algorithm. The algorithm uses this target as a benchmark to carry out state prediction and rolling optimization, so that the final output execution control command simultaneously satisfies the road adhesion constraint and the road curvature constraint, ensuring that the acceleration output is always within the allowable range of the vehicle's actual dynamics and that no control command exceeds the vehicle's driving limit. When estimating the road surface adhesion coefficient in real time, the vehicle's wheel speed signal, longitudinal acceleration signal, and output torque signal of the drive motor or braking system are monitored. Based on the vehicle's dynamics, the wheel slip ratio is calculated using the wheel speed signal. Combining the longitudinal acceleration signal and the output torque signal, the longitudinal force between the tire and the road surface is calculated using a back-calculation method. Based on the wheel slip ratio and the calculated longitudinal force, the tire characteristic curve is queried or the tire model is used to estimate the current road surface adhesion coefficient in real time. The estimated road surface adhesion coefficient is then low-pass filtered to eliminate noise interference, resulting in a stable estimate for control decisions.

[0084] In practical implementation, when dynamically limiting the desired acceleration adjustment based on the road surface adhesion coefficient, a preset mapping table is first consulted based on the real-time detected or estimated road surface adhesion coefficient to obtain the acceleration and deceleration limits under the current road conditions. For example, when the road surface adhesion coefficient is 0.8, the maximum allowable acceleration is 2.5 m / s², and the maximum allowable deceleration is -3.0 m / s². If the road surface adhesion coefficient drops to 0.4, the maximum allowable acceleration becomes 1.2 m / s², and the maximum allowable deceleration becomes -1.8 m / s². The desired acceleration adjustment output by the fuzzy controller is then superimposed with the actual acceleration of the vehicle to obtain the unconstrained original value of the desired acceleration. For example, if the fuzzy controller outputs +0.5 m / s², and the current actual acceleration is -0.2 m / s², the unconstrained original value of the desired acceleration is +0.3 m / s². The unconstrained original value of the desired acceleration is then compared with the acceleration limit and deceleration limit. If +0.3 m / s² does not exceed the range of [-1.8 m / s², 1.2 m / s²], the constrained desired acceleration value is taken as +0.3 m / s². If the unconstrained original value of the desired acceleration is +1.5 m / s², it is constrained to 1.2 m / s², see Table 1.

[0085] Table 1: Acceleration Limits for Different Road Surface Adhesion Coefficients

[0086] Road surface adhesion coefficient Maximum permissible acceleration (m / s²) Maximum permissible deceleration (m / s²) 0.9 2.8 -3.5 0.6 1.8 -2.2 0.3 0.9 -1.1

[0087] In some embodiments, the road adhesion coefficient is estimated by monitoring the vehicle's wheel speed signal, longitudinal acceleration signal, and drive motor output torque signal. Based on the vehicle's dynamics, the wheel slip ratio is calculated using the wheel speed signal, and the calculation formula is as follows:

[0088]

[0089] in: For wheel slip ratio, The angular velocity of the wheel. The effective rolling radius of the wheel, This represents the vehicle's longitudinal velocity. Combined with the longitudinal acceleration signal. With output torque signal The longitudinal force between the tire and the road surface is calculated using the reverse calculation method. ,in For vehicle mass. Based on wheel slip ratio. With the calculated longitudinal force By querying tire characteristic curves or using tire models, the current road adhesion coefficient can be estimated in real time. For example, when the slip ratio is 15%, the longitudinal force is 4000 N, and the road adhesion coefficient obtained from the curve is approximately 0.65. The estimated road adhesion coefficient is then low-pass filtered to eliminate noise interference, resulting in a stable estimate used for control decisions.

[0090] Optionally, if the vehicle is equipped with a brake pressure sensor, the brake pressure signal can be used to assist in calculating the longitudinal force, improving the estimation accuracy. When the road surface adhesion coefficient changes abruptly, for example, rapidly decreasing from 0.7 to 0.4, referring to Table 1, the maximum permissible deceleration immediately adjusts from -2.8 m / s² to -1.1 m / s², thus quickly constraining the desired acceleration value in dynamic limiting processing and preventing slippage and loss of control. It can be understood that by continuously monitoring signals such as wheel speed and torque and updating the road surface adhesion coefficient in real time, it can be ensured that the dynamic limiting boundary always matches the current road conditions.

[0091] In one embodiment of the present invention, the vehicle position and speed predicted at each sampling moment in the prediction time domain are subtracted from the reference position and reference speed in the corresponding reference state sequence to obtain the position deviation and speed deviation; weight coefficients are assigned to the position deviation, speed deviation, and rate of change of the control quantity in the objective function, and these weight coefficients are adaptively adjusted according to the current vehicle speed and comprehensive risk index; an objective function is constructed, which is the sum of the weighted squares of the position deviation, the weighted squares of the speed deviation, and the weighted squares of the rate of change of the control quantity at all sampling moments in the prediction time domain; the constraints of the optimization problem include the allowable fluctuation range of the desired acceleration value after constraints, the physical limits of the vehicle drive system and braking system, and the lower limit constraint of the position calculated based on the safe distance model; The adaptive adjustment of weight coefficients is achieved by setting a set of basic weight coefficients, including basic weights for position deviation, speed deviation, and control change rate. A weight adjustment coefficient table indexed by vehicle speed range and comprehensive risk index level is established. Based on the current real-time vehicle speed and the calculated comprehensive risk index, the weight adjustment coefficient table is queried to obtain the position weight adjustment factor, speed weight adjustment factor, and control change rate weight adjustment factor. The position deviation basic weight and the position weight adjustment factor are multiplied to obtain the final applied position deviation weight coefficient. The speed deviation basic weight and the speed weight adjustment factor are multiplied to obtain the final applied speed deviation weight coefficient. The control change rate basic weight and the control change rate weight adjustment factor are multiplied to obtain the final applied control change rate weight coefficient.

[0092] In practical implementation, when constructing an optimization problem with state deviation and control variable change rate as objective functions, the predicted vehicle position and speed at each sampling moment in the prediction time domain are subtracted from the reference position and speed in the corresponding reference state sequence to obtain the position deviation and speed deviation. For example, with a prediction time domain of 3 seconds and a sampling interval of 0.1 seconds, if the predicted vehicle position at a certain sampling moment is 205 meters and the reference position is 203 meters, the difference between the two is a position deviation of 2 meters; if the predicted vehicle speed is 19.5 meters per second and the reference speed is 19.0 meters per second, the difference between the two is a speed deviation of 0.5 meters per second. Weighting coefficients are assigned to the position deviation, speed deviation, and control variable change rate, and these weighting coefficients are adaptively adjusted according to the current vehicle speed and comprehensive risk index. The constructed objective function is expressed as the sum of the weighted squares of the position deviation, weighted squares of the speed deviation, and weighted squares of the control variable change rate at all sampling moments in the prediction time domain, and the mathematical expression is:

[0093]

[0094] in: The objective function value, To predict the total number of sampling steps in the time domain, For sampling time index, For the first Step position deviation, For the first Step speed deviation, For the first The rate of change of the control quantity at each step For the first The positional deviation weighting coefficient of the step. For the first Step speed deviation weighting coefficient, For the first The control variable change rate weighting coefficient is used for each step. The constraints of the optimization problem include the allowable fluctuation range of the desired acceleration value after constraints, the physical limits of the vehicle's drive and braking systems, and the lower position limit constraint calculated based on the safe distance model. For example, the maximum drive torque of the drive motor is 3000 Nm, the maximum braking force of the braking system is 5000 Nm, and the lower position limit calculated by the safe distance model is to maintain a distance of at least 20 meters from the vehicle in front.

[0095] In some embodiments, adaptive adjustment of weighting coefficients is achieved by setting a basic set of weighting coefficients, which includes basic weights for position deviation, speed deviation, and control rate of change. A weighting adjustment coefficient table indexed by vehicle speed range and comprehensive risk index level is established. Based on the current real-time vehicle speed and the calculated comprehensive risk index, this weighting adjustment coefficient table is queried to obtain the position weighting adjustment factor, speed weighting adjustment factor, and control rate of change weighting adjustment factor. The basic weight for position deviation is multiplied by the position weighting adjustment factor to obtain the final applied position deviation weighting coefficient; the basic weight for speed deviation is multiplied by the speed weighting adjustment factor to obtain the final applied speed deviation weighting coefficient; and the basic weight for control rate of change is multiplied by the control rate of change weighting adjustment factor to obtain the final applied control rate of change weighting coefficient. For example, if the basic weight for position deviation is set to 0.6, the basic weight for speed deviation to 0.3, and the basic weight for control change rate to 0.1; when the vehicle speed is in the range of 70-90 km / h and the comprehensive risk index is at a high risk level, the position weight adjustment factor is 1.2, the speed weight adjustment factor is 0.8, and the control change rate weight adjustment factor is 1.5, as shown in Table 2.

[0096] Table 2: Weight Adjustment Coefficients under Different Operating Conditions

[0097] Speed ​​range (km / h) Overall risk level Position weight adjustment factor Speed ​​weight adjustment factor Control change rate weighting adjustment factor 50-70 Low risk 1.0 1.0 1.0 70-90 High risk 1.2 0.8 1.5 >90 Extremely high risk 1.5 0.5 2.0

[0098] Optionally, the set of basic weight coefficients can be preset according to different control modes. For example, in comfort mode, the basic weight of the control change rate is set larger to prioritize smoothness. When the vehicle speed is low and the comprehensive risk index shows low risk, such as in urban congestion, the weight adjustment factors tend to be balanced, maintaining the basic weight ratio and focusing on overall stability. It can be understood that through dynamic adjustment of the weight coefficients, the objective function can flexibly respond to the needs of different driving scenarios, increasing the weight of safety-related state deviations in emergency situations and emphasizing the smoothness of control in stable conditions.

[0099] In one embodiment of the present invention, the input comprehensive risk index, the output expected acceleration adjustment amount, and the final actual acceleration response and following stability evaluation of the vehicle are recorded in each control cycle. When the actual following distance is lower than the comfort threshold for a long time or the acceleration change rate exceeds the smoothness threshold, the current control effect is determined to be unsatisfactory. For the fuzzy set where the comprehensive risk index corresponding to the unsatisfactory control effect is located, the conclusion part of the corresponding fuzzy inference rule in the fuzzy controller is slightly modified. The direction of the modification is to make the output expected acceleration adjustment amount more inclined to be smooth or conservative, and the magnitude of the modification amount is positively correlated with the duration of the unsatisfactory degree. The modified fuzzy rule is updated to the rule base of the fuzzy controller for subsequent control decisions.

[0100] In practice, the online fine-tuning of fuzzy rules based on historical control data is achieved by recording the input, output, and actual performance data for each control cycle. Specifically, the system records the comprehensive risk index input in each control cycle, the expected acceleration adjustment output by the fuzzy controller, and the final vehicle's actual acceleration response and following stability evaluation. For example, in a certain control cycle, the recorded input comprehensive risk index is 0.85, the expected acceleration adjustment output is -1.2 m / s², the actual vehicle acceleration response is -1.0 m / s², and the following stability evaluation index is an acceleration change rate of 0.8 m / s³. When the actual following distance is consistently below the comfort threshold or the acceleration change rate exceeds the smoothness threshold, the current control effect is deemed unsatisfactory. For example, if the actual following distance is consistently below the set comfort threshold of 15 meters for 10 consecutive control cycles, or the acceleration change rate consistently exceeds 1.0 m / s³, the rule correction process is triggered.

[0101] In some embodiments, for the fuzzy set where the comprehensive risk index corresponds to the unsatisfactory control effect, the conclusion part of the corresponding fuzzy inference rule in the fuzzy controller is slightly modified. For example, if the comprehensive risk index mainly falls in the "medium risk" fuzzy set when the control effect is unsatisfactory, the original conclusion of the corresponding fuzzy rule is "the expected acceleration adjustment is negative and large". The modification direction is to make the output expected acceleration adjustment more inclined to be smooth or conservative, that is, to change it to "the expected acceleration adjustment is negative and medium". The magnitude of the modification is positively correlated with the duration of the unsatisfactory degree, which can be expressed by the formula:

[0102]

[0103] in: This represents the amount of correction to the rule's conclusion. This is the proportionality coefficient. The duration of the undesirable condition is defined as follows: If the undesirable duration is 10 seconds and the scaling factor is set to 0.05, the correction amount is 0.5, corresponding to shifting the membership center of the fuzzy rule conclusion by 0.5 units in a conservative direction. The corrected fuzzy rule is then updated in the rule base of the fuzzy controller for subsequent control decisions. For example, in the next control cycle, when a similar comprehensive risk index is encountered again, the fuzzy controller will output a smoother expected acceleration adjustment amount based on the updated rule.

[0104] Optionally, in addition to the rate of change of acceleration, the stability evaluation index can also include a passenger comfort score, calculated from vibration data collected by in-vehicle sensors. If the undesirable condition persists for a long time, such as exceeding 20 seconds, the correction amount is increased proportionally to accelerate the convergence of the rules to a better state. It can be understood that by continuously recording operational data and dynamically adjusting fuzzy rules, the control system can gradually adapt to specific driving scenarios or driver preferences, improving long-term control performance.

[0105] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications 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 protection scope of the present invention.

Claims

1. A speed control method for unmanned vehicles based on fuzzy control, characterized in that, The method includes: The vehicle-mounted perception system acquires real-time information on the road environment and traffic participants in front of the autonomous vehicle. The road environment information includes the radius of curvature, the road surface adhesion coefficient, and the slope angle. The traffic participant information includes the relative distance, relative speed, and movement trend of the target object. The road environment information and traffic participant information are fused and processed to calculate a comprehensive risk index that characterizes the urgency of the current car-following scenario; The comprehensive risk index is input into a preset fuzzy controller, which outputs an initial expected acceleration adjustment amount based on a preset membership function and fuzzy inference rules. The desired acceleration adjustment amount is dynamically limited based on the road surface adhesion coefficient to generate a constrained desired acceleration value; An improved model predictive control algorithm is invoked, with the constrained desired acceleration value as the core control objective. Combined with the vehicle dynamics model, the final execution control command is obtained. The execution control command is used to directly regulate the vehicle's drive system and braking system. The improved model predictive control algorithm performs pre-compensation on the vehicle state in the prediction time domain based on the radius of curvature.

2. The speed control method for unmanned vehicles based on fuzzy control according to claim 1, characterized in that, The road environment information and traffic participant information are fused and processed to calculate a comprehensive risk index characterizing the urgency of the current car-following scenario, including: Extract the relative distance and relative speed between the vehicle and the vehicle in front from the traffic participant information; Based on the vehicle braking model, the minimum safe distance is calculated according to the relative speed and the current road surface adhesion coefficient. Calculate the ratio of the relative distance to the minimum safe distance to obtain the first-level risk factor; Analyze the movement trend of the target object. If the target object has the intention to cut in, decelerate, or stop, then calculate the second-level risk factor based on the intensity of the intention and the remaining reaction time. Combining the radius of curvature and slope angle, the third-level risk factor is obtained by querying the pre-set curve and slope risk mapping table; A weighted fusion strategy is adopted to fuse the first-level risk factor, the second-level risk factor, and the third-level risk factor. The weight of each factor in the weighted fusion strategy is dynamically adjusted according to the current vehicle speed and the road surface adhesion coefficient. The fusion result is the comprehensive risk index.

3. The speed control method for unmanned vehicles based on fuzzy control according to claim 1, characterized in that, The improved model predictive control algorithm is invoked, with the constrained desired acceleration value as the core control objective. Combined with the vehicle dynamics model, the final control commands are obtained, including: A simplified vehicle dynamics model incorporating longitudinal kinematics and tire force constraints is established, and the constrained desired acceleration value is used as the reference acceleration of the simplified vehicle dynamics model at the start of the prediction time domain. Within each control cycle of the improved model predictive control algorithm, based on the current vehicle state, the vehicle dynamics simplified model is used to predict the vehicle position and speed sequence at multiple future sampling times. The predicted vehicle position and velocity sequence is compared with the reference state sequence generated based on the expected acceleration value after the constraints, and an optimization problem is constructed with the state deviation and the rate of change of control quantity as the objective function. In the process of solving the optimization problem, the predicted lateral displacement is constrained based on the current radius of curvature, and the vehicle gravity component is compensated based on the current slope angle. Solve the optimization problem to obtain the optimal future control input sequence, and use the first control quantity in the optimal future control input sequence as the execution control command at the current moment, and output it to the vehicle's drive system and braking system.

4. The speed control method for unmanned vehicles based on fuzzy control according to claim 1, characterized in that, The improved model predictive control algorithm pre-compensates for the vehicle state in the prediction time domain based on the radius of curvature, including: In the state prediction module of the improved model predictive control algorithm, the prediction time domain is divided into multiple sub-intervals; Based on the radius of curvature, calculate the theoretical change in centripetal acceleration caused by the curve in each sub-interval when the vehicle travels along the predetermined trajectory. The theoretical centripetal acceleration change is converted into a compensation amount for the longitudinal desired acceleration. A negative compensation amount indicates that deceleration needs to be done in advance, while a positive compensation amount indicates that appropriate acceleration is allowed. In the rolling optimization calculation, the compensation amount is superimposed on the reference acceleration sequence generated based on the desired acceleration value after the constraint to form a pre-compensated reference state sequence. The pre-compensated reference state sequence and the state sequence predicted by the vehicle dynamics model are used for optimization, so that the generated control commands can actively adapt to changes in the curvature of the road ahead.

5. The speed control method for unmanned vehicles based on fuzzy control according to claim 1, characterized in that, Based on the road surface adhesion coefficient, the desired acceleration adjustment amount is dynamically limited to generate a constrained desired acceleration value, including: Based on the real-time detected or estimated road adhesion coefficient, the preset maximum allowable acceleration and maximum allowable deceleration mapping table is queried to obtain the acceleration limit and deceleration limit under the current road conditions. The desired acceleration adjustment output by the fuzzy controller is superimposed with the actual acceleration of the current vehicle to obtain the unconstrained original value of the desired acceleration. Compare the original value of the desired acceleration with the acceleration limit value and the deceleration limit value; If the original value of the desired acceleration is greater than the acceleration limit value, then the constrained desired acceleration value is set as the acceleration limit value; If the original value of the desired acceleration is less than the deceleration limit value, then the constrained desired acceleration value is set as the deceleration limit value; If the original value of the desired acceleration is between the acceleration limit value and the deceleration limit value, then the constrained desired acceleration value is equal to the original value of the desired acceleration.

6. The speed control method for unmanned vehicles based on fuzzy control according to claim 5, characterized in that, The method further includes a step of real-time estimation of the road surface adhesion coefficient: Monitor the vehicle's wheel speed signal, longitudinal acceleration signal, and output torque signal of the drive motor or braking system; Based on the vehicle's dynamics, the wheel slip ratio is calculated using the wheel speed signal; By combining the longitudinal acceleration signal and the output torque signal, the longitudinal force between the tire and the road surface is calculated using a reverse calculation method. Based on the wheel slip ratio and the calculated longitudinal force, the tire characteristic curve can be queried or the tire model can be used to estimate the current road adhesion coefficient in real time. The estimated road surface adhesion coefficient is subjected to low-pass filtering to eliminate noise interference and obtain a stable estimate for control decision-making.

7. The speed control method for unmanned vehicles based on fuzzy control according to claim 3, characterized in that, The predicted vehicle position and velocity sequence is compared with a reference state sequence generated based on the desired acceleration value after the constraints, and an optimization problem is constructed with the state deviation and the rate of change of control quantity as objective functions, including: The position deviation and speed deviation are obtained by subtracting the vehicle position and speed predicted at each sampling time in the prediction time domain from the reference position and reference speed in the reference state sequence at the corresponding time. Weighting coefficients are assigned to the position deviation, speed deviation, and rate of change of the control quantity, and the weighting coefficients are adaptively adjusted according to the current vehicle speed and the comprehensive risk index. Construct an objective function, which is the sum of the weighted squared position deviation, the weighted squared velocity deviation, and the squared rate of change of the weighted control quantity at all sampling times in the prediction time domain; The constraints of the optimization problem include the allowable fluctuation range of the desired acceleration value after the constraints, the physical limits of the vehicle drive system and braking system, and the lower limit constraint of the position calculated based on the safe distance model.

8. The speed control method for unmanned vehicles based on fuzzy control according to claim 7, characterized in that, The weighting coefficients are adaptively adjusted based on the current vehicle speed and the comprehensive risk index, including: Define a set of basic weight coefficients, including basic weights for position deviation, velocity deviation, and control rate of change; Establish a weight adjustment coefficient table, which is indexed by vehicle speed range and comprehensive risk index level; Based on the current real-time vehicle speed and the calculated comprehensive risk index, the weight adjustment coefficient table is consulted to obtain the position weight adjustment factor, speed weight adjustment factor and control rate of change weight adjustment factor. Multiply the base weight of the position deviation by the position weight adjustment factor to obtain the final applied position deviation weight coefficient; Multiply the speed deviation base weight by the speed weight adjustment factor to obtain the final applied speed deviation weight coefficient; Multiply the basic weight of the control change rate by the control change rate weight adjustment factor to obtain the final applied control change rate weight coefficient.

9. The speed control method for unmanned vehicles based on fuzzy control according to claim 1, characterized in that, The method also includes an online fine-tuning step for fuzzy rules based on historical control data: Record the input of the comprehensive risk index, the output of the expected acceleration adjustment amount, and the final actual acceleration response and following stability evaluation of the vehicle in each control cycle; When the actual following distance is below the comfort threshold for an extended period or the rate of change of acceleration exceeds the smoothness threshold, the current control effect is deemed unsatisfactory. For the fuzzy set where the comprehensive risk index is located when the control effect is not ideal, the conclusion part of the corresponding fuzzy inference rule in the fuzzy controller is slightly modified. The direction of the correction is to make the desired acceleration adjustment in the output more smooth or conservative, and the magnitude of the correction is positively correlated with the duration of the non-ideal degree. The revised fuzzy rules are updated in the rule base of the fuzzy controller for subsequent control decisions.

10. A speed control system for an unmanned vehicle based on fuzzy control, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the speed control method for unmanned vehicles based on fuzzy control as described in any one of claims 1 to 9.

Images